Donate
Independent, objective, nonpartisan research

R 1105RVR

Authors

R 1105RVR

Tagged with:

Publication PDFs

Database

This is the content currently stored in the post and postmeta tables.

View live version

object(Timber\Post)#3742 (44) { ["ImageClass"]=> string(12) "Timber\Image" ["PostClass"]=> string(11) "Timber\Post" ["TermClass"]=> string(11) "Timber\Term" ["object_type"]=> string(4) "post" ["custom"]=> array(5) { ["_wp_attached_file"]=> string(13) "R_1105RVR.pdf" ["wpmf_size"]=> string(7) "1778899" ["wpmf_filetype"]=> string(3) "pdf" ["wpmf_order"]=> string(1) "0" ["searchwp_content"]=> string(307598) " The Impact of Provider Choice on Workers’ Compensation Costs and Outcomes Richard A. Victor Peter S. Barth David Neumark With the assistance of Te-Chun Liu WC-05-14 November 2005 Workers Compensation Research Institute Cambridge, Massachusetts Public Policy Institute of California San Francisco, California copyright 2005 © by the workers compensation research institute and public policy institute of california all rights reserved. Library of Congress Cataloging-in-Publication Data Victor, Richard A. The impact of provider choice on workers’ compensation costs and outcomes / Richard Victor, Peter Barth, David Neumark; with the assistance of Te-Chun Liu. p. cm. “WC-05-14.” A study that utilizes data taken from interviews in 2002 and 2003 of employees in California, Texas, Massachusetts, and Pennsylvania. Includes bibliographical references. ISBN 1-931906-39-4 1. Workers’ compensation — United States — Costs. 2. Medical care, Cost of — United States. 3. Physician services utilization — United States. I. Barth, Peter S., 1937II. Neumark, David. III. Workers Compensation Research Institute (Cambridge, Mass.) IV. Title. HD7103.65.U6V494 2005 368.4’101’0973 — dc22 2005049410 publications of the workers compensation research institute do not necessarily reflect the opinions or policies of the institute’s research sponsors. ppic does not take or support positions on any ballot measure or on any local, state, or federal legislation, nor does it endorse, support, or oppose any political parties or candidates for public office. research publications reflect the views of the authors and do not necessarily reflect the views of the staff, officers, or board of directors of the public policy institute of california. short sections of text, not to exceed three paragraphs, may be quoted without written permission provided that full attribution is given to the source and the above copyright notice is included. Acknowledgments This study benefited greatly from the contributions of a number of our colleagues at the Workers Compensation Research Institute (WCRI), the Public Policy Institute of California (PPIC), and elsewhere. We especially appreciate the thoughtful comments of technical reviewers Dr. Leslie Boden of Boston University, Dr. Jeffrey Harris of J. Harris Associates, Inc., and Dr. Allan Hunt of the Upjohn Institute. We are also grateful for comments from Mark Baldassare, Jon Haveman, Joyce Peterson, and Fred Silva, all of PPIC, and seminar participants at PPIC. We also wish to thank Linda Carrubba for her excellent assistance in preparing the draft with precision and good humor, and Jill McNamee, who managed the review and publication process. We are indebted to Barbara McGowran for editing our prose to improve its readability and accuracy, and to Jan Cocker for proofreading the final report. Of course, any errors in the study remain our responsibility. Richard A. Victor Cambridge, Massachusetts Peter S. Barth Storrs, Connecticut David Neumark San Francisco, California November 2005 iii Table of Contents List of Tables ix List of Figures xiii Executive Summary xv 1. Introduction Policy Context / 3 Objectives and Scope of This Study / 4 Organization of the Report / 7 3 2. Key Concepts, Data, and Methods Key Concepts / 9 outcomes / 9 costs / 10 provider choice / 10 other terms / 11 Data Sources / 11 Measuring Provider Choice / 12 Cost and Outcome Measures / 18 Empirical Methods / 20 basic model and control variables / 20 variation in state workers’ compensation systems / 23 appropriate statistical models for each dependent variable / 24 equality of effects of provider choice across states / 25 causal inferences about policy changes regarding provider choice / 25 9 v vi t a b l e o f c o n t e n t s 3. Impact of Employee or Employer Choice of Provider: Main Results Summary of Findings / 34 Impact of Employee or Employer Choice of Provider on Costs and Outcomes: Main Results / 35 33 4. Employee Choice of Prior Provider or New Provider, or Employer Choice of Provider: Main Results 39 Patterns of Choosing New and Prior Providers / 40 Summary of Findings / 41 Impact on Costs and Outcomes of Employee-Selected Prior Provider or New Provider, or Employer Choice of Provider: Main Results / 41 Revisiting the Question of Unmeasured Residual Injury Severity / 45 5. Impact of Provider Choice on Costs and Outcomes: Results for Individual States California / 48 Texas / 52 Massachusetts / 55 Pennsylvania / 58 47 6. Worker Satisfaction with Health Care Correlates of Satisfaction and Overall Health Care Received / 62 Correlates of Satisfaction and Provider Choice / 65 Satisfaction with Care: Prior or New Provider / 67 61 7. Discussion and Policy Implications Summary of Results / 70 Interpretative Caveats / 71 Implications for Public Policy / 72 69 table of contents vii Technical Appendix A: Literature Review Technical Appendix B: Variables Used in Study: Definitions and Descriptive Statistics Technical Appendix C: Discussion of Construction and Validity of Health Status, Recovery, and Perceived Severity Measures Technical Appendix D: Discussion of Survey Response Rates and Response Bias Technical Appendix E: Statistical Methods Technical Appendix F: Selected State System Features Technical Appendix G: Tests of Pooling versus Individual State Regressions Technical Appendix H: Full Regression Results References About the Authors Related PPIC Publication Related WCRI Publications 77 81 89 97 107 111 121 129 145 149 150 151 List of Tables 2.1 Pattern of Providers Who Delivered Nonemergency Care / 13 2.2 Who Chose the Respondent’s Primary Provider? / 15 2.3 Provider Choice, by Type of Provider / 16 2.4 Costs and Health Outcomes, by State / 19 2.5 Substantial Return to Work and Duration of Period Out of Work / 20 2.6 Satisfaction with Overall Health Care / 21 2.7 Determinants of Provider Choice, Four States Combined / 29 3.1 Impact of Employee Choice Compared with Employer Choice / 36 3.2 Satisfaction with Overall Care, by Who Selected the Provider and by State / 37 4.1 Employee Choice of Prior Provider or New Provider / 40 4.2 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Four States Combined / 42 4.3 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider / 45 4.4 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Four States Combined, Excluding Severity and Injury Measures / 46 5.1 Impact of Employee Choice Compared with Employer Choice, California / 49 5.2 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, California / 50 5.3 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, California / 51 5.4 Impact of Employee Choice Compared with Employer Choice, Texas / 53 ix x list of tables 5.5 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Texas / 54 5.6 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Texas / 55 5.7 Impact of Employee Choice Compared with Employer Choice, Massachusetts / 56 5.8 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Massachusetts / 57 5.9 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Massachusetts / 57 5.10 Impact of Employee Choice Compared with Employer Choice, Pennsylvania / 58 5.11 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Pennsylvania / 59 5.12 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Pennsylvania / 60 6.1 Satisfaction with Care, by Degree of Recovery of Physical Health / 64 6.2 Worker’s Perception of Timing of Return to Work and Satisfaction with Care / 64 6.3 Second Absence Due to Injury and Satisfaction with Care / 65 6.4 Completeness of Recovery, by Who Selected the Provider / 66 6.5 Worker’s Perception of Timing of Return to Work, by Who Selected the Provider / 66 6.6 Second Absence Due to Injury, by Who Chose the Provider / 67 B.1 Definitions of Variables / 82 B.2 Descriptive Statistics / 85 C.1 Incurred Costs for Medical Care, by Perceived Injury Severity / 95 C.2 Comparison of Two Surveys on General Health / 96 D.1 Attempted Telephone Interviews and Valid Phone Numbers / 99 D.2 Disposition of Cases with Valid Phone Numbers / 100 D.3 Analysis of Representativeness / 101 D.4 Analysis of Response Bias / 105 F.1 Coverage under the State Workers’ Compensation Laws, 2004 / 112 F.2 Medical Cost Containment Strategies, 2004 / 113 list of tables xi F.3 Statutory Benefit Parameters for Weekly Temporary and Permanent Partial Disability Payments / 115 F.4 Waiting Period and Limits on Duration of Temporary Disability Benefits, 2002 / 118 F.5 Comparative Statistics on Costs, Injury Frequency, and Other Metrics of Workers’ Compensation Systems / 119 G.1 Two-Way Classification of Provider Choice, Effects of Provider Choice on Outcomes, Tests of Pooling Restrictions / 122 G.2 Three-Way Classification of Provider Choice, Effects of Provider Choice on Outcomes, Tests of Pooling Restrictions / 123 G.3 Impact of Employee Choice Compared with Employer Choice / 125 G.4 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice / 126 H.1 Determinants of Employee vs. Employer Choice of Provider, by State, Odds Ratios from Logit Models / 130 H.2 Determinants of Employee Choice of Prior and New Providers vs. Employer Choice of Provider, by State, Odds Ratios from Multinomial Logit Models / 132 H.3 Effects of Provider Choice on Medical and Indemnity Benefits / 135 H.4 Effects of Provider Choice on Duration and Substantial Return to Work / 138 H.5 Effects of Provider Choice on Recovery and Satisfaction / 141 List of Figures 2.1 Provider Choice Questions / 14 C.1 Perceived Injury Severity and Recovery of Physical Health and Functioning: An Example from the Texas Results / 93 xiii Executive Summary With workers’ compensation medical payments rising rapidly in many states (Telles, Wang, and Tanabe, 2004), policymakers have intensified their efforts to modify state laws to try to reduce those costs, while avoiding actions that might impair the outcomes experienced by injured workers. One of the actions often debated is giving employers more influence or direct control over the selection of providers. The health care provider plays many critical roles in the outcome of a workers’ compensation case. Those roles can include giving information that bears directly on most aspects of a claim for medical and indemnity benefits; diagnosing the condition and assessing its cause, which can affect the compensability of the claim; prescribing and providing a course of treatment and disability management practices, which can influence whether the worker returns to work and how quickly; assessing whether the worker’s condition has reached maximum medical improvement, whether the worker is left with a permanent impairment or disability, and the extent of the impairment; and judging whether a preexisting condition contributed to the degree of impairment. From the perspective of either the employer or the worker, any of these decisions by the health care provider can be sufficiently important to warrant being able to control the selection decision. Thus, the selection of that provider is an important matter for all parties of interest. Workers and their advocates have argued that the choice of the treating doctor or provider should be left to the worker. At a minimum, they argue that workers should be treated by those whom they trust and whose interests align with the workers’ — interests that encourage prompt return to work, but only as medically indicated, and the fullest restoration possible of physical capacity (Ellenberger, 1992). In contrast, employer advocates believe the choice of provider should be made by the employer, arguing that employer choice ensures that incentives exist for keeping the costs of care reasonable and appropriate (Morrison, 1990), employer choice helps avoid excessive services and treatments, and providers familiar with the employer’s workplace can use that knowledge to expedite return to work (National Federation of Independent Business Research Foundation and Na- xv xvi e x e c u t i v e s u m m a r y tional Foundation for Unemployment Compensation and Workers’ Compensation, n.d.). From the late 1980s to the early 1990s — a period of rising costs — a number of states modified “employee choice” laws to require that workers select providers from within approved networks of providers created by employers. The important role of provider choice in workers’ compensation public policy debates was highlighted recently by the passage of Senate Bill (SB) 899 in California. Until recently, California employers had the right to select the initial provider, unless the employee had predesignated a provider; but after 30 days, the worker had the right to change to a medical provider of his or her own choice. However, SB 899 changed the rules regarding provider choice.1 In particular, employers are now allowed to establish networks composed of occupational and nonoccupational physicians, and the legislation grants to the employer (or the insurer) the sole right to decide which medical providers are in the network. Further, the worker’s right to choose a physician after 30 days no longer applies if a network is established that complies with the law, unless the worker has predesignated a physician under particular conditions. As long as employers establish networks, which many are expected to do, California workers will have less flexibility to choose their providers — especially new providers. Objectives of This Study The purpose of this study is to determine if measurable costs and outcomes in workers’ compensation cases are affected by who selects the health care provider. The costs and outcomes we study include medical and indemnity costs, the duration of time out of work, the likelihood that the worker had a sustainable return to work, the worker’s perception of the degree of recovery from the work injury, and the worker’s overall satisfaction with the health care received. This study has at least four important advantages over the few previous studies that have attempted to answer these questions. First, it utilizes data taken from employee interviews conducted in 2002 and 2003 in four states: California, Texas, Massachusetts, and Pennsylvania. Workers were asked to identify who selected 1 This was one of many reforms, some of which were included in legislation passed the previous year (SB 228), addressing the rapid escalation in workers’ compensation costs in California that began in 1999 (Neumark, 2005). executive summary xvii their health care providers. In contrast, the previous studies classify provider choice based on state statutory provisions, despite the fact that the statutory provisions are imperfectly related to actual choice of provider exercised by a worker. Second, we focus on the primary provider of medical care, who is often different from the initial provider playing a subsidiary role. Third, we link the data from the interviews to claims data supplied by the claims payors, providing information on factors that include medical and indemnity costs, medical treatments, and employer attributes, among others. A more complete picture of the claim from the vantage point of both the worker and the employer should help to better establish the consequences of provider choice. Fourth, the existing studies are limited to estimating the effects of provider choice on costs, whereas we look at a wider set of outcomes of concern to policymakers and stakeholders. Finally, the interview data also indicate whether the primary provider had previously treated the worker for an unrelated condition. As noted earlier, recent legislation in California recognized the distinction between employee choice of a prior provider and a new provider. Thus, our findings with regard to the consequences of employee choice of prior providers, employee choice of new providers, and employer choice are especially salient to assessing the likely consequences of this very important component of workers’ compensation reforms in California. Those reforms are just beginning to be implemented and will likely be questioned and reexamined as the state implements the reforms enacted to deal with the workers’ compensation crisis that emerged in California at the beginning of this decade. Summary of Results These are among the most important findings of this study: ■ Comparing cases in which the worker selected the primary provider with otherwise similar cases in which the employer selected the provider, we found that costs were generally higher and return-to-work outcomes poorer when the worker selected the provider. Workers reported higher rates of satisfaction with overall care but similar perceived recovery of physical health. ■ Compared with cases in which the employer selected the provider, cases in which the worker selected a provider who had treated the worker previously for an unrelated condition (prior provider) may have had higher xviii e x e c u t i v e s u m m a r y costs, but the evidence was weak. Worker outcomes did not appear to be very different between cases with employee-selected prior providers and those with employer-selected providers, except that satisfaction with overall care was higher when the worker saw a prior provider. ■ Compared with cases in which the employer selected the provider, cases in which the worker selected a provider who had not treated him or her previously (new provider) had much higher costs and poorer return-to-work outcomes, generally no differences in physical recovery, and higher levels of satisfaction with overall care. ■ Comparing cases in which the employee selected a prior provider with similar cases in which the employee chose a new provider, we found that the worker treated by a new provider was less likely to return to work, returned to work more slowly if he or she did return, had lower levels of satisfaction with overall care, and experienced no better physical recovery. Medical costs were similar in both cases, but indemnity costs per claim were higher for a worker treated by a new provider, although this evidence was not as strong statistically as the other results. These primary findings come from the combined four-state sample, because the combined data lead to a larger sample and more precise estimates than do the data from the individual states. However, we also examine results from individual states, even though they are less precise and the statistical tests are less powerful. The following are among the findings from the state-by-state analyses: ■ In the two states with higher-than-typical medical payments, California and Texas, comparing cases in which the worker selected the primary provider with otherwise similar cases in which the employer selected the provider, we found that costs were generally higher and return-to-work outcomes poorer when workers selected their providers, although workers reported higher rates of satisfaction with overall care and similar perceived recovery of physical health. ■ The findings for the combined four-state sample previously described — comparing employee-selected prior providers, new providers, and employer-selected providers — were especially strong for California and Texas, although in Texas, both cases with employee-selected prior providers and those with new providers had higher costs than cases with employer-selected providers. executive summary xix Implications for Public Policy How do these results inform public policy debates regarding choice of provider in workers’ compensation? We found some evidence to support both those who advocate for employee choice and those who advocate for employer choice. As described here, however, based on our findings, it appears possible to improve the design of provider choice laws to lower costs and improve return-to-work outcomes without adversely affecting physical recovery from workplace injuries. First, we found that when the worker chose the provider, costs were higher, recovery of health outcomes was not better, and return-to-work outcomes were often worse than when the employer selected the provider. This finding suggests that employers, on average, may be well positioned to select good-quality, lower-cost providers — or at least better positioned than many workers. The finding also suggests that employers, in practice, are not generally selecting inferior-quality providers; although there may be exceptions, they do not appear to be frequent enough to affect the overall results. Second, we found that when workers select new providers — those they had not been treated by previously — costs were higher and return to work outcomes were poorer. This evidence suggests that state laws that grant employers greater influence over the choice of provider should lead to lower costs and better returnto-work outcomes than laws that allow workers to select providers whom they have not seen previously. Third, we found that when workers selected providers with whom they had a preexisting clinical relationship, the costs and most outcomes were not dramatically different than when the employer selected the provider. However, when workers selected providers — either prior or new — they expressed higher levels of satisfaction with care. We are not surprised by this finding regarding workers choosing prior providers, because a key issue is the likelihood that a worker will be seen by a provider who has the appropriate training and skills, is trusted by the worker, and delivers appropriate care. More surprising, though, is that workers also expressed greater satisfaction when they selected new providers (relative to employers choosing). We explored whether this greater satisfaction appeared to be related to dimensions of physical recovery not captured in our data or assistance in remaining out of work beyond the necessary time following an injury, but we were able to rule out such explanations. There may, however, be alternative explanations related to empowerment, trust, or the process of care that leaves workers more satisfied with their new-provider choices, even though costs and return-to-work outcomes appear to be worse and physical recovery no better. xx e x e c u t i v e s u m m a r y The results for California paralleled those for the larger sample in providing some evidence suggesting that the costs were higher and return-to-work outcomes were worse when workers selected providers with whom they had no prior relationship. This suggests that the recent legislative changes in California — which significantly expanded the limits on worker choice of provider but retain an exception where there is a preexisting provider relationship — may have struck an appropriate balance. The predesignation exception to the worker’s choice of provider among an employer-designated network of providers bears some resemblance to the prior-provider category that we analyze. However, this prior-provider category is broader than the California requirements for predesignation — for example, the prior provider has to be the personal physician of the worker under a nonoccupational group health insurance plan offered by the employer. Thus, the results are potentially quite informative about the likely effects of the changes in provider choice recently enacted in California, but they do not provide a direct test of the impact of the reforms. Such a direct test will not be possible until some time after the reforms are implemented, probably at least a couple of years from now. The Impact of Provider Choice on Workers’ Compensation Costs and Outcomes 1 Introduction Policy Context Because health care costs in workers’ compensation have grown rapidly and have become an increasingly important proportion of system benefits, more attention has focused on the choice of provider (National Academy of Social Insurance, 2004). Selection of the health care provider, who can critically affect the outcomes of a workers’ compensation case, is therefore important for all parties interested in workers’ compensation. The provider typically gives information with a direct bearing on most aspects of a claim for medical and indemnity benefits. The provider might diagnose the worker’s condition and shed light on its source, which affects the compensability of the claim. The provider prescribes and, in most cases, provides a course of treatment and disability management practices, the results of which could affect whether the worker returns to work, the duration of the worker’s time out of work, and the conditions under which the return to work occurs. In most jurisdictions, the provider assesses whether the worker’s condition has reached maximum medical improvement and if the worker is left with a permanent impairment or disability. In some jurisdictions, an important component of the health care professional’s assessment is also whether a preexisting condition contributed to the degree of impairment. The rating of the degree of permanent impairment, if any, can significantly affect the amount of indemnity benefits paid under that claim. From the perspective of either the insurer or the worker, any of these evaluations by the health care provider can be sufficiently important to warrant being able to control the selection of provider. 3 4 the impact of provider choice on costs and outcomes Rising costs have also forced policymakers to recognize that the issue of provider choice is more complex than simply deciding which party chooses the initial provider. First, a person undergoing medical treatment can become dissatisfied with that treatment and want to change providers. Alternatively, an insurer or employer might believe the type or quality of treatment being given or planned is inappropriate, or at least inconsistent with the goal of returning the worker to employment expeditiously. Second, the increased use of specialized treatment means that multiple providers are increasingly likely to be involved, so that the initial choice of provider may sometimes be less important than the choice of another provider. Thus, aside from who selects the initial provider, policies regulating provider choice must also specify the circumstances under which a change of provider is permitted and the procedures for doing so. Workers and their advocates have argued that provider choice should be left to the worker.1 At a minimum, they argue, a worker should be treated by people he or she trusts and whose interests are consistent with the worker’s — interests that encourage prompt return to work, but only as medically indicated, and the fullest possible restoration of physical capacity (Ellenberger, 1992). In contrast, employer advocates argue that provider choice should rest with the employer, because without employer choice there is “little incentive to see that the costs of care remain reasonable and appropriate” (Morrison, 1990). Employer advocates also argue, “Employer selection of the treating physician serves to direct injured workers away from those providers who provide excessive services and treatment procedures,” and to “retain those providers familiar with the operations of the employer and who can expedite return-to-work based on that knowledge” (National Federation of Independent Business Research Foundation and National Foundation for Unemployment Compensation and Workers’ Compensation, n.d.). Objectives and Scope of This Study The purpose of this study is to determine whether measurable costs and outcomes in workers’ compensation cases are affected by who selects the health care pro- 1 This is not to suggest that it is solely workers who have supported employee choice. The organization of workers’ compensation state administrators — the International Association of Industrial Accident Boards and Commissions — at one time published a list of standards for the states in which it endorsed the standard of worker choice. However, the organization no longer publishes a listing of standards. introduction 5 vider. The costs and outcomes we study include medical and indemnity costs, the duration of time out of work, the likelihood that the worker returned to substantial employment, the worker’s own perception of the degree of recovery from the work injury, and the worker’s overall satisfaction with the health care received. Only a handful of studies have attempted to ask these questions, and most of those have focused only on costs. Further, most previous studies only consider the relationship between costs and states’ statutory provisions about choice, with states simply categorized as either “employer choice” or “employee choice” jurisdictions. However, a state’s statutory classification can have little to do with the actual choice of primary provider reported by the worker. What can we conclude from past studies? First, while most studies appear to conclude that employer choice is associated with lower medical payments in workers’ compensation, the findings are not uniform. The variation in conclusions may stem from differences in states and years studied and from the use of crude measures of provider choice. Second, very little work has focused on the impact of provider choice on outcomes or cost measures other than medical payments — such as duration of time out of work, indemnity benefits, physical recovery, and worker satisfaction with care. Third, it has been rare to control for the many other factors that likely affect outcomes. Fourth, no study appears to have considered and analyzed the significance of whether the injured employee had been treated previously by the provider who gave primary care in the workers’ compensation claim. Finally, studies from even a few years earlier were done when network arrangements were less common. Because employer-selected providers are more likely to participate in such plans now than they previously did, the relationships between provider choice and workers’ compensation costs and outcomes may have changed. Prior studies on provider choice are summarized in more detail in Technical Appendix A. This study has at least four advantages over previous studies. First, it utilizes data taken from employee interviews conducted in 2002 and 2003 in four states — California, Texas, Massachusetts, and Pennsylvania — which asked workers to identify who selected their health care providers.2 Worker identification of provider is a critical factor in this study, because studies by Lewis (1992), Barth and 2 In contrast to the studies relying only on state-level variation, we have information on provider choice as reported by workers. We categorize a case as employer choice if either an employer or an insurer made the choice; we categorize a case as employee choice if the selection of the provider was made by the worker, by a friend or family member of the worker, or by the worker’s attorney. 6 the impact of provider choice on costs and outcomes Victor (2003), and Victor, Barth, and Liu (2003) have shown that in many instances employees actually choose their providers in employer choice states, and employers select workers’ providers in states categorized as employee choice.3 Analyzing the outcomes of cases on the basis of who actually chose the provider, and not simply whether a state law existed to mandate employee or employer choice, is more informative about the impact of provider choice and about policies governing that choice. A second advantage of this study is that we focus on the primary provider. In our interviews, we asked each worker both who selected the initial provider and who selected the “primary medical provider.” If an interviewed worker reported receiving treatment from multiple providers, we asked which one was primary, defined as “the medical professional that made the decisions about the care that the worker needed and either provided that care or directed the worker to someone who could provide it.”4 The third significant advantage of this study is that we link the interview data to claims data supplied by the claims payors, providing information on factors such as medical and indemnity costs, medical treatments, and employer attributes, among others. A more complete picture of the claim from the perspective of both the worker and the employer should help to better establish the consequences of provider choice. Fourth, a potentially important and unique feature of this study is that the interview also indicated whether the primary provider had previously treated the worker for an unrelated condition. Suspecting that a previous provider-patient relationship might affect some of the outcomes that we measured, we gathered the necessary data to test that hypothesis. The distinction between employee choice of a prior provider and employee choice of a new provider has been highlighted in recent workers’ compensation reforms in California (Senate Bill 899) and in legislative proposals in Texas (Senate Bill 5 and House Bill 7). Until recently in California, the employer had the right to select the initial provider, unless the employee had predesignated a provider; after 30 days, the worker had the right to change to a medical provider of his or her choice. However, the most recent reform legislation changed the rules regarding provider choice. In particular, employers are allowed to establish networks composed of 3 This can occur because the law only gives one party the right to choose the provider, which can be ceded to the other party. 4 A description of the survey, the methods used, and the data employed is presented in Victor, Barth, and Liu (2003, Chapter 1). introduction 7 both occupational and nonoccupational physicians, and the legislation grants to the employer (or the insurer) the sole right to decide which medical providers are in the network. Further, the worker’s right to choose his or her physician after 30 days no longer applies if a network is established that complies with the law, unless the worker has predesignated a physician under particular conditions — most importantly, that the physician was previously the worker’s primary provider of medical care under an employer-provided group health plan.5 In general, as long as employers establish networks, which many are expected to do, workers will have less scope to choose their physicians. Most importantly, a worker’s ability to look for a new physician after an injury will be severely curtailed. Thus, our findings on the consequences of employee choice of prior providers, employee choice of new providers, and employer choice are especially salient to assessing the likely consequences of this very important component of workers’ compensation reforms in California — reforms that are just beginning to be implemented and which will likely be questioned and reexamined as the state implements the reforms enacted to deal with the workers’ compensation crisis that emerged in California at the beginning of this decade. Although the study focuses on provider choice, many other factors affect costs and outcomes. A large number of those factors are included in this study, but others are beyond its scope. For example, worker outcomes can be affected by the clinical services delivered but can also be affected by the process of care — patient education, emphasis on functional recovery, release to modified duty — and other disability management practices. Even though these factors are influenced by the provider, our goal in this report is to estimate the effects of provider choice on costs and outcomes. It remains for future work to understand exactly what differences regarding treatment and other decisions made by providers underlie differences in costs and outcomes associated with provider choice. Organization of the Report Chapter 2 provides a description of the key concepts and data sources used in this study, the measures used to categorize the choice decision, and an explanation of the cost and outcome measures utilized. Chapter 2 also explains the empirical methods used to analyze the data and discusses issues regarding the interpreta- 5 For more details, see Neumark (2005). 8 the impact of provider choice on costs and outcomes tion of the estimates resulting from those methods. More detailed information on statistical methods and results is presented for interested readers in a series of technical appendices. In Chapter 3, we examine the results regarding cost and outcome differences between cases in which the employer chose the provider and those in which the employee chose. In Chapter 4, we conduct similar analyses but examine the role of provider choice more finely, distinguishing between cases in which the employee selected a provider who had previously treated the employee, cases in which the employee selected a new provider, and cases in which the employer chose the provider. Chapters 3 and 4 present combined data for the four states. In Chapter 5, we replicate much of the analysis of Chapters 3 and 4 but do so for each of the four states separately. In Chapter 6, we focus on our findings regarding worker reported satisfaction with the medical care that was received. Chapter 7 presents our conclusions and the policy implications that we draw from the findings of the study — in general and with respect to California and Texas in particular. 2 Key Concepts, Data, and Methods This chapter first describes the key concepts and the data used in the study, defines the key cost and outcome variables, and discusses how we constructed certain control variables. Most important, we describe how we measured the provider choice variables. We then provide an explanation of the statistical models used, consider some issues related to the statistical analysis, and discuss the interpretation of the estimates that the statistical methods yield. Key Concepts We begin by defining the terms used frequently in the report. A list of all the variables used in our statistical analyses, along with variable means, are presented in Appendix B. outcomes Substantial return to work: The worker returned to work and remained working for at least one month before any subsequent absence from work. Duration of time out of work: The number of weeks reported by the worker from the time of injury to the time of the first substantial return to work. Recovery of physical health: A measure derived from the SF-12® instrument for 9 10 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s quantifying health status.1 This reflects the worker’s perceptions of the recovery achieved after the injury, not a clinically based measure. For more information on the derivation of this measure, its strengths and limits, and evidence regarding its validity, see Technical Appendix C. Satisfaction with overall health care: The worker’s rating of his or her overall health care on the following scale: very satisfied, somewhat satisfied, somewhat dissatisfied, very dissatisfied. costs Medical payments: Payments per claim to health care providers for medical care. Indemnity benefits: Payments per claim to workers under state statute, partly to replace lost earnings. provider choice Employee choice: The worker reported that the provider was selected by him- or herself, a family member or friend, or the worker’s attorney. Employer choice: The worker reported that the provider was selected by the employer or insurer. Primary provider: If the worker was treated by multiple providers, he or she was asked which one was the primary provider — that is, the medical professional “who made the decisions about the care that you needed and either provided it or directed you to someone who could provide it.” The primary provider is the initial provider in cases with only a single provider. Prior provider: A provider who treated the worker previously for an unrelated condition. New provider: A provider who did not treat the worker previously for an unrelated condition. 1 SF-12® is a registered trademark of the Medical Outcomes Trust. As a standardized measure of health and functioning, the SF-12® has been used and normed extensively since it was developed (see Ware, Keller, and Kosinski, 1998). key concepts, data, and methods 11 other terms Worker: The individual surveyed to provide information for this study. Each worker we interviewed had a compensable claim involving more than seven days of lost work time (for which the worker received at least some payment). Case or claim: All cases in the study involved more than seven days of lost time. Injury severity: A measure derived from the SF-12® instrument for quantifying health status. Like the recovery measure, this reflects the worker’s perceptions of the severity of his or her injury, not a clinically based measure. Overnight hospitalization: A variable that measures whether the worker was admitted for an inpatient stay, judged by whether the worker received “room and board” or “intensive care” based on the hospital service billing (revenue) code. We derive this variable from the medical services data, not the interviews. Major surgery: A variable that measures whether the worker received surgical services. We use the term major surgery to distinguish these services from other medical services that are also commonly referred to as surgical services but are really medical treatments using invasive techniques — like debriding a wound or certain types of injections. We derive this variable from the medical services data, not the interviews. Data Sources One key data source for this report is the Workers Compensation Research Institute (WCRI) Detailed Benchmark/Evaluation (DBE) database, which contains more than 16 million workers’ compensation claims with representative data in at least a dozen large states. The second key data source is the set of telephone interviews conducted on behalf of WCRI by the Center for Survey Research and Analysis at the University of Connecticut as part of a study to compare worker outcomes in California, Massachusetts, Pennsylvania, and Texas for a subset of cases drawn from the WCRI DBE database. Approximately 750 interviews were completed in each state with workers who had experienced more than seven days of lost time from work. The published report from that study describes the sampling methods, response rates, validity of the outcome measures, and any sampling or response bias (Victor, Barth, and Liu, 2003). That report shows that any such biases are at most small. Technical Appendix D provides an abridged discussion of those issues. The telephone interviews supplement the claims data with information on choice of provider — a central factor in the analysis in this report — as well as satisfaction with health care, worker and employer characteristics, re- 12 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s turn to work, and self-reported information on health status from which we derived measures of severity of injury and recovery of physical health. Measuring Provider Choice Cases included. It is useful to understand the structure of the survey questions used to define provider choice. Some workers received care at the workplace, in an ambulance, or at a hospital emergency room. Because provider choice is not an issue in such cases, these workers are excluded from this study, unless they received subsequent treatment from a provider. Of course, we include any worker who received initial treatment at a medical doctor or chiropractor’s office, clinic, hospital, and other such facility. Primary provider. The central focus of this study is the choice of the primary provider — according to the worker, the one who made the decisions about the care that the worker needed and either provided that care or directed the worker to someone who could provide it. We asked respondents about the number of providers who treated them. In cases with only a single nonemergency provider (about 15–25 percent of cases), the initial provider was necessarily the primary provider. Some 75–85 percent of workers received care from more than one provider. Among those workers, the primary provider was also the worker’s initial provider in nearly 60 percent of cases, according to the worker, and was a different provider in about 40 percent of cases (Table 2.1). New or prior provider. We also asked each worker if the provider identified as primary had previously treated the worker for a different condition. If so, we defined the provider as a prior provider. If the provider had not previously treated the worker for a different condition, we labeled that provider as a new provider. Figure 2.1 shows the possible patterns that were followed by workers in the selection and classification of providers, and in the bottom half of Table 2.1 we report the breakdown. Provider specialty. We recognize that a worker can have many health care providers, but for the sake of clarity in interviewing and to keep the survey at a reasonable length, our questions related to those the worker identified as the initial provider and the primary provider. We also recognize that the number of types of specialties involved in treating injured individuals can be very large. We chose not key concepts, data, and methods 13 Table 2.1 Pattern of Providers Who Delivered Nonemergency Care Percentage of Workers Combined CA TX MA Single provider 20.7 15.4 21.5 23.8 Multiple providers 79.3 84.6 78.5 76.2 Among workers who received nonemergency care from multiple providers Initial provider was primary provider 57.9 59.9 53.3 54.8 Initial provider was not primary provider 42.1 40.1 46.7 45.2 PA 22.5 77.5 62.1 37.9 to probe this issue, other than to distinguish among physicians, chiropractors, and physical therapists. Again, one reason for that decision was the length of the survey; a second was that the worker might not know a physician’s precise specialty. Who chose the primary provider? Regardless of the number of providers a worker received treatment from, we asked each respondent to identify who chose the primary provider. For the purposes of this study, if the worker said that he or she chose the provider or that the choice was made by a family member, a friend, or the worker’s attorney, we regarded this as “employee choice.”2 If the worker said that the employer or insurer selected the provider, we categorized it as “employer choice.” If a medical center, medical provider, or “someone else” was seen by the worker to have chosen the provider, we excluded the case from this study because it was ambiguous whether the worker or employer selected the referring medical center or medical provider. The distributions of these choices are displayed in Table 2.2. In Chapter 3, we report results using the two-way classification of provider choice (employer chose or employee chose). In Chapter 4, we report results using the three-way classification of provider choice (employer chose, employee chose a prior provider, or employee chose a new provider). Whether the primary provider 2 We classified the choice of the provider as the employee’s choice if the attorney chose the physician because this strikes us as an accurate characterization of the choice. It is important to emphasize that attorney involvement and attorney choice of provider are not the same thing. Indeed, as Table B.2 in Technical Appendix B shows, in many cases in which the employers chose the physicians, attorneys were involved (18.5 percent of employer choice cases and about 24.0 percent of employee choice cases had attorney involvement, based on our classification). 14 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Figure 2.1 Provider Choice Questions Treated at doctor's office, medical clinic, hospital, or chiropractor's office Treated at workplace, in ambulance, or in hospital emergency room Additional treatment No additional treatment One provider Two or more providers Prior provider New provider Employee chose initial provider Employer chose initial provider Other chose initial provider Employee chose initial provider Initial provider was primary provider Employer chose initial provider Other chose initial provider Initial provider was not primary provider Prior provider New provider Employee chose primary provider Employer Other chose primary chose primary provider provider Two-way classification: employer chose or employee chose Three-way classification: employer chose, employee chose prior provider, or employee chose new provider Prior provider New provider key concepts, data, and methods 15 Table 2.2 Who Chose the Respondent’s Primary Provider? Percentage of Workers Combined CA TX MA Employee chose You (respondent) Family member Friend Your attorney Employer chose Your employer Insurance company Medical professional/ hospital/clinic Someone else Number of cases Number of cases with either employee or employer choice 41.4 36.9 1.9 1.3 1.3 37.5 31.7 5.8 17.7 3.3 2,513 1,960 33.8 52.7 51.0 28.4 46.8 46.3 0.7 2.9 2.8 1.6 2.0 1.4 3.1 1.0 0.5 48.3 27.0 19.4 41.0 21.4 14.4 7.3 5.6 5.0 13.8 16.7 25.1 4.0 3.6 4.5 665 609 542 538 481 376 PA 31.3 29.4 1.5 0.2 0.3 50.7 45.4 5.3 16.3 1.6 697 565 was the initial provider or not, we asked each worker if the primary provider was a new provider or a prior provider (Figure 2.1).3 Overall, from the survey data set of more than 2,500 cases with treatment from providers (that is, cases with more than emergency treatment), we constructed a subset of about 1,960 cases for which we could classify provider choice using both the two-way and three-way classifications.4 Table 2.2 shows the number of cases 3 We did not ask the prior/new question for initial provider of any worker for whom the initial provider was not primary. This was one of many compromises made in the design of the survey to reduce the scope to fit into the time constraints of the interview. We could have used those respondents for the two-way classification, but we did not because the three-way classification is of equal interest. 4 Of the total number of completed interviews (nearly 2,800), 232 were not included in this study because the respondents’ care was limited to emergency treatment only, and 44 others did not answer the provider choice question. Among the remaining group (2,513), 596 were dropped because the worker reported that the provider was chosen by a medical center, another medical professional, “someone else,” or “don’t know/refused to answer.” 16 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 2.3 Provider Choice, by Type of Provider Physician Percentage of Cases Chiropractic Physical Therapy Four states combined Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose California Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose Texas Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose Massachusetts Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose Pennsylvania Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose 91 89 93 85 89 88 87 88 86 84 93 81 92 75 90 88 93 93 94 92 93 95 97 93 89 34 82 61 10 3 37 35 10 1 10 1 11 2 3 11 25 15 2 71 20 3 45 32 42 51 42 34 33 23 30 16 19 Other 3 1 1 1 1 4 2 1 2 2 0 1 0 2 0 6 1 1 0 1 2 0 0 0 1 key concepts, data, and methods 17 by state. The table also reveals striking similarities in patterns of provider choice in California and Pennsylvania on the one hand and in Texas and Massachusetts on the other, but differences also appear between the two sets of states. In Texas and Massachusetts, the law in effect at the time of the study gives the worker the choice of initial provider and relatively free reign to change providers. In California and Pennsylvania, the effective law allows the employer to designate the provider for the first 30 days and 90 days, respectively, after which the worker can change providers.5 The influence of these policies is reflected in the higher incidence of employer choice in the latter two states and, conversely, the higher incidence of employee choice in Texas and Massachusetts. Choice of the primary provider, by type of provider. Regardless of who chose the provider, the overwhelming majority of workers reported that a physician was the primary provider — the provider that made the decisions about the care that the worker needed and either provided it or directed the worker to some who could provide it. As shown in Table 2.3, in all four states combined, employers infrequently chose chiropractors as primary providers — only 2–3 percent of cases in each of the four states. In Pennsylvania and Massachusetts, workers selected chiropractors as their primary providers in about 3 percent of cases — similar to when employers chose the providers. In California and Texas, however, workers selected chiropractors as primary providers much more often; in those two states, when workers chose their primary providers, they selected chiropractors at least 10 percent of the time.6 In California, workers were equally likely to select chiropractors who had previously treated them or a new chiropractor. In Texas, workers were much more likely to see chiropractors who had never previously treated them. Sometimes workers identified physical therapists as their primary providers. This occurred most often when the employer chose the provider or when the worker said that a medical center or a medical provider selected the primary provider. The latter is not surprising, because physical therapy typically requires a doctor’s referral. 5 There are some exceptions to those rules. For example, in some states, when the employer has established an approved network, the worker must select within the network. See Tanabe and Murray (2001). 6 Of course, one factor that likely affects the frequency of use of chiropractors is their supply (although this is also driven by demand). Large differences exist across states in the numbers of chiropractors per 100,000 population. As of 1995, the rates of chiropractors per 100,000 persons were 33.2 in California, 26.8 in Texas, 21.7 in Pennsylvania, and 20.3 in Massachusetts (Coulter and Shekelle, 1997). 18 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Cost and Outcome Measures We study most of the key outcomes of workplace injuries that should be of interest to policymakers: costs, return to work, recovery of physical health, and satisfaction with care. The two cost measures that we study are indemnity benefits and medical payments per claim. Both measures are derived from payors’ records about what payments were actually made as of 29 to 31 months after the injury. The WCRI DBE database standardizes definitions of those measures across payors and across states. Table 2.4 shows average indemnity benefit payments and medical payments per claim for each state. Medical payments per claim were significantly higher in California and Texas than in Massachusetts and Pennsylvania. Indemnity benefits per claim were higher in California than in the other three states. Another important outcome is the extent to which the worker recovered his or her physical health after the injury. The measure we use for this study is derived from worker responses to the most widely used instrument for measuring general health status — the SF-12® survey. The focus is on physical health, not mental health. Because the SF-12® scores for physical health are quite insensitive to even extreme variations in the mental health scores, we compute the physical health scores holding the mental health scores constant.7 In the interviews, we asked workers to recall their health status at three points in time — the month before the injury, the week after the injury, and the month before the interview. The recovery variable is the difference between the worker’s self-reported health status after the injury and the same measure at the time of interview.8 Because this measure is based on workers’ perceptions, we refer to this 7 This approach and related sensitivity analysis is discussed in Victor, Barth, and Liu (2003) and in an abridged form in Technical Appendix C. 8 The recovery measure we use is the change from one week after the injury to the interview; in most cases, this change is positive, but that is not imposed on the data since a worker’s health could worsen. In addition, the severity control we use in the regression models that follow is similarly defined as a change in levels — in this case, from before the injury to one week after. Again, we do not impose that the worker’s health had to worsen, although it did in almost every case. We also experimented with specifications defining each of these variables as relative measures — that is, we defined the percentage recovery relative to health status one week after the injury and the percentage severity relative to health status before the injury. The results were very similar. We have some preference for the specification with changes in levels, because we do not think a full recovery from a very minor injury should be treated symmetrically to a full recovery from a very serious injury. Put another way, we think it is important that the regression estimates of effects of provider choice on recovery reflect a large “penalty” for serious injuries that are not followed by substantial recoveries, even if they are also associated with near-complete recoveries for very minor injuries. key concepts, data, and methods 19 Table 2.4 Costs and Health Outcomes, by State Combined CA Average medical payment per claim Average indemnity benefit per claim Average recovery scorea Average severity scorea $8,713 $12,709 19.2 29.0 $9,950 $15,444 17.6 29.0 TX $11,729 $10,188 15.0 28.2 MA $4,946 $13,874 24.1 29.8 PA $7,594 $11,358 21.0 29.0 a Respondents’ SF-12® scores are scaled scores from 0 to 100, where 100 is the best health. The recovery score is the difference between the SF-12® value at the time of interview and the score one week after injury. The severity score is the difference between the score for the time four weeks before the injury occurred and the score one week after the injury. The mean value of the preinjury scores for respondents was about 54 or 55, depending on the state. variable as “perceived recovery.”9 Table 2.4 shows that the average perceived injury severity was very similar across all three states but that the average perceived recovery was better in Massachusetts and Pennsylvania. We also study whether the worker returned to work for at least one continuous month at any time between the injury and the interview. We call this a “substantial return to work.” In addition, we measure the duration of time out of work, as reported by the worker as of the date of the interview — approximately 3.0 to 3.5 years postinjury. Recall that all cases sampled had more than seven days of lost time. Table 2.5 shows the mean and median durations reported by the workers and the percentages not reporting substantial returns to work. Of course, both measures are closely related to indemnity benefits but not perfectly correlated because some workers receive permanent disability benefits that are not related to actual wage loss and others receive lump-sum settlements that terminate the employers’ liability regardless of whether the workers remain out of work. 9 Victor, Barth, and Liu (2003) provide extensive discussion of potential concerns about recall bias and other limitations, as well as evidence of validity of the health status measure from which both recovery and injury severity are derived. Technical Appendix C in the current report summarizes that discussion. 20 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 2.5 Substantial Return to Work and Duration of Period Out of Work Combined CA TX MA Percentage of workers who did not have substantial returns to work 19 Among workers who had substantial returns to work Average duration of time out of work (weeks) 13 Median duration of time out of work (weeks) 6 19 27 18 14 14 14 868 PA 13 11 6 Note: Only workers who had substantial returns to work were asked, “How many weeks was it from the time you first stopped working because of your injury and the first time that you returned to work for one full month?” Only cases in which the employee or employer chose the primary provider are included. We also measure satisfaction with care. In the survey, we asked a widely used set of questions about satisfaction with the timeliness of care, the provider, and overall care. The variable used in this study examines satisfaction with overall care. The specific question was “Now think about all of the medical care you received from the first treatment for your injury until now. Were you satisfied or dissatisfied with the medical care you received overall?” Table 2.6 shows the distribution of responses among the four choices offered. Empirical Methods This section sketches out the statistical methods and models used to assess the impact on outcomes related to provider choice. More details appear in Technical Appendix E. In subsequent chapters, we attempt to describe the findings from our analyses in a way that does not require the reader to have mastery of these models. Nontechnical readers should feel free to skip to Chapter 3. basic model and control variables The empirical approach uses statistical models to estimate the impact of provider choice on a variety of workers’ compensation costs and outcomes. We study key concepts, data, and methods 21 Table 2.6 Satisfaction with Overall Health Care Percentage of Workers Combined CA TX MA Very or somewhat satisfied 82 80 80 85 Very satisfied 52 47 51 56 Somewhat satisfied 29 33 29 29 Very or somewhat dissatisfied 18 20 20 15 Somewhat dissatisfied 8 10 9 6 Very dissatisfied 10 10 11 8 PA 83 57 26 17 8 9 Note: Only cases in which the employee or employer chose the primary provider are included. provider choice using a two-way classification (employee versus employer choice) and a three-way classification (employee choice of new provider versus employee choice of prior provider versus employer choice). The statistical models control for other influences on workers’ compensation costs and outcomes to avoid attributing the other influences to the impact of provider choice. We also take a number of steps, discussed in some detail in this chapter, to account for possible difficulties in measuring the severity of workplace injuries that might lead to incorrect conclusions. More formally, the framework is based on a standard regression model for a cost or outcome variable generically denoted Yis, where i indexes individuals and s indexes states, of the form: Yis = α + CHOICEis β + WORKERisγ + FIRMisδ + INJURYisθ + STATEsκ + TREATMENTis λ + εis. (2.1) As indicated in the preceding section, our dependent or outcome variables come in different forms: continuous (for example, the cost measures), in two categories (substantial return to work), and in more than two categories (satisfaction). We therefore have to use different statistical methods for different dependent variables, as discussed in more detail later in this section and especially in Technical Appendix E. The provider choice variables, which may be one dummy vari- 22 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s able corresponding to the two-way classification or two dummy variables corresponding to the three-way classification, are included in the vector CHOICE. In any model of workers’ compensation costs or outcomes, it is essential to include characteristics of workers (WORKER) and the workplace (FIRM), because both types of characteristics have been shown to affect costs and outcomes. For example, older workers have been found to be less likely to return to work; workers with less education are likely to have greater difficulty in the labor market; and workers employed by firms in certain industries (such as construction) may have unique return-to-work problems to surmount (see, for example, Galizzi and Boden, 1996). The list of variables included in WORKER includes demographics, education, wages and whether the individual was an hourly worker, tenure at the time of injury, and whether the worker elected to have the interview conducted in Spanish. Workplace characteristics include firm size and an industry breakdown, details of which are provided in the tables discussed later. Naturally, we would expect costs, return to work, recovery, and satisfaction to depend in important ways on the characteristics of the injury. We have several types of measures. The first is a classification of injury type: back pain; nonback sprain or strain; fracture; inflammation, laceration, or contusion; and a residual category of other injuries, based on the diagnostic (ICD-9) codes assigned by the providers.10 A second measure captures the worker’s perceived injury severity. This measure is constructed from the worker’s answers to the SF-12® instrument, paralleling what we did for the measures of perceived recovery (see the earlier discussion). Specifically, we measure injury severity by the difference in workers’ responses to the SF-12® questions regarding their health status between the month before the injury and the week after the injury. The construction and validity of this measure is discussed in Technical Appendix C. The inclusion of the worker, workplace, and injury characteristics in a model of how provider choice affects outcomes is unambiguous because those variables can be associated with both provider choice and the costs and outcomes we study, but not for reasons underlying any causal relationship between provider choice and outcomes. For example, compared with other workers, older males may have 10 In some cases, workers are assigned multiple diagnosis codes during the course of their disability. In such cases, we define a primary diagnosis code based on the code that receives the greatest expenditure. Also, in some cases, diagnosis codes are missing in the database. In these cases, we use information from the payor about the nature of injury and part of body to assign the case to the appropriate injury group. key concepts, data, and methods 23 worse medical outcomes because age inhibits recovery. Yet older males may also — because of greater affluence, access to health insurance, and possibly even previous injuries — be most likely to have chosen primary providers whom they have seen previously. Without controlling for age and sex in such a case, we might incorrectly infer that the older worker’s choice of a prior provider resulted in or caused worse medical outcomes. Similarly, more-severe injuries may make it more likely, at least in some states, that the employee chose the provider, simply because in California, for example, the employee during the sample period had the right to choose a physician 30 days after first receiving treatment, and more-severe injuries are more likely to pass the 30-day window. variation in state workers’ compensation systems However, the worker, workplace, and injury characteristics may not be sufficient as controls. First, as noted earlier, our data come from four states — California, Texas, Massachusetts, and Pennsylvania. Workers’ compensation systems vary widely across all states, including the four in our study. For example, the study states differ markedly on matters such as the frequency and sources of disputes, the methods used to terminate temporary disability benefits, the criteria used to rate permanent disability benefits, the use of networks to provide medical care, and so on. We know that outcomes related to cost, return to work, and other measures vary substantially across states (Telles, Wang, and Tanabe, 2004). We also know, as discussed earlier, that the states differ regarding the prevalence of employee and employer choice: California and Pennsylvania are states where initially the employer has the right to choose the provider, while Massachusetts and Texas are considered to be employee choice states. Technical Appendix F provides an overview of some of the more important interstate differences that are relevant for this study. Given these facts, if we use across-state variation in choice and outcomes to identify β in equation (2.1), we may incorrectly attribute differences in outcomes associated with other features of the states’ workers’ compensation systems to variation in individual choice of provider. Consequently, we report all specifications, including dummy variables for the states (STATE), in which case the effects of provider choice are identified solely from within-state differences associated with this choice. The potential downside of this is that we effectively throw out the variation in provider choice that is driven by differences in state workers’ compensation systems, which is plausibly the most exogenous source of variation in 24 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s provider choice.11 We examined estimates both excluding and including the state dummy variables to see whether we could find results that are robust to this specification choice and therefore could draw firmer conclusions. In general, we found that results including or excluding the state dummy variables were similar. Because we think it most important to control for omitted variation in state workers’ compensation systems, we report estimates based on specifications that include the state dummy variables. appropriate statistical models for each dependent variable We need to use different statistical models because our dependent variables take different forms. For the three costs and outcomes variables that are continuous (indemnity benefits, medical payments, and recovery of physical health), equation (2.1) is estimated as a linear regression, in which case the estimated coefficient of a variable simply measures how the outcome changes with a one-unit increase in the variable. For the other outcomes variables, we cannot use linear regression. In each case, there is some choice regarding exactly which type of model to use. We have chosen to use a set of models for which the estimated coefficients have a very similar interpretation. The return-to-work outcome is dichotomous; either the person returned to substantial employment or not, and we estimate this using a logit model. The model for the duration of time before a substantial return to work occurred has to be estimated using survival models, to account for the possible truncation of the out-of-work period. That is, it is possible that at the time of the survey, some individuals have still not returned to substantial employment. If so, all we know is that the out-of-work period lasted at least up to the time of the survey. Finally, the satisfaction outcome is also discrete (like return to work), but takes on one of four values: very satisfied, somewhat satisfied, somewhat dissatisfied, and very dissatisfied. Further, these values are ordered, given that the satisfaction responses can be ranked clearly. To study this outcome, we use an ordered logit model. 