Dengue Dynamics: Bayesian Spatio-Temporal Analysis using Intrinsic Conditional Autoregressive Model

Abstract

In data-limited regions such as the Philippines, exploring socio-economic and environmental factors beyond traditional mosquito vector studies is crucial to understanding disease dynamics. This study employed a Bayesian spatio-temporal approach to model dengue transmission in Region 10, incorporating time-varying coefficients and a latent spatial effect with an Intrinsic Conditional Autoregressive (ICAR) prior. Three models were evaluated: Model 1 included only the intercept and spatial autocorrelation term, Model 2 included time-varying covariates without spatial effects, and Model 3 incorporated both time-varying coefficients and spatial effects. Model 3 outperformed the other models in disease modeling (WAIC = 24,520.7), providing clearer insights into the relationships between key covariates and dengue incidence. In Model 3, the inclusion of the spatial effect  $\phi_i$ resulted in a shift in the relationships between key covariates and dengue incidence. The relationship between CMCI and dengue cases became negative, while elevation showed a positive association. Coastal proximity also became more negatively associated with dengue risk. Model 3 revealed that few urban areas had persistent high  $\phi_i$ values, indicating that unmeasured factors continue to influence spatial variation in dengue risk. Conversely, many other municipalities showed a reduction in the magnitude of  $\phi_i$ suggesting that covariates in Model 3 better explained the spatial variation. Graphs and plots of observed and fitted values, as well as a plot of relative risk over time, demonstrated the model’s performance across both time and space. The inclusion of spatial autocorrelation and time-varying effects in Model 3 improved model fit and provided a more nuanced understanding of dengue dynamics, highlighting the importance of considering both spatial and temporal components in disease modeling.