Cluster randomised trials in implementation research : intracluster correlation coefficients, sample size and reporting considerations
This thesis examines aspects of the application of the cluster randomised trial design in implementation research. In particular, it focuses on three main themes: a) the intracluster correlation coefficient (ICC) in cluster randomised trials of implementation research and factors affecting its magnitude; b) the impact of the ICC on sample size calculations; and c) reporting considerations for cluster randomised trials generally, and for ICCs in particular. The results show that empirical estimates of ICCs vary in size and certain factors - particularly the type of variable and the study setting - influence the magnitude of the ICC. When reporting an ICC, three aspects were found to be important: a) a description of the dataset, b) information on how the ICC was calculated; and c) information on the accuracy of the ICC. General reporting considerations for cluster randomised trials were also considered and found to require explanations for: a) the rationale for adopting a cluster design; b) how the effects of clustering were incorporated into the sample size calculations; c) how the effects of clustering were incorporated into the analysis; and d) the flow of both clusters and individuals through the trial, from assignment to analysis. A revised CONSORT statement accommodating the specific features of a cluster randomised trial was presented. The work outlined in this thesis shows that the adoption of a cluster randomised trial in implementation research has a number of important implications that researchers should consider when planning future trials.