Many studies have demonstrated that uninsured American

Transcription

1 Health Insurance Coverage And Mortality Among The Near-Elderly Uninsured near-elderly people are at much greater risk of premature death than their insured peers are. by J. Michael McWilliams, Alan M. Zaslavsky, Ellen Meara, and John Z. Ayanian ABSTRACT: Uninsured near-elderly people may be particularly at risk for adverse health outcomes. We compared mortality of a nationally representative cohort of insured and uninsured near-elderly people with stratification by race; income; and the presence of diabetes, hypertension, or heart disease, using propensity-score methods to adjust for numerous characteristics. Lacking health insurance was associated with substantially higher adjusted mortality among adults who were white; had low incomes; or had diabetes, hypertension, or heart disease. Expanding coverage to the near-elderly uninsured may greatly improve health outcomes for these groups. Many studies have demonstrated that uninsured American adults receive less appropriate care and fewer needed health services than their insured peers. 1 Near-elderly people who are uninsured represent a particularly vulnerable population. 2 The risks of experiencing major health problems and incurring substantial medical expenses increase dramatically for people ages 55 64, so the consequences of lacking insurance may be more severe. 3 Furthermore, near-elderly uninsured people often face higher premiums when acquiring health insurance and thus tend to be uninsured for longer periods than younger adults. 4 Projected increases in the number of near-elderly uninsured have motivated proposals to make coverage more affordable for this group. 5 A Medicare buy-in option allowing people to purchase Medicare coverage before reaching age sixty-five, with subsidies for those with low incomes, was proposed by the Clinton administration and more recently by Sen. John Edwards (D-NC) in the Democratic presidential primaries. Health savings accounts J. Michael McWilliams is an internal medicine resident and John Ayanian (ayanian@hcp.med.harvard.edu) is an associate professor of medicine in the Division of General Medicine and Primary Care, Department of Medicine, Brigham and Women s Hospital, in Boston, Massachusetts. Alan Zaslavsky is a professor of statistics in the Department of Health Care Policy, Harvard Medical School, Boston, where Ellen Meara is an assistant professor of health economics and Ayanian, an associate professor of health care policy. HEALTH AFFAIRS ~ Volume 23, Number DOI /hlthaff Project HOPE The People-to-People Health Foundation, Inc.

2 DataWatch (HSAs), enacted in the Medicare Prescription Drug, Improvement and Modernization Act (MMA) of 2003, allow individuals and employers to make tax-deductible contributions toward medical expenses. President George W. Bush has proposed tax credits for the low-income uninsured who buy nongroup coverage. As a presidential candidate, Sen. John Kerry (D-MA) has proposed tax credits for uninsured and near-elderly people buying coverage in a new group insurance option based on the Federal Employees Health Benefits Program (FEHBP). 6 Expansions in coverage are likely to increase the use of important clinical services for the near-elderly uninsured, but the effects of lacking coverage on health outcomes continue to be debated because of the analytic challenges of inferring causal effects from observational data. 7 In the few natural or randomized experiments on this topic, worse blood pressure control was evident among lowerincome people with hypertension who lost coverage or who were assigned less extensive coverage. 8 Observational studies have found increased mortality among uninsured adults relative to their insured peers with specific conditions such as cancer, myocardial infarction, or HIV infection, but only two national studies have compared all-cause mortality among insured and uninsured people. 9 Furthermore, both of these national studies ended seventeen years ago, and subsequent advances in medical care may have improved outcomes among people with better access to health services. 