Bayesian Predictive Approach to Early Termination for Enriched Enrollment Randomized Withdrawal Trials

Abstract

When assessing chronic pain, in certain settings the enriched enrollment randomized withdrawal (EERW) design may offer advantages over traditional trial designs in characterizing the treatment effect in a clinically relevant way. The EERW design by definition includes two distinct phases: an enriched enrollment phase during which subjects initially receive open-label treatment with the test drug, and a double-blind randomized withdrawal phase during which apparent responders are randomized to receive test drug or placebo. The response rate during the enriched enrollment phase provides useful information on the effectiveness of the test drug, and interim monitoring of the response rate during the enriched enrollment phase can help terminate the trial early when evidence accumulates to demonstrate that the treatment is ineffective. This article reviews the method of Bayesian predictive probability for observing a sufficient magnitude of response rate at the end of enriched enrollment phase given the observed data at an interim look. The method is applied to derive futility stopping rules, and the sensitivity of the futility stopping rules is examined based upon the choice of prior distributions. The operating characteristics of these stopping rules are compared to those based on observed response rate using simulated examples.