We conducted statistical analyses by using SAS for Windows software, version 7.0 (SAS Institute, Inc., Cary, North Carolina), for data management. We used SUDAAN software (Research Triangle Institute, Research Triangle Park, North Carolina) to obtain point estimates and SEs based on sampling weights to produce national estimates accounting for the complex survey design. We used Taylor series linearization for variance estimation. We computed the percentage of respondents who reported receipt of each measure. We examined the diabetes care measures by age, sex, race or ethnicity, education, insulin use, and health insurance status because our previous analysis had variations by these factors (18). However, insulin users were not asked to fast; hence, we did not examine LDL levels by insulin use. We used multiple logistic regression and predictive margins to estimate the probability of receiving or meeting the care measure after controlling for all known potential confounders. Predictive margins are a type of direct standardization, where the predicted values from the logistic regression models are averaged over the covariate distribution in the population (30). This statistic has several advantages over the odds ratio: It is not influenced if the outcome is not rare; a comparison group is not required; and it provides a measure of absolute difference rather than relative difference. We included an interaction term between time and each measure in the models to allow estimation of the probability for each period. To assess the difference in the percentage change between the 2 comparison groups, we tested the interaction term of each demographic characteristic and clinical variable (age, sex, race or ethnicity, education, insulin use, and health insurance status) with time.