Pathways To Access: Health Insurance, The Health Care Delivery System, And Racial/Ethnic Disparities,

Transcription

1 Pathways To Access: Health Insurance, The Health Care Delivery System, And Racial/Ethnic Disparities, Insurance matters, but so do other factors, when it comes to explaining differences in levels of access among racial and ethnic groups. by Samuel H. Zuvekas and Gregg S. Taliaferro ABSTRACT: We examine the roles that insurance coverage, the delivery system, and external factors play in explaining persistent disparities in access among racial and ethnic groups of all ages. Using data from the Medical Expenditure Panel Surveys and regression-based decomposition methods, we find that our measures of health care system capacity explain little and that while insurance clearly matters, external factors are equally important. Employment, job characteristics, and marital status are key determinants of disparities in access to insurance but are difficult for health policy to affect directly. Much of existing disparities remains unexplained, presenting a challenge to developing policies to eliminate them. Health care treatment and outcomes are influenced by a person s ability to access the health care system. Despite marked improvements in the nation s overall health, disparities in access persist across racial and ethnic groups. 1 Members of these groups continue to have poorer access to highquality health care services and different patterns of use than those of whites. These disparities include using fewer preventive services, being less likely to have a usual source of care, and being more likely to lack health insurance. Because the nation s health is intertwined with the well-being of all Americans, policymakers and health care providers have set a goal to eliminate racial and ethnic disparities in health. Indeed, this is one of the central goals of the Healthy People 2010 campaign. 2 The 2001 Institute of Medicine (IOM) report on health care quality further highlights access as a key component of a high-quality health care system. 3 Sam Zuvekas is a senior economist at the Center for Cost and Financing Studies, Agency for Healthcare Research and Quality, in Rockville, Maryland. Gregg Taliaferro is a sociologist there. HEALTH AFFAIRS ~ Volume 22, Number Project HOPE The People-to-People Health Foundation, Inc.

2 Research Challenge The reasons for racial and ethnic disparities in access and health care use are not fully understood. Variations in health insurance coverage are the best-studied explanation and a key area of emphasis for recent health policy reforms. Whites are more likely than any other group to have insurance coverage. 4 Further disparities are found when examining the type of health insurance held. Blacks and Hispanics are much less likely to be privately insured and more likely to be publicly insured than whites are. Even when the analysis is restricted to workers, disparities persist. Declining rates of eligibility appear to be a major source of the problem for Hispanic workers, while take-up rates seem to be the driving force behind decreases in employment-based insurance coverage in general. 5 Several recent studies suggest that variations in coverage may be only a small part of the explanation. 6 Disparities in access and use persist even among wellinsured minority racial and ethnic groups. Factors such as income, education, and health status that lead to variations in insurance coverage may also directly affect access. Other frequently cited sources of disparity are local variations in the supply of health care providers, cultural and linguistic differences, discrimination (whether intentional or not), perceptions of bias, and differences in beliefs and attitudes about health care. Some of these factors are more difficult to study; their existence underscores the complexity of the problem. That racial and ethnic disparities in access to high-quality health care not only exist but also appear to be growing over the past couple of decades is cause for serious concern. 7 The complexity of the problem suggests that there is no quick and easy solution. A better understanding of health care disparities is needed if programs are to be targeted effectively. In particular, research is needed to determine the extent to which disparities are explained by factors purely within the health care delivery system, how much is explained by factors largely outside the domain of health policy such as socioeconomic or demographic differences, and how much falls in between (such as the employment-related health insurance system). In this study we present recent data on trends in racial and ethnic differences in access to and use of health care services and examine in detail the sources of the disparities that continue to persist. In doing so, we bring together and build upon two strands of the literature on disparities to consider a wide range of factors and pathways to access to care. We build on previous decomposition studies to consider other factors besides insurance and income, most notably differences in local health care delivery system capacity and education. 8 We also build upon studies of disparities in employment-related coverage among workers, to more explicitly consider the relative contributions of differences in employment, offers of insurance, and take-up rates in explaining access to employer-provided coverage, which remains a key pathway to health care in the United States. 9 We focus mainly on access to ambulatory care but draw larger inferences about implications for health care and health outcomes and disparities in general. 140 March/April 2003

