A Conservative Approach to Leveraging External Evidence for Effective Clinical Trial Design

Mainstream methods for clinical trial design do not yet use prior probabilities of clinical hypotheses, mainly due to a concern that poor priors may lead to weak designs. To address this concern, we illustrate a conservative approach to trial design ensuring that the frequentist operational characteristics of the primary trial outcome are stronger than the design prior. Compared to current approaches to Bayesian design, we focus on defining a sample size cost commensurate to the prior to ensure against the possibility of prior-data conflict. Our approach is ethical, in that it calls for quantification of the level of clinical equipoise at design stage and requires the design to be appropriate to disturb this initial equipoise by a pre-specified amount. Four examples are discussed, illustrating the design of phase II-III trials with binary or time to event endpoints. Sample sizes are shown to be conductive to strong levels of overall evidence, whether positive or negative, increasing the conclusiveness of the design and associated trial outcome. Levels of negative evidence provided by standard group sequential designs are found negligible, underscoring the importance of complementing traditional efficacy boundaries with futility rules.