10 Therefore, in a nationally representative, longitudinal cohort of 8,736 nearelderly people, we compared mortality over eight years between those who were privately insured and those who were uninsured in To minimize potential confounding due to marked differences in numerous observed demographic, socioeconomic, health, and behavioral characteristics, we used rigorous propensity-score methods whose results are less likely to be biased than those of standard regression models. We also assessed whether the association between insurance coverage and mortality varied by race and ethnicity; income; or the presence of diabetes, hypertension, or heart disease. Study Data And Methods Study population. We analyzed publicly available data from the Health and Retirement Study (HRS), a nationally representative, longitudinal survey sponsored by the National Institute on Aging and conducted by the Institute for Social Research at the University of Michigan. 11 This study included noninstitutionalized adults in the forty-eight contiguous United States who were born during , with oversampling of blacks and Hispanics and Florida residents. In 1992 initial English or Spanish interviews were conducted in 7,702 households (82 percent response rate), yielding 9,825 participants. Among participants who completed the initial interview in 1992, we excluded those who reported public coverage at this interview, because disability influences eligibility for coverage by all large governmental programs available to this 224 July/August 2004

3 age group, including Medicare, Medicaid, the Civilian Health and Medical Program of the Uniformed Services (CHAMPUS, now known as TriCare), and the Department of Veterans Affairs (VA). Adults with public insurance were likely to have qualifying medical conditions (for example, end-stage renal disease) or disabilities not fully measured by the HRS that could have biased our results. Recipients of CHAMPUS or VA benefits who also reported having private health insurance were included, because their primary source of coverage was likely to be private insurance unrelated to disability. This study used publicly available, anonymous data, so the Human Studies Committee of Harvard Medical School deemed it exempt from review. Study variables. Participants were classified as insured if they reported private (employer-based or individually purchased) health insurance in 1992, and otherwise as uninsured. During the 1992 interview, participants also reported all study variables included in adjusted analyses. Participants reported to be deceased by household contacts through 2000 (n = 613) or whose vital status could not be determined from household contacts in 2000 (n = 562) were submitted by the HRS for matching to the National Death Index (NDI) for Participants with definite matches to the NDI (n = 605) were considered deceased. Participants with no match (n = 333) or possible matches (n = 237) were considered alive in Very few of these participants were reported deceased by household contacts (thirteen or 3.9 percent of nonmatches, and eight or 3.4 percent of possible matches). 12 Statistical analysis. To control for substantial differences in observed characteristics between insured and uninsured participants, we used propensity-score methods. 13 We used logistic regression to predict whether participants had private health insurance in 1992 as a function of twenty-seven variables. In addition to those listed in Exhibit 1, these variables included household size, census region, selfreported recent change in health, work limits imposed by health, job stress, physical effort required by job, daily alcohol consumption, exercise habits, expected probability of survival to age seventy-five, and the number of hospital stays in the prior twelve months. 14 The estimated probability of being insured in 1992, the propensity score, was used to derive individual weights equal to the probability of belonging to the opposite insurance group. We used Cox proportional hazards survival analyses to conduct our principal comparison of mortality by insurance status in Absolute differences in unadjusted and adjusted eight-year mortality rates were also determined and tested for significance using chi-square tests. Comparisons of mortality by insurance status were stratified by race and ethnicity, household income, and the presence or absence of diabetes, hypertension, or heart disease. Additional variables that may have explained mortality differences between insured and uninsured Hispanic adults (U.S. nativity, nonresponse in 2000, language of interview, years lived in the United States, ethnic identifica- HEALTH AFFAIRS ~ Volume 23, Number 4 225