3 Study Methods To study recent trends in disparities and to understand the factors explaining these disparities, we use the rich data from the Medical Expenditure Panel Survey (MEPS). MEPS began in 1996 and annually collects detailed information on health care use and spending, access to care, health status, health insurance coverage, demographic and socioeconomic characteristics, and employment and job characteristics for nationally representative samples of the civilian, noninstitutionalized U.S. population. 10 We supplement the MEPS data with matching local area variables from the Area Resource File (ARF). The MEPS and ARF data allow us to examine a variety of potential factors, both internal and external to the health care system, which might explain disparities in access. We examine in some detail disparities in access to employer-sponsored coverage, because this is the primary means by which Americans gain access to health care, but we also consider other key factors such as delivery system capacity, income, and education. Analytical methods. We first present descriptive statistics of the differences in access and use between blacks, Hispanics, and whites, using z-score tests at the.05 percent level to test for significant differences. We then use the Oaxaca-Blinder regression-based method to decompose each total disparity into the percentage attributable to differences in observed characteristics between groups and the percentage attributable to unexplained factors. 11 This method further allows us to decompose the percentages of the total disparity that are associated with differences in (1) demographic characteristics; (2) socioeconomic status; (3) health status; (4) employment and job characteristics; (5) health insurance coverage; (6) health care system capacity; and (7) county-level demographic characteristics and economic climate. 12 Where appropriate (such as for children), we substituted socioeconomic and job characteristics of either the policyholder or the head of household. We use reduced-form regression models, so that the decompositions should be interpreted as associations and not necessarily causal relationships. We then present descriptive statistics of patterns of health insurance coverage and use an algebraic method to decompose differences in employment-related health insurance coverage into the percentages attributable to differences in employment, offer rates from jobs, and take-up rates. Finally, we use Oaxaca-Blinder decompositions to examine the factors explaining disparities that persist even among the privately insured. Scope of sample. We include data from MEPS files covering calendar years 1996 (N = 21,571), 1998 (N = 23,565), and 1999 (N = 22,953). MEPS access data for 1997 are incomplete and thus are omitted. We present statistics on disparities for all three years, but we focus on the factors explaining disparities in 1998, because it is the most recent year for which all of the necessary data are available. The sample scope varies by specific analyses. We include all age groups in looking at disparities in the access and use measures. However, we limit our sample to adults ages in examining sources of health insurance, because they represent the bulk of people HEALTH AFFAIRS ~ Volume 22, Number 2 141

4 Research Challenge who acquire employer coverage for themselves and their families. 13 Finally, we decompose disparities among people with private insurance coverage during all of 1998, but no Medicaid, Medicare, or other public coverage. Sample sizes for each analysis are given in the exhibits. Access and use. We examine disparities in access using three measures from the MEPS Access to Care supplement. Whether a person has a usual source of care is a standard benchmark of regular access to ambulatory care; this question is asked as follows: Is there a particular doctor s office, clinic, health center, or other place that the individual usually goes when sick or in need of health advice? We also examine differences in access using two subjective family-level measures of (1) whether the family reported that any family member had difficulty or delays or did not get needed health care, and (2) whether the family was very satisfied with their ability to get medical treatment when needed. Although not specifically limited, these subjective measures likely reflect primarily family experiences with ambulatory care, since that is the setting where people most often receive care. We include two measures of any non emergency room (ER) ambulatory treatment from office-based providers and outpatient departments of hospitals, and the number of visits to these providers. We exclude ER visits because they may themselves indicate access problems. However, we note that disparities in ambulatory care use diminish only slightly if ER visits are included. Personal characteristics. We include demographic measures of age, sex, marital status, and family size and socioeconomic measures of education and family income relative to the federal poverty level. Health status measures include indicators for poor or fair, good, and excellent health from a five-item self-reported health scale and an indicator for any activity of daily living (ADL) limitation. Employment and job characteristics. Employment is represented by whether a person was employed, and employed full time (thirty-five hours or more). Industry, occupation, employer-group size, and other benefits (sick pay, paid vacations, and retirement plans) are used to explain differences in whether employers offer insurance and serve as proxies for the generosity of coverage. We include information on whether an employer offered insurance and whether a person held employer coverage. All employment and job characteristics are measured at the time of the second interview for each year to coincide with the MEPS access questions. Health insurance. We include indicators of Medicare, full- or part-year private insurance coverage, and full- or part-year coverage by Medicaid or other public coverage. In some analyses we measure health insurance coverage at the time of the second interview for each year to coincide with the access measures and with employment and job characteristics. Health care system capacity. We use four variables derived from the ARF to measure county-level health care system capacity: number of physicians per capita, hospital beds per capita, and two measures of whether the entire county or part of the county was designated a health professional shortage area by the Health Re- 142 March/April 2003

