Avaliação de dados do sistema de saúde e elaboração de ferramentas de triagem

Abstract

Neglected tropical diseases, such as dengue and leptospirosis, pose a significant public health challenge, particularly impacting impoverished populations. Dengue, an arbovirus transmitted by mosquitoes of the Aedes aegypti genus, is continuously expanding globally. Due to its socioeconomic impact, there is a need for the development of tools that assist in both diagnosis and the control of vector spread. Leptospirosis, despite being a globally important disease, exhibits high levels of underreporting. It is a condition with low media visibility and receives limited investment from public authorities and pharmaceutical industries. In Brazil, one way to understand the dynamics of these diseases is through the monitoring of Health Systems, which compile notification forms and make them available anonymously for epidemiological studies. This thesis comprises five scientific articles with the following objectives, respectively: Evaluate the relationship between dengue cases and deaths and their correlation with social vulnerability; Train machine learning models for dengue case screening; Analyze the completeness of leptospirosis notification data in Rio Grande do Sul; Apply machine learning models to screen leptospirosis cases using patient identification data and clinical signs; Evaluate the spatiotemporal dynamics of dengue cases and Aedes aegypti eggs in Belo Horizonte. All studies utilized data from passive surveillance systems. Article 1: Dengue incidence and lethality rates and their relationship with gender, age, education, skin color, and social vulnerability were studied using chi-square tests, ordinary least squares (OLS) analyses, and geographically weighted regression (GWR). The results demonstrated a direct correlation between the increase in dengue cases and areas with higher social vulnerability and a larger population of working-age individuals. Additionally, the disease was found to be more severe in self-declared black individuals, the elderly, and males. Article 2: Data from the National Notifiable Diseases Information System for the states of Rio de Janeiro and Minas Gerais in 2016 and 2019 were used to train machine learning algorithms. Decision Trees and Multilayer Perceptron (MLP) algorithms achieved the best results in decision metrics, with an accuracy of 98%. Article 3: An ecological study assessed the completeness of leptospirosis data in municipalities in Rio Grande do Sul. It was observed that 79.29% of municipalities left some notification forms with the "final classification" field blank or filled with the number 8 (ignored). Information related to education, occupation, previous cases in humans or animals, and laboratory diagnosis showed the highest rates of incompleteness. Article 4: Data from leptospirosis notification forms in the state of Rio Grande do Sul from 2007 to 2019 were used. The chi-square technique was employed to select the most relevant attributes. The Random Forest classification model achieved an accuracy of 0.6638. Article 5: Data on the number of eggs and cases collected by surveillance in Belo Horizonte were analyzed. It was found that the Barreiro, Norte, and Nordeste districts presented risk clusters for cases and an increase in the number of eggs. In the evaluation of temporal trends, a decrease in cases was observed from 2016, but there was an increase in the number of eggs. These studies aim to contribute to the understanding and control of these diseases, assisting the surveillance system with effective tools for diagnosis, monitoring, and intervention.