Use of machine learning methods to predict Covid-19 prognosis in hospitalized patients: a systematic review

Abstract

Introduction: The COVID-19 pandemic has depleted human and financial resources from health systems. Thus, the use of artificial intelligence in patient care can be an effective strategy in the pandemic. Objective:To analyze the use of Machine Learning (ML) to predict death, ICU admission and use of mechanical ventilation (MV) in patients hospitalized with COVID-19. Method:Systematic review following PRISMA. Databases consulted: PUBMED, SCIELO, IEEE, COCHRANE, BVS and SCOPUS. Were included: primary studies; COVID-19 confirmed by RT-PCR; hospitalized patients; use of ML to predict one of the predefined prognoses. Simulations, studies of patients with specific comorbidities and studies without the number of patients were excluded. Results:18 studies were included in the review, with some articles analyzing the outcomes of death, ICU admission and use of MV separately, and others evaluating the combined outcomes. Among the articles, 22 values of area under the curve (AUC) were obtained, with the highest and lowest values: 1 and 0.66The ML techniques used clinical, laboratory and/or imaging criteria. Conclusion:The models used showed good results. ML can help predict the outcome of patients hospitalized with COVID-19, improving care and resource allocation.

Ähnliche Arbeiten