Using Confidence Distributions in Final and Interim Analyses for Single-Arm Studies or Platform Trials Consisting of Single-Arm Studies.

Abstract

Confidence distributions are a frequentist alternative to the Bayesian posterior distribution. These confidence distributions have received more attention in the recent past because of their simplicity. In rare diseases, oncology, or in pediatric drug development, single-arm trials, or platform trials consisting of a series of single-arm trials are increasingly being used, both to establish proof-of-concept and to provide pivotal evidence for a marketing application. Often, these single-arm trials are designed as two-stage designs, or they include sequential or continuous monitoring approaches. They are analyzed using standard frequentist, Bayesian, or other methods. In this paper, we describe how to define analysis strategies based on confidence distributions for such single-arm trials or for platform trials that consist of a series of single arm trials. We focus on binary endpoints and show how to define the corresponding decision rules for final and interim analyses and how to derive their operating characteristics exactly, for example, without simulation. Our approach uses predictive probabilities rather than conditional probabilities (as with stochastic curtailment) to define the interim decision rules. It can be applied to platform, basket, and umbrella trials that consist of a series of single-arm trials but also to stand-alone single arm trials.