Investigating a sequence of randomized phase II trials to discover promising treatments

We consider clinical trial strategies to study diseases in which there is rapidly developing technology. We assume the availability of a limited number of patients for screening treatments over a time horizon, and that availability of new treatments for test is staggered over time. We assume further that patient response is binary and rapidly observable. We consider the strategy of conducting a sequence of two-armed randomized clinical trials. We carry over the treatment with the larger number of observed successes on the current trial to the next trial for comparison with a new treatment, with this process repeated at each step. For a fixed total number of patients (N), the number of trials one may conduct in sequence (k) is inversely related to the sample size per trial (2n), N = 2nk. We investigate how k and n influence (a) the expected success probability for the treatment selected at the end, and (b) the expected number of total successes for the N patients. The ultimate objective is to select one treatment, the winner at stage k, to test against a standard regimen in a randomized comparative phase III trial.