Internal Pilots for Cluster-Randomized Trials

• Activity Number: 253
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Ashutosh Ranjan*+ and Leslie A McClure and Christopher S Coffey
• Companies: The University of Alabama at Birmingham and The University of Alabama at Birmingham and University of Iowa
• Keywords: cluster-randomized trials ; internal pilot ; sample size re-estimation ; type I error ; weighted linear model

Abstract

Although internal pilots for randomized clinical trials have existed for some time now and are well accepted in practice, there is a need for methods for sample size re-estimation in cluster-randomized trials where a social unit, rather than an individual, is randomized to treatment conditions. Sample size re-estimation in terms of adjusting the number of clusters (e.g., families) midway through the trial will give the investigators a better chance to realize the aim of trial with sufficient power while preserving the type I error. Currently, methods exist for a particular class of linear mixed models with no missing or mistimed data and a compound symmetric covariance structure, along with Gaussian errors. We generalize this concept by converting the data from a cluster trial into a weighted univariate linear model and developing exact internal pilot theory for weighted model. We discuss the impact of this method on type I error rates across a variety of planned sample sizes.