Cluster versus individual randomization in adolescent tobacco and alcohol studies: illustrations for design decisions

The decision to randomize by clusters of subjects such as a classroom or clinic versus individual randomization where some contamination may occur is examined within the framework of sample size issues. Estimates for background rates and intraclass correlations are also provided for adolescent tobacco and alcohol outcomes derived from a recent study using cluster randomization.

A ratio of adjusted sample sizes is derived which is a function of the intraclass correlation and cluster size for cluster randomization and total amount of contamination for individual randomization. Using estimated incidence rates and intraclass correlations, we provide a comparison of sample sizes for two plausible study outcomes.

Small clusters such as a family or small classroom tend to have stronger within cluster dependence and cluster randomization would be clearly favoured over individual randomization. For moderately sized clusters, if contamination levels are likely to be high then cluster randomization would be a better choice. However in some situations where lower levels of contamination are expected, individual randomization may be preferred. With larger clusters, individual randomization should be considered when contamination rates are expected to be low.

Investigators must carefully consider the choice of cluster randomization versus individual randomization in the context of likely contamination. In this paper we provided a basis for making this decision as well as examples to illustrate these decisions, and parameter estimates that will be especially useful for investigators in adolescent tobacco and alcohol studies.