Proposal for fully sequential multiarm trials with correlated arms

Abstract

We focus on the design of multiarm multistage (MAMS) clinical trials, using ideas from simulation optimization, biostatistics, and health economics. From a trial design perspective, we build on the trend of comparing multiple treatments with a single control by allowing for more than two arms in a trial, and we allow for arbitrarily many stages of sampling by using a diffusion approximation that allows for adaptive stopping rules. From a simulation perspective, our techniques extend the correlated knowledge-gradient concept, which has been used in one-stage lookahead (knowledge gradient) procedures, to Bayesian fully sequential selection procedures.