11 Looking at the issue in a different way, provider choice is correlated with unobservables in equation (2.1) — for example, unmeasured variation in injuries. If we thought that state dummy variables could be excluded from the generic model given by equation (2.1), and we also thought that state of residence affected choice because of features of states’ workers’ compensation systems, then the state dummy variables would provide a natural set of instrumental variables for provider choice. However, given that other features of state workers’ compensation systems likely affect outcomes, the exclusion restriction is invalid, precluding this instrumental variables strategy. key concepts, data, and methods 25 equality of effects of provider choice across states Another issue is whether we can combine, or pool, the data across the four states to obtain the most precise estimates of the impact of provider choice. Given that we have only about 400 to 550 observations per state, this pooling is highly desirable. However, it could be inappropriate and lead to biased estimates if the effects of provider choice on the outcomes we study vary significantly across states. We tested for this and did not find evidence against the restrictions implied in combining the data and estimating a common set of effects of provider choice.12 We emphasize the combined estimates in the chapters that follow, although for completeness we also report results for the states individually (in Chapter 5). causal inferences about policy changes regarding provider choice The central goal of this report is to estimate the effect of provider choice on workers’ compensation costs and worker outcomes. The evidence we report is of interest from a policy perspective to the extent that it helps assess the effects of policy choices regarding provider choice, such as California’s most recent workers’ compensation reforms that restrict employee choice of a new physician. Does the evidence that we assemble here speak to the effects of policy changes, and if so, under what conditions? In an ideal world, to estimate the effects of policy changes, we would randomly assign injured workers to different provider choice “regimes” (for example, with choice assigned to either the worker or the employer) and then observe the outcomes. Because workers would be randomly assigned, in a large sample we could rule out any differences across workers in the different choice regimes as alternative explanations of differences in outcomes associated with provider choice. That is, we could confidently draw causal inferences about the effects of provider choice 12 To test this, for each analysis we conduct we also test for differences in the parameters describing the effects of provider choice, as well as the coefficients of the other variables in the model. We do this by interacting each of these variables with the state dummy variables, estimating these full models, and then separately testing the constraints that the provider choice coefficients are the same across states, and that the other coefficients are the same across states. We never reject the first set of restrictions; we sometimes reject the latter, but we verify that the provider choice estimates are insensitive to allowing the effects of the other control variables to differ across states. A full description of the testing and results is presented in Technical Appendix G. 26 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s that would be informative about the effects of policy changes. Of course the real world falls short of this experimental idea. We instead observe injured workers seeing different providers — some chosen by them, whether prior or new providers, and some chosen by employers — and we observe the outcomes. But we have to be concerned that there are characteristics of workers associated with both provider choice and with workers’ compensation outcomes that result in misleading inferences about the effects of provider choice. As one concrete example, to which we will return later, suppose that the most severe injuries tend to result in workers ending up with new providers, chosen by them, as their primary providers. This might occur because in search of solutions for the most severe injuries, workers are motivated to seek out particular providers (such as specialists). In this scenario, if we simply compared outcomes such as costs and time away from work across these workers and workers for whom the employers chose the providers, we would find that for the former group, costs were higher and return-to-work outcomes worse. We could then be led to the incorrect conclusion that worker choice of a new provider causes higher costs and worse return to work, whereas the relationship arises only because the most severely injured workers selected into the employee choice/new provider group. We address this potential problem in a few ways. Most important, as explained earlier, for all our analyses we include controls for numerous detailed characteristics of workers, workplace characteristics, and injury characteristics. Indeed, we would argue that the data used in this report yield far more detailed sets of control variables than past work. (The full list of control variables is listed in Table 2.7.) The relation between these control variables, including the severity controls, are not necessarily causal, but where statistically significant, are at least associative. Second, in Table 2.7, we report estimates of models for provider choice, to explore which variables are in fact associated with choice. In these models, we report odds ratios for the employee choice options relative to employer choice.13 A coefficient estimate greater than 1, when statistically significantly different from 1, implies that the variable associated with that coefficient boosts the likelihood of employee choice.14 13 Results by state are reported in Technical Appendix H. 14 Note that throughout the report, when we refer to estimated odds ratios, we simply refer to “statistically significant” as a shorthand for “statistically significantly different from 1,” which is equivalent to the statement that the difference between the estimate and 1 is statistically significantly different from zero. key concepts, data, and methods 27 Notes on Statistical Significance: ■ In all tables reporting results from our statistical models, we note with symbols whether or not the estimates are statistically significantly different from zero. We report the level of statistical significance for each estimate, focusing on statistical significance at the 5 percent and 10 percent levels, which are indicated with two asterisks (**) and one (*), respectively. Statistical significance is important because all estimates to some extent reflect the randomness of which individuals were included in the sample. Statistical significance at the 5 percent level, for example, means that there is a 95 percent chance that the estimated effect is different from zero; conversely, the chance is 5 percent (or 1 in 20) chance that an estimated effect looks different from zero when in fact the true effect is zero. The lower the level of statistical significance (5 percent versus 10 percent), the greater the confidence we have that an estimated effect is in fact different from zero. ■ Estimates significant at the 20 percent level are often not noted in other work, because the chance is 20 percent (or 1 in 5) that the estimated effect is in fact zero. However, we report these results to distinguish effects that are still considerably more likely to be different from zero than not — sometimes termed “marginally significant.” These estimates are indicated with a dagger (†). ■ It is important to interpret carefully results that are classified as “not statistically significant” or, more simply, “not significant.” An estimated effect that is not significant is not the same as concluding that the effect is zero. Under the assumptions of the statistical models we use, the estimate we obtain is the best estimate of the effect. However, the estimate cannot be established as statistically significant, meaning that there is a reasonable probability that the nonzero estimate could have been obtained even if the true effect is zero. It turns out that quite a few variables are significantly related to provider choice — which would not be the case with random assignment of choice. For example, older workers are significantly more likely to choose their own providers in the two-way model, and their own prior providers in the three-way model. We also found that choosing a prior provider is positively correlated with firm size, and there are some interindustry differences in choice, probably both reflective of dif- 28 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s ferences in health insurance coverage. We found that persons with the lowest levels of educational attainment and those interviewed in Spanish are less likely to choose their own providers or prior providers. We have no reason to believe that single males or persons with lower education or Spanish speakers are more likely to have more-severe injuries, conditional on the controls. However, we found they are more likely not to know or to have prior providers because of a lack of health insurance. More significantly, certain types of injuries, especially back injuries, are significantly more likely to be associated with employee choice of provider than is the reference category of inflammation, laceration, or contusion. On the other hand, it is interesting to note that perceived severity is not associated with employee provider choice, although it must be remembered that because the model includes such variables as type of injury, what is captured by the estimated coefficient of severity is the effect of variation in severity for the same type of injury. Note also that major surgery is significantly positively associated with employee choice, although as discussed later in this section, surgery may to some extent be an outcome of employee choice rather than a measure of the severity of the injury. The estimates in Table 2.7 certainly indicate that the assignment of workers and their injuries to provider choice regimes is not random, which is no surprise. What the estimates cannot tell us, however, is whether the inclusion of all the control variables listed in Table 2.7 in our models for workers’ compensation costs and outcomes capture enough of the variation in other determinants of these costs and outcomes that we are confident that the regression models capture the causal effects of provider choice, or instead whether there is still residual unmeasured variation in severity of injury or other factors that is related to provider choice. However, the fact that greater severity is not independently associated with a higher likelihood of employee choice makes it more plausible that we are estimating causal effects of provider choice. Our third approach to obtaining estimates that provide evidence on the causal effects of provider choice involves including additional control variables related to severity. In particular, the claims database includes information on the treatment of the injury, including whether the treatment included an overnight hospitalization and major surgery; these are captured in the variable TREATMENT in equation (2.1). These potential control variables present a double-edged sword. On the plus side, they are likely to capture additional variation in the severity of the injury that is not picked up in the other variables that capture nature and severity of injury. For example, some fractures, even if viewed by the respondent as entailing the same severity, may result in overnight hospitalization for a variety of reasons related to the injury, and therefore we would expect higher medical payments. On key concepts, data, and methods 29 Table 2.7 Determinants of Provider Choice, Four States Combined Two-Way Provider Choice Classificationa Employee vs. Employer Three-Way Provider Choice Classificationb Employee/Prior vs. Employer Employee/New vs. Employer State Pennsylvania California Texas Massachusetts Worker characteristics Age Male Married Wage Hourly worker Tenure Less than high school Some high school High school graduate Some college College graduate Postgraduate Survey in Spanish Workplace characteristics Firm size,≤50 Firm size, 51–250 Firm size, 251–1,000 Firm size,>1,000+ High-risk services Low-risk services Clerical/professional services — 1.140 3.759** 4.507** 1.007† 1.085 1.039 0.972† 0.416** 1.015** 0.582† 0.814 — 0.856 0.882 1.211 0.329** — 0.904 0.887 1.109 0.709† 1.022 — — 1.236 3.160** 5.498** 1.011* 0.942 1.117 0.973 0.411** 1.017** 0.434* 0.848 — 0.868 0.790 1.328 0.193** — 0.926 0.843 1.590** 0.548** 0.886 — — 1.033 4.226** 3.777** 1.006 1.226† 0.979 0.972 0.409** 1.012† 0.690 0.807 — 0.848 0.996 1.089 0.440** — 0.893 0.952 0.773 0.899 1.223 — 30 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 2.7 Determinants of Provider Choice, Four States Combined (continued) Two-Way Provider Choice Classificationa Employee vs. Employer Three-Way Provider Choice Classificationb Employee/Prior vs. Employer Employee/New vs. Employer Manufacturing Construction Trade Other industries Injury characteristics Back pain Nonback sprain or strain Fracture Inflammation, laceration, or contusion Other injuries Severity Treatment characteristics Overnight hospitalization Major surgery Attorney involvement N 0.562** 0.802 0.944 0.749 1.614** 1.148 1.394† — 1.536* 0.996 0.980 1.376** 1.553** 1,960 0.436** 0.520* 0.803 0.644 1.509† 1.095 1.341 — 1.669* 0.991* 1.134 1.462** 1.506** 1,951 0.728 1.153 1.128 0.894 1.658** 1.148 1.406 — 1.416 1.002 0.842 1.329* 1.592** 1,951 a Odds ratios from logit model are shown, relative to employer choice. The odds ratio measures the effect of the variable on the probability of the type of employee choice indicated in the column, relative to the probability of employer choice. b Odds ratios from multinomial logit model are shown, relative to employer choice. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. key concepts, data, and methods 31 the minus side, the treatment variables also reflect outcomes of the medical decision-making process and thus to some extent directly reflect the choice of provider. Because the treatment variables in part capture costs and outcomes, their inclusion may amount to what is often referred to as “overcontrolling” for injury severity.15 That is, they may capture not only remaining differences in severity but also outcomes of provider choice that we more appropriately want to think about as effects of provider choice but will not capture when the treatment variables are included.16 Under this interpretation, excluding the treatment variables runs the risk of having important unmeasured heterogeneity in injuries, which if associated with provider choice may lead to choice-related differences in costs and outcomes that are too large. On the other hand, including the treatment variables is likely to generate estimates that understate the differences associated with provider choice. In this case, the truth would lie somewhere in between the estimates including and excluding the treatment variables. Consequently, we present both sets of estimates to determine the outcomes for which the resulting range of estimates is tight enough to provide information on the effects of provider choice. When the estimates differ, readers more concerned that our injury and severity measures leave potentially important differences in severity unmeasured may be more inclined to emphasize the estimates that include the treatment variables, and vice versa. Finally, a fourth approach we take to the problem of unmeasured severity is to assess how sensitive the estimates are to omitting from the model variables measuring severity or the nature of the injury. If the estimates are not very sensitive, this suggests that additional unmeasured variation in severity when these variables are included cannot play much of a role. Of course, even with all these efforts, we cannot definitively rule out the possibility that even with the treatment variables included, there is unmeasured variation in injury severity that might affect, for example, costs or return to work. 15 We consider including these variables, but not attorney involvement, because hospitalization and surgery are sometimes likely to be dictated by medical exigencies. At the same time, we recognize that it is possible that attorney involvement exacerbates the effects of employee choice of provider. This raises very interesting questions about how costs and outcomes — and their relationship to provider choice — might change were policies relating to use of attorneys in workers’ compensation cases altered. That question, however, is well beyond the scope of this study. 16 To see the problem of overcontrolling in an extreme form, if we simply put the dependent variable on the right side of the equation, we would obtain a perfect fit, and nothing else, including provider choice, would explain variation in outcomes. 32 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s This implies that, ultimately, we cannot arrive at an definitive answer regarding the causal effects of provider choice from these data, because at its core, that is a question about differences between workers that we cannot measure, in contrast to those we can measure. In our view, the extensive set of control variables that we have, coupled with the results from the various analyses just described, allows us to be reasonably confident that we are identifying the causal effects of provider choice. This means that it is appropriate to think of our estimates as indicating what would happen if policies regarding provider choice were changed, for example, to restrict employee choice. We believe this is particularly true of the specifications that we regard as likely overcontrolling for injury characteristics by including the hospitalization and surgery variables. At the same time, we recognize that our evidence falls short of experimental standards, which of course leaves open the possibility that experimental evidence could lead to different conclusions. 3 Impact of Employee or Employer Choice of Provider: Main Results In this chapter, we examine the impact of the actual choice of primary provider (regardless of who the state law permits to choose) on costs and outcomes. Recall that by primary provider, we mean the provider that the worker said was the medical professional who made the decisions about the care that he or she needed, whether the care was provided directly or the provider directed the worker to someone who provided the care. We examine the impact on medical and indemnity costs, which are the focus of much policy discussion. However, policymakers need to know about not only the cost differences associated with provider choice but also the effects of provider choice on outcomes of care that are associated with quality, including return to work, duration of lost time, recovery of health, and patient satisfaction. Certainly, cost reductions associated with employer choice would be viewed less positively if they were accompanied by reductions in the quality of care. We also recognize that these outcomes are influenced by factors besides the quality of health care. Here and in Chapter 4, we present the main results of the study — relying on the combined sample pooling the data from all four states, while controlling for important differences among the states in policies and practices that affect the outcomes that we study. In Chapter 5, we discuss the results estimated separately for each state. We emphasize the findings from the combined sample because with the larger sample size, the results are more precise and statistically powerful. As discussed in Chapter 2 and described in detail in Technical Appendix G, our tests 33 34 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s do not indicate statistically significant differences in the effects of provider choice across states, indicating that the best approach is to combine the data from the four states and rely on the more precise estimates that result. We examine here the most common way of characterizing provider choice — employee choice compared with employer choice. This is the standard way of thinking about provider choice in existing research and in much of the policy debate. Chapter 4 examines a three-way classification of provider choice, presenting comparisons of the differences in costs and outcomes for cases in which the employee chose a provider who previously treated the worker for an unrelated condition, in which the employee chose a provider who is new to the worker, and in which the employer chose the provider. Given that state workers’ compensation policies can influence not only whether employees choose their provider but also under what conditions — in particular, whether that provider must be a predesignated provider who previously treated the worker — it is important to understand the findings presented here and in Chapter 4 to derive the full implications of our analysis for public policy. Summary of Findings When workers rather than employers selected their primary providers, on average, costs were higher and return-to-work outcomes were poorer. Physical recovery was unaffected, although workers were more satisfied with their health care. More specifically, compared with cases in which the employer chose the provider, cases in which the employee selected the provider had: ■ Medical payments that were 10–21 percent higher and indemnity benefits that were 8–15 percent higher (although only the higher estimate for indemnity benefits is statistically significant). ■ Odds of returning to work and remaining at work for at least one continuous month that were 16–19 percent lower (although the evidence is statistically weaker than the other findings).1 ■ Time out of work that was 23–32 percent longer. ■ Very similar reported recoveries of physical health. 1 The outcomes for events that do or do not occur are often summarized this way. Formally, the statement that the “odds” are, for example, 10 percent higher means that the relative probability of the event is 1.10. impact of employee or employer choice of provider 35 ■ Likelihood of reporting a higher level of satisfaction with their care that was 57–59 percent higher. Impact of Employee or Employer Choice of Provider on Costs and Outcomes: Main Results This section examines the results using the combined data for the four states while controlling for material differences among the states and cases. We report the results for two specifications: the first might undercontrol for severity by excluding the treatment variables, the second likely overcontrols for severity by including them. The true results probably lie somewhere between the reported results for the two models. When the results from the two models are similar, the reader should have greater confidence in the conclusions; and when the results are significant, even when they include the hospitalization and surgery controls, it seems particularly unlikely that associations between provider choice and outcomes are in fact attributable to unmeasured variation in injury severity rather than the effects of provider choice. We found that cases in which the worker rather than the employer selected the primary provider were associated with higher costs and poorer return-to-work outcomes, with no differences in physical recovery but with higher satisfaction, compared with cases in which the employer selected the provider. As Table 3.1 shows, medical payments were 10–21 percent higher, significant in both cases.2 Lower health care costs when the employer selects the provider can occur for several reasons. Employers might refer injured workers through their medical networks, in which providers have been preselected on the basis of quality or price discounts through negotiations between networks and providers. These arrangements can provide certain advantages in terms of cost, access, or adherence to recommended treatment protocols. The results for indemnity costs also suggest higher costs when the worker chooses the provider, although the evidence is weaker. In particular, the estimates from model 1 — which excludes the hospitalization and surgery controls — indicate that indemnity benefits were 15 percent higher when the worker chose the 2 Note that in the tables in the main text, we report only the effects of provider choice on the outcomes. Full regression results for the combined sample of all states, for the model with the two-way and three-way classifications of provider choice, are given in Technical Appendix H. 36 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 3.1 Impact of Employee Choice Compared with Employer Choice Model 1: Without Treatment Model 2: With Treatment Controls (percent) Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 21** ($1,868) 15* ($1,908) 32** –19† 0 57** 10* ($903) 8 ($978) 23** –16† 1 59** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. In model 1, for example, when the worker chose the provider, medical payments were $1,868 greater per case (on average) than when the employer chose the provider; or the duration of time out of work was 32 percent longer when the worker chose the provider than when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Table 2.2 provides the breakdown of observations by state. Observations are weighted to be representative of claims within each state; see the discussion on weighting in Technical Appendix D. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. provider, while the difference falls by nearly half and becomes statistically insignificant in model 2 — which includes the controls. The results consistently indicate that employee choice of provider is associated with slower return to work. Reported time from injury until initial substantial return to employment was 23–32 percent longer when the employee chose, and substantial return to work was 16–19 percent less likely in the three years after the injury, although the latter results are only marginally significant. (Recall that a substantial return to work is one that lasts for at least one month without having to stop working again due to the injury.) Note that for the return-to-work variables, the range of estimates for models 1 and 2 is tighter, and the statistical significance of the results is not weakened by including the hospitalization and surgery controls, bolstering our confidence in these results and in a causal interpretation of the effect of provider choice. Interestingly, despite the differences in costs and time out of work, there was no impact of employee or employer choice of provider Table 3.2 Satisfaction with Overall Care, by Who Selected the Provider and by State California Employee Chose (percent) Employer Chose (percent) Texas Employee Chose (percent) Employer Chose (percent) Massachusetts Employee Chose (percent) Employer Chose (percent) Pennsylvania Employee Chose (percent) Employer Chose (percent) Very or somewhat satisfied 84 78 80 82 86 83 87 80 Very satisfied 55 41 57 39 60 46 65 51 Somewhat satisfied 29 37 23 43 26 37 22 29 Very or somewhat dissatisfied 17 22 21 17 14 17 13 20 Somewhat dissatisfied 9 10 9 9 7 6 8 8 Very dissatisfied 8 12 12 8 7 11 5 12 37 38 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s difference in the perceived recovery of physical health between workers who selected their providers and workers whose employers selected the providers; any estimated differences in recovery were trivially small and are thus not statistically significant. Yet workers who chose their providers were much more likely to be more satisfied with their overall medical care, with nearly 60 percent higher odds of reporting a higher level of satisfaction. In interpreting the impact on satisfaction, it is useful to note that a large share of workers reported being very or somewhat satisfied with their overall care regardless of who chose their providers. This is true overall and in each of the four states (Table 3.2). Though this study is unique in evaluating satisfaction with care by the choice of the provider, the ranges reported of those who were satisfied (very and somewhat) are very much in line with other studies, both in workers’ compensation and in general health care.3 Chapter 6 discusses possible reasons why we found higher levels of satisfaction with health care when workers selected their providers despite no difference in recovery, and that chapter attempts to untangle the question of whether the higher satisfaction reflects other dimensions of the quality of medical care. 3 See Victor, Barth, and Liu (2003, pp. 115–117). For example, a national study reported that 55 percent of persons receiving medical treatment in the two years before the survey were “extremely or very satisfied” with the quality of the medical care they received (Employee Benefit Research Institute, Consumer Health Education Council, and Matthew Greenwald and Associates, 2002). 4 Employee Choice of Prior Provider or New Provider, or Employer Choice of Provider: Main Results Some workers we interviewed for this study had established relationships with providers before they were injured; others did not. This chapter examines whether the costs and outcomes differ among otherwise similar cases in which (1) the worker selected as the primary provider a “prior provider,” defined as someone who treated the worker before the injury for an unrelated condition; (2) the worker selected as primary provider a “new provider,” defined as someone who had not previously treated the worker; and (3) the employer selected the primary provider. In our view, the results from this chapter are more informative than those from the previous chapter from the perspective of assessing the implications of provider choice for public policy, because public policy can restrict only one type of employee choice of provider — such as California’s recent workers’ compensation reforms that restrict the worker’s ability to choose a new provider. We present estimates that first compare the costs and outcomes of cases with an employee-selected prior provider with those with an employer-selected provider and then compare the costs and outcomes of cases with an employeeselected new provider with those with an employer-selected provider. In addition, we present comparisons of the costs and outcomes of cases with employee-selected prior and employee-selected new providers. 39 40 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 4.1 Employee Choice of Prior Provider or New Provider Percentage of Workers (of those who chose the primary provider) Combined CA TX MA PA Prior provider New provider 45.5 54.5 50.3 36.3 52.4 45.2 49.7 63.7 47.6 54.8 Patterns of Choosing New and Prior Providers Among the cases in which the worker chose the primary provider, a prior provider was chosen about half the time in California, Massachusetts, and Pennsylvania but only about one-third of the time in Texas (Table 4.1). We cannot be certain why Texas workers were less likely to select prior providers as their primary providers. However, it is reasonable to suppose that injured workers who are not covered by health insurance are less likely to have established relationships with health care providers. We do not know from our survey whether injured workers had health insurance coverage. However, we do know that the population in Texas is much less likely to have health insurance coverage than are persons in the other three states.1 Additionally, a recent survey found that injured workers in Texas were less likely to have received general medical care recently than were injured workers in other states, consistent with fewer being covered by nonoccupational health insurance.2 Another possible reason for the relatively low rate of utilization of prior providers is that we found low rates of use among those with lower levels of education, and Texas workers are generally less educated than workers in the other three states,3 although low utilization may also reflect health insurance differences, because workers with less education are less likely to have health insurance. 1 For the period 2001–2003, the proportions of the population not covered by health insurance in the four study states and in the United States were California, 18.7 percent; Massachusetts, 9.6 percent; Pennsylvania, 10.7 percent; Texas, 24.6 percent (highest in the nation); and United States, 15.1 percent (DeNavas-Walt, Proctor, and Mills, 2004). 2 Research and Oversight Council on Workers’ Compensation and Med-FX, LLC (2001, p. 55). 3 See Technical Appendix Table B.2. employee or employer choice of provider: main results 41 Summary of Findings Compared with cases in which the employer chose the provider, we find that: ■ When the worker selected a provider who had treated the worker previously for an unrelated condition, there is some evidence that medical benefits were higher; however, indemnity benefits did not differ from those related to cases in which the employer chose the provider. There is little evidence of worse return-to-work outcomes. Recovery was unaffected by employee choice of a prior provider, but employee choice of a prior provider was associated with much higher satisfaction with overall care. ■ In contrast, when the worker selected a new provider (one that had not previously treated the worker), medical and indemnity costs were higher and return-to-work outcomes were poorer. Again, there was no difference in physical recovery, but satisfaction with care was greater. Impact on Costs and Outcomes of Employee-Selected Prior Provider or New Provider, or Employer Choice of Provider: Main Results This section presents the results in detail. We report estimates using the combined data from the four states, while controlling for material differences among states and cases. Again, we report the results for two specifications: the first might undercontrol for severity by excluding the treatment variables, while the second likely overcontrols for severity by including them. The true results probably lie somewhere between the results of the two models. When the results are similar in the two models, or where the results remain strong for the second model, the reader should have greater confidence in the conclusions. The evidence does not point to substantive differences in costs, return to work, or physical recovery between cases in which the employer chose the provider and those in which the employee chose a prior provider, although employee choice of a prior provider is associated with greater satisfaction with overall medical care. The results are presented in the first two columns of Table 4.2. In model 1, medical payments and duration of time until first substantial return to employment were higher. However, the differences become considerably smaller (shrinking by about two-thirds) and statistically insignificant in model 2, with the hospitalization and surgery controls added. Satisfaction with care is much more likely to be higher for either model. 42 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 4.2 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Four States Combined Employee Chose a Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Employee Chose a New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 22** ($1,924) 9 ($1,116) 17† –4 –3 86** 7 ($629) –1 (–$162) 7 3 –1 89** 20** ($1,745) 20** ($2,538) 48** –28** 2 38** 12* ($1,052) 15† ($1,879) 40** –28** 3 39** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. In model 1, for example, when the worker chose a prior provider, medical payments were $1,924 greater per case (on average) than when the employer chose the provider; or the duration of time out of work was 17 percent longer when the worker chose a prior provider than when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Specifically, as Table 4.2 shows, compared with cases in which the employer selected the provider, cases in which the employee chose a prior provider had: ■ Medical payments that were 7–22 percent higher, although the results are not statistically significant in one of the two models used. ■ Indemnity benefits that are not statistically different (estimates ranged from 1 percent lower to 9 percent higher). ■ Substantial return-to-work outcomes that are not statistically different (estimates of return-to-work rates that ranged from 4 percent lower to 3 percent higher). employee or employer choice of provider: main results 43 ■ Durations that were 7–17 percent longer, although the results were only weakly significant in one model and not significant in the other. ■ Recovery of physical health that was similar. ■ Much higher likelihood that the worker rated the care with higher satisfac- tion. By contrast, as the third and fourth columns of Table 4.2 show, comparing cases in which the employer chose the provider to those in which the worker selected a new provider provides stronger evidence that employee choice leads to higher medical and indemnity costs and poorer return-to-work outcomes. The statistical evidence is weaker in model 2 for indemnity benefits but remains quite strong for medical benefits and for both return-to-work measures. Again, employee choice of provider (in this case, a new provider) is not associated with better physical recovery, although it is associated with higher satisfaction. Specifically, as Table 4.2 shows, compared with cases in which the employer selected the provider, cases in which the employee chose a new provider (who had not provided prior treatment for an unrelated condition) had: ■ Medical payments that were 12–20 percent higher. ■ Indemnity benefits that were higher by 15–20 percent. ■ Much lower rates of substantial return to work and longer durations of time out of work, with workers 28 percent less likely to return to work and remain there for at least one continuous month compared with employerselected cases and their time until return to work 40–48 percent longer. ■ Recovery of physical health that was similar. ■ Nearly 40 percent higher likelihood that the worker reported a higher level of satisfaction with care. When the employee chose a prior provider, the outcomes for indemnity benefits, duration, substantial return to work, and recovery were generally close to those found when the employer made the selection, except perhaps for medical costs. However, when the employee chose a new provider, the costs were higher and most outcomes poorer than when the employer selected the provider. In either case, workers were more likely to report higher levels of satisfaction when they chose the primary provider — and especially so with a prior provider. The results in Table 4.2 suggest that the findings in Chapter 3 regarding higher costs and worse return to work associated with employee choice overall were driven, in large part, by employee choice of a new provider. That is, there are po- 44 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s tentially important differences in the costs and outcomes of cases in which the worker selected a prior provider compared with those in which the worker selected a new provider. Table 4.3 uses the estimates underlying Table 4.2 to show the differences in costs and outcomes between when the worker selected a prior versus a new primary provider and indicates which differences are statistically significant. The table reports the impact of the employee choosing a new provider compared with the employee choosing a prior provider.4 We see that medical payments were similar but indemnity benefits were 11–16 percent higher when the employee selected a new provider, although the latter difference is at best weakly statistically significant. With respect to return to work, the differences are sharper, with employee choice of a new provider associated with significantly poorer return-to-work outcomes. In particular, when the worker selected a new provider, the odds of having a substantial return to work were 26–30 percent lower, and the duration of time out of work was 26–30 percent longer. Finally, satisfaction was lower when the worker selected a new provider than when the worker selected a prior provider. Yet, as we have found throughout this report, choice was unrelated to physical recovery. Chapter 6 returns to the issue of provider choice, recovery, and worker satisfaction. 4 In other words, in Table 4.2, we report the differences associated with comparisons between the two types of employee choice, on the one hand, and employer choice, on the other. The results reveal considerably sharper differences between employee choice of a new provider and employer choice, compared with between employee choice of a prior provider and employer choice. Table 4.2 does not, however, address whether the differences associated with the two types of employee choice are significantly different from each other. If they are not, then arguably our best estimates come from the simpler models presented in Chapter 3, which constrain the effects of the two types of employee choice to be the same. The results in Table 4.3 come from including in the regression models [see equation (2.1)] a dummy variable for either type of employee choice and an interaction between this dummy variable and a dummy variable for employee choice of a new provider. The estimated coefficient of the latter interaction measures the difference between the two types of employee choice, and a test of its statistical significance tells us whether the two types of employee choice have significantly different effects. Table 4.3 reports these latter differences and their statistical significance. Thus, for example, the result for duration in the model 1 column means that durations were on average 26 percent longer with employee choice of a new provider than with employee choice of a prior provider, and this difference is significant at the 5 percent level. Note that this 26 percent figure is not simply the difference between the estimates for duration reported in Table 4.2 for the two types of employee choice for model 1, because the numbers reported in Tables 4.2 and 4.3 are calculated from the exponentials of the regression coefficients. employee or employer choice of provider: main results 45 Table 4.3 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction –2 (–$180) 11 ($1,422) 26** –26* 5 –26** 5 ($422) 16† ($2,040) 30** –30** 4 –26** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Revisiting the Question of Unmeasured Residual Injury Severity As discussed in Chapter 2, the question arises whether the estimates presented so far (and in the next chapter) reflect only provider choice or also reflect unmeasured residual variation in injury severity that is associated with provider choice. We noted that, especially in the models that control for treatment (model 2 in the tables), we are quite confident that the estimates reflect causal effects of provider choice. However, as a way of shedding a little more light on this question, Table 4.4 reports results in which, in a sense, we go in the opposite direction to what we did when we added the treatment variables. In particular, we begin here with the model 1 estimates and then successively drop the perceived severity variable, and then drop the “type of injury” variables as well. If unmeasured injury severity accounted for large shares of the apparent effects of provider choice on workers’ compensation outcomes, then when we drop the perceived severity measure, the effects of provider choice should appear even larger. However, as indicated in the first two columns of Table 4.4 for em- 46 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 4.4 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Four States Combined, Excluding Severity and Injury Measures Employee Chose a Prior Provider Employee Chose a New Provider Model 1: Without Treatment Controls (percent) Omit Severity Variable (percent) Omit Severity and Injury Variables (percent) Model 1: Without Treatment Controls (percent) Omit Severity Variable (percent) Omit Severity and Injury Variables (percent) Medical payments 22** ($1,924) 21** ($1,817) 22**($1,940) 20** ($1,745) 21** ($1,811) 23** ($2,027) Indemnity benefits 9 ($1,116) 7 ($916) 9 ($1,116) 20** ($2,538) 21** ($2,651) 24** ($3,093) Duration 17† 15 17† 48** 47** 52** Substantial return to work −4 −3 −7 −28** −28** −31** Recovery −3 −5 −5 2 3 3 Satisfaction 86** 87** 83** 38** 36** 32** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables, and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. ployee choice of a prior provider, and in the fourth and fifth columns for employee choice of a new provider, the estimates scarcely change when the perceivedseverity variable is omitted, casting doubt on an important role for unmeasured severity in the model 1 estimates. Taking this one step further, in the last column for each type of employee choice, we even drop the injury type variables, which surely capture information on the nature and severity of the injury. Here, especially for choice of a new provider, the estimated effects on costs and return to work grow (in absolute value), but only slightly. The implication is that it is unlikely that unmeasured injury severity materially distorts the estimated effects of provider choice that we find. 5 Impact of Provider Choice on Costs and Outcomes: Results for Individual States In this chapter, we consider evidence from each of the four study states on the impact of provider choice on workers’ compensation costs and outcomes. Because the sample sizes are much larger for the analysis using the combined data, we recognize that those results are more precise and statistically powerful and that the results for individual states are less informative. Nonetheless, the estimates presented here can help confirm which states show overall patterns of results similar to the results for the combined sample. The structure of this chapter parallels the analyses reported in Chapters 3 and 4. First we show the differences in costs and outcomes where the employee or the employer chose the primary provider, and then we analyze whether there appear to be differences when the employee selected a new or a prior primary provider. The workers’ compensation systems in the four study states have some similarities and some differences. Technical Appendix F provides a summary of some of the relevant attributes of the workers’ compensation systems in each of the four states. Those attributes include costs, litigiousness, benefit levels, and a variety of factors associated with medical care, including provider choice rules, fee schedules, and limits on who can treat injured workers. We provide Technical Appendix F for readers who would like to understand some of the institutional features that might shape the provider decision or the consequences of that decision. Citations to sources for the information summarized here are contained in the appendix. 47 48 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Here are some of the important differences: ■ Coverage was mandatory with a few exceptions in each of the states except Texas, where employers can elect not to participate in workers’ compensation. According to a recent study by the Texas Department of Insurance (2004), 24 percent of the workforce was not covered by workers’ compensation insurance in 2004. ■ The average cost of a claim is much higher in California and Texas than in Massachusetts and Pennsylvania. In California, both medical and indemnity benefit costs are high. In Texas, medical costs and the duration of disability are the main cost drivers. In Pennsylvania, medical and indemnity costs are typical, while in Massachusetts, medical costs per claim are significantly lower than the other three states. ■ All four states have medical fee schedules that regulate nonhospital provider fees. In 2001, the fee schedule levels were, on average, much lower in Massachusetts than in the other three states. This was true for all the major services provided to injured workers. ■ States with the lowest fee schedules, Massachusetts and Pennsylvania, have the lowest network penetration rates. This occurs because opportunities to obtain network discounts are limited when regulated prices are already low. ■ Litigation is relatively infrequent in Texas but much more common in the other states, especially in California. ■ A much larger percentage of the general population is covered by nonoccupational health insurance in Massachusetts and Pennsylvania than in California and Texas. As we did in Chapters 3 and 4, we report here the results for two models, excluding and including the treatment variables as additional proxies for severity, and for the two-way and three-way classifications of provider choice. California A comparison of the findings for California (Table 5.1) with the more precise and statistically powerful multistate results (see Table 3.1) reveals similarities. When the employee chose the primary provider, costs were higher, return-to-work outcomes were poorer, recovery was no different, and satisfaction with overall care was higher. There are also marked differences between when the employee chose a impact of provider choice on costs and outcomes 49 Table 5.1 Impact of Employee Choice Compared with Employer Choice, California Model 1: Without Treatment Controls (percent) Model 2: Without Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 22* ($2,210) 26** ($3,946) 48** −16 −2 34† 10 ($1,031) 18† ($2,840) 39** −12 −1 37* Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state; see the discussion on weighting in Technical Appendix D. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. new provider and when the employer chose the provider, as shown in Table 5.2 (which can be compared with Table 4.2 for the combined sample). As Table 5.1 indicates, in cases in which the worker selected the provider, medical payments were 10–22 percent higher, although only the higher estimate is statistically significant. Indemnity benefits were 18–26 percent higher, and only the higher estimate is strongly significant. In these cases, the worker was 12–16 percent less likely to have a substantial return to work in the 3.0 to 3.5 years between injury and interview compared with cases in which the employer selected the provider, but the difference is not statistically significant. Workers reported a 39–48 percent longer time span between injury and initial return to substantial employment when the employee chose the provider. We found little difference in perceived recovery of physical health between cases in which the employee chose the primary provider and those in which the employer chose. When workers selected their providers, they were 34–37 percent more likely to report a higher level of satisfaction with care. 50 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 5.2 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, California Employee Chose a Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Employee Chose a New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 18 ($1,778) 2 ($367) 11 48 −10 24 9 ($944) −2 (−$373) 10 25† ($2,507) 47** ($7,345) 102** 9 ($940) 37** ($5,796) 77** 54 −44* −9 6 25 42† −41* 7 48* Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. As shown in Table 5.2, when, the differences in costs and outcomes are not statistically significant different between cases in which the worker chose a prior provider and those in which the employer chose the provider.1 In contrast, comparing employee choice of a new provider to employer choice, indemnity benefits were 37–47 percent higher, both statistically significant. Medical payments were 1 A prior provider is defined as a provider who previously treated the worker for an unrelated condition, regardless of whether the worker predesignated that provider under the prevailing California law. Under the old law, a predesignated provider need not have previously treated the worker; under the new law, a predesignated provider must be a personal physician under a group health plan who agrees to the predesignation. Our measure of prior provider is different from either statutory definition and, in some respects, broader. impact of provider choice on costs and outcomes 51 Table 5.3 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, California Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 7 ($729) 45** ($6,978) 82** –62** 16† 14 0 (–$5) 40** ($6,170) 61** –61** 16† 19 Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. also higher by 9–25 percent, but not significantly so at the lower estimate and only marginally significant for the higher estimate. Workers who saw new providers were much less likely (41–44 percent) to have substantial returns to work and had much longer (77–102 percent) periods until substantial returns to employment. Workers who selected new providers were more likely (42–48 percent) to report higher levels of satisfaction with care than when employers chose the providers, although as has proven to be common throughout our analyses, there is no evidence of differences in physical recovery. The results also reveal important differences in the costs and outcomes of cases in which the worker selected a new provider compared with cases in which the worker chose a prior provider. Table 5.3 shows these differences and indicates which differences are statistically significant in California. In particular, indemnity benefits were about 40–45 percent higher when the worker selected a new provider rather than a prior provider. Medical payments were not significantly different. Indemnity benefits were higher in part because workers who saw new providers were much less likely (61–62 percent) to have substantial returns to work and had 52 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s longer durations of time out of work (61–82 percent). Table 5.3 also shows that workers who saw new providers had 16 percent better recoveries of physical health — although this is only marginally significant in either of the two California models. Note that the latter result is the one of the few pieces of evidence in the entire report that gives any indication of better physical recovery when the employee chooses a new provider, whereas the evidence of higher costs and worse return to work is ubiquitous. Although our overall conclusions suggest, therefore, that employee choice of a new provider raises costs without improving physical recovery, it is important to keep in mind that for California, we have to be a little more cautious in reaching the conclusion that employee choice of a new provider does not deliver better recovery than employee choice of a prior provider, given the results on recovery shown in Table 5.3. However, this evidence is only marginally statistically significant. Texas Like the findings for California, those for Texas are in many ways similar to the more precise and statistically powerful multistate results. When the employee chose the primary provider, costs were higher, return-to-work outcomes appear to be poorer, recovery was no different, and satisfaction with overall care was higher. For Texas, some of these results hold for employee choice of a prior provider as well as a new provider, although more so for choice of a new provider. Differences also exist between cases with an employee-selected prior provider and those with an employee-selected new provider. In particular, with a new provider, return-towork outcomes were poorer and satisfaction with overall care was lower. However, we found no material differences in perceived recovery or costs. As Table 5.4 indicates, in cases in which the worker selected the provider, medical payments were 22–24 percent higher, and indemnity benefits were 21 percent higher. In these cases, the worker was 21–24 percent less likely to have a substantial return to work in the 3.0 to 3.5 years between injury and interview than was the worker whose employer selected the provider, and the duration of time until return to substantial employment was 28–31 percent longer, although the returnto-work differences generally are not statistically significant. We found little difference in perceived recovery of physical health between cases in which the employee or the employer chose the primary provider. The worker who selected the provider was 57–58 percent more likely to report higher satisfaction with overall medical care. impact of provider choice on costs and outcomes 53 Table 5.4 Impact of Employee Choice Compared with Employer Choice, Texas Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 24* ($2,836) 21* ($2,152) 31† –21 –3 58** 22** ($2,558) 21** ($2,169) 28 –24 –3 57** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. When the worker selected a prior provider, compared with cases in which the employer chose, indemnity benefits were 16–22 percent higher, while medical payments were 21–31 percent higher, although most of these estimates are at best marginally significant (Table 5.5). Workers reported similar recovery of physical health, regardless of who selected their providers, and return-to-work differences are relatively small and statistically insignificant. Workers were much more likely to report higher satisfaction when treated by prior providers that they chose compared with employer-selected providers. Comparing cases in which the worker selected a new provider to cases with an employer-selected provider, costs were higher and return-to-work outcomes poorer when the worker chose, although some of these estimates are only marginally significant. Satisfaction with care was about 23 percent higher, but the difference is not statistically significant. Recovery of physical health was similar. As Table 5.5 indicates, medical payments and indemnity benefits per case were 19–23 percent higher when the worker chose a new provider. In such cases, the worker was 31–36 percent less likely to have a substantial return to work. Compared with 54 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 5.5 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Texas Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 31* ($3,639) 22† ($2,273) 10 12 2 159** 21† ($2,421) 16 ($1,613) 7 20 2 160** 19† ($2,220) 22* ($2,504) 19† ($1,926) 23** ($2,348) 44† 42† –31† –36† –5 –5 24 23 Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. cases in which the employer chose the provider, the duration of time until a substantial initial return to employment was about 43 percent longer. Table 5.6 compares Texas cases in which the worker selected a prior provider with cases in which a new provider was selected. The evidence indicates that return-to-work and satisfaction outcomes were poorer in cases in which the employee selected a new provider rather than a prior provider. A worker in this group was 38–47 percent less likely to have a substantial return to work. When the worker selected a new provider, the duration of time until return to substantial employment was 31–33 percent longer, although these differences are not significant. Additionally, a worker choosing a new provider was 52–53 percent less likely to report a higher level of satisfaction with care; that is, the worker was much more likely to report lower satisfaction with care compared with a worker who chose a prior provider. impact of provider choice on costs and outcomes 55 Table 5.6 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Texas Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction −12 (−$1,419) −3 (−$347) 31 −38† −7 −52** 1 ($83) 7 ($735) 33 −47* −7 −53** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Massachusetts The results for Massachusetts parallel the more precise and statistically powerful multistate results along some dimensions only, and the differences associated with employee versus employer choice are generally not statistically significant. The same is true when we look at differences between employee choice of prior and new providers. The strongest result for Massachusetts (and the only statistically significant one) is that in cases in which the worker selected the provider, the worker was much more likely to report higher satisfaction with care than when the employer chose. Medical payments in Massachusetts are unlikely to be much different regardless of whether the employee or employer selects the provider. As in the other three states, in cases in which the worker chose the provider, the worker was less likely to have a substantial return to work and had a longer time out of work. The estimated coefficients on the return-to-work outcomes are similar to those reported earlier, but they are not statistically significant. As in the other states, per- 56 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 5.7 Impact of Employee Choice Compared with Employer Choice, Massachusetts Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 10 ($514) 11 ($1,486) 19 −33 −1 83** −2 (−$106) 7 ($956) 17 −33 0 74** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. ceived recovery of physical health was not affected by who selected the provider. These results are reported in Table 5.7. Turning to employee choice of prior and new providers, the estimates reveal no consistent pattern of higher costs associated with either type of employee choice of provider (Table 5.8). The estimates are generally consistent with employee choice of a new provider worsening return-to-work outcomes, although again the estimates are not statistically significant. The only strong result is that in cases in which workers selected prior or new providers, they were much more likely to report higher satisfaction with care. Not surprising, given the lack of statistical significance in these findings, we also see no statistically significant differences between employee choice of prior or new providers. Table 5.9 shows that there are no statistically significant differences in costs or outcomes between cases in which the worker chose a prior provider compared with those in which the worker chose a new provider. Worker satisfaction appears to be higher when a prior provider was used and return-to-work outcomes worse, but neither of these differences is statistically significant. impact of provider choice on costs and outcomes 57 Table 5.8 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Massachusetts Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 19 ($955) 19 ($2,672) 13 −38 −1 99** −1 (−$61) 11 ($1,526) 6 −38 0 92** 2 ($119) 3 ($393) 26 −27 −1 74** −3 (−$171) 3 ($387) 28 −28 −1 65* Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Table 5.9 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Massachusetts Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction −17 (−$836) −16 (−$2,279) 11 17 0 −13 −2 (−$110) −8 (−$1,139) 20 16 0 −14 Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. 58 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Pennsylvania The results for Pennsylvania provide less evidence that is similar to the more precise and statistically powerful multistate results. The strongest result for Pennsylvania is that in cases in which the worker selected the provider, the worker was much more likely to report higher satisfaction with medical care than when the employer chose (Table 5.10). Medical payments in Pennsylvania are unlikely to be much different regardless of whether the employee or employer selects the provider. As in the other states, in cases in which the worker chose the provider, the worker was less likely to have a substantial return to work than when the employer chose; but in Pennsylvania, these effects are not statistically significant. Again, as in the other states, perceived recovery of physical health was not materially affected by choice of provider, looking at the two-way classification of employee versus employer choice. The Pennsylvania results parallel the multistate results when we look at whether workers selected new providers or employers chose, but the results are somewhat different when we look at when employees chose prior providers com- Table 5.10 Impact of Employee Choice Compared with Employer Choice, Pennsylvania Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 13 ($997) −6 (−$683) 22† −22 4 72** −3 (−$210) −19 (−$2,113) 16 −19 6 72** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. impact of provider choice on costs and outcomes 59 Table 5.11 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Pennsylvania Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments 3 ($234) −14 (−$1,041) 22† ($1,709) 8 ($579) Indemnity benefits −28 (−$3,144) −39† (−$4,448) 14 ($1,586) 1 ($62) Duration 14 9 32* 26† Substantial return to work −16 −7 −29 −31 Recovery −3 −1 11† 12* Satisfaction 87** 85** 58** 60** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. pared with when employers chose the providers. As before, when workers selected prior providers, they were much more likely to report higher satisfaction with care (Table 5.11). However, there is some marginally significant evidence that indemnity costs are lower when workers choose prior providers, even though the estimates indicate worse return-to-work outcomes (which are relatively small and not significant). When the worker selected a new provider compared with the employer selecting the provider, there is no consistent statistically significant difference in costs. We did find longer duration and lower rates of return to work, although only the former is statistically significant and only in model 1. We also, as before, found higher rates of satisfaction with overall care. Finally, employee choice in Pennsylvania may improve physical recovery. When we compare the costs and outcomes of cases with worker-selected prior or new providers (Table 5.12), we find that both indemnity and medical payments may be higher with new providers, but the evidence is at best marginally sig- 60 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 5.12 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Pennsylvania Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 19 ($1,475) 42† ($4,730) 16 −16 14* −15 21 ($1,620) 40† ($4,510) 16 −25 14† −14 Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. nificant. New providers were associated with 14 percent better physical recoveries. The result is not consistent with what we have found in the combined data, although as noted earlier, we found a similar — albeit also weak — result for California. None of the other differences (return to work or satisfaction) are statistically significant. 6 Worker Satisfaction with Health Care One of the strong and consistent findings in this study is that worker satisfaction with the overall health care received was higher when the worker chose the primary provider. In this chapter, we probe the possible reasons underlying this finding. We are particularly interested in trying to understand the higher satisfaction associated with employee choice of provider because the evidence shows that employee choice is not associated with better physical recovery as reported by the worker. The survey asked workers, “Now think about all the medical care you received from the first treatment for your injury until now. Were you satisfied or dissatisfied with the medical care you received overall?” This summary question enabled respondents to take account of their satisfaction with multiple aspects of health care delivery, and overall the survey found that workers generally reported high levels of satisfaction with their care.1 However, we also found that employee choice of primary provider was associated with higher rates of satisfaction than was employer choice. 1 The survey also asked workers about other metrics of satisfaction: satisfaction with the providers; with the availability, timing, and kind of care sought; with their desire to change providers at some point during treatment; and with the process of providing care. The patterns of responses on these metrics can be found in Barth and Victor (2003) and Victor, Barth, and Liu (2003). 61 62 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s In this chapter, we examine evidence on several conjectures regarding the higher satisfaction associated with employee choice of provider: 1. Some employee-selected providers might achieve better physical recoveries for their patients, and workers might be more satisfied with their better recoveries. 2. Some employer-selected providers might tend to rush workers back to work prematurely; therefore, workers who return before they are ready might have lower satisfaction with care. 3. Some workers might prefer not to go back to work, or prefer to delay their return to work beyond the time that they may be physically able to return, and worker-selected providers might be more likely to support their preferences; therefore, these workers might be more satisfied with such care even if physical recovery is no better. 4. Workers might have certain expectations about the processes of care (for example, speed of first visit, time spent with provider, and bedside manner), and employee-selected providers might be more likely to meet those expectations, regardless of physical recovery. 5. Some workers might experience an “empowerment effect” when they select their own providers, and by itself this effect might lead to higher levels of satisfaction regardless of physical recovery. 6. Some workers might suspect that employer-selected providers are more concerned with satisfying the needs of the employer than of the worker. Such a suspicion could result in a lower degree of trust and hence lower satisfaction with the treatment, even if recovery is not affected. The data allow us to examine conjectures 1, 2, and 3 but not the other possibilities. In the sections that follow, we discuss evidence on each of the first three conjectures and rule them out. This leaves open the possibility that the impact of provider choice may be the result of satisfaction with the process (rather than outcomes) of care, an “empowerment effect,” or an attitude about the perceived motivation of the provider. What unifies these three conjectures is that they have more to do with the process of the medical care, not the outcomes, especially the physical recovery of the injured worker. Correlates of Satisfaction and Overall Health Care Received We begin by examining a number of factors correlated with satisfaction, as suggested by conjectures 1, 2, and 3 listed in the previous section. Then we examine worker satisfaction with health care 63 whether each is affected by provider choice. The evidence is very clear that higher satisfaction with overall care is strongly related to (1) a better recovery of physical health, (2) a worker’s belief that he or she was not sent back to work too soon, and (3) having a sustained return to work, uninterrupted by subsequent periods of lost time due to the injury. However, it turns out that these correlates of satisfaction are not associated with who chose the provider, which is the evidence that rules out the first three conjectures in our list. We begin by categorizing the recovery scores of respondents. Recovery is classified into groups based on the change in the standardized SF-12® scores measured soon after the injury and at the time of the interview. (Recovery is defined in Chapter 2 and in Technical Appendix C.) We defined the groups as follows: change of at most −2 points (complete recovery); −2 to −10 points (somewhat incomplete); −10 to −20 points (more incomplete recovery); and exceeding −20 points (very incomplete). As Table 6.1 shows, workers who reported less recovery of physical health were less likely to report higher satisfaction with their overall health care. Of those reporting complete recoveries, 91 percent said that they were very or somewhat satisfied with their care. In contrast, among workers reporting the least complete recoveries, only 67 percent said that they were very or somewhat satisfied.2 Although satisfaction and recovery are strongly correlated, we previously reported the findings that provider choice did not affect recovery of physical health. Thus, we rule out the first conjecture listed in the previous section. Workers who believed that they returned to work too soon were less satisfied with their care. Table 6.2 shows that 91 percent of workers who reported that they returned to work at the “right time” were very or somewhat satisfied with care. Those who said they returned to work “too soon” were less satisfied — 74 percent reported being very or somewhat satisfied with care. Sixteen percent of those who reported returning to work too soon were very dissatisfied with their overall care compared with only 4 percent of those who said they returned to work at the right time. These workers may have thought that their health care providers were responsible for having them return to work prematurely. 2 It may seem surprising that 2 out of 3 persons with very incomplete recoveries told us that they were very or somewhat satisfied with the care that they received. Yet this may be explained when one considers that the workers expressed satisfaction with the care that they received and not with their conditions following their injuries and treatments. That is, someone with a more serious medical condition may not expect a complete recovery, and the worker may measure satisfaction against his or her expectations. 