Comparing conditional autoregressive models for Bayesian spatial mapping of dengue cases in Indonesia

Dengue Haemorrhagic Fever (DHF) remains a public health burden in Indonesia with substantial provincial variation. We modelled province-level DHF counts in 2023 using Bayesian spatial conditional autoregressive Poisson models with population offsets. Predictors were average annual temperature (per 1°C) and the number of public health workers (province-level count). Spatial dependence was supported by Moran's I=0.4689 (p=0.021). We fitted models using Besag-York-Mollié (BYM) and Leroux priors via Markov chain Monte Carlo and compared fit using the Deviance Information Criterion (DIC) and the Watanabe-Akaike Information Criterion (WAIC). In the BYM model, temperature was associated with lower risk (RR=0.90; 95% CrI: 0.76 to 1.07), with uncertainty including unity, whereas workforce density was associated with higher reported risk (RR=1.05; 95% CrI: 1.03 to 1.07). Estimates were similar under the Leroux prior (temperature RR=0.89; 95% CrI: 0.74 to 1.07; workforce RR=1.04; 95% CrI: 1.02 to 1.07), and BYM showed marginally better fit. Risk mapping indicated elevated burden in parts of Kalimantan and eastern Indonesia. Findings may inform geographically targeted surveillance and vector control; the workforce association should be interpreted cautiously because it may reflect reporting capacity or reactive deployment.