Medical Expenditure Panel Survey (MEPS)
MEPS is a longitudinal survey that provides nationally representative estimates of healthcare use and expenditures for the U.S. civilian non-institutionalized population. The 2000 MEPS consisted of 12,280 households, comprised of 23,839 individuals, from the 1998–1999 National Health Interview Survey. For purposes of this analysis, year 2000 data were collapsed for a cross-sectional sample of the population. After excluding individuals less than 20 years of age, over 14,900 adults were included in these analyses.
Variables
Patient sex served as the main independent variable. Men and women were stratified into five age groups. For adjustment purposes, other independent variables were selected based on the Andersen model and prior studies demonstrating influence on receipt of preventive services [5, 6, 8, 9, 14, 17, 23]. Selected covariates were: race/ethnicity, education, marital status, income, insurance, usual source of care, perceived health status, and region of country. Number of ambulatory care visits was operationalized as whether or not individuals had one or more outpatient or office-based visits in the previous year.
From the MEPS data file, we selected items representing five preventive services applicable to both men and women: whether a doctor has checked respondent's blood pressure within last two years, duration since respondent's last cholesterol measurement, whether a doctor has advised respondent (if a smoker) to quit smoking within the past 12 months, duration since last blood stool test home kit, duration since last sigmoidoscopy or colonoscopy, and duration since last routine checkup. Answers to these items were by self-report.
For our blood pressure screening variable, we excluded respondents who reported that they had been previously diagnosed with hypertension. Those remaining who reported that they had a blood pressure measurement within the past two years were considered to have received this preventive service. For our cholesterol screening variable, we excluded respondents already diagnosed with hyperlipidemia and considered those who reported a cholesterol measurement within the last five years to have received this preventive service. To create a colorectal cancer screening variable, we considered those over age 50 who reported a blood stool test home kit within the past year, or a sigmoidoscopy or colonoscopy within the past five years to have received appropriate screening [24]. We considered those who reported a checkup within the past two years to have received this service. These intervals are also commonly used in other studies of receipt of preventive services [6, 11, 14, 17].
Statistical Analysis
The statistical software STATA 8.0 (STATA Corporation, College Station, TX) was used for all analyses with incorporation of appropriate sampling weights, primary sampling units, and strata to account for the complex survey design. Weighted percentages of men and women within each age group as well as within categories of selected covariates were determined. For bivariate analyses, weighted percentages were determined by 2 × 2 tables and tested for significance using chi-square comparing sex to each outcome, and then were stratified by age category. Unadjusted odds ratios and 95% confidence intervals were determined from logistic regression models without covariates.
Multivariable analyses were conducted using separate logistic regression models for each outcome, with sex as the main independent variable adjusted for covariates. The five age categories were modeled using four dummy variables with the oldest age category left out as the reference category. To evaluate for interaction between sex and age (i.e., stratification by age), interaction terms between sex and each of the four age category dummy variables were added to the models. A four degree-of-freedom Wald test was used to test for the interaction effect. A significant test means that the odds ratios comparing men to women differed by age group. An exponentiation of the linear combination of the beta estimates was used to estimate adjusted odds ratios for the main exposure (sex) and each of the four interaction terms (sex by age dummy variables). The variance-covariance matrix was used to estimate 95% confidence intervals around each odds ratio [25, 26]. Finally, to test whether interaction between sex and age on receipt of preventive services was accounted for by group differences in outpatient care-seeking, the Wald test was re-run after the ambulatory care visit variable was added to the models.