Multi-armed bandits in clinical trials🔗
Abstract🔗
Bayesian bandit problems have been described in the theoretical statistics literature since 1933. I will say how I got into the area as a graduate student and some of my contributions to the bandit theory. My motivation was clinical trials, even though I came to learn that this application was summarily rejected by all clinical trialists… at the time. I will give you a few-minute tour through several decades of my life where I worked within the conventional clinical trial establishment trying to change the paradigm. In the last decade, I have had more than a modicum of success. I will describe actual national and global clinical trials that I have designed and that are being conducted today. I call them Bayesian adaptive platform trials, which is a euphemism for multi-armed bandits. They are getting a surprising amount of support from regulators. For example, the Director of the Center for Drugs at the FDA says, "These have to be the future."