Design and analysis of trials for mosquito control interventions with contamination

The movement of mosquitoes leads to a contamination of the effect of an intervention targeting mosquito-borne diseases because people living nearby might benefit from a reduced density of infectious mosquitoes. It is usually attempted to avoid these diffuse effects when testing new interventions. In this thesis however, we seek to understand these contamination effects and build the theoretical basis to estimate them since they may provide valuable information about the intervention effectiveness and should be considered.
We use partial differential equations to describe the dynamics of an Aedes aegypti population under a modified male mosquito release on an island to inform the trial design of such an intervention. By invoking optimal control theory, we assess the optimized release strategy under the constraint of a limited availability of modified males to achieve elimination as fast as possible. Our findings show that, to eliminate Aedes aegypti from a single location, the optimal release strategy is to initially release a high number of modified males and to subsequently release fewer mosquitoes proportionally to the decreasing female population. The best approach for elimination on the whole island is to target high mosquito density areas first and then move the focus in both directions along the periphery of the island until all areas have been covered. Sufficient release intensity has to be retained in the already targeted areas throughout this process to prevent reintroduction.
We then shift our focus to the analysis of trials targeting Anopheles mosquitoes. Informed by a model that describes the dispersion of mosquito with Gaussian kernels, we propose a nonlinear random effects model based on a sigmoid function of the distance to the nearest discordant household to analyze cluster randomized trials or stepped wedge cluster randomized trials of malaria interventions. This model approach leads to a closed-form contamination range that quantifies the measurable extent of the contamination. In a simulation study, we find that this approach indeed provides unbiased and precise estimates of effectiveness if an appropriate number of households is not affected by the contamination range. We extend this model to provide an estimate of the intervention effectiveness as a function of the intervention coverage at each household, that we define based on the estimated contamination range.
This methodological development is then applied to three trials of malaria interventions: the Navrongo trial on the use of insecticide treated nets, the SolarMal trial on the impact of mass trapping of mosquitoes with odor-baited traps and the AvecNet trial on the effect of adding pyriproxyfen to long-lasting insecticidal nets. These three trials were conducted in different countries with different settlement patterns and were testing different malaria interventions. In our reanalyses we find that a sigmoid analysis yields a similar estimate of effectiveness compared to what was found in the original analysis. Furthermore, in all three trials the contamination range is around 100 to 200 meters,
which is much less than the maximal distance Anopheles mosquitoes can fly.
For contamination effects to be estimable, the trial must be designed to collect information from zones where contamination is likely. We use the gained insights to give guidance on how to plan trials to allow for contamination and develop algorithmic approaches for cluster construction that allow for cluster boundaries to pass villages and hence enable the estimation of the contamination range. We conclude by connecting the work on the optimized release strategy for modified male mosquitoes with the analysis of trials with contamination by proposing a trial design that accounts for the dispersal of modified males and tests both the short-term effectiveness and the potential for elimination.
The work reported here creates a solid foundation for measuring and understanding the effects of contamination in trials of mosquito-borne diseases. Cluster size can be reduced to the minimum determined by operational factors or contamination effects, without the need for clusters to correspond to separated villages. This reduces the required number of participants in trials of malaria interventions and should lead to more cost-efficient trials and a better understanding of the indirect effects of interventions in protecting nearby nonusers.