|
Would Safety-Net Expansions Offset Reduced Access Resulting From Lost Insurance Coverage? Race/Ethnicity Differences
This study simulated whether increased community health center (CHC) funding under the Bush administration narrowed racial/ethnic gaps in access to care among low-income people. Expanded CHC funding resulted in small increases in access to care, more so for minorities than for whites. Spanish-speaking Hispanics had the largest improvements in access in the simulation. However, minorities experienced bigger drops in insurance coverage. The net result was no improvements in the access measures for Spanish-speaking Hispanics and slight decreases in access for whites, English-speaking Hispanics, and African Americans. Access gaps either remained the same or worsened slightly for English-speaking Hispanics and African Americans relative to whites.
RACIAL AND ETHNIC MINORITIES consistently have higher rates of uninsurance than whites, by almost twofold for African Americans and roughly threefold for Hispanics.1 Recent data indicate that the gap is getting larger. Between 2000 and 2004, the percentage of nonelderly people without insurance increased 3.1 percentage points for African Americans, 2.2 percentage points for Hispanics, and 1.8 percentage points for white non-Hispanics.2 Alongside these significant differences, national reports on health care disparities document minorities’ consistently lower levels of access to care.3 Coinciding with increased uninsurance, the Bush administration expanded federal funding for community health centers (CHCs) by more than 50 percent—roughly $250 million—between 2001 and 2006.4 The president’s 2006 State of the Union address called for continued expansion, setting a goal of establishing a CHC "in every high-poverty county in America that can support one."5 This policy was motivated in part by the desire to maintain or expand access to care for people who lose or cannot afford private insurance coverage, since CHCs typically use grant funds to provide care to the uninsured. CHCs also disproportionately serve minority patients: In 2001, 35 percent of CHC users were Hispanics, and 25 percent were African Americans.6 This study estimates the effects of CHCs and insurance coverage on access to care by race and ethnicity, and it simulates whether expanded funding for CHCs narrowed the gaps in access to care between whites and minority groups. Since racial and ethnic minorities use CHCs more than non-Hispanic whites do, federal investments in CHCs should narrow differences in access. However, a growing gap in insurance coverage between whites and racial/ethnic minorities might frustrate this goal. Previous studies. Marsha Lillie-Blanton and Catherine Hoffman reviewed four multivariate analyses of the role of health insurance coverage in explaining racial/ethnic disparities in access to health care.7 With one exception, these studies found that differences in health insurance coverage between whites, Hispanics, and African Americans accounted for a sizable share of the differences in having a usual source of care. Two of these studies also examined the effects of area resources and found that they explained very little of the difference in access to care.8 Another unpublished study also found only a weak relationship between a county’s broadly defined safety-net resources and access to care.9 These studies shared two major limitations: They used indirect measures of safety-net capacity, and, more importantly, they did not account for possible endogeneity bias in both insurance coverage and safety-net capacity. Endogeneity bias refers to the difficulty of disentangling the direction of causation between insurance coverage, CHC capacity, and access to care. For example, CHCs might be more likely to locate in areas with poor access or high rates of uninsurance, or greater local CHC capacity might influence some people to forgo insurance coverage. In previous work, Jack Hadley and Peter Cunningham showed that safety-net availability, measured by the distance to the nearest safety-net provider, is in fact endogenous and that failure to adjust for endogeneity understated the estimated impact of safety-net proximity on access.10 Using an appropriate statistical method, they found that greater proximity to safety-net providers has a statistically significant and positive effect on several measures of access to care.11 In a study of the effects of insurance coverage and safety-net capacity on having a usual source of care, Cunningham and Hadley used an improved measure of safety-net availability (total grant revenues received by federally qualified CHCs within a five-mile radius of survey respondents) and treated both insurance coverage and safety-net capacity as endogenous.12 After adjusting for endogeneity bias, their study concluded that both insurance coverage and safety-net capacity have positive effects on having a usual source of care. Current analysis. The analysis on which we report here follows and extends our prior work. We based our simulations on statistical models that treat both insurance coverage and safety-net capacity as endogenous variables but used an improved measure of safety-net capacity: total CHC grant revenues per poor person within a five-mile radius of survey respondents. This measure captured both the absolute capacity of nearby CHCs and the potential demand for care from CHCs by poor people. We extended this earlier work by allowing all effects to vary by race and ethnicity and by distinguishing between English- and Spanish-speaking Hispanics (based on whether the interview was conducted in Spanish or English) in estimating the access models and comparing the effects of coverage and CHC capacity.13
