Validating a Bayesian Spatio-Temporal Model to Predict La Crosse Virus Human Incidence in the Appalachian Mountain Region, USA.

La Crosse virus (LACV) is a rare cause of pediatric encephalitis, yet identifying and mitigating transmission foci is critical to detecting additional cases. Neurologic disease disproportionately occurs among children, and survivors often experience substantial, life-altering chronic disability. Despite its severe clinical impact, public health resources to detect and mitigate transmission are lacking. This study aimed to design a Bayesian modelling approach to effectively identify and predict LACV incidence for geospatially informed public health interventions. A Bayesian negative binomial spatio-temporal regression model best fit the data and demonstrated high accuracy. Nine variables were statistically significant in predicting LACV incidence for the Appalachian Mountain Region. Proportion of children, proportion of developed open space, and proportion of barren land were positively associated with LACV incidence, while vapor pressure deficit index, year, and proportions of developed high intensity land, evergreen forest, hay pasture, and woody wetland were negatively associated with LACV incidence. Model prediction error was low, less than 2%, indicating high accuracy in predicting annual LACV human incidence at the county level. In summary, this study demonstrates the utility of Bayesian negative binomial spatio-temporal regression models for predicting rare but medically important LACV human cases. Future studies could examine more granular models for predicting LACV cases from localized variables such as mosquito control efforts, local reservoir host density and local weather fluctuations.