Keywords: Second Generation p-value, Adaptive Monitoring Designs, Bayesian Adaptive Designs
The Food and Drug Administration is committed to “facilitate the advancement and use of complex adaptive, Bayesian, and other novel clinical trial designs” (Prescription Drug User Fee Act VI). We introduce a novel design based on the Second-generation p-value (SGPV; Blume, 2018), which indicates when the data are compatible with the alternative hypothesis, the null hypothesis, or when inconclusive. This requires that the minimum clinically meaningful effect size is specified upfront. Our design uses this information to reduce the false discovery rate by ignoring statistically significant results for clinically meaningless effect sizes. SGPVs are easy to implement and outperform traditional approaches based on adjusting the p-value for multiple comparisons or looks. Our novel design permits sequential monitoring and interim examinations of multiple endpoints. The trial halts when the data support either convincingly superior or uninteresting clinical results and, to reduce bias, when is affirmed by a subsequent validation monitoring examination. In extensive simulations and the currently active clinical trial REACH, we compare our method’s performance in terms of error rates, false discovery rates, bias, and average stopping times to interval null Bayesian Adaptive designs (Kruschke, 2013) and provide recommendations on implementing the validation monitoring step required for stopping.