5 sources and Services Administration (HRSA). Other local area characteristics. We include measures of region and whether the person resides in an urban area (standard metropolitan statistical area, or SMSA) drawn from MEPS and the percentage of county residents who are respectively black, white, and Hispanic from the ARF. ARF county-level measures of poverty rates, unemployment, and a wage index provide information about local economic conditions, which might explain differences in health insurance coverage and stresses on the health care system. Study Results Disparities in access and use. We see in Exhibit 1 both the racial and ethnic disparities and their persistence between 1996 and The percentage with a usual source of care remained virtually unchanged over the period, with about 5 percent more blacks than whites and 16 percent more Hispanics than whites lacking a EXHIBIT 1 Changes In Access To And Use Of Services, By Race And Ethnicity (All Ages), Access/use measure Usual source of care (sample size) Black Hispanic White All Any non-er ambulatory use (sample size) Black Hispanic White All Number of visits if any non-er ambulatory use (sample size) Black Hispanic White All Family experienced delays, difficulties, or unmet need (sample size) Black Hispanic White All Very satisfied family can get care they need (sample size) Black Hispanic White All 21, % a,b 69.5 a,b , % a,b 69.2 a,b , % a,b 69.0 a,b , % a 60.7 a , % a 57.6 a , % a 59.3 a , a 5.9 a , % a 5.8 a , a 5.3 a , % b 15.5 a,b , % b 12.7 a,b , % a,b 12.4 b , a,b 62.3 a,b , a,b 59.8 a,b , b,c 61.0 a,b SOURCE: Medical Expenditure Panel Survey (MEPS), a Significantly different from white at the.05 level. b Black-Hispanic difference significant at the.05 level. c Significantly different from 1996 at the.05 level. HEALTH AFFAIRS ~ Volume 22, Number 2 143

6 Research Challenge source of care. Although the disparity among Hispanics did not grow recently, as it did over the previous two decades, neither has it diminished. 14 Blacks and Hispanics were much less likely to use non-er ambulatory treatment than whites were and made fewer visits than whites did. Better health status is an unlikely explanation for this lower use, because (where differences are statistically significant) whites tend to report slightly better health. This pattern is even more pronounced among children under age eighteen, for whom variation in health care needs is less. About 76 percent of white children had an ambulatory visit in the study period, compared with 55 percent of black children and 60 percent of Hispanic children (data not shown). The two family questions on perceptions of access paint a more complicated picture. Blacks, despite their lower ambulatory use and lower access to a usual source of care, were less likely than whites were to report that anyone in the family experienced problems in getting needed care. By 1999 blacks were also equally likely to report that they were very satisfied their family could get the health care they needed, closing a disparity that existed in In contrast, Hispanics were more likely to report problems with family members getting needed treatment, although this difference was not statistically significant in However, the disparity between Hispanics and whites in satisfaction with the family s ability to get care was large and did not change between 1996 and The relatively larger gap between Hispanics and others in being very satisfied with their family s ability to get care, as opposed to reports of barriers getting access to treatment, may reflect different aspects of perceptions of access difficulties. The large gaps between Hispanics and whites in the measures of having a usual source of care and non-er ambulatory care use are reflected in the more subjective measures of perceptions of access problems. But blacks appear not to share those perceptions of access problems, despite large gaps in health care use and a smaller gap in having a usual source of care. Explaining disparities. We present the results of decompositions of selected disparities between blacks and whites and Hispanics and whites in Exhibit 2. For ease of graphical presentation, we aggregate several categories (demographic, socioeconomic, health status, area demographic, and economic characteristics) into the other characteristic category and present this along with health insurance, health care system capacity, and unexplained differences. The differences in health insurance coverage are important factors in explaining disparities in access but not the only factors. Health insurance explained 42 percent of the five-percentage-point black-white disparity in having a usual source of care but only 24 percent of the fifteen-percentage-point Hispanic-white disparity in this measure and about one-third of the large Hispanic-white disparity in satisfaction with the family s ability to get care. Differences in health insurance explained even smaller proportions of the disparities in any use and number of visits. Our findings concerning the role of health insurance are consistent with 144 March/April 2003

