Spatial nonparametric Bayesian Poisson hurdle model for analyzing zero-inflated tick data

Abstract

Severe fever with thrombocytopenia syndrome, an infectious tick-borne disease, has been increasingly reported in South Korea. Understanding the epidemiology of its vectors will help predict the spatio-seasonal abundance of ticks and manage future health risks related to this disease. We develop a spatial nonparametric Bayesian Poisson hurdle random effects model (snp-BayesPHM) to investigate the spatio-seasonal distribution of different life stages of Haemaphysalis longicornis ( H. longicornis ). Specifically, we characterize the spatial variation in seasonal patterns and identify the key factors that may underlie spatial heterogeneity. The proposed model consists of two parts: a binary part that models the probability of tick presence and a positive part that models nonzero tick abundance. Through a hierarchical structure, the model incorporates location-specific random effects and environmental covariates, while spatial dependence across locations is introduced through a local Dirichlet process indexed by geographical coordinates. To identify the underlying subgroup structure, we assume a local Dirichlet process mixture of normals for the joint vectors of random effects and location-specific covariates. We compare the proposed model with the nonspatial nonparametric Bayesian Poisson hurdle random effects model (np-BayesPHM) using the deviance information criterion (DIC). The proposed model yields lower DIC values for both life stages, indicating better model fit. It also reveals strong seasonal patterns in both tick presence and nonzero tick abundance, as well as distinct spatial clustering patterns for the nymph and adult stages.