Behavioral Carry-Over Effects and Power Considerations in Crossover Trials

Abstract

The carry-over effect remains an outstanding concern in crossover trials when washout periods cannot sufficiently diminish their impacts. This can occur in comparative effectiveness research where carry-over effects are behavioral rather than biological. We first investigate crossover trials with and without carry-over effects under the potential outcome framework. We find that when carry-over effects exist and satisfy some sign condition, the classic estimator underestimates the treatment effect, which does not inflate the type I error of one-sided tests but decreases the power and leads to a power trade-off between crossover and parallel trials. We derive the condition under which crossover trials do not have type I error inflation and are more powerful than parallel trials. We further develop covariate adjustment methods for crossover trials and illustrate their performance using data from a real analgesic study and a hypothetical HIV prevention trial. We also compare the total investigation times and the total trial times of crossover and parallel trials.