7 EXHIBIT 2 Decomposing Racial And Ethnic Disparities In Access To And Use Of Services (All Ages), 1998 Black-white disparities Usual source of care b Any non-er ambulatory care use c Health insurance Other characteristics Health care system capacity Unexplained factors a Hispanic-white disparities Usual source of care d Any non-er ambulatory care use e Satisfaction with family s ability to get care f SOURCE: Oaxaca-Blinder decompositions of 1998 full-year Medical Expenditure Panel Survey (MEPS) data. NOTES: R 2 statistics of.19,.24, and.09, respectively, for blacks (n = 3,338), Hispanics (n = 5,489), and whites (n = 12,989) in usual source of care decomposition regressions. R 2 statistics of.21,.21, and.12, respectively, for blacks (n = 3,431), Hispanics (n = 5,585), and whites (n = 13,216) in any non emergency room (ER) ambulatory care use decomposition regressions. R 2 statistics of.19 and.12, respectively, for Hispanics (n = 1,648) and whites (n = 5,717) in satisfaction with family s ability to get care decomposition regressions. a Difference in intercept and other coefficient estimates between groups. b Disparity is 5.4 percentage points. c Disparity is 15.7 percentage points. d Disparity is 16.9 percentage points. e Disparity is 18.9 percentage points. f Disparity is 16.8 percentage points. those reported by Robin Weinick and colleagues, especially the Hispanic-white disparities in having a usual source of care and ambulatory use. 15 Our measures of variation in health care system capacity explain almost none of the differences between groups. The coefficient estimates for these measures were almost never statistically significant in the separate black, Hispanic, and white regressions. This suggests either that they are poor measures of health care capacity as it relates to specific people because they are aggregate, county-level measures, or that differences in capacity levels do not explain access and use differences across individuals. Differences in other observable factors often explained as much as or more than did health insurance and our measures of health care system capacity combined. For example, these differences explained 53 percent of the black-white disparity in having a usual source of care, with local area demographic and economic indicators (28 percent), income (17 percent), and demographic characteristics (16 percent) being the most important factors, while education explained little. In contrast, education appears to play a substantial role in the disparities in ambulatory care use, explaining about 20 percent of the Hispanic-white and 10 percent of the black-white gaps, while income explains little. Demographic characteristics, es- HEALTH AFFAIRS ~ Volume 22, Number 2 145

8 Research Challenge pecially marital status, were the other substantial determinants of utilization disparities. No single factor stood out in explaining the large Hispanic-white disparity in satisfaction with the family s ability to get needed care. Health status differences played almost no role in explaining the disparities. The unexplained portions of the Oaxaca-Blinder decompositions were large for most of the disparities. For example, two-thirds of the Hispanic-white disparity in having a usual source of care, ten percentage points, cannot be explained by differences in observable characteristics. Only for the black-white disparity in having a usual source of care did the differences in observed characteristics explain most of the disparity. We also decomposed (not shown) the large disparities between blacks and Hispanics in having a usual source of care and satisfaction with their family s getting care. Differences in observable factors explain only 40 percent of the Hispanicblack disparity in usual source of care and 30 percent of the satisfaction disparity, with health insurance by far the most important observable factor. Pathways to health insurance coverage. Although it is not the only determinant, health insurance remains an important pathway to health care and a major focus for health policy. We look at three aspects of access to health insurance coverage: (1) access to private insurance through employment; (2) access to private insurance through marriage; and (3) the role of public insurance in mitigating differences in access to private health insurance. Differences in the pattern of coverage among single and married men and women ages are shown in Exhibit 3. We see that for both married and single people, private employer-group coverage is most prevalent among whites. The gap shrinks considerably for married versus unmarried blacks. While employer coverage is greater among married than among single Hispanics, the gap between married Hispanics and married whites is greatest. More than 80 percent of married whites are covered by private employer-group coverage, compared with slightly more than half of married Hispanics. Public insurance programs make up some of the differences for blacks and Hispanics, but the number of uninsured Hispanics remains very high. What accounts for the large racial and ethnic differences in private health insurance coverage? Whether a person or family holds private employer-group coverage is simply a product of whether they are employed, whether insurance is offered through their job, and whether they take the insurance that is offered. We decompose these different effects for married and single blacks, Hispanics, and whites, respectively, in Exhibit 4. For married people, we combine the information for husband and wife, so that we look for whether or not at least one spouse was employed, offered health insurance, and held insurance from a job that offered coverage. 16 Differences in employment, offer, and take-up rates are small between black and white couples, so they have similar rates of holding some employergroup coverage in the family. 17 Hispanic couples were much less likely than white 146 March/April 2003

