Título: Predicting Hospitalization Risk of Suspected COVID-19 Patients Using a Machine Learning Methodology
Autores: Alesson M. Torres, Amanda de A B Silva and Karla Figueiredo
Resumo:
New SARS-CoV-2 variants and the delay in immunization hinders COVID-19 mitigation. The pressure on healthcare systems caused by the disease has been a challenge for hospital managers. Thus, rapid assessment of hospitalization risk can support the management of hospital resources, increasing treatment opportunities for patients at higher risk, especially in case of high demands for hospitalizations. In this retrospective study, we propose a hospitalization prediction model based on classical Machine Learning algorithms: Logistic Regression (LR), k-Nearest Neighbors (KNN), Decision Tree (DT), and Support Vector Classifier (SVC). Our proposed classifier combines the results of the two algorithms, which have the best accuracies in order to reduce true-positive misclassification. We included 6,967 COVID-19 patients from São Paulo, Brazil, who had been admitted at the Hospital Sírio Libanês from February 26th to December 28th, 2020. The achieved accuracy was 90.4%, with only 7% of false-negative patients (NPV=0.82). Our results suggest that the prediction model proposed can be an efficient tool in screening COVID-19 hospitalization risk patients.
Palavras-chave:
COVID-19 Diagnosis, Machine Learning, Hospitalization Risk.
Páginas: 6
Código DOI: 10.21528/CBIC2021-146
Artigo em pdf: CBIC_2021_paper_146.pdf
Arquivo BibTeX: CBIC_2021_146.bib