64 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 6.1 Satisfaction with Care, by Degree of Recovery of Physical Health Complete Recovery (percent) Somewhat Incomplete Recovery (percent) More Incomplete Recovery (percent) Very Incomplete Recovery (percent) Very or somewhat satisfied 91 86 78 67 Very satisfied 65 53 44 36 Somewhat satisfied 26 33 34 31 Very or somewhat dissatisfied 9 14 22 33 Somewhat dissatisfied 5 6 11 15 Very dissatisfied 4 8 11 18 Note: Recovery categories based on changes in standardized SF-12® scores. Complete recovery: change of at least −2 points; somewhat incomplete: change of −2 to −10 points; more incomplete recovery: change of −10 to −20 points; very incomplete: change of more than −20 points. Table 6.2 Worker’s Perception of Timing of Return to Work and Satisfaction with Care Very Satisfied (percent) Somewhat Satisfied (percent) Somewhat Dissatisfied (percent) Very Dissatisfied (percent) Right time Too soon 63 42 28 32 6 11 4 16 A worker who had a second absence due to his or her injury was also likely to be less satisfied with the health care received. Seventeen percent of workers who had a substantial return to work reported that they experienced a second significant absence (lasting more than one week). Respondents may have perceived the second absence as the result of a premature return to work or returning without adequate work limitations prescribed. Table 6.3 shows that 88 percent of workers who reported no second absence due to the injury said that they were very or somewhat satisfied with care. Those who had a second absence were less worker satisfaction with health care 65 Table 6.3 Second Absence Due to Injury and Satisfaction with Care No Second Absence (percent) Second Absence (percent) Very satisfied Somewhat satisfied Somewhat dissatisfied Very dissatisfied 60 28 6 6 35 32 14 19 satisfied — 67 percent reported being very or somewhat satisfied with care, and 33 percent were somewhat or very dissatisfied. Correlates of Satisfaction and Provider Choice Are workers with less complete recoveries more likely to have providers selected by employers? We have seen that workers who have less complete recoveries have lower levels of satisfaction with care. However, there is no significant difference in recovery between cases in which the employee or employer selected the provider. Hence, it is unlikely that the large impact of employee choice on satisfaction is due to employee-selected providers achieving better physical results for workers. We show this in two ways. First, in Tables 3.1 and 4.2, we showed that provider choice had no statistically significant effect on recovery, controlling for a large number of injury, worker, and employer characteristics that could affect recovery. Second, in Table 6.4, we show a simpler comparison with no controls for other factors. Cases in which the employer selected the provider were equally likely to report an incomplete recovery as cases in which the employee selected the provider. Are workers who report that they returned to work “too soon” more likely to have providers selected by employers? As shown in Table 6.2, workers who believed that they returned to work too soon were less satisfied with their care. However, workers who chose their providers were equally likely to report that they returned to work too soon as workers whose employers chose the providers (Table 6.5). Thus, it is unlikely that the large impact of employee choice on satisfaction is due to employer-selected providers rushing workers back to work prematurely. We 66 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 6.4 Completeness of Recovery, by Who Selected the Provider Complete Recovery (percent) Somewhat Incomplete Recovery (percent) More Incomplete Recovery (percent) Employee-selected provider Employer-selected provider 41 41 19 21 14 13 Very Incomplete Recovery (percent) 27 25 Note: Recovery categories based on changes in standardized SF-12® scores. Complete recovery: change of at least −2 points; somewhat incomplete: change of −2 to −10 points; more incomplete recovery: change of −10 to −20 points; very incomplete: change of more than −20 points. Table 6.5 Worker’s Perception of Timing of Return to Work, by Who Selected the Provider Employee Chose (percent) Employer Chose (percent) Right time Too soon 63 37 65 35 confirmed this conclusion using multivariate statistical techniques that control for the factors described in Chapter 2. Is a worker who reported a second significant absence due to the injury, after having achieved a substantial return to work, more likely to have an employer-selected provider? As seen in Table 6.3, a worker who had a second absence was likely to be less satisfied with his or her care. However, a worker who chose the provider was equally likely to have a second absence as a worker whose employer chose the provider (Table 6.6). Therefore, it is unlikely that the large impact of employee choice on satisfaction is due to employer-selected providers returning workers to work prematurely, leading those workers to experience second absences. We confirmed this conclusion using multivariate statistical techniques that control for the factors described in Chapter 2. worker satisfaction with health care 67 Table 6.6 Second Absence Due to Injury, by Who Chose the Provider Employee Chose (percent) Employer Chose (percent) No second absence Second absence 82 18 82 18 Is the greater satisfaction with care when employees select their providers a re- flection of employee-selected providers who support the desires of some workers to return to work more slowly? Some workers might prefer not to return to work or to return to work more slowly. If employee-selected providers support this desire more often than employer-selected providers, we would expect that these workers to be more likely to report that they returned to work “too soon” when they saw employer-selected providers. We find no evidence that, on average, higher satisfaction with care from employee-selected providers derives from those providers supporting the preferences of some workers to return to work more slowly. As Table 6.5 showed, the percentage of workers reporting that they went back to work too soon was similar, regardless of who chose the provider. Satisfaction with Care: Prior or New Provider In considering the source of worker satisfaction or dissatisfaction with the health care received, we also note another strong and consistent finding. Comparing the satisfaction rates of workers who selected primary providers who had previously treated them for an unrelated condition with the satisfaction rate of workers who selected new providers, the former was significantly higher (see Table 4.3). Satisfaction was also higher for both types of employee choice compared with employer choice (see Table 4.2). We explored all the questions considered earlier in this chapter for the expanded three-way classification of provider choice to assess whether the higher satisfaction with employee choice of a prior or new provider, relative to employer choice, or higher satisfaction with employee choice of a prior rather than a new provider was attributable to any of the conjectures outlined at the beginning of this chapter. However, paralleling the analysis of provider choice based on the 68 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s two-way classification of employee versus employer choice, we found no evidence that the higher satisfaction with employee choice of either prior or new providers (or prior relative to new providers) derived from better physical recoveries, differences in the extent to which workers were rushed back to work, or greater ability to delay return to work. Again, we are led to the conclusion that differences in satisfaction are more likely due to factors such as the manner in which care was delivered, empowerment of the worker, and trust rather than more objective medical outcomes. 7 Discussion and Policy Implications With workers’ compensation medical payments high and rising rapidly in many states (Telles, Wang, and Tanabe, 2004), policymakers have intensified their efforts to modify state laws to try to reduce these costs while avoiding actions that might impair the outcomes experienced by injured workers. One of the actions often debated is giving employers more influence or direct control over the selection of providers. As discussed in Chapter 1, providers play many important roles in workers’ compensation cases and is believed to have a large impact on workers’ compensation costs and on outcomes for workers. During the period of rising costs between the late 1980s and early 1990s, several states modified “employee choice” laws to require that workers select providers from within approved networks of providers created by the employers. In California, an important cost containment provision of the 2004 legislative changes requires a worker to choose a provider from employer-selected networks of providers, unless the worker predesignates a provider who has previously treated the worker under a qualifying employer-sponsored group health plan. Previous studies of provider choice have focused primarily on the impacts on medical costs. Most reported that employee choice was associated with higher costs, as has this study. This study, however, is one of the first that rigorously examines not only costs but also worker outcomes associated with whether employers or workers actually select the provider, apart from statutory mandates on choice. It is also unique in looking at a variety of outcomes indicative of return to work and the quality of care and in focusing on the primary provider. Finally — and perhaps most relevant in light of the recent workers’ compensation reforms in 69 70 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s California — it is the only study that examines how the costs and outcomes of treatment differ when workers chose providers who previously treated them and when workers choose new providers. Summary of Results These are among the more important findings of this study: ■ Comparing cases in which the worker selected the primary provider with otherwise similar cases in which the employer selected the provider, we found that costs were generally higher and return-to-work outcomes poorer when the worker selected the provider, although workers reported higher rates of satisfaction with overall care but similar perceived recovery of physical health. ■ Compared with cases in which the employer selected the provider, cases in which the worker selected a provider who had treated the worker previously for an unrelated condition (“prior provider”) may have had higher costs, but the evidence was weak. Worker outcomes did not appear to be very different between cases with an employee-selected prior provider and those with an employer-selected provider, except that satisfaction with overall care was higher when the worker saw a prior provider. ■ Compared with cases in which the employer selected the provider, cases in which the worker selected a provider who had not treated him or her previously (“new provider”) had much higher costs and poorer return-to-work outcomes, generally no differences in physical recovery, and higher levels of satisfaction with overall care. ■ Comparing cases in which the employee selected a prior provider with similar cases in which the employee chose a new provider, we found that the worker treated by a new provider was less likely to return to work, returned to work more slowly if he or she did return, had lower levels of satisfaction with overall care, and experienced no better physical recovery. Medical payments were similar in both cases, but indemnity benefits per claim were higher for the worker treated by a new provider, although this evidence is statistically weaker than the other results. As discussed in Chapters 3 and 4, the primary findings summarized here come from the combined sample from four states (California, Texas, Massachusetts, and Pennsylvania), because compared with the individual states, the combined sample discussion and policy implications 71 size is much larger and the estimates are more precise and statistically powerful. We also present the individual state results, despite the fact that they are less precise and the statistical tests are less powerful. ■ In the two states with higher-than-typical medical payments, California and Texas, comparing cases in which the worker selected the primary provider with otherwise similar cases in which the employer selected the provider, we found some evidence of higher costs and poorer return-to-work outcomes when workers selected their providers, although workers reported higher rates of satisfaction with overall care and similar perceived recovery of physical health. ■ Costs for care were not significantly different between employer-selected and employee-selected providers in the two states (Massachusetts and Pennsylvania) with medical payments that were typical or lower than typical among states. However, duration of lost time may have been longer in Pennsylvania when workers select their providers. Satisfaction with care was higher when workers selected their providers, and there was no difference in perceived recovery of physical health. ■ The findings previously described for the combined sample were especially strong for California and Texas — although in Texas, cases with employeeselected prior providers and new providers had higher costs than cases with employer-selected providers. ■ In Pennsylvania, when the worker selected a prior provider, it appears that indemnity costs were lower than when the employer chose the provider, although the difference is at best weakly significant. However, paralleling other results, duration was longer when the worker selected a new provider compared with when the employer selected the provider. Moreover, workers reported higher levels of satisfaction when they chose their providers. Pennsylvania is the only state for which there is some statistically reliable evidence that employee choice of new providers may lead to slightly better physical recovery, and even then the evidence is relatively weak. Interpretative Caveats We are mindful of the need for care in interpreting and using the results from this study. First, only four states are included, and a wider set of states would add information that either reinforces the findings or is less consistent with them. Moreover, the focus of this study is who actually chose the primary providers in specific 72 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s cases, rather than the impact of the state legal provision about choice of initial provider or the laws about ongoing control of provider choice. State laws do appear to influence the actual choice, but there is hardly a perfect correspondence. Also, the reader is cautioned that the California and Pennsylvania laws and practices in effect during the sample period were not strong versions of employer choice laws. In both states, the employer retained the right to select the provider for only a limited period, after which the worker could change providers. Therefore, one needs to be careful about extrapolating from our findings the impact of changing state laws about who controls the choice of provider. The study includes both closed and open claims. However, cases are slower to close in California than in most other states. Consequently, the underlying data on paid costs understate the ultimate costs in California more than in the other states. This study does not address whether this understatement is associated with provider choice. If so, the magnitudes of the effects of provider choice on costs in this study could be biased. We do not explore the relationship between the choice of the initial provider and outcomes. Nor do we analyze the role of medical networks in the selection of providers and the costs and outcomes that result. For example, it seems more likely that a provider selected by an employer would be drawn from a network than would a provider selected by a worker. In future studies, we may address these issues. Finally, although we regard this study as an important addition to a relatively sparse empirical literature on a very important public policy issue, it is just one study. Additional research on other states and using other data sources and approaches will be useful to see if these results are robust, if they are supported in other contexts, whether provider choice has different effects in certain types of states but not in others, and how provider choice affects outcomes other than costs. With these caveats in mind, we conclude by discussing the implications for public policy that we can draw from our findings. Implications for Public Policy How do our results apply to public policy debates regarding choice of provider in workers’ compensation? The major themes of the policy debate often involve employer advocates arguing for employer choice on the grounds that (1) employers are better positioned to select good-quality providers because they, or the insurers, have better information; or (2) employee choice invites workers and attorneys discussion and policy implications 73 to game the system by selecting providers willing to extend the duration of time out of work in order to achieve financial gain for the providers or the workers. Employee advocates argue for worker choice of provider on the grounds that (1) workers are best positioned to select the most appropriate providers for their individual situations; and (2) some employers take advantage of the power to choose providers by selecting lower-cost providers who deliver inferior care or who may have a propensity to send workers back to work prematurely. This study finds some evidence to support both sides of the argument. As described in this report, however, based on our findings, it appears possible to improve the design of provider choice laws to lower costs and improve return-to-work outcomes without adversely affecting physical recovery from workplace injuries. First, we found that when workers chose their providers, costs were higher, recovery of health outcomes were not better, and return-to-work outcomes were often worse compared with when employers selected the providers. This finding suggests that employers, on average, may be well positioned to select good-quality, lower-cost providers — or at least better positioned than many workers. The finding also suggests that employers, in practice, are not generally selecting inferiorquality providers; although there may be exceptions, they are not frequent enough to affect the overall results. Second, we found that when workers selected prior providers, the costs and outcomes were not dramatically different from when employers selected providers. This evidence suggests that state laws that grant employers greater influence over the choice of provider should lead to lower costs and better return-to-work outcomes than laws that allow workers to select providers whom they have not seen previously — consistent with recent legislative changes made in California in Senate Bill 899. However, when workers selected providers — either prior or new — they expressed higher levels of satisfaction with care. We are not surprised by this finding regarding workers choosing prior providers, because the key issue is the likelihood that the worker will be seen by a provider who has the appropriate training and skills, is trusted by the worker, and delivers appropriate care. When a worker sees a provider with whom he or she has a preexisting relationship, a sense of trust may have already been established. In addition, the need for some diagnostics may be avoided, and the confidence that the worker has in a prior provider might increase the odds of compliance with treatment as well as ensure more open communication with the provider. The worker may also be confident that the provider has the necessary training and skills to treat the particular injury. When a worker needs to find a new provider, the employer or insurer is often 74 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s well positioned to identify the appropriate provider and help the worker get access in a timely manner, because insurers and employers typically have more experience and superior information about the practices of various providers. More surprising, though, is that workers also expressed greater satisfaction when they selected new providers (relative to employers choosing). We explored whether this greater satisfaction was related to dimensions of physical recovery not captured in our data or assistance in remaining out of work beyond the necessary time following an injury, but we were able to rule out such explanations. There may, however, be alternative explanations relating to empowerment or trust that leave workers more satisfied with choice of new providers even though costs and return-to-work outcomes are worse and physical recovery no better, perhaps stemming from distrust of providers chosen by employers and thus dissatisfaction with employer-selected providers. One possibility revolves around the process by which a worker accesses a health care provider following an injury. It may be that the worker does not have a preexisting relationship with a provider and does not wish to accept the employer’s recommendation. Alternatively, the worker may have a preexisting relationship with another provider but does not see that provider as having the requisite training or skills to treat the particular injury. Arguably, these types of considerations regarding higher satisfaction when employees choose their providers are of less interest to policymakers concerned with striking a balance between reducing the costs of workers’ compensation and ensuring adequate medical care and indemnity benefits. Overall, then, we view the higher costs and worse return-to-work outcomes associated with employee choice of new providers, coupled with no better physical recovery, as most consistent with some of the arguments in favor of employer choice. It remains to some extent a matter of speculation as to why costs appear to be higher and recovery no better when the employee chooses the provider — especially a new provider. For example, a worker trying to choose a new provider may not have adequate information about provider quality and may lack leverage to gain access to high-quality providers; the providers considered high quality may not be taking new patients or may be scheduling with significant delays. In contrast, the employer or insurer (or network), through its purchasing power, might be able to help the worker “jump the line” in when the employer chooses the provider. We cannot say for sure, but our results are consistent with situations in which the worker without a preexisting provider relationship is forced to participate in a search process with inadequate information about quality providers and inadequate leverage to gain access to those providers — almost a lottery-like pro- discussion and policy implications 75 cess compared with the situation in which the worker has a preexisting provider relationship or follows the employer’s recommendation. There are other reasons why an employer-selected provider might produce better outcomes at lower costs for a worker without a preexisting provider relationship. First, a provider selected by the employer might be more knowledgeable about the working environment and therefore better equipped to recommend sound return-to-work conditions. Second, many employers participate in medical network arrangements, which conduct provider credentialing and obtain fee discounts. Third, a new provider who is unfamiliar with the worker might feel compelled to practice defensively, thereby contributing to higher treatment costs and possibly to higher indemnity costs as well. Fourth, the findings about higher costs and poorer return-to-work outcomes with employee choice were stronger in the two study states with higher-than-typical medical payments and longer-thantypical durations of disability — California and Texas. The results were weaker in the two states with typical or lower-than-typical medical payments — Massachusetts and Pennsylvania. It would not be surprising if employee/employer choice is a much more important leverage point in higher cost states. The conclusions of this study regarding employee choice of new providers and prior providers are particularly salient for California. The results for California provide some evidence suggesting that the costs are higher and recovery no better when workers select providers with whom they have no prior relationship. Therefore, it is possible that the recent legislative changes struck an appropriate balance by significantly expanding the limits on worker choice of provider but retaining an exception where there is a preexisting provider relationship. However, it is important to understand the technical differences between the study’s definition of a prior provider and the new California statutory definition authorizing predesignation of a provider. The recently enacted California statute (Section 4600 of the California Labor Code) provides the following: ■ Unless the employer established a medical provider network, after 30 days from the date the injury is reported the employee may be treated by a physician of his or her own choice. [Sec. 4600(c)] ■ If an employee has notified his or her employer in writing before the date of injury that he or she has a personal physician, the employee shall have the right to be treated by that physician if the employer provides nonoccupational health insurance. [Sec. 4600 (d)(1)] ■ A personal physician shall meet all of the following conditions: ■ The physician is the employee’s regular physician. [Sec. 4600 (d)(2)(a)] 76 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s ■ The physician is the employee’s primary care physician and has previously directed the medical treatment of the employee, and retains the employee’s medical records. [Sec. 4600 (d)(2)(b)] ■ The physician agrees to be predesignated. [Sec. 4600 (d)(2)(c)] ■ The maximum percentage of all employees who are covered under the pre- designation provision at any time in the state is 7 percent. [Sec 4600 (d)(6)]1 Thus, the predesignation exception to the workers’ choice of providers among an employer-designated network of providers bears some resemblance to the “prior provider” category analyzed in this report. However, compared with the California requirements for predesignation, our prior provider category is in some respects broader because it does not require the prior provider to be the personal physician of the worker under a nonoccupational group health insurance plan offered by the employer. Consequently, while suggesting that the reforms in California may have struck an appropriate balance, the evidence does not constitute a direct evaluation of the particular way in which employee choice of providers has been changed under the reforms. One particular point to note for Texas: it was the only state where we found higher costs regardless of whether the employee selected a prior provider or a new provider. It may be that the potential lesson for California about distinguishing between employee choice of prior and new providers is less applicable for Texas. 1 How this provision will be applied, if it is, is unclear to observers with whom we have conferred. In addition, the rule could be read as applying to 7 percent of an employer’s workers. Technical Appendix A: Literature Review In general, studies of workers’ compensation outcomes based on the choice of provider have reached mixed conclusions. Potential explanations for this variation in results include weaknesses that we try to rectify in this report. First, different results are often the result of different data sources covering different periods and states. Moreover, in most studies, the choice of provider was not known with respect to specific claims, and the degree to which the worker might have changed providers or had multiple providers was not considered because of data limitations. Rather, studies prior to ours characterized provider choice for all cases in a state as employer choice, for example, if the state statute gave the employer the right to select the provider — regardless of the actual choice made in each case. Based on average annual changes in medical payments in workers’ compensation cases in 41 states from 1965 to 1985, Boden and Fleischman (1989) found little relationship at the state level between the state’s approach to provider choice and the rate of medical cost growth. During the period in question, eight states changed their laws — two switched to employee choice and six switched to employer choice. Boden and Fleischman did not find evidence that changing the method of choice was correlated with cost changes after the change was made. Nor was there evidence that states that remained employer choice states over the 20 years studied tended to have lower rates of medical cost growth. Subsequently, Victor and Fleischman (1990) employed multivariate analytical methods and concluded that the choice of provider does affect medical payments. Using data from state rating bureaus and state funds (excluding self-insurers), the researchers examined the impact of a change in provider choice in Illinois (after 1975) and Texas (after 1973). They report that medical payments in Illinois rose 8–11 percent following a change to employee choice in the short run, and when the full impact of the change was absorbed, medical payments rose 19–49 percent. In Texas, the short-term effect was 4–6 percent, while the ultimate impact was es- 77 78 t e c h n i c a l a p p e n d i x a timated to be 7–29 percent. In their report, Victor and Fleischman emphasize the tentative nature of these results, partly because they used aggregate (not claimlevel) data and had a small sample, problems that the current study overcomes. In a later paper, Boden (1992) reports that of eight states analyzed, three might have seen costs affected by changes in provider choice approaches, but in the other five states, there was “no evidence that these changes triggered changes in medical payments” (p. 45). This issue is also addressed in a report by Durbin and Appel (1991). After studying average state medical payments between 1965 and 1984 and employing multivariate analysis across the states, they report that states with employer choice had 15 percent lower average medical payments in 1965 and that the difference widened to 36.5 percent in 1984. Their results also suggest that physician choice has a greater impact on medical payments than do fee schedules. The most data-intensive study of the issue of provider choice was conducted by Pozzebon (1994), whose findings differ from those of Durbin and Appel. She relied on data from almost 32,000 closed claims obtained from the National Council on Compensation Insurance, Inc. (NCCI) for 17 states for 1979–1987. Using medical payments per claim as the dependent variable, Pozzebon created four variables as her statutory choice measures: initial choice was limited, changing the provider was limited, no limits were placed on employee initial choice or on subsequent changes, and both initial choice and subsequent choice were limited. She found that where the employee’s initial choice was constrained, “Restrictions on initial choice increase health costs in workers’ compensation programs by 11–16 percent, a large and statistically significant effect” (p. 161). Limits on changing the provider subsequent to the initial choice were also found to be correlated with higher medical payments. However, Pozzebon acknowledges that these findings could result from higher costs leading to policies to limit change, rather than the cost-reducing effects of policies limiting choice. Pozzebon’s somewhat unexpected findings do not seem attributable simply to the source of the data used. In a 1996 study, Durbin, Corro, and Helvacian also used NCCI data and found that employer choice was associated with lower costs of medical benefits. However, the sample in that study was more limited, including 1,300 claims each for four states with 1987 as the injury year and closing dates between 1988 and 1992. A more recent study found that network penetration rates were higher when the state law authorized employer choice of initial provider (Victor, Wang, and Borba, 2002). The authors used estimates from other studies showing that higher network penetration was associated with lower medical payments to compute “a rough estimate” that changing a state’s law from employer to employee control of the decision to change providers increased medical payments by 7–10 percent. technical appendix a 79 The study used data from nine states (including the four states covered in this report) and controlled for worker characteristics, injury and claim type, industry, wage, and the prevalence of networks in the state, to estimate the impact of provider choice laws on network penetration rates. The only study of provider choice in workers’ compensation utilizing rigorous experimental methods compared experimental and control groups, with the former group treated in a managed care framework and the latter group allowed to select their own providers in a traditional fee-for-service arrangement (Washington Department of Labor and Industries and University of Washington Department of Health Services, 1997). Firms, not individual workers, were placed in the experimental or control group. The study tracked 1,354 injury cases with treatment in managed care and 1,708 cases from firms in the control group. For our purposes, this study has one significant drawback — namely, the differences between the groups were more than simply a matter of who selected the provider. Among other differences was the method of payment to the providers for either group. However, the study is also an important extension of earlier studies because the outcomes analyzed were more extensive than simply the medical payments per case. The Washington study found that workers in managed care settings had medical payments that were 27–32 percent below those in the traditional employee choice fee-for-service model. The study also compared rates of injured workers who received “time loss costs,” role-functioning scores (self-reported measures of how well individuals were able to carry out activities related to personal and social roles), and self-reported opinions on the progress of recovery and on overall outcomes. Workers were surveyed both at six weeks and six months after their injury. Workers treated in a managed care setting reported statistically significant lower role-functioning scores at six weeks and at six months. Workers reported significantly lower rates of satisfaction with their treatment, their attending physician, and with their overall access to care at six weeks if they were treated in the managed care group. However, at the six-month interview, statistically significant lower rates of satisfaction were found only with regard to overall access to care. At six weeks and at six months, workers in the managed care group reported less progress on recovery, and the difference is statistically significant. However, at six months, the study found no differences in the two groups with regard to pain, mental health status, and physical functioning. This study points to the multiplicity of outcomes that warrant attention in studies of provider choice. What can we conclude from this review of the literature? First, although most studies appear to conclude that employer choice is associated with lower medical payments in workers’ compensation, the findings are not unchallenged. This 80 t e c h n i c a l a p p e n d i x a should hardly be surprising, because the states and the years selected have varied and the measures of choice have tended to be crude. Very little work related to choice of provider has focused on outcomes or cost measures other than medical payments — such as duration of time out of work, indemnity benefits, physical recovery, and worker satisfaction with care. Rarely have many other factors that likely affect outcomes, such as worker and employer characteristics, been controlled for in these studies. No study appears to have considered and analyzed the significance of whether the injured employee had been treated previously by the provider who gave primary care in the workers’ compensation claim. Finally, studies done even a few years before this study were done when network arrangements were less common. Since employer-selected providers are more likely to participate in such plans now than they did previously, the relevance of some of those earlier studies may have diminished. Technical Appendix B: Variables Used in Study: Definitions and Descriptive Statistics 82 t e c h n i c a l a p p e n d i x b Table B.1 Definitions of Variables Dependent variables Indemnity benefits Medical benefits Substantial return to work Duration of disability Recovery Satisfaction Provider choice Independent variables State Pennsylvania California Texas Massachusetts Employer chose Employee chose Employee chose, prior Employee chose, new The indemnity payment the worker received. The amount the insurer paid for the worker’s medical treatment. A dummy variable. The value is 1 if the worker was able to return to work and stay for one full month. The number of weeks from the time of the injury to the first substantial return to work. Any worker who did not have a substantial return to work was assigned 156 weeks. Worker’s perceived recovery. The difference between SF-12® score in the week after the injury and the score at the time of the interview. An ordinal categorical variable. The question is about the satisfaction level with the medical care the worker received overall: 1 is “very satisfied,” 2 is “somewhat satisfied,” 3 is “somewhat dissatisfied,” 4 is “very dissatisfied.” Two-way or three-way classification. Two-way classification corresponds to employee or employer choice. Three-way classification corresponds to employee choice of prior provider, employee choice of new provider, or employer choice. A dummy variable. The value is 1 if the claim was from Pennsylvania. A dummy variable. The value is 1 if the claim was from California. A dummy variable. The value is 1 if the claim was from Texas. A dummy variable. The value is 1 if the claim was from Massachusetts. A dummy variable. The value is 1 if the employer or the insurance company chose the provider. A dummy variable. The value is 1 if the worker, a family member or friend, or the worker’s attorney chose the provider. A dummy variable. The value is 1 if the worker, a family member or friend, or the worker’s attorney chose the provider and the worker was previously treated by this provider for another medical condition. A dummy variable. The value is 1 if the worker, a family member or friend, or the worker’s attorney chose the provider and the worker was not previously treated by this provider for another medical condition. Table B.1 Definitions of Variables (continued) Worker characteristics Age Male Married Wage Hourly worker Tenure Less than high school Some high school High school graduate Some college College graduate Postgraduate Survey in Spanish Workplace characteristics Firm size ≤ 50 Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Clerical/professional Manufacturing Construction Trade Other industries Worker’s age at the time of the injury. A dummy variable. The value is 1 if the worker is male. A dummy variable. The value is 1 if the worker was married at the time of the injury. Worker’s average weekly wage. The unit is $100; that is, the variable would be 6 if the average weekly wage was $600. A dummy variable. The value is 1 if the worker was paid an hourly wage in the month before the injury. The number of years the worker was employed at the job before the injury. A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is grade school or less (0–8). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is some high school (9–11). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is high school (12). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is some college (1–3 years). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is college graduate (4 years). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is postgraduate (professional, masters, or doctorate). A dummy variable. The value is 1 if the worker preferred to be interviewed in Spanish. A dummy variable. The value is 1 if the number of people who were employed at the location where the worker was working at the time of the injury was between 1 and 50. A dummy variable. The value is 1 if the number of people who were employed at the location where the worker was working at the time of the injury was between 51 and 250. A dummy variable. The value is 1 if the number of people who were employed at the location where the worker was working at the time of the injury was between 251 and 1,000. A dummy variable. The value is 1 if the number of people who were employed at the location where the worker was working at the time of the injury was more than 1,000. A dummy variable. The value is 1 if the worker worked in high-risk service industries at the time of the injury. A dummy variable. The value is 1 if the worker worked in low-risk service industries at the time of the injury. A dummy variable. The value is 1 if the worker worked in the clerical/professional occupations in any industry at the time of the injury. A dummy variable. The value is 1 if the worker worked in the manufacturing industry at the time of the injury. A dummy variable. The value is 1 if the worker worked in the construction industry at the time of the injury. A dummy variable. The value is 1 if the worker worked in the trade industry at the time of the injury. A dummy variable. The value is 1 if the worker worked in other industries at the time of the injury. technical appendix b 83 continued 84 t e c h n i c a l a p p e n d i x b Table B.1 Definitions of Variables (continued) Injury characteristics Back pain Nonback sprain or strain Fracture Inflammation, laceration, or contusion Other injuries Severity Treatment characteristics Overnight hospitalization Major surgery Attorney involvement A dummy variable. The value is 1 if the type of injury is back pain. A dummy variable. The value is 1 if the type of injury is nonback sprain or strain. A dummy variable. The value is 1 if the type of injury is fracture. A dummy variable. The value is 1 if the type of injury is inflammation, laceration, or contusion. A dummy variable. The value is 1 if the type of injury is not one of the above. Worker’s perceived severity. The difference between SF-12 score during the four weeks before the injury and the score during the week after the injury. A dummy variable. The value is 1 if the worker received “room and board” or “intensive care” based on the revenue code. A dummy variable. The value is 1 if the total payment for significant surgical services was greater than 0. A dummy variable. The value is 1 if the worker was represented by a lawyer when trying to collect workers’ compensation. Table B.2 Descriptive Statistics Combined Employee chose Employer chose Prior New California Employee chose Employer chose Prior New Texas Employee Employer chose chose Prior New Massachusetts Employee Employer chose chose Prior New Pennsylvania Employee Employer chose chose Prior New technical appendix b Claims in each state (%) Pennsylvania California Texas Massachusetts Worker characteristics Age (mean) Male (%) Married (%) Weekly wage (mean) Hourly worker (%) Tenure (mean, years) Less than high school (%) Some high school (%) High school graduate (%) Some college (%) College graduate (%) Postgraduate (%) Survey in Spanish (%) 20.9 23.8 24.2 31.1 43.6 59.5 69.7 $632 81.4 9.9 1.6 8.7 44.4 30.3 9.5 5.5 1.3 21.2 19.7 35.4 23.7 42.1 67.5 67.0 $609 83.5 8.8 3.2 9.3 44.7 28.7 10.6 3.5 3.6 37.5 34.1 17.1 11.3 41.5 62.5 65.6 $612 92.7 8.7 5.6 11.0 41.9 29.0 9.7 2.8 7.1 —— —— —— —— — — — — 44.3 57.2 72.9 $659 78.0 12.2 1.9 9.2 30.0 37.8 13.0 8.1 1.7 43.0 61.5 64.2 $646 79.4 9.9 3.5 6.4 29.6 33.6 14.5 12.3 5.6 40.5 55.0 57.4 $601 89.7 7.2 9.7 10.9 31.0 31.6 12.3 4.6 11.5 —— —— —— —— — — — — 41.7 57.1 65.7 $578 85.2 7.1 1.0 5.9 43.0 34.4 9.0 6.8 3.5 42.0 65.6 70.5 $544 81.8 7.5 4.2 12.4 40.9 31.8 10.1 0.6 6.9 39.5 67.1 70.4 $479 90.7 6.3 10.2 18.7 36.8 27.2 5.5 1.6 17.1 —— —— —— —— — — — — 43.8 63.3 68.8 $666 77.0 9.8 2.6 9.9 47.0 28.6 8.7 3.1 0.0 40.9 70.6 63.7 $706 89.1 8.3 2.4 7.4 44.5 33.7 8.7 3.3 0.0 42.4 74.0 68.8 $714 95.7 8.4 3.0 9.0 44.4 31.0 10.7 1.9 2.0 —— —— —— —— — — — — 44.6 59.1 71.8 $614 87.7 10.7 0.4 9.5 59.0 19.2 7.1 4.7 0.0 43.0 72.8 67.5 $573 84.0 10.3 2.3 8.9 65.4 13.4 10.0 0.0 0.0 43.1 63.7 70.0 $652 95.6 11.2 0.8 8.4 53.4 26.8 8.8 1.8 0.0 continued 85 86 t e c h n i c a l a p p e n d i x b Table B.2 Descriptive Statistics (continued) Combined California Employee chose Employer chose Employee chose Employer chose Prior New Prior New Texas Employee Employer chose chose Prior New Massachusetts Employee Employer chose chose Prior New Pennsylvania Employee Employer chose chose Prior New Workplace characteristics (%) Firm size≤50 48.5 Firm size 51–250 27.1 Firm size 251–1,000 13.4 Firm size Ͼ 1,000 11.0 High-risk services 24.1 Low-risk services 13.1 Clerical/professional services 12.8 Manufacturing 19.6 Construction 7.1 Trade 18.3 Other industries 5.1 Injury characteristics Back pain (%) 34.6 Nonback sprain or strain (%) 34.9 Fracture (%) 8.6 Inflammation, laceration, or contusion (%) 5.1 Other injuries (%) 16.8 Severity (mean)a 28.5 Treatment characteristics (%) Overnight hospitalization 9.7 Major surgery 37.8 Attorney involvement 24.1 52.6 26.2 15.5 5.8 26.1 11.6 7.6 20.6 11.3 17.9 4.9 38.1 33.1 9.0 5.7 14.1 29.8 8.6 34.8 23.6 47.2 29.5 16.0 7.2 27.6 8.3 6.3 28.8 8.4 14.7 5.9 30.0 39.7 8.6 8.5 13.1 28.7 7.7 31.3 18.5 50.7 60.7 30.0 25.0 9.6 11.8 9.8 2.5 25.7 24.7 11.5 16.2 10.7 12.0 11.4 12.1 8.3 11.5 26.9 17.5 5.5 5.9 53.1 27.0 12.8 7.1 27.8 8.8 9.7 24.6 6.0 17.8 5.3 40.4 39.0 32.4 37.8 7.1 7.1 4.9 6.4 15.3 9.7 27.6 28.7 28.3 44.7 7.3 9.2 10.5 29.7 3.0 9.9 31.8 35.3 36.5 42.1 5.2 26.9 25.6 50.1 53.6 20.5 24.7 16.1 15.7 13.3 6.1 31.0 27.0 17.5 10.5 19.3 10.2 7.8 21.3 4.2 7.0 15.9 17.8 4.4 6.1 49.6 28.1 16.0 6.3 25.7 8.2 5.7 34.4 8.7 14.6 2.6 38.1 41.3 29.1 34.7 11.8 8.7 7.7 3.6 13.3 11.6 29.7 29.9 35.1 28.5 10.3 10.1 16.1 25.0 13.5 8.6 39.5 31.8 8.6 16.3 11.5 31.7 12.3 51.8 50.7 28.0 27.2 11.1 14.5 9.1 7.6 20.6 24.1 12.1 12.7 9.5 4.8 25.6 21.3 10.2 17.1 16.8 16.9 5.1 3.2 42.8 36.0 15.0 6.2 35.0 7.5 1.7 22.9 17.2 12.9 2.8 34.7 39.2 33.9 28.1 8.5 9.0 4.8 7.9 18.1 15.8 28.9 32.0 36.7 31.9 10.4 5.4 15.6 28.5 5.9 4.8 35.3 30.5 31.0 24.6 6.2 22.7 21.9 39.7 45.5 30.1 28.6 17.8 19.7 12.4 6.2 18.7 28.0 10.1 8.5 11.2 2.2 34.2 25.7 4.8 12.3 15.6 19.6 5.4 3.8 42.0 30.6 19.4 8.1 26.1 8.1 5.1 31.2 7.6 12.7 9.1 23.8 31.0 45.8 31.5 6.8 11.4 2.8 5.9 20.7 20.1 27.8 28.3 27.3 42.8 8.4 8.1 13.4 29.6 16.4 12.2 45.6 44.0 17.6 17.2 8.5 37.1 13.8 Table B.2 Descriptive Statistics (continued) Combined Employee chose Employer chose California Employee chose Employer chose Prior New Prior New Texas Employee Employer chose chose Prior New Massachusetts Employee Employer chose chose Prior New Pennsylvania Employee Employer chose chose Prior New Satisfaction with overall care Very satisfied (%) 62.9 Somewhat satisfied (%) 22.4 Somewhat dissatisfied (%) 7.9 Very dissatisfied (%) 6.7 N 458 55.5 26.5 8.4 9.6 597 45.3 34.8 8.7 11.2 896 54.2 54.6 27.8 30.3 10.2 7.8 7.8 7.3 106 122 41.2 36.7 10.5 11.7 306 66.2 50.5 20.1 24.2 4.9 11.4 8.8 13.8 114 213 39.5 42.8 9.5 8.2 152 63.4 57.6 23.0 28.1 6.9 6.8 6.7 7.5 144 133 46.2 36.8 5.7 11.3 98 68.5 62.4 18.0 25.2 10.3 5.5 3.1 6.9 94 129 51.3 29.0 7.7 12.0 340 technical appendix b a Based on respondents’ SF-12® scores, which are scaled from 0 to 100, where 100 is the best health. The severity score is the difference between the score for the time four weeks before the injury occurred and the score one week after the injury. 87 Technical Appendix C: Discussion of Construction and Validity of Health Status, Recovery, and Perceived Severity Measures This appendix summarizes material from an earlier WCRI report (Victor, Barth, and Liu, 2003), which describes and analyzes the survey data used in the study. Here we provide evidence on the construction and validity of two important measures used in this study: perceived recovery of physical health and perceived injury severity. Both are derived from the SF-12® — the most widely used instrument for measuring health status. The validity of the measures was assessed based on internal consistency, consistency across states, consistency with other published studies, and plausible correlations with other measures, especially other markers of injury severity. In general, the results indicate that the measures are consistent with other studies and exhibit the patterns expected of valid measures. To construct the measures of health status, the survey asked workers about the following:1 ■ General health: “In general, would you say that your health was excellent, very good, good, fair, or poor in the four weeks before your injury?” 1 There are some differences in the wording used in the SF-12® and the WCRI questions that seek to use the SF-12®. These wording differences may lead to some differences in how the questions are answered. However, it is unlikely that these specific differences would have a major effect on the results. 89 90 t e c h n i c a l a p p e n d i x c ■ Limits on activities: “During a typical day in the four weeks before your injury, how limited were you in performing moderate activities such as moving a table, pushing a vacuum cleaner, bowling, or playing golf? Were you limited a lot, limited a little, or not limited at all?” ■ Limits on climbing stairs: “During a typical day in the four weeks before your injury, how limited were you in climbing several flights of stairs? Were you limited a lot, limited a little, or not limited at all?” ■ Amount accomplished: “During the four weeks before your injury, did you accomplish less than you would like with your work or other regular daily activities as a result of your physical health?” (Note that daily activities include activities outside work.) ■ Physical limitations: “During the four weeks before your injury, were you limited in the kind of work or other regular daily activities you did as a result of your physical health?” ■ Pain: “During the four weeks before your injury, how much did pain in general interfere with your normal work, including both work outside the home and housework? Would you say not at all, a little bit, a moderate amount, quite a bit, or an extreme amount?” Workers’ responses to questions regarding these measures were combined into a single scaled score ranging from zero to 100 (a higher score reflects better physical health and functioning), based on the SF-12® survey questions and scoring methodology.2 In the interview (conducted 3.0 to 3.5 years postinjury), each worker was asked to recall and report his or her health and functioning status regarding each of the previously listed dimensions from the perspective of three points in time: before 2 The methodology for scoring the SF-12® is widely accepted and well documented. A description is found in Ware, Turner-Bowker, Kosinski, and Gandek (2002). Survey questions related to mental and emotional health and functioning were treated differently, as discussed in Victor, Barth, and Liu (2003). In particular, the mental health questions were only asked at the time of the interview, in part because of special concerns about the difficulty of recalling mental health status, and also because asking all 12 questions about three periods would have significantly lengthened the time needed to conduct the entire survey. This poses a problem because the overall scores for physical health and functioning based on the SF-12® require information on mental health. The assumption was made that the mental health scores at the time of the interview prevailed at the other times (pre- and immediately postinjury). The study cited above explored the sensitivity of the SF-12® scores to using different extreme (best-case and worst-case) assumptions about mental health at these two times and found the physical health and functioning scores were very insensitive to the mental health responses. Thus, any bias from not having information on mental health pre- and postinjury should be trivial. technical appendix c 91 the injury (preinjury), just after the injury (postinjury), and at the time of interview. This information was used to create two measures: ■ Perceived recovery of physical health and functioning: measured by calculating the difference between the composite health and functioning status score at interview and the score based on health and functioning postinjury. ■ Perceived injury severity: measured by calculating the difference between the postinjury and preinjury composite health and functioning status scores. All empirical measures of severity and recovery have some inherent limitations, and these are no exception. First, it is important to emphasize that these measures are based on workers’ self-reports and may be different from health status measures based on clinical findings. Second, the reader should keep in mind that the health status reported “postinjury” is the worker’s recollection about his or her health status one week after the injury. The perceived severity measure focuses on the short-term physical consequences of the injury. Most conditions improve over time, but some may worsen before they improve. For cases in which health status worsened after the first week, both perceived severity and recovery may be understated. Third, the at-interview health status reported by some workers may have been affected by nonwork-related events — for example, automobile or other accidents or transitory or chronic diseases — that occurred or worsened after the work injury and were unrelated to the injury. In such cases, as well, perceived recovery is understated. Other workers in the study may have experienced such events before injury, thus affecting their preinjury health scores, and some may have improved after injury (along dimensions unrelated to their injuries). When this has occurred, perceived injury severity is understated and perceived recovery may be overstated. However, there is no obvious reason that these sources of mismeasurement of severity or recovery should be related to provider choice, suggesting that they do not influence our key findings. How Plausible Are the Survey Results on Health and Functioning? Because the recall concern is potentially important, a variety of analyses were used to assess how plausible these health and functioning results might be. Other than the evidence on validity that is presented in the earlier WCRI study, there is no di- 92 t e c h n i c a l a p p e n d i x c rectly relevant evidence from prior studies about the validity or invalidity of the approach used in the WCRI survey. No other study has assessed the validity of the retrospective use of the SF-12®, as done by WCRI (telephone conversation with Mark Kosinski, Senior Scientist, QualityMetric, Inc., August 31, 2005). In a study by Ware, et al. (1996) that used retrospective recall of health status at one, two and four years, the authors concluded that “Retrospective evaluations of prior health status and self-reported evaluations of changes over time may yield useful information about outcomes when baseline assessments are not available.” The study examined 1,466 patients with chronic health conditions, but not traumatic (as in workers’ compensation). Although they found some mismeasurement in the retrospective self-reported measures, they found a high correlation between the retrospective self-reported changes and measured changes of health status. They also found that the correlation was similar regardless of the recall period. The retrospectively reported changes were influenced by the patient’s health status at the time of interview, biasing the magnitudes of the retrospectively reported changes. A number of the workers’ compensation studies cited in the literature review in Chapter 2 relied on retrospective recall of the SF-12® measures (Research and Oversight Council on Workers’ Compensation (ROC) and Med-FX, LLC, 2001). The only study that assessed the validity of retrospective recall used a somewhat different recall approach. That study asked patients to recall retrospectively the change in their health as indicated by several of the SF-12® questions, and found that this approach appeared to be valid (Damiano, Pastores, and Ware, 1998). These analyses included: ■ Internal consistency ■ Consistency across states ■ Consistency with other studies ■ Plausible correlations with other measures The perceived physical health and functioning measures showed plausible patterns and reconciled well with values reported in other studies. The average preinjury scaled scores for physical health and functioning in the four states (54– 55) were remarkably consistent across the states and higher than average for the general U.S. population (50).3 One would expect the scores for an employed pop- 3 See Ware, Keller, and Kosinski (1998). technical appendix c 93 SF-12® Score Figure C.1 Perceived Injury Severity and Recovery of Physical Health and Functioning: An Example from the Texas Results 60 54 50 Severity 50 41 40 Recovery 30 27 20 10 0 Preinjury Postinjury At interview U.S. population Notes: All workers surveyed experienced more than 7 days of lost time. SF-12® scores range from 0 to 100. A higher score indicates better health. SF-12® is a registered trademark of the Medical Outcomes Trust. Source of average figures for U.S. population: Ware, Keller, and Kosinski (1998). ulation to be higher than that for the general population. Further, the scores were similar to those found in a special study of a healthy population (Airey et al., 1999, Table 3.12). In all four states, the average postinjury scores were lower than the preinjury scores (reflecting injury severity), and the average at-interview scores were higher than the postinjury scores (reflecting perceived recovery); only in a handful of cases did severity or recovery have the unexpected sign. In addition, the greater the perceived injury severity, the higher the medical payments paid or expected to be paid for the claim. Figure C.1 shows the SF-12® scores for workers’ perceived physical health and functioning for Texas. The pattern is typical of the four states. The average SF-12® score reported for Texas workers at interview was 41 points, lower than the average of 50 for the U.S. population. This is as expected because a population of injured workers should have lower average scores than the general population. In addition, the Texas score compares well with a study sponsored by the Research and Oversight Council on Workers’ Compensation (ROC) and Med-FX, LLC (2001), which found a physical health and functioning score of 38 points for a 94 t e c h n i c a l a p p e n d i x c sample of workers injured in 1998 when they were surveyed in 2000. That study found a score of 42 for workers from four other states who were similarly surveyed. The mental health and functioning score resulting from our survey (48) also compares well with that from the ROC survey (44). A later ROC survey of workers with back, neck, or shoulder injuries who were interviewed 21 to 33 months after their injuries found a score of 39 points — quite similar to the Texas score in the current study (Shields, Baroni, and Lu, 2003). Within each state, one would expect that workers reporting more-severe injuries would receive more medical care and that the care would be more expensive. Thus, one would expect to see higher medical payments for workers who reported more-severe injuries (that is, workers with larger reductions in their SF-12® perceived physical health and functioning scores from pre- to postinjury). To examine this, respondents were divided into three perceived injury severity groups based on the differences in their pre- and postinjury SF-12® scores:4 ■ Least severe injuries: postinjury scores minus preinjury scores equals –5.0 to –19.9 points. ■ Moderately severe injuries: postinjury scores minus preinjury scores equals –20.0 to –34.9 points. ■ Most severe injuries: postinjury scores minus preinjury scores equals –35.0 to –45.0 points. Table C.1 shows that medical payments were strongly correlated with respondents’ perceived injury severity. For example, workers who perceived the most severe injuries had incurred medical payments that were about double those of the least severe injury group. As previously discussed, this result is what one would expect and adds to the plausibility of the survey measures and results. To further validate the survey measures, the results obtained from one of the SF-12® questions regarding general health can be compared with the results from a similar question in another national survey — the Behavioral Risk Factor Sur- 4 The authors urge caution regarding this nomenclature. The categories of least severe, moderately severe, and most severe injuries are derived from the difference between the worker’s SF-12® scores from before the injury to just after the injury. On the other hand, the categories of less serious and more serious claims, as discussed later, are financial classifications based on data from the workers’ compensation claim and were used to select and stratify our sample. technical appendix c 95 Table C.1 Incurred Costs for Medical Care, by Perceived Injury Severity California Texas Massachusetts Mean incurred costs for medical care Least severe $13,774 Moderately severe $15,550 Most severe $25,754 Median incurred costs for medical care Least severe $5,381 Moderately severe $5,178 Most severe $10,372 $9,266 $12,236 $17,407 $3,459 $6,100 $8,851 $3,231 $6,767 $9,552 $1,425 $1,606 $3,063 Pennsylvania $6,163 $8,145 $11,960 $3,091 $3,676 $4,154 Notes: All workers surveyed experienced more than 7 days of lost time. Average incurred medical payments per claim are for 1998 injuries for Texas and for 1999 injuries for other states. All costs reflect an average of 36 months’ experience after the injury. Key: Least severe means reductions in SF-12® scores of 5.0–19.9 points; moderately severe, reductions of 20.0–34.9 points; and most severe, reductions of 35.0–45.0 points. veillance System (BRFSS) survey for 2000 (Centers for Disease Control and Prevention, 2002).5 These comparisons are reported in Table C.2. In all four states, the health status reported in the BRFSS is systematically lower than the preinjury health status of the injured workers in the WCRI survey. This difference may reflect the nature of either the survey instrument or the populations surveyed. As indicated previously, it is expected that a survey of an employed population would yield higher scores than a survey of the general population. This is reinforced by how similar our reported SF-12® scores are to those found by the ROC study of Texas (ROC and Med-FX, 2001). However, it is possible that the differences reflect recall biases in using the question retrospectively. Notwithstanding the systematic differences discussed here, it is reassuring that both surveys found that Massachusetts is the state where people report the best health and that Texas is the state where workers report the worst health, among these four states. 5 Our survey asked, “In general, would you say that your health is excellent, very good, good, fair, or poor?” The BRFSS question asked of 5,002 adults in Texas and nationwide, “How is your general health: excellent, very good, good, fair, or poor?” 96 t e c h n i c a l a p p e n d i x c Table C.2 Comparison of Two Surveys on General Health SelfReported General Health California WCRI Survey, Injured Workers, Preinjury (percent) BRFSS, General State Population (percent) Texas WCRI Survey, Injured Workers, Preinjury (percent) BRFSS, General State Population (percent) Excellent 38 24 38 20 Very good 36 32 33 28 Good 22 28 21 33 Fair 3 12 6 15 Poor 1424 Massachusetts WCRI Survey, Injured Workers, Preinjury (percent) BRFSS, General State Population (percent) 43 28 35 34 19 26 39 03 Pennsylvania WCRI Survey, Injured Workers, Preinjury (percent) BRFSS, General State Population (percent) 40 21 35 35 22 30 2 10 14 Key: BRFSS: Behavioral Risk Factor Surveillance System. Sources: WCRI DBE database; Centers for Disease Control and Prevention (2002). Technical Appendix D: Discussion of Survey Response Rates and Response Bias After a very brief description of survey methods, this appendix focuses on response rates and the representativeness of the respondents. A more complete description of the survey and an evaluation of it can be found in Victor, Barth, and Liu (2003). WCRI developed the survey in collaboration with the University of Connecticut’s Center for Survey Research and Analysis (CSRA), which conducted the survey. The survey instrument was first fielded in Texas in 2002. A year later, it was fielded in California, Massachusetts, and Pennsylvania. For the latter surveys, some questions were modified slightly and some skip patterns altered, in light of the experience gained in the Texas survey effort. CSRA conducted telephone interviews of between 629 and 754 workers in each state who were injured during portions of 1998 in Texas and during 1999 in the other states, had more than seven days of lost work time, and received workers’ compensation income benefits. The surveys were conducted in 2002 and 2003 — on average about 3.0 to 3.5 years after the injury. We selected this time lag because we wanted to know the intermediate-term consequences of injury — in particular, the recovery of health and functioning and return to work. We interviewed fewer people in Massachusetts than in the other states because the refusal rates were higher in Massachusetts. The telephone surveys were conducted seven days a week, in both day and evening hours. The survey instrument was translated into Spanish and administered in that language when requested by the worker. In Massachusetts, Pennsylvania, and Texas, state agencies provided the workers’ names and contact information we needed to draw the sample. In California, the state agency requested that WCRI use data from insurers and employers to draw the sample. All workers were assured in writing and on the telephone that 97 98 t e c h n i c a l a p p e n d i x d their responses were confidential and would not be shared with the state agencies, their employers, or the insurers. To increase the statistical power of the survey analysis, we used a random sample stratified along two dimensions: the financial seriousness of the claim and the insurance market segment.1 This improves the statistical power since self-insurance is less common than insurance arrangements, and more-serious claims are less common than less serious ones. Representativeness and Response Bias Table D.1 shows the results of the telephone survey calls. The surveyors began with a sample of workers, of which 53–62 percent could be located. For the others, either the telephone number supplied did not work or the person answering the call claimed not to know of, or know the whereabouts of, the person to be interviewed. These rates of invalid contacts were in line with those found in other studies that interviewed injured workers. In our case, 38–47 percent of workers sampled had invalid phone numbers 3.0 to 3.5 years after the injury. In a related study, surveyors found invalid telephone numbers for 28 percent of injured workers who belonged to the same craft labor union, where injuries occurred in 1995 and 1996 and the survey was conducted between November 1997 and October 1999 (Borba and Parry, 2000). An April 1998 survey of Texas workers injured in 1996 yielded 39 percent disconnected or wrong phone numbers (Research and Oversight Council on Workers’ Compensation, 1998). Clearly, the longer the period from the injury date to the survey, the harder it is to track down respondents. Not surprisingly, workers for whom the phone numbers were invalid had personal or claim characteristics that suggested they had less severe injuries and were more mobile than respondents. Workers with invalid phone numbers had medical payments that were 11–31 percent lower than those of respondents, and they were more likely to have cuts and bruises (less severe injuries), less likely to have fractures or nonback sprains (more-severe injuries), and less likely to have surgery. Workers with invalid phone numbers also had shorter durations of temporary disability and lower indemnity benefits, on average, and their claims were less likely to be open. Workers with invalid phone numbers also appeared to be more 1 “Insurance market segment” refers to the three sources of financing workers’ compensation liability: private insurance carriers, the state fund, and self-insurance. technical appendix d 99 Table D.1 Attempted Telephone Interviews and Valid Phone Numbers Type of Disposition California Texas Massachusetts Pennsylvania Total number sampled Percentage with valid phone numbers 4,110 53 2,678 55 3,114 53 3,384 62 mobile than respondents because they were more likely to be single, male, and younger. Further, workers with invalid phone numbers had much lower preinjury wages and less tenure on the job. Clearly, workers with invalid phone numbers had some differences relative to both respondents and refusers. However, these differences were not large. Of the remaining workers for whom valid phone numbers existed, the surveyors found and completed interviews with 34–51 percent. This represents a response rate of 51–61 percent in Texas, 33–48 percent in California, 37–46 percent in Massachusetts, and 36–45 percent in Pennsylvania, depending on the method used to compute the response rate (see Table D.2).2 Given the number of invalid phone numbers and the number of workers with valid phone numbers who refused to participate or who could not be contacted, three questions arise about representativeness and response bias, which we answer in the next section: 1. How do the respondents compare with all workers with claims paid in their respective workers’ compensation systems (with more than seven days of lost time)? 2. How do the respondents compare with refusals? 3. How do the respondents compare with those who could not be contacted because they did not have valid phone numbers? 