4 DataWatch EXHIBIT 1 Characteristics Of Near-Elderly People With And Without Insurance In 1992 Unadjusted analysis Adjusted analysis Characteristic Insured (n = 7,199) Uninsured (n = 1,537) Insured (n = 7,199) Uninsured (n = 1,537) Mean age (years) Female (%) 55.6 (0.04) (0.09) (0.06) (0.11) 55.7 Race or ethnic group (%) Non-Hispanic white Non-Hispanic black Hispanic Other Married (%) Veteran (%) Education (%) Not a high school graduate High school graduate Some college or college graduate Employed full time (%) Annual household income (%) $21,000 $21,001 $38,450 $38,451 $61,224 $61, Total household wealth (%) $36,000 $36,001 $101,545 $101,546 $234,321 $234, Self-reported health status (%) Excellent Very good Good Fair Poor Current smoker (%) Overweight (%) a Doctor visit in past 12 months (%) Mean number of difficulties with ADLs b Mean number of self-reported mobility limitations b Mean number of self-reported chronic diseases b CESD score for depression c 0.09 (0.01) 2.28 (0.05) 1.04 (0.01) 1.41 (0.02) 0.24 (0.02) 3.17 (0.09) 1.22 (0.04) 0.82 (0.05) 0.16 (0.02) 2.78 (0.07) 1.12 (0.02) 1.11 (0.03) 0.16 (0.01) 2.79 (0.09) 1.12 (0.04) 1.11 (0.05) SOURCE: Authors estimates using data from the Health and Retirement Study. NOTES: Standard errors are in parentheses for selected characteristics. Insured and uninsured adults differed significantly (p.05) across all characteristics except age in unadjusted comparisons. After propensity-score adjustment, all observed characteristics were very closely balanced between insured and uninsured adults (p >.95). a Women with body-mass index 27.3 kg/m 2 and men with body-mass index 27.8 kg/m 2. b Based on six activities of daily living (ADLs), such as bathing and eating, and eleven mobility functions such as walking a block or climbing a flight of stairs. Chronic diseases include hypertension, diabetes, heart disease, chronic lung disease, cancer, arthritis, stroke, ulcers, and psychiatric disorders. c Adapted from the Center for Epidemiologic Studies Depression Scale (CESD) score; more negative scores indicate better mental health. 226 July/August 2004

5 tion [Mexican, Puerto Rican, Cuban, or other], and an interaction between language and self-reported health) were included in the propensity-score analysis for Hispanics. In these stratified analyses, separate propensity-score models were fit for each stratum to determine appropriate weights for adjustment. In addition, we performed a sensitivity analysis to assess whether differences in unobserved characteristics between insured and uninsured adults might explain observed differences in mortality. 15 All analyses accounted for the complex survey design. Study Results Characteristics of study cohort. Of the 9,825 participants interviewed in 1992, we excluded 953 (9.7 percent) adults with public health insurance and 136 (1.4 percent) with missing data on insurance coverage. Of the remaining 8,736 adults, 7,199 (82.4 percent) were privately insured and 1,537 (17.6 percent) were uninsured in Among adults privately insured in 1992, 10.9 percent reported they were uninsured in at least one biennial survey through Among adults who were uninsured in 1992, the proportion of respondents who reported being publicly or privately insured rose progressively in the ensuing four surveys (46.6 percent, 58.4 percent, 66.1 percent, and 74.5 percent), as nearly half reached age sixty-five and became eligible for Medicare by Insured and uninsured adults in 1992 differed significantly across almost all observed characteristics in unadjusted comparisons (Exhibit 1). After propensityscore adjustment, all observed characteristics were very closely balanced between insured and uninsured adults. Mortality. Exhibit 2 shows unadjusted eight-year mortality rates for uninsured and insured adults, and rates adjusted for the propensity to be insured. Significant differences in adjusted eight-year mortality rates were evident among white adults; adults with low incomes; and those with diabetes, hypertension, or heart disease, but not among black adults, Hispanic adults, adults with higher incomes, or those without diabetes or cardiovascular disease (Exhibits 3 and 4). Mortality was significantly greater for uninsured adults than insured adults in an unadjusted proportional hazards analysis (hazard ratio [HR]: 1.83; 95 percent confidence interval [CI]: 1.46, 2.29; p <.001) and remained significantly greater after adjustment for propensity scores (HR: 1.43; 95 percent CI: 1.10, 1.85; p =.009). In stratified survival analyses, insurance coverage was associated with significantly lower adjusted mortality in white adults (HR: 1.57; 95 percent CI: 1.16, 2.12), adults in the lowest income quartile (HR: 1.53; 95 percent CI: 1.11, 2.12), and adults with diabetes, hypertension, or heart disease (HR: 1.56; 95 percent CI: 1.15, 2.10) (all p <.01), but not in adults with higher incomes (HR: 1.27; 95 percent CI: 0.78, 2.06) or without these conditions (HR: 1.22; 95 percent CI: 0.82, 1.80). Mortality did not differ statistically between uninsured and insured non-hispanic black adults (HR: 1.08; 95 percent CI: 0.67, 1.75; p =.73) or uninsured and insured Hispanic adults (HR: 0.48; 95 percent CI: 0.18, 1.28; p =.14). HEALTH AFFAIRS ~ Volume 23, Number 4 227

