West Nile virus (WNV) is the leading cause of domestically acquired arthropod-borne viral disease in the United States; however, there is considerable inter-annual variation in the number of human cases. As a consequence, effective allocation of public health resources is challenging and often reactive, a circumstance that highlights the need for accurate, real-time forecasts of the burden of disease.
Recently, we showed that accurate and reliable predictions of seasonal West Nile virus (WNV) outbreaks can be made using a mathematical model representing WNV transmission dynamics among mosquitoes and birds, as well as spill-over to humans. The mathematical model is optimized using a data assimilation method and two observed data streams: mosquito infection rates and reported human WNV cases. This coupled model-inference framework has previously been used to generate retrospective ensemble forecasts of WNV for 12 geographically diverse United States counties over multiple seasons. These retrospective forecasts were then used to calibrate estimation of real-time forecast expected accuracies for various predicted features: outbreak peak timing, peak magnitude, and total number of infected mosquitoes for the season and the number of human cases in the next 4 weeks and over the season. Weekly forecasts of WNV were generated in real time using this calibrated system for 4 California Counties during the 2018 outbreak season.
Results/Conclusions
Here we present the forecasting framework, evaluation of the real time forecasts, and discuss limitation of the current real-time monitoring network. Overall the real-time forecasts were able to estimate accurately the peak timing, peak magnitude, and total number of infected mosquitoes for the season in real-time prior to the peak of infected mosquitoes. Forecasts of human WNV cases over the next 4-weeks were able to provide accurate prediction intervals of future observations. From week 27 to week 36, a time period in which, historically the majority of human cases have been reported in the following 4-weeks, the 4-week ahead 50% prediction interval captured 55% of observations. However, seasonal forecasts of human WNV cases were not as accurate during this time period: the 50% prediction intervals only captured 35% of the total number of human WNV cases for the season.
WNV forecasting is potentially an important evidence-based decision support tool for public health officials and mosquito abatement districts; however, to operationalize real-time forecasting, more resources are needed to reduce human case reporting lags between illness onset and case confirmation.