A comparative phase II clinical trials procedure for choosing the best of three treatments

In some clinical trials one can employ adaptive designs advantageously, although in practice such techniques are rarely used, in part due to their inherent complexity. A simple and practicable decision-theoretic approach for the case of three treatments with binary responses is considered, using equal allocation to remaining treatments and, once eliminated, a treatment cannot be re-employed. Having specified the overall number of patients treated within and beyond the comparative stages of the trial, the goal is to maximize the expected total number of those successfully treated. Investigation of the method involves a computer program that can handle arbitrarily large numbers of patients. It is shown empirically that the decision procedure behaves only marginally worse than if the truly superior treatment had been known and had been given to all patients. Implementation of the method uses a minimax approach that removes dependence on prior parameters. Primarily an identification procedure, one advantage of this approach over traditional hypothesis testing methods is the potential to detect small improvements in treatment efficacy. The intended application is to assist in treatment selection during phase II trials, especially with rapid responses and when the disease involved is serious enough that design-motivating ethical considerations become paramount.