Background and Objective: To describe the association between values for a proportion and the intraclass correlation coefficient (ICC). Methods: Analysis of data obtained from the General Practice Research Database (GPRD) for variation between United Kingdom general practices and results from a Health Technology Assessment (HTA) review for a range of outcomes in community and health services settings. Results: There were 188 ICCs from the GPRD, the median prevalence was 13.1% (interquartile range IQR 3.5 to 28.4%) and median ICC 0.051 (IQR 0.011 to 0.094). There were 136 ICCs from the HTA review, with median prevalence 6.5% (IQR 0.4 to 20.7%) and median ICC 0.006 (IQR 0.0003 to 0.036). There was a linear association of log ICC with log prevalence in both datasets (GPRD, regression coefficient 0.61, 95% confidence interval 0.53 to 0.69, P < 0.001; HTA, 0.91, 0.81 to 1.01, P < 0.001). When the prevalence was 1% the predicted ICC was 0.008 from the GPRD or 0.002 from the HTA, but when the prevalence was 40% the predicted ICC was 0.075 (GPRD) or 0.046 (HTA). Conclusion: The prevalence of an outcome may be used to make an informed assumption about the magnitude of the intraclass correlation coefficient.
Bibliographical noteFunding Information:
This work was supported by the UK Medical Research Council and the Charitable Foundation of Guy's and St Thomas' Hospitals. We thank the journal's reviewers for contributions to the revision of the paper.
- Binary data
- Cluster randomization
- Cluster sampling
- Intraclass correlation coefficient