USING MACHINE LEARNING TO PREDICT DEATHS IN PATIENTS HOSPITALIZED FOR SARS: INTEGRATIVE REVIEW

Abstract

Machine Learning (ML) plays an important role in healthcare, providing data on patient diagnosis and prognosis through predictive models. The present work aimed to synthesize the available knowledge about the current applications of AM in predicting deaths due to Severe Acute Respiratory Syndrome (SARS). This integrative review was developed in six stages, in which data were extracted from the PubMed and Virtual Health Library databases between April and May 2023, using the PRISMA structure for documentation and the CASP tool for evaluating articles. The search strategy used returned 383 articles after excluding duplicates. After applying the inclusion and exclusion criteria, 54 articles were chosen for full reading, leaving 36 articles in the final sample. In the results, supervised ML techniques stood out, with the algorithms XGBoost, Random Forest, Logistic Regression and SVM (Support Vector Machine) showing promising results, with AUC-ROC above 80%. Among the variables most used in studies, demographic, clinical and laboratory data stand out. Finally, the lack of tools for practical application of models in hospital environments limits their use, as most studies focus only on comparing ML algorithms for model generation.