A confidence interval approach for difference and ratio of normal means in self-designing clinical trials

Abstract

In self-designing clinical trials, confidence intervals are derived for the difference and the ratio of normal means, where the results of the independent study stages are combined using the weighted inverse normal method. The confidence intervals always hold the predefined nominal confidence level. During the course of the Self-designing trial, the sample sizes as well as the number of study stages can be determined simultaneously in a completely adaptive way. Self-designing may be considered as the limit case of adaptive group sequential designing of O'Brien and Fleming type when the full significance level is shifted to the last stage. We consider the effect measures difference and ratio of normal means, where the latter has not yet been considered in group sequential trials so far. Concrete rules are derived for updating sample sizes and assigning weights to the stages of the trial. The clinical trial may be originally designed either to show non-inferiority or superiority. But, in each interim analysis, it is possible to change the planning from showing non-inferiority to showing superiority or vice versa. The performance of the Self-designing and the resulting confidence intervals are demonstrated in real-data examples for both considered effect measures showing both kinds of switching during an ongoing trial.