Predictive checking for Bayesian interim analyses in clinical trials

Bayesian methodologies have been used for interim analyses of clinical trial data. In Bayesian interim analyses, decisions regarding the continuation of a trial are guided by a Bayesian model or indices, e.g., the predictive probability derived from it that specifies the conditions under which the clinical trial results might be judged sufficiently convincing to allow early stopping. Thus, its appropriateness for making such decisions depends on whether the model or the indices are reliable. In this paper we describe the use of both prior- and posterior- predictive checking approaches as a diagnostic tool for assessing the reliability of the model or indices on which the decision making is based. The proposed approach is illustrated with three examples, one of which is a simulation.