9 EXHIBIT 3 Distribution Of Insurance Coverage, By Marital Status And Race And Ethnicity, Among People Ages 25 54, 1998 Characteristic Black Hispanic White Married Female Male Insurance coverage -single females Private employer group Other private Public Uninsured Insurance coverage -single males Private employer group Other private Public Uninsured Insurance coverage -married females Private employer group Other private Public Uninsured Insurance coverage -married males Private employer group Other private Public Uninsured 35.0% a 44.3 a 64.4% a 1.7 a 23.7 a a 1.1 a 18.3 a 33.3 a 74.9 a 1.5 a 7.0 a 16.6 a 74.6 a 1.1 a 8.4 a 15.9 a 40.2 a 5.8 a 23.2 a 30.9 a 43.4 a a 53.8 a a 34.6 a 51.5 a 1.9 a 5.5 a 41.1 a 67.8% SOURCE: Medical Expenditure Panel Survey (MEPS), second interview (Round 2/Round 4), 1998 (n = 8,990). a Significantly different from whites at the.05 level. couples were to have at least one spouse offered coverage through a job and were also less likely to take up offered coverage. We see that 68 percent of the difference in overall employer coverage rates between married Hispanics and whites is attributable to differences in offer rates. Alan Monheit and Jessica Vistnes find that the lower offer rates among Hispanics are attributable primarily to the types of jobs they hold. 18 During our study years, single people were less likely than their married counterparts were to be employed and, if employed, less likely to be offered health insurance. Regression-based decompositions confirm that some single-married differences are attributable to observed characteristics (for example, single people are younger), but these cannot explain all of the differences. Having a family unquestionably changes the calculus of the desirability of having health insurance coverage. Singles are also likely to have other unobserved characteristics that may make them less desirable both as marriage partners and in the labor market. But being married also increases the probability of having coverage simply because there are two people with the chance to work and to be offered health insurance coverage. The much lower rates of marriage among blacks, then, decreases their access to private health insurance coverage. HEALTH AFFAIRS ~ Volume 22, Number 2 147

10 Research Challenge EXHIBIT 4 Decomposition Of Differences In Employer-Group Health Insurance Coverage, By Race And Ethnicity, Head Of Household Ages 25 54, 1998 Black Hispanic White Married Hold coverage Any employment Offer from current job Took offer Married decomposition of difference in held coverage vs. white Employment Offer from current job Took offer 82.4% % a 94.3 a 67.4 a 89.2 a 82.6% b b b c c c Single Hold coverage Any employment Offer from current job Took offer 44.9 a 72.8 a a 36.1 a 70.6 a 57.5 a 88.8 a Single decomposition of difference in held coverage vs. white Employment Offer from current job Took offer c c c SOURCE: Medical Expenditure Panel Survey (MEPS), second interview (Round 2/Round 4), 1998 (n = 6,036 households with heads ages 25 54). a Significantly different from white at the.05 level. b Difference is not statistically significant. c Not applicable. Lower employment among single blacks accounts for almost three-quarters of the disparity in employment-related insurance coverage compared with single whites. For single Hispanics, lower employment and offer rates are equally important in explaining the disparity in employment-related insurance. Explaining disparities among the insured. Health insurance coverage is a key pathway to health care services, but disparities persist even among the insured. We use Oaxaca-Blinder decompositions to increase understanding of the source of these disparities. We focus on people who were covered by private insurance all year long and did not have Medicaid, Medicare, or any other source of public health insurance. Looking at Exhibit 5, we see that even with stable private coverage, a disparity of 3.7 percentage points for blacks and 4.5 percentage points for Hispanics remains in access to a usual source of care. The gaps in the percentage using non-er ambulatory care services and the number of visits (not shown) are even larger. Also, Hispanics with stable private insurance are still 6.5 percentage points less likely than whites are to report that they are very satisfied with their family s ability to get needed health care. Our measures of health care delivery system capacity explain little of the remaining disparities for blacks and Hispanics with stable private insurance coverage. Similarly, our proxies for generosity of coverage employer-group size, industry, occupation, and other employee benefits explain little. 148 March/April 2003

