Filling surveillance gaps: Bayesian INLA models for predicting tick distributions in data-sparse regions

Abstract

Fragmented and geographically uneven surveillance limits the ability to target tick monitoring and control in many livestock systems. We evaluated whether heterogeneous published surveillance could be converted into district-level, uncertainty-aware abundance predictions for Rhipicephalus microplus and Hyalomma anatolicum across Punjab and Khyber Pakhtunkhwa, Pakistan. Species-specific tick counts were standardized by the number of animals examined and sampling duration and expressed as ticks per 100,000 animal-months. Environmental covariates were summarized using principal component analysis, with livestock density included separately. Bayesian Gaussian models, with and without a district-level Besag-York-Mollie 2 spatial effect, were fitted in R using Integrated Nested Laplace Approximation (INLA) and compared. Leave-one-district-out validation was performed to assess out-of-sample predictive performance. Spatial models improved model fit for both species. For R. microplus, PC1 and PC4 showed credible negative associations with abundance, whereas no fixed-effect 95% credible interval excluded zero for H. anatolicum. Predicted R. microplus abundance was more spatially heterogeneous, with high but uncertain estimates in several northern districts, whereas H. anatolicum predictions were more homogeneous and relatively higher toward southern Punjab. Mapping posterior uncertainty identified districts where additional standardized surveillance would be most informative. This approach provides a transferable framework for extracting spatial surveillance information from heterogeneous veterinary data while explicitly representing uncertainty.