Using high-resolution mosquito surveillance data and a mechanistic metapopulation model to predict within-city spread of dengue fever
Dengue fever is a major public health risk in Brazil, with over three million human cases reported between 2000 and 2007 (Cardoso et al. 2011). The dengue virus is spread by the mosquito vector Aedes aegypti, and as such, prevention of dengue fever is accomplished primarily through municipal mosquito control programs. While these programs have been successful in preventing dengue cases (Pepin et al. 2013), how to best use mosquito surveillance data to inform control efforts remains an open question. At the municipal scale, connecting patterns of mosquito abundance to human disease risk is complicated by human movement and heterogeneous mixing between human and vector populations in space. As a result, mosquito control policies would benefit greatly from the development of mechanisitic models that can explicitly incorporate human-vector transmission processes and human movement. In this work, we adapt an existing single population model (Wearing and Rohani 2006) into a metapopulation context, where neighborhoods within a city are coupled by human movement. We combine this model with high-resolution mosquito surveillance data and a statistical framework that can predict observed weekly case reports from the city of Vitoria, Brazil.
The metapopulation model is able to successfully capture seasonal dengue dynamics at the city scale. This model performs better than both models that assume city-wide homogeneous mixing between humans and mosquitoes, and models that treat each neighborhood independently. However, metapopulation models that model human movement among neighborhoods as a decaying function of distance do not capture the neighborhood-level heterogeneity in dengue dynamics. One possible explanation for this is that distance alone does not provide enough structure for human movement patterns and allows for too much mixing among neighborhoods. We can incorporate additional structure through a framework that produces a sparse human movement matrix, that is, one in which many between-neighborhood movement rates are exactly zero. Such methods are currently being developed and will provide greater insight into the patterns of human-mosquito mixing within Vitoria. Capturing these patterns will in turn help to better predict human disease risk from mosquito surveillance data and thereby better inform the allocation of mosquito control efforts within the city.