Monthly mean HbA1c levels across the study groups. The data are represented as fitted means derived from a linear mixed model that included the covariables age, sex, race/ethnicity, diabetes duration, and baseline HbA1c level. The error bars represent SEs. The group effect was significant for comparison of trends in monthly mean HbA1c levels between control vs. responder (P = 0.014), as well as control vs. nonresponder groups (P = 0.0026), but not for responder vs. nonresponder (P = 0.254).