Use of Bayesian Predictive Power in drug development: What to watch out for


In this talk I will discuss two aspects relevant to compute and track Bayesian predictive power (BPP) for a clinical trial: First, I will illustrate how the choice of prior impacts the distribution of power values and that summarizing this distribution by its mean – which is what we typically do – might not be ideal and misleading. Secondly, BPP can be updated after performing an interim analysis of the trial of interest, be it blinded or unblinded to the effect size. In general, BBP will decrease if we do not stop a trial at an efficacy interim analysis and will increase if we do not stop a trial at a futility interim analysis.

Toronto, CA

Paper on priors

Paper on update after interim