11 EXHIBIT 5 Decomposing Racial And Ethnic Disparities In Access To And Use Of Services Among The Privately Insured (All Ages), 1998 Black-white disparities Usual source of care b Any non-er ambulatory care use c Hispanic-white disparities Usual source of care d Any non-er ambulatory care use e Satisfaction with family s ability to get care f Health care system capacity Other characteristics Unexplained factors a SOURCE: Oaxaca-Blinder decompositions of 1998 full-year Medical Expenditure Panel Survey (MEPS) data. NOTES: Sample is limited to people with private health insurance and no Medicaid, Medicare, or other public coverage during theentire1998calendaryear.r 2 statistics of.21,.19, and.06, respectively, for blacks (n = 1,234), Hispanics (n = 1,751), and whites (n = 7,439) in usual source of care decomposition regressions. R 2 statistics of.23,.17, and.09, respectively, for blacks (n = 1,256), Hispanics (n = 1,772), and whites (n = 7,496) in any non emergency room (ER) ambulatory care use decomposition regressions. R 2 statistics of.19 and.07, respectively, for Hispanics (n = 350) and whites (n = 2,495) in satisfaction with family s ability to get care decomposition regressions. a Difference in intercept and other coefficient estimates between groups. b Disparity is 3.7 percentage points. c Disparity is 13.8 percentage points. d Disparity is 4.5 percentage points. e Disparity is 15.3 percentage points. f Disparity is 6.5 percentage points. Other observed differences are associated with disparities in access and use. For example, together they account for 68 percent of remaining gaps in having a usual source of care for blacks, with demographic characteristics (22 percent), especially marital status, and income (19 percent) playing the most important roles. Income was also the most important factor in explaining the Hispanic-white disparity in having a usual source of care, but only 17 percent overall can be explained by differences in observable attributes. Demographic characteristics, education, and income, to a lesser extent, all play important roles in explaining the large gaps in ambulatory care use. For example, differences in education explain about 17 percent of the Hispanic-white gap. However, unexplained differences still remain large in the disparities among the privately insured, ranging from 24 percent for the black disparity in having a usual source of care to 83 percent for Hispanics in this same measure. Discussion Sources of disparities. The stubborn persistence of disparities in access and use of health care services among racial and ethnic groups, especially Hispanics, re- HEALTH AFFAIRS ~ Volume 22, Number 2 149

