Modeling impure clusters in a cluster randomized controlled trial

Cluster randomized controlled trials (CRCT) can be susceptible to a wide range of methodological problems. Many of these problems are not commonly recognized by researchers. This article is focused on one potential problem, the issue of impure clustering (multiple patient membership) within the CRCT structures and how it can lead to possible misunderstanding and bias in the results of the trial. A solution to this problem is presented using a multiple membership random effects model (MMREM). A simulated example of a three-level CRCT is presented and modeled with and without multiple patient membership data. Results indicate underestimation of higher level variances, and overestimation of lower-level variances, while also indicating underestimation of level predictors where the multiple membership occurs.