6 DataWatch EXHIBIT 2 Unadjusted And Adjusted Eight-Year Mortality Rates Among Insured And Uninsured Near-Elderly People, By Insurance Status In 1992 In a sensitivity analysis, we found that the presence of an unobserved factor similar to smoking in prevalence (approximately 25 percent of the study cohort) and its association with insurance status (relative risk of being uninsured equal to 1.66) would have to be associated with a relative eight-year mortality risk of 2.65 for the association between insurance status and mortality to become non- EXHIBIT 3 Adjusted Eight-Year Mortality Rates Among Insured And Uninsured Near-Elderly People, By Race And Ethnicity In July/August 2004

7 EXHIBIT 4 Adjusted Eight-Year Mortality Rates Among Insured And Uninsured Near-Elderly People, By Income And The Presence Of Diabetes Or Cardiovascular Disease In 1992 significant when further adjusted for this unmeasured factor (HR: 1.28; 95 percent CI: 0.99, 1.67; p =.06). In comparison, smoking was associated with a relative eight-year mortality risk of Discussion And Policy Implications This nationally representative study demonstrated that uninsured near-elderly people are at much greater risk of premature death than their insured peers. Although consistent with the conclusions of two previous national studies, our findings provide more recent estimates of mortality differences between privately insured and uninsured adults. 16 Based on these adjusted eight-year mortality rates and an estimated 3.5 million uninsured people ages in 2002, more than 105,000 excess deaths in the next eight years (more than 13,000 annually) may be attributable to the present lack of insurance coverage among the near-elderly. 17 This estimate would place uninsurance third on a list of leading causes of death for this age group, below only heart disease and cancer. 18 This rapidly growing age group is expected to more than double to 61.9 million (about 20 percent of the U.S. population) by Taking this growth into consideration and assuming a stable uninsurance rate (13 percent), the annual number of excess deaths attributable to the lack of health insurance may exceed 30,000 by 2015, more than the combined number of deaths attributable to stroke, diabetes, and lung disease in this age group. HEALTH AFFAIRS ~ Volume 23, Number 4 229

8 DataWatch By focusing on the near-elderly, we assessed an age group that faces greater risks of acute and chronic illnesses than younger people and is thus more likely to benefit from effective medical care. Findings from our stratified analyses indicate that the increased mortality of uninsured adults was concentrated among those with low incomes or with diabetes or cardiovascular disease, which underscores the importance of effective medical care for these groups. However, near-elderly people with low incomes or chronic illness typically face the greatest obstacles to obtaining private health insurance if they do not qualify for public coverage. 20 Our findings suggest that expanding coverage to the near-elderly uninsured may greatly reduce mortality for these vulnerable groups. However, each of the major policy options for expanding coverage in this age group must address major challenges. A Medicare buy-in option could reduce the number of near-elderly uninsured people but would require sizable premium subsidies to do so, since uninsured adults tend to have lower incomes. 21 HSAs and tax deductions to purchase nongroup insurance primarily benefit higher-income people with high marginal tax rates. Furthermore, chronically ill people face higher deductibles and premiums in the nongroup insurance market compared with those of healthy people, which pose substantial barriers for those with low incomes even when tax credits are available. Tax credits alone, unless much greater than those proposed in the Bush administration s federal budget for 2005, are therefore unlikely to reduce substantially the number of uninsured near-elderly people. 22 Tax credits proposed by Senator Kerry in his presidential campaign are more generous and are coupled with a group insurance option that would likely be more affordable than nongroup insurance, and thus potentially more effective in expanding coverage. 