A time-to-event three-outcome design for randomized phase II cancer trials

Tumor response, a binary variable, has historically been the main measure of antitumor activity for many cancer phase II single-arm trials. Simon two-stage designs are often used. Sargent et al. proposed a three-outcome trial design in this setting which requires smaller sample sizes. For many new, molecularly targeted therapies, however, tumor response may not be the most reliable endpoint for measuring anti-tumor activity. Increasingly, time-to-event endpoints, such as progression-free survival (PFS), are used in the phase II setting. When such endpoints are the primary measure of efficacy, a randomized concurrently controlled study design is usually required. Given limited resources for phase II, studies are often underpowered with relatively large type I and II error rates, and it is sometimes unavoidable to have a "gray" decision zone after phase II where a clear decision regarding further development actions cannot be made without additional information. Compared with an underpowered standard two-outcome study, a three-outcome design prompts clinical trialists to contemplate the likelihood of landing in the "gray" zone at the trial design stage and choose study design parameters more appropriately. We propose a three-outcome design, with or without interim analyses, for randomized comparative phase II trials when a time-to-event endpoint is used.