Estimating Intraclass Correlation Coefficient Using Data from Pilot Study of Two Clusters

• Activity Number: 175
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Zhiying You*+ and Lorraine Robbins
• Companies: Michigan State University and Michigan State University
• Keywords: intraclass correlation coefficient ; cluster randomized trials ; estimation ; pilot study

Abstract

An estimated value of the intraclass correlation coefficient (ICC) is among those design parameters that are required in the design of cluster randomized trials specifically for the calculation of power and sample size requirement. In practice, a small pilot study is usually conducted to provide such information for the design of a larger study. However, some pilot studies may be too small to provide the required information. For example, if a pilot study has only two clusters with one in the control and the other in the intervention, then the ICC cannot be estimated by using outcome variable measured after intervention, because the inter-cluster variance cannot be distinguished from potential intervention effect. In this study, we propose estimate the ICC by using bootstrapping based on the baseline measures of outcome variables from such pilot studies with only two clusters. Simulation study is performed to examine the behavior of the estimate. The results indicate that the proposed approach serves as an option of estimating the ICC when a pilot study has only two clusters under some conditions, and results are subjected to limited data.