23 The incremental cost of providing equivalent care to uninsured and insured Americans would likely be exceeded by the economic value of sizable gains in health capital for people without insurance. 24 Mortality was surprisingly similar for insured and uninsured blacks in our study, which suggests that insurance coverage alone may not reduce mortality for near-elderly blacks. Even with health insurance, black adults face greater barriers to care and receive lower-quality health care than whites, which may result from inadequate communication or stereotyping, subconscious biases, or clinical uncertainty among health care providers. 25 Insurance may also be insufficient to overcome lifelong risk factors for ill health and mortality, including income inequality and broader discrimination experienced by black Americans. 26 Adjusted mortality was nearly identical for white and Hispanic adults who were insured, but mortality tended to be lower among uninsured Hispanics relative to insured Hispanics. In general, Hispanic adults have a worse socioeconomic profile but experience lower mortality than non-hispanic adults in the United States. 27 Several theories regarding this epidemiological paradox, such as better unobserved health among new immigrants, may also explain the trend toward lower observed mortality among uninsured Hispanic adults. 28 For example, be- 230 July/August 2004

9 cause acculturation at the group level occurs over several generations, variables that we controlled for, such as language and time spent in the United States, may not have fully captured differences in health behavior between insured and uninsured Hispanic adults. 29 The HRS may also have selectively enrolled uninsured Hispanics with better health or access to care than those who were not enrolled, thereby underestimating the actual mortality of this group. Our analysis adjusted for numerous variables not present in the most extensive prior study of this topic, including activities of daily living (ADLs) and physical functioning, presence of chronic diseases or depressive symptoms, marital status, veteran status, geographic region, wealth, and job stress. 30 All of these variables differed significantly by insurance status and have been independently associated with mortality. 31 Despite the breadth of variables included in this study and the use of rigorous propensity-score methods, unobserved factors could have explained our findings. However, we adjusted for numerous variables directly related to disability (for example, job status, work limits imposed by health, physical functioning, chronic conditions, and expected mortality), greatly reducing the effect of actual or impending disability as a potential confounder of the increased mortality associated with lacking insurance. Our sensitivity analysis further demonstrated that the confounding effect of unmeasured variables would have to be even greater than the impact of smoking on mortality in our study for the increased mortality of uninsured adults to become statistically nonsignificant. Our study had other potential limitations. Self-reported data were used for statistical adjustment; the accuracy of these data for insured and uninsured adults should be evaluated in future studies. In addition, our analysis focused on insurance status in 1992 and did not address the subsequent gains and losses in coverage experienced by many participants. Because continuity of health insurance may have a dose-related effect on health, changes in coverage after 1992, particularly those related to Medicare eligibility, may have attenuated the effect of observed insurance status on mortality. 32 Our study, however, had limited statistical power to compare the impact of gaining Medicare coverage on the mortality of those who were previously insured or uninsured. Future research could assess this effect in a larger cohort of people followed well beyond age sixty-five. Our study demonstrated greatly increased mortality among the nearelderly uninsured relative to their privately insured peers. This finding was evident among white adults; those with low incomes; and those with diabetes, hypertension, or heart disease, which suggests that these groups are most likely to experience health benefits of expanding insurance coverage for uninsured people over age fifty. Our study also indicates that expanding health insurance coverage alone may not be sufficient to reduce the increased mortality experienced by near-elderly blacks. Reforms to expand coverage, such as a Medicare buy-in HEALTH AFFAIRS ~ Volume 23, Number 4 231

