Uso de machine learning para previsão da evolução de casos de SRAG incluindo casos de COVID-19 considerando variáveis clínicas e demográficas

Abstract

This work presents an exploratory analysis of the public database about Severe Acute Respiratory Syndrome (SARS) and the use of supervised learning methods to predict the evolution of patients' cases to death or cure. Two databases were created to compare the results and select relevant variables by the classification models, one containing the clinical variables and the other containing clinical and demographic variables. The classifiers used were k-nearest neighbors (KNN), Decision Trees, Logistic Regression, Random Forests, Neural Networks and XGBoost. XGBoost had the best area under the curve (AUC) result: 0.86 and the main attributes for classification were the use of invasive ventilation support, age over 70 years and admission to the ICU, the main demographic attributes were living in the northeast and north regions. The database with demographic variables presented results slightly superior to the results obtained with the clinical variables alone and allowed us to understand which sociodemographic characteristics are more relevant in the outcome of SARS cases.