Power Analysis of Cutoff-Based Randomized Clinical Trials

Abstract

A cutoff-based randomized clinical trial couples cutoff-based assignment on an appropriate covariate with random assignment to help balance ethical and scientific concerns in certain situations. A statistical power algorithm based on the Fisher Z method is developed that is particular to and inclusive of cutoff-based random clinical trials and the single cutoff-point (regression-discontinuity) design, which has no randomization. This article quantifies power and sample size estimates for varying levels of randomization and cutoff-based assignment. Although more randomization engenders greater statistical power, less randomization requires a much larger increase in sample size for small treatment effects.