A Bayesian hierarchical spatial model for West Nile Virus in New York City: evaluating an approach to handle large spatial data sets

In 1999 West Nile Virus (WNV) was first detected in the Western Hemisphere in New York City, NY and is now responsible for approximately 1.8 million human infections, 360000 illnesses, and 1308 deaths. Our objectives are to: 1) develop a geostatistical model to assess which covariates influence WNV distribution, 2) assess the predictive performance of the developed model, and 3) explore approaches to reduce the computational burden associated with fitting spatial models to large data sets. We found a positive effect of land cover type, housing density, and distance to wetlands, and a negative effect of slope on the prevalence of WNV in New York City. We found predicted WNV occurrence to be highest in Staten Island. We believe the use of hierarchical modeling provides a useful approach to exploring the prevalence and distribution of WNV, but does not come without a cost regarding the difficulty of optimal implementation.