2 The calculation of the response rate depends on how one chooses to interpret the validity of the live telephone numbers. In Texas, for example, if one assumes that all the live telephone numbers were valid, the response rate was 50.8 percent. If one assumes that the busy signals, no answers, and others were invalid at the same rate as the numbers overall, the response rate was 55.0 percent. If one assumes that the busy signals, no answers, and others were all invalid numbers, the response rate was 61.1 percent. Considering that more than 25 attempts were made for each phone number, the 61.1 percent response rate would not be unreasonable, although a more conservative choice is the 55.0 percent response rate. 100 t e c h n i c a l a p p e n d i x d Table D.2 Disposition of Cases with Valid Phone Numbers Type of Disposition Percentage of Cases with Valid Phone Numbers California Texas Massachusetts Pennsylvania Completed surveys Other valid phone numbers Refused Answering machine Busy signal No answer Othera 34 66 21 14 1 5 25 51 49 20 12 1 12 3 38 62 36 7 0 5 13 37 63 27 15 1 6 13 a The two biggest contributors to the “Other” category were persons who spoke foreign languages (other than Spanish) and persons who were called and asked us to call back at another time. Before we attempted to call people in this category a second time, we had reached the desired number of completed interviews. representativeness Comparing the characteristics of the respondents and their claims with the characteristics of workers in each state, the evidence indicates that the respondents were reasonably representative of workers with workers’ compensation claims in each state. For example, the average medical cost per claim for respondents was very similar to that for all workers with more than seven days of lost time in each state system. The average medical cost per claim among respondents was 2–16 percent different than medical payments among all workers with claims in the four states.3 Table D.3 shows that the survey respondents were reasonably representative of the population of injured workers whose claims were paid. It shows the similari- 3 The average medical payment per claim for each state is taken from CompScopeTM Benchmarks: Multistate Comparisons, 4th Edition, which is based on large samples of payor claim data — 33–67 percent of the claims in each state. These data come from a diverse group of payors that represent the various insurance market segments in each state. The costs measures are externally validated again data from the states and rating bureaus, and WCRI reports that they reconcile within 10 percent of the external figure (Telles, Wang, and Tanabe, 2004). Table D.3 Analysis of Representativeness California Average Respondents for Statea Texas Average Respondents for Statea Massachusetts Average Respondents for Statea Pennsylvania Average Respondents for Statea technical appendix d Worker characteristics Age (mean years) Female (percentage of claims) Single (percentage of claims) Tenure with employer (mean years) Weekly wage (mean) Industry (percentage of cases) Manufacturing Construction Clerical/professional Trade High-risk services Low-risk services Other Type of injury (percentage of cases) Back sprains and strains Fractures Inflammations, lacerations, contusions Nonback sprains and strains Other 40 41 51 6 $504 15 7 10 14 18 9 29 31 7 10 31 22 42 44 37 8 $599 15 7 9 13 19 9 28 30 10 9 36 15 39 36 39 5 $450 19 8 8 15 21 8 21 31 9 11 27 23 41 37 32 6 $519 23 7 9 14 24 12 12 32 13 8 28 20 39 27 51 5 $541 16 13 5 16 21 8 21 30 9 15 25 22 42 28 34 7 $690 18 15 5 12 21 9 20 32 13 9 29 17 40 32 44 7 $520 22 6 4 10 23 8 27 27 11 12 30 21 43 34 32 10 $624 25 7 4 13 21 7 23 26 11 8 38 17 continued 101 102 t e c h n i c a l a p p e n d i x d Table D.3 Analysis of Representativeness (continued) California Average Respondents for Statea Texas Average Respondents for Statea Massachusetts Average Respondents for Statea Pennsylvania Average Respondents for Statea Claim costs and characteristics Medical payment (mean) $10,506 $11,144 Indemnity payment (mean) $14,171 $15,553 Open claims (percentage of claims) 39 37 PPD or lump-sum payment (percentage of claims) 53 50 Lump-sum payment (percentage of claims) 22 18 Defense attorney involved (percentage of claims) 29 22 Vocational rehabilitation services (percentage of claims) 31 34 PPD or lump-sum payment (mean) $11,924 $13,400 Lump-sum payment (mean) $13,218 $15,897 Duration of temporary disability (mean weeks) 29 25 Type of medical treatment received (percentage of claims) Major surgery 20 26 Chiropractic care 12 12 $11,617 $9,523 26 55 10 9 6 $6,165 $5,181 24 29 22 $11,341 $9,385 26 58 6 4 3 $5,969 $3,955 20 31 20 $4,937 $10,791 13 23 19 17 5 $14,008 $16,482 25 18 9 $5,717 $13,156 15 17 15 19 7 $19,036 $21,264 23 27 9 $7,978 $10,735 16 17 15 19 6 $21,921 $23,410 22 26 6 $7,755 $11,181 13 9 9 15 7 $32,789 $33,182 21 36 6 Note: All values in table for claims with more than 7 days of lost time. a The “Average for State” values are from the WCRI Detailed Benchmark/Evaluation (DBE) database; each value is weighted to represent the claims in the system of that state. Key: PPD: permanent partial disability. technical appendix d 103 ties and differences for characteristics of workers, their injuries, and their claims. For example, the respondents and the injured-worker population within each state were of similar age and gender, had similar industry and injury mixes (with a few moderate exceptions), had similar medical payments and somewhat higher indemnity benefits, and had similar rates of chiropractic care. The main differences were that the respondents tended to have characteristics associated with being less difficult to find or more willing to talk about their injuries because they considered their injuries more significant than did nonrespondents. The average respondent was more likely to be married, have longer job tenure, earn a higher preinjury wage, be in a low-risk service industry, and have a fracture or nonback sprain (that is, a more severe injury), rather than a cut or bruise (that is, a less severe injury). Respondents were slightly less likely to have defense attorneys involved and to have shorter durations of temporary disability. In the two wage-loss states (Massachusetts and Pennsylvania), respondents were also less likely to have a permanent partial disability (PPD) or lump-sum payment resolve the case, but the PPD or lump-sum payment was likely to be larger, indicating a more serious case.4 None of these differences was large enough to engender concern about the representativeness of the respondent data. The four states are much more similar in their industry mix among injured workers than they are for the state economies as a whole. Of course, most of these measures are explanatory variables in our models, so we control for any differences among claims and states. refusals and response bias Workers who refused to be interviewed appear to have had less severe injuries than did respondents. One way to analyze this is to compare the average medical cost per claim of refusals with that of respondents. Medical payments reflect differences in the severity of the injury, the nature of providers used, and the attributes of the workers that influence the demand for medical care, among a variety of other factors. Refusals had medical payments that were 15–28 percent lower than those of respondents in three states, and only 3 percent lower in Pennsylvania. It is not surprising to find that, on average, workers who were unwilling to spend time on interviews had injuries that were less severe and thus less important to them. Table D.4 reinforces this conclusion. Compared with respondents, refusals had 4 For a definition and examples of wage-loss systems, see Barth and Niss (1999). 104 t e c h n i c a l a p p e n d i x d similar marital status, wages, and industry mix. However, there were a number of indicators that refusals were more likely to have relatively less serious claims (that is, lower indemnity benefits), either similar or lower durations of disability and a lower fraction of cases with PPD payments, more claims with cuts and bruises (less serious claims), and less frequent surgery. There was some concern, before conducting the survey, that workers still being represented by attorneys at the time of the interviews would be more likely to refuse to answer our questions, on advice of counsel. Although this could not be tested directly, Table D.3 shows that cases in three of the four states were slightly less likely to involve a defense attorney than the estimate for the state as a whole, which is consistent with the concern. However, Table D.4 shows that cases with defense attorneys involved were slightly more likely to refuse in California and Pennsylvania and slightly less likely to refuse in Massachusetts and Texas.5 weighting the responses The data come from payors from three different market segments — the voluntary insurance market, the residual market (market of last resort) or state fund, and the self-insured market. In addition, claims were divided into two levels of financial seriousness, and the more serious claims were oversampled, because they are relatively rare. Consequently, in all of our analyses, weights are applied to data from each of the six strata based on market segment and financial seriousness to make the claims representative of claims in each state. However, in our regression analyses, we do not weight the data by state to make the sample representative of claims in the four states. Doing so would, naturally, apply much higher weights to observations from California and Texas. In the regression analysis, this latter type of weighting would only matter if parameters differ across states, which is something we independently investigate in a number of ways. Overall, the results suggest that by not weighting across states we, if anything, understate the strength of our conclusions, because the results on the effects of provider choice are strongest for California and Texas, the two largest states. 5 We use defense attorney involvement as a proxy for worker attorney involvement, assuming that a case with a defense attorney involved is more likely to be one in which the worker has retained counsel. Table D.4 Analysis of Response Bias California Texas Massachusetts Pennsylvania Respondents Invalid Refusals Respondents Invalid Refusals Respondents Invalid Refusals Respondents Invalid Refusals Phone Phone Phone Phone Numbers Numbers Numbers Numbers technical appendix d Worker characteristics Age (mean years) 42 Female (percentage of claims) 44 Single (percentage of claims) 37 Tenure with employer (mean years) 8 Weekly wage (mean) $599 Industry (percentage of cases) Manufacturing 15 Construction 7 Clerical/ professional 9 Trade 13 High-risk services 19 Low-risk services 9 Other 28 Type of injury (percentage of cases) Back sprains and strains 30 Fractures 10 Inflammations, lacerations, contusions 9 Nonback sprains and strains 36 Other 15 37 41 37 40 47 41 5 $509 15 9 7 14 21 7 28 7 $619 11 8 9 14 23 7 28 34 30 97 11 8 30 37 15 17 41 37 32 6 $519 23 7 9 14 24 12 12 32 13 8 28 20 36 40 31 36 47 36 3 $427 23 9 6 16 27 9 10 6 $527 23 7 10 13 24 13 9 33 35 10 12 14 12 25 22 19 20 42 28 34 7 $690 18 15 5 12 21 9 20 32 13 9 29 17 36 41 23 22 54 34 4 $568 19 12 3 15 24 7 20 7 $709 19 13 4 13 25 7 19 43 34 32 10 $624 25 7 4 13 21 7 23 35 33 10 10 12 12 29 32 14 14 26 11 8 38 17 36 43 30 28 52 30 5 $514 22 7 3 13 28 7 21 9 $619 22 8 6 11 27 5 22 33 31 12 10 12 11 31 33 13 15 105 continued 106 t e c h n i c a l a p p e n d i x d Table D.4 Analysis of Response Bias (continued) California Texas Massachusetts Pennsylvania Respondents Invalid Refusals Respondents Invalid Refusals Respondents Invalid Refusals Respondents Invalid Refusals Phone Phone Phone Phone Numbers Numbers Numbers Numbers Claim costs and characteristics Medical payment (mean) $11,144 $9,414 $9,421 Indemnity payment $15,553 $14,289 $14,236 Open claims (percentage of claims) 37 28 35 PPD or lump-sum payment (percentage of claims) 50 42 49 Lump-sum payment (percentage of claims) 18 20 19 Defense attorney involved (percentage of claims) 22 25 27 Vocational rehabilitation services (percentage of claims) 34 33 32 PPD or lump-sum payment (mean) $13,400 $14,362 $11,436 Lump-sum payment (mean) $15,897 $15,592 $12,883 Duration of temporary disability (mean weeks) 25 25 25 Type of medical treatment received (percentage of claims) Major surgery 26 21 26 Chiropractic care 12 13 10 $11,341 $9,385 26 58 6 4 3 $5,969 $3,955 20 31 20 $8,540 $6,901 $8,962 $7,615 16 22 43 45 76 31 23 $5,295 $5,733 $5,411 $5,056 17 17 23 26 20 16 $5,717 $13,156 $3,925 $4,108 $11,000 $11,038 $7,755 $11,181 $6,867 $7,488 $9,371 $11,332 15 11 12 13 11 15 17 19 14 15 17 13 9 10 9 9 10 8 19 22 16 15 19 17 7 $19,036 $21,264 54 $21,629 $22,083 $22,484 $23,381 7 $32,789 $33,182 67 $28,604 $33,437 $29,523 $34,492 23 19 18 21 18 20 27 17 19 36 24 29 9 99 6 66 Note: All values in table for claims with more than 7 days of lost time. Key: PPD: permanent partial disability. Technical Appendix E: Statistical Methods This appendix explains the different statistical models used in our analysis and the interpretation of the results. We need to use different statistical models to study the alternative dependent variables because our dependent variables take differ- ent forms. For the three cost and outcome variables that are continuous (indemnity benefits, medical payments, and recovery of physical health), equation (2.1) is es- timated as a linear regression, in which case the estimated coefficient of a variable simply measures how the outcome changes with a one-unit increase in the vari- able. However, we always report the recovery results in terms of the percentage of recovery, relative to the worker’s health status before the injury. For the other out- come variables, we cannot use linear regression. In each case, there is some choice regarding exactly which type of model to use. We have chosen to use a set of mod- els for which the estimated coefficients have a very similar interpretation. The return-to-work outcome is dichotomous, and we estimate a logit model. In the logit model, Yis in equation (2.1) is replaced by an unobserved variable Y* is for the unobserved propensity to return to work. The discrete indicator Yis is then observed to equal 1 if Y* is > 0, and to equal zero otherwise. Using ZisΘ as a short- hand for the parameters and variables in equation (2.1), and assuming that the cumulative distribution for εis is the logistic, we have P(Yis = 1) = exp(ZisΘ)/[1 + exp(ZisΘ)] P(Yis = 0) = 1/[1 + exp(ZisΘ)]. (E.1) The parameters Θ are estimated by maximum likelihood. Finally, we present a transformation of the logit coefficients that is more easily interpretable. Specifically, the two expressions in equation (E.1) imply that 107 108 t e c h n i c a l a p p e n d i x e P(Yis = 1)/P(Yis = 0) = exp(ZisΘ), (E.2) which in turn implies that exp(Θk) — where Θk is the coefficient on a particular variable Zk in Z — measures the multiplicative effect on the relative probability P(Yis = 1)/P(Yis = 0) of a one-unit increase on Zk.1 For example, if exp(Θk) equals 1.2, then a one-unit increase in Zk increases the relative probability of P(Yis = 1) by 20 percent.2 The model for the duration of time out of work has to be estimated using survival models to account for the possible truncation of the period out of work. That is, it is possible that, at the time of the survey, some individuals are still in the midst of a period out of work, in which case all we know is that the period out of work lasts at least up to the time of the survey. In this framework, the outcome measure is Tis, the length of the period out of work. We estimate an accelerated failure time model, in which Tis = exp(ZisΘ + σεis). (E.3) In a standard regression model, we would assume a distribution for ε, typically normal, in which case least squares gives the maximum likelihood estimates. However, there are censored periods in which we do not know the ultimate value of Tis but only that it is at least as large as tis, because the period is still ongoing at the time of the survey. In addition, it is common in these models to fix the variance of ε at 1, and allow σ to be a parameter that is estimated. In this setting, we build the likelihood function for two types of observations. For the uncensored observations, we have an expression for the probability of observing a period of length Tis, or f(Tis). For the censored observations, all we know is that period out of work lasts at least as long as tis. The probability of this event is 1 minus the cumulative distribution function for tis, or the survivor function for tis, which we denote 1 This is also often described as measuring how the “odds” of the outcome Yis = 1 change with Z, where the odds are the relative probability. (For example, if the probability of event A is .75, and the probability of event B is .25, then the odds ratio for event A relative to event B is 3 = .75/.25.) To see this in the simplest example, suppose that Z contains an intercept and a single dummy variable, or ZisΘ = μ + D, where D is the dummy variable. Then the odds ratio when D = 0 is exp(μ), the odds ratio when D = 1 is exp(μ + ), and the ratio of the second to the first is exp( ). The same is true when D is continuous, and we think in terms of a one-unit increase in D. 2 In each case, we report the standard error such that the t-statistic is the same as on the original coefficient of the logit model. technical appendix e 109 S(tis). The density and survival functions are related through the hazard function h(tis) = f(tis)/S(tis). All that remains is to specify a distribution for ε in equation (E.3). We assume a logistic distribution for ε (a log-logistic distribution for Tis), in which case the survivor function is S(tis) = 1/[1+ {exp(−ZisΘ)tis}1/σ]. (E.4) The hazard function is more complicated, but an appealing feature of the function that results for the log-logistic distribution is that it is flexible and can be increasing monotonically, decreasing monotonically, or first increasing and then decreasing, depending on the value of σ.3 Finally, a nice feature of the log-logistic distribution is that an expression very similar to that for the logit model results; specifically, we have S(tis)/{1- S(tis)} = exp[Zis(Θ/σ) − (1/σ)ln(tis)], (E.5) which implies that exp(Θk/σ), computed from the coefficient on a particular variable Zk in Z, measures the effect of a one-unit increase in Zk on the ratio of the probabilities of the period lasting at least as long as any time t.4 This parallels the earlier interpretations of the parameters for the logit and multinomial logit models. However, it is also the case that exp(Θk/σ) equals the ratio of the expected duration when the corresponding variable Zk is one unit higher to when it is not, and therefore 100 × (exp(Θk/σ) − 1) measures the percentage by which the expected duration is longer with this change in Zk. We report these percentages in the tables.5 3 This contrasts with some more widely used distributional assumptions that impose more restrictions on the hazard function. 4 The implication of this is that the values of the regressors Z and the parameters Θ exert a proportional shift on the odds ratio in equation (E.5) for all values of t. That is, for any two individuals, who have different Z’s but the same values of Θ, the odds of the spell lasting longer than t are constant for any t. 5 In working with duration models, sometimes attention is given to the problem of unobserved heterogeneity. We do not think this is critical in our context for two reasons. First, because we have very detailed controls we do not have reason to believe that there is an important role for unmeasured heterogeneity. More importantly, the unique problem that unobserved heterogeneity introduces in duration models is bias in the estimates of parameters measuring duration dependence, because one cannot easily identify whether, for example, the probability of escaping from some status decreases over time because of duration dependence or because the sample increasingly shifts toward those likely to have (footnote continued on next page) 110 t e c h n i c a l a p p e n d i x e Finally, the satisfaction outcome is also discrete but takes on four values: very satisfied, somewhat satisfied, somewhat dissatisfied, and very dissatisfied. These values are ordered, given that the satisfaction responses can clearly be ranked. To study the outcome, an ordered discrete choice model is used. The framework is similar to that for the logit model, with Y* is now interpreted as the unobserved continuous measure of satisfaction, which follows the model Y* is = ZisΘ + εis. Now, though, the individual responds with the lowest category, Yis = 1, if Y* is < ω1, the next category, Yis = 2, if ω1 ≤ Y* is < ω2, etc., and the highest category, Yis = 4, if ω3 ≤ Y*is, with ω1 < ω2 < ω3 (the ω’s are unknown parameters to be estimated). Assuming again that the cumulative distribution function of ε is logistic, then the probability of each of these outcomes can clearly be written as a function of the same expressions used in the logit model. For example, we have P(Yis = 1) = P(Y*is < ω1) = 1/[1 + exp(ZisΘ − ω1)], (E.6) and P(Yis = 2) = P(Y*is < ω2) − P(Y*is < ω1) = {1/[1 + exp(ZisΘ − ω2)]} − {1/[1 + exp(ZisΘ − ω1)]}. (E.7) In this way the probability of each response can be written, and the likelihood function constructed. Note that in this case the relative probability of the response being in any category j+1 or higher relative to j is P(Yis ≥ j+1))/P(Yis = j) = exp(ZisΘ − ωj), (E.8) so that, paralleling the logit model, exp(Θk) measures the effect of a one-unit increase in Zk on the log of the relative probability P(Yis ≥ j+1))/P(Yis = j), or the relative probability of reporting a higher level of satisfaction. long durations. By extension, bias will also be transmitted to coefficients of variables in duration models that vary with time. However, we have neither of these. Rather, we simply have time-invariant controls that are unlikely to be affected by unobserved heterogeneity any more than would coefficients in a standard regression model. So rather than use statistical tricks to address this problem in duration models (such as assuming a functional form for the heterogeneity and integrating out), we prefer to use the data to try to address the issue. The place we think this is most important is with respect to the problem of unobserved injury severity, which was discussed in detail in the main text of the report. Technical Appendix F: Selected State System Features 112 t e c h n i c a l a p p e n d i x f Table F.1 Coverage under the State Workers’ Compensation Laws, 2004 General provisions Private employment California Texas Massachusetts Compulsory. Voluntary (see note). Selected exemptions from coverage Farm laborers Compulsory (see note). California Texas Massachusetts None. Domestic workers Generally exempt for migrant, seasonal, and other farm workers under payroll limit or for farms with fewer than three workers. None. California Texas Massachusetts Exempt if employed fewer than 52 hours in preceding 90 days or if earned less than $100. Exempt for employment incidental to personal residence. Exempt if employed fewer than 16 hours weekly. Pennsylvania Compulsory. Pennsylvania Exempt unless employer is otherwise covered, pays more than $1,200 in wages per year, or furnishes employment for more than 30 days. Pennsylvania Exempt. Massachusetts: Among employment exceptions to compulsory coverage are commission-paid salespersons and independent taxi drivers. Texas: In a 2004 survey of employers, the Texas Department of Insurance (2004) found that 38 percent of employers in Texas did not carry workers’ compensation coverage; that group — mostly smaller companies — employed about 24 percent of the workforce in the state. Texas self-insures employees of the state highway department, the University of Texas, Texas A&M University, and state employees. Other public employers are self-insured, risk-pool insured, or insured through commercial regulated carriers. The level of nonparticipation has been declining, from 44 percent in 1993, to 39 percent in 1996, and down to 35 percent in 2001. Under HB 2600, Article 16, employers who do not have workers’ compensation insurance cannot take a preinjury waiver of an injured worker’s right to sue for damages under the common law. Previously, a worker who signed a waiver could not sue his or her nonsubscribing employer in case of an injury. Sources: Texas Department of Insurance, 2004; U.S. Chamber of Commerce, 2004. Table F.2 Medical Cost Containment Strategies, 2004 Who can be a treating provider (2001) California Texas Treating provider can be medical doctor, psychologist, osteopath, chiropractor, podiatrist, dentist, optometrist, acupuncturist. Treating provider can be medical doctor, osteopath, chiropractor, podiatrist, dentist, optometrist. Initial choice of provider California Employer for first 30 days, unless worker predesignated a treating physician (see note). Employee change of provider California Once after 30 days, or after 90 to 365 days if HCO arrangement (see note). Medical fee schedule California Yes. Texas Worker from commission-approved list; a nonsubscriber typically writes into the plan that the employer chooses the provider (see note). Texas With commission’s approval using stated criteria. Texas Yes (see note). Massachusetts State statutes do not define who can be designated a treating provider. Some examples included in the state’s fee schedule for workers’ compensation providers are psychologist, medical doctor, osteopath, chiropractor, podiatrist, dentist, optometrist, independent nurse practitioner, physician assistant, certified nurse anesthetist (CRNA), licensed social worker (LICSW), physical therapist, and occupational therapist. Massachusetts Worker; if a PPA exists, worker may be required to have first appointment with provider from plan. Massachusetts Once within the same specialty. Massachusetts Yes. Pennsylvania Treating provider can be nonphysician. Pennsylvania Employer directs choice for first 90 days by posting list of six or more designated health care providers; worker choice if no panel is posted. Pennsylvania After 90 days without restriction (if a panel is posted) or at any time (if no panel is posted). Pennsylvania Yes. continued 113 technical appendix f 114 t e c h n i c a l a p p e n d i x f Table F.2 Medical Cost Containment Strategies, 2004 (continued) Hospital payment regulation California Texas Massachusetts Yes. DRG-based fee schedule for inpatient services. Treatment guidelines California Yes. Reimbursement paid on per diem basis. Texas Yes. Hospital-specific percentage discounts established annually; alternative rates and services can be negotiated. Massachusetts Yes, since 1995; advisory; nine guidelines in place (see note). Optional as of June 17, 2001. If elected, guidelines have to be nationally recognized, scientifically valid, and outcome based (see note). Yes; used in conjunction with UR program; 28 guidelines in place, developed through consensus-based, multidisciplinary effort. Pennsylvania Yes. Based on 113% of Medicare plus pass-through costs. Pennsylvania No. California: Senate Bill 899 allows employers to establish medical treatment networks, effective January 1, 2005; injured workers who do not predesignate a treating physician must receive care only through the network. Under SB 899, an employee may be treated by a predesignated physician from the date of injury if all of the requirements for predesignation are met. AB 749 reduced the offer of two HCOs to one. If HCO or personal physician is not predesignated prior to injury, employee will be treated by the HCO selected by the employer. SB 228 required the Commission on Health and Safety and Workers’ Compensation (CHSWC) to conduct a study and evaluation of existing treatment utilization standards by July 1, 2004, and to issue a report of its findings by October 1, 2004. Further, the legislation required the administrative director in consultation with CHSWC to adopt a medical treatment utilization schedule by December 1, 2004, based on CHSWC study recommendations. All employers are required to adopt utilization review systems consistent with the utilization schedule. Texas: HB 2600 called for an introduction of regional provider networks. At present, however, injured workers are not required to choose a doctor participating in a network but must choose a provider who is on the Commission’s Approved Doctor List (ADL). As of September 1, 2003, all health care providers practicing in the workers’ compensation system must be on the ADL, which means that they must be trained, have applied, and been approved to practice workers’ compensation. In practice, the fee schedule is reviewed every four to six years. A new Medical Fee Guideline was adopted in April 2002, but a legal challenge delayed its implementation until August 1, 2003. The Workers’ Compensation Commission is considering adopting treatment/loss time guidelines. Key: DRG: diagnostic related group; HCO: health care organization; PPA preferred-provider arrangement; UR: utilization review. Source: Tanabe and Murray, 2001. Table F.3 Statutory Benefit Parameters for Weekly Temporary and Permanent Partial Disability Payments Effective Date TTD Benefit Rate Maximum Benefit Minimum Benefit (not to exceed AWW or percentage of AWW, as noted) California (see note) 7/1/99–6/30/00 Massachusetts 10/1/99–9/30/00 Pennsylvania (see note) 1/1/99–12/31/99 Texas (see note) 9/1/98–8/31/99 662 3% of AWW 60% of AWW 662 3% of AWW 70%; if hourly wage more than $8.50 per hour, then 75% of AWW (for 26 weeks) $490.00 $749.69 $588.00 $523.00 $126.00 or worker’s AWW, whichever is less $149.93 or worker’s AWW, whichever is less $294.00 or 90% of worker’s AWW, whichever is less $78.00 technical appendix f California: Prior to AB 749, the maximum and minimum benefits were changed by periodic legislation, rather than by automatic annual adjustments tied to annual changes in the state average weekly wage. Where changes were made, the legislative changes took effect on July 1 of the relevant years. California has three benefit tiers: Workers receive 100 percent of their AWW up to $126, then $126 up to an AWW of $189, and then two-thirds of their AWW up to the maximum. Under legislation signed into law February 15, 2002, maximum temporary disability benefits were increased to $602 a week effective January 2003 and to $840 a week by 2006. Pennsylvania: If the statutory benefit rate is less than 50 percent of the statewide average weekly wage (SAWW), the benefit must be calculated using the lower of 50 percent of the SAWW or 90 percent of the worker’s AWW. The minimum benefit is the point at which benefits computed using the statutory rate are subject to recalculation. Annual increases in benefits go into effect January 1. Texas: Temporary total disability benefits are called temporary income benefits in Texas. For workers who earn less than $8.50 an hour, the benefit rate is 75 percent of their AWW for the first 26 weeks; the benefit rate reverts to 70 percent after 26 weeks. The minimum weekly benefit for temporary disability is 15 percent of the statewide average weekly wage for manufacturing production workers. Key: AWW: average weekly wage; TTD: temporary total disability. Source: State workers’ compensation statutes. continued 115 116 t e c h n i c a l a p p e n d i x f Table F.3 Statutory Benefit Parameters for Weekly Temporary and Permanent Partial Disability Payments (continued) Effective Date PPD Benefit Rate (percentage of AWW) Maximum Benefit Minimum Benefit (not to exceed AWW or percentage of AWW, as noted) California (see note) 7/1/99–6/30/00 Massachusetts (see note) 10/1/99–9/30/00 6623% of AWW n.a. Pennsylvania (see note) 1/1/99–12/31/99 Texas (see note) 9/1/98–8/31/99 6623% of AWW 70% of AWW $140.00–$230.00 For scheduled benefits, statutory amount based on the SAWW at time of injury $588.00 $366.10 $70.00 n.a. $294.00 or 90% of worker’s AWW, whichever is less $78.00 Table F.3 Statutory Benefit Parameters for Weekly Temporary and Permanent Partial Disability Payments (continued) California: For a disability rating of greater than 70 percent, workers receive a lifetime pension, an additional benefit paid for life. The benefit is 1.5 percent of the worker’s AWW for each percentage-point rating over 60 percent, up to the maximum earnings limit for the date of injury. PPD weekly payments, now $140 a week, will increase to $230 a week in 2006. Legislation passed in April 2004 made a number of changes to permanent disability benefits, including increasing benefits paid to severely injured workers (72 percent rating or higher) by nearly doubling the number of weeks per rating point and reducing the benefits paid to workers with a rating of 15 percent or less by reducing the number of weeks per rating point. Massachusetts: Massachusetts does not pay benefits for unscheduled permanent disability. Instead, the state pays benefits for wage loss or loss of wageearning capacity through partial disability benefits. These benefits are paid at 60 percent of the difference between a worker’s preinjury and actual wages or earning capacity, but not more than 75 percent of what the worker would receive for total disability benefits if eligible, or two times the SAWW. Pennsylvania: Table entries are for scheduled benefits only. Scheduled benefits are called specific-loss benefits in Pennsylvania. There are two different periods of payments under specific loss: for the healing period and for the specific loss itself. By statute, benefits are paid for the healing period before benefits are paid for the specific loss. The healing period ends when the worker returns to work at the preinjury wage or the period specified in the statute ends. Pennsylvania does not pay benefits for unscheduled permanent disability. Instead, the state pays benefits for wage loss or loss of wage-earning capacity through partial disability benefits. Those benefits are paid at 6623 percent of the difference between the preinjury and current actual or imputed wages, subject to the total disability maximum. If the benefit at the statutory rate is less than 50 percent of the SAWW, the benefit must be calculated using the lower of 50 percent of the SAWW or 90 percent of the worker’s AWW. The minimum benefit column in the table lists the point at which benefits computed using the statutory rate are subject to recalculation. Texas: PPD benefits in Texas are called impairment income benefits (IIBs). A worker may receive a supplemental income benefit (SIB) when IIBs end. Four conditions must be met: (1) the worker’s impairment rating must be at least 15 percent, (2) the worker has not taken an advance payment of benefits due (commutation), (3) the worker has not returned to work or is unable to earn at least 80 percent of the preinjury weekly wage, and (4) the worker has made a good-faith effort to find suitable work. The SIB is calculated at 80 percent of the difference between 80 percent of the worker’s average weekly wage and the worker’s earnings over the reporting period and cannot exceed 70 percent of the SAWW. Eligibility for SIB terminates at 401 weeks after date of injury. Key: AWW: average weekly wage (preinjury); n.a.: not applicable; PPD: permanent partial disability; SAWW: statewide average weekly wage. Sources: State statutes; California Division of Workers’ Compensation; Massachusetts Department of Industrial Accidents; Pennsylvania Bureau of Workers’ Compensation; Texas Workers’ Compensation Commission. technical appendix f 117 118 t e c h n i c a l a p p e n d i x f Table F.4 Waiting Period and Limits on Duration of Temporary Disability Benefits, 2002 Waiting period before income benefits commence California Texas Massachusetts Pennsylvania After 3 days of lost After 7 days of lost time. time. After 5 days of lost time. Limitations on duration of weekly temporary disability benefits After 7 days of lost time. California Texas Massachusetts Pennsylvania 5 years or until RTW or condition determined permanent and stationary (see note). 104 weeks of benefits, or until RTW, maximum medical improvement, or physician approves RTW and worker has bona fide offer of employment at preinjury wage. 156 weeks from injury, or until RTW or treating or impartial physician approves RTW and worker refuses suitable job. Duration of disability, unless adjudicated or agreed to; since June 24, 1996, 104 weeks for workers with less than 50% permanent impairment; no limit for workers with greater than 50% impairment. California: The 2004 legislation set a limit of 104 weeks of paid temporary disability within two years of the first temporary payment except for specified injuries that usually require extended recuperation (the limit had been 5 years). There is no explicit cost-of-living adjustment for benefits. However, if any temporary total disability (TTD) payment is made for two or more years from the date of injury, the amount of the benefit is adjusted to the TTD rate in effect at that time, based on the worker’s average weekly wage. Key: RTW: return to work. technical appendix f 119 Table F.5 Comparative Statistics on Costs, Injury Frequency, and Other Metrics of Workers’ Compensation Systems CA MA PA TX Comparative Source Benchmark Injuries/illnesses per 100 workers (2001) All cases Lost-time cases WC claim cost per worker (2000 policy year) WC cost per claim (2000 claims with experience to 2003) Total benefits Medical Indemnity Litigiousness Percentage of cases with defense attorney involvement (2000) Average payment to defense attorney (a proxy for hours billed/intensity of litigation) Percentage of people without health insurance (non-WC, 2001–2003) Fee schedule (percentage different from state Medicare rates, 2001) Overall Office visits Surgery Physical medicine Radiology 6.0 3.3 1,292 $51,212 $25,996 $25,216 29.7 4,063 18.7 12 –10 36 –1 14 5.1 2.5 388 $24,320 $6,913 $17,407 19.8 2,436 9.6 –13 –36 –4 –3 –12 n.a. n.a. 547 $35,921 $16,468 $19,453 19.8 3,500 10.7 17 –6 42 8 27 4.9 2.5 521 $41,755 $27,757 $13,998 11.1 1,884 24.6 38 –8 75 26 60 US 5.7 2.8 Median of 46 states 460 Median of 46 states $29,458 $14,611 $14,847 Median of 12 large states 24.7 2,415 15.1 44 9 80 9 66 a b b c d e continued 120 t e c h n i c a l a p p e n d i x f Table F.5 Comparative Statistics on Costs, Injury Frequency, and Other Metrics of Workers’ Compensation Systems (continued) CA MA PA TX Comparative Source Benchmark Network penetration rate (percentage of payments to network providers, 2000) Overall Office visits Surgery Physical medicine Radiology None 49 24 23 30 52 25 21 33 53 19 21 33 44 31 23 27 55 25 33 35 f a U.S. Department of Labor, Bureau of Labor Statistics, 2001. b Derived from NCCI data. c Telles, Wang, and Tanabe, 2005. d DeNavas-Walt, Proctor, and Mills, 2004. e Eccleston, Laszlo, Zhao, and Watson, 2002. f Wang and Zhao, 2003. Key: n.a.: not available; WC: workers’ compensation. Technical Appendix G: Tests of Pooling versus Individual State Regressions The main estimates reported in Tables 3.1 and 4.2 pool the data across the four states. The justification for doing this is the additional statistical precision we get from the larger sample, compared with analyzing each state separately. Pooling observations across the four states, however, restricts the coefficients of the models we estimate to be the same in each state, with the exception of the intercept, which is allowed to differ via the state dummy variables. To see this, note that another version of estimating the models separately by state is to allow a full set of interactions of the control variables with the state dummy variables. In the generic case, equation (2.1) would become Yis = α + CHOICEisβ + CHOICEis·STATEsβ′ + WORKERisγ + WORKERis·STATEsγ′ + FIRMisδ + FIRMis·STATEsδ′ + INJURYisθ + INJURYis·STATEsθ′ + STATEsκ + TREATMENTisλ + TREATMENTis·STATEsλ′+ εis.1 (G.1) The combined model used in Tables 3.1 and 4.2 assumes that β′, γ′, δ′, θ′, and λ′ are zero, in which case equation (G.1) reduces to equation (2.1). We can, however, test this set of restrictions by estimating the expanded model in equation (G.1) and then testing these restrictions explicitly. 1 In the case of a linear regression model, as we estimate for indemnity and medical payments and recovery, using estimates of this expanded regression model to obtain the effects for each state would give us exactly the same coefficient estimates as we obtain when we estimate the models separately by state. Only the standard errors of the estimates differ slightly. 121 122 t e c h n i c a l a p p e n d i x g Table G.1 Two-Way Classification of Provider Choice, Effects of Provider Choice on Outcomes, Tests of Pooling Restrictions Indemnity Benefits Medical Benefits Duration of Temporary Disability Substantial Return to Work Perceived Recovery Satisfaction Model 1: Without treatment controls Control coefficients equal across states Provider choice coefficients equal across states Model 2: With treatment controls Control coefficients equal across states Provider choice coefficients equal across states (1) 0.0490 0.4048 0.0017 0.2426 (2) 0.1494 0.5744 0.0139 0.2443 (3) <0.005 0.8309 <0.0001 0.8508 (4) 0.8732 0.9706 0.8152 0.9484 (5) <0.0001 0.8229 0.0006 0.6809 (6) 0.4611 0.6530 0.4254 0.7579 Note: P-values are reported for the restrictions specified in the leftmost-column. Table G.2 Three-Way Classification of Provider Choice, Effects of Provider Choice on Outcomes, Tests of Pooling Restrictions Indemnity Benefits Medical Benefits Duration of Temporary Disability Substantial Return to Work Perceived Recovery Satisfaction Model 1: Without treatment controls Control coefficients equal across states Provider choice coefficients equal across states Model 2: With treatment controls Control coefficients equal across states Provider choice coefficients equal across states (1) (2) (3) 0.0593 0.1212 0.1661 0.7096 0.0024 0.1900 0.0127 0.5893 <0.005 0.7295 <0.0001 0.9269 (4) (5) 0.8887 0.5288 <0.0001 0.4626 0.8248 0.5429 0.0005 0.4012 (6) 0.5484 0.2823 0.5088 0.2912 technical appendix g Note: P-values are reported for the restrictions specified in the left-most column. 123 124 t e c h n i c a l a p p e n d i x g We do this for the two-way and three-way classification of choice. We also consider this set of restrictions separately for the effects of provider choice (CHOICE), which is our direct concern, as well as the effects of the other control variables. The results of these statistical tests are reported in Tables G.1 and G.2, for the two-way and three-way classifications, respectively. We report the p-values from the tests. A p-value below 0.05 means that the set of tested coefficients are jointly significant at the five-percent level, and a p-value below 0.1 means that the set of tested coefficients are jointly significant at the 10 percent level. The results from these tests are quite clear. There is not a single case in which we reject the restrictions that the coefficients of the provider choice variables are equal across states at the ten-percent level, and in most cases the evidence against this restriction is very weak (as reflected in p-values closer to 1). This suggests that the estimates we obtained by pooling across states and restricting all coefficients except the intercept to be the same across states are not biased by imposing invalid restrictions.2 There is only one type of evidence suggesting that this conclusion may be unwarranted. Specifically, in contrast to the test of the restrictions on the provider choice coefficients, we frequently reject the restrictions that the coefficients of the other control variables are equal. These findings suggest that we reestimate the models continuing to restrict the coefficients of the provider choice variables to be the same across states, while freeing up the other coefficients; that is, in models corresponding to equation (G.1), we only restrict β′ to be zero. This is of interest, because it is possible that incorrectly restricting the coefficients of the other control variables to be equal across states could, in principle, bias the estimates of the provider choice coefficients. The results from this approach are reported in Tables G.3 and G.4. In each case, the top panel repeats the estimates from the analysis in the main text in which all slope coefficients are constrained to be equal across states. The bottom panel then reports the results freeing up the coefficients other than those for provider choice, which are the specifications to which the statistical tests in Tables G.1 and G.2 2 We also checked whether we obtained sharper evidence of differences across states if we group the four states into two pairs — states with employer choice and states with employee choice. However, as the results from Chapter 5 suggest, the estimates were generally more similar for California and Texas, on the one hand, and Massachusetts and Pennsylvania, on the other. Since in each of these pairs there is one employee choice state and one employer choice state, it is clear that when we group by the choice regime, there is also no (and even less) evidence of differences in effects of whether the worker or employer actually chose the provider. technical appendix g 125 Table G.3 Impact of Employee Choice Compared with Employer Choice Model 1: Without Treatment Controls (percent) Model 2 With Treatment Controls (percent) Table 3.1 estimates Medical payments 21** ($1,868) 10* ($903) Indemnity benefits 15* ($1,908) 8 ($978) Duration 32** 23** Substantial return to work –19† –16† Perceived recovery 01 Satisfaction 57** 59** Estimates including control variables interacted with state dummy variables Medical payments 19** ($1,689) 9† ($813) Indemnity benefits 13† ($1,693) 7 ($851) Duration 31** 25** Substantial return to work –21* –20† Perceived recovery 01 Satisfaction 58** 58** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. In model 1, for example, when the worker chose the provider, medical payments were $1,868 greater per case (on average) than when the employer chose the provider; or the duration of time out of work was 32 percent longer when the worker chose the provider than when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Table 2.2 provides the breakdown of observations by state. Observations are weighted to be representative of claims within each state; see the discussion on weighting in Technical Appendix D. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. 126 t e c h n i c a l a p p e n d i x g Table G.4 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice Employee Chose a Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Employee Chose a New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Table 4.2 estimates Medical payments 22** ($1,924) 7 ($629) 20** ($1,745) 12* ($1,052) Indemnity benefits 9 ($1,116) –1 (–$162) 20** ($2,538) 15† ($1,879) Duration 17† 7 48** 40** Substantial return to work –4 3 –28** –28** Perceived recovery –3 –1 2 3 Satisfaction 86** 89** 38** 39** Estimates including control variables interacted with state dummy variables Medical payments 19** ($1,693) 6 ($536) 19** ($1,615) 11† ($969) Indemnity benefits 5 ($642) –3 (–$441) 20** ($2,540) 15† ($1,875) Duration 14 9 48** 42** Substantial return to work –3 3 –33** –34** Perceived recovery –3 –1 3 3 Satisfaction 80** 79** 42** 42** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. In model 1, for example, when the worker chose a prior provider, medical payments were $1,924 greater per case (on average) than when the employer chose the provider; or the duration of time out of work was 17 percent longer when the worker chose a prior provider than when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. technical appendix g 127 generally lead. It turns out, though, that the less restrictive estimates in the bottom panels are scarcely changed, and certainly none of the conclusions are. Thus, restricting the effects of provider choice to be equal across states is consonant with the data, and freeing up other restrictions that are rejected by the data does not change the results for the effects of provider choice on the workers’ compensation outcomes we study. Technical Appendix H: Full Regression Results 129 130 t e c h n i c a l a p p e n d i x h Table H.1 Determinants of Employee vs. Employer Choice of Provider, by State, Odds Ratios from Logit Models California Texas Massachusetts Pennsylvania State Pennsylvania California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school High school graduate Some college College graduate Postgraduate Survey in Spanish Workplace controls Firm size≤50 Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Clerical/professional services —— —— —— —— — — — — 0.999 1.197 1.457* 1.019 0.592* 1.039** 0.476 0.864 — 1.132 0.980 1.747 0.539 1.019* 0.953 0.893 1.056 0.674 1.001 0.301** 0.718 — 1.100 1.397 1.024 0.777 0.983† 0.902 0.959 0.947 0.213** 1.027† 1.878 0.818 — 1.070 0.800 1.106 <0.001 — 0.903 0.789 0.919 1.095 1.508 — — 0.703 0.631† 1.011 0.609 0.797 — — 0.616† 0.633 1.066 0.259† 0.678 — — — — — 1.011 1.227 0.919 0.891** 0.152** 0.999 1.683 0.800 — 0.491** 0.701 0.820 — 1.120 1.090 1.372 0.789 0.900 — technical appendix h 131 Table H.1 Determinants of Employee vs. Employer Choice of Provider, by State, Odds Ratios from Logit Models (continued) California Texas Massachusetts Pennsylvania Manufacturing Construction Trade Other industries Injury controls Back pain Nonback sprain or strain Fracture Inflammation, laceration, or contusion Other injuries Severity Treatment controls Overnight hospitalization Major surgery Attorney involvement N 0.497† 1.657 1.427 1.108 1.965† 0.990 1.097 — 1.478 0.984* 0.848 1.263 2.243** 538 0.351** 0.359† 0.687 1.537 1.678 1.346 1.504 — 1.189 1.011 0.654 1.315 1.256 481 0.555 0.290 0.631 0.684 0.953 0.851 1.044 — 0.760 1.019* 0.808 2.002** 1.182 376 0.795 1.322 1.048 0.456 1.421 1.277 1.629 — 1.950† 0.983** 1.623† 1.191 1.387 565 ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Table H.2 Determinants of Employee Choice of Prior and New Providers vs. Employer Choice of Provider, by State, Odds Ratios from Multinomial Logit Models 132 t e c h n i c a l a p p e n d i x h California Prior New Texas Prior New Massachusetts Prior New Pennsylvania Prior New State Pennsylvania California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school High school graduate Some college College graduate Postgraduate Survey in Spanish — — —— — — —— — — —— — — —— — — — — ——— ——— ——— ——— 1.005 0.984 1.747** 1.012 0.491* 1.049** 0.490 1.054 — 1.271 1.002 1.184 0.287† 0.999 1.387 1.172 1.021 0.647 1.029* 0.425 0.663 — 1.002 1.010 2.344† 0.789 1.012 0.734 0.825 1.050 1.145 1.004 0.116** 0.439† — 0.946 1.145 4.058† 0.503 1.021* 1.106 0.969 1.047 0.552† 1.002 0.422† 0.870 — 1.188 1.607 0.094† 0.799 0.991 0.892 1.093 0.905* 0.146** 1.038** 1.189 1.004 — 1.080 0.799 0.913 <0.001 0.978† 0.933 0.793 0.995 0.391† 1.016 4.344 0.760 — 1.129 0.855 1.320 <0.001 1.018† 0.818 0.938 0.949 0.244** 0.989 0.546 0.981 — 0.615† 0.595 1.618 1.006 1.902** 0.942 0.821** 0.088** 1.006 2.636 0.664 — 0.370** 0.812 <0.001** Table H.2 Determinants of Employee Choice of Prior and New Providers vs. Employer Choice of Provider, by State, Odds Ratios from Multinomial Logit Models (continued) California Prior New Texas Prior New Massachusetts Prior New Pennsylvania Prior New technical appendix h Workplace controls Firm size≤50 Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Clerical/professional services Manufacturing Construction Trade Other industries — 1.190 0.893 1.925† 1.186 1.272 — 0.477 1.665 2.124† 1.237 — 0.727 0.737 0.339† 1.005 1.837 — 0.562 1.763 1.024 1.166 — 0.666 0.736 1.575 0.456† 0.825 — 0.104** 0.180** 0.549 0.915 — 0.753 0.657 0.753 0.733 0.817 — 0.638 0.563 0.876 2.196 — 0.593† 0.523† 1.218 0.201† 0.488 — 0.455 0.185† 0.487 0.625 — 0.604† 0.731 0.874 0.384 1.022 — 0.674 0.442 0.851 0.704 — 1.171 0.944 2.030† 0.344* 0.576 — 0.518 0.487 0.661 0.269* — 1.104 1.326 0.742 2.527 2.096 — 1.722 3.863† 2.099 0.936 continued 133 134 t e c h n i c a l a p p e n d i x h Table H.2 Determinants of Employee Choice of Prior and New Providers vs. Employer Choice of Provider, by State, Odds Ratios from Multinomial Logit Models (continued) California Prior New Texas Prior New Massachusetts Prior New Pennsylvania Prior New Injury controls Back pain Nonback sprain or strain Fracture Inflammation, laceration, or contusion Other injuries Severity Treatment controls Overnight hospitalization Major surgery Attorney involvement N 1.849 0.808 1.036 — 1.713 0.979* 1.760 1.096 1.047 — 1.167 0.990 0.752 0.442† 1.117 — 0.606 1.001 2.596* 2.445† 1.896 — 1.796 1.015† 0.451 1.162 1.215 1.303 2.015** 2.454** 534 0.828 0.539 1.817† 1.110 0.691 1.773† 479 1.379 1.221 1.553 — 1.165 1.009 0.664 0.566 0.763 1.810 2.376 1.881 — 0.491 1.038** — 3.431* 0.986† 1.366 0.819 1.618 — 1.384 0.982* 0.923 0.671 2.239** 1.909* 1.533 0.885 375 2.348** 1.040 0.986 1.543† 1.315 1.467 563 ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Table H.3 Effects of Provider Choice on Medical and Indemnity Benefits Medical Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Provider choice Employee chose Employee chose prior provider Employee chose new provider State California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school Some college College graduate 21** — — 29** 40** –42** 1** –5 13* 3** 17† –1† 7 19† –9 –10 10* — — 31** 41** –23** 0† –12* 8† 2† 13† 0 5 20* –4 –6 — 22** 20** 29** 39** –42** 1** –5 12* 3** 16† –1† 6 19† –10 –11 — 7 12* 31** 41** –23** 0* –13** 7 2† 13† 0 5 20* –5 –7 Indemnity Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 15* — — 31** –12 5 2** 6 –2 7** 3 –1† 33† 38** 10 10 8 — — 28** –12 13 1** 1 –6 6** 1 0 30† 38** 12† 9 —— 9 –1 20** 15† 31** –13 5 28** –13 14 2** 5 –3 7** 2 –1† 32† 37** 10 10 1** 1 –6 6** 0 0 29 38** 11 9 continued 135 technical appendix h Table H.3 Effects of Provider Choice on Medical and Indemnity Benefits (continued) 136 t e c h n i c a l a p p e n d i x h Medical Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Indemnity Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Worker controls (continued) Postgraduate Survey in Spanish Workplace controls Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Manufacturing Construction Trade Other industries Injury controls Back pain Nonback sprain or strain Fracture Other injuries Severity –27† 46** 1 0 16 –15 3 1 14 7 –17 65** 54** 34** 42** 2** –18 40** 2 –3 23** –12 –2 –14 –1 3 –12 52** 13 26* –6 1** –27† 46** 1 0 16 –15 5 2 15 9 –16 65** 54** 34** 42** 2** –18 40** –25 39* 1 –3 23** –12 –1 –13 –1 5 –11 10 25** 7 –3 –6 7 31† –1 4 52** 71** 13 40** 26* 13 –5 27† 1** 2** –22 32† 11 23** 12 0 –8 –3 23 –2 8 60** 12 7 –5 2** –25 39* 10 24** 8 –3 –5 7 31† 0 5 71** 40** 13 28† 2** –21 31† 11 23** 14 0 –7 –3 23 –1 8 60** 12 7 –5 2** Table H.3 Effects of Provider Choice on Medical and Indemnity Benefits (continued) Medical Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Indemnity Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Treatment controls Overnight hospitalization Major surgery N — — 1,954 148** 118** 1,954 — — 1,945 145** 118** 1,945 — — 1,951 89** 73** 1,951 — — 1,942 88** 73** 1,942 technical appendix h Notes: The average medical payment is $8,713 in the two-way classification and $8,688 in the three-way classification. The average indemnity payment is $12,709 in the two-way classification and $12,714 in the three-way classification. We divide the coefficients by the average payments to get the percentage effect. Some claims have missing values for some measures such as married, wage, hourly worker, tenure, education, firm size, industry, overnight hospitalization, or major surgery. To include these claims in the regressions, dummy variables are created to indicate the missing information, but we do not show the coefficients of these dummy variables in the table. The omitted reference categories are employer choice; Pennsylvania; high school graduate, firm size ≤ 50; professional/clerical services; and inflammation, laceration, or confusion. An intercept is also included but not reported. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. 137 Table H.4 Effects of Provider Choice on Duration and Substantial Return to Work 138 t e c h n i c a l a p p e n d i x h Duration (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Substantial Return to Work (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Provider choice Employee chose Employee chose prior provider Employee chose new provider State California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school Some college College graduate 32** — — 35** 53** 24* 1** –6 9 0 29** 0 337** 34* –6 5 23** — — 32** 53** 31** 1** –10 5 0 30** 0 314** 35** –4 3 — 17† 48** 36** 52** 25* 1** –6 8 1 30** 0 334** 33* –7 3 — 7 40** 33** 51** 31** 1** –11† 4 0 30** 0 310** 34** –6 1 –19† — — –37** –56** –31* –3** 28* –1 0 –18 0 –74** –43** 29† 19 –16† — — –35** –57** –36** –3** 34** 1 0 –17 0 –74** –42** 28† 22 — –4 –28** –37** –55** –32* –3** 31* 0 0 –18 0 –74** –43** 31* 21 — 3 –28** –35** –56** –37** –3** 37** 2 0 –17 0 –74** –42** 30† 23 Table H.4 Effects of Provider Choice on Duration and Substantial Return to Work (continued) Duration (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Substantial Return to Work (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 technical appendix h Postgraduate Survey in Spanish Workplace controls Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Manufacturing Construction Trade Other industries Injury controls Back pain Nonback sprain or strain Fracture Other injuries Severity –26† –25† –27† –26† 52 52 51 51 50† 40 47† 37 –15 –13 –14 –11 –6 –6 –6 –6 –2 –2 0 –1 2 1 1 0 –14 –14 –13 –13 –14 –8 –13 –5 80* 74* 77* 70* 0 11 0 10 8 6 7 5 10 16 9 15 22 25 20 23 4 1300405 26 23 23 20 14 15 14 16 16 24 17 24 –4 –6 –6 –7 0 10 0 9 3 2 4 3 54** 41** 55** 41** –39* –33† –39* –33† 20 –4 21 –3 11 23 10 23 19 12 19 12 –7 0 –8 –1 9 –17 10 –16 –14 2 –15 0 2** 2** 2** 2** –1 0 0 0 139 continued 140 t e c h n i c a l a p p e n d i x h Table H.4 Effects of Provider Choice on Duration and Substantial Return to Work (continued) Duration (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Substantial Return to Work (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Treatment controls Overnight hospitalization Major surgery N — — 1,829 140** 73** 1,829 — — 1,820 138** 73** 1,820 — — 1,829 –56** –18 1,829 — — 1,820 –56** –18 1,820 Notes: To get the percentage effect, we take 100 × (ecoefficient – 1) for duration and 100 × (odds ratio – 1) for substantial return to work. Some claims have missing values for some measures such as age, married, wage, hourly worker, tenure, education, firm size, industry, overnight hospitalization, or major surgery. To include these claims in the regressions, dummy variables are created to indicate the missing information, but we do not show the coefficients of these dummy variables in the table. The omitted reference categories are employer choice; Pennsylvania; high school graduate, firm size≤50; professional/clerical services; and inflammation, laceration, or contusion. An intercept is also included but not reported. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Table H.5 Effects of Provider Choice on Recovery and Satisfaction Recovery (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Provider choice Employee chose Employee chose prior provider Employee chose new provider State California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school Some college College graduate 0 — — –12** –22** 15** –1** 11** 2 0 0 1** –19** –14** 4 3 1 — — –10** –22** 14** –1** 11** 2 0 0 1** –18** –14** 4 3 — –3 2 –12** –22** 15** –1** 11** 2 0 0 1** –19** –14** 4 3 — –1 3 –10** –22** 14** –1** 11** 3 0 1 1** –18** –14** 4 3 Satisfaction (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 57** — — –28** –29** –13 0 60** 19* –1 –21† 1** 28 11 7 26† 59** — — –28** –29** –10 0 60** 19* –1† –21† 1** 27 11 8 28† —— 86** 89** 38** 39** –29** –29** –13 –28** –29** –11 0 63** 20* –1† –21† 1** 29 11 7 27† 0 64** 20* –1† –21† 1** 27 10 8 29† continued 141 technical appendix h Table H.5 Effects of Provider Choice on Recovery and Satisfaction (continued) Recovery (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Worker controls (continued) Postgraduate Survey in Spanish Workplace controls Firm size 51–250 Firm size 251–1,000 Firm size≥1,000 High-risk services Low-risk services Manufacturing Construction Trade Other industries Injury controls Back pain Nonback sprain or strain Fracture Other injuries Severity 7 –38** 1 0 –14** 2 13* 4 2 1 6 –26** –16** 7 –8 3** 7 –36** 1 –1 –15** 1 13* 5 2 1 5 –24** –14** 8 –5 3** 7 –38** 2 –1 –14** 2 13* 4 1 1 6 –26** –16** 7 –8 3** 7 –37** 1 –1 –15** 1 13* 5 2 1 5 –25** –14** 7 –5 3** Satisfaction (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 18 –41* 8 –1 0 –7 7 –31* 18 2 –38* –45** –37** –22 –37** –2** 19 –38* 8 0 1 –6 7 –30* 20 1 –37* –44** –35** –21 –35** –2** 16 –39* 8 0 –2 –9 6 –31* 19 1 –38* –45** –37** –22 –38** –2** 18 –37* 8 1 –1 –8 7 –29* 21 0 –37* –44** –35** –21 –36** –2** 142 t e c h n i c a l a p p e n d i x h Table H.5 Effects of Provider Choice on Recovery and Satisfaction (continued) Recovery (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Satisfaction (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Treatment controls Overnight hospitalization Major surgery N — — 1,956 –18** –4 1,956 — — 1,947 –18** –4 1,947 — — 1,941 1 –6 1,941 — — 1,932 1 –7 1,932 technical appendix h Notes: The average recovery score is 19.23 in the two-way classification and 19.26 in the three-way classification. We divide the coefficients by the average score to get the percentage effect for recovery. We take 100 × (odds ratio – 1) to get the percentage effect for satisfaction. Some claims have missing values for some measures such as age, married, wage, hourly worker, tenure, education, firm size, industry, overnight hospitalization, or major surgery. To include these claims in the regressions, dummy variables are created to indicate the missing information, but we do not show the coefficients of these dummy variables in the table. The omitted reference categories are employer choice; Pennsylvania; high school graduate, firm size≤50; professional/clerical services; and inflammation, laceration, or contusion. An intercept is also included but not reported. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. 143 References Airey, C., S. Bruster, B. Erens, S. Lilley, K. Pickering, and L. Pitson. 1999. National surveys of NHS patients: General practice 1998. London: NHS Executive. Barth, P., and M. Niss. 1999. Permanent partial disability benefits: Interstate differences. Cambridge, MA: Workers Compensation Research Institute. Barth, P., and R. Victor. 2003. Outcomes for injured workers in Texas. Cambridge, MA: Workers Compensation Research Institute. Boden, L. 1992. Workers’ compensation medical costs: A special case. In J. Greenwood and A. Tarico, eds., Workers’ compensation health care cost containment. Horsham, PA: LRP Publications. Boden, L., and C. Fleischman. 1989. Medical costs in workers’ compensation: Trends and interstate comparisons. Cambridge, MA: Workers Compensation Research Institute. Borba, P., and T. Parry. 2000 (October). An evaluation of the comprehensive and organized managed care program: Final report. Unpublished report prepared for the Robert Wood Johnson Foundation and the Joint Industry Board of the Electrical Industry. Centers for Disease Control and Prevention. 2002 (March 28). Behavioral risk factor surveillance system: Historical questions. Available at http:// apps.nccd.cdc.gov/brfssQuest/. Coulter, I., and P. Shekelle. 1997 (December). Supply, distribution, and utilization of chiropractors in the United States. In D. Cherkin and R. Mootz, Chiropractic in the United States: Training, practice, and research. AHCPR Publication No. 98-N002. Washington, DC: Agency for Health Care Policy and Research. Damiano, A., G. Pastores, and J. Ware, Jr., 1998. The health-related quality of life of adults with Gaucher’s disease receiving enzyme replacement. Quality of Life Research 7:373–386. DeNavas-Walt, C., B. Proctor, and R. Mills. 2004. Income, poverty, and health insurance coverage in the United States, 2003. Washington, DC: U.S. Government Printing Office. 145 146 r e f e r e n c e s Durbin, D., and D. Appel. 1991. The impact of fee schedules and employer choice of physician. NCCI Digest 6(3): 39–59. Durbin, D., D. Corro, and N. Helvacian. 1996. Workers’ compensation medical expenditures: Price vs. quantity. Journal of Risk and Insurance 63(1): 13–33. Eccleston, S., A. Laszlo, X. Zhao, and M. Watson. 2002. Benchmarks for designing workers’ compensation medical fee schedules, 2001–2002. Cambridge, MA: Workers Compensation Research Institute. Ellenberger, J. 1992. Labor’s perspective on health care reform. In J. Greenwood and A. Tarico, eds., Workers’ compensation health care cost containment. Horsham, PA: LRP Publications. Employee Benefit Research Institute, Consumer Health Education Council, and Matthew Greenwald and Associates. 2002. 2002 Health confidence survey: Summary of findings. Available at http://www.ebri.org/hcs/2002/hcs02sof.pdf. Galizzi, M., and L. Boden. 1996. What are the most important factors shaping return to work? Evidence from Wisconsin. Cambridge, MA: Workers Compensation Research Institute. Lewis, J. 1992. Legislative reform efforts and the medical benefit. In J. Greenwood and A. Tarico, eds., Workers’ compensation health care cost containment. Horsham, PA: LRP Publications. Morrison, J. 1990. Medical cost containment for workers’ compensation. Journal of Risk and Insurance 57(4): 646–653. National Academy of Social Insurance. 2004. Workers’ compensation: Benefits, coverage, and costs, 2002. Washington, DC. National Federation of Independent Business Research Foundation and National Foundation for Unemployment Compensation and Workers’ Compensation. n.d. Legislative guide to workers’ compensation insurance reform in the states. Washington, DC. Neumark, D. 2005 (January). The workers’ compensation crisis in California: A primer. California Economic Policy 1(1): 1–20. Available at http:// www.ppic.org/main/publication.asp?I=583. Pozzebon, S. 1994. Medical cost containment in workers’ compensation. Industrial and Labor Relations Review 48(1): 153–167. Research and Oversight Council on Workers’ Compensation (ROC). 1998 (August). An analysis of workers who were fired or laid off after a work-related injury. Austin, TX. Research and Oversight Council on Workers’ Compensation (ROC) and MedFX, LLC. 2001. Striking the balance: An analysis of the cost and quality of medical care in the Texas workers’ compensation system. A report to the 77th Texas Legislature. Austin, TX. references 147 Shields, J., D. Baroni, and X. Lu. 2003 (Summer, Special Edition). Post-injury health status of Texas workers with soft-tissue injuries. Texas Monitor 8(4): 1–7. Tanabe, R., and S. Murray. 2001. Managed care and medical cost containment in workers’ compensation: A national inventory, 2001–2002. Cambridge, MA: Workers Compensation Research Institute. Telles, C., D. Wang, and R. Tanabe. 2004. CompScopeTM benchmarks: Multistate comparisons, 4th edition. Cambridge, MA: Workers Compensation Research Institute. Telles, C., D. Wang, and R. Tanabe. 2005. CompScopeTM benchmarks, 5th edition. Cambridge, MA: Workers Compensation Research Institute. Texas Department of Insurance. 2004. Employer participation in the Texas workers’ compensation system: 2004 estimates. Austin, TX. U.S. Chamber of Commerce. 2004. 2004 analysis of workers’ compensation laws. Washington, D.C. U.S. Department of Labor, Bureau of Labor Statistics. 2001. State occupational injuries, illnesses, and fatalities. Available at http://www.bls.gov/iif/oshstate.htm. Victor, R., P. Barth, and T. Liu. 2003. Outcomes for injured workers in California, Massachusetts, Pennsylvania, and Texas. Cambridge, MA: Workers Compensation Research Institute. Victor, R., and C. Fleischman. 1990 (June). How choice of provider and recessions affect medical costs in workers’ compensation. Cambridge, MA: Workers Compensation Research Institute. Victor, R., D. Wang, and P. Borba. 2002. Provider choice laws, network involvement, and medical costs. Cambridge, MA: Workers Compensation Research Institute. Ware, J., M. Kosinski, and W. Rogers. 1996. The accuracy of retrospective evaluations of physical, mental, and general health status among patients with chronic conditions (abstract of unpublished study). The Health Institute, New England Medical Center. Boston, MA. Available at http://www.sf-36.org/cgibin/discuss/msg.cgi?msg=710. Ware, J., S. Keller, and M. Kosinski. 1998. SF-12®: How to score the SF-12® physical and mental health summary scale. Lincoln, RI: QualityMetric, Inc. Ware, J., D. Turner-Bowker, M. Kosinski, and B. Gandek. 2002. SF-12v2TM: How to score version 2 of the SF-12® health survey. Lincoln, RI: QualityMetric, Inc. Washington Department of Labor and Industries and University of Washington Department of Health Services. 1997. Workers’ compensation managed care pilot project: Final report to the legislature. Seattle, Washington. About the Authors Dr. Richard A. Victor, executive director of WCRI, helped establish the Institute in 1983. He received his J.D. and a Ph.D. in economics from the University of Michigan, where he was the George Humphrey Fellow in Law and Economic Policy. He then spent seven years conducting research at the Rand Corporation in both Washington, D.C., and Santa Monica, California. At Rand, Dr. Victor was a principal researcher at the Institute for Civil Justice. Dr. Victor is the author of numerous books and articles on workers’ compensation issues. Dr. Peter S. Barth is currently a Professor of Economics Emeritus at the University of Connecticut, where he formerly served for six years as the Department Head. His undergraduate degree is from Columbia University, and he earned his Ph.D. at the University of Michigan. He has prepared studies on permanent partial disability in workers’ compensation and on workers’ compensation systems in Connecticut, Texas, California, and Florida for the Workers Compensation Research Institute. Much of his published work has dealt with compensating workers disabled by occupational diseases. He also has done studies of workers’ compensation programs in British Columbia, Ontario and Victoria, Australia. In addition, he has served as a consultant to numerous organizations and advised government agencies, state and federal in the U.S., and abroad on workers’ compensation and related issues. Dr. David Neumark is a senior fellow in economics at the Public Policy Institute of California and a research associate of the National Bureau of Economic Research. He has published numerous studies on school-to-work, workplace segregation, sex discrimination, the economics of gender and the family, affirmative action, aging, minimum wages, and living wages. He is on the editorial boards of Industrial Relations, Contemporary Economic Policy, and Economics of Education Review. He has also held positions as professor of economics at Michigan State University, assistant professor of economics at the University of Pennsylvania, and economist at the Federal Reserve Board. He holds a Ph.D. in economics from Harvard University. 149 Related PPIC Publication “The Workers’ Compensation Crisis in California: A Primer” California Economic Policy. David Neumark. Volume 1, Number 1, January 2005. 150 Related WCRI Publications Provider Choice Laws, Network Involvement, and Medical Costs. Richard A. Victor, Dongchun Wang, and Philip Borba. December 2002. WC-02–05. The Impact of Initial Treatment by Network Providers on Workers’ Compensation Medical Costs and Disability Payments. Sharon E. Fox, Richard A. Victor, Xiaoping Zhao, and Igor Polevoy. August 2001. DM-01–01. The Impact of Workers’ Compensation Networks on Medical and Disability Payments. William G. Johnson, Marjorie L. Baldwin, and Steven C. Marcus. November 1999. WC-99–5. 151" } ["___content":protected]=> string(104) "

R 1105RVR

