^{1}

^{2}

^{3}

^{1}

^{2}

^{*}

^{4}

Cameron Simmons is an investigator in the Eliminate Dengue Project, which aims to use

Dengue is the most important arboviral infection of humans. In endemic countries the scale of the dengue disease burden imparts an economic cost

Novel vector control approaches to limit dengue virus (DENV) transmission include release of

A critical challenge for all entomological approaches to control of vector-borne disease is how best to demonstrate efficacy in reducing disease transmission

Cluster randomised trials (CRTs) are the gold standard design to provide evidence on the efficacy of an intervention that has community-wide impact

We consider the incidence of DENV-seroconversions during a trial as a suitable primary endpoint and DENV-naïve children aged 2–5 years living in each cluster as an optimal “sentinel” cohort for serological surveillance. Young children are less likely to spend substantial periods of time outside of their residence and local community (and hence outside of the “treatment umbrella”) than more mobile older children and adults. In addition, DENV-prevalence in older children is higher and those remaining naïve and hence eligible for the study are potentially less representative of the full population (for example, for socio-economic reasons).

Two alternative designs are considered. The first is the classical parallel two-armed cluster randomised trial (PCRT) in which each recruited cluster is randomised to intervention or control, and the intervention is implemented simultaneously across the relevant clusters. Thus the control clusters provide contemporaneous controls for the intervention clusters. The other design considered is a stepped wedge cluster randomised trial (SWCRT) in which each cluster is assigned to the control treatment initially and clusters are subsequently crossed-over to the intervention in a random selection at fixed time points until eventually all clusters are under treatment

SWCRTs have been most frequently used for evaluating interventions during routine implementation such as the evaluation of a vaccine on the community level following a successful individual randomised trial. From a logistic perspective, they are attractive, because the intervention can be rolled out in a step-wise fashion and evaluated. As clusters are their own controls, SWCRTs are less sensitive to between-cluster variation and thus might require a lower sample size compared to parallel designs

Cluster size and cluster separation are important considerations in the design of all CRTs, but they require particular attention in trials of vector control interventions, for which entomological and community considerations need be taken into account. Entomological considerations include the dispersal of

Sample size requirements for CRTs of a

A mixed-effects Poisson-regression model fitted to these data gave coefficients of variation (cv, i.e., SD/mean) for yearly DENV infection incidence of 0.27 for between-school variation, 0.57 for annual variation, and 0.85 for residual variation (i.e., variation that cannot be explained by systematic spatial or temporal variation, respectively, and corresponds to localized school and year specific variation). A detailed description of the model used to derive these coefficients of variation can be found in

We then used the incidence and variability data reported above to simulate hypothetical PCRT and SWCRT trials. Additional assumptions for the trial simulations were a study duration of 3 years and a surveillance cohort of 100 children in each cluster. We varied the intervention effect between a 40% and an 80% decrease of DENV seroconversion in intervention clusters compared to controls. Allowing for the fact that some children in intervention clusters will experience infections outside of the intervention area, we regard an effect of a 50%–60% reduction as realistic in our target population. Details regarding the set-up of the simulation study and the statistical analysis of simulated trials are provided in

Sample size requirements for the two designs and for varying treatment effects are shown in

Total number of clusters required for a PCRT (black lines) or a SWCRT (blue lines) depending on the size of the intervention effect. Solid lines correspond to 90% power, dashed lines to 80% power. Simulations are based on parameters determined from the Kamphaeng Phet dengue cohort (Thailand) (described in

A parallel cluster-randomised trial is the design of choice for testing novel entomological methods of dengue control. Under realistic assumptions we show it to require a substantially lower sample size than a stepped wedge design. Sample size requirements for a parallel design are relatively modest; our example gave a minimum sample size of 20 clusters (ten per study arm) with each cluster providing 100 person-years of follow-up per year and a follow-up duration of 3 years. Although careful planning and substantial funding are required to run such a trial, the benefits of having a robust evidence-base from which to promote programmatic roll-out and/or further optimisation of the strategy should prove invaluable.

Statistical appendix containing: (1) a diagram of a parallel two-arm cluster randomised trial (PCRT) and a stepped wedge cluster randomised trial (SWCRT), (2) details regarding the determination of coefficients of variation for the Thailand data, and (3) details regarding the simulation study to compare PCRT versus SWCRT designs.

(DOCX)