Cluster-randomized trials (CRTs) require accounting for within-cluster correlation when determining sample size and statistical power. Therefore, appropriate specification of intracluster correlation coefficients (ICCs) is crucial for planning CRTs. However, identifying suitable ICC values can be challenging because they may vary across outcomes, populations, settings, and cluster definitions, while estimates from prior studies may be limited or not always directly applicable to a new trial. The challenge is greater for multiple-period CRTs, such as stepped-wedge and longitudinal designs, in which outcomes are correlated both within clusters and across time, and multiple correlation parameters may be required.

This webinar will provide a practical framework for selecting and estimating correlation parameters for CRT design. The presenter will discuss how prior trials, historical data, routinely collected health data, and empirical ICC repositories can inform design assumptions. For multiple-period CRTs, commonly used correlation structures and their associated parameters will be introduced, including within-period ICCs, between-period ICCs, and cluster autocorrelation coefficients. The presenter will also discuss approaches for estimating these parameters and deriving plausible values for more complex correlation structures when only simpler correlation estimates are available.

This webinar will be captioned in real time. Individuals needing reasonable accommodations should e-mail [email protected]. Requests should be made at least 5 business days before the event. 

This event is open to the public and there will be an opportunity to ask questions at the end of the presentation. This event will be recorded and made publicly available on the @NIHODP YouTube channel within a few weeks.