The primary hypothesis was that canagliflozin 300 mg is statistically superior to placebo in reducing HbA1c from baseline to week 26. Key secondary hypotheses were statistical superiority of canagliflozin 100 mg to placebo in HbA1c- lowering effect at week 26 and non-inferiority of canagliflozin 300 mg or both canagliflozin doses to sitagliptin 100 mg in reducing HbA1c from baseline to week 52. Primary efficacy analysis was performed in the modified intent-to-treat (mITT) population (randomised participants who received ≥1 dose of study drug) using a last observation carried forward (LOCF) approach. Assuming a group difference of 0.5% (5.5 mmol/mol) between canagliflozin and placebo and a common SD of 1.0% (10.9 mmol/mol) for change in HbA1c, and using a two-sample, two-sided t test with a type I error rate of 0.05, an estimated 86 participants per group were required to achieve 90% power to demonstrate statistical superiority of canagliflozin to placebo. To support superiority and non-inferiority objectives for the primary endpoint in the mITT population and for supportive analysis in the per-protocol population (mITT participants who completed the study, did not receive rescue therapy and had no major protocol violations), an estimated 360 randomised participants were needed for each active treatment group and 180 for the placebo group, assuming a 35% discontinuation rate at week 52 and with a 2:2:2:1 randomisation ratio for canagliflozin 100 mg, canagliflozin 300 mg, sitagliptin 100 mg and placebo.
Primary efficacy analyses were performed in the mITT population according to randomised treatment assignment using LOCF to impute missing data; for participants who received rescue therapy, the last post-baseline value before rescue was used. Safety analyses were performed in the same population according to the predominant treatment received; in this study, the mITT and safety populations were identical. Only data from participants randomised to sitagliptin 100 mg on day 1 (i.e. not including participants who switched from placebo to sitagliptin at week 26) were included in efficacy comparisons at week 52. Safety analyses over 52 weeks included participants who received canagliflozin 100 mg or 300 mg or sitagliptin and those who switched from placebo to sitagliptin after 26 weeks (placebo/sitagliptin group).
An analysis of covariance (ANCOVA) model with treatment and stratification factor as fixed effects and corresponding baseline value as a covariate was used to assess primary and continuous secondary endpoints. Least squares (LS) mean differences between groups and two-sided 95% CIs were estimated. The categorical secondary endpoint was analysed with a logistic model with treatment and stratification factor as fixed effects and baseline HbA1c as a covariate. Assessment of non-inferiority of canagliflozin to sitagliptin was based on a pre-specified margin of 0.3% for the upper limit of the two-sided 95% CI for the comparison. If non-inferiority was demonstrated, then superiority was assessed based on an upper bound of the 95% CI around the between-group differences of <0.0%.
Comparisons were performed for canagliflozin vs placebo at week 26 and vs sitagliptin at week 52 based on pre-specified hierarchical testing sequences implemented to strongly control overall type I error due to multiplicity. At week 26, statistical tests were interpreted at a two-sided significance level of 5% for all endpoints except change in systolic BP, HDL-cholesterol and triacylglycerol. These were grouped together into two separate families (one each for canagliflozin 100 mg and 300 mg) and each family was tested using the Hochberg procedure at the 2.5% significance level. Comparisons of canagliflozin with sitagliptin at week 52 were initiated after statistical superiority of canagliflozin 100 mg and 300 mg to placebo in HbA1c lowering at week 26 was established; statistical tests at week 52 were interpreted at a two-sided significance level of 5% for all endpoints. The p values are reported for pre-specified comparisons only.
Primary efficacy analyses were performed in the mITT population according to randomised treatment assignment using LOCF to impute missing data; for participants who received rescue therapy, the last post-baseline value before rescue was used. Safety analyses were performed in the same population according to the predominant treatment received; in this study, the mITT and safety populations were identical. Only data from participants randomised to sitagliptin 100 mg on day 1 (i.e. not including participants who switched from placebo to sitagliptin at week 26) were included in efficacy comparisons at week 52. Safety analyses over 52 weeks included participants who received canagliflozin 100 mg or 300 mg or sitagliptin and those who switched from placebo to sitagliptin after 26 weeks (placebo/sitagliptin group).
An analysis of covariance (ANCOVA) model with treatment and stratification factor as fixed effects and corresponding baseline value as a covariate was used to assess primary and continuous secondary endpoints. Least squares (LS) mean differences between groups and two-sided 95% CIs were estimated. The categorical secondary endpoint was analysed with a logistic model with treatment and stratification factor as fixed effects and baseline HbA1c as a covariate. Assessment of non-inferiority of canagliflozin to sitagliptin was based on a pre-specified margin of 0.3% for the upper limit of the two-sided 95% CI for the comparison. If non-inferiority was demonstrated, then superiority was assessed based on an upper bound of the 95% CI around the between-group differences of <0.0%.
Comparisons were performed for canagliflozin vs placebo at week 26 and vs sitagliptin at week 52 based on pre-specified hierarchical testing sequences implemented to strongly control overall type I error due to multiplicity. At week 26, statistical tests were interpreted at a two-sided significance level of 5% for all endpoints except change in systolic BP, HDL-cholesterol and triacylglycerol. These were grouped together into two separate families (one each for canagliflozin 100 mg and 300 mg) and each family was tested using the Hochberg procedure at the 2.5% significance level. Comparisons of canagliflozin with sitagliptin at week 52 were initiated after statistical superiority of canagliflozin 100 mg and 300 mg to placebo in HbA1c lowering at week 26 was established; statistical tests at week 52 were interpreted at a two-sided significance level of 5% for all endpoints. The p values are reported for pre-specified comparisons only.
