Application of Relational Bayesian Networks to Vector Control in New York City

Abstract

We propose a novel way of rapidly formulating hypotheses describing the efficacy of West Nile virus (WNV) control campaigns using relational Bayesian Networks (RBN). RBN takes advantage of both quantitative information and expert opinion. We used multiyear, geo-referenced, and temporal data about dead birds, mosquitoes and humans, to infer probable relationships using proprietary software (CleverSet® Modeler) and discover the RBN that best fit the data for our initial models. Additional expert knowledge was later utilized for augmenting subsequent models. The final RBM was created after relating all the variables of interest that correlate to the occurrence of human cases, positive birds, and positive mosquitoes. The findings of this research suggest that WNV positive birds are valuable indicators of WNV activity. The most probable core organisms are bird-feeding Culex and ground-dwelling birds. WNV preventive measures such as public education, source reduction, and larviciding may be supplemented by adult mosquito control by ground spraying. To effectively interrupt the chain of transmission, careful effort must be devoted to ensuring that fewer mosquitoes remain the day after spray.