Sequential Elimination in Multi‐arm Multi‐stage Selection Trials

Abstract

The main objective of a phase II randomized multi‐arm selection trial is to identify the most promising treatment among multiple competing experimental regimens, when it truly exists. A multi‐arm multi‐stage (MAMS) design could be used to allow for preplanned adaptations based on the interim data. In particular, response‐adaptive randomization (AR) aims to steer patients away from the inferior treatments by updating the allocation probability throughout. Two AR approaches are the Bayesian AR and a frequentist alternative known as sequential elimination. The latter closes arms that are noticeably worse to the best arm at interim evaluations, and hence channel more patients to the more promising open treatment arms. A Bayesian futility monitoring rule based on comparison to the historical response rate of standard treatment could also be incorporated. The two AR approaches and Bayesian MAMS are compared against a single‐stage pick‐the‐winner selection design. There are clear advantages of using such adaptive selection designs compared to a single‐stage design, with the ethical attractiveness of allocating more patients to better performing arms. The improved performance is even more evident if there is a clear winner or if all arms are futile. It is particularly important to incorporate futility monitoring to improve the design's overall performance.