Dichotomizing a continuous outcome in cluster randomized trials: impact on power

Area of Science:

• Biostatistics
• Clinical Trials Methodology

Background:

• Cluster randomized trials (CRTs) involve randomizing groups, not individuals.
• Statistical power in CRTs is influenced by the intra-class correlation coefficient (ICC), representing within-cluster similarity.
• Higher ICC values decrease power for a fixed sample size.

Purpose of the Study:

• To assess the impact of dichotomizing continuous outcomes on statistical power in cluster randomized trials.
• To evaluate whether the reduction in ICC from dichotomization compensates for information loss.

Main Methods:

• A simulation study was conducted to evaluate the effect of dichotomization on ICC in CRTs with variable cluster sizes.
• An analytical formula for power was derived using generalized estimating equations after dichotomizing a continuous outcome.
• The accuracy of the analytical formula was assessed via simulation, comparing equal and varying cluster sizes.

Main Results:

• Dichotomizing a continuous outcome in CRTs leads to decreased statistical power.
• The reduction in ICC due to dichotomization does not offset the power loss from reduced information.
• Power loss is mitigated when the initial continuous-outcome ICC is high or the dichotomized outcome prevalence approaches 50%.

Conclusions:

• Dichotomizing continuous outcomes in CRTs generally reduces statistical power.
• Researchers should carefully consider the trade-offs between information loss and ICC reduction when dichotomizing outcomes in CRTs.
• Strategies to mitigate power loss include retaining continuous outcomes or ensuring outcome prevalence is near 50% if dichotomization is necessary.