All analyses were conducted by using SAS software (SAS Institute, Inc., Cary, North Carolina) for data management and SUDAAN (Research Triangle Institute, Research Triangle Park, North Carolina) to account for the complex sampling scheme, the unequal probability of selection, the oversampling of certain demographic groups, and nonresponse factors (32–33). Level of care was estimated by the percentage, with corresponding 95% CIs, of respondents who reported each preventive care practice. We estimated prevalence by groups of sex, age, ethnicity, education, health insurance, type of treatment, and duration of diabetes. Multiple logistic regression and computation of predictive margins were used to estimate the probability of receiving DQIP indicators when we controlled for all other independent variables. Predictive margins are a type of direct standardization in which the predicted values from the logistic regression models are averaged over the covariate distribution in the population (34). Predictive margins and their standard errors from logistic regression models are provided by the current version of SUDAAN (33). Taylor linearization is used in SUDAAN for calculating these standard errors. Odds ratios are usually used to display the result from logistic regression models. However, predictive margins are easier to interpret than odds ratios, and they do not require designating one of the groups as the referent group.