10 DataWatch program or tax credits to purchase insurance, may produce sizable health benefits if they provide affordable coverage for the near-elderly uninsured, particularly those with low incomes or chronic illness. The authors are grateful to Robert Wolf for assistance with statistical programming. This study was supported by the Primary Care Research Fund of Brigham and Women s Hospital. NOTES 1. Institute of Medicine, Coverage Matters: Insurance and Health Care (Washington: National Academies Press, 2001); IOM, Care without Coverage: Too Little, Too Late (Washington: National Academies Press, 2002); J.Z. Ayanian et al., Unmet Health Needs of Uninsured Adults in the United States, Journal of the American Medical Association 284, no. 16 (2000): ; and J. Hadley, Sicker and Poorer The Consequences of Being Uninsured: A Review of the Research on the Relationship between Health Insurance, Medical Care Use, Health, Work, and Income, Medical Care Research and Review 60, Suppl. 2 (2003): 3S 75S. 2. K. Davis, Uninsured in an Era of Managed Care, Health Services Research 31, no. 6 (1997): R.W. Johnson and S. Crystal, Health Insurance Coverage at Midlife: Characteristics, Costs, and Dynamics, Health Care Financing Review 18, no. 3 (1997): ; and N. Brennan, Health Insurance Coverage of the Near Elderly, National Survey of America s Families, Series B, no. B-12 (Washington: Urban Institute, 2000). 4. IOM, Coverage Matters; Brennan, Health Insurance Coverage of the Near Elderly; and C. Schoen et al., Counting on Medicare: Perspectives and Concerns of Americans Ages Fifty to Seventy, Pub. no. 406 (New York: Commonwealth Fund, 2000). 5. K. Kinsella and V.A. Velkoff, An Aging World: 2001, U.S. Census Bureau, Series P95/01-1 (Washington: U.S. Government Printing Office, 2001); R.W. Johnson, M. Moon, and A.J. Davidoff, A Medicare Buy-In for the Near Elderly: Design Issues and Potential Effects on Coverage, Pub. no (Menlo Park, Calif.: Henry J. Kaiser Family Foundation, 2002); D.G. Shea, P.F. Short, and M.P. Powell, Betwixt and Between: Targeting Coverage Reforms to Those Approaching Medicare, Health Affairs 20, no. 1 (2001): ; and J. Sheils and Y.J. Chen, Medicare Buy-In Options: Estimating Coverage and Costs, Pub. no. 441 (New York: Commonwealth Fund, 2001). 6. Under Bush s proposal, uninsured single adults would be eligible for credits worth up to $1,000, and families with two or more children, up to $3,000 per year, with eligibility phasing out at annual incomes between $15,000 and $30,000 for individuals and $25,000 and $60,000 for families. Kerry s proposal would provide tax credits to uninsured individuals for premiums exceeding 6 percent of income, with additional credits for small employers and their workers, people ages 55 64, and unemployed workers. See S.R. Collins, K. Davis, and J.M. Lambrew, Health Care Reform Returns to the National Agenda: The 2004 Presidential Candidates Proposals, Pub. no. 671 (New York: Commonwealth Fund, 2004). 7. J.M. McWilliams et al., Impact of Medicare Coverage on Basic Clinical Services for Previously Uninsured Adults, Journal of the American Medical Association 290, no. 6 (2003): ; and H. Levy and D. Meltzer, What Do We Really Know about Whether Health Insurance Affects Health? in Health Policy and the Uninsured, ed. C.G. McLaughlin (Washington: Urban Institute Press, 2004). 8. N. Lurie et al., Termination of Medi-Cal Benefits: A Follow-Up Study One Year Later, New England Journal of Medicine 314, no. 19 (1986): ; and J.P. Newhouse and the Insurance Experiment Group, Free for All? Lessons from the RAND Health Insurance Experiment (Cambridge, Mass.: Harvard University Press, 1993). 9. J.Z. Ayanian et al., The Relation between Health Insurance Coverage and Clinical Outcomes among Women with Breast Cancer, New England Journal of Medicine 329, no. 5 (1993): ; R.G. Roetzheim et al., Effects of Health Insurance and Race on Breast Carcinoma Treatments and Outcomes, Cancer 89, no. 11 (2000): ; J.G. Canto et al., Payer Status and the Utilization of Hospital Resources in Acute Myocardial Infarction, Archives of Internal Medicine 160, no. 6 (2000): ; D.P. Goldman et al., Effects of Insurance on Mortality in an HIV-Positive Population in Care, Journal of the American Statistical Association 96, no. 455 (2001): ; P. Franks, C.M. Clancy, and M.R. Gold, Health Insurance and Mortality: Evidence from a National Cohort, Journal of the American Medical Association 270, no. 6 (1993): ; and P.D. Sorlie et al., Mortality in the Uninsured Compared with That in Persons with Public and Private Health Insurance, Archives of Internal Medicine 154, no. 21 (1994): D.M. Cutler and M. McClellan, Is Technological Change in Medicine Worth It? Health Affairs 20, no. 5 (2001): 11 29; D.M. Cutler, Declining Disability among the Elderly, Health Affairs 20, no. 6 (2001): 11 27; and S. Glied and S.E. Little, The Uninsured and the Benefits of Medical Progress, Health Affairs 22, no July/August 2004