12 Research Challenge mains a vexing problem for health policymakers. This paper asks if we can identify the internal and external factors that are the source of these disparities for ambulatory care. Policymakers need to know which factors are most important in explaining disparities and where health policy reforms may be most efficient in reducing them. Not surprisingly, our decomposition analyses suggest that the answer to our central question can we identify the source of disparities is both yes and no. Health insurance. Clearly, health insurance is important. Differences in insurance coverage explained up to one-third of Hispanic-white disparities and two-fifths of black-white disparities in having a usual source of care. Increasing health insurance coverage would no doubt increase access for all Americans and reduce racial and ethnic disparities. However, achieving the goal of increased health insurance coverage remains a challenge for health policy. Our analyses suggest that the disparities in access to employment-related coverage for blacks can be traced to lower employment among single blacks and lower marriage rates. For Hispanics, the lower levels of employment among singles also reduce access to employment-related coverage. Incremental insurance reforms, within the context of a predominantly employer-based system, are unlikely to have direct impacts on these factors. The types of jobs that both single and married Hispanics hold are much less likely to offer insurance than is true for jobs held by other ethnic groups, which further contributes to disparities in access to health insurance. Again, health policy cannot change the types of jobs that Hispanics hold, but it might increase offers of coverage through existing jobs. While health insurance clearly matters, we find that it explains only a portion of disparities, as other studies have found. 19 Moreover, these disparities persist even among insured blacks and, especially, Hispanics. This suggests that health insurance coverage alone cannot eliminate disparities. Health care delivery system. We were surprised that our measures of health care system capacity explain so little of the large disparities in access and use. Increasing delivery system capacity and enhancing the safety net, especially in underserved areas, has been a long-standing focus of policy for increasing access and reducing disparities. However, county-level capacity measures almost certainly do not capture well the delivery system faced by each person, and how this systematically differs by racial and ethnic groups within counties. Efforts to carefully target resources and interventions where they are needed most within neighborhoods for example, school-based programs or increasing the capabilities of local health care systems to provide culturally competent care to diverse populations may be quite effective in reducing disparities. Unfortunately, it is difficult for large-scale national data sets such as MEPS to adequately assess the impact of these types of programs in reducing disparities. However, MEPS is in the process of being geocoded down to the census tract level, which holds the promise of incorporating better measures of local delivery system capacity to study their impact on access and disparities. 150 March/April 2003

13 Since attitudes and preferences play a large role in use and access, they may also help to explain disparities. Income and education. Our decompositions revealed that income and education are two important policy-relevant factors associated with disparities. Income was relatively more important explaining disparities in having a usual source of care and family perceptions of access, while education was relatively more important in explaining disparities in ambulatory care use. Both income and education may act as a proxy for a wide range of attributes that we cannot measure, including differences in attitudes and care-seeking behavior (for example, mistrust of the system), as well as affordability of care and insurance generosity. While health policy cannot directly affect income and education, it can help mitigate problems of affordability or better understanding when care is needed. Explaining the unexplained. Much of the disparities among blacks, Hispanics, and whites remain unexplained even after differences in a large number of characteristics are controlled for. However, there are a number of potential explanations for these remaining disparities, which offer the hope that the most important factors can be identified and effective strategies devised to reduce or eliminate their role in disparities. We note just a few potential factors here. Perhaps most importantly, we cannot control for systematic differences in attitudes toward risk, the health care system, and care-seeking behavior, as well as other cultural issues. Since attitudes and preferences play a large role in health care use and access, they may also help to explain disparities. Second, we used employer-group size, industry, occupation, and other proxies with limited power to examine the extent to which the generosity of insurance coverage explained remaining racial and ethnic disparities among even insured populations. Detailed benefit data from employers abstracted in MEPS for 1996, and possibly in future years, can provide better measures of benefit generosity as well as better information about managed care attributes of private plans. Other factors that are beyond the scope of this paper include the sheer diversity of the Hispanic American population and its implications for thinking about issues of access. Of course, outright discrimination or subtle biases may also be important. Implications of disparities for health. Ultimately, we care about disparities in access to ambulatory care, because we care about improving the quality of health care and the health of all Americans. For example, having access to a usual source of care is associated with use of preventive care services, which are expected to improve health outcomes in the long run. Also, while our study is primarily focused on disparities in access to ambulatory care, our results have implications for other settings. It is likely that access problems with ambulatory care serve as a marker for more general problems with access to health care. Moreover, reduced access to ambulatory care services may exacerbate chronic or acute conditions, leading to hospi- HEALTH AFFAIRS ~ Volume 22, Number 2 151

