West Nile virus (WNV) has been the dominant cause of mosquito-borne illness in the United States since its introduction into North America in 1999. There are no vaccines nor medications to prevent or treat illness in people, so surveillance, prevention, and control remain the best options to protect the public. Mosquito surveillance for WNV is a central component of the public health response, but this approach is labor intensive and limited by practical constraints on the number of locations that can be sampled. To address this limitation, Joseph McMillan and colleagues developed a validated machine learning model that uses freely available weather, land cover, and demographic data in combination with mosquito surveillance data from the state of Connecticut from 2001 to 2020 to predict the risk of WNV to humans in areas of the Northeast US with and without mosquito surveillance operations. Variables found to be predictive of WNV risk in the region include a mix of high temperature and precipitation during the current and prior months, current and prior drought conditions, high human population density, and the prevalence of urbanized land cover. According to the model, WNV risk is consistently highest in the urbanized southern and eastern edge of the Northeast, especially during July–September. This model can further generate monthly risk maps of WNV, which, according to the authors, advances the region’s capacity to predict and communicate risks in the absence of active mosquito surveillance, not unlike how weather forecasts inform health hazards associated with extreme weather.
PNAS Nexus
Using mosquito and arbovirus data to computationally predict West Nile virus in unsampled areas of the Northeast United States
19-Aug-2025