Predictive versus posterior probabilities for phase II trial monitoring

Bayesian monitoring of clinical trials is typically based on posterior or predictive probabilities. In the first case, the decision rules are based on the posterior probability that the experimental treatment shows the required performance, given the interim data. In the second case, the idea is to evaluate the predictive probability of observing a positive result if the trial were to continue to its pre-specified maximum sample size. In this paper, we compare the two strategies when applied to a single-arm phase II trial based on binary efficacy and toxicity endpoints.