The study used data from the household component (HC) of the 2000 MEPS, a survey sponsored by the Agency for Healthcare Research and Quality (AHRQ) and the National Center for Health Statistics (NCHS). We used the 2000 full-year consolidated data files (HC-050) from the MEPS HC survey, which contained questions on demographic characteristics, health conditions, health status, and use and expenditure of medical services. The overall response to the 2000 MEPS was 65.8% (17).

In addition, the 2000 MEPS consolidation file included disease-specific information from a series of surveys that included diabetes. In the beginning of calendar year 2000, MEPS conducted a new series of interviews as part of AHRQ's focus on quality of health care, asking questions about several specific medical conditions such as diabetes, asthma, high blood pressure, heart disease (including coronary heart disease, angina, and myocardial infarction), stroke, emphysema, and joint pain  (17). HC respondents received A Survey About Your Diabetes Care — The Diabetes Care Survey, a self-administered paper-and-pencil questionnaire, if they responded yes to a question about whether their diabetes had been diagnosed by a health care professional.

The Diabetes Care Survey is a series of questions (18) about diabetes management and includes the number of times respondents reported having a glycated hemoglobin A1c (HbA1c) test, the number of times they reported having their feet checked for sores or irritation in 2000, and the last time they reported having an eye examination. Respondents were also asked to report on diabetes treatment (diet, oral medications, and insulin) and complications (eye problems and kidney problems caused by diabetes). Lower extremity amputation, a serious complication of diabetes, and type and duration of diabetes were not included in the Diabetes Care Survey. The response rate for the Diabetes Care Survey was 91.4% (17).

Out of 24,791 respondents to the 2000 MEPS, 1021 adults (18 years and older) with diabetes responded to the Diabetes Care Survey. Race in this study was defined using two variables in MEPS indicating race (categorized as American Indian, Aleut Eskimo, Asian and Pacific Islander, black, and white) and ethnicity (categorized as Hispanic, non-Hispanic African American, and other). The race variable in this study was categorized into three groups including non-Hispanic white, non-Hispanic African American, and Hispanic. We excluded American Indians, Aleut Eskimos, and Asian Americans and Pacific Islanders because the number of respondents (37 out of 1021 total adult Diabetes Care Survey respondents, or 3.6%) was too small to analyze. Thus, the final sample size was 984 adults (18 years and older) with diabetes, of whom 540 (54.9%) were white, 210 (21.3%) African American, and 234 (23.8%) Hispanic. Population weights from the full-year sample were used to generate estimates for the national population of adults with diabetes as of the index date December 31, 2000. The population weighted study sample represents 16,808 adults with diabetes in 2000.

Health care use included ambulatory care visits and prescription drug fills. Ambulatory care visits in both office-based settings and hospital outpatient settings were included. Physician and nonphysician visits were included in ambulatory care visits. Nonphysicians included nurse practitioners, physician assistants, chiropractors, podiatrists, physical and occupational therapists, and social workers. Prescription fills were defined as all prescribed medications purchased in 2000.

Health care costs in MEPS were defined as direct payments rather than charges by providers, including out-of-pocket costs and third-party payments (17). Administrative costs (i.e., costs not directly related to patient care), over-the-counter drugs, and payments for alternative care services were not included in MEPS. The three components of health care costs in this study included ambulatory care costs, prescription drug costs, and total health care costs.

Ambulatory care costs included payment for providers in office-based settings and hospital-based settings. The cost of prescription fills includes all amounts paid out-of-pocket and by third-party payers for prescribed medications purchased in 2000. Total health care costs were defined as the sum of payments for care for ambulatory care visits, hospital inpatient stays with zero-night admission, emergency department visits, dental care, home health care, and other care including vision aids, medical supplies and equipment, and prescription medications.

Process and outcome measures specific to diabetes care included management, treatment, and diabetes-related complications. Diabetes care management included whether the respondents had received an HbA1c test, had had their feet checked for sores or irritation, and had received an eye examination. Diabetes treatment included diet modification, oral medications, and insulin therapy reported by Diabetes Care Survey respondents. The variables for diabetes complications included eye and kidney problems caused by diabetes. In this study, all measures were binary (yes or no) variables using the original MEPS variables in the Diabetes Care Survey for statistical analysis. Positive answers were coded 1; for example, having received any HbA1c test was coded as 1 if a Diabetes Care Survey respondent reported having received one or more than one HbA1c test; we recorded 0 otherwise.

Table 1 shows the definitions of dependent and independent variables. The primary independent variable of interest is race. Covariates included age, sex, education, marital status, living in a standard metropolitan statistical area (MSA), income, health insurance, having a usual source of care (USC) provider, self-rated health status, employment, and comorbidity in multivariate analyses. Having a USC is a commonly used measure of access to care.

The data were analyzed using bivariate and multivariate methods with Stata 7 (Stata Corp, College Station, Tex). Chi-square (Χ²) tests were used to compare racial and ethnic differences in individual characteristics. Logistic regression was used to examine differences in diabetes care among the three racial and ethnic groups (white, African American, and Hispanic). For multivariate analyses of utilization measures, which are count variables, we used count data methods (19). Because utilization data are usually not normally distributed and tend to have a long heavy right tail, distributions do not satisfy the assumptions for ordinary least squares regression, which include normality, homoscedasticity, and independence of error terms. We used negative binomial regression models to analyze health care use outcomes (20-23). Incidence rate ratios (IRRs) from negative binomial regression models were used to compare incidence rates of ambulatory care use and prescription fills among three racial and ethnic groups. For example, if the IRR for ambulatory care visits among African Americans is 0.8, the interpretation is that being African American decreases the expected number of visits compared with whites by a factor of 0.8, holding other variables constant. In other words, being African American decreases the expected number of ambulatory care visits by 20% (20).

