Affiliation:
1. Texas Tech University
2. Indiana University
3. The Connecticut Agricultural Experiment Station
Abstract
Abstract
Background: Predicting risk of West Nile virus (WNV) to humans in spaces without mosquito surveillance data is a key limitation of many WNV surveillance programs. To address this knowledge gap, we analyzed 20 years (2001 – 2020) of statewide, point-level mosquito and WNV surveillance data from Connecticut (CT), USA, using boosted regression trees (BRT) and generalized linear models (GLMs) to determine the most influential climate variables, land cover classes, and seasonality factors (such as Month of collection) associated with Culex pipiens abundance and WNV presence/absence in tested mosquito pools.
Methods: Candidate Cx. pipiens collection models were assessed based on explained deviance and root mean square error then optimized using a backward selection process. We then used predicted Cx. pipiens abundance in models of WNV presence/absence to predict WNV detection probabilities throughout CT. We validated these WNV predictions by testing the association between predicted WNV detection probabilities in mosquitoes and observed WNV incidence in mosquitoes from 2021 – 2022 and in humans from 2001 – 2022 using binomial-error generalized linear mixed effects models.
Results: Predicted mosquito WNV detection probabilities in unsampled spaces were significantly associated with the odds of a mosquito pool testing positive as well as a human case occurring within the geopolitical boundaries of a town.
Conclusion: This methodology has broad utility in the US and abroad to inform the public of risk of WNV quickly and easily in mosquitoes using only a few online and easily accessible data sources.
Publisher
Research Square Platform LLC