29 - Forecasting West Nile virus incidence with environmental drivers

Abstract: West Nile virus (WNV) is a zoonotic virus and thus intrinsically linked to the environmental conditions that support the vector mosquitoes and birds necessary for its propagation. Population dynamics of the vectors and hosts and the seasonality of WNV disease in humans underscore the link between transmission biology and human epidemiology. However, the role of environment in determining substantial year-to-year variability in yearly WNV incidence is unclear. In 2020 and 2022, the Centers for Disease Control and Prevention coordinated WNV Forecasting Challenges to assess the ability of models to forecast county-level incidence across the United States. The majority of submitted forecasts integrated climate data with some also using data on land use, mosquitoes, birds. Despite including these relevant factors in carefully built models, a simple model based only on historical human WNV data that used no year-specific information performed as well as the best submitted forecast in 2020 and was second only to the ensemble in 2022. Moreover, amongst submitted forecasts we found mixed evidence about the value of environmental data. Evaluation of additional forecast models using standard time series and machine learning methods with environmental covariates similarly indicated limited forecasting skill, even when models were fitted within climate regions. Together, these results indicate that despite clear biological links to the environment, environmental data were not sufficient to improve yearly forecasts across the US with the models assessed here. These findings also highlight potential avenues to improve forecasts, such as using different approaches for high incidence versus low incidence locations and tailoring models more specifically to local dynamics where the combinations of avian and mosquito species and their environmental determinants may be highly varied.