Methods for the analysis of incidence rates in cluster randomized trials

Abstract

The techniques provide a straightforward approach to the analysis of incidence rates in cluster randomized trials. Both the unadjusted analysis and the analysis adjusting for confounders are shown to be robust, even for very small numbers of clusters, in situations that are likely to arise in randomized trials.

Efficacy of indoor residual spraying with dichlorodiphenyltrichloroethane against malaria in Gambian communities with high usage of long-lasting insecticidal mosquito nets: a cluster-randomised controlled trial

“…All subsequent analyses used incidence rates calculated over both malaria seasons and censoring at the first attack. These two-year incidence rates were calculated for each cluster and the mean rate ratio calculated by study arm with confidence intervals obtained using the approximations given by Bennett 36 . Time to first malaria attack was examined by a survival analysis approach using Kaplan-Meier curves to compare the probability of subjects in the two arms becoming infected as the malaria transmission seasons progressed and significance was calculated using a log-rank test.…”

The clinical effectiveness and cost-effectiveness of telephone triage for managing same-day consultation requests in general practice: a cluster randomised controlled trial comparing general practitioner-led and nurse-led management systems with usual care (the ESTEEM trial)

“…Cluster-level SDs were reported where appropriate, as this parameter approximates the coefficient of variation in underlying cluster rates in certain models. 57,58 To derive an ICC for the primary outcome, a linear hierarchical model was also fitted. Additional analyses were conducted on the POM, and on secondary measures derived from the POM, using the hierarchical GLM methods above.…”

“…The trial was a rigorously conducted multicentre cluster RCT, using both quantitative and qualitative methods and in full conformity with the CONSORT guidelines. 57 It was a definitive study, including a large number of both practices and patients with few exclusion criteria and a spread of geographical locations, increasing the external validity of the trial and the generalisability of the findings. Although we succeeded in recruiting large numbers of practices and patients from diverse geographical locations, enhancing the generalisability of our results, our findings may be less applicable to practices serving populations with greater ethnic diversity, or those located in inner-city areas with very high levels of deprivation.…”

A simple approach to test for interaction between intervention and an individual‐level variable in community randomized trials

“…The type I error rates and power follow the nominal level closely, although the power is somewhat reduced in case of only four communities per trial arm. The robustness of the t-test has also been shown by previous simulation studies (Bennett et al 2002) and theoretical exploration (Heeren & D'Agostino 1987). Use of the Wilcoxon test was less powerful, but there was no advantage in terms of type I error rates.…”

“…Useful though they are, when the number of communities per trial arm is small (£20) and the number of individuals per community is large, methods based on analysis of individual data values are not always valid and model diagnostics are sometimes not feasible (Donner & Klar 2000;Omar & Thompson 2000;Feng et al 2001;Murray et al 2004). In contrast, the summary statistics approach is valid even when the number of communities is smaller than about 20 per trial arm (Bennett et al 2002), but problematic when the number of subjects per community is small (Omar & Thompson 2000). As a small number of communities per intervention arm and a large number of observations per community are common characteristics of many CRTs, the summary statistics approach is a useful choice.…”