Epidemiological data often include characteristics such as spatial and/or temporal dependencies and excess zero counts, which pose modeling challenges. Excess zeros in such data may arise from imperfect detection and/or relative rareness of the disease in a given location. Here, we studied the spatio-temporal variation in annual Lyme disease cases in Virginia from 2001-2016 and modeled the disease with a spatio-temporal hierarchical Bayesian model. Using observed ecological and environmental covariates, we constructed a predictive model for the disease spread over space and time, including spatial and temporal random effects. We considered several different models and found that the negative binomial hurdle model performs the best for such epidemiological data. Among the various ecological predictors, the North-South (V component) of winds and relative humidity significantly contributed to predicting the Lyme cases. Our model results provide important insights on the spread of the disease in Virginia and the proposed modeling framework offers epidemiologists and health policymakers a useful tool for improving disease preparedness and control plans for the future.
