Combining statistical and dynamical modelling to guide the design of cluster randomised trials for malaria.

Abstract

Cluster randomised trials (CRTs) remain key for evaluating the community-wide impact of interventions against infectious diseases such as malaria. Randomising by cluster prevents contamination and enables both the direct and indirect effects of the intervention to be estimated. Although these trials are extremely informative, they can be logistically demanding and costly to carry out, which means it is important that these trials are well powered. Here, we present a framework for planning CRTs that measure malaria prevalence as the outcome using an established mathematical model of malaria transmission. In this way, we explicitly consider the epidemiology of the individual trial clusters. The framework can be used alongside a baseline prevalence survey to help inform the sample size calculations for the trial. We use a case study to illustrate the framework, where we simulate a CRT in which a next-generation pyrethroid-pyrrole insecticide-treated net (ITN) is compared against a standard pyrethroid-only ITN. We show how the malaria endemicity of the trial location and timing of the follow-up surveys can affect the results obtained. We also highlight how other active interventions against malaria can reduce study power by increasing the amount of between-cluster heterogeneity in malaria prevalence.