|
Abstract:
|
Phase II cancer trials are traditionally designed as single-arm, with the null hypothesis chosen ahead of time, presumably from the literature. This approach, often criticized for its lack of rigorous control, endures in practice, partly because a concurrent control arm will substantially increase the sample size. An unbalanced randomization with more patients randomized to the treatment arm is advocated as a way to keep the sample size in check, but when the resulting trial is analyzed using a traditional statistical test, this planned imbalance actually reduces power. We propose a hybrid-approach where an imbalanced randomization scheme is implemented but the data for the control arm are augmented with historical information using a Bayesian approach. This has the advantage of higher number patients assigned to the experimental arm, desirable for a Phase II program, and has the potential to better control Type I and Type II errors. We will review the use of power priors and related methods to evaluate the gains from this procedure. While similar literature focuses on power improvements, we will also show implications for Type I error control.
|
