Modelos de aprendizado de máquina para predição de óbito de pacientes em hemodiálise

Abstract

In Brazil, 133,464 people with Chronic Kidney Disease (CKD) are on hemodialysis with mortality of 19.5% per year. The mandatory biomarkers for monitoring these patients are determined by the 2014 Ministry of Health Clinical Guideline. More than thirty biomarkers are evaluated annually, with a third being assessed monthly. No critical assessment of the predictive use value of these biomarkers has been carried out in the country using Machine Learning (ML). Therefore, the present project aims to develop AM models that allow predicting the outcome of death in hemodialysis patients based on routine biomarkers. The aim is to generate monitoring instruments that make it possible to evaluate the predictive effectiveness of clinical examinations in order to improve patients’ quality of life and contribute to the cost management of the Unified Health System (SUS). In this work, data from incident hemodialysis patients from a retrospective cohort study carried out from 2012 to 2019 in 23 dialysis units in five Brazilian states are used. The features used in the development of the models are biomarkers, patient profile variables and clinical outcomes. Different ML approaches and algorithms will be tested using the Decision Tree, Random Forest, Logistic Regression and XGBoost ML models in order to identify the best predictive model.