Aplicação de modelos lineares generalizados mistos na classificação de surtos de dengue em unidades de saúde da cidade de Belo Horizonte

Abstract

The aim of this study is to apply generalized linear mixed models (GLMMs) to forecast dengue outbreaks in health units in Belo Horizonte, Minas Gerais, Brazil, including information on climate variables and space-time effects. An outbreak is defined when the total number of dengue reports equal to or exceeding the 95th percentile of the historical series of monthly dengue reports in the health unit. The applied GLMM include two random effects and fifteen fixed effects. Results showed that climate variables such as temperature, precipitation, and humidity have a significant effect on the odds of dengue outbreaks. For some of them, it was estimated that the measurement in one, two, or even three previous months could affect the chance of outbreak in the current month. In the final model, only the random effect related to the month of occurrence was shown to be relevant for explaining the variability of dengue outbreaks. To test the quality of the classifications, the accuracy, sensitivity, and specificity measures were used, which returned predictive capacity above 91%.