Last Updated: 17/10/2018
Design of trials for pathogen elimination
Objectives
In field trials of new interventions against mosquitoes and the diseases they transmit, there are often spill-over effects between areas that receive the intervention and others that do not, because of mosquito movement.
This project is developing ways of designing trials to make the best use of data on spill-over effects to better understand the potential impacts of interventions (including odour-baited traps, and novel bednets), including effects of mosquito movement.
It will develop the required statistical methods for allowing for contamination in trial design, for analyzing the extent of contamination and using this for causal inference, and generalization of intervention impacts to non-trial settings. This methodological development will entail deriving:
- Point and interval estimates of the extent of contamination in CRTs and SWCRTs based on measuring gradients in outcomes across boundaries between trial arms. Some trials have assessed these gradients but without accounting for the correlation structure of the data. There is a need for practical analytical approaches that account for this correlation structure in both estimating the gradient, and in estimating power to test if the gradient is non-zero.
- Algorithm(s) for optimising cluster size and randomization strategies for CRTs and SWCRTs in the presence of contamination. These will be based both on considerations of statistical power of outcome measures, including measures of contamination effect and in the case of SWCRTs, of how the evidence for causality is influenced by stratification in the trial randomization.
Swiss Tropical and Public Health Institute (Swiss TPH), Switzerland
Cluster-randomized trials (CRTs) are used to evaluate health interventions that have effects at the community level. In such trials, the effects of the intervention may spill-over into adjacent areas of control clusters leading to contamination effects. Trials of interventions against infectious diseases increasingly aim to evaluate the potential to interrupt pathogen transmission at maximum scale-up. To achieve this, there is a need for stepped-wedge designs. In stepped-wedge cluster-randomised trials (SWCRTs), assignment of clusters to the intervention, and hence the contamination effects, are time dependent.
Established statistical methods seek to minimize effects of contamination, and hence neither measure it, nor exploit the information that it provides. Rather than aiming to avoid contamination, the proposal is that contamination effects may provide valuable evidence about the intervention effectiveness. Contamination, assessed via sub-cluster spatial variation in outcomes and patterns of outcomes across cluster boundaries, should be both measured as a trial outcome and used to make inferences about the properties of the intervention when deployed in non-trial settings. However, while intervention assignment in such trials is randomly assigned, spatial configuration is not. Measures of the contamination effect, therefore, depend jointly on factors over which trial participants are randomized and factors where randomization plays no role, and this raises novel issues in causal inference.
Jan 2016 — Dec 2018
$339,000
Related Resources
