Sample Size and Power Estimates for Cluster-Randomized Trials That Use a Stepped Wedge Design

• Activity Number: 132
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Korean International Statistical Society
• Author(s): David Thompson*
• Keywords: sample size and power ; stepped wedge design

Abstract

The stepped wedge design (Hussey and Hughes, 2007), an attractive alternative for cluster-randomized trials, is a type of crossover design in which participants in each cluster receive the intervention after a non-intervention period of varying length. Graphs that depict the phased intervention resemble a wedge with steps at each time of initiation. Woertman et al. (2013) offer an approach to sample size calculation that multiplies conventional estimates by a design effect or correction factor that considers cluster size, within-cluster correlation (ICC), number of steps, and the number of pre- and post-intervention observations. The present study demonstrates a direct calculation of power using an approach articulated by Stroup (1999) and expanded by Littell et al. (2006). Design elements (number of steps, number and size of clusters) are specified in an exemplar dataset, which is presented to a mixed model that calculates the non-centrality parameters needed to estimate power for specific alternate hypotheses. The demonstration uses SAS PROC MIXED for that purpose, after using SAS PROC IML to pre-specify complex within-cluster covariance structures.