Cluster-crossover design: a method for limiting clusters level effect in community-intervention studies

Abstract

The cluster-crossover design can be used for clinical trials comparing two or more interventions in a naturally formed study population, i.e. a cluster. This design differs from that of a crossover study, in that the treatment sequence is allocated at the cluster level. A cluster-crossover study can thus be considered as a cluster-randomized controlled study with additional periodic cluster-randomization(s) or treatment permutation(s) during the study. The data must be analyzed with hierarchical models with random effects in order to allow for different outcome probabilities in each period, cluster and cluster-period. Original data from two published field studies of hospital infection control based on this design are used here to illustrate the impact of different statistical models on the interpretation of the results.