A Predictive Approach for Monitoring Multiple Outcomes in Phase II Clinical Trials

Most phase II trials are designed as one-arm studies based on a binary endpoint of interest, that represents treatment efficacy. Monitoring strategies can be adopted to perform interim evaluations before data collection has been completed. The essential aim is to ensure early termination of the trial if the experimental treatment is unlikely to provide the desired level of efficacy. Although phase II trials are mainly focused on evaluation of efficacy, many authors consider more ethical and informative to gather also information about toxicity during this phase. In this paper, we present Bayesian monitoring rules for single-arm phase II trials based on posterior predictive probabilities, that jointly consider both binary efficacy and toxicity endpoints. At any interim stage, given the current data and the prior distribution, the Dirichlet-Multinomial distribution provides the predictive probability of each possible combination of future efficacy and toxicity outcomes. It is exploited to compute the predictive probability that the trial will yield a positive outcome, if it continues to the planned end. Stopping rules based on this predictive probability are examined as the critical boundaries vary and under different scenarios.