Emprego de técnicas de machine learning na predição de morte em pacientes hospitalizados com COVID-19: uma revisão sistemática

Abstract

COVID-19 pandemic has been consuming human and material resources from health systems, pointing the need to optimize these processes. Thus, artificial intelligence techniques emerge as potential allies to better manage care. Aim: Analyze the use of Machine Learning (ML) to predict death of hospitalized patients with COVID-19. Methods: Systematic review following PRISMA guidelines. Bases searched: PUBMED, COCHRANE, SCIELO, IEEE, SCOPUS and BVS. Were included primary studies analyzing hospitalized patients with COVID-19 confirmed by RT-PCR, in which ML was used to predict death prognosis. Were excluded simulations, studies containing patients with specific comorbidities and studies without number of patients. Results: 9 studies were included. Highest AUC (Area Under ROC curve) found was 1.00 and lowest 0.66. ML systems used clinical, laboratorial and/or imaging parameters. Conclusion: Models analyzed revealed great results and may help to predict the outcome of patients hospitalized with COVID-19, improving care and resources allocation.