A Comparison of Sample Size Calculations for Cluster Randomized Crossover Trials with a Binary Outcome

Abstract

A simplified sample size calculation for a cluster randomized crossover trial with a binary survival outcome (the “ T-BOSS ” method) was compared to the closed form equation and simulation methods currently available from Connolly, et. al. (2013, Canadian Journal of Cardiology, 29, p.652) and Reich, et. al. (2012, PLoS ONE, 7(4), p.E35564), respectively. When there was no period effect present, the T-BOSS method was consistently conservative with a sample size that is 11.4%, 14.8%, and 37.8% larger than the closed form sample size for intracluster correlations of 0.01, 0.04, and 0.20, respectively. When there was a period effect present, T-BOSS was nearly always anti-conservative unless the period effect was small or the average cluster size was very small (less than 20 patients at most for the settings we explored). The simulations were fairly close to the closed form equation, though showed more variability.