Sample size calculations for stepped wedge trials using design effects are only approximate in some circumstances

Abstract

Estimation of sample size and power for stepped wedge cluster randomised trials can be determined by one of a number of related methods. These include exact analytical approaches, design effects or simulation. A recent paper compared the design effect to the analytical method. There were some differences between the two approaches. We show here that these differences occur because the design effect approach is only technically correct when there is an equal number of clusters crossing over at each step. Findings: The design effect for the stepped wedge cluster randomised trial is only appropriate when there is an equal number of clusters switching at each step.