Spatiotemporal modeling of relative risk of dengue disease in Colombia

Abstract

Spatiotemporal modeling of relative risk of dengue disease provides useful risk maps for surveillance and forecasting. The objective of the study was to generate smoothed estimates of relative risk applying hierarchical Bayesian spatiotemporal models, including covariates derived from satellite images containing land surface temperature (LST) and normalized difference vegetation index, for the period January 2009–December 2015, in a medium–sized Colombian city. Our models are based on the spatiotemporal interaction modeling framework of relative risk, where the interaction effects are unstructured, temporal, spatial or inseparable, at a small spatial and temporal scales. We fitted the models using Markov chain MonteCarlo Simulations, selecting the best model using leave-one-out cross-validation and widely applicable information criteria. Our best model was the inseparable spatiotemporal interaction-effects plus LST with constant coefficient model. We found a weak, positive association between LST and cases of dengue. We discussed the strengths and weaknesses of our spatiotemporal models given the spatial and temporal resolution selected in the study.