Comparing the power of the discontinuation design to that of the classic randomized design on time-to-event endpoints

Abstract

The discontinuation design has been proposed as an alternative to the classic randomized design for evaluating the effect of an experimental agent on time-to-disease progression and survival duration. With this design, all enrolled patients are treated with an experimental agent for a fixed course of therapy. Those patients with progressive disease at or before the end of this fixed period are removed from trial while those with stable disease or better are randomized to continued treatment with the experimental agent or standard of care. Simulations presented in this paper demonstrate that for realistic situations, the loss in information on patients enrolled but not randomized in the discontinuation design is of sufficient magnitude that it is underpowered as compared to the classic design of randomizing all enrolled subjects.