Data. The analysis used data from three rounds of the CTS survey, conducted in 1998–99, 2000–01, and 2003. The CTS is a nationally representative telephone survey conducted primarily in sixty randomly selected communities in thirty-four states and the District of Columbia.14 Since CHCs primarily serve low-income people, the analysis sample was limited to nonelderly people with incomes below 300 percent of the federal poverty level. The total sample size is 47,992 people categorized into four racial/ethnic groups: non-Hispanic whites, Hispanics interviewed in English, Hispanics interviewed in Spanish, and African Americans.15 We distinguish between English- and Spanish-speaking Hispanics because prior research indicates that Spanish-speaking Hispanics have a much higher uninsurance rate, are much more likely to be noncitizens, and generally have less access to public insurance or jobs that offer private insurance than is true for English-speaking Hispanics.16 Spanish speakers might also be more likely than English speakers to use CHCs, which often provide on-site interpreters and might be more culturally sensitive.17 Access measures and statistical estimation. We analyzed two measures of access to care: whether the person has a usual source of care, and whether the person had any ambulatory care visits with a health professional in the previous year. Having a usual source of care is an important indicator of whether a person is likely to be able to obtain care if needed, while having had at least one ambulatory care visit is a marker for whether the person was able to realize a perceived need for care by actually seeing a health care provider. CHC capacity is defined as the amount of CHC grant revenue per poor person (family income less than 100 percent of poverty) within five miles of the survey respondent.18 We identified all CHCs within five miles of each respondent using information on the latitudes and longitudes of CHCs’ and survey respondents’ five-digit ZIP codes. We then summed the CHCs’ grant revenues (obtained from the Uniform Data System maintained by the Health Resources and Services Administration, Bureau of Primary Health Care) within a five-mile radius of each survey respondent as an indicator of CHCs’ financial capacity to provide care to uninsured people.19 Total CHC grant revenue was then divided by the number of poor people within the same radius, to adjust the capacity measure for the potential demand for CHC services. We simulated the effects of changes in CHC funding and insurance coverage on access to care from the parameters of multivariate regression models that control for the effects of several important sociodemographic characteristics: age, education, self-reported general health status, marital status and the presence of children, attitudes toward risk, and type of community (small metropolitan or non-metropolitan versus large metropolitan). To allow the results to vary with race/ethnicity, all of the models’ independent variables were interacted with dummy variables for race/ethnicity (English-speaking Hispanic, Spanish-speaking Hispanic, and African American relative to non-Hispanic white).20 We estimated the models’ parameters using instrumental variable estimation to test and adjust for potential endogeneity bias.21
Levels of access by race/ethnicity, insurance status, and CHC capacity. Regardless of race, ethnicity, and the level of nearby CHC capacity, insured people have better access to care than uninsured people (Exhibit 1  ). In no case, moreover, is the level of access for the uninsured with a high level of CHC capacity greater than for the insured with no nearby CHCs. Within each of the two insurance-status groups, non-Hispanic whites tend to have the best levels of access, while Spanish-speaking Hispanics have the worst.
Holding insurance status fixed, the association between the access measures and the level of CHC capacity is more ambiguous. Among the uninsured, non-Hispanic whites and English-speaking Hispanics with no CHCs within five miles appear to have better access than similar people with high levels of CHC availability. However, for Spanish-speaking Hispanics and African Americans, the proportion with an ambulatory care visit increases monotonically as CHC capacity increases. However, the usual-source-of-care measure suggests that there is little difference between no CHC availability and a low level of CHC availability. These ambiguities are indicative of endogeneity bias. Since CHCs are more likely to locate in areas with poor access to care, the observed relationship between CHCs and access probably understates the true effect. The association between insurance status and access looks more clear-cut. However, its magnitude could also be biased if people who have a strong preference for having a usual source of care or who anticipate using medical services are more likely to seek insurance coverage. Multivariate regression estimates. The statistical tests indicated the presence of significant endogeneity bias and supported the use of instrumental variable analysis to estimate the multivariate regression models. The resulting parameter estimates are generally much larger in magnitude than parameters estimated without the adjustment for endogeneity bias.22 The multivariate results indicate (1) that the uninsured are much less likely than the insured to report having either a usual source of care or an ambulatory care visit in the past year, and (2) that people who live in areas with greater CHC capacity are more likely than those not living in such areas to have a usual source of care and an ambulatory care visit in the past year. The multivariate results also imply that insurance status has the same effects on access in each of the racial and ethnic groups. However, CHC capacity appears to have a bigger impact on minorities’ reporting that they have a usual source of care than it does for white non-Hispanics. The CHC capacity measure shows smaller differences across racial and ethnic groups in affecting ambulatory care visits: The only statistically significant difference is between African Americans and non-Hispanic whites. Simulations of changes in insurance status and CHC capacity. To provide information about associations between variations in insurance status and CHC capacity and the relative magnitudes of changes in the access measures, we used the multivariate regression models to simulate the effects of four hypothetical policy changes: (1) reduce the uninsurance rate by 5 percent for each racial/ethnic group; (2) eliminate uninsurance (enact universal coverage); (3) increase aggregate CHC grants by about $250 million, which mimics the Bush administration’s increase in CHC funding between 2001 and 2004, and (4) increase uninsurance by amounts similar to the national changes between 2000 and 2004 and simultaneously increase CHC capacity by $250 million. (This fourth simulation approximates the national experience since 2000 and addresses the primary policy question raised here.)
