The primary hypothesis for this study was that canagliflozin 300 mg was noninferior to sitagliptin 100 mg in reducing A1C from baseline to week 52. The primary analysis was based on the modified intent-to-treat population (all randomized subjects who received one or more doses of study drug) with a last observation carried forward approach to impute missing data at the end point. Assuming no difference between canagliflozin and sitagliptin in A1C-lowering efficacy and a common SD of 1.0% with respect to change in A1C, it was estimated that 234 subjects per treatment group would provide ∼90% power to demonstrate the noninferiority of canagliflozin compared with sitagliptin. In addition, per-protocol analysis (subjects completing the 52-week study and without protocol deviations that could impact efficacy assessment) was conducted to further support the noninferiority assessment. To provide 90% power for the per-protocol analysis, assuming a discontinuation rate of 35% over 52 weeks, the sample size was increased to 360 subjects per treatment group.
Safety analyses and the primary efficacy analysis were conducted using the modified intent-to-treat population. The last observation carried forward approach was used for the primary analysis of efficacy data. All statistical tests were interpreted at a two-sided significance level of 5%, and all CIs were interpreted at a two-sided confidence level of 95%. Primary and continuous secondary end points were assessed using an ANCOVA model, including treatment and stratification factors as fixed effects and the corresponding baseline value as a covariate. The least squares (LS) mean differences and two-sided 95% CIs were estimated for the comparisons of canagliflozin versus sitagliptin. Noninferiority of canagliflozin to sitagliptin was assessed based on a prespecified margin of 0.3% for the upper limit of the two-sided 95% CI for the comparison in the primary last observation carried forward analysis. If noninferiority was demonstrated, then superiority was assessed, as determined by an upper bound of the 95% CI around the between-group difference (canagliflozin minus sitagliptin) of <0.0%. A prespecified hierarchical testing sequence was implemented to strongly control overall type I error attributable to multiplicity; P values are reported for prespecified comparisons only.
For subgroup analysis, descriptive statistics and 95% CIs for the change from baseline in A1C were provided for subgroups of subjects with baseline A1C of <8.0% (64 mmol/mol), ≥8.0% (64 mmol/mol) to <9.0% (75 mmol/mol), and ≥9.0% (75 mmol/mol). For indices of βCF, descriptive statistics and 95% CIs for the changes from baseline were provided; comparisons of canagliflozin with sitagliptin for changes from baseline at week 52 were assessed using an ANCOVA model with treatment and stratification factors as fixed effects and the corresponding baseline value as a covariate.
Safety analyses and the primary efficacy analysis were conducted using the modified intent-to-treat population. The last observation carried forward approach was used for the primary analysis of efficacy data. All statistical tests were interpreted at a two-sided significance level of 5%, and all CIs were interpreted at a two-sided confidence level of 95%. Primary and continuous secondary end points were assessed using an ANCOVA model, including treatment and stratification factors as fixed effects and the corresponding baseline value as a covariate. The least squares (LS) mean differences and two-sided 95% CIs were estimated for the comparisons of canagliflozin versus sitagliptin. Noninferiority of canagliflozin to sitagliptin was assessed based on a prespecified margin of 0.3% for the upper limit of the two-sided 95% CI for the comparison in the primary last observation carried forward analysis. If noninferiority was demonstrated, then superiority was assessed, as determined by an upper bound of the 95% CI around the between-group difference (canagliflozin minus sitagliptin) of <0.0%. A prespecified hierarchical testing sequence was implemented to strongly control overall type I error attributable to multiplicity; P values are reported for prespecified comparisons only.
For subgroup analysis, descriptive statistics and 95% CIs for the change from baseline in A1C were provided for subgroups of subjects with baseline A1C of <8.0% (64 mmol/mol), ≥8.0% (64 mmol/mol) to <9.0% (75 mmol/mol), and ≥9.0% (75 mmol/mol). For indices of βCF, descriptive statistics and 95% CIs for the changes from baseline were provided; comparisons of canagliflozin with sitagliptin for changes from baseline at week 52 were assessed using an ANCOVA model with treatment and stratification factors as fixed effects and the corresponding baseline value as a covariate.
