Modelos de regressão para dados de áreas com valores extremos no estudo da incidência de dengue em Minas Gerais

Abstract

Dengue fever has become a growing threat in recent years, with frequent outbreaks and overwhelming healthcare systems. The spread of the virus, transmitted by Aedes aegypti, is strongly related to environmental and urban factors, making the use of statistical methods essential to understand its dissemination and implement more effective actions to combat it. Traditional linear regression models, although widely used, assume the independence of observations and do not consider spatial autocorrelation, which, when present, can compromise the accuracy of estimates in geographically interdependent areas. To overcome this limitation, spatial models such as Simultaneous Autoregressive Regression (SAR) and Conditional Autoregressive Regression (CAR) have emerged as solutions. These models are able to capture spatial patterns that are normally present in area data. However, these models are sensitive to outliers, which generally impair the quality of the analyses. Geographically Weighted Regression (GWR), which incorporates spatial variation through local estimation of coefficients, has been proposed as an alternative but can also be influenced by outliers. Robust Geographically Weighted Regression (RGWR) has been proposed as an improved solution that can provide greater robustness against outliers and, consequently, lead to more effective estimates. This study applies RGWR to investigate the incidence of dengue fever in the state of Minas Gerais, in the year of 2023, aiming to improve the accuracy of estimates in the face of outliers, which are very common in this type of data. The variables used in the study were: dengue fever incidence as the dependent variable and adequate sanitation, urbanized area, street urbanization, and street tree planting as independent variables. The results showed that RGWR can provide a more accurate approach to understanding the spatial distribution of dengue in data from areas with outliers which can contribute to more efficient and effective management of public health actions.