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

Area of Science:

• Epidemiology
• Spatial Analysis
• Public Health

Background:

• Dengue Haemorrhagic Fever (DHF) presents a significant public health challenge in Indonesia, characterized by notable provincial disparities.
• Understanding the spatial distribution and risk factors of DHF is crucial for effective public health interventions.

Purpose of the Study:

• To model province-level DHF counts in Indonesia for 2023.
• To investigate the association of average annual temperature and public health workforce density with DHF risk.
• To identify geographical areas with elevated DHF burden.

Main Methods:

• Bayesian spatial conditional autoregressive Poisson models with population offsets were employed.
• Besag-York-Mollié (BYM) and Leroux priors were utilized with Markov chain Monte Carlo methods.
• Deviance Information Criterion (DIC) and Watanabe-Akaike Information Criterion (WAIC) were used for model comparison.

Main Results:

• Spatial dependence in DHF distribution was statistically significant (Moran's I=0.4689, p=0.021).
• Average annual temperature showed a non-significant association with lower DHF risk (RR=0.90; 95% CrI: 0.76 to 1.07).
• Public health workforce density was associated with higher reported DHF risk (RR=1.05; 95% CrI: 1.03 to 1.07), potentially reflecting reporting capacity.

Conclusions:

• Elevated DHF risk was mapped in parts of Kalimantan and eastern Indonesia.
• Findings suggest the need for geographically targeted surveillance and vector control strategies.
• The association between workforce density and DHF risk requires cautious interpretation, possibly indicating reporting or deployment factors rather than direct causality.