" ["_permalink":protected]=> string(116) "https://www.ppic.org/publication/the-impact-of-provider-choice-on-workers-compensation-costs-and-outcomes/r_1105rvr/" ["_next":protected]=> array(0) { } ["_prev":protected]=> array(0) { } ["_css_class":protected]=> NULL ["id"]=> int(8453) ["ID"]=> int(8453) ["post_author"]=> string(1) "1" ["post_content"]=> string(0) "" ["post_date"]=> string(19) "2017-05-20 02:37:43" ["post_excerpt"]=> string(0) "" ["post_parent"]=> int(3656) ["post_status"]=> string(7) "inherit" ["post_title"]=> string(9) "R 1105RVR" ["post_type"]=> string(10) "attachment" ["slug"]=> string(9) "r_1105rvr" ["__type":protected]=> NULL ["_wp_attached_file"]=> string(13) "R_1105RVR.pdf" ["wpmf_size"]=> string(7) "1778899" ["wpmf_filetype"]=> string(3) "pdf" ["wpmf_order"]=> string(1) "0" ["searchwp_content"]=> string(307598) " The Impact of Provider Choice on Workers’ Compensation Costs and Outcomes Richard A. Victor Peter S. Barth David Neumark With the assistance of Te-Chun Liu WC-05-14 November 2005 Workers Compensation Research Institute Cambridge, Massachusetts Public Policy Institute of California San Francisco, California copyright 2005 © by the workers compensation research institute and public policy institute of california all rights reserved. Library of Congress Cataloging-in-Publication Data Victor, Richard A. The impact of provider choice on workers’ compensation costs and outcomes / Richard Victor, Peter Barth, David Neumark; with the assistance of Te-Chun Liu. p. cm. “WC-05-14.” A study that utilizes data taken from interviews in 2002 and 2003 of employees in California, Texas, Massachusetts, and Pennsylvania. Includes bibliographical references. ISBN 1-931906-39-4 1. Workers’ compensation — United States — Costs. 2. Medical care, Cost of — United States. 3. Physician services utilization — United States. I. Barth, Peter S., 1937II. Neumark, David. III. Workers Compensation Research Institute (Cambridge, Mass.) IV. Title. HD7103.65.U6V494 2005 368.4’101’0973 — dc22 2005049410 publications of the workers compensation research institute do not necessarily reflect the opinions or policies of the institute’s research sponsors. ppic does not take or support positions on any ballot measure or on any local, state, or federal legislation, nor does it endorse, support, or oppose any political parties or candidates for public office. research publications reflect the views of the authors and do not necessarily reflect the views of the staff, officers, or board of directors of the public policy institute of california. short sections of text, not to exceed three paragraphs, may be quoted without written permission provided that full attribution is given to the source and the above copyright notice is included. Acknowledgments This study benefited greatly from the contributions of a number of our colleagues at the Workers Compensation Research Institute (WCRI), the Public Policy Institute of California (PPIC), and elsewhere. We especially appreciate the thoughtful comments of technical reviewers Dr. Leslie Boden of Boston University, Dr. Jeffrey Harris of J. Harris Associates, Inc., and Dr. Allan Hunt of the Upjohn Institute. We are also grateful for comments from Mark Baldassare, Jon Haveman, Joyce Peterson, and Fred Silva, all of PPIC, and seminar participants at PPIC. We also wish to thank Linda Carrubba for her excellent assistance in preparing the draft with precision and good humor, and Jill McNamee, who managed the review and publication process. We are indebted to Barbara McGowran for editing our prose to improve its readability and accuracy, and to Jan Cocker for proofreading the final report. Of course, any errors in the study remain our responsibility. Richard A. Victor Cambridge, Massachusetts Peter S. Barth Storrs, Connecticut David Neumark San Francisco, California November 2005 iii Table of Contents List of Tables ix List of Figures xiii Executive Summary xv 1. Introduction Policy Context / 3 Objectives and Scope of This Study / 4 Organization of the Report / 7 3 2. Key Concepts, Data, and Methods Key Concepts / 9 outcomes / 9 costs / 10 provider choice / 10 other terms / 11 Data Sources / 11 Measuring Provider Choice / 12 Cost and Outcome Measures / 18 Empirical Methods / 20 basic model and control variables / 20 variation in state workers’ compensation systems / 23 appropriate statistical models for each dependent variable / 24 equality of effects of provider choice across states / 25 causal inferences about policy changes regarding provider choice / 25 9 v vi t a b l e o f c o n t e n t s 3. Impact of Employee or Employer Choice of Provider: Main Results Summary of Findings / 34 Impact of Employee or Employer Choice of Provider on Costs and Outcomes: Main Results / 35 33 4. Employee Choice of Prior Provider or New Provider, or Employer Choice of Provider: Main Results 39 Patterns of Choosing New and Prior Providers / 40 Summary of Findings / 41 Impact on Costs and Outcomes of Employee-Selected Prior Provider or New Provider, or Employer Choice of Provider: Main Results / 41 Revisiting the Question of Unmeasured Residual Injury Severity / 45 5. Impact of Provider Choice on Costs and Outcomes: Results for Individual States California / 48 Texas / 52 Massachusetts / 55 Pennsylvania / 58 47 6. Worker Satisfaction with Health Care Correlates of Satisfaction and Overall Health Care Received / 62 Correlates of Satisfaction and Provider Choice / 65 Satisfaction with Care: Prior or New Provider / 67 61 7. Discussion and Policy Implications Summary of Results / 70 Interpretative Caveats / 71 Implications for Public Policy / 72 69 table of contents vii Technical Appendix A: Literature Review Technical Appendix B: Variables Used in Study: Definitions and Descriptive Statistics Technical Appendix C: Discussion of Construction and Validity of Health Status, Recovery, and Perceived Severity Measures Technical Appendix D: Discussion of Survey Response Rates and Response Bias Technical Appendix E: Statistical Methods Technical Appendix F: Selected State System Features Technical Appendix G: Tests of Pooling versus Individual State Regressions Technical Appendix H: Full Regression Results References About the Authors Related PPIC Publication Related WCRI Publications 77 81 89 97 107 111 121 129 145 149 150 151 List of Tables 2.1 Pattern of Providers Who Delivered Nonemergency Care / 13 2.2 Who Chose the Respondent’s Primary Provider? / 15 2.3 Provider Choice, by Type of Provider / 16 2.4 Costs and Health Outcomes, by State / 19 2.5 Substantial Return to Work and Duration of Period Out of Work / 20 2.6 Satisfaction with Overall Health Care / 21 2.7 Determinants of Provider Choice, Four States Combined / 29 3.1 Impact of Employee Choice Compared with Employer Choice / 36 3.2 Satisfaction with Overall Care, by Who Selected the Provider and by State / 37 4.1 Employee Choice of Prior Provider or New Provider / 40 4.2 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Four States Combined / 42 4.3 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider / 45 4.4 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Four States Combined, Excluding Severity and Injury Measures / 46 5.1 Impact of Employee Choice Compared with Employer Choice, California / 49 5.2 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, California / 50 5.3 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, California / 51 5.4 Impact of Employee Choice Compared with Employer Choice, Texas / 53 ix x list of tables 5.5 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Texas / 54 5.6 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Texas / 55 5.7 Impact of Employee Choice Compared with Employer Choice, Massachusetts / 56 5.8 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Massachusetts / 57 5.9 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Massachusetts / 57 5.10 Impact of Employee Choice Compared with Employer Choice, Pennsylvania / 58 5.11 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Pennsylvania / 59 5.12 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Pennsylvania / 60 6.1 Satisfaction with Care, by Degree of Recovery of Physical Health / 64 6.2 Worker’s Perception of Timing of Return to Work and Satisfaction with Care / 64 6.3 Second Absence Due to Injury and Satisfaction with Care / 65 6.4 Completeness of Recovery, by Who Selected the Provider / 66 6.5 Worker’s Perception of Timing of Return to Work, by Who Selected the Provider / 66 6.6 Second Absence Due to Injury, by Who Chose the Provider / 67 B.1 Definitions of Variables / 82 B.2 Descriptive Statistics / 85 C.1 Incurred Costs for Medical Care, by Perceived Injury Severity / 95 C.2 Comparison of Two Surveys on General Health / 96 D.1 Attempted Telephone Interviews and Valid Phone Numbers / 99 D.2 Disposition of Cases with Valid Phone Numbers / 100 D.3 Analysis of Representativeness / 101 D.4 Analysis of Response Bias / 105 F.1 Coverage under the State Workers’ Compensation Laws, 2004 / 112 F.2 Medical Cost Containment Strategies, 2004 / 113 list of tables xi F.3 Statutory Benefit Parameters for Weekly Temporary and Permanent Partial Disability Payments / 115 F.4 Waiting Period and Limits on Duration of Temporary Disability Benefits, 2002 / 118 F.5 Comparative Statistics on Costs, Injury Frequency, and Other Metrics of Workers’ Compensation Systems / 119 G.1 Two-Way Classification of Provider Choice, Effects of Provider Choice on Outcomes, Tests of Pooling Restrictions / 122 G.2 Three-Way Classification of Provider Choice, Effects of Provider Choice on Outcomes, Tests of Pooling Restrictions / 123 G.3 Impact of Employee Choice Compared with Employer Choice / 125 G.4 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice / 126 H.1 Determinants of Employee vs. Employer Choice of Provider, by State, Odds Ratios from Logit Models / 130 H.2 Determinants of Employee Choice of Prior and New Providers vs. Employer Choice of Provider, by State, Odds Ratios from Multinomial Logit Models / 132 H.3 Effects of Provider Choice on Medical and Indemnity Benefits / 135 H.4 Effects of Provider Choice on Duration and Substantial Return to Work / 138 H.5 Effects of Provider Choice on Recovery and Satisfaction / 141 List of Figures 2.1 Provider Choice Questions / 14 C.1 Perceived Injury Severity and Recovery of Physical Health and Functioning: An Example from the Texas Results / 93 xiii Executive Summary With workers’ compensation medical payments rising rapidly in many states (Telles, Wang, and Tanabe, 2004), policymakers have intensified their efforts to modify state laws to try to reduce those costs, while avoiding actions that might impair the outcomes experienced by injured workers. One of the actions often debated is giving employers more influence or direct control over the selection of providers. The health care provider plays many critical roles in the outcome of a workers’ compensation case. Those roles can include giving information that bears directly on most aspects of a claim for medical and indemnity benefits; diagnosing the condition and assessing its cause, which can affect the compensability of the claim; prescribing and providing a course of treatment and disability management practices, which can influence whether the worker returns to work and how quickly; assessing whether the worker’s condition has reached maximum medical improvement, whether the worker is left with a permanent impairment or disability, and the extent of the impairment; and judging whether a preexisting condition contributed to the degree of impairment. From the perspective of either the employer or the worker, any of these decisions by the health care provider can be sufficiently important to warrant being able to control the selection decision. Thus, the selection of that provider is an important matter for all parties of interest. Workers and their advocates have argued that the choice of the treating doctor or provider should be left to the worker. At a minimum, they argue that workers should be treated by those whom they trust and whose interests align with the workers’ — interests that encourage prompt return to work, but only as medically indicated, and the fullest restoration possible of physical capacity (Ellenberger, 1992). In contrast, employer advocates believe the choice of provider should be made by the employer, arguing that employer choice ensures that incentives exist for keeping the costs of care reasonable and appropriate (Morrison, 1990), employer choice helps avoid excessive services and treatments, and providers familiar with the employer’s workplace can use that knowledge to expedite return to work (National Federation of Independent Business Research Foundation and Na- xv xvi e x e c u t i v e s u m m a r y tional Foundation for Unemployment Compensation and Workers’ Compensation, n.d.). From the late 1980s to the early 1990s — a period of rising costs — a number of states modified “employee choice” laws to require that workers select providers from within approved networks of providers created by employers. The important role of provider choice in workers’ compensation public policy debates was highlighted recently by the passage of Senate Bill (SB) 899 in California. Until recently, California employers had the right to select the initial provider, unless the employee had predesignated a provider; but after 30 days, the worker had the right to change to a medical provider of his or her own choice. However, SB 899 changed the rules regarding provider choice.1 In particular, employers are now allowed to establish networks composed of occupational and nonoccupational physicians, and the legislation grants to the employer (or the insurer) the sole right to decide which medical providers are in the network. Further, the worker’s right to choose a physician after 30 days no longer applies if a network is established that complies with the law, unless the worker has predesignated a physician under particular conditions. As long as employers establish networks, which many are expected to do, California workers will have less flexibility to choose their providers — especially new providers. Objectives of This Study The purpose of this study is to determine if measurable costs and outcomes in workers’ compensation cases are affected by who selects the health care provider. The costs and outcomes we study include medical and indemnity costs, the duration of time out of work, the likelihood that the worker had a sustainable return to work, the worker’s perception of the degree of recovery from the work injury, and the worker’s overall satisfaction with the health care received. This study has at least four important advantages over the few previous studies that have attempted to answer these questions. First, it utilizes data taken from employee interviews conducted in 2002 and 2003 in four states: California, Texas, Massachusetts, and Pennsylvania. Workers were asked to identify who selected 1 This was one of many reforms, some of which were included in legislation passed the previous year (SB 228), addressing the rapid escalation in workers’ compensation costs in California that began in 1999 (Neumark, 2005). executive summary xvii their health care providers. In contrast, the previous studies classify provider choice based on state statutory provisions, despite the fact that the statutory provisions are imperfectly related to actual choice of provider exercised by a worker. Second, we focus on the primary provider of medical care, who is often different from the initial provider playing a subsidiary role. Third, we link the data from the interviews to claims data supplied by the claims payors, providing information on factors that include medical and indemnity costs, medical treatments, and employer attributes, among others. A more complete picture of the claim from the vantage point of both the worker and the employer should help to better establish the consequences of provider choice. Fourth, the existing studies are limited to estimating the effects of provider choice on costs, whereas we look at a wider set of outcomes of concern to policymakers and stakeholders. Finally, the interview data also indicate whether the primary provider had previously treated the worker for an unrelated condition. As noted earlier, recent legislation in California recognized the distinction between employee choice of a prior provider and a new provider. Thus, our findings with regard to the consequences of employee choice of prior providers, employee choice of new providers, and employer choice are especially salient to assessing the likely consequences of this very important component of workers’ compensation reforms in California. Those reforms are just beginning to be implemented and will likely be questioned and reexamined as the state implements the reforms enacted to deal with the workers’ compensation crisis that emerged in California at the beginning of this decade. Summary of Results These are among the most important findings of this study: ■ Comparing cases in which the worker selected the primary provider with otherwise similar cases in which the employer selected the provider, we found that costs were generally higher and return-to-work outcomes poorer when the worker selected the provider. Workers reported higher rates of satisfaction with overall care but similar perceived recovery of physical health. ■ Compared with cases in which the employer selected the provider, cases in which the worker selected a provider who had treated the worker previously for an unrelated condition (prior provider) may have had higher xviii e x e c u t i v e s u m m a r y costs, but the evidence was weak. Worker outcomes did not appear to be very different between cases with employee-selected prior providers and those with employer-selected providers, except that satisfaction with overall care was higher when the worker saw a prior provider. ■ Compared with cases in which the employer selected the provider, cases in which the worker selected a provider who had not treated him or her previously (new provider) had much higher costs and poorer return-to-work outcomes, generally no differences in physical recovery, and higher levels of satisfaction with overall care. ■ Comparing cases in which the employee selected a prior provider with similar cases in which the employee chose a new provider, we found that the worker treated by a new provider was less likely to return to work, returned to work more slowly if he or she did return, had lower levels of satisfaction with overall care, and experienced no better physical recovery. Medical costs were similar in both cases, but indemnity costs per claim were higher for a worker treated by a new provider, although this evidence was not as strong statistically as the other results. These primary findings come from the combined four-state sample, because the combined data lead to a larger sample and more precise estimates than do the data from the individual states. However, we also examine results from individual states, even though they are less precise and the statistical tests are less powerful. The following are among the findings from the state-by-state analyses: ■ In the two states with higher-than-typical medical payments, California and Texas, comparing cases in which the worker selected the primary provider with otherwise similar cases in which the employer selected the provider, we found that costs were generally higher and return-to-work outcomes poorer when workers selected their providers, although workers reported higher rates of satisfaction with overall care and similar perceived recovery of physical health. ■ The findings for the combined four-state sample previously described — comparing employee-selected prior providers, new providers, and employer-selected providers — were especially strong for California and Texas, although in Texas, both cases with employee-selected prior providers and those with new providers had higher costs than cases with employer-selected providers. executive summary xix Implications for Public Policy How do these results inform public policy debates regarding choice of provider in workers’ compensation? We found some evidence to support both those who advocate for employee choice and those who advocate for employer choice. As described here, however, based on our findings, it appears possible to improve the design of provider choice laws to lower costs and improve return-to-work outcomes without adversely affecting physical recovery from workplace injuries. First, we found that when the worker chose the provider, costs were higher, recovery of health outcomes was not better, and return-to-work outcomes were often worse than when the employer selected the provider. This finding suggests that employers, on average, may be well positioned to select good-quality, lower-cost providers — or at least better positioned than many workers. The finding also suggests that employers, in practice, are not generally selecting inferior-quality providers; although there may be exceptions, they do not appear to be frequent enough to affect the overall results. Second, we found that when workers select new providers — those they had not been treated by previously — costs were higher and return to work outcomes were poorer. This evidence suggests that state laws that grant employers greater influence over the choice of provider should lead to lower costs and better returnto-work outcomes than laws that allow workers to select providers whom they have not seen previously. Third, we found that when workers selected providers with whom they had a preexisting clinical relationship, the costs and most outcomes were not dramatically different than when the employer selected the provider. However, when workers selected providers — either prior or new — they expressed higher levels of satisfaction with care. We are not surprised by this finding regarding workers choosing prior providers, because a key issue is the likelihood that a worker will be seen by a provider who has the appropriate training and skills, is trusted by the worker, and delivers appropriate care. More surprising, though, is that workers also expressed greater satisfaction when they selected new providers (relative to employers choosing). We explored whether this greater satisfaction appeared to be related to dimensions of physical recovery not captured in our data or assistance in remaining out of work beyond the necessary time following an injury, but we were able to rule out such explanations. There may, however, be alternative explanations related to empowerment, trust, or the process of care that leaves workers more satisfied with their new-provider choices, even though costs and return-to-work outcomes appear to be worse and physical recovery no better. xx e x e c u t i v e s u m m a r y The results for California paralleled those for the larger sample in providing some evidence suggesting that the costs were higher and return-to-work outcomes were worse when workers selected providers with whom they had no prior relationship. This suggests that the recent legislative changes in California — which significantly expanded the limits on worker choice of provider but retain an exception where there is a preexisting provider relationship — may have struck an appropriate balance. The predesignation exception to the worker’s choice of provider among an employer-designated network of providers bears some resemblance to the prior-provider category that we analyze. However, this prior-provider category is broader than the California requirements for predesignation — for example, the prior provider has to be the personal physician of the worker under a nonoccupational group health insurance plan offered by the employer. Thus, the results are potentially quite informative about the likely effects of the changes in provider choice recently enacted in California, but they do not provide a direct test of the impact of the reforms. Such a direct test will not be possible until some time after the reforms are implemented, probably at least a couple of years from now. The Impact of Provider Choice on Workers’ Compensation Costs and Outcomes 1 Introduction Policy Context Because health care costs in workers’ compensation have grown rapidly and have become an increasingly important proportion of system benefits, more attention has focused on the choice of provider (National Academy of Social Insurance, 2004). Selection of the health care provider, who can critically affect the outcomes of a workers’ compensation case, is therefore important for all parties interested in workers’ compensation. The provider typically gives information with a direct bearing on most aspects of a claim for medical and indemnity benefits. The provider might diagnose the worker’s condition and shed light on its source, which affects the compensability of the claim. The provider prescribes and, in most cases, provides a course of treatment and disability management practices, the results of which could affect whether the worker returns to work, the duration of the worker’s time out of work, and the conditions under which the return to work occurs. In most jurisdictions, the provider assesses whether the worker’s condition has reached maximum medical improvement and if the worker is left with a permanent impairment or disability. In some jurisdictions, an important component of the health care professional’s assessment is also whether a preexisting condition contributed to the degree of impairment. The rating of the degree of permanent impairment, if any, can significantly affect the amount of indemnity benefits paid under that claim. From the perspective of either the insurer or the worker, any of these evaluations by the health care provider can be sufficiently important to warrant being able to control the selection of provider. 3 4 the impact of provider choice on costs and outcomes Rising costs have also forced policymakers to recognize that the issue of provider choice is more complex than simply deciding which party chooses the initial provider. First, a person undergoing medical treatment can become dissatisfied with that treatment and want to change providers. Alternatively, an insurer or employer might believe the type or quality of treatment being given or planned is inappropriate, or at least inconsistent with the goal of returning the worker to employment expeditiously. Second, the increased use of specialized treatment means that multiple providers are increasingly likely to be involved, so that the initial choice of provider may sometimes be less important than the choice of another provider. Thus, aside from who selects the initial provider, policies regulating provider choice must also specify the circumstances under which a change of provider is permitted and the procedures for doing so. Workers and their advocates have argued that provider choice should be left to the worker.1 At a minimum, they argue, a worker should be treated by people he or she trusts and whose interests are consistent with the worker’s — interests that encourage prompt return to work, but only as medically indicated, and the fullest possible restoration of physical capacity (Ellenberger, 1992). In contrast, employer advocates argue that provider choice should rest with the employer, because without employer choice there is “little incentive to see that the costs of care remain reasonable and appropriate” (Morrison, 1990). Employer advocates also argue, “Employer selection of the treating physician serves to direct injured workers away from those providers who provide excessive services and treatment procedures,” and to “retain those providers familiar with the operations of the employer and who can expedite return-to-work based on that knowledge” (National Federation of Independent Business Research Foundation and National Foundation for Unemployment Compensation and Workers’ Compensation, n.d.). Objectives and Scope of This Study The purpose of this study is to determine whether measurable costs and outcomes in workers’ compensation cases are affected by who selects the health care pro- 1 This is not to suggest that it is solely workers who have supported employee choice. The organization of workers’ compensation state administrators — the International Association of Industrial Accident Boards and Commissions — at one time published a list of standards for the states in which it endorsed the standard of worker choice. However, the organization no longer publishes a listing of standards. introduction 5 vider. The costs and outcomes we study include medical and indemnity costs, the duration of time out of work, the likelihood that the worker returned to substantial employment, the worker’s own perception of the degree of recovery from the work injury, and the worker’s overall satisfaction with the health care received. Only a handful of studies have attempted to ask these questions, and most of those have focused only on costs. Further, most previous studies only consider the relationship between costs and states’ statutory provisions about choice, with states simply categorized as either “employer choice” or “employee choice” jurisdictions. However, a state’s statutory classification can have little to do with the actual choice of primary provider reported by the worker. What can we conclude from past studies? First, while most studies appear to conclude that employer choice is associated with lower medical payments in workers’ compensation, the findings are not uniform. The variation in conclusions may stem from differences in states and years studied and from the use of crude measures of provider choice. Second, very little work has focused on the impact of provider choice on outcomes or cost measures other than medical payments — such as duration of time out of work, indemnity benefits, physical recovery, and worker satisfaction with care. Third, it has been rare to control for the many other factors that likely affect outcomes. Fourth, no study appears to have considered and analyzed the significance of whether the injured employee had been treated previously by the provider who gave primary care in the workers’ compensation claim. Finally, studies from even a few years earlier were done when network arrangements were less common. Because employer-selected providers are more likely to participate in such plans now than they previously did, the relationships between provider choice and workers’ compensation costs and outcomes may have changed. Prior studies on provider choice are summarized in more detail in Technical Appendix A. This study has at least four advantages over previous studies. First, it utilizes data taken from employee interviews conducted in 2002 and 2003 in four states — California, Texas, Massachusetts, and Pennsylvania — which asked workers to identify who selected their health care providers.2 Worker identification of provider is a critical factor in this study, because studies by Lewis (1992), Barth and 2 In contrast to the studies relying only on state-level variation, we have information on provider choice as reported by workers. We categorize a case as employer choice if either an employer or an insurer made the choice; we categorize a case as employee choice if the selection of the provider was made by the worker, by a friend or family member of the worker, or by the worker’s attorney. 6 the impact of provider choice on costs and outcomes Victor (2003), and Victor, Barth, and Liu (2003) have shown that in many instances employees actually choose their providers in employer choice states, and employers select workers’ providers in states categorized as employee choice.3 Analyzing the outcomes of cases on the basis of who actually chose the provider, and not simply whether a state law existed to mandate employee or employer choice, is more informative about the impact of provider choice and about policies governing that choice. A second advantage of this study is that we focus on the primary provider. In our interviews, we asked each worker both who selected the initial provider and who selected the “primary medical provider.” If an interviewed worker reported receiving treatment from multiple providers, we asked which one was primary, defined as “the medical professional that made the decisions about the care that the worker needed and either provided that care or directed the worker to someone who could provide it.”4 The third significant advantage of this study is that we link the interview data to claims data supplied by the claims payors, providing information on factors such as medical and indemnity costs, medical treatments, and employer attributes, among others. A more complete picture of the claim from the perspective of both the worker and the employer should help to better establish the consequences of provider choice. Fourth, a potentially important and unique feature of this study is that the interview also indicated whether the primary provider had previously treated the worker for an unrelated condition. Suspecting that a previous provider-patient relationship might affect some of the outcomes that we measured, we gathered the necessary data to test that hypothesis. The distinction between employee choice of a prior provider and employee choice of a new provider has been highlighted in recent workers’ compensation reforms in California (Senate Bill 899) and in legislative proposals in Texas (Senate Bill 5 and House Bill 7). Until recently in California, the employer had the right to select the initial provider, unless the employee had predesignated a provider; after 30 days, the worker had the right to change to a medical provider of his or her choice. However, the most recent reform legislation changed the rules regarding provider choice. In particular, employers are allowed to establish networks composed of 3 This can occur because the law only gives one party the right to choose the provider, which can be ceded to the other party. 4 A description of the survey, the methods used, and the data employed is presented in Victor, Barth, and Liu (2003, Chapter 1). introduction 7 both occupational and nonoccupational physicians, and the legislation grants to the employer (or the insurer) the sole right to decide which medical providers are in the network. Further, the worker’s right to choose his or her physician after 30 days no longer applies if a network is established that complies with the law, unless the worker has predesignated a physician under particular conditions — most importantly, that the physician was previously the worker’s primary provider of medical care under an employer-provided group health plan.5 In general, as long as employers establish networks, which many are expected to do, workers will have less scope to choose their physicians. Most importantly, a worker’s ability to look for a new physician after an injury will be severely curtailed. Thus, our findings on the consequences of employee choice of prior providers, employee choice of new providers, and employer choice are especially salient to assessing the likely consequences of this very important component of workers’ compensation reforms in California — reforms that are just beginning to be implemented and which will likely be questioned and reexamined as the state implements the reforms enacted to deal with the workers’ compensation crisis that emerged in California at the beginning of this decade. Although the study focuses on provider choice, many other factors affect costs and outcomes. A large number of those factors are included in this study, but others are beyond its scope. For example, worker outcomes can be affected by the clinical services delivered but can also be affected by the process of care — patient education, emphasis on functional recovery, release to modified duty — and other disability management practices. Even though these factors are influenced by the provider, our goal in this report is to estimate the effects of provider choice on costs and outcomes. It remains for future work to understand exactly what differences regarding treatment and other decisions made by providers underlie differences in costs and outcomes associated with provider choice. Organization of the Report Chapter 2 provides a description of the key concepts and data sources used in this study, the measures used to categorize the choice decision, and an explanation of the cost and outcome measures utilized. Chapter 2 also explains the empirical methods used to analyze the data and discusses issues regarding the interpreta- 5 For more details, see Neumark (2005). 8 the impact of provider choice on costs and outcomes tion of the estimates resulting from those methods. More detailed information on statistical methods and results is presented for interested readers in a series of technical appendices. In Chapter 3, we examine the results regarding cost and outcome differences between cases in which the employer chose the provider and those in which the employee chose. In Chapter 4, we conduct similar analyses but examine the role of provider choice more finely, distinguishing between cases in which the employee selected a provider who had previously treated the employee, cases in which the employee selected a new provider, and cases in which the employer chose the provider. Chapters 3 and 4 present combined data for the four states. In Chapter 5, we replicate much of the analysis of Chapters 3 and 4 but do so for each of the four states separately. In Chapter 6, we focus on our findings regarding worker reported satisfaction with the medical care that was received. Chapter 7 presents our conclusions and the policy implications that we draw from the findings of the study — in general and with respect to California and Texas in particular. 2 Key Concepts, Data, and Methods This chapter first describes the key concepts and the data used in the study, defines the key cost and outcome variables, and discusses how we constructed certain control variables. Most important, we describe how we measured the provider choice variables. We then provide an explanation of the statistical models used, consider some issues related to the statistical analysis, and discuss the interpretation of the estimates that the statistical methods yield. Key Concepts We begin by defining the terms used frequently in the report. A list of all the variables used in our statistical analyses, along with variable means, are presented in Appendix B. outcomes Substantial return to work: The worker returned to work and remained working for at least one month before any subsequent absence from work. Duration of time out of work: The number of weeks reported by the worker from the time of injury to the time of the first substantial return to work. Recovery of physical health: A measure derived from the SF-12® instrument for 9 10 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s quantifying health status.1 This reflects the worker’s perceptions of the recovery achieved after the injury, not a clinically based measure. For more information on the derivation of this measure, its strengths and limits, and evidence regarding its validity, see Technical Appendix C. Satisfaction with overall health care: The worker’s rating of his or her overall health care on the following scale: very satisfied, somewhat satisfied, somewhat dissatisfied, very dissatisfied. costs Medical payments: Payments per claim to health care providers for medical care. Indemnity benefits: Payments per claim to workers under state statute, partly to replace lost earnings. provider choice Employee choice: The worker reported that the provider was selected by him- or herself, a family member or friend, or the worker’s attorney. Employer choice: The worker reported that the provider was selected by the employer or insurer. Primary provider: If the worker was treated by multiple providers, he or she was asked which one was the primary provider — that is, the medical professional “who made the decisions about the care that you needed and either provided it or directed you to someone who could provide it.” The primary provider is the initial provider in cases with only a single provider. Prior provider: A provider who treated the worker previously for an unrelated condition. New provider: A provider who did not treat the worker previously for an unrelated condition. 1 SF-12® is a registered trademark of the Medical Outcomes Trust. As a standardized measure of health and functioning, the SF-12® has been used and normed extensively since it was developed (see Ware, Keller, and Kosinski, 1998). key concepts, data, and methods 11 other terms Worker: The individual surveyed to provide information for this study. Each worker we interviewed had a compensable claim involving more than seven days of lost work time (for which the worker received at least some payment). Case or claim: All cases in the study involved more than seven days of lost time. Injury severity: A measure derived from the SF-12® instrument for quantifying health status. Like the recovery measure, this reflects the worker’s perceptions of the severity of his or her injury, not a clinically based measure. Overnight hospitalization: A variable that measures whether the worker was admitted for an inpatient stay, judged by whether the worker received “room and board” or “intensive care” based on the hospital service billing (revenue) code. We derive this variable from the medical services data, not the interviews. Major surgery: A variable that measures whether the worker received surgical services. We use the term major surgery to distinguish these services from other medical services that are also commonly referred to as surgical services but are really medical treatments using invasive techniques — like debriding a wound or certain types of injections. We derive this variable from the medical services data, not the interviews. Data Sources One key data source for this report is the Workers Compensation Research Institute (WCRI) Detailed Benchmark/Evaluation (DBE) database, which contains more than 16 million workers’ compensation claims with representative data in at least a dozen large states. The second key data source is the set of telephone interviews conducted on behalf of WCRI by the Center for Survey Research and Analysis at the University of Connecticut as part of a study to compare worker outcomes in California, Massachusetts, Pennsylvania, and Texas for a subset of cases drawn from the WCRI DBE database. Approximately 750 interviews were completed in each state with workers who had experienced more than seven days of lost time from work. The published report from that study describes the sampling methods, response rates, validity of the outcome measures, and any sampling or response bias (Victor, Barth, and Liu, 2003). That report shows that any such biases are at most small. Technical Appendix D provides an abridged discussion of those issues. The telephone interviews supplement the claims data with information on choice of provider — a central factor in the analysis in this report — as well as satisfaction with health care, worker and employer characteristics, re- 12 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s turn to work, and self-reported information on health status from which we derived measures of severity of injury and recovery of physical health. Measuring Provider Choice Cases included. It is useful to understand the structure of the survey questions used to define provider choice. Some workers received care at the workplace, in an ambulance, or at a hospital emergency room. Because provider choice is not an issue in such cases, these workers are excluded from this study, unless they received subsequent treatment from a provider. Of course, we include any worker who received initial treatment at a medical doctor or chiropractor’s office, clinic, hospital, and other such facility. Primary provider. The central focus of this study is the choice of the primary provider — according to the worker, the one who made the decisions about the care that the worker needed and either provided that care or directed the worker to someone who could provide it. We asked respondents about the number of providers who treated them. In cases with only a single nonemergency provider (about 15–25 percent of cases), the initial provider was necessarily the primary provider. Some 75–85 percent of workers received care from more than one provider. Among those workers, the primary provider was also the worker’s initial provider in nearly 60 percent of cases, according to the worker, and was a different provider in about 40 percent of cases (Table 2.1). New or prior provider. We also asked each worker if the provider identified as primary had previously treated the worker for a different condition. If so, we defined the provider as a prior provider. If the provider had not previously treated the worker for a different condition, we labeled that provider as a new provider. Figure 2.1 shows the possible patterns that were followed by workers in the selection and classification of providers, and in the bottom half of Table 2.1 we report the breakdown. Provider specialty. We recognize that a worker can have many health care providers, but for the sake of clarity in interviewing and to keep the survey at a reasonable length, our questions related to those the worker identified as the initial provider and the primary provider. We also recognize that the number of types of specialties involved in treating injured individuals can be very large. We chose not key concepts, data, and methods 13 Table 2.1 Pattern of Providers Who Delivered Nonemergency Care Percentage of Workers Combined CA TX MA Single provider 20.7 15.4 21.5 23.8 Multiple providers 79.3 84.6 78.5 76.2 Among workers who received nonemergency care from multiple providers Initial provider was primary provider 57.9 59.9 53.3 54.8 Initial provider was not primary provider 42.1 40.1 46.7 45.2 PA 22.5 77.5 62.1 37.9 to probe this issue, other than to distinguish among physicians, chiropractors, and physical therapists. Again, one reason for that decision was the length of the survey; a second was that the worker might not know a physician’s precise specialty. Who chose the primary provider? Regardless of the number of providers a worker received treatment from, we asked each respondent to identify who chose the primary provider. For the purposes of this study, if the worker said that he or she chose the provider or that the choice was made by a family member, a friend, or the worker’s attorney, we regarded this as “employee choice.”2 If the worker said that the employer or insurer selected the provider, we categorized it as “employer choice.” If a medical center, medical provider, or “someone else” was seen by the worker to have chosen the provider, we excluded the case from this study because it was ambiguous whether the worker or employer selected the referring medical center or medical provider. The distributions of these choices are displayed in Table 2.2. In Chapter 3, we report results using the two-way classification of provider choice (employer chose or employee chose). In Chapter 4, we report results using the three-way classification of provider choice (employer chose, employee chose a prior provider, or employee chose a new provider). Whether the primary provider 2 We classified the choice of the provider as the employee’s choice if the attorney chose the physician because this strikes us as an accurate characterization of the choice. It is important to emphasize that attorney involvement and attorney choice of provider are not the same thing. Indeed, as Table B.2 in Technical Appendix B shows, in many cases in which the employers chose the physicians, attorneys were involved (18.5 percent of employer choice cases and about 24.0 percent of employee choice cases had attorney involvement, based on our classification). 14 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Figure 2.1 Provider Choice Questions Treated at doctor's office, medical clinic, hospital, or chiropractor's office Treated at workplace, in ambulance, or in hospital emergency room Additional treatment No additional treatment One provider Two or more providers Prior provider New provider Employee chose initial provider Employer chose initial provider Other chose initial provider Employee chose initial provider Initial provider was primary provider Employer chose initial provider Other chose initial provider Initial provider was not primary provider Prior provider New provider Employee chose primary provider Employer Other chose primary chose primary provider provider Two-way classification: employer chose or employee chose Three-way classification: employer chose, employee chose prior provider, or employee chose new provider Prior provider New provider key concepts, data, and methods 15 Table 2.2 Who Chose the Respondent’s Primary Provider? Percentage of Workers Combined CA TX MA Employee chose You (respondent) Family member Friend Your attorney Employer chose Your employer Insurance company Medical professional/ hospital/clinic Someone else Number of cases Number of cases with either employee or employer choice 41.4 36.9 1.9 1.3 1.3 37.5 31.7 5.8 17.7 3.3 2,513 1,960 33.8 52.7 51.0 28.4 46.8 46.3 0.7 2.9 2.8 1.6 2.0 1.4 3.1 1.0 0.5 48.3 27.0 19.4 41.0 21.4 14.4 7.3 5.6 5.0 13.8 16.7 25.1 4.0 3.6 4.5 665 609 542 538 481 376 PA 31.3 29.4 1.5 0.2 0.3 50.7 45.4 5.3 16.3 1.6 697 565 was the initial provider or not, we asked each worker if the primary provider was a new provider or a prior provider (Figure 2.1).3 Overall, from the survey data set of more than 2,500 cases with treatment from providers (that is, cases with more than emergency treatment), we constructed a subset of about 1,960 cases for which we could classify provider choice using both the two-way and three-way classifications.4 Table 2.2 shows the number of cases 3 We did not ask the prior/new question for initial provider of any worker for whom the initial provider was not primary. This was one of many compromises made in the design of the survey to reduce the scope to fit into the time constraints of the interview. We could have used those respondents for the two-way classification, but we did not because the three-way classification is of equal interest. 4 Of the total number of completed interviews (nearly 2,800), 232 were not included in this study because the respondents’ care was limited to emergency treatment only, and 44 others did not answer the provider choice question. Among the remaining group (2,513), 596 were dropped because the worker reported that the provider was chosen by a medical center, another medical professional, “someone else,” or “don’t know/refused to answer.” 16 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 2.3 Provider Choice, by Type of Provider Physician Percentage of Cases Chiropractic Physical Therapy Four states combined Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose California Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose Texas Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose Massachusetts Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose Pennsylvania Employer chose Employee chose Employee chose — prior Employee chose — new Medical center/professional chose 91 89 93 85 89 88 87 88 86 84 93 81 92 75 90 88 93 93 94 92 93 95 97 93 89 34 82 61 10 3 37 35 10 1 10 1 11 2 3 11 25 15 2 71 20 3 45 32 42 51 42 34 33 23 30 16 19 Other 3 1 1 1 1 4 2 1 2 2 0 1 0 2 0 6 1 1 0 1 2 0 0 0 1 key concepts, data, and methods 17 by state. The table also reveals striking similarities in patterns of provider choice in California and Pennsylvania on the one hand and in Texas and Massachusetts on the other, but differences also appear between the two sets of states. In Texas and Massachusetts, the law in effect at the time of the study gives the worker the choice of initial provider and relatively free reign to change providers. In California and Pennsylvania, the effective law allows the employer to designate the provider for the first 30 days and 90 days, respectively, after which the worker can change providers.5 The influence of these policies is reflected in the higher incidence of employer choice in the latter two states and, conversely, the higher incidence of employee choice in Texas and Massachusetts. Choice of the primary provider, by type of provider. Regardless of who chose the provider, the overwhelming majority of workers reported that a physician was the primary provider — the provider that made the decisions about the care that the worker needed and either provided it or directed the worker to some who could provide it. As shown in Table 2.3, in all four states combined, employers infrequently chose chiropractors as primary providers — only 2–3 percent of cases in each of the four states. In Pennsylvania and Massachusetts, workers selected chiropractors as their primary providers in about 3 percent of cases — similar to when employers chose the providers. In California and Texas, however, workers selected chiropractors as primary providers much more often; in those two states, when workers chose their primary providers, they selected chiropractors at least 10 percent of the time.6 In California, workers were equally likely to select chiropractors who had previously treated them or a new chiropractor. In Texas, workers were much more likely to see chiropractors who had never previously treated them. Sometimes workers identified physical therapists as their primary providers. This occurred most often when the employer chose the provider or when the worker said that a medical center or a medical provider selected the primary provider. The latter is not surprising, because physical therapy typically requires a doctor’s referral. 5 There are some exceptions to those rules. For example, in some states, when the employer has established an approved network, the worker must select within the network. See Tanabe and Murray (2001). 6 Of course, one factor that likely affects the frequency of use of chiropractors is their supply (although this is also driven by demand). Large differences exist across states in the numbers of chiropractors per 100,000 population. As of 1995, the rates of chiropractors per 100,000 persons were 33.2 in California, 26.8 in Texas, 21.7 in Pennsylvania, and 20.3 in Massachusetts (Coulter and Shekelle, 1997). 18 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Cost and Outcome Measures We study most of the key outcomes of workplace injuries that should be of interest to policymakers: costs, return to work, recovery of physical health, and satisfaction with care. The two cost measures that we study are indemnity benefits and medical payments per claim. Both measures are derived from payors’ records about what payments were actually made as of 29 to 31 months after the injury. The WCRI DBE database standardizes definitions of those measures across payors and across states. Table 2.4 shows average indemnity benefit payments and medical payments per claim for each state. Medical payments per claim were significantly higher in California and Texas than in Massachusetts and Pennsylvania. Indemnity benefits per claim were higher in California than in the other three states. Another important outcome is the extent to which the worker recovered his or her physical health after the injury. The measure we use for this study is derived from worker responses to the most widely used instrument for measuring general health status — the SF-12® survey. The focus is on physical health, not mental health. Because the SF-12® scores for physical health are quite insensitive to even extreme variations in the mental health scores, we compute the physical health scores holding the mental health scores constant.7 In the interviews, we asked workers to recall their health status at three points in time — the month before the injury, the week after the injury, and the month before the interview. The recovery variable is the difference between the worker’s self-reported health status after the injury and the same measure at the time of interview.8 Because this measure is based on workers’ perceptions, we refer to this 7 This approach and related sensitivity analysis is discussed in Victor, Barth, and Liu (2003) and in an abridged form in Technical Appendix C. 8 The recovery measure we use is the change from one week after the injury to the interview; in most cases, this change is positive, but that is not imposed on the data since a worker’s health could worsen. In addition, the severity control we use in the regression models that follow is similarly defined as a change in levels — in this case, from before the injury to one week after. Again, we do not impose that the worker’s health had to worsen, although it did in almost every case. We also experimented with specifications defining each of these variables as relative measures — that is, we defined the percentage recovery relative to health status one week after the injury and the percentage severity relative to health status before the injury. The results were very similar. We have some preference for the specification with changes in levels, because we do not think a full recovery from a very minor injury should be treated symmetrically to a full recovery from a very serious injury. Put another way, we think it is important that the regression estimates of effects of provider choice on recovery reflect a large “penalty” for serious injuries that are not followed by substantial recoveries, even if they are also associated with near-complete recoveries for very minor injuries. key concepts, data, and methods 19 Table 2.4 Costs and Health Outcomes, by State Combined CA Average medical payment per claim Average indemnity benefit per claim Average recovery scorea Average severity scorea $8,713 $12,709 19.2 29.0 $9,950 $15,444 17.6 29.0 TX $11,729 $10,188 15.0 28.2 MA $4,946 $13,874 24.1 29.8 PA $7,594 $11,358 21.0 29.0 a Respondents’ SF-12® scores are scaled scores from 0 to 100, where 100 is the best health. The recovery score is the difference between the SF-12® value at the time of interview and the score one week after injury. The severity score is the difference between the score for the time four weeks before the injury occurred and the score one week after the injury. The mean value of the preinjury scores for respondents was about 54 or 55, depending on the state. variable as “perceived recovery.”9 Table 2.4 shows that the average perceived injury severity was very similar across all three states but that the average perceived recovery was better in Massachusetts and Pennsylvania. We also study whether the worker returned to work for at least one continuous month at any time between the injury and the interview. We call this a “substantial return to work.” In addition, we measure the duration of time out of work, as reported by the worker as of the date of the interview — approximately 3.0 to 3.5 years postinjury. Recall that all cases sampled had more than seven days of lost time. Table 2.5 shows the mean and median durations reported by the workers and the percentages not reporting substantial returns to work. Of course, both measures are closely related to indemnity benefits but not perfectly correlated because some workers receive permanent disability benefits that are not related to actual wage loss and others receive lump-sum settlements that terminate the employers’ liability regardless of whether the workers remain out of work. 9 Victor, Barth, and Liu (2003) provide extensive discussion of potential concerns about recall bias and other limitations, as well as evidence of validity of the health status measure from which both recovery and injury severity are derived. Technical Appendix C in the current report summarizes that discussion. 20 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 2.5 Substantial Return to Work and Duration of Period Out of Work Combined CA TX MA Percentage of workers who did not have substantial returns to work 19 Among workers who had substantial returns to work Average duration of time out of work (weeks) 13 Median duration of time out of work (weeks) 6 19 27 18 14 14 14 868 PA 13 11 6 Note: Only workers who had substantial returns to work were asked, “How many weeks was it from the time you first stopped working because of your injury and the first time that you returned to work for one full month?” Only cases in which the employee or employer chose the primary provider are included. We also measure satisfaction with care. In the survey, we asked a widely used set of questions about satisfaction with the timeliness of care, the provider, and overall care. The variable used in this study examines satisfaction with overall care. The specific question was “Now think about all of the medical care you received from the first treatment for your injury until now. Were you satisfied or dissatisfied with the medical care you received overall?” Table 2.6 shows the distribution of responses among the four choices offered. Empirical Methods This section sketches out the statistical methods and models used to assess the impact on outcomes related to provider choice. More details appear in Technical Appendix E. In subsequent chapters, we attempt to describe the findings from our analyses in a way that does not require the reader to have mastery of these models. Nontechnical readers should feel free to skip to Chapter 3. basic model and control variables The empirical approach uses statistical models to estimate the impact of provider choice on a variety of workers’ compensation costs and outcomes. We study key concepts, data, and methods 21 Table 2.6 Satisfaction with Overall Health Care Percentage of Workers Combined CA TX MA Very or somewhat satisfied 82 80 80 85 Very satisfied 52 47 51 56 Somewhat satisfied 29 33 29 29 Very or somewhat dissatisfied 18 20 20 15 Somewhat dissatisfied 8 10 9 6 Very dissatisfied 10 10 11 8 PA 83 57 26 17 8 9 Note: Only cases in which the employee or employer chose the primary provider are included. provider choice using a two-way classification (employee versus employer choice) and a three-way classification (employee choice of new provider versus employee choice of prior provider versus employer choice). The statistical models control for other influences on workers’ compensation costs and outcomes to avoid attributing the other influences to the impact of provider choice. We also take a number of steps, discussed in some detail in this chapter, to account for possible difficulties in measuring the severity of workplace injuries that might lead to incorrect conclusions. More formally, the framework is based on a standard regression model for a cost or outcome variable generically denoted Yis, where i indexes individuals and s indexes states, of the form: Yis = α + CHOICEis β + WORKERisγ + FIRMisδ + INJURYisθ + STATEsκ + TREATMENTis λ + εis. (2.1) As indicated in the preceding section, our dependent or outcome variables come in different forms: continuous (for example, the cost measures), in two categories (substantial return to work), and in more than two categories (satisfaction). We therefore have to use different statistical methods for different dependent variables, as discussed in more detail later in this section and especially in Technical Appendix E. The provider choice variables, which may be one dummy vari- 22 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s able corresponding to the two-way classification or two dummy variables corresponding to the three-way classification, are included in the vector CHOICE. In any model of workers’ compensation costs or outcomes, it is essential to include characteristics of workers (WORKER) and the workplace (FIRM), because both types of characteristics have been shown to affect costs and outcomes. For example, older workers have been found to be less likely to return to work; workers with less education are likely to have greater difficulty in the labor market; and workers employed by firms in certain industries (such as construction) may have unique return-to-work problems to surmount (see, for example, Galizzi and Boden, 1996). The list of variables included in WORKER includes demographics, education, wages and whether the individual was an hourly worker, tenure at the time of injury, and whether the worker elected to have the interview conducted in Spanish. Workplace characteristics include firm size and an industry breakdown, details of which are provided in the tables discussed later. Naturally, we would expect costs, return to work, recovery, and satisfaction to depend in important ways on the characteristics of the injury. We have several types of measures. The first is a classification of injury type: back pain; nonback sprain or strain; fracture; inflammation, laceration, or contusion; and a residual category of other injuries, based on the diagnostic (ICD-9) codes assigned by the providers.10 A second measure captures the worker’s perceived injury severity. This measure is constructed from the worker’s answers to the SF-12® instrument, paralleling what we did for the measures of perceived recovery (see the earlier discussion). Specifically, we measure injury severity by the difference in workers’ responses to the SF-12® questions regarding their health status between the month before the injury and the week after the injury. The construction and validity of this measure is discussed in Technical Appendix C. The inclusion of the worker, workplace, and injury characteristics in a model of how provider choice affects outcomes is unambiguous because those variables can be associated with both provider choice and the costs and outcomes we study, but not for reasons underlying any causal relationship between provider choice and outcomes. For example, compared with other workers, older males may have 10 In some cases, workers are assigned multiple diagnosis codes during the course of their disability. In such cases, we define a primary diagnosis code based on the code that receives the greatest expenditure. Also, in some cases, diagnosis codes are missing in the database. In these cases, we use information from the payor about the nature of injury and part of body to assign the case to the appropriate injury group. key concepts, data, and methods 23 worse medical outcomes because age inhibits recovery. Yet older males may also — because of greater affluence, access to health insurance, and possibly even previous injuries — be most likely to have chosen primary providers whom they have seen previously. Without controlling for age and sex in such a case, we might incorrectly infer that the older worker’s choice of a prior provider resulted in or caused worse medical outcomes. Similarly, more-severe injuries may make it more likely, at least in some states, that the employee chose the provider, simply because in California, for example, the employee during the sample period had the right to choose a physician 30 days after first receiving treatment, and more-severe injuries are more likely to pass the 30-day window. variation in state workers’ compensation systems However, the worker, workplace, and injury characteristics may not be sufficient as controls. First, as noted earlier, our data come from four states — California, Texas, Massachusetts, and Pennsylvania. Workers’ compensation systems vary widely across all states, including the four in our study. For example, the study states differ markedly on matters such as the frequency and sources of disputes, the methods used to terminate temporary disability benefits, the criteria used to rate permanent disability benefits, the use of networks to provide medical care, and so on. We know that outcomes related to cost, return to work, and other measures vary substantially across states (Telles, Wang, and Tanabe, 2004). We also know, as discussed earlier, that the states differ regarding the prevalence of employee and employer choice: California and Pennsylvania are states where initially the employer has the right to choose the provider, while Massachusetts and Texas are considered to be employee choice states. Technical Appendix F provides an overview of some of the more important interstate differences that are relevant for this study. Given these facts, if we use across-state variation in choice and outcomes to identify β in equation (2.1), we may incorrectly attribute differences in outcomes associated with other features of the states’ workers’ compensation systems to variation in individual choice of provider. Consequently, we report all specifications, including dummy variables for the states (STATE), in which case the effects of provider choice are identified solely from within-state differences associated with this choice. The potential downside of this is that we effectively throw out the variation in provider choice that is driven by differences in state workers’ compensation systems, which is plausibly the most exogenous source of variation in 24 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s provider choice.11 We examined estimates both excluding and including the state dummy variables to see whether we could find results that are robust to this specification choice and therefore could draw firmer conclusions. In general, we found that results including or excluding the state dummy variables were similar. Because we think it most important to control for omitted variation in state workers’ compensation systems, we report estimates based on specifications that include the state dummy variables. appropriate statistical models for each dependent variable We need to use different statistical models because our dependent variables take different forms. For the three costs and outcomes variables that are continuous (indemnity benefits, medical payments, and recovery of physical health), equation (2.1) is estimated as a linear regression, in which case the estimated coefficient of a variable simply measures how the outcome changes with a one-unit increase in the variable. For the other outcomes variables, we cannot use linear regression. In each case, there is some choice regarding exactly which type of model to use. We have chosen to use a set of models for which the estimated coefficients have a very similar interpretation. The return-to-work outcome is dichotomous; either the person returned to substantial employment or not, and we estimate this using a logit model. The model for the duration of time before a substantial return to work occurred has to be estimated using survival models, to account for the possible truncation of the out-of-work period. That is, it is possible that at the time of the survey, some individuals have still not returned to substantial employment. If so, all we know is that the out-of-work period lasted at least up to the time of the survey. Finally, the satisfaction outcome is also discrete (like return to work), but takes on one of four values: very satisfied, somewhat satisfied, somewhat dissatisfied, and very dissatisfied. Further, these values are ordered, given that the satisfaction responses can be ranked clearly. To study this outcome, we use an ordered logit model. 