A Simple Method for Predicting a Binomial Response Rate in a Bayesian Interim Analysis of a Single-Arm Trial When Observing Responses and Failures Requires Prolonged Follow-Up

• Activity Number: 127
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Gary Hantsbarger*+
• Companies: Astellas
• Keywords: single-arm study ; Bayesian futility analyses ; Dirichlet distribution

Abstract

Initial investigation of the efficacy of a drug may involve a single-arm study in which the response rate of subjects receiving the drug is compared to a pre-defined minimum rate, below which the drug is deemed not worth further study. During recruitment, interim futility analyses may be used to determine whether the probability of a successful trial is high enough to justify continuing the study. When response or failure is quickly determined, Bayesian interim analyses for such studies typically use a beta distribution as a prior for the response rate. However, if observing a response requires prolonged follow-up, this model is not adequate. A Bayesian interim analysis method which takes into account both time to failure and time to response, while respecting the binomial nature of the final analysis, will be described. The method is based on decomposing a beta prior for a binomial response/failure probability into a Dirichlet distribution describing probabilities of failure and response among multiple time intervals.