Modelagem estatístico-computacional para predição de casos de dengue no município de Joinville (SC)

Abstract

This study aimed to model and predicted dengue incidence in the city of Joinville, southern Brazil, from 2020 to 2023 using statistical count models applied to neighborhood-level aggregated data. An ecological, quantitative, and observational multiple-group design was conducted, with neighborhoods serving as the unit of analysis across the historical series. The dataset included reported dengue cases, population estimates, and environmental indicators, particularly vegetation proportion. After exploratory analysis, Poisson and Negative Binomial regression models were fitted, both incorporating a population offset. Strong overdispersion rendered the Poisson model inadequate, while the Negative Binomial model showed superior performance, with lower AIC, more homogeneous residuals, and consistent halfnormal plots. Results indicated a significant yearly increase in dengue incidence, a strong inverse association between vegetation cover and case counts, and substantial spatial heterogeneity, with higher relative risks concentrated in peripheral neighborhoods. The findings demonstrate that the Negative Binomial model is statistically and epidemiologically more appropriate for capturing the spatiotemporal dynamics of dengue transmission in Joinville and provides a valuable tool for epidemiological surveillance and vector-control planning.