Design and Analysis of Cluster Randomization Trials

Abstract

In the great majority of clinical trials, randomization is performed at the level of the individual subject. In certain trials, however, the appropriate unit of randomization is some aggregate of individuals. This form of randomization is known as cluster randomization or group randomization. Cluster randomization is employed by necessity in trials in which the intervention is by nature designed to be applied at the cluster level. Examples of this type of trial include trials of community-based interventions, such as the Community Health Trial for Smoking Cessation (COMMIT) (Gail et al., 1992; COMMIT Research Group, 1995), and trials of schoolbased interventions, such as the Child and Adolescent Trial for Cardiovascular Health (CATCH) (Zucker et al., 1995; Luepker et al., 1996). Cluster randomization is also sometimes employed for convenience of trial administration in trials in which the intervention is applied at the level of the individual. An example of this situation is the trial design in which clinics are assigned to the various experimental arms, with all patients in a given clinic receiving the treatment towhich the clinic was assigned (Simon, 1981). The cluster randomization design poses special issues in the areas of design, logistics, sample size calculation, and statistical analysis. This point