11 Looking at the issue in a different way, provider choice is correlated with unobservables in equation (2.1) — for example, unmeasured variation in injuries. If we thought that state dummy variables could be excluded from the generic model given by equation (2.1), and we also thought that state of residence affected choice because of features of states’ workers’ compensation systems, then the state dummy variables would provide a natural set of instrumental variables for provider choice. However, given that other features of state workers’ compensation systems likely affect outcomes, the exclusion restriction is invalid, precluding this instrumental variables strategy. key concepts, data, and methods 25 equality of effects of provider choice across states Another issue is whether we can combine, or pool, the data across the four states to obtain the most precise estimates of the impact of provider choice. Given that we have only about 400 to 550 observations per state, this pooling is highly desirable. However, it could be inappropriate and lead to biased estimates if the effects of provider choice on the outcomes we study vary significantly across states. We tested for this and did not find evidence against the restrictions implied in combining the data and estimating a common set of effects of provider choice.12 We emphasize the combined estimates in the chapters that follow, although for completeness we also report results for the states individually (in Chapter 5). causal inferences about policy changes regarding provider choice The central goal of this report is to estimate the effect of provider choice on workers’ compensation costs and worker outcomes. The evidence we report is of interest from a policy perspective to the extent that it helps assess the effects of policy choices regarding provider choice, such as California’s most recent workers’ compensation reforms that restrict employee choice of a new physician. Does the evidence that we assemble here speak to the effects of policy changes, and if so, under what conditions? In an ideal world, to estimate the effects of policy changes, we would randomly assign injured workers to different provider choice “regimes” (for example, with choice assigned to either the worker or the employer) and then observe the outcomes. Because workers would be randomly assigned, in a large sample we could rule out any differences across workers in the different choice regimes as alternative explanations of differences in outcomes associated with provider choice. That is, we could confidently draw causal inferences about the effects of provider choice 12 To test this, for each analysis we conduct we also test for differences in the parameters describing the effects of provider choice, as well as the coefficients of the other variables in the model. We do this by interacting each of these variables with the state dummy variables, estimating these full models, and then separately testing the constraints that the provider choice coefficients are the same across states, and that the other coefficients are the same across states. We never reject the first set of restrictions; we sometimes reject the latter, but we verify that the provider choice estimates are insensitive to allowing the effects of the other control variables to differ across states. A full description of the testing and results is presented in Technical Appendix G. 26 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s that would be informative about the effects of policy changes. Of course the real world falls short of this experimental idea. We instead observe injured workers seeing different providers — some chosen by them, whether prior or new providers, and some chosen by employers — and we observe the outcomes. But we have to be concerned that there are characteristics of workers associated with both provider choice and with workers’ compensation outcomes that result in misleading inferences about the effects of provider choice. As one concrete example, to which we will return later, suppose that the most severe injuries tend to result in workers ending up with new providers, chosen by them, as their primary providers. This might occur because in search of solutions for the most severe injuries, workers are motivated to seek out particular providers (such as specialists). In this scenario, if we simply compared outcomes such as costs and time away from work across these workers and workers for whom the employers chose the providers, we would find that for the former group, costs were higher and return-to-work outcomes worse. We could then be led to the incorrect conclusion that worker choice of a new provider causes higher costs and worse return to work, whereas the relationship arises only because the most severely injured workers selected into the employee choice/new provider group. We address this potential problem in a few ways. Most important, as explained earlier, for all our analyses we include controls for numerous detailed characteristics of workers, workplace characteristics, and injury characteristics. Indeed, we would argue that the data used in this report yield far more detailed sets of control variables than past work. (The full list of control variables is listed in Table 2.7.) The relation between these control variables, including the severity controls, are not necessarily causal, but where statistically significant, are at least associative. Second, in Table 2.7, we report estimates of models for provider choice, to explore which variables are in fact associated with choice. In these models, we report odds ratios for the employee choice options relative to employer choice.13 A coefficient estimate greater than 1, when statistically significantly different from 1, implies that the variable associated with that coefficient boosts the likelihood of employee choice.14 13 Results by state are reported in Technical Appendix H. 14 Note that throughout the report, when we refer to estimated odds ratios, we simply refer to “statistically significant” as a shorthand for “statistically significantly different from 1,” which is equivalent to the statement that the difference between the estimate and 1 is statistically significantly different from zero. key concepts, data, and methods 27 Notes on Statistical Significance: ■ In all tables reporting results from our statistical models, we note with symbols whether or not the estimates are statistically significantly different from zero. We report the level of statistical significance for each estimate, focusing on statistical significance at the 5 percent and 10 percent levels, which are indicated with two asterisks (**) and one (*), respectively. Statistical significance is important because all estimates to some extent reflect the randomness of which individuals were included in the sample. Statistical significance at the 5 percent level, for example, means that there is a 95 percent chance that the estimated effect is different from zero; conversely, the chance is 5 percent (or 1 in 20) chance that an estimated effect looks different from zero when in fact the true effect is zero. The lower the level of statistical significance (5 percent versus 10 percent), the greater the confidence we have that an estimated effect is in fact different from zero. ■ Estimates significant at the 20 percent level are often not noted in other work, because the chance is 20 percent (or 1 in 5) that the estimated effect is in fact zero. However, we report these results to distinguish effects that are still considerably more likely to be different from zero than not — sometimes termed “marginally significant.” These estimates are indicated with a dagger (†). ■ It is important to interpret carefully results that are classified as “not statistically significant” or, more simply, “not significant.” An estimated effect that is not significant is not the same as concluding that the effect is zero. Under the assumptions of the statistical models we use, the estimate we obtain is the best estimate of the effect. However, the estimate cannot be established as statistically significant, meaning that there is a reasonable probability that the nonzero estimate could have been obtained even if the true effect is zero. It turns out that quite a few variables are significantly related to provider choice — which would not be the case with random assignment of choice. For example, older workers are significantly more likely to choose their own providers in the two-way model, and their own prior providers in the three-way model. We also found that choosing a prior provider is positively correlated with firm size, and there are some interindustry differences in choice, probably both reflective of dif- 28 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s ferences in health insurance coverage. We found that persons with the lowest levels of educational attainment and those interviewed in Spanish are less likely to choose their own providers or prior providers. We have no reason to believe that single males or persons with lower education or Spanish speakers are more likely to have more-severe injuries, conditional on the controls. However, we found they are more likely not to know or to have prior providers because of a lack of health insurance. More significantly, certain types of injuries, especially back injuries, are significantly more likely to be associated with employee choice of provider than is the reference category of inflammation, laceration, or contusion. On the other hand, it is interesting to note that perceived severity is not associated with employee provider choice, although it must be remembered that because the model includes such variables as type of injury, what is captured by the estimated coefficient of severity is the effect of variation in severity for the same type of injury. Note also that major surgery is significantly positively associated with employee choice, although as discussed later in this section, surgery may to some extent be an outcome of employee choice rather than a measure of the severity of the injury. The estimates in Table 2.7 certainly indicate that the assignment of workers and their injuries to provider choice regimes is not random, which is no surprise. What the estimates cannot tell us, however, is whether the inclusion of all the control variables listed in Table 2.7 in our models for workers’ compensation costs and outcomes capture enough of the variation in other determinants of these costs and outcomes that we are confident that the regression models capture the causal effects of provider choice, or instead whether there is still residual unmeasured variation in severity of injury or other factors that is related to provider choice. However, the fact that greater severity is not independently associated with a higher likelihood of employee choice makes it more plausible that we are estimating causal effects of provider choice. Our third approach to obtaining estimates that provide evidence on the causal effects of provider choice involves including additional control variables related to severity. In particular, the claims database includes information on the treatment of the injury, including whether the treatment included an overnight hospitalization and major surgery; these are captured in the variable TREATMENT in equation (2.1). These potential control variables present a double-edged sword. On the plus side, they are likely to capture additional variation in the severity of the injury that is not picked up in the other variables that capture nature and severity of injury. For example, some fractures, even if viewed by the respondent as entailing the same severity, may result in overnight hospitalization for a variety of reasons related to the injury, and therefore we would expect higher medical payments. On key concepts, data, and methods 29 Table 2.7 Determinants of Provider Choice, Four States Combined Two-Way Provider Choice Classificationa Employee vs. Employer Three-Way Provider Choice Classificationb Employee/Prior vs. Employer Employee/New vs. Employer State Pennsylvania California Texas Massachusetts Worker characteristics Age Male Married Wage Hourly worker Tenure Less than high school Some high school High school graduate Some college College graduate Postgraduate Survey in Spanish Workplace characteristics Firm size,≤50 Firm size, 51–250 Firm size, 251–1,000 Firm size,>1,000+ High-risk services Low-risk services Clerical/professional services — 1.140 3.759** 4.507** 1.007† 1.085 1.039 0.972† 0.416** 1.015** 0.582† 0.814 — 0.856 0.882 1.211 0.329** — 0.904 0.887 1.109 0.709† 1.022 — — 1.236 3.160** 5.498** 1.011* 0.942 1.117 0.973 0.411** 1.017** 0.434* 0.848 — 0.868 0.790 1.328 0.193** — 0.926 0.843 1.590** 0.548** 0.886 — — 1.033 4.226** 3.777** 1.006 1.226† 0.979 0.972 0.409** 1.012† 0.690 0.807 — 0.848 0.996 1.089 0.440** — 0.893 0.952 0.773 0.899 1.223 — 30 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 2.7 Determinants of Provider Choice, Four States Combined (continued) Two-Way Provider Choice Classificationa Employee vs. Employer Three-Way Provider Choice Classificationb Employee/Prior vs. Employer Employee/New vs. Employer Manufacturing Construction Trade Other industries Injury characteristics Back pain Nonback sprain or strain Fracture Inflammation, laceration, or contusion Other injuries Severity Treatment characteristics Overnight hospitalization Major surgery Attorney involvement N 0.562** 0.802 0.944 0.749 1.614** 1.148 1.394† — 1.536* 0.996 0.980 1.376** 1.553** 1,960 0.436** 0.520* 0.803 0.644 1.509† 1.095 1.341 — 1.669* 0.991* 1.134 1.462** 1.506** 1,951 0.728 1.153 1.128 0.894 1.658** 1.148 1.406 — 1.416 1.002 0.842 1.329* 1.592** 1,951 a Odds ratios from logit model are shown, relative to employer choice. The odds ratio measures the effect of the variable on the probability of the type of employee choice indicated in the column, relative to the probability of employer choice. b Odds ratios from multinomial logit model are shown, relative to employer choice. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. key concepts, data, and methods 31 the minus side, the treatment variables also reflect outcomes of the medical decision-making process and thus to some extent directly reflect the choice of provider. Because the treatment variables in part capture costs and outcomes, their inclusion may amount to what is often referred to as “overcontrolling” for injury severity.15 That is, they may capture not only remaining differences in severity but also outcomes of provider choice that we more appropriately want to think about as effects of provider choice but will not capture when the treatment variables are included.16 Under this interpretation, excluding the treatment variables runs the risk of having important unmeasured heterogeneity in injuries, which if associated with provider choice may lead to choice-related differences in costs and outcomes that are too large. On the other hand, including the treatment variables is likely to generate estimates that understate the differences associated with provider choice. In this case, the truth would lie somewhere in between the estimates including and excluding the treatment variables. Consequently, we present both sets of estimates to determine the outcomes for which the resulting range of estimates is tight enough to provide information on the effects of provider choice. When the estimates differ, readers more concerned that our injury and severity measures leave potentially important differences in severity unmeasured may be more inclined to emphasize the estimates that include the treatment variables, and vice versa. Finally, a fourth approach we take to the problem of unmeasured severity is to assess how sensitive the estimates are to omitting from the model variables measuring severity or the nature of the injury. If the estimates are not very sensitive, this suggests that additional unmeasured variation in severity when these variables are included cannot play much of a role. Of course, even with all these efforts, we cannot definitively rule out the possibility that even with the treatment variables included, there is unmeasured variation in injury severity that might affect, for example, costs or return to work. 15 We consider including these variables, but not attorney involvement, because hospitalization and surgery are sometimes likely to be dictated by medical exigencies. At the same time, we recognize that it is possible that attorney involvement exacerbates the effects of employee choice of provider. This raises very interesting questions about how costs and outcomes — and their relationship to provider choice — might change were policies relating to use of attorneys in workers’ compensation cases altered. That question, however, is well beyond the scope of this study. 16 To see the problem of overcontrolling in an extreme form, if we simply put the dependent variable on the right side of the equation, we would obtain a perfect fit, and nothing else, including provider choice, would explain variation in outcomes. 32 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s This implies that, ultimately, we cannot arrive at an definitive answer regarding the causal effects of provider choice from these data, because at its core, that is a question about differences between workers that we cannot measure, in contrast to those we can measure. In our view, the extensive set of control variables that we have, coupled with the results from the various analyses just described, allows us to be reasonably confident that we are identifying the causal effects of provider choice. This means that it is appropriate to think of our estimates as indicating what would happen if policies regarding provider choice were changed, for example, to restrict employee choice. We believe this is particularly true of the specifications that we regard as likely overcontrolling for injury characteristics by including the hospitalization and surgery variables. At the same time, we recognize that our evidence falls short of experimental standards, which of course leaves open the possibility that experimental evidence could lead to different conclusions. 3 Impact of Employee or Employer Choice of Provider: Main Results In this chapter, we examine the impact of the actual choice of primary provider (regardless of who the state law permits to choose) on costs and outcomes. Recall that by primary provider, we mean the provider that the worker said was the medical professional who made the decisions about the care that he or she needed, whether the care was provided directly or the provider directed the worker to someone who provided the care. We examine the impact on medical and indemnity costs, which are the focus of much policy discussion. However, policymakers need to know about not only the cost differences associated with provider choice but also the effects of provider choice on outcomes of care that are associated with quality, including return to work, duration of lost time, recovery of health, and patient satisfaction. Certainly, cost reductions associated with employer choice would be viewed less positively if they were accompanied by reductions in the quality of care. We also recognize that these outcomes are influenced by factors besides the quality of health care. Here and in Chapter 4, we present the main results of the study — relying on the combined sample pooling the data from all four states, while controlling for important differences among the states in policies and practices that affect the outcomes that we study. In Chapter 5, we discuss the results estimated separately for each state. We emphasize the findings from the combined sample because with the larger sample size, the results are more precise and statistically powerful. As discussed in Chapter 2 and described in detail in Technical Appendix G, our tests 33 34 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s do not indicate statistically significant differences in the effects of provider choice across states, indicating that the best approach is to combine the data from the four states and rely on the more precise estimates that result. We examine here the most common way of characterizing provider choice — employee choice compared with employer choice. This is the standard way of thinking about provider choice in existing research and in much of the policy debate. Chapter 4 examines a three-way classification of provider choice, presenting comparisons of the differences in costs and outcomes for cases in which the employee chose a provider who previously treated the worker for an unrelated condition, in which the employee chose a provider who is new to the worker, and in which the employer chose the provider. Given that state workers’ compensation policies can influence not only whether employees choose their provider but also under what conditions — in particular, whether that provider must be a predesignated provider who previously treated the worker — it is important to understand the findings presented here and in Chapter 4 to derive the full implications of our analysis for public policy. Summary of Findings When workers rather than employers selected their primary providers, on average, costs were higher and return-to-work outcomes were poorer. Physical recovery was unaffected, although workers were more satisfied with their health care. More specifically, compared with cases in which the employer chose the provider, cases in which the employee selected the provider had: ■ Medical payments that were 10–21 percent higher and indemnity benefits that were 8–15 percent higher (although only the higher estimate for indemnity benefits is statistically significant). ■ Odds of returning to work and remaining at work for at least one continuous month that were 16–19 percent lower (although the evidence is statistically weaker than the other findings).1 ■ Time out of work that was 23–32 percent longer. ■ Very similar reported recoveries of physical health. 1 The outcomes for events that do or do not occur are often summarized this way. Formally, the statement that the “odds” are, for example, 10 percent higher means that the relative probability of the event is 1.10. impact of employee or employer choice of provider 35 ■ Likelihood of reporting a higher level of satisfaction with their care that was 57–59 percent higher. Impact of Employee or Employer Choice of Provider on Costs and Outcomes: Main Results This section examines the results using the combined data for the four states while controlling for material differences among the states and cases. We report the results for two specifications: the first might undercontrol for severity by excluding the treatment variables, the second likely overcontrols for severity by including them. The true results probably lie somewhere between the reported results for the two models. When the results from the two models are similar, the reader should have greater confidence in the conclusions; and when the results are significant, even when they include the hospitalization and surgery controls, it seems particularly unlikely that associations between provider choice and outcomes are in fact attributable to unmeasured variation in injury severity rather than the effects of provider choice. We found that cases in which the worker rather than the employer selected the primary provider were associated with higher costs and poorer return-to-work outcomes, with no differences in physical recovery but with higher satisfaction, compared with cases in which the employer selected the provider. As Table 3.1 shows, medical payments were 10–21 percent higher, significant in both cases.2 Lower health care costs when the employer selects the provider can occur for several reasons. Employers might refer injured workers through their medical networks, in which providers have been preselected on the basis of quality or price discounts through negotiations between networks and providers. These arrangements can provide certain advantages in terms of cost, access, or adherence to recommended treatment protocols. The results for indemnity costs also suggest higher costs when the worker chooses the provider, although the evidence is weaker. In particular, the estimates from model 1 — which excludes the hospitalization and surgery controls — indicate that indemnity benefits were 15 percent higher when the worker chose the 2 Note that in the tables in the main text, we report only the effects of provider choice on the outcomes. Full regression results for the combined sample of all states, for the model with the two-way and three-way classifications of provider choice, are given in Technical Appendix H. 36 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 3.1 Impact of Employee Choice Compared with Employer Choice Model 1: Without Treatment Model 2: With Treatment Controls (percent) Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 21** ($1,868) 15* ($1,908) 32** –19† 0 57** 10* ($903) 8 ($978) 23** –16† 1 59** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. In model 1, for example, when the worker chose the provider, medical payments were $1,868 greater per case (on average) than when the employer chose the provider; or the duration of time out of work was 32 percent longer when the worker chose the provider than when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Table 2.2 provides the breakdown of observations by state. Observations are weighted to be representative of claims within each state; see the discussion on weighting in Technical Appendix D. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. provider, while the difference falls by nearly half and becomes statistically insignificant in model 2 — which includes the controls. The results consistently indicate that employee choice of provider is associated with slower return to work. Reported time from injury until initial substantial return to employment was 23–32 percent longer when the employee chose, and substantial return to work was 16–19 percent less likely in the three years after the injury, although the latter results are only marginally significant. (Recall that a substantial return to work is one that lasts for at least one month without having to stop working again due to the injury.) Note that for the return-to-work variables, the range of estimates for models 1 and 2 is tighter, and the statistical significance of the results is not weakened by including the hospitalization and surgery controls, bolstering our confidence in these results and in a causal interpretation of the effect of provider choice. Interestingly, despite the differences in costs and time out of work, there was no impact of employee or employer choice of provider Table 3.2 Satisfaction with Overall Care, by Who Selected the Provider and by State California Employee Chose (percent) Employer Chose (percent) Texas Employee Chose (percent) Employer Chose (percent) Massachusetts Employee Chose (percent) Employer Chose (percent) Pennsylvania Employee Chose (percent) Employer Chose (percent) Very or somewhat satisfied 84 78 80 82 86 83 87 80 Very satisfied 55 41 57 39 60 46 65 51 Somewhat satisfied 29 37 23 43 26 37 22 29 Very or somewhat dissatisfied 17 22 21 17 14 17 13 20 Somewhat dissatisfied 9 10 9 9 7 6 8 8 Very dissatisfied 8 12 12 8 7 11 5 12 37 38 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s difference in the perceived recovery of physical health between workers who selected their providers and workers whose employers selected the providers; any estimated differences in recovery were trivially small and are thus not statistically significant. Yet workers who chose their providers were much more likely to be more satisfied with their overall medical care, with nearly 60 percent higher odds of reporting a higher level of satisfaction. In interpreting the impact on satisfaction, it is useful to note that a large share of workers reported being very or somewhat satisfied with their overall care regardless of who chose their providers. This is true overall and in each of the four states (Table 3.2). Though this study is unique in evaluating satisfaction with care by the choice of the provider, the ranges reported of those who were satisfied (very and somewhat) are very much in line with other studies, both in workers’ compensation and in general health care.3 Chapter 6 discusses possible reasons why we found higher levels of satisfaction with health care when workers selected their providers despite no difference in recovery, and that chapter attempts to untangle the question of whether the higher satisfaction reflects other dimensions of the quality of medical care. 3 See Victor, Barth, and Liu (2003, pp. 115–117). For example, a national study reported that 55 percent of persons receiving medical treatment in the two years before the survey were “extremely or very satisfied” with the quality of the medical care they received (Employee Benefit Research Institute, Consumer Health Education Council, and Matthew Greenwald and Associates, 2002). 4 Employee Choice of Prior Provider or New Provider, or Employer Choice of Provider: Main Results Some workers we interviewed for this study had established relationships with providers before they were injured; others did not. This chapter examines whether the costs and outcomes differ among otherwise similar cases in which (1) the worker selected as the primary provider a “prior provider,” defined as someone who treated the worker before the injury for an unrelated condition; (2) the worker selected as primary provider a “new provider,” defined as someone who had not previously treated the worker; and (3) the employer selected the primary provider. In our view, the results from this chapter are more informative than those from the previous chapter from the perspective of assessing the implications of provider choice for public policy, because public policy can restrict only one type of employee choice of provider — such as California’s recent workers’ compensation reforms that restrict the worker’s ability to choose a new provider. We present estimates that first compare the costs and outcomes of cases with an employee-selected prior provider with those with an employer-selected provider and then compare the costs and outcomes of cases with an employeeselected new provider with those with an employer-selected provider. In addition, we present comparisons of the costs and outcomes of cases with employee-selected prior and employee-selected new providers. 39 40 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 4.1 Employee Choice of Prior Provider or New Provider Percentage of Workers (of those who chose the primary provider) Combined CA TX MA PA Prior provider New provider 45.5 54.5 50.3 36.3 52.4 45.2 49.7 63.7 47.6 54.8 Patterns of Choosing New and Prior Providers Among the cases in which the worker chose the primary provider, a prior provider was chosen about half the time in California, Massachusetts, and Pennsylvania but only about one-third of the time in Texas (Table 4.1). We cannot be certain why Texas workers were less likely to select prior providers as their primary providers. However, it is reasonable to suppose that injured workers who are not covered by health insurance are less likely to have established relationships with health care providers. We do not know from our survey whether injured workers had health insurance coverage. However, we do know that the population in Texas is much less likely to have health insurance coverage than are persons in the other three states.1 Additionally, a recent survey found that injured workers in Texas were less likely to have received general medical care recently than were injured workers in other states, consistent with fewer being covered by nonoccupational health insurance.2 Another possible reason for the relatively low rate of utilization of prior providers is that we found low rates of use among those with lower levels of education, and Texas workers are generally less educated than workers in the other three states,3 although low utilization may also reflect health insurance differences, because workers with less education are less likely to have health insurance. 1 For the period 2001–2003, the proportions of the population not covered by health insurance in the four study states and in the United States were California, 18.7 percent; Massachusetts, 9.6 percent; Pennsylvania, 10.7 percent; Texas, 24.6 percent (highest in the nation); and United States, 15.1 percent (DeNavas-Walt, Proctor, and Mills, 2004). 2 Research and Oversight Council on Workers’ Compensation and Med-FX, LLC (2001, p. 55). 3 See Technical Appendix Table B.2. employee or employer choice of provider: main results 41 Summary of Findings Compared with cases in which the employer chose the provider, we find that: ■ When the worker selected a provider who had treated the worker previously for an unrelated condition, there is some evidence that medical benefits were higher; however, indemnity benefits did not differ from those related to cases in which the employer chose the provider. There is little evidence of worse return-to-work outcomes. Recovery was unaffected by employee choice of a prior provider, but employee choice of a prior provider was associated with much higher satisfaction with overall care. ■ In contrast, when the worker selected a new provider (one that had not previously treated the worker), medical and indemnity costs were higher and return-to-work outcomes were poorer. Again, there was no difference in physical recovery, but satisfaction with care was greater. Impact on Costs and Outcomes of Employee-Selected Prior Provider or New Provider, or Employer Choice of Provider: Main Results This section presents the results in detail. We report estimates using the combined data from the four states, while controlling for material differences among states and cases. Again, we report the results for two specifications: the first might undercontrol for severity by excluding the treatment variables, while the second likely overcontrols for severity by including them. The true results probably lie somewhere between the results of the two models. When the results are similar in the two models, or where the results remain strong for the second model, the reader should have greater confidence in the conclusions. The evidence does not point to substantive differences in costs, return to work, or physical recovery between cases in which the employer chose the provider and those in which the employee chose a prior provider, although employee choice of a prior provider is associated with greater satisfaction with overall medical care. The results are presented in the first two columns of Table 4.2. In model 1, medical payments and duration of time until first substantial return to employment were higher. However, the differences become considerably smaller (shrinking by about two-thirds) and statistically insignificant in model 2, with the hospitalization and surgery controls added. Satisfaction with care is much more likely to be higher for either model. 42 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 4.2 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Four States Combined Employee Chose a Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Employee Chose a New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 22** ($1,924) 9 ($1,116) 17† –4 –3 86** 7 ($629) –1 (–$162) 7 3 –1 89** 20** ($1,745) 20** ($2,538) 48** –28** 2 38** 12* ($1,052) 15† ($1,879) 40** –28** 3 39** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. In model 1, for example, when the worker chose a prior provider, medical payments were $1,924 greater per case (on average) than when the employer chose the provider; or the duration of time out of work was 17 percent longer when the worker chose a prior provider than when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Specifically, as Table 4.2 shows, compared with cases in which the employer selected the provider, cases in which the employee chose a prior provider had: ■ Medical payments that were 7–22 percent higher, although the results are not statistically significant in one of the two models used. ■ Indemnity benefits that are not statistically different (estimates ranged from 1 percent lower to 9 percent higher). ■ Substantial return-to-work outcomes that are not statistically different (estimates of return-to-work rates that ranged from 4 percent lower to 3 percent higher). employee or employer choice of provider: main results 43 ■ Durations that were 7–17 percent longer, although the results were only weakly significant in one model and not significant in the other. ■ Recovery of physical health that was similar. ■ Much higher likelihood that the worker rated the care with higher satisfac- tion. By contrast, as the third and fourth columns of Table 4.2 show, comparing cases in which the employer chose the provider to those in which the worker selected a new provider provides stronger evidence that employee choice leads to higher medical and indemnity costs and poorer return-to-work outcomes. The statistical evidence is weaker in model 2 for indemnity benefits but remains quite strong for medical benefits and for both return-to-work measures. Again, employee choice of provider (in this case, a new provider) is not associated with better physical recovery, although it is associated with higher satisfaction. Specifically, as Table 4.2 shows, compared with cases in which the employer selected the provider, cases in which the employee chose a new provider (who had not provided prior treatment for an unrelated condition) had: ■ Medical payments that were 12–20 percent higher. ■ Indemnity benefits that were higher by 15–20 percent. ■ Much lower rates of substantial return to work and longer durations of time out of work, with workers 28 percent less likely to return to work and remain there for at least one continuous month compared with employerselected cases and their time until return to work 40–48 percent longer. ■ Recovery of physical health that was similar. ■ Nearly 40 percent higher likelihood that the worker reported a higher level of satisfaction with care. When the employee chose a prior provider, the outcomes for indemnity benefits, duration, substantial return to work, and recovery were generally close to those found when the employer made the selection, except perhaps for medical costs. However, when the employee chose a new provider, the costs were higher and most outcomes poorer than when the employer selected the provider. In either case, workers were more likely to report higher levels of satisfaction when they chose the primary provider — and especially so with a prior provider. The results in Table 4.2 suggest that the findings in Chapter 3 regarding higher costs and worse return to work associated with employee choice overall were driven, in large part, by employee choice of a new provider. That is, there are po- 44 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s tentially important differences in the costs and outcomes of cases in which the worker selected a prior provider compared with those in which the worker selected a new provider. Table 4.3 uses the estimates underlying Table 4.2 to show the differences in costs and outcomes between when the worker selected a prior versus a new primary provider and indicates which differences are statistically significant. The table reports the impact of the employee choosing a new provider compared with the employee choosing a prior provider.4 We see that medical payments were similar but indemnity benefits were 11–16 percent higher when the employee selected a new provider, although the latter difference is at best weakly statistically significant. With respect to return to work, the differences are sharper, with employee choice of a new provider associated with significantly poorer return-to-work outcomes. In particular, when the worker selected a new provider, the odds of having a substantial return to work were 26–30 percent lower, and the duration of time out of work was 26–30 percent longer. Finally, satisfaction was lower when the worker selected a new provider than when the worker selected a prior provider. Yet, as we have found throughout this report, choice was unrelated to physical recovery. Chapter 6 returns to the issue of provider choice, recovery, and worker satisfaction. 4 In other words, in Table 4.2, we report the differences associated with comparisons between the two types of employee choice, on the one hand, and employer choice, on the other. The results reveal considerably sharper differences between employee choice of a new provider and employer choice, compared with between employee choice of a prior provider and employer choice. Table 4.2 does not, however, address whether the differences associated with the two types of employee choice are significantly different from each other. If they are not, then arguably our best estimates come from the simpler models presented in Chapter 3, which constrain the effects of the two types of employee choice to be the same. The results in Table 4.3 come from including in the regression models [see equation (2.1)] a dummy variable for either type of employee choice and an interaction between this dummy variable and a dummy variable for employee choice of a new provider. The estimated coefficient of the latter interaction measures the difference between the two types of employee choice, and a test of its statistical significance tells us whether the two types of employee choice have significantly different effects. Table 4.3 reports these latter differences and their statistical significance. Thus, for example, the result for duration in the model 1 column means that durations were on average 26 percent longer with employee choice of a new provider than with employee choice of a prior provider, and this difference is significant at the 5 percent level. Note that this 26 percent figure is not simply the difference between the estimates for duration reported in Table 4.2 for the two types of employee choice for model 1, because the numbers reported in Tables 4.2 and 4.3 are calculated from the exponentials of the regression coefficients. employee or employer choice of provider: main results 45 Table 4.3 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction –2 (–$180) 11 ($1,422) 26** –26* 5 –26** 5 ($422) 16† ($2,040) 30** –30** 4 –26** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Revisiting the Question of Unmeasured Residual Injury Severity As discussed in Chapter 2, the question arises whether the estimates presented so far (and in the next chapter) reflect only provider choice or also reflect unmeasured residual variation in injury severity that is associated with provider choice. We noted that, especially in the models that control for treatment (model 2 in the tables), we are quite confident that the estimates reflect causal effects of provider choice. However, as a way of shedding a little more light on this question, Table 4.4 reports results in which, in a sense, we go in the opposite direction to what we did when we added the treatment variables. In particular, we begin here with the model 1 estimates and then successively drop the perceived severity variable, and then drop the “type of injury” variables as well. If unmeasured injury severity accounted for large shares of the apparent effects of provider choice on workers’ compensation outcomes, then when we drop the perceived severity measure, the effects of provider choice should appear even larger. However, as indicated in the first two columns of Table 4.4 for em- 46 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 4.4 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Four States Combined, Excluding Severity and Injury Measures Employee Chose a Prior Provider Employee Chose a New Provider Model 1: Without Treatment Controls (percent) Omit Severity Variable (percent) Omit Severity and Injury Variables (percent) Model 1: Without Treatment Controls (percent) Omit Severity Variable (percent) Omit Severity and Injury Variables (percent) Medical payments 22** ($1,924) 21** ($1,817) 22**($1,940) 20** ($1,745) 21** ($1,811) 23** ($2,027) Indemnity benefits 9 ($1,116) 7 ($916) 9 ($1,116) 20** ($2,538) 21** ($2,651) 24** ($3,093) Duration 17† 15 17† 48** 47** 52** Substantial return to work −4 −3 −7 −28** −28** −31** Recovery −3 −5 −5 2 3 3 Satisfaction 86** 87** 83** 38** 36** 32** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables, and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. ployee choice of a prior provider, and in the fourth and fifth columns for employee choice of a new provider, the estimates scarcely change when the perceivedseverity variable is omitted, casting doubt on an important role for unmeasured severity in the model 1 estimates. Taking this one step further, in the last column for each type of employee choice, we even drop the injury type variables, which surely capture information on the nature and severity of the injury. Here, especially for choice of a new provider, the estimated effects on costs and return to work grow (in absolute value), but only slightly. The implication is that it is unlikely that unmeasured injury severity materially distorts the estimated effects of provider choice that we find. 5 Impact of Provider Choice on Costs and Outcomes: Results for Individual States In this chapter, we consider evidence from each of the four study states on the impact of provider choice on workers’ compensation costs and outcomes. Because the sample sizes are much larger for the analysis using the combined data, we recognize that those results are more precise and statistically powerful and that the results for individual states are less informative. Nonetheless, the estimates presented here can help confirm which states show overall patterns of results similar to the results for the combined sample. The structure of this chapter parallels the analyses reported in Chapters 3 and 4. First we show the differences in costs and outcomes where the employee or the employer chose the primary provider, and then we analyze whether there appear to be differences when the employee selected a new or a prior primary provider. The workers’ compensation systems in the four study states have some similarities and some differences. Technical Appendix F provides a summary of some of the relevant attributes of the workers’ compensation systems in each of the four states. Those attributes include costs, litigiousness, benefit levels, and a variety of factors associated with medical care, including provider choice rules, fee schedules, and limits on who can treat injured workers. We provide Technical Appendix F for readers who would like to understand some of the institutional features that might shape the provider decision or the consequences of that decision. Citations to sources for the information summarized here are contained in the appendix. 47 48 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Here are some of the important differences: ■ Coverage was mandatory with a few exceptions in each of the states except Texas, where employers can elect not to participate in workers’ compensation. According to a recent study by the Texas Department of Insurance (2004), 24 percent of the workforce was not covered by workers’ compensation insurance in 2004. ■ The average cost of a claim is much higher in California and Texas than in Massachusetts and Pennsylvania. In California, both medical and indemnity benefit costs are high. In Texas, medical costs and the duration of disability are the main cost drivers. In Pennsylvania, medical and indemnity costs are typical, while in Massachusetts, medical costs per claim are significantly lower than the other three states. ■ All four states have medical fee schedules that regulate nonhospital provider fees. In 2001, the fee schedule levels were, on average, much lower in Massachusetts than in the other three states. This was true for all the major services provided to injured workers. ■ States with the lowest fee schedules, Massachusetts and Pennsylvania, have the lowest network penetration rates. This occurs because opportunities to obtain network discounts are limited when regulated prices are already low. ■ Litigation is relatively infrequent in Texas but much more common in the other states, especially in California. ■ A much larger percentage of the general population is covered by nonoccupational health insurance in Massachusetts and Pennsylvania than in California and Texas. As we did in Chapters 3 and 4, we report here the results for two models, excluding and including the treatment variables as additional proxies for severity, and for the two-way and three-way classifications of provider choice. California A comparison of the findings for California (Table 5.1) with the more precise and statistically powerful multistate results (see Table 3.1) reveals similarities. When the employee chose the primary provider, costs were higher, return-to-work outcomes were poorer, recovery was no different, and satisfaction with overall care was higher. There are also marked differences between when the employee chose a impact of provider choice on costs and outcomes 49 Table 5.1 Impact of Employee Choice Compared with Employer Choice, California Model 1: Without Treatment Controls (percent) Model 2: Without Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 22* ($2,210) 26** ($3,946) 48** −16 −2 34† 10 ($1,031) 18† ($2,840) 39** −12 −1 37* Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state; see the discussion on weighting in Technical Appendix D. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. new provider and when the employer chose the provider, as shown in Table 5.2 (which can be compared with Table 4.2 for the combined sample). As Table 5.1 indicates, in cases in which the worker selected the provider, medical payments were 10–22 percent higher, although only the higher estimate is statistically significant. Indemnity benefits were 18–26 percent higher, and only the higher estimate is strongly significant. In these cases, the worker was 12–16 percent less likely to have a substantial return to work in the 3.0 to 3.5 years between injury and interview compared with cases in which the employer selected the provider, but the difference is not statistically significant. Workers reported a 39–48 percent longer time span between injury and initial return to substantial employment when the employee chose the provider. We found little difference in perceived recovery of physical health between cases in which the employee chose the primary provider and those in which the employer chose. When workers selected their providers, they were 34–37 percent more likely to report a higher level of satisfaction with care. 50 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 5.2 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, California Employee Chose a Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Employee Chose a New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 18 ($1,778) 2 ($367) 11 48 −10 24 9 ($944) −2 (−$373) 10 25† ($2,507) 47** ($7,345) 102** 9 ($940) 37** ($5,796) 77** 54 −44* −9 6 25 42† −41* 7 48* Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. As shown in Table 5.2, when, the differences in costs and outcomes are not statistically significant different between cases in which the worker chose a prior provider and those in which the employer chose the provider.1 In contrast, comparing employee choice of a new provider to employer choice, indemnity benefits were 37–47 percent higher, both statistically significant. Medical payments were 1 A prior provider is defined as a provider who previously treated the worker for an unrelated condition, regardless of whether the worker predesignated that provider under the prevailing California law. Under the old law, a predesignated provider need not have previously treated the worker; under the new law, a predesignated provider must be a personal physician under a group health plan who agrees to the predesignation. Our measure of prior provider is different from either statutory definition and, in some respects, broader. impact of provider choice on costs and outcomes 51 Table 5.3 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, California Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 7 ($729) 45** ($6,978) 82** –62** 16† 14 0 (–$5) 40** ($6,170) 61** –61** 16† 19 Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. also higher by 9–25 percent, but not significantly so at the lower estimate and only marginally significant for the higher estimate. Workers who saw new providers were much less likely (41–44 percent) to have substantial returns to work and had much longer (77–102 percent) periods until substantial returns to employment. Workers who selected new providers were more likely (42–48 percent) to report higher levels of satisfaction with care than when employers chose the providers, although as has proven to be common throughout our analyses, there is no evidence of differences in physical recovery. The results also reveal important differences in the costs and outcomes of cases in which the worker selected a new provider compared with cases in which the worker chose a prior provider. Table 5.3 shows these differences and indicates which differences are statistically significant in California. In particular, indemnity benefits were about 40–45 percent higher when the worker selected a new provider rather than a prior provider. Medical payments were not significantly different. Indemnity benefits were higher in part because workers who saw new providers were much less likely (61–62 percent) to have substantial returns to work and had 52 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s longer durations of time out of work (61–82 percent). Table 5.3 also shows that workers who saw new providers had 16 percent better recoveries of physical health — although this is only marginally significant in either of the two California models. Note that the latter result is the one of the few pieces of evidence in the entire report that gives any indication of better physical recovery when the employee chooses a new provider, whereas the evidence of higher costs and worse return to work is ubiquitous. Although our overall conclusions suggest, therefore, that employee choice of a new provider raises costs without improving physical recovery, it is important to keep in mind that for California, we have to be a little more cautious in reaching the conclusion that employee choice of a new provider does not deliver better recovery than employee choice of a prior provider, given the results on recovery shown in Table 5.3. However, this evidence is only marginally statistically significant. Texas Like the findings for California, those for Texas are in many ways similar to the more precise and statistically powerful multistate results. When the employee chose the primary provider, costs were higher, return-to-work outcomes appear to be poorer, recovery was no different, and satisfaction with overall care was higher. For Texas, some of these results hold for employee choice of a prior provider as well as a new provider, although more so for choice of a new provider. Differences also exist between cases with an employee-selected prior provider and those with an employee-selected new provider. In particular, with a new provider, return-towork outcomes were poorer and satisfaction with overall care was lower. However, we found no material differences in perceived recovery or costs. As Table 5.4 indicates, in cases in which the worker selected the provider, medical payments were 22–24 percent higher, and indemnity benefits were 21 percent higher. In these cases, the worker was 21–24 percent less likely to have a substantial return to work in the 3.0 to 3.5 years between injury and interview than was the worker whose employer selected the provider, and the duration of time until return to substantial employment was 28–31 percent longer, although the returnto-work differences generally are not statistically significant. We found little difference in perceived recovery of physical health between cases in which the employee or the employer chose the primary provider. The worker who selected the provider was 57–58 percent more likely to report higher satisfaction with overall medical care. impact of provider choice on costs and outcomes 53 Table 5.4 Impact of Employee Choice Compared with Employer Choice, Texas Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 24* ($2,836) 21* ($2,152) 31† –21 –3 58** 22** ($2,558) 21** ($2,169) 28 –24 –3 57** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. When the worker selected a prior provider, compared with cases in which the employer chose, indemnity benefits were 16–22 percent higher, while medical payments were 21–31 percent higher, although most of these estimates are at best marginally significant (Table 5.5). Workers reported similar recovery of physical health, regardless of who selected their providers, and return-to-work differences are relatively small and statistically insignificant. Workers were much more likely to report higher satisfaction when treated by prior providers that they chose compared with employer-selected providers. Comparing cases in which the worker selected a new provider to cases with an employer-selected provider, costs were higher and return-to-work outcomes poorer when the worker chose, although some of these estimates are only marginally significant. Satisfaction with care was about 23 percent higher, but the difference is not statistically significant. Recovery of physical health was similar. As Table 5.5 indicates, medical payments and indemnity benefits per case were 19–23 percent higher when the worker chose a new provider. In such cases, the worker was 31–36 percent less likely to have a substantial return to work. Compared with 54 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 5.5 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Texas Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 31* ($3,639) 22† ($2,273) 10 12 2 159** 21† ($2,421) 16 ($1,613) 7 20 2 160** 19† ($2,220) 22* ($2,504) 19† ($1,926) 23** ($2,348) 44† 42† –31† –36† –5 –5 24 23 Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. cases in which the employer chose the provider, the duration of time until a substantial initial return to employment was about 43 percent longer. Table 5.6 compares Texas cases in which the worker selected a prior provider with cases in which a new provider was selected. The evidence indicates that return-to-work and satisfaction outcomes were poorer in cases in which the employee selected a new provider rather than a prior provider. A worker in this group was 38–47 percent less likely to have a substantial return to work. When the worker selected a new provider, the duration of time until return to substantial employment was 31–33 percent longer, although these differences are not significant. Additionally, a worker choosing a new provider was 52–53 percent less likely to report a higher level of satisfaction with care; that is, the worker was much more likely to report lower satisfaction with care compared with a worker who chose a prior provider. impact of provider choice on costs and outcomes 55 Table 5.6 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Texas Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction −12 (−$1,419) −3 (−$347) 31 −38† −7 −52** 1 ($83) 7 ($735) 33 −47* −7 −53** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Massachusetts The results for Massachusetts parallel the more precise and statistically powerful multistate results along some dimensions only, and the differences associated with employee versus employer choice are generally not statistically significant. The same is true when we look at differences between employee choice of prior and new providers. The strongest result for Massachusetts (and the only statistically significant one) is that in cases in which the worker selected the provider, the worker was much more likely to report higher satisfaction with care than when the employer chose. Medical payments in Massachusetts are unlikely to be much different regardless of whether the employee or employer selects the provider. As in the other three states, in cases in which the worker chose the provider, the worker was less likely to have a substantial return to work and had a longer time out of work. The estimated coefficients on the return-to-work outcomes are similar to those reported earlier, but they are not statistically significant. As in the other states, per- 56 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 5.7 Impact of Employee Choice Compared with Employer Choice, Massachusetts Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 10 ($514) 11 ($1,486) 19 −33 −1 83** −2 (−$106) 7 ($956) 17 −33 0 74** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. ceived recovery of physical health was not affected by who selected the provider. These results are reported in Table 5.7. Turning to employee choice of prior and new providers, the estimates reveal no consistent pattern of higher costs associated with either type of employee choice of provider (Table 5.8). The estimates are generally consistent with employee choice of a new provider worsening return-to-work outcomes, although again the estimates are not statistically significant. The only strong result is that in cases in which workers selected prior or new providers, they were much more likely to report higher satisfaction with care. Not surprising, given the lack of statistical significance in these findings, we also see no statistically significant differences between employee choice of prior or new providers. Table 5.9 shows that there are no statistically significant differences in costs or outcomes between cases in which the worker chose a prior provider compared with those in which the worker chose a new provider. Worker satisfaction appears to be higher when a prior provider was used and return-to-work outcomes worse, but neither of these differences is statistically significant. impact of provider choice on costs and outcomes 57 Table 5.8 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Massachusetts Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 19 ($955) 19 ($2,672) 13 −38 −1 99** −1 (−$61) 11 ($1,526) 6 −38 0 92** 2 ($119) 3 ($393) 26 −27 −1 74** −3 (−$171) 3 ($387) 28 −28 −1 65* Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Table 5.9 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Massachusetts Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction −17 (−$836) −16 (−$2,279) 11 17 0 −13 −2 (−$110) −8 (−$1,139) 20 16 0 −14 Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. 58 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Pennsylvania The results for Pennsylvania provide less evidence that is similar to the more precise and statistically powerful multistate results. The strongest result for Pennsylvania is that in cases in which the worker selected the provider, the worker was much more likely to report higher satisfaction with medical care than when the employer chose (Table 5.10). Medical payments in Pennsylvania are unlikely to be much different regardless of whether the employee or employer selects the provider. As in the other states, in cases in which the worker chose the provider, the worker was less likely to have a substantial return to work than when the employer chose; but in Pennsylvania, these effects are not statistically significant. Again, as in the other states, perceived recovery of physical health was not materially affected by choice of provider, looking at the two-way classification of employee versus employer choice. The Pennsylvania results parallel the multistate results when we look at whether workers selected new providers or employers chose, but the results are somewhat different when we look at when employees chose prior providers com- Table 5.10 Impact of Employee Choice Compared with Employer Choice, Pennsylvania Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 13 ($997) −6 (−$683) 22† −22 4 72** −3 (−$210) −19 (−$2,113) 16 −19 6 72** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. impact of provider choice on costs and outcomes 59 Table 5.11 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice, Pennsylvania Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments 3 ($234) −14 (−$1,041) 22† ($1,709) 8 ($579) Indemnity benefits −28 (−$3,144) −39† (−$4,448) 14 ($1,586) 1 ($62) Duration 14 9 32* 26† Substantial return to work −16 −7 −29 −31 Recovery −3 −1 11† 12* Satisfaction 87** 85** 58** 60** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a prior or new provider compared with when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. pared with when employers chose the providers. As before, when workers selected prior providers, they were much more likely to report higher satisfaction with care (Table 5.11). However, there is some marginally significant evidence that indemnity costs are lower when workers choose prior providers, even though the estimates indicate worse return-to-work outcomes (which are relatively small and not significant). When the worker selected a new provider compared with the employer selecting the provider, there is no consistent statistically significant difference in costs. We did find longer duration and lower rates of return to work, although only the former is statistically significant and only in model 1. We also, as before, found higher rates of satisfaction with overall care. Finally, employee choice in Pennsylvania may improve physical recovery. When we compare the costs and outcomes of cases with worker-selected prior or new providers (Table 5.12), we find that both indemnity and medical payments may be higher with new providers, but the evidence is at best marginally sig- 60 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 5.12 Differences in Costs and Outcomes between Employee Choice of New Provider and Prior Provider, Pennsylvania Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Medical payments Indemnity benefits Duration Substantial return to work Recovery Satisfaction 19 ($1,475) 42† ($4,730) 16 −16 14* −15 21 ($1,620) 40† ($4,510) 16 −25 14† −14 Notes: The results are interpreted as the difference in costs or outcomes when the employee chose a new provider compared with when the employee chose a prior provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. nificant. New providers were associated with 14 percent better physical recoveries. The result is not consistent with what we have found in the combined data, although as noted earlier, we found a similar — albeit also weak — result for California. None of the other differences (return to work or satisfaction) are statistically significant. 6 Worker Satisfaction with Health Care One of the strong and consistent findings in this study is that worker satisfaction with the overall health care received was higher when the worker chose the primary provider. In this chapter, we probe the possible reasons underlying this finding. We are particularly interested in trying to understand the higher satisfaction associated with employee choice of provider because the evidence shows that employee choice is not associated with better physical recovery as reported by the worker. The survey asked workers, “Now think about all the medical care you received from the first treatment for your injury until now. Were you satisfied or dissatisfied with the medical care you received overall?” This summary question enabled respondents to take account of their satisfaction with multiple aspects of health care delivery, and overall the survey found that workers generally reported high levels of satisfaction with their care.1 However, we also found that employee choice of primary provider was associated with higher rates of satisfaction than was employer choice. 1 The survey also asked workers about other metrics of satisfaction: satisfaction with the providers; with the availability, timing, and kind of care sought; with their desire to change providers at some point during treatment; and with the process of providing care. The patterns of responses on these metrics can be found in Barth and Victor (2003) and Victor, Barth, and Liu (2003). 61 62 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s In this chapter, we examine evidence on several conjectures regarding the higher satisfaction associated with employee choice of provider: 1. Some employee-selected providers might achieve better physical recoveries for their patients, and workers might be more satisfied with their better recoveries. 2. Some employer-selected providers might tend to rush workers back to work prematurely; therefore, workers who return before they are ready might have lower satisfaction with care. 3. Some workers might prefer not to go back to work, or prefer to delay their return to work beyond the time that they may be physically able to return, and worker-selected providers might be more likely to support their preferences; therefore, these workers might be more satisfied with such care even if physical recovery is no better. 4. Workers might have certain expectations about the processes of care (for example, speed of first visit, time spent with provider, and bedside manner), and employee-selected providers might be more likely to meet those expectations, regardless of physical recovery. 5. Some workers might experience an “empowerment effect” when they select their own providers, and by itself this effect might lead to higher levels of satisfaction regardless of physical recovery. 6. Some workers might suspect that employer-selected providers are more concerned with satisfying the needs of the employer than of the worker. Such a suspicion could result in a lower degree of trust and hence lower satisfaction with the treatment, even if recovery is not affected. The data allow us to examine conjectures 1, 2, and 3 but not the other possibilities. In the sections that follow, we discuss evidence on each of the first three conjectures and rule them out. This leaves open the possibility that the impact of provider choice may be the result of satisfaction with the process (rather than outcomes) of care, an “empowerment effect,” or an attitude about the perceived motivation of the provider. What unifies these three conjectures is that they have more to do with the process of the medical care, not the outcomes, especially the physical recovery of the injured worker. Correlates of Satisfaction and Overall Health Care Received We begin by examining a number of factors correlated with satisfaction, as suggested by conjectures 1, 2, and 3 listed in the previous section. Then we examine worker satisfaction with health care 63 whether each is affected by provider choice. The evidence is very clear that higher satisfaction with overall care is strongly related to (1) a better recovery of physical health, (2) a worker’s belief that he or she was not sent back to work too soon, and (3) having a sustained return to work, uninterrupted by subsequent periods of lost time due to the injury. However, it turns out that these correlates of satisfaction are not associated with who chose the provider, which is the evidence that rules out the first three conjectures in our list. We begin by categorizing the recovery scores of respondents. Recovery is classified into groups based on the change in the standardized SF-12® scores measured soon after the injury and at the time of the interview. (Recovery is defined in Chapter 2 and in Technical Appendix C.) We defined the groups as follows: change of at most −2 points (complete recovery); −2 to −10 points (somewhat incomplete); −10 to −20 points (more incomplete recovery); and exceeding −20 points (very incomplete). As Table 6.1 shows, workers who reported less recovery of physical health were less likely to report higher satisfaction with their overall health care. Of those reporting complete recoveries, 91 percent said that they were very or somewhat satisfied with their care. In contrast, among workers reporting the least complete recoveries, only 67 percent said that they were very or somewhat satisfied.2 Although satisfaction and recovery are strongly correlated, we previously reported the findings that provider choice did not affect recovery of physical health. Thus, we rule out the first conjecture listed in the previous section. Workers who believed that they returned to work too soon were less satisfied with their care. Table 6.2 shows that 91 percent of workers who reported that they returned to work at the “right time” were very or somewhat satisfied with care. Those who said they returned to work “too soon” were less satisfied — 74 percent reported being very or somewhat satisfied with care. Sixteen percent of those who reported returning to work too soon were very dissatisfied with their overall care compared with only 4 percent of those who said they returned to work at the right time. These workers may have thought that their health care providers were responsible for having them return to work prematurely. 2 It may seem surprising that 2 out of 3 persons with very incomplete recoveries told us that they were very or somewhat satisfied with the care that they received. Yet this may be explained when one considers that the workers expressed satisfaction with the care that they received and not with their conditions following their injuries and treatments. That is, someone with a more serious medical condition may not expect a complete recovery, and the worker may measure satisfaction against his or her expectations. 