We performed a natural logarithmic transformation of the cost variables to compensate for the nonnormal distribution and high degree of right skewing. In the regression model using the log transformation for costs, we can interpret the coefficients (β) as percent changes for a change in a dummy variable from zero to one by calculating exponentiation of the β coefficient − 1 (exp [β] − 1) (24,25).

All analyses were adjusted for the complex survey design used in MEPS (17). The 2000 MEPS diabetes weights from the Diabetes Care Survey were used for the diabetes care analyses to yield valid estimates. The MEPS person-level weights were used for health care use and cost analyses to yield valid national estimates for individuals with diabetes (17). The University of Washington Human Subjects Committee determined this study to be exempt from human subjects review.

Table 2 presents demographic characteristics for adults with diabetes by racial and ethnic group. Among adults with diabetes, we found statistically significant differences across racial and ethnic groups in age, sex, education, marital status, income, insurance, and comorbidity. The mean age of adults with diabetes was 60 (61 among whites, 59 among African Americans, and 57 among Hispanics, P = .006). The proportion of women was highest among African Americans (66%), followed by whites (53%) and Hispanics (51%) (P = .04). African Americans had the lowest proportion of married individuals (36%), followed by Hispanics (53%) and whites (62%) (P < .001). Whites had the highest proportion of individuals with a high school education or more (77%), followed by African Americans (68%) and Hispanics (51%) (P < .001). More than half of African Americans (55%) were low income, compared with 45% of Hispanics and 32% of whites (P < .001). Most whites (97%) and African Americans (93%) had health insurance at least some of the time in 2000, but only 80% of Hispanics did (P < .001). Whites (72%) were significantly more likely to have private insurance than African Americans (50%) or Hispanics (40%) (P < .001), whereas African Americans (24%) and Hispanics (28%) were significantly more likely to be enrolled in Medicaid than whites (8%) (P < .001). The proportion of individuals having a USC was similar among the racial and ethnic groups (96% for whites, 94% for African Americans, and 90% for Hispanics, P = .12). Seventy-one percent of African Americans were concomitantly diagnosed with high blood pressure, compared with 64% of whites and 49% of Hispanics (P = .01). Whites had the highest proportion of concurrent heart disease with diabetes (35%), compared with 23% among African Americans and 14% among Hispanics (P < .001).

Table 3 presents unadjusted diabetes process-of-care measures by racial and ethnic group. Most of the diabetes care measures, including management, complications, and treatment, were not significantly different among racial and ethnic groups. However, dilated-eye examination rates were significantly different among the groups (95% among whites, 90% among African Americans, and 86% among Hispanics, P = .02). The probability of having eye problems caused by diabetes also differed significantly across the three groups (23% among whites, 33% among African Americans, and 33% among Hispanics, P = .03). 

Table 4 presents adjusted odds ratios (ORs) for diabetes care by racial and ethnic group. After adjusting for covariates, Hispanics were more likely than whites to have eye problems from diabetes complications (OR, 1.56; 95% CI, 1.03–2.56). We found no significant differences in other measures of diabetes management, complications, and treatment and among racial and ethnic groups after adjustment.

Table 5 summarizes unadjusted health care use and costs. The probability of using ambulatory care was significantly different among groups (P = .04), but we found no significant difference in the mean number of ambulatory care visits. Whites were more likely to use ambulatory care (96%), followed by African Americans (92%) and Hispanics (89%). We found no significant difference in the probability of having a prescription fill among the groups. However, whites filled more prescriptions (38.3 fills), followed by African Americans (29.8 fills) and Hispanics (29.8 fills) (P = .002).

We found significant differences in ambulatory care costs and prescription drug costs among the three groups. Whites had the highest ambulatory care costs ($1783), followed by African Americans ($1654) and Hispanics ($1028) (P = .03). The prescription drug costs among the three groups were $1886 for whites, $1419 for Hispanics, and $1392 for African Americans (P < .001). Although we found no significant difference in total health care costs among the racial and ethnic groups, the proportion by level of payment sources to total expenditure was different. Approximately one third of total health care costs was paid by private insurance among whites. Medicaid paid more health care costs for African Americans ($985) and Hispanics ($803) than whites ($374) (P = .01).

Table 6 presents adjusted health care use and costs. Compared with whites, African Americans were less likely to have ambulatory care use (IRR, 0.71; 95% CI, 0.58–0.87) and prescription fills (IRR, 0.73; 95% CI, 0.63–0.83). After adjustment, we found no significant differences between whites and Hispanics in the probability of having ambulatory care use and prescription fills.

Compared with whites, African Americans had 25% lower total health care costs (P = .008), 51% lower ambulatory care costs (P < .001), and 36% lower prescription drug costs (P < .001). Hispanics also had 31% lower total health care costs (P = .006) and 58% lower ambulatory care costs (P = .003) than whites.