The primary and continuous key secondary
and exploratory end points were
evaluated using ANCOVA, with treatment
as the fixed effect and baseline value as the
covariate, to derive a least squares estimate
of the treatment difference in mean change
with corresponding two-sided 95% CI.
Proportions were analyzed using logistic
regression with adjustment for baseline
values as described by Zhang et al. (15).
Statistical noninferiority of dapagliflozin
versus glipizide was established if the upper
limit of the 95% CI for the treatment
difference in mean HbA1c change from
baseline to week 52 was ,0.35% (noninferiority
margin). For graphic presentation
of HbA1c and TBW over the 52-week
treatment period, the change from baseline
(last observation carried forward [LOCF])
was analyzed at each interval using
ANCOVA with treatment as the fixed effect
and baseline value as the covariate.
The Kaplan-Meier method was used to analyze
time to onset of patient discontinuation
because of poor glycemic control.