64 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 6.1 Satisfaction with Care, by Degree of Recovery of Physical Health Complete Recovery (percent) Somewhat Incomplete Recovery (percent) More Incomplete Recovery (percent) Very Incomplete Recovery (percent) Very or somewhat satisfied 91 86 78 67 Very satisfied 65 53 44 36 Somewhat satisfied 26 33 34 31 Very or somewhat dissatisfied 9 14 22 33 Somewhat dissatisfied 5 6 11 15 Very dissatisfied 4 8 11 18 Note: Recovery categories based on changes in standardized SF-12® scores. Complete recovery: change of at least −2 points; somewhat incomplete: change of −2 to −10 points; more incomplete recovery: change of −10 to −20 points; very incomplete: change of more than −20 points. Table 6.2 Worker’s Perception of Timing of Return to Work and Satisfaction with Care Very Satisfied (percent) Somewhat Satisfied (percent) Somewhat Dissatisfied (percent) Very Dissatisfied (percent) Right time Too soon 63 42 28 32 6 11 4 16 A worker who had a second absence due to his or her injury was also likely to be less satisfied with the health care received. Seventeen percent of workers who had a substantial return to work reported that they experienced a second significant absence (lasting more than one week). Respondents may have perceived the second absence as the result of a premature return to work or returning without adequate work limitations prescribed. Table 6.3 shows that 88 percent of workers who reported no second absence due to the injury said that they were very or somewhat satisfied with care. Those who had a second absence were less worker satisfaction with health care 65 Table 6.3 Second Absence Due to Injury and Satisfaction with Care No Second Absence (percent) Second Absence (percent) Very satisfied Somewhat satisfied Somewhat dissatisfied Very dissatisfied 60 28 6 6 35 32 14 19 satisfied — 67 percent reported being very or somewhat satisfied with care, and 33 percent were somewhat or very dissatisfied. Correlates of Satisfaction and Provider Choice Are workers with less complete recoveries more likely to have providers selected by employers? We have seen that workers who have less complete recoveries have lower levels of satisfaction with care. However, there is no significant difference in recovery between cases in which the employee or employer selected the provider. Hence, it is unlikely that the large impact of employee choice on satisfaction is due to employee-selected providers achieving better physical results for workers. We show this in two ways. First, in Tables 3.1 and 4.2, we showed that provider choice had no statistically significant effect on recovery, controlling for a large number of injury, worker, and employer characteristics that could affect recovery. Second, in Table 6.4, we show a simpler comparison with no controls for other factors. Cases in which the employer selected the provider were equally likely to report an incomplete recovery as cases in which the employee selected the provider. Are workers who report that they returned to work “too soon” more likely to have providers selected by employers? As shown in Table 6.2, workers who believed that they returned to work too soon were less satisfied with their care. However, workers who chose their providers were equally likely to report that they returned to work too soon as workers whose employers chose the providers (Table 6.5). Thus, it is unlikely that the large impact of employee choice on satisfaction is due to employer-selected providers rushing workers back to work prematurely. We 66 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s Table 6.4 Completeness of Recovery, by Who Selected the Provider Complete Recovery (percent) Somewhat Incomplete Recovery (percent) More Incomplete Recovery (percent) Employee-selected provider Employer-selected provider 41 41 19 21 14 13 Very Incomplete Recovery (percent) 27 25 Note: Recovery categories based on changes in standardized SF-12® scores. Complete recovery: change of at least −2 points; somewhat incomplete: change of −2 to −10 points; more incomplete recovery: change of −10 to −20 points; very incomplete: change of more than −20 points. Table 6.5 Worker’s Perception of Timing of Return to Work, by Who Selected the Provider Employee Chose (percent) Employer Chose (percent) Right time Too soon 63 37 65 35 confirmed this conclusion using multivariate statistical techniques that control for the factors described in Chapter 2. Is a worker who reported a second significant absence due to the injury, after having achieved a substantial return to work, more likely to have an employer-selected provider? As seen in Table 6.3, a worker who had a second absence was likely to be less satisfied with his or her care. However, a worker who chose the provider was equally likely to have a second absence as a worker whose employer chose the provider (Table 6.6). Therefore, it is unlikely that the large impact of employee choice on satisfaction is due to employer-selected providers returning workers to work prematurely, leading those workers to experience second absences. We confirmed this conclusion using multivariate statistical techniques that control for the factors described in Chapter 2. worker satisfaction with health care 67 Table 6.6 Second Absence Due to Injury, by Who Chose the Provider Employee Chose (percent) Employer Chose (percent) No second absence Second absence 82 18 82 18 Is the greater satisfaction with care when employees select their providers a re- flection of employee-selected providers who support the desires of some workers to return to work more slowly? Some workers might prefer not to return to work or to return to work more slowly. If employee-selected providers support this desire more often than employer-selected providers, we would expect that these workers to be more likely to report that they returned to work “too soon” when they saw employer-selected providers. We find no evidence that, on average, higher satisfaction with care from employee-selected providers derives from those providers supporting the preferences of some workers to return to work more slowly. As Table 6.5 showed, the percentage of workers reporting that they went back to work too soon was similar, regardless of who chose the provider. Satisfaction with Care: Prior or New Provider In considering the source of worker satisfaction or dissatisfaction with the health care received, we also note another strong and consistent finding. Comparing the satisfaction rates of workers who selected primary providers who had previously treated them for an unrelated condition with the satisfaction rate of workers who selected new providers, the former was significantly higher (see Table 4.3). Satisfaction was also higher for both types of employee choice compared with employer choice (see Table 4.2). We explored all the questions considered earlier in this chapter for the expanded three-way classification of provider choice to assess whether the higher satisfaction with employee choice of a prior or new provider, relative to employer choice, or higher satisfaction with employee choice of a prior rather than a new provider was attributable to any of the conjectures outlined at the beginning of this chapter. However, paralleling the analysis of provider choice based on the 68 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s two-way classification of employee versus employer choice, we found no evidence that the higher satisfaction with employee choice of either prior or new providers (or prior relative to new providers) derived from better physical recoveries, differences in the extent to which workers were rushed back to work, or greater ability to delay return to work. Again, we are led to the conclusion that differences in satisfaction are more likely due to factors such as the manner in which care was delivered, empowerment of the worker, and trust rather than more objective medical outcomes. 7 Discussion and Policy Implications With workers’ compensation medical payments high and rising rapidly in many states (Telles, Wang, and Tanabe, 2004), policymakers have intensified their efforts to modify state laws to try to reduce these costs while avoiding actions that might impair the outcomes experienced by injured workers. One of the actions often debated is giving employers more influence or direct control over the selection of providers. As discussed in Chapter 1, providers play many important roles in workers’ compensation cases and is believed to have a large impact on workers’ compensation costs and on outcomes for workers. During the period of rising costs between the late 1980s and early 1990s, several states modified “employee choice” laws to require that workers select providers from within approved networks of providers created by the employers. In California, an important cost containment provision of the 2004 legislative changes requires a worker to choose a provider from employer-selected networks of providers, unless the worker predesignates a provider who has previously treated the worker under a qualifying employer-sponsored group health plan. Previous studies of provider choice have focused primarily on the impacts on medical costs. Most reported that employee choice was associated with higher costs, as has this study. This study, however, is one of the first that rigorously examines not only costs but also worker outcomes associated with whether employers or workers actually select the provider, apart from statutory mandates on choice. It is also unique in looking at a variety of outcomes indicative of return to work and the quality of care and in focusing on the primary provider. Finally — and perhaps most relevant in light of the recent workers’ compensation reforms in 69 70 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s California — it is the only study that examines how the costs and outcomes of treatment differ when workers chose providers who previously treated them and when workers choose new providers. Summary of Results These are among the more important findings of this study: ■ Comparing cases in which the worker selected the primary provider with otherwise similar cases in which the employer selected the provider, we found that costs were generally higher and return-to-work outcomes poorer when the worker selected the provider, although workers reported higher rates of satisfaction with overall care but similar perceived recovery of physical health. ■ Compared with cases in which the employer selected the provider, cases in which the worker selected a provider who had treated the worker previously for an unrelated condition (“prior provider”) may have had higher costs, but the evidence was weak. Worker outcomes did not appear to be very different between cases with an employee-selected prior provider and those with an employer-selected provider, except that satisfaction with overall care was higher when the worker saw a prior provider. ■ Compared with cases in which the employer selected the provider, cases in which the worker selected a provider who had not treated him or her previously (“new provider”) had much higher costs and poorer return-to-work outcomes, generally no differences in physical recovery, and higher levels of satisfaction with overall care. ■ Comparing cases in which the employee selected a prior provider with similar cases in which the employee chose a new provider, we found that the worker treated by a new provider was less likely to return to work, returned to work more slowly if he or she did return, had lower levels of satisfaction with overall care, and experienced no better physical recovery. Medical payments were similar in both cases, but indemnity benefits per claim were higher for the worker treated by a new provider, although this evidence is statistically weaker than the other results. As discussed in Chapters 3 and 4, the primary findings summarized here come from the combined sample from four states (California, Texas, Massachusetts, and Pennsylvania), because compared with the individual states, the combined sample discussion and policy implications 71 size is much larger and the estimates are more precise and statistically powerful. We also present the individual state results, despite the fact that they are less precise and the statistical tests are less powerful. ■ In the two states with higher-than-typical medical payments, California and Texas, comparing cases in which the worker selected the primary provider with otherwise similar cases in which the employer selected the provider, we found some evidence of higher costs and poorer return-to-work outcomes when workers selected their providers, although workers reported higher rates of satisfaction with overall care and similar perceived recovery of physical health. ■ Costs for care were not significantly different between employer-selected and employee-selected providers in the two states (Massachusetts and Pennsylvania) with medical payments that were typical or lower than typical among states. However, duration of lost time may have been longer in Pennsylvania when workers select their providers. Satisfaction with care was higher when workers selected their providers, and there was no difference in perceived recovery of physical health. ■ The findings previously described for the combined sample were especially strong for California and Texas — although in Texas, cases with employeeselected prior providers and new providers had higher costs than cases with employer-selected providers. ■ In Pennsylvania, when the worker selected a prior provider, it appears that indemnity costs were lower than when the employer chose the provider, although the difference is at best weakly significant. However, paralleling other results, duration was longer when the worker selected a new provider compared with when the employer selected the provider. Moreover, workers reported higher levels of satisfaction when they chose their providers. Pennsylvania is the only state for which there is some statistically reliable evidence that employee choice of new providers may lead to slightly better physical recovery, and even then the evidence is relatively weak. Interpretative Caveats We are mindful of the need for care in interpreting and using the results from this study. First, only four states are included, and a wider set of states would add information that either reinforces the findings or is less consistent with them. Moreover, the focus of this study is who actually chose the primary providers in specific 72 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s cases, rather than the impact of the state legal provision about choice of initial provider or the laws about ongoing control of provider choice. State laws do appear to influence the actual choice, but there is hardly a perfect correspondence. Also, the reader is cautioned that the California and Pennsylvania laws and practices in effect during the sample period were not strong versions of employer choice laws. In both states, the employer retained the right to select the provider for only a limited period, after which the worker could change providers. Therefore, one needs to be careful about extrapolating from our findings the impact of changing state laws about who controls the choice of provider. The study includes both closed and open claims. However, cases are slower to close in California than in most other states. Consequently, the underlying data on paid costs understate the ultimate costs in California more than in the other states. This study does not address whether this understatement is associated with provider choice. If so, the magnitudes of the effects of provider choice on costs in this study could be biased. We do not explore the relationship between the choice of the initial provider and outcomes. Nor do we analyze the role of medical networks in the selection of providers and the costs and outcomes that result. For example, it seems more likely that a provider selected by an employer would be drawn from a network than would a provider selected by a worker. In future studies, we may address these issues. Finally, although we regard this study as an important addition to a relatively sparse empirical literature on a very important public policy issue, it is just one study. Additional research on other states and using other data sources and approaches will be useful to see if these results are robust, if they are supported in other contexts, whether provider choice has different effects in certain types of states but not in others, and how provider choice affects outcomes other than costs. With these caveats in mind, we conclude by discussing the implications for public policy that we can draw from our findings. Implications for Public Policy How do our results apply to public policy debates regarding choice of provider in workers’ compensation? The major themes of the policy debate often involve employer advocates arguing for employer choice on the grounds that (1) employers are better positioned to select good-quality providers because they, or the insurers, have better information; or (2) employee choice invites workers and attorneys discussion and policy implications 73 to game the system by selecting providers willing to extend the duration of time out of work in order to achieve financial gain for the providers or the workers. Employee advocates argue for worker choice of provider on the grounds that (1) workers are best positioned to select the most appropriate providers for their individual situations; and (2) some employers take advantage of the power to choose providers by selecting lower-cost providers who deliver inferior care or who may have a propensity to send workers back to work prematurely. This study finds some evidence to support both sides of the argument. As described in this report, however, based on our findings, it appears possible to improve the design of provider choice laws to lower costs and improve return-to-work outcomes without adversely affecting physical recovery from workplace injuries. First, we found that when workers chose their providers, costs were higher, recovery of health outcomes were not better, and return-to-work outcomes were often worse compared with when employers selected the providers. This finding suggests that employers, on average, may be well positioned to select good-quality, lower-cost providers — or at least better positioned than many workers. The finding also suggests that employers, in practice, are not generally selecting inferiorquality providers; although there may be exceptions, they are not frequent enough to affect the overall results. Second, we found that when workers selected prior providers, the costs and outcomes were not dramatically different from when employers selected providers. This evidence suggests that state laws that grant employers greater influence over the choice of provider should lead to lower costs and better return-to-work outcomes than laws that allow workers to select providers whom they have not seen previously — consistent with recent legislative changes made in California in Senate Bill 899. However, when workers selected providers — either prior or new — they expressed higher levels of satisfaction with care. We are not surprised by this finding regarding workers choosing prior providers, because the key issue is the likelihood that the worker will be seen by a provider who has the appropriate training and skills, is trusted by the worker, and delivers appropriate care. When a worker sees a provider with whom he or she has a preexisting relationship, a sense of trust may have already been established. In addition, the need for some diagnostics may be avoided, and the confidence that the worker has in a prior provider might increase the odds of compliance with treatment as well as ensure more open communication with the provider. The worker may also be confident that the provider has the necessary training and skills to treat the particular injury. When a worker needs to find a new provider, the employer or insurer is often 74 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s well positioned to identify the appropriate provider and help the worker get access in a timely manner, because insurers and employers typically have more experience and superior information about the practices of various providers. More surprising, though, is that workers also expressed greater satisfaction when they selected new providers (relative to employers choosing). We explored whether this greater satisfaction was related to dimensions of physical recovery not captured in our data or assistance in remaining out of work beyond the necessary time following an injury, but we were able to rule out such explanations. There may, however, be alternative explanations relating to empowerment or trust that leave workers more satisfied with choice of new providers even though costs and return-to-work outcomes are worse and physical recovery no better, perhaps stemming from distrust of providers chosen by employers and thus dissatisfaction with employer-selected providers. One possibility revolves around the process by which a worker accesses a health care provider following an injury. It may be that the worker does not have a preexisting relationship with a provider and does not wish to accept the employer’s recommendation. Alternatively, the worker may have a preexisting relationship with another provider but does not see that provider as having the requisite training or skills to treat the particular injury. Arguably, these types of considerations regarding higher satisfaction when employees choose their providers are of less interest to policymakers concerned with striking a balance between reducing the costs of workers’ compensation and ensuring adequate medical care and indemnity benefits. Overall, then, we view the higher costs and worse return-to-work outcomes associated with employee choice of new providers, coupled with no better physical recovery, as most consistent with some of the arguments in favor of employer choice. It remains to some extent a matter of speculation as to why costs appear to be higher and recovery no better when the employee chooses the provider — especially a new provider. For example, a worker trying to choose a new provider may not have adequate information about provider quality and may lack leverage to gain access to high-quality providers; the providers considered high quality may not be taking new patients or may be scheduling with significant delays. In contrast, the employer or insurer (or network), through its purchasing power, might be able to help the worker “jump the line” in when the employer chooses the provider. We cannot say for sure, but our results are consistent with situations in which the worker without a preexisting provider relationship is forced to participate in a search process with inadequate information about quality providers and inadequate leverage to gain access to those providers — almost a lottery-like pro- discussion and policy implications 75 cess compared with the situation in which the worker has a preexisting provider relationship or follows the employer’s recommendation. There are other reasons why an employer-selected provider might produce better outcomes at lower costs for a worker without a preexisting provider relationship. First, a provider selected by the employer might be more knowledgeable about the working environment and therefore better equipped to recommend sound return-to-work conditions. Second, many employers participate in medical network arrangements, which conduct provider credentialing and obtain fee discounts. Third, a new provider who is unfamiliar with the worker might feel compelled to practice defensively, thereby contributing to higher treatment costs and possibly to higher indemnity costs as well. Fourth, the findings about higher costs and poorer return-to-work outcomes with employee choice were stronger in the two study states with higher-than-typical medical payments and longer-thantypical durations of disability — California and Texas. The results were weaker in the two states with typical or lower-than-typical medical payments — Massachusetts and Pennsylvania. It would not be surprising if employee/employer choice is a much more important leverage point in higher cost states. The conclusions of this study regarding employee choice of new providers and prior providers are particularly salient for California. The results for California provide some evidence suggesting that the costs are higher and recovery no better when workers select providers with whom they have no prior relationship. Therefore, it is possible that the recent legislative changes struck an appropriate balance by significantly expanding the limits on worker choice of provider but retaining an exception where there is a preexisting provider relationship. However, it is important to understand the technical differences between the study’s definition of a prior provider and the new California statutory definition authorizing predesignation of a provider. The recently enacted California statute (Section 4600 of the California Labor Code) provides the following: ■ Unless the employer established a medical provider network, after 30 days from the date the injury is reported the employee may be treated by a physician of his or her own choice. [Sec. 4600(c)] ■ If an employee has notified his or her employer in writing before the date of injury that he or she has a personal physician, the employee shall have the right to be treated by that physician if the employer provides nonoccupational health insurance. [Sec. 4600 (d)(1)] ■ A personal physician shall meet all of the following conditions: ■ The physician is the employee’s regular physician. [Sec. 4600 (d)(2)(a)] 76 t h e i m p a c t o f p r o v i d e r c h o i c e o n c o s t s a n d o u t c o m e s ■ The physician is the employee’s primary care physician and has previously directed the medical treatment of the employee, and retains the employee’s medical records. [Sec. 4600 (d)(2)(b)] ■ The physician agrees to be predesignated. [Sec. 4600 (d)(2)(c)] ■ The maximum percentage of all employees who are covered under the pre- designation provision at any time in the state is 7 percent. [Sec 4600 (d)(6)]1 Thus, the predesignation exception to the workers’ choice of providers among an employer-designated network of providers bears some resemblance to the “prior provider” category analyzed in this report. However, compared with the California requirements for predesignation, our prior provider category is in some respects broader because it does not require the prior provider to be the personal physician of the worker under a nonoccupational group health insurance plan offered by the employer. Consequently, while suggesting that the reforms in California may have struck an appropriate balance, the evidence does not constitute a direct evaluation of the particular way in which employee choice of providers has been changed under the reforms. One particular point to note for Texas: it was the only state where we found higher costs regardless of whether the employee selected a prior provider or a new provider. It may be that the potential lesson for California about distinguishing between employee choice of prior and new providers is less applicable for Texas. 1 How this provision will be applied, if it is, is unclear to observers with whom we have conferred. In addition, the rule could be read as applying to 7 percent of an employer’s workers. Technical Appendix A: Literature Review In general, studies of workers’ compensation outcomes based on the choice of provider have reached mixed conclusions. Potential explanations for this variation in results include weaknesses that we try to rectify in this report. First, different results are often the result of different data sources covering different periods and states. Moreover, in most studies, the choice of provider was not known with respect to specific claims, and the degree to which the worker might have changed providers or had multiple providers was not considered because of data limitations. Rather, studies prior to ours characterized provider choice for all cases in a state as employer choice, for example, if the state statute gave the employer the right to select the provider — regardless of the actual choice made in each case. Based on average annual changes in medical payments in workers’ compensation cases in 41 states from 1965 to 1985, Boden and Fleischman (1989) found little relationship at the state level between the state’s approach to provider choice and the rate of medical cost growth. During the period in question, eight states changed their laws — two switched to employee choice and six switched to employer choice. Boden and Fleischman did not find evidence that changing the method of choice was correlated with cost changes after the change was made. Nor was there evidence that states that remained employer choice states over the 20 years studied tended to have lower rates of medical cost growth. Subsequently, Victor and Fleischman (1990) employed multivariate analytical methods and concluded that the choice of provider does affect medical payments. Using data from state rating bureaus and state funds (excluding self-insurers), the researchers examined the impact of a change in provider choice in Illinois (after 1975) and Texas (after 1973). They report that medical payments in Illinois rose 8–11 percent following a change to employee choice in the short run, and when the full impact of the change was absorbed, medical payments rose 19–49 percent. In Texas, the short-term effect was 4–6 percent, while the ultimate impact was es- 77 78 t e c h n i c a l a p p e n d i x a timated to be 7–29 percent. In their report, Victor and Fleischman emphasize the tentative nature of these results, partly because they used aggregate (not claimlevel) data and had a small sample, problems that the current study overcomes. In a later paper, Boden (1992) reports that of eight states analyzed, three might have seen costs affected by changes in provider choice approaches, but in the other five states, there was “no evidence that these changes triggered changes in medical payments” (p. 45). This issue is also addressed in a report by Durbin and Appel (1991). After studying average state medical payments between 1965 and 1984 and employing multivariate analysis across the states, they report that states with employer choice had 15 percent lower average medical payments in 1965 and that the difference widened to 36.5 percent in 1984. Their results also suggest that physician choice has a greater impact on medical payments than do fee schedules. The most data-intensive study of the issue of provider choice was conducted by Pozzebon (1994), whose findings differ from those of Durbin and Appel. She relied on data from almost 32,000 closed claims obtained from the National Council on Compensation Insurance, Inc. (NCCI) for 17 states for 1979–1987. Using medical payments per claim as the dependent variable, Pozzebon created four variables as her statutory choice measures: initial choice was limited, changing the provider was limited, no limits were placed on employee initial choice or on subsequent changes, and both initial choice and subsequent choice were limited. She found that where the employee’s initial choice was constrained, “Restrictions on initial choice increase health costs in workers’ compensation programs by 11–16 percent, a large and statistically significant effect” (p. 161). Limits on changing the provider subsequent to the initial choice were also found to be correlated with higher medical payments. However, Pozzebon acknowledges that these findings could result from higher costs leading to policies to limit change, rather than the cost-reducing effects of policies limiting choice. Pozzebon’s somewhat unexpected findings do not seem attributable simply to the source of the data used. In a 1996 study, Durbin, Corro, and Helvacian also used NCCI data and found that employer choice was associated with lower costs of medical benefits. However, the sample in that study was more limited, including 1,300 claims each for four states with 1987 as the injury year and closing dates between 1988 and 1992. A more recent study found that network penetration rates were higher when the state law authorized employer choice of initial provider (Victor, Wang, and Borba, 2002). The authors used estimates from other studies showing that higher network penetration was associated with lower medical payments to compute “a rough estimate” that changing a state’s law from employer to employee control of the decision to change providers increased medical payments by 7–10 percent. technical appendix a 79 The study used data from nine states (including the four states covered in this report) and controlled for worker characteristics, injury and claim type, industry, wage, and the prevalence of networks in the state, to estimate the impact of provider choice laws on network penetration rates. The only study of provider choice in workers’ compensation utilizing rigorous experimental methods compared experimental and control groups, with the former group treated in a managed care framework and the latter group allowed to select their own providers in a traditional fee-for-service arrangement (Washington Department of Labor and Industries and University of Washington Department of Health Services, 1997). Firms, not individual workers, were placed in the experimental or control group. The study tracked 1,354 injury cases with treatment in managed care and 1,708 cases from firms in the control group. For our purposes, this study has one significant drawback — namely, the differences between the groups were more than simply a matter of who selected the provider. Among other differences was the method of payment to the providers for either group. However, the study is also an important extension of earlier studies because the outcomes analyzed were more extensive than simply the medical payments per case. The Washington study found that workers in managed care settings had medical payments that were 27–32 percent below those in the traditional employee choice fee-for-service model. The study also compared rates of injured workers who received “time loss costs,” role-functioning scores (self-reported measures of how well individuals were able to carry out activities related to personal and social roles), and self-reported opinions on the progress of recovery and on overall outcomes. Workers were surveyed both at six weeks and six months after their injury. Workers treated in a managed care setting reported statistically significant lower role-functioning scores at six weeks and at six months. Workers reported significantly lower rates of satisfaction with their treatment, their attending physician, and with their overall access to care at six weeks if they were treated in the managed care group. However, at the six-month interview, statistically significant lower rates of satisfaction were found only with regard to overall access to care. At six weeks and at six months, workers in the managed care group reported less progress on recovery, and the difference is statistically significant. However, at six months, the study found no differences in the two groups with regard to pain, mental health status, and physical functioning. This study points to the multiplicity of outcomes that warrant attention in studies of provider choice. What can we conclude from this review of the literature? First, although most studies appear to conclude that employer choice is associated with lower medical payments in workers’ compensation, the findings are not unchallenged. This 80 t e c h n i c a l a p p e n d i x a should hardly be surprising, because the states and the years selected have varied and the measures of choice have tended to be crude. Very little work related to choice of provider has focused on outcomes or cost measures other than medical payments — such as duration of time out of work, indemnity benefits, physical recovery, and worker satisfaction with care. Rarely have many other factors that likely affect outcomes, such as worker and employer characteristics, been controlled for in these studies. No study appears to have considered and analyzed the significance of whether the injured employee had been treated previously by the provider who gave primary care in the workers’ compensation claim. Finally, studies done even a few years before this study were done when network arrangements were less common. Since employer-selected providers are more likely to participate in such plans now than they did previously, the relevance of some of those earlier studies may have diminished. Technical Appendix B: Variables Used in Study: Definitions and Descriptive Statistics 82 t e c h n i c a l a p p e n d i x b Table B.1 Definitions of Variables Dependent variables Indemnity benefits Medical benefits Substantial return to work Duration of disability Recovery Satisfaction Provider choice Independent variables State Pennsylvania California Texas Massachusetts Employer chose Employee chose Employee chose, prior Employee chose, new The indemnity payment the worker received. The amount the insurer paid for the worker’s medical treatment. A dummy variable. The value is 1 if the worker was able to return to work and stay for one full month. The number of weeks from the time of the injury to the first substantial return to work. Any worker who did not have a substantial return to work was assigned 156 weeks. Worker’s perceived recovery. The difference between SF-12® score in the week after the injury and the score at the time of the interview. An ordinal categorical variable. The question is about the satisfaction level with the medical care the worker received overall: 1 is “very satisfied,” 2 is “somewhat satisfied,” 3 is “somewhat dissatisfied,” 4 is “very dissatisfied.” Two-way or three-way classification. Two-way classification corresponds to employee or employer choice. Three-way classification corresponds to employee choice of prior provider, employee choice of new provider, or employer choice. A dummy variable. The value is 1 if the claim was from Pennsylvania. A dummy variable. The value is 1 if the claim was from California. A dummy variable. The value is 1 if the claim was from Texas. A dummy variable. The value is 1 if the claim was from Massachusetts. A dummy variable. The value is 1 if the employer or the insurance company chose the provider. A dummy variable. The value is 1 if the worker, a family member or friend, or the worker’s attorney chose the provider. A dummy variable. The value is 1 if the worker, a family member or friend, or the worker’s attorney chose the provider and the worker was previously treated by this provider for another medical condition. A dummy variable. The value is 1 if the worker, a family member or friend, or the worker’s attorney chose the provider and the worker was not previously treated by this provider for another medical condition. Table B.1 Definitions of Variables (continued) Worker characteristics Age Male Married Wage Hourly worker Tenure Less than high school Some high school High school graduate Some college College graduate Postgraduate Survey in Spanish Workplace characteristics Firm size ≤ 50 Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Clerical/professional Manufacturing Construction Trade Other industries Worker’s age at the time of the injury. A dummy variable. The value is 1 if the worker is male. A dummy variable. The value is 1 if the worker was married at the time of the injury. Worker’s average weekly wage. The unit is $100; that is, the variable would be 6 if the average weekly wage was $600. A dummy variable. The value is 1 if the worker was paid an hourly wage in the month before the injury. The number of years the worker was employed at the job before the injury. A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is grade school or less (0–8). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is some high school (9–11). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is high school (12). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is some college (1–3 years). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is college graduate (4 years). A dummy variable. The value is 1 if the highest level of education the worker completed and got credit for is postgraduate (professional, masters, or doctorate). A dummy variable. The value is 1 if the worker preferred to be interviewed in Spanish. A dummy variable. The value is 1 if the number of people who were employed at the location where the worker was working at the time of the injury was between 1 and 50. A dummy variable. The value is 1 if the number of people who were employed at the location where the worker was working at the time of the injury was between 51 and 250. A dummy variable. The value is 1 if the number of people who were employed at the location where the worker was working at the time of the injury was between 251 and 1,000. A dummy variable. The value is 1 if the number of people who were employed at the location where the worker was working at the time of the injury was more than 1,000. A dummy variable. The value is 1 if the worker worked in high-risk service industries at the time of the injury. A dummy variable. The value is 1 if the worker worked in low-risk service industries at the time of the injury. A dummy variable. The value is 1 if the worker worked in the clerical/professional occupations in any industry at the time of the injury. A dummy variable. The value is 1 if the worker worked in the manufacturing industry at the time of the injury. A dummy variable. The value is 1 if the worker worked in the construction industry at the time of the injury. A dummy variable. The value is 1 if the worker worked in the trade industry at the time of the injury. A dummy variable. The value is 1 if the worker worked in other industries at the time of the injury. technical appendix b 83 continued 84 t e c h n i c a l a p p e n d i x b Table B.1 Definitions of Variables (continued) Injury characteristics Back pain Nonback sprain or strain Fracture Inflammation, laceration, or contusion Other injuries Severity Treatment characteristics Overnight hospitalization Major surgery Attorney involvement A dummy variable. The value is 1 if the type of injury is back pain. A dummy variable. The value is 1 if the type of injury is nonback sprain or strain. A dummy variable. The value is 1 if the type of injury is fracture. A dummy variable. The value is 1 if the type of injury is inflammation, laceration, or contusion. A dummy variable. The value is 1 if the type of injury is not one of the above. Worker’s perceived severity. The difference between SF-12 score during the four weeks before the injury and the score during the week after the injury. A dummy variable. The value is 1 if the worker received “room and board” or “intensive care” based on the revenue code. A dummy variable. The value is 1 if the total payment for significant surgical services was greater than 0. A dummy variable. The value is 1 if the worker was represented by a lawyer when trying to collect workers’ compensation. Table B.2 Descriptive Statistics Combined Employee chose Employer chose Prior New California Employee chose Employer chose Prior New Texas Employee Employer chose chose Prior New Massachusetts Employee Employer chose chose Prior New Pennsylvania Employee Employer chose chose Prior New technical appendix b Claims in each state (%) Pennsylvania California Texas Massachusetts Worker characteristics Age (mean) Male (%) Married (%) Weekly wage (mean) Hourly worker (%) Tenure (mean, years) Less than high school (%) Some high school (%) High school graduate (%) Some college (%) College graduate (%) Postgraduate (%) Survey in Spanish (%) 20.9 23.8 24.2 31.1 43.6 59.5 69.7 $632 81.4 9.9 1.6 8.7 44.4 30.3 9.5 5.5 1.3 21.2 19.7 35.4 23.7 42.1 67.5 67.0 $609 83.5 8.8 3.2 9.3 44.7 28.7 10.6 3.5 3.6 37.5 34.1 17.1 11.3 41.5 62.5 65.6 $612 92.7 8.7 5.6 11.0 41.9 29.0 9.7 2.8 7.1 —— —— —— —— — — — — 44.3 57.2 72.9 $659 78.0 12.2 1.9 9.2 30.0 37.8 13.0 8.1 1.7 43.0 61.5 64.2 $646 79.4 9.9 3.5 6.4 29.6 33.6 14.5 12.3 5.6 40.5 55.0 57.4 $601 89.7 7.2 9.7 10.9 31.0 31.6 12.3 4.6 11.5 —— —— —— —— — — — — 41.7 57.1 65.7 $578 85.2 7.1 1.0 5.9 43.0 34.4 9.0 6.8 3.5 42.0 65.6 70.5 $544 81.8 7.5 4.2 12.4 40.9 31.8 10.1 0.6 6.9 39.5 67.1 70.4 $479 90.7 6.3 10.2 18.7 36.8 27.2 5.5 1.6 17.1 —— —— —— —— — — — — 43.8 63.3 68.8 $666 77.0 9.8 2.6 9.9 47.0 28.6 8.7 3.1 0.0 40.9 70.6 63.7 $706 89.1 8.3 2.4 7.4 44.5 33.7 8.7 3.3 0.0 42.4 74.0 68.8 $714 95.7 8.4 3.0 9.0 44.4 31.0 10.7 1.9 2.0 —— —— —— —— — — — — 44.6 59.1 71.8 $614 87.7 10.7 0.4 9.5 59.0 19.2 7.1 4.7 0.0 43.0 72.8 67.5 $573 84.0 10.3 2.3 8.9 65.4 13.4 10.0 0.0 0.0 43.1 63.7 70.0 $652 95.6 11.2 0.8 8.4 53.4 26.8 8.8 1.8 0.0 continued 85 86 t e c h n i c a l a p p e n d i x b Table B.2 Descriptive Statistics (continued) Combined California Employee chose Employer chose Employee chose Employer chose Prior New Prior New Texas Employee Employer chose chose Prior New Massachusetts Employee Employer chose chose Prior New Pennsylvania Employee Employer chose chose Prior New Workplace characteristics (%) Firm size≤50 48.5 Firm size 51–250 27.1 Firm size 251–1,000 13.4 Firm size Ͼ 1,000 11.0 High-risk services 24.1 Low-risk services 13.1 Clerical/professional services 12.8 Manufacturing 19.6 Construction 7.1 Trade 18.3 Other industries 5.1 Injury characteristics Back pain (%) 34.6 Nonback sprain or strain (%) 34.9 Fracture (%) 8.6 Inflammation, laceration, or contusion (%) 5.1 Other injuries (%) 16.8 Severity (mean)a 28.5 Treatment characteristics (%) Overnight hospitalization 9.7 Major surgery 37.8 Attorney involvement 24.1 52.6 26.2 15.5 5.8 26.1 11.6 7.6 20.6 11.3 17.9 4.9 38.1 33.1 9.0 5.7 14.1 29.8 8.6 34.8 23.6 47.2 29.5 16.0 7.2 27.6 8.3 6.3 28.8 8.4 14.7 5.9 30.0 39.7 8.6 8.5 13.1 28.7 7.7 31.3 18.5 50.7 60.7 30.0 25.0 9.6 11.8 9.8 2.5 25.7 24.7 11.5 16.2 10.7 12.0 11.4 12.1 8.3 11.5 26.9 17.5 5.5 5.9 53.1 27.0 12.8 7.1 27.8 8.8 9.7 24.6 6.0 17.8 5.3 40.4 39.0 32.4 37.8 7.1 7.1 4.9 6.4 15.3 9.7 27.6 28.7 28.3 44.7 7.3 9.2 10.5 29.7 3.0 9.9 31.8 35.3 36.5 42.1 5.2 26.9 25.6 50.1 53.6 20.5 24.7 16.1 15.7 13.3 6.1 31.0 27.0 17.5 10.5 19.3 10.2 7.8 21.3 4.2 7.0 15.9 17.8 4.4 6.1 49.6 28.1 16.0 6.3 25.7 8.2 5.7 34.4 8.7 14.6 2.6 38.1 41.3 29.1 34.7 11.8 8.7 7.7 3.6 13.3 11.6 29.7 29.9 35.1 28.5 10.3 10.1 16.1 25.0 13.5 8.6 39.5 31.8 8.6 16.3 11.5 31.7 12.3 51.8 50.7 28.0 27.2 11.1 14.5 9.1 7.6 20.6 24.1 12.1 12.7 9.5 4.8 25.6 21.3 10.2 17.1 16.8 16.9 5.1 3.2 42.8 36.0 15.0 6.2 35.0 7.5 1.7 22.9 17.2 12.9 2.8 34.7 39.2 33.9 28.1 8.5 9.0 4.8 7.9 18.1 15.8 28.9 32.0 36.7 31.9 10.4 5.4 15.6 28.5 5.9 4.8 35.3 30.5 31.0 24.6 6.2 22.7 21.9 39.7 45.5 30.1 28.6 17.8 19.7 12.4 6.2 18.7 28.0 10.1 8.5 11.2 2.2 34.2 25.7 4.8 12.3 15.6 19.6 5.4 3.8 42.0 30.6 19.4 8.1 26.1 8.1 5.1 31.2 7.6 12.7 9.1 23.8 31.0 45.8 31.5 6.8 11.4 2.8 5.9 20.7 20.1 27.8 28.3 27.3 42.8 8.4 8.1 13.4 29.6 16.4 12.2 45.6 44.0 17.6 17.2 8.5 37.1 13.8 Table B.2 Descriptive Statistics (continued) Combined Employee chose Employer chose California Employee chose Employer chose Prior New Prior New Texas Employee Employer chose chose Prior New Massachusetts Employee Employer chose chose Prior New Pennsylvania Employee Employer chose chose Prior New Satisfaction with overall care Very satisfied (%) 62.9 Somewhat satisfied (%) 22.4 Somewhat dissatisfied (%) 7.9 Very dissatisfied (%) 6.7 N 458 55.5 26.5 8.4 9.6 597 45.3 34.8 8.7 11.2 896 54.2 54.6 27.8 30.3 10.2 7.8 7.8 7.3 106 122 41.2 36.7 10.5 11.7 306 66.2 50.5 20.1 24.2 4.9 11.4 8.8 13.8 114 213 39.5 42.8 9.5 8.2 152 63.4 57.6 23.0 28.1 6.9 6.8 6.7 7.5 144 133 46.2 36.8 5.7 11.3 98 68.5 62.4 18.0 25.2 10.3 5.5 3.1 6.9 94 129 51.3 29.0 7.7 12.0 340 technical appendix b a Based on respondents’ SF-12® scores, which are scaled from 0 to 100, where 100 is the best health. The severity score is the difference between the score for the time four weeks before the injury occurred and the score one week after the injury. 87 Technical Appendix C: Discussion of Construction and Validity of Health Status, Recovery, and Perceived Severity Measures This appendix summarizes material from an earlier WCRI report (Victor, Barth, and Liu, 2003), which describes and analyzes the survey data used in the study. Here we provide evidence on the construction and validity of two important measures used in this study: perceived recovery of physical health and perceived injury severity. Both are derived from the SF-12® — the most widely used instrument for measuring health status. The validity of the measures was assessed based on internal consistency, consistency across states, consistency with other published studies, and plausible correlations with other measures, especially other markers of injury severity. In general, the results indicate that the measures are consistent with other studies and exhibit the patterns expected of valid measures. To construct the measures of health status, the survey asked workers about the following:1 ■ General health: “In general, would you say that your health was excellent, very good, good, fair, or poor in the four weeks before your injury?” 1 There are some differences in the wording used in the SF-12® and the WCRI questions that seek to use the SF-12®. These wording differences may lead to some differences in how the questions are answered. However, it is unlikely that these specific differences would have a major effect on the results. 89 90 t e c h n i c a l a p p e n d i x c ■ Limits on activities: “During a typical day in the four weeks before your injury, how limited were you in performing moderate activities such as moving a table, pushing a vacuum cleaner, bowling, or playing golf? Were you limited a lot, limited a little, or not limited at all?” ■ Limits on climbing stairs: “During a typical day in the four weeks before your injury, how limited were you in climbing several flights of stairs? Were you limited a lot, limited a little, or not limited at all?” ■ Amount accomplished: “During the four weeks before your injury, did you accomplish less than you would like with your work or other regular daily activities as a result of your physical health?” (Note that daily activities include activities outside work.) ■ Physical limitations: “During the four weeks before your injury, were you limited in the kind of work or other regular daily activities you did as a result of your physical health?” ■ Pain: “During the four weeks before your injury, how much did pain in general interfere with your normal work, including both work outside the home and housework? Would you say not at all, a little bit, a moderate amount, quite a bit, or an extreme amount?” Workers’ responses to questions regarding these measures were combined into a single scaled score ranging from zero to 100 (a higher score reflects better physical health and functioning), based on the SF-12® survey questions and scoring methodology.2 In the interview (conducted 3.0 to 3.5 years postinjury), each worker was asked to recall and report his or her health and functioning status regarding each of the previously listed dimensions from the perspective of three points in time: before 2 The methodology for scoring the SF-12® is widely accepted and well documented. A description is found in Ware, Turner-Bowker, Kosinski, and Gandek (2002). Survey questions related to mental and emotional health and functioning were treated differently, as discussed in Victor, Barth, and Liu (2003). In particular, the mental health questions were only asked at the time of the interview, in part because of special concerns about the difficulty of recalling mental health status, and also because asking all 12 questions about three periods would have significantly lengthened the time needed to conduct the entire survey. This poses a problem because the overall scores for physical health and functioning based on the SF-12® require information on mental health. The assumption was made that the mental health scores at the time of the interview prevailed at the other times (pre- and immediately postinjury). The study cited above explored the sensitivity of the SF-12® scores to using different extreme (best-case and worst-case) assumptions about mental health at these two times and found the physical health and functioning scores were very insensitive to the mental health responses. Thus, any bias from not having information on mental health pre- and postinjury should be trivial. technical appendix c 91 the injury (preinjury), just after the injury (postinjury), and at the time of interview. This information was used to create two measures: ■ Perceived recovery of physical health and functioning: measured by calculating the difference between the composite health and functioning status score at interview and the score based on health and functioning postinjury. ■ Perceived injury severity: measured by calculating the difference between the postinjury and preinjury composite health and functioning status scores. All empirical measures of severity and recovery have some inherent limitations, and these are no exception. First, it is important to emphasize that these measures are based on workers’ self-reports and may be different from health status measures based on clinical findings. Second, the reader should keep in mind that the health status reported “postinjury” is the worker’s recollection about his or her health status one week after the injury. The perceived severity measure focuses on the short-term physical consequences of the injury. Most conditions improve over time, but some may worsen before they improve. For cases in which health status worsened after the first week, both perceived severity and recovery may be understated. Third, the at-interview health status reported by some workers may have been affected by nonwork-related events — for example, automobile or other accidents or transitory or chronic diseases — that occurred or worsened after the work injury and were unrelated to the injury. In such cases, as well, perceived recovery is understated. Other workers in the study may have experienced such events before injury, thus affecting their preinjury health scores, and some may have improved after injury (along dimensions unrelated to their injuries). When this has occurred, perceived injury severity is understated and perceived recovery may be overstated. However, there is no obvious reason that these sources of mismeasurement of severity or recovery should be related to provider choice, suggesting that they do not influence our key findings. How Plausible Are the Survey Results on Health and Functioning? Because the recall concern is potentially important, a variety of analyses were used to assess how plausible these health and functioning results might be. Other than the evidence on validity that is presented in the earlier WCRI study, there is no di- 92 t e c h n i c a l a p p e n d i x c rectly relevant evidence from prior studies about the validity or invalidity of the approach used in the WCRI survey. No other study has assessed the validity of the retrospective use of the SF-12®, as done by WCRI (telephone conversation with Mark Kosinski, Senior Scientist, QualityMetric, Inc., August 31, 2005). In a study by Ware, et al. (1996) that used retrospective recall of health status at one, two and four years, the authors concluded that “Retrospective evaluations of prior health status and self-reported evaluations of changes over time may yield useful information about outcomes when baseline assessments are not available.” The study examined 1,466 patients with chronic health conditions, but not traumatic (as in workers’ compensation). Although they found some mismeasurement in the retrospective self-reported measures, they found a high correlation between the retrospective self-reported changes and measured changes of health status. They also found that the correlation was similar regardless of the recall period. The retrospectively reported changes were influenced by the patient’s health status at the time of interview, biasing the magnitudes of the retrospectively reported changes. A number of the workers’ compensation studies cited in the literature review in Chapter 2 relied on retrospective recall of the SF-12® measures (Research and Oversight Council on Workers’ Compensation (ROC) and Med-FX, LLC, 2001). The only study that assessed the validity of retrospective recall used a somewhat different recall approach. That study asked patients to recall retrospectively the change in their health as indicated by several of the SF-12® questions, and found that this approach appeared to be valid (Damiano, Pastores, and Ware, 1998). These analyses included: ■ Internal consistency ■ Consistency across states ■ Consistency with other studies ■ Plausible correlations with other measures The perceived physical health and functioning measures showed plausible patterns and reconciled well with values reported in other studies. The average preinjury scaled scores for physical health and functioning in the four states (54– 55) were remarkably consistent across the states and higher than average for the general U.S. population (50).3 One would expect the scores for an employed pop- 3 See Ware, Keller, and Kosinski (1998). technical appendix c 93 SF-12® Score Figure C.1 Perceived Injury Severity and Recovery of Physical Health and Functioning: An Example from the Texas Results 60 54 50 Severity 50 41 40 Recovery 30 27 20 10 0 Preinjury Postinjury At interview U.S. population Notes: All workers surveyed experienced more than 7 days of lost time. SF-12® scores range from 0 to 100. A higher score indicates better health. SF-12® is a registered trademark of the Medical Outcomes Trust. Source of average figures for U.S. population: Ware, Keller, and Kosinski (1998). ulation to be higher than that for the general population. Further, the scores were similar to those found in a special study of a healthy population (Airey et al., 1999, Table 3.12). In all four states, the average postinjury scores were lower than the preinjury scores (reflecting injury severity), and the average at-interview scores were higher than the postinjury scores (reflecting perceived recovery); only in a handful of cases did severity or recovery have the unexpected sign. In addition, the greater the perceived injury severity, the higher the medical payments paid or expected to be paid for the claim. Figure C.1 shows the SF-12® scores for workers’ perceived physical health and functioning for Texas. The pattern is typical of the four states. The average SF-12® score reported for Texas workers at interview was 41 points, lower than the average of 50 for the U.S. population. This is as expected because a population of injured workers should have lower average scores than the general population. In addition, the Texas score compares well with a study sponsored by the Research and Oversight Council on Workers’ Compensation (ROC) and Med-FX, LLC (2001), which found a physical health and functioning score of 38 points for a 94 t e c h n i c a l a p p e n d i x c sample of workers injured in 1998 when they were surveyed in 2000. That study found a score of 42 for workers from four other states who were similarly surveyed. The mental health and functioning score resulting from our survey (48) also compares well with that from the ROC survey (44). A later ROC survey of workers with back, neck, or shoulder injuries who were interviewed 21 to 33 months after their injuries found a score of 39 points — quite similar to the Texas score in the current study (Shields, Baroni, and Lu, 2003). Within each state, one would expect that workers reporting more-severe injuries would receive more medical care and that the care would be more expensive. Thus, one would expect to see higher medical payments for workers who reported more-severe injuries (that is, workers with larger reductions in their SF-12® perceived physical health and functioning scores from pre- to postinjury). To examine this, respondents were divided into three perceived injury severity groups based on the differences in their pre- and postinjury SF-12® scores:4 ■ Least severe injuries: postinjury scores minus preinjury scores equals –5.0 to –19.9 points. ■ Moderately severe injuries: postinjury scores minus preinjury scores equals –20.0 to –34.9 points. ■ Most severe injuries: postinjury scores minus preinjury scores equals –35.0 to –45.0 points. Table C.1 shows that medical payments were strongly correlated with respondents’ perceived injury severity. For example, workers who perceived the most severe injuries had incurred medical payments that were about double those of the least severe injury group. As previously discussed, this result is what one would expect and adds to the plausibility of the survey measures and results. To further validate the survey measures, the results obtained from one of the SF-12® questions regarding general health can be compared with the results from a similar question in another national survey — the Behavioral Risk Factor Sur- 4 The authors urge caution regarding this nomenclature. The categories of least severe, moderately severe, and most severe injuries are derived from the difference between the worker’s SF-12® scores from before the injury to just after the injury. On the other hand, the categories of less serious and more serious claims, as discussed later, are financial classifications based on data from the workers’ compensation claim and were used to select and stratify our sample. technical appendix c 95 Table C.1 Incurred Costs for Medical Care, by Perceived Injury Severity California Texas Massachusetts Mean incurred costs for medical care Least severe $13,774 Moderately severe $15,550 Most severe $25,754 Median incurred costs for medical care Least severe $5,381 Moderately severe $5,178 Most severe $10,372 $9,266 $12,236 $17,407 $3,459 $6,100 $8,851 $3,231 $6,767 $9,552 $1,425 $1,606 $3,063 Pennsylvania $6,163 $8,145 $11,960 $3,091 $3,676 $4,154 Notes: All workers surveyed experienced more than 7 days of lost time. Average incurred medical payments per claim are for 1998 injuries for Texas and for 1999 injuries for other states. All costs reflect an average of 36 months’ experience after the injury. Key: Least severe means reductions in SF-12® scores of 5.0–19.9 points; moderately severe, reductions of 20.0–34.9 points; and most severe, reductions of 35.0–45.0 points. veillance System (BRFSS) survey for 2000 (Centers for Disease Control and Prevention, 2002).5 These comparisons are reported in Table C.2. In all four states, the health status reported in the BRFSS is systematically lower than the preinjury health status of the injured workers in the WCRI survey. This difference may reflect the nature of either the survey instrument or the populations surveyed. As indicated previously, it is expected that a survey of an employed population would yield higher scores than a survey of the general population. This is reinforced by how similar our reported SF-12® scores are to those found by the ROC study of Texas (ROC and Med-FX, 2001). However, it is possible that the differences reflect recall biases in using the question retrospectively. Notwithstanding the systematic differences discussed here, it is reassuring that both surveys found that Massachusetts is the state where people report the best health and that Texas is the state where workers report the worst health, among these four states. 5 Our survey asked, “In general, would you say that your health is excellent, very good, good, fair, or poor?” The BRFSS question asked of 5,002 adults in Texas and nationwide, “How is your general health: excellent, very good, good, fair, or poor?” 96 t e c h n i c a l a p p e n d i x c Table C.2 Comparison of Two Surveys on General Health SelfReported General Health California WCRI Survey, Injured Workers, Preinjury (percent) BRFSS, General State Population (percent) Texas WCRI Survey, Injured Workers, Preinjury (percent) BRFSS, General State Population (percent) Excellent 38 24 38 20 Very good 36 32 33 28 Good 22 28 21 33 Fair 3 12 6 15 Poor 1424 Massachusetts WCRI Survey, Injured Workers, Preinjury (percent) BRFSS, General State Population (percent) 43 28 35 34 19 26 39 03 Pennsylvania WCRI Survey, Injured Workers, Preinjury (percent) BRFSS, General State Population (percent) 40 21 35 35 22 30 2 10 14 Key: BRFSS: Behavioral Risk Factor Surveillance System. Sources: WCRI DBE database; Centers for Disease Control and Prevention (2002). Technical Appendix D: Discussion of Survey Response Rates and Response Bias After a very brief description of survey methods, this appendix focuses on response rates and the representativeness of the respondents. A more complete description of the survey and an evaluation of it can be found in Victor, Barth, and Liu (2003). WCRI developed the survey in collaboration with the University of Connecticut’s Center for Survey Research and Analysis (CSRA), which conducted the survey. The survey instrument was first fielded in Texas in 2002. A year later, it was fielded in California, Massachusetts, and Pennsylvania. For the latter surveys, some questions were modified slightly and some skip patterns altered, in light of the experience gained in the Texas survey effort. CSRA conducted telephone interviews of between 629 and 754 workers in each state who were injured during portions of 1998 in Texas and during 1999 in the other states, had more than seven days of lost work time, and received workers’ compensation income benefits. The surveys were conducted in 2002 and 2003 — on average about 3.0 to 3.5 years after the injury. We selected this time lag because we wanted to know the intermediate-term consequences of injury — in particular, the recovery of health and functioning and return to work. We interviewed fewer people in Massachusetts than in the other states because the refusal rates were higher in Massachusetts. The telephone surveys were conducted seven days a week, in both day and evening hours. The survey instrument was translated into Spanish and administered in that language when requested by the worker. In Massachusetts, Pennsylvania, and Texas, state agencies provided the workers’ names and contact information we needed to draw the sample. In California, the state agency requested that WCRI use data from insurers and employers to draw the sample. All workers were assured in writing and on the telephone that 97 98 t e c h n i c a l a p p e n d i x d their responses were confidential and would not be shared with the state agencies, their employers, or the insurers. To increase the statistical power of the survey analysis, we used a random sample stratified along two dimensions: the financial seriousness of the claim and the insurance market segment.1 This improves the statistical power since self-insurance is less common than insurance arrangements, and more-serious claims are less common than less serious ones. Representativeness and Response Bias Table D.1 shows the results of the telephone survey calls. The surveyors began with a sample of workers, of which 53–62 percent could be located. For the others, either the telephone number supplied did not work or the person answering the call claimed not to know of, or know the whereabouts of, the person to be interviewed. These rates of invalid contacts were in line with those found in other studies that interviewed injured workers. In our case, 38–47 percent of workers sampled had invalid phone numbers 3.0 to 3.5 years after the injury. In a related study, surveyors found invalid telephone numbers for 28 percent of injured workers who belonged to the same craft labor union, where injuries occurred in 1995 and 1996 and the survey was conducted between November 1997 and October 1999 (Borba and Parry, 2000). An April 1998 survey of Texas workers injured in 1996 yielded 39 percent disconnected or wrong phone numbers (Research and Oversight Council on Workers’ Compensation, 1998). Clearly, the longer the period from the injury date to the survey, the harder it is to track down respondents. Not surprisingly, workers for whom the phone numbers were invalid had personal or claim characteristics that suggested they had less severe injuries and were more mobile than respondents. Workers with invalid phone numbers had medical payments that were 11–31 percent lower than those of respondents, and they were more likely to have cuts and bruises (less severe injuries), less likely to have fractures or nonback sprains (more-severe injuries), and less likely to have surgery. Workers with invalid phone numbers also had shorter durations of temporary disability and lower indemnity benefits, on average, and their claims were less likely to be open. Workers with invalid phone numbers also appeared to be more 1 “Insurance market segment” refers to the three sources of financing workers’ compensation liability: private insurance carriers, the state fund, and self-insurance. technical appendix d 99 Table D.1 Attempted Telephone Interviews and Valid Phone Numbers Type of Disposition California Texas Massachusetts Pennsylvania Total number sampled Percentage with valid phone numbers 4,110 53 2,678 55 3,114 53 3,384 62 mobile than respondents because they were more likely to be single, male, and younger. Further, workers with invalid phone numbers had much lower preinjury wages and less tenure on the job. Clearly, workers with invalid phone numbers had some differences relative to both respondents and refusers. However, these differences were not large. Of the remaining workers for whom valid phone numbers existed, the surveyors found and completed interviews with 34–51 percent. This represents a response rate of 51–61 percent in Texas, 33–48 percent in California, 37–46 percent in Massachusetts, and 36–45 percent in Pennsylvania, depending on the method used to compute the response rate (see Table D.2).2 Given the number of invalid phone numbers and the number of workers with valid phone numbers who refused to participate or who could not be contacted, three questions arise about representativeness and response bias, which we answer in the next section: 1. How do the respondents compare with all workers with claims paid in their respective workers’ compensation systems (with more than seven days of lost time)? 2. How do the respondents compare with refusals? 3. How do the respondents compare with those who could not be contacted because they did not have valid phone numbers? 