11 (2003): S.G. Heeringa and J.H. Connor, Technical Description of the Health and Retirement Survey Sample Design, May 1996, hrsonline.isr.umich.edu/intro/sho_uinfo.php?hfyle=ref023&xtyp=5 (17 May 2004). 12. Because we used data released early by the HRS that included possible matches of subjects to the NDI, we performed a sensitivity analysis that included as deaths the small number of participants reported deceased by household contacts but not definitely matched to the NDI. The addition of these deaths had only a slight effect on overall adjusted mortality associated with being uninsured. 13. D.B. Rubin, Estimating Causal Effects from Large Data Sets using Propensity Scores, Annals of Internal Medicine 127, no. 8S (1997): ; L.E. Braitman and P.R. Rosenbaum, Rare Outcomes, Common Treatments: Analytic Strategies using Propensity Scores, Annals of Internal Medicine 137, no. 8 (2002): ; and M.B. Landrum and J.Z. Ayanian, Causal Effect of Ambulatory Specialty Care on Mortality following Myocardial Infarction: A Comparison of Propensity Score and Instrumental Variable Analyses, Health Services and Outcomes Research Methodology 2, no. 3 (2001): An expanded version of Exhibit 1 is available online, at content.healthaffairs.org/cgi/content/full/23/4/ 223/DC P.R. Rosenbaum, Observational Studies (New York: Springer-Verlag, 1995). 16. Franks et al., Health Insurance and Mortality ; and Sorlie et al., Mortality in the Uninsured. 17. U.S. Census Bureau, Health Insurance Coverage: 2002 (Washington: U.S. GPO, September 2003). 18. R.N. Anderson and B.L. Smith, Deaths: Leading Causes for 2001, National Vital Statistics Reports 52, no. 9 (Hyattsville, Md.: National Center for Health Statistics, 2003). 19. Kinsella and Velkoff, An Aging World. 20. IOM, Coverage Matters; Johnson, Health Insurance Coverage at Midlife ; Johnson et al., A Medicare Buy-In for the Near Elderly; and Shea et al., Betwixt and Between. 21. Johnson et al., A Medicare Buy-In for the Near Elderly; Shea et al., Betwixt and Between ; and Sheils and Chen, Medicare Buy-in Options. 22. Johnson et al., A Medicare Buy-In for the Near Elderly; and J.R. Gabel, K. Dhont, and J. Pickreign, Are Tax Credits Alone the Solution to Affordable Health Insurance? Pub. no. 527 (New York: Commonwealth Fund, 2000). 23. Gabel et al., Are Tax Credits Alone the Solution? 24. IOM, Hidden Costs, Value Lost (Washington: National Academies Press, 2003); and W. Miller, E.R. Vigdor, and W.G. Manning, Covering the Uninsured: What Is It Worth? Health Affairs, 31 March 2004, content.healthaffairs.org/cgi/content/abstract/hlthaff.w4.157 (4 April 2004). 25. IOM, Coverage Matters; IOM, Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care (Washington: National Academies Press, 2002); and A.I. Balsa and T.G. McGuire, Prejudice, Uncertainty, and Stereotypes as Sources of Health Care Disparities, Journal of Health Economics 22, no. 1 (2003): J.Z. Ayanian, Heart Disease in Black and White, NewEnglandJournalofMedicine 329, no. 9 (1993): P.D. Sorlie et al., Mortality by Hispanic Status in the United States, Journal of the American Medical Association 270, no. 20 (1993): ; and A.F. Abraido-Lanza et al., The Latino Mortality Paradox: A Test of the Salmon Bias and Healthy Migrant Hypotheses, American Journal of Public Health 89, no. 10 (1999): Ibid; and R. Scribner, Paradox as Paradigm The Health Outcomes of Mexican Americans, American Journal of Public Health 86, no. 3 (1996): Scribner, Paradox as Paradigm. 30. Franks et al., Health Insurance and Mortality. 31. S.K. Inouye et al., Importance of Functional Measures in Predicting Mortality among Older Hospitalized Patients, Journal of the American Medical Association 279, no. 15 (1998): ; and A. Falk et al., Job Strain and Mortality in Elderly Men: Social Network, Support, and Influence as Buffers, American Journal of Public Health 82, no. 8 (1992): D.W. Baker et al., Lack of Health Insurance and Decline in Overall Health in Late Middle Age, New England Journal of Medicine 345, no. 15 (2001): ; and D.W. Baker et al., Loss of Health Insurance and the Risk for a Decline in Self-Reported Health and Physical Functioning, Medical Care 40, no. 11 (2002): HEALTH AFFAIRS ~ Volume 23, Number 4 233