Intracluster correlation adjustments to maintain power in cluster trials for binary outcomes.

Adequately powered sample size calculations for cluster randomized trials primarily depend on the event rate variability, effect size, average cluster size, and intracluster correlation (ICC). Furthermore, an ICC estimate depends on event rate variability among clusters, cluster size, and number of clusters. We evaluated the impact on ICC estimates of event rates, event rate variations, cluster size, and cluster size variations for different numbers of clusters. We also evaluated how the event rate changes at the end of the trial affect ICC estimates. We created one simulation exercise to investigate how different event rates, event rate variations, cluster size, and cluster size variations impact ICC estimates and 95% confidence intervals. A separate simulation exercise in four different trial scenarios examined the impact of an intervention or drug effect in the intervention group on ICC estimates, 95% confidence intervals, and on sample size. The first simulation results suggest that the ICC value depends upon the event rate and event rate variations in addition to the cluster size, cluster size variations, and number of clusters. The second simulation exercise suggested that adjusting the sample size will help to preserve the appropriate power at the end of the trial.

Duke Scholars

Author Hrishikesh Chakraborty Biostatistics & Bioinformatics