Cluster-randomized trials

Abstract

Cluster-randomized trials represent an important experimental design, supplementing ordinary randomized clinical trials. They are particularly relevant when evaluating interventions at the level of clinic, hospital, district or region. They are necessary when it is not feasible to randomize individual patients, and desirable when there may be contamination between clusters. But they also carry serious design and analysis implications, and the use of clusters as the unit of randomization must be justified. Sample sizes will usually need to be greatly increased, an adequate number of clusters is essential, and the statistical analysis must allow for the cluster design. And one should rigorously guard against selection bias.