Intraclass Correlation Coefficients for Cluster Randomized Trials with Pain Outcomes
A working document from the Pain Management Collaboratory Biostatistics/Design Workgroup
Prepared by: Charity G. Patterson, PhD, MSPH (University of Pittsburgh), Scott S. Coggeshall, PhD
(VA Puget Sound), Tom Greene, PhD (University of Utah)
Acknowledgement: Clair Smith, MS (University of Pittsburgh
Version 1.0, last updated December 10, 2019
Cluster Randomized Trials are particularly well suited to testing differences in a method or approach to patient care and are commonly used for testing interventions in real world settings, as in the case of the PMC’s 11 Pragmatic Trials. Sample size calculations for cluster randomized trials require an additional input, the intraclass correlation coefficient (ICC). The ICC estimates the correlation among individuals’ responses within the same cluster and can have a large impact on the sample sizes needed to appropriately power cluster randomized trials.
Unfortunately, to date, most trials have had to make assumptions about the ICC with limited prior evidence. This white paper encourages investigators to report their observed ICCs once trials are completed so as to help improve the design of future cluster trials in pain research.
In this working document, the Pain Management Collaboratory Biostatistics/Design Workgroup offers an overview of the ICC, further discussion on the impact of sample size calculations, and recommendations for researchers planning cluster randomized trials with pain outcomes.