Exploring local and global regression models to estimate the spatial variability of Zika and Chikungunya cases in Recife, Brazil

• Published in: Revista da Sociedade Brasileira de Medicina Tropical
• Main Authors: Rafael Silva dos Anjos, Ranyére Silva Nóbrega, Henrique dos Santos Ferreira, António Pais de Lacerda, Nuno de Sousa-Neves
• Format: Article in Journal/Newspaper
• Language: English
• Published: Sociedade Brasileira de Medicina Tropical (SBMT) 2020
• Subjects: Spatial Analysis; Public Health; Spatial Interaction; Arctic medicine. Tropical medicine; RC955-962; Arctic
• Online Access: https://doi.org/10.1590/0037-8682-0027-2020; https://doaj.org/article/1401e2060f524327b692e3646561ff95

Abstract INTRODUCTION: In this study, we aim to compare spatial statistic models to estimate the spatial distribution of Zika and Chikungunya infections in the city of Recife, Brazil. We also aim to establish the relationship between the diseases and the analyzed geographical conditions. METHODS: The models were defined by combining three categories: type of spatial unit, calculation of the dependent variable format, and estimation methods (Geographical Weighted Regression [GWR] and Ordinary Least Square [OLS]). We identified the most accurate model to estimate the spatial distribution of the diseases. After selecting the model that provided best results, the relationship between the geographical conditions and the incidence of the diseases was analyzed. RESULTS: It was observed that the matrix of 100 meters (as the spatial unit) showed the highest efficiency to estimate the diseases. The best results were observed in the models that utilized the kernel density estimation (as the calculation of the dependent variable). In all models, the GWR method showed the best results. By considering the OLS coefficient values, it was observed that all geographical conditions are related to the incidence of Zika and Chikungunya, while the GWR coefficient values showed where this relationship was more noticeable. CONCLUSIONS: The model that utilized the combination of the matrix of 100 meters, kernel density estimation (as the calculation of the dependent variable) and GWR method showed the highest efficiency in estimating the spatial distribution of the diseases. The coefficient values showed that all analyzed geographical conditions are related to the illnesses’ incidence.