Predictive probability of success in clinical drug development

Abstract

Predictive probability of success is a (subjective) Bayesian evaluation of the probability of a future successful event in a given state of information. In the context of pharmaceutical clinical drug development, successful events relate to the accrual of positive evidence on the therapy which is being developed, like demonstration of superior efficacy or ascertainment of safety. Positive evidence will usually be obtained via standard frequentist tools, according to the regulations imposed in the world of pharmaceutical development. Within a single trial, predictive probability of success can be identified with expected power, i.e. the evaluation of the success probability of the trial. Success means, for example, obtaining a significant result of a standard superiority test. Across trials, predictive probability of success can be the probability of a successful completio n of an entire part of clinical development, for example a successful phase III development in the presence of phase II data. Calculations of predictive probability of success in the presence of normal data with known variance will be illustrated, both for within-trial and across-trial predictions.