A comparison of multiple testing procedures for the gold standard non-inferiority trial.

For the three-arm parallel-group design, several procedures have been proposed in the literature to control for the multiple type I error when all groups are to be statistically compared. Mere statements regarding the rejection of the null hypotheses are not satisfactory, but instead confidence intervals are more desirable. For this purpose, the procedure by Koch and Röhmel (2004) is modified and the theoretical reasoning behind it is given. However, one combination of effects occurs where control of the multiple type I error cannot be guaranteed. This pathologic case is discussed theoretically and further investigated in a simulation study.

RCT Entities:   Population Interventions Outcomes