Effect of Insurance on Mortality in an HIV-Positive Population in Care Dana P. Goldman, Jayanta Bhattacharya, Daniel F. McCaffrey, Naihua Duan, Arleen A. Leibowitz, Geoffrey F. Joyce, and Sally C. Morton As policymakers consider expanding insurance coverage for individuals infected with human immunodeficiency virus (HIV), it is useful to ask if insurance has any affect on health outcomes and, if so, whether its magnitude has changed with recent efficacious but expensive treatments. By using data from a nationally representative cohort of HIV-infected (HIV+) persons receiving regular medical care, we estimate the impact of insurance on mortality in this population. A naïve single-equation model confirms the perverse result found by others in the literature—that insurance increases the probability of death for HIV+ patients. We attribute this finding to a correlation between unobserved health status and insurance status in the mortality equation for two reasons. First, the eligibility rules for Medicaid and Medicare require HIV+ patients to demonstrate a disability, almost always defined as advanced disease, to qualify. Second, if unobserved health status is the cause of the positive correlation, then including measures of HIV+ disease as controls should mitigate the effect. Including measures of immune function (CD4 lymphocyte counts) reduces the effect size by approximately 50%, although it does not change sign. To deal with this correlation, we develop a two-equation parametric model of both insurance and mortality. The effect of insurance on mortality is identified through the judicious use of state policy variables as instruments (variables related to insurance status but not mortality, except through insurance). The results from this model indicate that insurance does have a beneficial effect on outcomes, lowering the probability of 6-month mortality by 71% at baseline and 85% at follow-up. The larger effect at follow- up can be attributed to the recent introduction of effective therapies for HIV infection, which have magnified the returns to insurance for HIV+ patients (as measured by mortality rates). KEY WORDS: AIDS; Bivariate normal; Instrumental variable; Potential outcome; Probit; Switching regression. 1. INTRODUCTION Treatment of human immunodeficiency virus (HIV) infection poses tremendous challenges to our public and private health care financing system. With the advent of efficacious—but expensive—new therapies, it is possible that health outcomes for HIV-infected (HIV+) persons will be very responsive to the availability of insurance. The HIV Cost and Services Utilization Study (HCSUS) assembled the first nationally representative cohort of HIV-infected persons receiving regular medical care (Bozzette et al. 1998; Shapiro et al. 1999a). This dataset provides a unique opportunity to produce national estimates of the impact of insurance on health outcomes, particularly mortality. Measuring the efficacy of insurance in deferring mortal- ity among HIV+ patients is complicated by the association between disease severity and insurance availability. For exam- ple, in the early stages of the epidemic, HIV affected a relatively affluent homosexual male population. Many were Dana P. Goldman is Senior Economist, RAND, Santa Monica, CA 90407 (E-mail: Dgoldman@rand.org). Jayanta Bhattacharya is Associate Economist, RAND, Santa Monica, CA 90407 (E-mail: Jay@rand.org). Daniel F. McCaffrey is Statistician, RAND, Santa Monica, CA 90407 (E-mail: Danielm@rand.org). Naihua Duan is Professor in Residence, School of Medicine, UCLA, Los Angeles, CA 90095. Arleen A. Leibowitz is Pro- fessor, School of Public Policy and Social Research, UCLA, Los Angeles, CA 90095 (E-mail: Arleen@ucla.edu). Geoffrey F. Joyce is Associate Economist, RAND, Santa Monica, CA 90407 (E-mail: Gjoyce@rand.org). Sally C. Morton is Senior Statistician, RAND, Santa Monica, CA 90407 (E-mail: Smorton@rand.org). The HIV Cost and Services Utilization Study is being conducted under cooperative agreement U01HS08578 between RAND and the Agency for Health Care Policy and Research. Substantial additional funding for this cooperative agreement was provided by the Health Ser- vices Resources Administration, the National Institute for Mental Health, the National Institute on Drug Abuse, and the National Institutes of Health Office of Research on Minority Health through the National Institute for Dental Research. Additional support was provided by the Robert Wood Johnson Foundation, Merck and Company, Glaxo-Wellcome, Inc., the National Insti- tute on Aging, and the Office of the Assistant Secretary for Planning and Eval- uation in the U.S. Department of Health and Human Services. The authors thank Afshin Rastegar for research assistance and three anonymous reviewers and an associate editor for their useful comments. employed and had private insurance at the time of infection. However, as their health deteriorated, they were forced to leave their jobs and they lost their insurance. Eventually they qualified for public insurance through Medicaid or Medicare, but eligibility rules for these programs require HIV+ patients to demonstrate a disability—almost always associated with advanced disease. These patterns of insurance coverage pose a challenge for analysts. Indeed, in Lancaster and Intrator’s (1998) study of the hospitalization rates for HIV+ patients using longitudinal data from a large multisite study, they found the perverse result that health insurance increases the risk of death for HIV+ individuals. Lancaster and Intrator were appropriately suspi- cious of this result, calling it “striking,” and identified it as an area for further investigation. In this article, we attempt to estimate the true effect of insurance on mortality by modeling both simultaneously. Our article has applicability to a broader set of issues in medicine. Much clinical research is devoted to analyzing the effects of various treatments on health outcomes. Typi- cally, this is done with controlled clinical trials that random- ize patients to treatment group. However, for various reasons, including cost and ethical concerns, randomized trials often are infeasible or too narrowly defined to be useful to policy- makers. Researchers can circumvent these problems by using observational data in studies that evaluate outcomes. How- ever, as McClellan, McNeil, and Newhouse (1994) and others have argued, such studies may be subject to bias—differences in outcomes between treatment groups may reflect underlying differences in characteristics that are not measured and cannot be controlled for via modeling. In this quasi-experimental set- ting, analysts typically view these unobservable characteristics © 2001 American Statistical Association Journal of the American Statistical Association September 2001, Vol. 96, No. 455, Applications and Case Studies 883