14 Research Challenge talizations that might have been preventable. Reducing barriers to ambulatory and other health care may both improve the overall health of Americans and reduce racial and ethnic disparities. Additional caveats. In light of the large number of potential factors and explanations for disparities that we cannot address in this study, we note some important limitations of decomposition methods, which cannot control for these unobserved factors. Some of the unobserved differences in characteristics are simply reflected in the unexplained component of the total disparities between groups. That is, if we could add these characteristics to the regression models, less of the total disparity would remain unexplained, but the portion attributable to differences in each of the characteristics we already include would not change. But these unobserved differences might also bias the estimates that are attributable to observed characteristics. For example, the proportion attributable to insurance coverage might be overestimated in the classic case of adverse selection. Unmeasured differences in local health care delivery system might also be correlated with education and income, so that their true impact could be overstated. As a result, we view these reduced-form decompositions as associations rather than causal relationships. The authors are especially grateful to Steve Hill, Phil Cooper, Barbara Schone, and Jessica Vistnes for advice and assistance, and to three anonymous referees for many helpful suggestions. The views expressed in this paper are those of the authors, and no official endorsement by the Agency for Healthcare Research and Quality or the Department of Health and Human Services is intended or should be inferred. 152 March/April 2003

15 NOTES 1. R.M. Weinick, S.H. Zuvekas, and S.K. Drilea, Access to Health Care Sources and Barriers, 1996, MEPS Research Findings no. 3, Pub. no (Rockville, Md.: Agency for Healthcare Research and Quality, 1997); and N.A. Krause, S. Machlin, and B.L. Kass, Use of Health Services, 1996, MEPS Research Findings no. 7, Pub. no (Rockville, Md.: AHRQ, 1999). 2. U.S. Department of Health and Human Services, Healthy People 2010: Understanding and Improving Health, 2d ed. (Washington: U.S. Government Printing Office, November 2000). 3. Institute of Medicine, Crossing the Quality Chasm: A New Health System for the Twenty-first Century (Washington: National Academy Press, 2001). 4. L. Ammons, Demographic Profile of Health-Care Coverage in America in 1993, Journal of the National Medical Association 89, no. 11 (1997): ; P.W. Newachek, D.C. Hughes, and J.J. Stoddard, Children s Access to Primary Care: Differences by Race, Income, and Insurance Status, Pediatrics 97, no. 1 (1996): 26 32; and S.H. Long, Public versus Employment-Related Health Insurance: Experience and Implications for Black and Non-Black Americans, Milbank Quarterly 65, Suppl. 1 (1987): P.F. Cooper and B.S. Schone, More Offers, Fewer Takers for Employment-Based Health Insurance, , Health Affairs (Nov/Dec 1997): R.M. Weinick, S.H. Zuvekas, and J.W. Cohen, Racial and Ethnic Differences in Access and Use of Health Care Services, , Medical Care Research and Review 57, Suppl. 1 (2000): 36 54; and S.H. Zuvekas and R.M. Weinick, Changes in Access to Care, : The Role of Health Insurance, Part II, Health Services Research (April 1999): Ibid. 8. Ibid. 9. Cooper and Schone, More Offers, Fewer Takers ; and A.C. Monheit and J.P. Vistnes, Race/Ethnicity and Health Insurance Status: 1987 and 1996, Medical Care Research and Review 57, Suppl. 1 (2000): J.W. Cohen et al., The Medical Expenditure Panel Survey: A National Health Information Resource, Inquiry 33, no. 4 (1996): R.L. Oaxaca, Male-Female Wage Differentials in Urban Labor Markets, International Economic Review 14, no. 3 (1973): ; and A. Blinder, Wage Discrimination Reduced Form and Structural Estimates, Journal of Human Resources 8, no. 4 (1973): See R. Oaxaca and M.R. Ransom, Identification in Detailed Wage Decompositions, Review of Economics and Statistics (February 1999): We do not consider adults ages here because of their much more complicated dynamics related to transitions from childhood to their own families and from school to the workforce. Adults ages present a different set of complexities related to retirement decisions that are outside the scope of this paper. 14. Weinick et al., Racial and Ethnic Differences. 15. Ibid. 16. The vast majority of employers that offer health insurance coverage in the United States offer family coverage (according to unpublished MEPS Insurance Component data). 17. A detailed discussion of differences in the (higher) take-up rates reported here and elsewhere in the literature is available from the authors. Send to Sam Zuvekas at szuvekas@ahrq.gov. 18. Monheit and Vistnes, Race/Ethnicity and Health Insurance Status. 19. Weinick et al., Racial and Ethnic Differences ; and Zuvekas and Weinick, Changes in Access to Care, HEALTH AFFAIRS ~ Volume 22, Number 2 153