The simulations show that both reducing uninsurance and increasing CHC capacity would improve both measures of access (Exhibit 2
Spanish-speaking Hispanics appear to benefit the most from both policies. If uninsurance were eliminated, access would increase the most for Spanish-speaking Hispanics because they by far have the highest initial level of uninsurance. However, even these substantial improvements in access would reduce the gap between them and non-Hispanic whites by only about half. The final simulation estimates the combined consequences of an increase in uninsurance, which reduces access, and an increase in CHC capacity. The changes in uninsurance and CHC capacity were set to approximate actual national experience between 2000 and 2004. The simulations indicate that the expansion of CHC capacity generally offset the adverse effects on access of the increase in uninsurance rates between 2000 and 2004. For Spanish-speaking Hispanics, the CHC expansion appears to have fully offset the effect of increased uninsurance, with no change in either of the two access measures. However, the simulations indicate that the other racial and ethnic groups experienced small decreases in access (–0.1 to –0.3 percentage points).
This analysis suggests that policies to reduce uninsurance and expand CHC capacity will increase low-income people’s access to care, as measured by having a usual source of care and an ambulatory care visit in the past year. Moreover, minority groups’ access appears to be more responsive to changes in insurance coverage and CHC capacity than non-Hispanic whites’ access. Thus, if policy changes were targeted equally to all low-income people, the gaps in access between whites and nonwhites should be reduced. However, although eliminating uninsurance would have a sizable impact on minorities’ access levels, it would not eliminate access gaps completely. The simulations also suggest that the Bush administration’s CHC expansion might have offset much of the adverse effects on access of the recent increases in uninsurance. However, it is important to emphasize that the measures analyzed in this study represent only the most basic forms of access. Although they are important indicators of access to primary and first-contact care, they might not capture variations in access to specific types of care (such as specialists, prescription drugs, expensive therapeutic procedures, or hospital care) or the total amount of care received. Investigating the relationships between insurance status and CHC capacity on these additional access measures is important for obtaining a complete picture of the extent to which CHCs fully substitute for insurance coverage. Another limitation is that the simulations did not distinguish between effects in rural and urban areas. CHCs are less widely available in rural areas; when they do exist, they tend to treat more privately insured and fewer Medicaid or uninsured people than urban CHCs.23 (In the data used in this analysis, only 13.4 percent of nonmetropolitan residents lived within five miles of a CHC, compared with 42.1 percent of metropolitan residents; among those within the five-mile radius, 20.5 percent of the nonmetropolitan sample were uninsured and 16.9 percent were covered by Medicaid, compared with 29.8 percent uninsured and 25.6 percent with public coverage in the metropolitan sample.) Thus, the impact of expanded CHC funding in rural areas might be more sensitive to whether new CHCs are established, compared to urban areas where existing CHC capacity could be a more limiting factor. It is also important to emphasize that the simulations of expanding CHC capacity were relatively crude and did not attempt to replicate the actual changes in CHC capacity resulting from the Bush administration’s increases in CHC funding. Nor did the simulations distinguish between expansions of existing CHCs and funding new CHCs in areas that do not have any. Future research should examine the impacts on access to care in communities that specifically expanded CHC capacity under the Bush administration’s policy, distinguishing between rural and urban communities and between newly established CHCs and expansions of existing CHCs. In spite of these limitations, the simulations illustrate the importance of adjusting for the endogeneity bias, which tends to be inherent in nonrandomized observational data, when drawing policy implications from multivariate regression statistical analysis. Without the adjustment for endogeneity bias, the analysis would have implied that CHC capacity does not have a significant and positive effect on access, especially for racial and ethnic minorities. By using an appropriate statistical method, our analysis leads to the important policy implication that expanding CHC capacity would be valuable for helping to reduce access disparities even if insurance coverage expands. The findings also suggest that further expansions in CHC capacity will be needed to maintain access to care if insurance coverage continues to decline.
Jack Hadley (jhadley{at}hschange.org) is a senior fellow at the Center for Studying Health System Change (HSC) and a principal research associate at the Urban Institute in Washington, D.C. Peter Cunningham is a senior fellow at HSC. J. Lee Hargraves is a research associate professor in the Department of Family Medicine and Community Health at the University of Massachusetts Medical School in Worcester. The research for this study was supported by a grant from the Robert Wood Johnson Foundation to the Center for Studying Health System Change (HSC). The authors gratefully acknowledge James Reschovsky of HSC for his helpful comments on a preliminary draft of this paper, Cynthia Saiontz-Martinez of Social and Scientific Systems for excellent computer programming assistance, and Gretchen Nowlin of HSC for help with manuscript preparation.
This article has been cited by other articles:
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Top
Study Data And Methods
Study Findings
