Performance of randomization-based causal methods with and without integrating external data sources for adjusting overall survival in case of extensive treatment switches in placebo-controlled randomized oncology phase 3 trials

Abstract

ABSTRACT In recent placebo-controlled randomized phase 3 oncology trials, evaluation of overall survival with frequent crossover is crucial for regulatory and pricing decisions. The problem is that an intention-to-treat based analysis causes a substantial loss of power to detect causal survival effect without crossover, and performance of existing methods is not satisfactory. In this article, our aims were to evaluate properties of the existing and a proposed Bayesian power prior method where data from an external trial is available. Simulation results suggested that proposed method was the most powerful under typical scenarios where patients with better prognosis are likely to crossover.