Interim Monitoring of Cluster Randomized Trials: Benefits and Challenges

• Activity Number: 447
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Abigail Shoben*+ and Siobhan Brown
• Companies: Ohio State University and University of Washington
• Keywords: cluster randomized trials ; group sequential methods ; interim analysis ; ICC

Abstract

Cluster randomized trials (CRTs) are becoming increasingly common in medical research. Examples include randomizing emergency response teams to test a new device and randomizing schools to evaluate exercise promotion strategies. In a CRT, individual outcomes are correlated within the same cluster, and statistical analyses must account for this correlation. In many clinical trials ethical and financial reasons motivate the use of group sequential methodology to allow the trial to be analyzed prior to study completion, but implementing such interim analyses in CRTs requires additional considerations. In particular, standard methodology relies on estimating the fractional amount of statistical information available at an interim analysis. In a CRT, this information growth depends on the intracluster correlation coefficient (ICC), which may not be known or even well-estimated at an interim analysis. We illustrate the dependence of the information growth on the ICC and show the inflation of the type I error rate and dramatic loss of power that can occur when using a poor estimate of the information growth. We conclude by providing recommendations for future studies.