We used data from the 2006 Medical Expenditure Panel Survey (MEPS) and the 2000 US Census Summary File 1 to determine whether residential segregation was associated with healthcare use. MEPS is a longitudinal survey that covers the United States civilian noninstitutionalized population. The Agency Healthcare Research and Quality (AHRQ) fields the MEPS based on a sampling frame of the National Health Interview Survey. The survey consists of five interviews conducted over a two and one half year period. MEPS is unique in its ability to link data on individuals and households (including demographics, health status, employment, and income) to information on their use of health services. This information includes sources of payment for specific medical services, health insurance status, and the details of individual/household health plans. Each of the five interviews conducted for the panel asked about all health care utilization and associated expenditures for a specific period of time, and these periods cumulatively covered the two-year period.
There are 23,791 non-institutionalized adults age 18 and over in the 2006 MEPS. We restricted this sample to non-Hispanic Whites, Hispanic, and non-Hispanic African Americans living in MSAs for which we had zip code level data for a total of 17,514. (For ease of composition, we dropped the non-Hispanic suffix for Whites and African Americans.) We were unable to study Asians because there were only 48 Asians out of the 896 in the MEPS living in predominantly Asian zip codes. Sample includes 9,210 Whites, 3,286 African Americans and 5,018 Hispanics. We divided the sample by racial and ethnic composition of the respondent’s zip code: predominantly White, predominantly African American, predominantly Hispanics, and integrated/other. The definitions for each of these categories are listed below. displays the distribution of the sample adults.
Distribution of the Sample of Adults in Person Level reported MSAs in the 2006 Medical Expenditure Panel Survey by Racial Composition of Zip Code.
We used five measures of healthcare services use: office based physician visits, outpatient department physician visits, visits to nurses or physician’s assistants, visits to other health professionals, and having a usual source of care (USC). Physician visits included services provided by primary care and specialty physicians. Visits to nurses or physician’s assistants included services provided by nurses and nurse practitioners, physician’s assistants or midwives. Nurse and nurse practitioners provided the bulk of these services in this category. Other health professional visits included services provided by chiropractors, dentist, optometrists, podiatrists, physical therapists, occupational therapists, psychologists, social workers, acupuncturists, massage therapists, homeopathic/naturopathic/herbalists or alternative/complementary care professionals. Chiropractors provided most of the services in this category, followed by physical therapists and psychologists. Our models controlled for age, gender, marital status, poverty status, insurance status, income, educational attainment, employment status, region, and health status.
We linked the MEPS data to the race-ethnic composition of the zip code. Race-ethnic composition was measured with four indicator variables that denoted whether the zip code was predominantly (50% or more) White, predominantly African American, predominantly Hispanic, or Integrated/Other (no predominant group or predominantly other). We used dichotomous variables to define the racial/ethnic composition of a zip code because the distribution of the percentage of minority residents across zip codes is skewed. Of zip codes in MSAs, only 5.0% were predominantly African American and 3.9% were predominantly Hispanic; however, large proportions of African Americans (40.9%) and Hispanics (37.7%) in MSAs reside in predominantly African American and Hispanic zip codes, respectively. We combined the respondents’ race/ethnicity with the race/ethnic composition of the zip code to create 12 race/ethnicity-place indicator variables e.g., Whites living in predominantly White zip codes, African Americans living in predominantly White zip codes, and Hispanics living in predominantly White zip codes.
We estimated four sets of logistic regression analyses to compute the odds of having a visit or a usual source of care for African Americans and Hispanics relative to Whites. The first set of regressions is our base model that estimates disparities by race and ethnicity controlling only for individual-level characteristics. In this base model, Whites were the reference group and the coefficients of interest were the indicator variable for African American and Hispanic race/ethnic identity. We exponentiated the coefficients on the African American and Hispanic indicator variables to compute the odds of having a visit or usual source of care relative to Whites. The second set of models adds place indicators to the base specification. In this model, predominantly White zip codes were the reference place and the coefficients of interest were the indicator variables for predominantly African American, predominantly Hispanic and Integrated/Other zip codes. The third set of models estimated disparities by allowing interactions between race-ethnicity and place. These models tested whether the association of place was uniform across race and ethnic groups. For these models, instead using the African American and Hispanic indicator variables, we substituted eleven race/ethnicity-place indicator variables using Whites living in predominantly White zip codes as the reference category. The fourth set of models estimated disparities by race and ethnicity within place. We stratified the sample by racial and ethnic composition of the zip code and used the traditional specification to determine whether there were disparities in utilization within zip code type. We performed the analysis using the survey estimation procedures in STATA© 11 to control for strata and sampling units because the MEPS has a complex sampling design.