Machine learning models map the risk of West Nile virus in the northeastern US

Since its introduction to North America in 1999, the West Nile virus (WNV) has been the dominant cause of mosquito-borne diseases in the United States. There are no vaccines or medications to prevent or treat people's illnesses.

While WNV mosquito surveillance is a central component of public health responses, 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 predicts the risk of WNV in areas of the Northeastern United States using freely available weather, land cover, and demographic data, in conjunction with mosquito surveillance data from Connecticut from 2001 to 2020. Their research is published in pnas nexus.

Variables found to predict WNV risk in this region include mixing of current and previous moon temperatures and precipitation, current and previous drought conditions, and prevalence of high population density and urbanized land cover.

According to the model, the risk of WNV is consistently the highest in the urbanized south and eastern edges of the northeast, particularly from July to September. This model can generate more monthly risk maps for WNVs. This, according to the authors, is different from the way weather forecasts communicate health hazards associated with extreme weather, advancing the local ability to predict and communicate risks in the absence of aggressive mosquito surveillance.

Provided by PNAS Nexus