2 The calculation of the response rate depends on how one chooses to interpret the validity of the live telephone numbers. In Texas, for example, if one assumes that all the live telephone numbers were valid, the response rate was 50.8 percent. If one assumes that the busy signals, no answers, and others were invalid at the same rate as the numbers overall, the response rate was 55.0 percent. If one assumes that the busy signals, no answers, and others were all invalid numbers, the response rate was 61.1 percent. Considering that more than 25 attempts were made for each phone number, the 61.1 percent response rate would not be unreasonable, although a more conservative choice is the 55.0 percent response rate. 100 t e c h n i c a l a p p e n d i x d Table D.2 Disposition of Cases with Valid Phone Numbers Type of Disposition Percentage of Cases with Valid Phone Numbers California Texas Massachusetts Pennsylvania Completed surveys Other valid phone numbers Refused Answering machine Busy signal No answer Othera 34 66 21 14 1 5 25 51 49 20 12 1 12 3 38 62 36 7 0 5 13 37 63 27 15 1 6 13 a The two biggest contributors to the “Other” category were persons who spoke foreign languages (other than Spanish) and persons who were called and asked us to call back at another time. Before we attempted to call people in this category a second time, we had reached the desired number of completed interviews. representativeness Comparing the characteristics of the respondents and their claims with the characteristics of workers in each state, the evidence indicates that the respondents were reasonably representative of workers with workers’ compensation claims in each state. For example, the average medical cost per claim for respondents was very similar to that for all workers with more than seven days of lost time in each state system. The average medical cost per claim among respondents was 2–16 percent different than medical payments among all workers with claims in the four states.3 Table D.3 shows that the survey respondents were reasonably representative of the population of injured workers whose claims were paid. It shows the similari- 3 The average medical payment per claim for each state is taken from CompScopeTM Benchmarks: Multistate Comparisons, 4th Edition, which is based on large samples of payor claim data — 33–67 percent of the claims in each state. These data come from a diverse group of payors that represent the various insurance market segments in each state. The costs measures are externally validated again data from the states and rating bureaus, and WCRI reports that they reconcile within 10 percent of the external figure (Telles, Wang, and Tanabe, 2004). Table D.3 Analysis of Representativeness California Average Respondents for Statea Texas Average Respondents for Statea Massachusetts Average Respondents for Statea Pennsylvania Average Respondents for Statea technical appendix d Worker characteristics Age (mean years) Female (percentage of claims) Single (percentage of claims) Tenure with employer (mean years) Weekly wage (mean) Industry (percentage of cases) Manufacturing Construction Clerical/professional Trade High-risk services Low-risk services Other Type of injury (percentage of cases) Back sprains and strains Fractures Inflammations, lacerations, contusions Nonback sprains and strains Other 40 41 51 6 $504 15 7 10 14 18 9 29 31 7 10 31 22 42 44 37 8 $599 15 7 9 13 19 9 28 30 10 9 36 15 39 36 39 5 $450 19 8 8 15 21 8 21 31 9 11 27 23 41 37 32 6 $519 23 7 9 14 24 12 12 32 13 8 28 20 39 27 51 5 $541 16 13 5 16 21 8 21 30 9 15 25 22 42 28 34 7 $690 18 15 5 12 21 9 20 32 13 9 29 17 40 32 44 7 $520 22 6 4 10 23 8 27 27 11 12 30 21 43 34 32 10 $624 25 7 4 13 21 7 23 26 11 8 38 17 continued 101 102 t e c h n i c a l a p p e n d i x d Table D.3 Analysis of Representativeness (continued) California Average Respondents for Statea Texas Average Respondents for Statea Massachusetts Average Respondents for Statea Pennsylvania Average Respondents for Statea Claim costs and characteristics Medical payment (mean) $10,506 $11,144 Indemnity payment (mean) $14,171 $15,553 Open claims (percentage of claims) 39 37 PPD or lump-sum payment (percentage of claims) 53 50 Lump-sum payment (percentage of claims) 22 18 Defense attorney involved (percentage of claims) 29 22 Vocational rehabilitation services (percentage of claims) 31 34 PPD or lump-sum payment (mean) $11,924 $13,400 Lump-sum payment (mean) $13,218 $15,897 Duration of temporary disability (mean weeks) 29 25 Type of medical treatment received (percentage of claims) Major surgery 20 26 Chiropractic care 12 12 $11,617 $9,523 26 55 10 9 6 $6,165 $5,181 24 29 22 $11,341 $9,385 26 58 6 4 3 $5,969 $3,955 20 31 20 $4,937 $10,791 13 23 19 17 5 $14,008 $16,482 25 18 9 $5,717 $13,156 15 17 15 19 7 $19,036 $21,264 23 27 9 $7,978 $10,735 16 17 15 19 6 $21,921 $23,410 22 26 6 $7,755 $11,181 13 9 9 15 7 $32,789 $33,182 21 36 6 Note: All values in table for claims with more than 7 days of lost time. a The “Average for State” values are from the WCRI Detailed Benchmark/Evaluation (DBE) database; each value is weighted to represent the claims in the system of that state. Key: PPD: permanent partial disability. technical appendix d 103 ties and differences for characteristics of workers, their injuries, and their claims. For example, the respondents and the injured-worker population within each state were of similar age and gender, had similar industry and injury mixes (with a few moderate exceptions), had similar medical payments and somewhat higher indemnity benefits, and had similar rates of chiropractic care. The main differences were that the respondents tended to have characteristics associated with being less difficult to find or more willing to talk about their injuries because they considered their injuries more significant than did nonrespondents. The average respondent was more likely to be married, have longer job tenure, earn a higher preinjury wage, be in a low-risk service industry, and have a fracture or nonback sprain (that is, a more severe injury), rather than a cut or bruise (that is, a less severe injury). Respondents were slightly less likely to have defense attorneys involved and to have shorter durations of temporary disability. In the two wage-loss states (Massachusetts and Pennsylvania), respondents were also less likely to have a permanent partial disability (PPD) or lump-sum payment resolve the case, but the PPD or lump-sum payment was likely to be larger, indicating a more serious case.4 None of these differences was large enough to engender concern about the representativeness of the respondent data. The four states are much more similar in their industry mix among injured workers than they are for the state economies as a whole. Of course, most of these measures are explanatory variables in our models, so we control for any differences among claims and states. refusals and response bias Workers who refused to be interviewed appear to have had less severe injuries than did respondents. One way to analyze this is to compare the average medical cost per claim of refusals with that of respondents. Medical payments reflect differences in the severity of the injury, the nature of providers used, and the attributes of the workers that influence the demand for medical care, among a variety of other factors. Refusals had medical payments that were 15–28 percent lower than those of respondents in three states, and only 3 percent lower in Pennsylvania. It is not surprising to find that, on average, workers who were unwilling to spend time on interviews had injuries that were less severe and thus less important to them. Table D.4 reinforces this conclusion. Compared with respondents, refusals had 4 For a definition and examples of wage-loss systems, see Barth and Niss (1999). 104 t e c h n i c a l a p p e n d i x d similar marital status, wages, and industry mix. However, there were a number of indicators that refusals were more likely to have relatively less serious claims (that is, lower indemnity benefits), either similar or lower durations of disability and a lower fraction of cases with PPD payments, more claims with cuts and bruises (less serious claims), and less frequent surgery. There was some concern, before conducting the survey, that workers still being represented by attorneys at the time of the interviews would be more likely to refuse to answer our questions, on advice of counsel. Although this could not be tested directly, Table D.3 shows that cases in three of the four states were slightly less likely to involve a defense attorney than the estimate for the state as a whole, which is consistent with the concern. However, Table D.4 shows that cases with defense attorneys involved were slightly more likely to refuse in California and Pennsylvania and slightly less likely to refuse in Massachusetts and Texas.5 weighting the responses The data come from payors from three different market segments — the voluntary insurance market, the residual market (market of last resort) or state fund, and the self-insured market. In addition, claims were divided into two levels of financial seriousness, and the more serious claims were oversampled, because they are relatively rare. Consequently, in all of our analyses, weights are applied to data from each of the six strata based on market segment and financial seriousness to make the claims representative of claims in each state. However, in our regression analyses, we do not weight the data by state to make the sample representative of claims in the four states. Doing so would, naturally, apply much higher weights to observations from California and Texas. In the regression analysis, this latter type of weighting would only matter if parameters differ across states, which is something we independently investigate in a number of ways. Overall, the results suggest that by not weighting across states we, if anything, understate the strength of our conclusions, because the results on the effects of provider choice are strongest for California and Texas, the two largest states. 5 We use defense attorney involvement as a proxy for worker attorney involvement, assuming that a case with a defense attorney involved is more likely to be one in which the worker has retained counsel. Table D.4 Analysis of Response Bias California Texas Massachusetts Pennsylvania Respondents Invalid Refusals Respondents Invalid Refusals Respondents Invalid Refusals Respondents Invalid Refusals Phone Phone Phone Phone Numbers Numbers Numbers Numbers technical appendix d Worker characteristics Age (mean years) 42 Female (percentage of claims) 44 Single (percentage of claims) 37 Tenure with employer (mean years) 8 Weekly wage (mean) $599 Industry (percentage of cases) Manufacturing 15 Construction 7 Clerical/ professional 9 Trade 13 High-risk services 19 Low-risk services 9 Other 28 Type of injury (percentage of cases) Back sprains and strains 30 Fractures 10 Inflammations, lacerations, contusions 9 Nonback sprains and strains 36 Other 15 37 41 37 40 47 41 5 $509 15 9 7 14 21 7 28 7 $619 11 8 9 14 23 7 28 34 30 97 11 8 30 37 15 17 41 37 32 6 $519 23 7 9 14 24 12 12 32 13 8 28 20 36 40 31 36 47 36 3 $427 23 9 6 16 27 9 10 6 $527 23 7 10 13 24 13 9 33 35 10 12 14 12 25 22 19 20 42 28 34 7 $690 18 15 5 12 21 9 20 32 13 9 29 17 36 41 23 22 54 34 4 $568 19 12 3 15 24 7 20 7 $709 19 13 4 13 25 7 19 43 34 32 10 $624 25 7 4 13 21 7 23 35 33 10 10 12 12 29 32 14 14 26 11 8 38 17 36 43 30 28 52 30 5 $514 22 7 3 13 28 7 21 9 $619 22 8 6 11 27 5 22 33 31 12 10 12 11 31 33 13 15 105 continued 106 t e c h n i c a l a p p e n d i x d Table D.4 Analysis of Response Bias (continued) California Texas Massachusetts Pennsylvania Respondents Invalid Refusals Respondents Invalid Refusals Respondents Invalid Refusals Respondents Invalid Refusals Phone Phone Phone Phone Numbers Numbers Numbers Numbers Claim costs and characteristics Medical payment (mean) $11,144 $9,414 $9,421 Indemnity payment $15,553 $14,289 $14,236 Open claims (percentage of claims) 37 28 35 PPD or lump-sum payment (percentage of claims) 50 42 49 Lump-sum payment (percentage of claims) 18 20 19 Defense attorney involved (percentage of claims) 22 25 27 Vocational rehabilitation services (percentage of claims) 34 33 32 PPD or lump-sum payment (mean) $13,400 $14,362 $11,436 Lump-sum payment (mean) $15,897 $15,592 $12,883 Duration of temporary disability (mean weeks) 25 25 25 Type of medical treatment received (percentage of claims) Major surgery 26 21 26 Chiropractic care 12 13 10 $11,341 $9,385 26 58 6 4 3 $5,969 $3,955 20 31 20 $8,540 $6,901 $8,962 $7,615 16 22 43 45 76 31 23 $5,295 $5,733 $5,411 $5,056 17 17 23 26 20 16 $5,717 $13,156 $3,925 $4,108 $11,000 $11,038 $7,755 $11,181 $6,867 $7,488 $9,371 $11,332 15 11 12 13 11 15 17 19 14 15 17 13 9 10 9 9 10 8 19 22 16 15 19 17 7 $19,036 $21,264 54 $21,629 $22,083 $22,484 $23,381 7 $32,789 $33,182 67 $28,604 $33,437 $29,523 $34,492 23 19 18 21 18 20 27 17 19 36 24 29 9 99 6 66 Note: All values in table for claims with more than 7 days of lost time. Key: PPD: permanent partial disability. Technical Appendix E: Statistical Methods This appendix explains the different statistical models used in our analysis and the interpretation of the results. We need to use different statistical models to study the alternative dependent variables because our dependent variables take differ- ent forms. For the three cost and outcome variables that are continuous (indemnity benefits, medical payments, and recovery of physical health), equation (2.1) is es- timated as a linear regression, in which case the estimated coefficient of a variable simply measures how the outcome changes with a one-unit increase in the vari- able. However, we always report the recovery results in terms of the percentage of recovery, relative to the worker’s health status before the injury. For the other out- come variables, we cannot use linear regression. In each case, there is some choice regarding exactly which type of model to use. We have chosen to use a set of mod- els for which the estimated coefficients have a very similar interpretation. The return-to-work outcome is dichotomous, and we estimate a logit model. In the logit model, Yis in equation (2.1) is replaced by an unobserved variable Y* is for the unobserved propensity to return to work. The discrete indicator Yis is then observed to equal 1 if Y* is > 0, and to equal zero otherwise. Using ZisΘ as a short- hand for the parameters and variables in equation (2.1), and assuming that the cumulative distribution for εis is the logistic, we have P(Yis = 1) = exp(ZisΘ)/[1 + exp(ZisΘ)] P(Yis = 0) = 1/[1 + exp(ZisΘ)]. (E.1) The parameters Θ are estimated by maximum likelihood. Finally, we present a transformation of the logit coefficients that is more easily interpretable. Specifically, the two expressions in equation (E.1) imply that 107 108 t e c h n i c a l a p p e n d i x e P(Yis = 1)/P(Yis = 0) = exp(ZisΘ), (E.2) which in turn implies that exp(Θk) — where Θk is the coefficient on a particular variable Zk in Z — measures the multiplicative effect on the relative probability P(Yis = 1)/P(Yis = 0) of a one-unit increase on Zk.1 For example, if exp(Θk) equals 1.2, then a one-unit increase in Zk increases the relative probability of P(Yis = 1) by 20 percent.2 The model for the duration of time out of work has to be estimated using survival models to account for the possible truncation of the period out of work. That is, it is possible that, at the time of the survey, some individuals are still in the midst of a period out of work, in which case all we know is that the period out of work lasts at least up to the time of the survey. In this framework, the outcome measure is Tis, the length of the period out of work. We estimate an accelerated failure time model, in which Tis = exp(ZisΘ + σεis). (E.3) In a standard regression model, we would assume a distribution for ε, typically normal, in which case least squares gives the maximum likelihood estimates. However, there are censored periods in which we do not know the ultimate value of Tis but only that it is at least as large as tis, because the period is still ongoing at the time of the survey. In addition, it is common in these models to fix the variance of ε at 1, and allow σ to be a parameter that is estimated. In this setting, we build the likelihood function for two types of observations. For the uncensored observations, we have an expression for the probability of observing a period of length Tis, or f(Tis). For the censored observations, all we know is that period out of work lasts at least as long as tis. The probability of this event is 1 minus the cumulative distribution function for tis, or the survivor function for tis, which we denote 1 This is also often described as measuring how the “odds” of the outcome Yis = 1 change with Z, where the odds are the relative probability. (For example, if the probability of event A is .75, and the probability of event B is .25, then the odds ratio for event A relative to event B is 3 = .75/.25.) To see this in the simplest example, suppose that Z contains an intercept and a single dummy variable, or ZisΘ = μ + D, where D is the dummy variable. Then the odds ratio when D = 0 is exp(μ), the odds ratio when D = 1 is exp(μ + ), and the ratio of the second to the first is exp( ). The same is true when D is continuous, and we think in terms of a one-unit increase in D. 2 In each case, we report the standard error such that the t-statistic is the same as on the original coefficient of the logit model. technical appendix e 109 S(tis). The density and survival functions are related through the hazard function h(tis) = f(tis)/S(tis). All that remains is to specify a distribution for ε in equation (E.3). We assume a logistic distribution for ε (a log-logistic distribution for Tis), in which case the survivor function is S(tis) = 1/[1+ {exp(−ZisΘ)tis}1/σ]. (E.4) The hazard function is more complicated, but an appealing feature of the function that results for the log-logistic distribution is that it is flexible and can be increasing monotonically, decreasing monotonically, or first increasing and then decreasing, depending on the value of σ.3 Finally, a nice feature of the log-logistic distribution is that an expression very similar to that for the logit model results; specifically, we have S(tis)/{1- S(tis)} = exp[Zis(Θ/σ) − (1/σ)ln(tis)], (E.5) which implies that exp(Θk/σ), computed from the coefficient on a particular variable Zk in Z, measures the effect of a one-unit increase in Zk on the ratio of the probabilities of the period lasting at least as long as any time t.4 This parallels the earlier interpretations of the parameters for the logit and multinomial logit models. However, it is also the case that exp(Θk/σ) equals the ratio of the expected duration when the corresponding variable Zk is one unit higher to when it is not, and therefore 100 × (exp(Θk/σ) − 1) measures the percentage by which the expected duration is longer with this change in Zk. We report these percentages in the tables.5 3 This contrasts with some more widely used distributional assumptions that impose more restrictions on the hazard function. 4 The implication of this is that the values of the regressors Z and the parameters Θ exert a proportional shift on the odds ratio in equation (E.5) for all values of t. That is, for any two individuals, who have different Z’s but the same values of Θ, the odds of the spell lasting longer than t are constant for any t. 5 In working with duration models, sometimes attention is given to the problem of unobserved heterogeneity. We do not think this is critical in our context for two reasons. First, because we have very detailed controls we do not have reason to believe that there is an important role for unmeasured heterogeneity. More importantly, the unique problem that unobserved heterogeneity introduces in duration models is bias in the estimates of parameters measuring duration dependence, because one cannot easily identify whether, for example, the probability of escaping from some status decreases over time because of duration dependence or because the sample increasingly shifts toward those likely to have (footnote continued on next page) 110 t e c h n i c a l a p p e n d i x e Finally, the satisfaction outcome is also discrete but takes on four values: very satisfied, somewhat satisfied, somewhat dissatisfied, and very dissatisfied. These values are ordered, given that the satisfaction responses can clearly be ranked. To study the outcome, an ordered discrete choice model is used. The framework is similar to that for the logit model, with Y* is now interpreted as the unobserved continuous measure of satisfaction, which follows the model Y* is = ZisΘ + εis. Now, though, the individual responds with the lowest category, Yis = 1, if Y* is < ω1, the next category, Yis = 2, if ω1 ≤ Y* is < ω2, etc., and the highest category, Yis = 4, if ω3 ≤ Y*is, with ω1 < ω2 < ω3 (the ω’s are unknown parameters to be estimated). Assuming again that the cumulative distribution function of ε is logistic, then the probability of each of these outcomes can clearly be written as a function of the same expressions used in the logit model. For example, we have P(Yis = 1) = P(Y*is < ω1) = 1/[1 + exp(ZisΘ − ω1)], (E.6) and P(Yis = 2) = P(Y*is < ω2) − P(Y*is < ω1) = {1/[1 + exp(ZisΘ − ω2)]} − {1/[1 + exp(ZisΘ − ω1)]}. (E.7) In this way the probability of each response can be written, and the likelihood function constructed. Note that in this case the relative probability of the response being in any category j+1 or higher relative to j is P(Yis ≥ j+1))/P(Yis = j) = exp(ZisΘ − ωj), (E.8) so that, paralleling the logit model, exp(Θk) measures the effect of a one-unit increase in Zk on the log of the relative probability P(Yis ≥ j+1))/P(Yis = j), or the relative probability of reporting a higher level of satisfaction. long durations. By extension, bias will also be transmitted to coefficients of variables in duration models that vary with time. However, we have neither of these. Rather, we simply have time-invariant controls that are unlikely to be affected by unobserved heterogeneity any more than would coefficients in a standard regression model. So rather than use statistical tricks to address this problem in duration models (such as assuming a functional form for the heterogeneity and integrating out), we prefer to use the data to try to address the issue. The place we think this is most important is with respect to the problem of unobserved injury severity, which was discussed in detail in the main text of the report. Technical Appendix F: Selected State System Features 112 t e c h n i c a l a p p e n d i x f Table F.1 Coverage under the State Workers’ Compensation Laws, 2004 General provisions Private employment California Texas Massachusetts Compulsory. Voluntary (see note). Selected exemptions from coverage Farm laborers Compulsory (see note). California Texas Massachusetts None. Domestic workers Generally exempt for migrant, seasonal, and other farm workers under payroll limit or for farms with fewer than three workers. None. California Texas Massachusetts Exempt if employed fewer than 52 hours in preceding 90 days or if earned less than $100. Exempt for employment incidental to personal residence. Exempt if employed fewer than 16 hours weekly. Pennsylvania Compulsory. Pennsylvania Exempt unless employer is otherwise covered, pays more than $1,200 in wages per year, or furnishes employment for more than 30 days. Pennsylvania Exempt. Massachusetts: Among employment exceptions to compulsory coverage are commission-paid salespersons and independent taxi drivers. Texas: In a 2004 survey of employers, the Texas Department of Insurance (2004) found that 38 percent of employers in Texas did not carry workers’ compensation coverage; that group — mostly smaller companies — employed about 24 percent of the workforce in the state. Texas self-insures employees of the state highway department, the University of Texas, Texas A&M University, and state employees. Other public employers are self-insured, risk-pool insured, or insured through commercial regulated carriers. The level of nonparticipation has been declining, from 44 percent in 1993, to 39 percent in 1996, and down to 35 percent in 2001. Under HB 2600, Article 16, employers who do not have workers’ compensation insurance cannot take a preinjury waiver of an injured worker’s right to sue for damages under the common law. Previously, a worker who signed a waiver could not sue his or her nonsubscribing employer in case of an injury. Sources: Texas Department of Insurance, 2004; U.S. Chamber of Commerce, 2004. Table F.2 Medical Cost Containment Strategies, 2004 Who can be a treating provider (2001) California Texas Treating provider can be medical doctor, psychologist, osteopath, chiropractor, podiatrist, dentist, optometrist, acupuncturist. Treating provider can be medical doctor, osteopath, chiropractor, podiatrist, dentist, optometrist. Initial choice of provider California Employer for first 30 days, unless worker predesignated a treating physician (see note). Employee change of provider California Once after 30 days, or after 90 to 365 days if HCO arrangement (see note). Medical fee schedule California Yes. Texas Worker from commission-approved list; a nonsubscriber typically writes into the plan that the employer chooses the provider (see note). Texas With commission’s approval using stated criteria. Texas Yes (see note). Massachusetts State statutes do not define who can be designated a treating provider. Some examples included in the state’s fee schedule for workers’ compensation providers are psychologist, medical doctor, osteopath, chiropractor, podiatrist, dentist, optometrist, independent nurse practitioner, physician assistant, certified nurse anesthetist (CRNA), licensed social worker (LICSW), physical therapist, and occupational therapist. Massachusetts Worker; if a PPA exists, worker may be required to have first appointment with provider from plan. Massachusetts Once within the same specialty. Massachusetts Yes. Pennsylvania Treating provider can be nonphysician. Pennsylvania Employer directs choice for first 90 days by posting list of six or more designated health care providers; worker choice if no panel is posted. Pennsylvania After 90 days without restriction (if a panel is posted) or at any time (if no panel is posted). Pennsylvania Yes. continued 113 technical appendix f 114 t e c h n i c a l a p p e n d i x f Table F.2 Medical Cost Containment Strategies, 2004 (continued) Hospital payment regulation California Texas Massachusetts Yes. DRG-based fee schedule for inpatient services. Treatment guidelines California Yes. Reimbursement paid on per diem basis. Texas Yes. Hospital-specific percentage discounts established annually; alternative rates and services can be negotiated. Massachusetts Yes, since 1995; advisory; nine guidelines in place (see note). Optional as of June 17, 2001. If elected, guidelines have to be nationally recognized, scientifically valid, and outcome based (see note). Yes; used in conjunction with UR program; 28 guidelines in place, developed through consensus-based, multidisciplinary effort. Pennsylvania Yes. Based on 113% of Medicare plus pass-through costs. Pennsylvania No. California: Senate Bill 899 allows employers to establish medical treatment networks, effective January 1, 2005; injured workers who do not predesignate a treating physician must receive care only through the network. Under SB 899, an employee may be treated by a predesignated physician from the date of injury if all of the requirements for predesignation are met. AB 749 reduced the offer of two HCOs to one. If HCO or personal physician is not predesignated prior to injury, employee will be treated by the HCO selected by the employer. SB 228 required the Commission on Health and Safety and Workers’ Compensation (CHSWC) to conduct a study and evaluation of existing treatment utilization standards by July 1, 2004, and to issue a report of its findings by October 1, 2004. Further, the legislation required the administrative director in consultation with CHSWC to adopt a medical treatment utilization schedule by December 1, 2004, based on CHSWC study recommendations. All employers are required to adopt utilization review systems consistent with the utilization schedule. Texas: HB 2600 called for an introduction of regional provider networks. At present, however, injured workers are not required to choose a doctor participating in a network but must choose a provider who is on the Commission’s Approved Doctor List (ADL). As of September 1, 2003, all health care providers practicing in the workers’ compensation system must be on the ADL, which means that they must be trained, have applied, and been approved to practice workers’ compensation. In practice, the fee schedule is reviewed every four to six years. A new Medical Fee Guideline was adopted in April 2002, but a legal challenge delayed its implementation until August 1, 2003. The Workers’ Compensation Commission is considering adopting treatment/loss time guidelines. Key: DRG: diagnostic related group; HCO: health care organization; PPA preferred-provider arrangement; UR: utilization review. Source: Tanabe and Murray, 2001. Table F.3 Statutory Benefit Parameters for Weekly Temporary and Permanent Partial Disability Payments Effective Date TTD Benefit Rate Maximum Benefit Minimum Benefit (not to exceed AWW or percentage of AWW, as noted) California (see note) 7/1/99–6/30/00 Massachusetts 10/1/99–9/30/00 Pennsylvania (see note) 1/1/99–12/31/99 Texas (see note) 9/1/98–8/31/99 662 3% of AWW 60% of AWW 662 3% of AWW 70%; if hourly wage more than $8.50 per hour, then 75% of AWW (for 26 weeks) $490.00 $749.69 $588.00 $523.00 $126.00 or worker’s AWW, whichever is less $149.93 or worker’s AWW, whichever is less $294.00 or 90% of worker’s AWW, whichever is less $78.00 technical appendix f California: Prior to AB 749, the maximum and minimum benefits were changed by periodic legislation, rather than by automatic annual adjustments tied to annual changes in the state average weekly wage. Where changes were made, the legislative changes took effect on July 1 of the relevant years. California has three benefit tiers: Workers receive 100 percent of their AWW up to $126, then $126 up to an AWW of $189, and then two-thirds of their AWW up to the maximum. Under legislation signed into law February 15, 2002, maximum temporary disability benefits were increased to $602 a week effective January 2003 and to $840 a week by 2006. Pennsylvania: If the statutory benefit rate is less than 50 percent of the statewide average weekly wage (SAWW), the benefit must be calculated using the lower of 50 percent of the SAWW or 90 percent of the worker’s AWW. The minimum benefit is the point at which benefits computed using the statutory rate are subject to recalculation. Annual increases in benefits go into effect January 1. Texas: Temporary total disability benefits are called temporary income benefits in Texas. For workers who earn less than $8.50 an hour, the benefit rate is 75 percent of their AWW for the first 26 weeks; the benefit rate reverts to 70 percent after 26 weeks. The minimum weekly benefit for temporary disability is 15 percent of the statewide average weekly wage for manufacturing production workers. Key: AWW: average weekly wage; TTD: temporary total disability. Source: State workers’ compensation statutes. continued 115 116 t e c h n i c a l a p p e n d i x f Table F.3 Statutory Benefit Parameters for Weekly Temporary and Permanent Partial Disability Payments (continued) Effective Date PPD Benefit Rate (percentage of AWW) Maximum Benefit Minimum Benefit (not to exceed AWW or percentage of AWW, as noted) California (see note) 7/1/99–6/30/00 Massachusetts (see note) 10/1/99–9/30/00 6623% of AWW n.a. Pennsylvania (see note) 1/1/99–12/31/99 Texas (see note) 9/1/98–8/31/99 6623% of AWW 70% of AWW $140.00–$230.00 For scheduled benefits, statutory amount based on the SAWW at time of injury $588.00 $366.10 $70.00 n.a. $294.00 or 90% of worker’s AWW, whichever is less $78.00 Table F.3 Statutory Benefit Parameters for Weekly Temporary and Permanent Partial Disability Payments (continued) California: For a disability rating of greater than 70 percent, workers receive a lifetime pension, an additional benefit paid for life. The benefit is 1.5 percent of the worker’s AWW for each percentage-point rating over 60 percent, up to the maximum earnings limit for the date of injury. PPD weekly payments, now $140 a week, will increase to $230 a week in 2006. Legislation passed in April 2004 made a number of changes to permanent disability benefits, including increasing benefits paid to severely injured workers (72 percent rating or higher) by nearly doubling the number of weeks per rating point and reducing the benefits paid to workers with a rating of 15 percent or less by reducing the number of weeks per rating point. Massachusetts: Massachusetts does not pay benefits for unscheduled permanent disability. Instead, the state pays benefits for wage loss or loss of wageearning capacity through partial disability benefits. These benefits are paid at 60 percent of the difference between a worker’s preinjury and actual wages or earning capacity, but not more than 75 percent of what the worker would receive for total disability benefits if eligible, or two times the SAWW. Pennsylvania: Table entries are for scheduled benefits only. Scheduled benefits are called specific-loss benefits in Pennsylvania. There are two different periods of payments under specific loss: for the healing period and for the specific loss itself. By statute, benefits are paid for the healing period before benefits are paid for the specific loss. The healing period ends when the worker returns to work at the preinjury wage or the period specified in the statute ends. Pennsylvania does not pay benefits for unscheduled permanent disability. Instead, the state pays benefits for wage loss or loss of wage-earning capacity through partial disability benefits. Those benefits are paid at 6623 percent of the difference between the preinjury and current actual or imputed wages, subject to the total disability maximum. If the benefit at the statutory rate is less than 50 percent of the SAWW, the benefit must be calculated using the lower of 50 percent of the SAWW or 90 percent of the worker’s AWW. The minimum benefit column in the table lists the point at which benefits computed using the statutory rate are subject to recalculation. Texas: PPD benefits in Texas are called impairment income benefits (IIBs). A worker may receive a supplemental income benefit (SIB) when IIBs end. Four conditions must be met: (1) the worker’s impairment rating must be at least 15 percent, (2) the worker has not taken an advance payment of benefits due (commutation), (3) the worker has not returned to work or is unable to earn at least 80 percent of the preinjury weekly wage, and (4) the worker has made a good-faith effort to find suitable work. The SIB is calculated at 80 percent of the difference between 80 percent of the worker’s average weekly wage and the worker’s earnings over the reporting period and cannot exceed 70 percent of the SAWW. Eligibility for SIB terminates at 401 weeks after date of injury. Key: AWW: average weekly wage (preinjury); n.a.: not applicable; PPD: permanent partial disability; SAWW: statewide average weekly wage. Sources: State statutes; California Division of Workers’ Compensation; Massachusetts Department of Industrial Accidents; Pennsylvania Bureau of Workers’ Compensation; Texas Workers’ Compensation Commission. technical appendix f 117 118 t e c h n i c a l a p p e n d i x f Table F.4 Waiting Period and Limits on Duration of Temporary Disability Benefits, 2002 Waiting period before income benefits commence California Texas Massachusetts Pennsylvania After 3 days of lost After 7 days of lost time. time. After 5 days of lost time. Limitations on duration of weekly temporary disability benefits After 7 days of lost time. California Texas Massachusetts Pennsylvania 5 years or until RTW or condition determined permanent and stationary (see note). 104 weeks of benefits, or until RTW, maximum medical improvement, or physician approves RTW and worker has bona fide offer of employment at preinjury wage. 156 weeks from injury, or until RTW or treating or impartial physician approves RTW and worker refuses suitable job. Duration of disability, unless adjudicated or agreed to; since June 24, 1996, 104 weeks for workers with less than 50% permanent impairment; no limit for workers with greater than 50% impairment. California: The 2004 legislation set a limit of 104 weeks of paid temporary disability within two years of the first temporary payment except for specified injuries that usually require extended recuperation (the limit had been 5 years). There is no explicit cost-of-living adjustment for benefits. However, if any temporary total disability (TTD) payment is made for two or more years from the date of injury, the amount of the benefit is adjusted to the TTD rate in effect at that time, based on the worker’s average weekly wage. Key: RTW: return to work. technical appendix f 119 Table F.5 Comparative Statistics on Costs, Injury Frequency, and Other Metrics of Workers’ Compensation Systems CA MA PA TX Comparative Source Benchmark Injuries/illnesses per 100 workers (2001) All cases Lost-time cases WC claim cost per worker (2000 policy year) WC cost per claim (2000 claims with experience to 2003) Total benefits Medical Indemnity Litigiousness Percentage of cases with defense attorney involvement (2000) Average payment to defense attorney (a proxy for hours billed/intensity of litigation) Percentage of people without health insurance (non-WC, 2001–2003) Fee schedule (percentage different from state Medicare rates, 2001) Overall Office visits Surgery Physical medicine Radiology 6.0 3.3 1,292 $51,212 $25,996 $25,216 29.7 4,063 18.7 12 –10 36 –1 14 5.1 2.5 388 $24,320 $6,913 $17,407 19.8 2,436 9.6 –13 –36 –4 –3 –12 n.a. n.a. 547 $35,921 $16,468 $19,453 19.8 3,500 10.7 17 –6 42 8 27 4.9 2.5 521 $41,755 $27,757 $13,998 11.1 1,884 24.6 38 –8 75 26 60 US 5.7 2.8 Median of 46 states 460 Median of 46 states $29,458 $14,611 $14,847 Median of 12 large states 24.7 2,415 15.1 44 9 80 9 66 a b b c d e continued 120 t e c h n i c a l a p p e n d i x f Table F.5 Comparative Statistics on Costs, Injury Frequency, and Other Metrics of Workers’ Compensation Systems (continued) CA MA PA TX Comparative Source Benchmark Network penetration rate (percentage of payments to network providers, 2000) Overall Office visits Surgery Physical medicine Radiology None 49 24 23 30 52 25 21 33 53 19 21 33 44 31 23 27 55 25 33 35 f a U.S. Department of Labor, Bureau of Labor Statistics, 2001. b Derived from NCCI data. c Telles, Wang, and Tanabe, 2005. d DeNavas-Walt, Proctor, and Mills, 2004. e Eccleston, Laszlo, Zhao, and Watson, 2002. f Wang and Zhao, 2003. Key: n.a.: not available; WC: workers’ compensation. Technical Appendix G: Tests of Pooling versus Individual State Regressions The main estimates reported in Tables 3.1 and 4.2 pool the data across the four states. The justification for doing this is the additional statistical precision we get from the larger sample, compared with analyzing each state separately. Pooling observations across the four states, however, restricts the coefficients of the models we estimate to be the same in each state, with the exception of the intercept, which is allowed to differ via the state dummy variables. To see this, note that another version of estimating the models separately by state is to allow a full set of interactions of the control variables with the state dummy variables. In the generic case, equation (2.1) would become Yis = α + CHOICEisβ + CHOICEis·STATEsβ′ + WORKERisγ + WORKERis·STATEsγ′ + FIRMisδ + FIRMis·STATEsδ′ + INJURYisθ + INJURYis·STATEsθ′ + STATEsκ + TREATMENTisλ + TREATMENTis·STATEsλ′+ εis.1 (G.1) The combined model used in Tables 3.1 and 4.2 assumes that β′, γ′, δ′, θ′, and λ′ are zero, in which case equation (G.1) reduces to equation (2.1). We can, however, test this set of restrictions by estimating the expanded model in equation (G.1) and then testing these restrictions explicitly. 1 In the case of a linear regression model, as we estimate for indemnity and medical payments and recovery, using estimates of this expanded regression model to obtain the effects for each state would give us exactly the same coefficient estimates as we obtain when we estimate the models separately by state. Only the standard errors of the estimates differ slightly. 121 122 t e c h n i c a l a p p e n d i x g Table G.1 Two-Way Classification of Provider Choice, Effects of Provider Choice on Outcomes, Tests of Pooling Restrictions Indemnity Benefits Medical Benefits Duration of Temporary Disability Substantial Return to Work Perceived Recovery Satisfaction Model 1: Without treatment controls Control coefficients equal across states Provider choice coefficients equal across states Model 2: With treatment controls Control coefficients equal across states Provider choice coefficients equal across states (1) 0.0490 0.4048 0.0017 0.2426 (2) 0.1494 0.5744 0.0139 0.2443 (3) <0.005 0.8309 <0.0001 0.8508 (4) 0.8732 0.9706 0.8152 0.9484 (5) <0.0001 0.8229 0.0006 0.6809 (6) 0.4611 0.6530 0.4254 0.7579 Note: P-values are reported for the restrictions specified in the leftmost-column. Table G.2 Three-Way Classification of Provider Choice, Effects of Provider Choice on Outcomes, Tests of Pooling Restrictions Indemnity Benefits Medical Benefits Duration of Temporary Disability Substantial Return to Work Perceived Recovery Satisfaction Model 1: Without treatment controls Control coefficients equal across states Provider choice coefficients equal across states Model 2: With treatment controls Control coefficients equal across states Provider choice coefficients equal across states (1) (2) (3) 0.0593 0.1212 0.1661 0.7096 0.0024 0.1900 0.0127 0.5893 <0.005 0.7295 <0.0001 0.9269 (4) (5) 0.8887 0.5288 <0.0001 0.4626 0.8248 0.5429 0.0005 0.4012 (6) 0.5484 0.2823 0.5088 0.2912 technical appendix g Note: P-values are reported for the restrictions specified in the left-most column. 123 124 t e c h n i c a l a p p e n d i x g We do this for the two-way and three-way classification of choice. We also consider this set of restrictions separately for the effects of provider choice (CHOICE), which is our direct concern, as well as the effects of the other control variables. The results of these statistical tests are reported in Tables G.1 and G.2, for the two-way and three-way classifications, respectively. We report the p-values from the tests. A p-value below 0.05 means that the set of tested coefficients are jointly significant at the five-percent level, and a p-value below 0.1 means that the set of tested coefficients are jointly significant at the 10 percent level. The results from these tests are quite clear. There is not a single case in which we reject the restrictions that the coefficients of the provider choice variables are equal across states at the ten-percent level, and in most cases the evidence against this restriction is very weak (as reflected in p-values closer to 1). This suggests that the estimates we obtained by pooling across states and restricting all coefficients except the intercept to be the same across states are not biased by imposing invalid restrictions.2 There is only one type of evidence suggesting that this conclusion may be unwarranted. Specifically, in contrast to the test of the restrictions on the provider choice coefficients, we frequently reject the restrictions that the coefficients of the other control variables are equal. These findings suggest that we reestimate the models continuing to restrict the coefficients of the provider choice variables to be the same across states, while freeing up the other coefficients; that is, in models corresponding to equation (G.1), we only restrict β′ to be zero. This is of interest, because it is possible that incorrectly restricting the coefficients of the other control variables to be equal across states could, in principle, bias the estimates of the provider choice coefficients. The results from this approach are reported in Tables G.3 and G.4. In each case, the top panel repeats the estimates from the analysis in the main text in which all slope coefficients are constrained to be equal across states. The bottom panel then reports the results freeing up the coefficients other than those for provider choice, which are the specifications to which the statistical tests in Tables G.1 and G.2 2 We also checked whether we obtained sharper evidence of differences across states if we group the four states into two pairs — states with employer choice and states with employee choice. However, as the results from Chapter 5 suggest, the estimates were generally more similar for California and Texas, on the one hand, and Massachusetts and Pennsylvania, on the other. Since in each of these pairs there is one employee choice state and one employer choice state, it is clear that when we group by the choice regime, there is also no (and even less) evidence of differences in effects of whether the worker or employer actually chose the provider. technical appendix g 125 Table G.3 Impact of Employee Choice Compared with Employer Choice Model 1: Without Treatment Controls (percent) Model 2 With Treatment Controls (percent) Table 3.1 estimates Medical payments 21** ($1,868) 10* ($903) Indemnity benefits 15* ($1,908) 8 ($978) Duration 32** 23** Substantial return to work –19† –16† Perceived recovery 01 Satisfaction 57** 59** Estimates including control variables interacted with state dummy variables Medical payments 19** ($1,689) 9† ($813) Indemnity benefits 13† ($1,693) 7 ($851) Duration 31** 25** Substantial return to work –21* –20† Perceived recovery 01 Satisfaction 58** 58** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. In model 1, for example, when the worker chose the provider, medical payments were $1,868 greater per case (on average) than when the employer chose the provider; or the duration of time out of work was 32 percent longer when the worker chose the provider than when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Table 2.2 provides the breakdown of observations by state. Observations are weighted to be representative of claims within each state; see the discussion on weighting in Technical Appendix D. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. 126 t e c h n i c a l a p p e n d i x g Table G.4 Impact of Employee Choice of Prior and New Providers Compared with Employer Choice Employee Chose a Prior Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Employee Chose a New Provider Model 1: Without Treatment Controls (percent) Model 2: With Treatment Controls (percent) Table 4.2 estimates Medical payments 22** ($1,924) 7 ($629) 20** ($1,745) 12* ($1,052) Indemnity benefits 9 ($1,116) –1 (–$162) 20** ($2,538) 15† ($1,879) Duration 17† 7 48** 40** Substantial return to work –4 3 –28** –28** Perceived recovery –3 –1 2 3 Satisfaction 86** 89** 38** 39** Estimates including control variables interacted with state dummy variables Medical payments 19** ($1,693) 6 ($536) 19** ($1,615) 11† ($969) Indemnity benefits 5 ($642) –3 (–$441) 20** ($2,540) 15† ($1,875) Duration 14 9 48** 42** Substantial return to work –3 3 –33** –34** Perceived recovery –3 –1 3 3 Satisfaction 80** 79** 42** 42** Notes: The results are interpreted as the difference in costs or outcomes when the employee chose the provider compared with when the employer chose the provider. In model 1, for example, when the worker chose a prior provider, medical payments were $1,924 greater per case (on average) than when the employer chose the provider; or the duration of time out of work was 17 percent longer when the worker chose a prior provider than when the employer chose the provider. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. Observations are weighted to be representative of claims within each state. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. technical appendix g 127 generally lead. It turns out, though, that the less restrictive estimates in the bottom panels are scarcely changed, and certainly none of the conclusions are. Thus, restricting the effects of provider choice to be equal across states is consonant with the data, and freeing up other restrictions that are rejected by the data does not change the results for the effects of provider choice on the workers’ compensation outcomes we study. Technical Appendix H: Full Regression Results 129 130 t e c h n i c a l a p p e n d i x h Table H.1 Determinants of Employee vs. Employer Choice of Provider, by State, Odds Ratios from Logit Models California Texas Massachusetts Pennsylvania State Pennsylvania California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school High school graduate Some college College graduate Postgraduate Survey in Spanish Workplace controls Firm size≤50 Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Clerical/professional services —— —— —— —— — — — — 0.999 1.197 1.457* 1.019 0.592* 1.039** 0.476 0.864 — 1.132 0.980 1.747 0.539 1.019* 0.953 0.893 1.056 0.674 1.001 0.301** 0.718 — 1.100 1.397 1.024 0.777 0.983† 0.902 0.959 0.947 0.213** 1.027† 1.878 0.818 — 1.070 0.800 1.106 <0.001 — 0.903 0.789 0.919 1.095 1.508 — — 0.703 0.631† 1.011 0.609 0.797 — — 0.616† 0.633 1.066 0.259† 0.678 — — — — — 1.011 1.227 0.919 0.891** 0.152** 0.999 1.683 0.800 — 0.491** 0.701 0.820 — 1.120 1.090 1.372 0.789 0.900 — technical appendix h 131 Table H.1 Determinants of Employee vs. Employer Choice of Provider, by State, Odds Ratios from Logit Models (continued) California Texas Massachusetts Pennsylvania Manufacturing Construction Trade Other industries Injury controls Back pain Nonback sprain or strain Fracture Inflammation, laceration, or contusion Other injuries Severity Treatment controls Overnight hospitalization Major surgery Attorney involvement N 0.497† 1.657 1.427 1.108 1.965† 0.990 1.097 — 1.478 0.984* 0.848 1.263 2.243** 538 0.351** 0.359† 0.687 1.537 1.678 1.346 1.504 — 1.189 1.011 0.654 1.315 1.256 481 0.555 0.290 0.631 0.684 0.953 0.851 1.044 — 0.760 1.019* 0.808 2.002** 1.182 376 0.795 1.322 1.048 0.456 1.421 1.277 1.629 — 1.950† 0.983** 1.623† 1.191 1.387 565 ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Table H.2 Determinants of Employee Choice of Prior and New Providers vs. Employer Choice of Provider, by State, Odds Ratios from Multinomial Logit Models 132 t e c h n i c a l a p p e n d i x h California Prior New Texas Prior New Massachusetts Prior New Pennsylvania Prior New State Pennsylvania California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school High school graduate Some college College graduate Postgraduate Survey in Spanish — — —— — — —— — — —— — — —— — — — — ——— ——— ——— ——— 1.005 0.984 1.747** 1.012 0.491* 1.049** 0.490 1.054 — 1.271 1.002 1.184 0.287† 0.999 1.387 1.172 1.021 0.647 1.029* 0.425 0.663 — 1.002 1.010 2.344† 0.789 1.012 0.734 0.825 1.050 1.145 1.004 0.116** 0.439† — 0.946 1.145 4.058† 0.503 1.021* 1.106 0.969 1.047 0.552† 1.002 0.422† 0.870 — 1.188 1.607 0.094† 0.799 0.991 0.892 1.093 0.905* 0.146** 1.038** 1.189 1.004 — 1.080 0.799 0.913 <0.001 0.978† 0.933 0.793 0.995 0.391† 1.016 4.344 0.760 — 1.129 0.855 1.320 <0.001 1.018† 0.818 0.938 0.949 0.244** 0.989 0.546 0.981 — 0.615† 0.595 1.618 1.006 1.902** 0.942 0.821** 0.088** 1.006 2.636 0.664 — 0.370** 0.812 <0.001** Table H.2 Determinants of Employee Choice of Prior and New Providers vs. Employer Choice of Provider, by State, Odds Ratios from Multinomial Logit Models (continued) California Prior New Texas Prior New Massachusetts Prior New Pennsylvania Prior New technical appendix h Workplace controls Firm size≤50 Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Clerical/professional services Manufacturing Construction Trade Other industries — 1.190 0.893 1.925† 1.186 1.272 — 0.477 1.665 2.124† 1.237 — 0.727 0.737 0.339† 1.005 1.837 — 0.562 1.763 1.024 1.166 — 0.666 0.736 1.575 0.456† 0.825 — 0.104** 0.180** 0.549 0.915 — 0.753 0.657 0.753 0.733 0.817 — 0.638 0.563 0.876 2.196 — 0.593† 0.523† 1.218 0.201† 0.488 — 0.455 0.185† 0.487 0.625 — 0.604† 0.731 0.874 0.384 1.022 — 0.674 0.442 0.851 0.704 — 1.171 0.944 2.030† 0.344* 0.576 — 0.518 0.487 0.661 0.269* — 1.104 1.326 0.742 2.527 2.096 — 1.722 3.863† 2.099 0.936 continued 133 134 t e c h n i c a l a p p e n d i x h Table H.2 Determinants of Employee Choice of Prior and New Providers vs. Employer Choice of Provider, by State, Odds Ratios from Multinomial Logit Models (continued) California Prior New Texas Prior New Massachusetts Prior New Pennsylvania Prior New Injury controls Back pain Nonback sprain or strain Fracture Inflammation, laceration, or contusion Other injuries Severity Treatment controls Overnight hospitalization Major surgery Attorney involvement N 1.849 0.808 1.036 — 1.713 0.979* 1.760 1.096 1.047 — 1.167 0.990 0.752 0.442† 1.117 — 0.606 1.001 2.596* 2.445† 1.896 — 1.796 1.015† 0.451 1.162 1.215 1.303 2.015** 2.454** 534 0.828 0.539 1.817† 1.110 0.691 1.773† 479 1.379 1.221 1.553 — 1.165 1.009 0.664 0.566 0.763 1.810 2.376 1.881 — 0.491 1.038** — 3.431* 0.986† 1.366 0.819 1.618 — 1.384 0.982* 0.923 0.671 2.239** 1.909* 1.533 0.885 375 2.348** 1.040 0.986 1.543† 1.315 1.467 563 ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Table H.3 Effects of Provider Choice on Medical and Indemnity Benefits Medical Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Provider choice Employee chose Employee chose prior provider Employee chose new provider State California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school Some college College graduate 21** — — 29** 40** –42** 1** –5 13* 3** 17† –1† 7 19† –9 –10 10* — — 31** 41** –23** 0† –12* 8† 2† 13† 0 5 20* –4 –6 — 22** 20** 29** 39** –42** 1** –5 12* 3** 16† –1† 6 19† –10 –11 — 7 12* 31** 41** –23** 0* –13** 7 2† 13† 0 5 20* –5 –7 Indemnity Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 15* — — 31** –12 5 2** 6 –2 7** 3 –1† 33† 38** 10 10 8 — — 28** –12 13 1** 1 –6 6** 1 0 30† 38** 12† 9 —— 9 –1 20** 15† 31** –13 5 28** –13 14 2** 5 –3 7** 2 –1† 32† 37** 10 10 1** 1 –6 6** 0 0 29 38** 11 9 continued 135 technical appendix h Table H.3 Effects of Provider Choice on Medical and Indemnity Benefits (continued) 136 t e c h n i c a l a p p e n d i x h Medical Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Indemnity Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Worker controls (continued) Postgraduate Survey in Spanish Workplace controls Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Manufacturing Construction Trade Other industries Injury controls Back pain Nonback sprain or strain Fracture Other injuries Severity –27† 46** 1 0 16 –15 3 1 14 7 –17 65** 54** 34** 42** 2** –18 40** 2 –3 23** –12 –2 –14 –1 3 –12 52** 13 26* –6 1** –27† 46** 1 0 16 –15 5 2 15 9 –16 65** 54** 34** 42** 2** –18 40** –25 39* 1 –3 23** –12 –1 –13 –1 5 –11 10 25** 7 –3 –6 7 31† –1 4 52** 71** 13 40** 26* 13 –5 27† 1** 2** –22 32† 11 23** 12 0 –8 –3 23 –2 8 60** 12 7 –5 2** –25 39* 10 24** 8 –3 –5 7 31† 0 5 71** 40** 13 28† 2** –21 31† 11 23** 14 0 –7 –3 23 –1 8 60** 12 7 –5 2** Table H.3 Effects of Provider Choice on Medical and Indemnity Benefits (continued) Medical Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Indemnity Benefits (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Treatment controls Overnight hospitalization Major surgery N — — 1,954 148** 118** 1,954 — — 1,945 145** 118** 1,945 — — 1,951 89** 73** 1,951 — — 1,942 88** 73** 1,942 technical appendix h Notes: The average medical payment is $8,713 in the two-way classification and $8,688 in the three-way classification. The average indemnity payment is $12,709 in the two-way classification and $12,714 in the three-way classification. We divide the coefficients by the average payments to get the percentage effect. Some claims have missing values for some measures such as married, wage, hourly worker, tenure, education, firm size, industry, overnight hospitalization, or major surgery. To include these claims in the regressions, dummy variables are created to indicate the missing information, but we do not show the coefficients of these dummy variables in the table. The omitted reference categories are employer choice; Pennsylvania; high school graduate, firm size ≤ 50; professional/clerical services; and inflammation, laceration, or confusion. An intercept is also included but not reported. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. 137 Table H.4 Effects of Provider Choice on Duration and Substantial Return to Work 138 t e c h n i c a l a p p e n d i x h Duration (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Substantial Return to Work (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Provider choice Employee chose Employee chose prior provider Employee chose new provider State California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school Some college College graduate 32** — — 35** 53** 24* 1** –6 9 0 29** 0 337** 34* –6 5 23** — — 32** 53** 31** 1** –10 5 0 30** 0 314** 35** –4 3 — 17† 48** 36** 52** 25* 1** –6 8 1 30** 0 334** 33* –7 3 — 7 40** 33** 51** 31** 1** –11† 4 0 30** 0 310** 34** –6 1 –19† — — –37** –56** –31* –3** 28* –1 0 –18 0 –74** –43** 29† 19 –16† — — –35** –57** –36** –3** 34** 1 0 –17 0 –74** –42** 28† 22 — –4 –28** –37** –55** –32* –3** 31* 0 0 –18 0 –74** –43** 31* 21 — 3 –28** –35** –56** –37** –3** 37** 2 0 –17 0 –74** –42** 30† 23 Table H.4 Effects of Provider Choice on Duration and Substantial Return to Work (continued) Duration (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Substantial Return to Work (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 technical appendix h Postgraduate Survey in Spanish Workplace controls Firm size 51–250 Firm size 251–1,000 Firm size Ͼ 1,000 High-risk services Low-risk services Manufacturing Construction Trade Other industries Injury controls Back pain Nonback sprain or strain Fracture Other injuries Severity –26† –25† –27† –26† 52 52 51 51 50† 40 47† 37 –15 –13 –14 –11 –6 –6 –6 –6 –2 –2 0 –1 2 1 1 0 –14 –14 –13 –13 –14 –8 –13 –5 80* 74* 77* 70* 0 11 0 10 8 6 7 5 10 16 9 15 22 25 20 23 4 1300405 26 23 23 20 14 15 14 16 16 24 17 24 –4 –6 –6 –7 0 10 0 9 3 2 4 3 54** 41** 55** 41** –39* –33† –39* –33† 20 –4 21 –3 11 23 10 23 19 12 19 12 –7 0 –8 –1 9 –17 10 –16 –14 2 –15 0 2** 2** 2** 2** –1 0 0 0 139 continued 140 t e c h n i c a l a p p e n d i x h Table H.4 Effects of Provider Choice on Duration and Substantial Return to Work (continued) Duration (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Substantial Return to Work (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Treatment controls Overnight hospitalization Major surgery N — — 1,829 140** 73** 1,829 — — 1,820 138** 73** 1,820 — — 1,829 –56** –18 1,829 — — 1,820 –56** –18 1,820 Notes: To get the percentage effect, we take 100 × (ecoefficient – 1) for duration and 100 × (odds ratio – 1) for substantial return to work. Some claims have missing values for some measures such as age, married, wage, hourly worker, tenure, education, firm size, industry, overnight hospitalization, or major surgery. To include these claims in the regressions, dummy variables are created to indicate the missing information, but we do not show the coefficients of these dummy variables in the table. The omitted reference categories are employer choice; Pennsylvania; high school graduate, firm size≤50; professional/clerical services; and inflammation, laceration, or contusion. An intercept is also included but not reported. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. Table H.5 Effects of Provider Choice on Recovery and Satisfaction Recovery (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Provider choice Employee chose Employee chose prior provider Employee chose new provider State California Texas Massachusetts Worker controls Age Male Married Wage Hourly worker Tenure Less than high school Some high school Some college College graduate 0 — — –12** –22** 15** –1** 11** 2 0 0 1** –19** –14** 4 3 1 — — –10** –22** 14** –1** 11** 2 0 0 1** –18** –14** 4 3 — –3 2 –12** –22** 15** –1** 11** 2 0 0 1** –19** –14** 4 3 — –1 3 –10** –22** 14** –1** 11** 3 0 1 1** –18** –14** 4 3 Satisfaction (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 57** — — –28** –29** –13 0 60** 19* –1 –21† 1** 28 11 7 26† 59** — — –28** –29** –10 0 60** 19* –1† –21† 1** 27 11 8 28† —— 86** 89** 38** 39** –29** –29** –13 –28** –29** –11 0 63** 20* –1† –21† 1** 29 11 7 27† 0 64** 20* –1† –21† 1** 27 10 8 29† continued 141 technical appendix h Table H.5 Effects of Provider Choice on Recovery and Satisfaction (continued) Recovery (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Worker controls (continued) Postgraduate Survey in Spanish Workplace controls Firm size 51–250 Firm size 251–1,000 Firm size≥1,000 High-risk services Low-risk services Manufacturing Construction Trade Other industries Injury controls Back pain Nonback sprain or strain Fracture Other injuries Severity 7 –38** 1 0 –14** 2 13* 4 2 1 6 –26** –16** 7 –8 3** 7 –36** 1 –1 –15** 1 13* 5 2 1 5 –24** –14** 8 –5 3** 7 –38** 2 –1 –14** 2 13* 4 1 1 6 –26** –16** 7 –8 3** 7 –37** 1 –1 –15** 1 13* 5 2 1 5 –25** –14** 7 –5 3** Satisfaction (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 18 –41* 8 –1 0 –7 7 –31* 18 2 –38* –45** –37** –22 –37** –2** 19 –38* 8 0 1 –6 7 –30* 20 1 –37* –44** –35** –21 –35** –2** 16 –39* 8 0 –2 –9 6 –31* 19 1 –38* –45** –37** –22 –38** –2** 18 –37* 8 1 –1 –8 7 –29* 21 0 –37* –44** –35** –21 –36** –2** 142 t e c h n i c a l a p p e n d i x h Table H.5 Effects of Provider Choice on Recovery and Satisfaction (continued) Recovery (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Satisfaction (percent) Two-Way Classification Three-Way Classification Model 1 Model 2 Model 1 Model 2 Treatment controls Overnight hospitalization Major surgery N — — 1,956 –18** –4 1,956 — — 1,947 –18** –4 1,947 — — 1,941 1 –6 1,941 — — 1,932 1 –7 1,932 technical appendix h Notes: The average recovery score is 19.23 in the two-way classification and 19.26 in the three-way classification. We divide the coefficients by the average score to get the percentage effect for recovery. We take 100 × (odds ratio – 1) to get the percentage effect for satisfaction. Some claims have missing values for some measures such as age, married, wage, hourly worker, tenure, education, firm size, industry, overnight hospitalization, or major surgery. To include these claims in the regressions, dummy variables are created to indicate the missing information, but we do not show the coefficients of these dummy variables in the table. The omitted reference categories are employer choice; Pennsylvania; high school graduate, firm size≤50; professional/clerical services; and inflammation, laceration, or contusion. An intercept is also included but not reported. Model 1 excludes treatment variables (surgery and overnight hospitalization), and model 2 includes them. ** Statistically significant at the 5 percent level. * Statistically significant at the 10 percent level. † Statistically significant at the 20 percent level. 143 References Airey, C., S. Bruster, B. Erens, S. Lilley, K. Pickering, and L. Pitson. 1999. National surveys of NHS patients: General practice 1998. London: NHS Executive. Barth, P., and M. Niss. 1999. Permanent partial disability benefits: Interstate differences. Cambridge, MA: Workers Compensation Research Institute. Barth, P., and R. Victor. 2003. Outcomes for injured workers in Texas. Cambridge, MA: Workers Compensation Research Institute. Boden, L. 1992. Workers’ compensation medical costs: A special case. In J. Greenwood and A. Tarico, eds., Workers’ compensation health care cost containment. Horsham, PA: LRP Publications. Boden, L., and C. Fleischman. 1989. Medical costs in workers’ compensation: Trends and interstate comparisons. Cambridge, MA: Workers Compensation Research Institute. Borba, P., and T. Parry. 2000 (October). An evaluation of the comprehensive and organized managed care program: Final report. Unpublished report prepared for the Robert Wood Johnson Foundation and the Joint Industry Board of the Electrical Industry. Centers for Disease Control and Prevention. 2002 (March 28). Behavioral risk factor surveillance system: Historical questions. Available at http:// apps.nccd.cdc.gov/brfssQuest/. Coulter, I., and P. Shekelle. 1997 (December). Supply, distribution, and utilization of chiropractors in the United States. In D. Cherkin and R. Mootz, Chiropractic in the United States: Training, practice, and research. AHCPR Publication No. 98-N002. Washington, DC: Agency for Health Care Policy and Research. Damiano, A., G. Pastores, and J. Ware, Jr., 1998. The health-related quality of life of adults with Gaucher’s disease receiving enzyme replacement. Quality of Life Research 7:373–386. DeNavas-Walt, C., B. Proctor, and R. Mills. 2004. Income, poverty, and health insurance coverage in the United States, 2003. Washington, DC: U.S. Government Printing Office. 145 146 r e f e r e n c e s Durbin, D., and D. Appel. 1991. The impact of fee schedules and employer choice of physician. NCCI Digest 6(3): 39–59. Durbin, D., D. Corro, and N. Helvacian. 1996. Workers’ compensation medical expenditures: Price vs. quantity. Journal of Risk and Insurance 63(1): 13–33. Eccleston, S., A. Laszlo, X. Zhao, and M. Watson. 2002. Benchmarks for designing workers’ compensation medical fee schedules, 2001–2002. Cambridge, MA: Workers Compensation Research Institute. Ellenberger, J. 1992. Labor’s perspective on health care reform. In J. Greenwood and A. Tarico, eds., Workers’ compensation health care cost containment. Horsham, PA: LRP Publications. Employee Benefit Research Institute, Consumer Health Education Council, and Matthew Greenwald and Associates. 2002. 2002 Health confidence survey: Summary of findings. Available at http://www.ebri.org/hcs/2002/hcs02sof.pdf. Galizzi, M., and L. Boden. 1996. What are the most important factors shaping return to work? Evidence from Wisconsin. Cambridge, MA: Workers Compensation Research Institute. Lewis, J. 1992. Legislative reform efforts and the medical benefit. In J. Greenwood and A. Tarico, eds., Workers’ compensation health care cost containment. Horsham, PA: LRP Publications. Morrison, J. 1990. Medical cost containment for workers’ compensation. Journal of Risk and Insurance 57(4): 646–653. National Academy of Social Insurance. 2004. Workers’ compensation: Benefits, coverage, and costs, 2002. Washington, DC. National Federation of Independent Business Research Foundation and National Foundation for Unemployment Compensation and Workers’ Compensation. n.d. Legislative guide to workers’ compensation insurance reform in the states. Washington, DC. Neumark, D. 2005 (January). The workers’ compensation crisis in California: A primer. California Economic Policy 1(1): 1–20. Available at http:// www.ppic.org/main/publication.asp?I=583. Pozzebon, S. 1994. Medical cost containment in workers’ compensation. Industrial and Labor Relations Review 48(1): 153–167. Research and Oversight Council on Workers’ Compensation (ROC). 1998 (August). An analysis of workers who were fired or laid off after a work-related injury. Austin, TX. Research and Oversight Council on Workers’ Compensation (ROC) and MedFX, LLC. 2001. Striking the balance: An analysis of the cost and quality of medical care in the Texas workers’ compensation system. A report to the 77th Texas Legislature. Austin, TX. references 147 Shields, J., D. Baroni, and X. Lu. 2003 (Summer, Special Edition). Post-injury health status of Texas workers with soft-tissue injuries. Texas Monitor 8(4): 1–7. Tanabe, R., and S. Murray. 2001. Managed care and medical cost containment in workers’ compensation: A national inventory, 2001–2002. Cambridge, MA: Workers Compensation Research Institute. Telles, C., D. Wang, and R. Tanabe. 2004. CompScopeTM benchmarks: Multistate comparisons, 4th edition. Cambridge, MA: Workers Compensation Research Institute. Telles, C., D. Wang, and R. Tanabe. 2005. CompScopeTM benchmarks, 5th edition. Cambridge, MA: Workers Compensation Research Institute. Texas Department of Insurance. 2004. Employer participation in the Texas workers’ compensation system: 2004 estimates. Austin, TX. U.S. Chamber of Commerce. 2004. 2004 analysis of workers’ compensation laws. Washington, D.C. U.S. Department of Labor, Bureau of Labor Statistics. 2001. State occupational injuries, illnesses, and fatalities. Available at http://www.bls.gov/iif/oshstate.htm. Victor, R., P. Barth, and T. Liu. 2003. Outcomes for injured workers in California, Massachusetts, Pennsylvania, and Texas. Cambridge, MA: Workers Compensation Research Institute. Victor, R., and C. Fleischman. 1990 (June). How choice of provider and recessions affect medical costs in workers’ compensation. Cambridge, MA: Workers Compensation Research Institute. Victor, R., D. Wang, and P. Borba. 2002. Provider choice laws, network involvement, and medical costs. Cambridge, MA: Workers Compensation Research Institute. Ware, J., M. Kosinski, and W. Rogers. 1996. The accuracy of retrospective evaluations of physical, mental, and general health status among patients with chronic conditions (abstract of unpublished study). The Health Institute, New England Medical Center. Boston, MA. Available at http://www.sf-36.org/cgibin/discuss/msg.cgi?msg=710. Ware, J., S. Keller, and M. Kosinski. 1998. SF-12®: How to score the SF-12® physical and mental health summary scale. Lincoln, RI: QualityMetric, Inc. Ware, J., D. Turner-Bowker, M. Kosinski, and B. Gandek. 2002. SF-12v2TM: How to score version 2 of the SF-12® health survey. Lincoln, RI: QualityMetric, Inc. Washington Department of Labor and Industries and University of Washington Department of Health Services. 1997. Workers’ compensation managed care pilot project: Final report to the legislature. Seattle, Washington. About the Authors Dr. Richard A. Victor, executive director of WCRI, helped establish the Institute in 1983. He received his J.D. and a Ph.D. in economics from the University of Michigan, where he was the George Humphrey Fellow in Law and Economic Policy. He then spent seven years conducting research at the Rand Corporation in both Washington, D.C., and Santa Monica, California. At Rand, Dr. Victor was a principal researcher at the Institute for Civil Justice. Dr. Victor is the author of numerous books and articles on workers’ compensation issues. Dr. Peter S. Barth is currently a Professor of Economics Emeritus at the University of Connecticut, where he formerly served for six years as the Department Head. His undergraduate degree is from Columbia University, and he earned his Ph.D. at the University of Michigan. He has prepared studies on permanent partial disability in workers’ compensation and on workers’ compensation systems in Connecticut, Texas, California, and Florida for the Workers Compensation Research Institute. Much of his published work has dealt with compensating workers disabled by occupational diseases. He also has done studies of workers’ compensation programs in British Columbia, Ontario and Victoria, Australia. In addition, he has served as a consultant to numerous organizations and advised government agencies, state and federal in the U.S., and abroad on workers’ compensation and related issues. Dr. David Neumark is a senior fellow in economics at the Public Policy Institute of California and a research associate of the National Bureau of Economic Research. He has published numerous studies on school-to-work, workplace segregation, sex discrimination, the economics of gender and the family, affirmative action, aging, minimum wages, and living wages. He is on the editorial boards of Industrial Relations, Contemporary Economic Policy, and Economics of Education Review. He has also held positions as professor of economics at Michigan State University, assistant professor of economics at the University of Pennsylvania, and economist at the Federal Reserve Board. He holds a Ph.D. in economics from Harvard University. 149 Related PPIC Publication “The Workers’ Compensation Crisis in California: A Primer” California Economic Policy. David Neumark. Volume 1, Number 1, January 2005. 150 Related WCRI Publications Provider Choice Laws, Network Involvement, and Medical Costs. Richard A. Victor, Dongchun Wang, and Philip Borba. December 2002. WC-02–05. The Impact of Initial Treatment by Network Providers on Workers’ Compensation Medical Costs and Disability Payments. Sharon E. Fox, Richard A. Victor, Xiaoping Zhao, and Igor Polevoy. August 2001. DM-01–01. The Impact of Workers’ Compensation Networks on Medical and Disability Payments. William G. Johnson, Marjorie L. Baldwin, and Steven C. Marcus. November 1999. WC-99–5. 151" ["post_date_gmt"]=> string(19) "2017-05-20 09:37:43" ["comment_status"]=> string(4) "open" ["ping_status"]=> string(6) "closed" ["post_password"]=> string(0) "" ["post_name"]=> string(9) "r_1105rvr" ["to_ping"]=> string(0) "" ["pinged"]=> string(0) "" ["post_modified"]=> string(19) "2017-05-20 02:37:43" ["post_modified_gmt"]=> string(19) "2017-05-20 09:37:43" ["post_content_filtered"]=> string(0) "" ["guid"]=> string(51) "http://148.62.4.17/wp-content/uploads/R_1105RVR.pdf" ["menu_order"]=> int(0) ["post_mime_type"]=> string(15) "application/pdf" ["comment_count"]=> string(1) "0" ["filter"]=> string(3) "raw" ["status"]=> string(7) "inherit" ["attachment_authors"]=> bool(false) }