Repository of Research and Investigative Information

Repository of Research and Investigative Information

Ilam University of Medical Sciences

Comparison of Machine Learning Tools for the Prediction of ICU Admission in COVID-19 Hospitalized Patients

Wed May 22 09:24:12 2024

(2022) Comparison of Machine Learning Tools for the Prediction of ICU Admission in COVID-19 Hospitalized Patients. Shiraz E Medical Journal. ISSN 17351391 (ISSN)

Full text not available from this repository.

Official URL: https://www.scopus.com/inward/record.uri?eid=2-s2....

Abstract

Background: The rapid coronavirus disease 2019 (COVID-19) outbreak has overwhelmed many healthcare systems worldwide and put them at the edge of collapsing. As the capacity of intensive care units (ICUs) is limited, deciding on the proper allocation of required resources is crucial. Objectives: This study aimed to create a machine learning (ML)-based predictive model of ICU admission among COVID-19 in-hospital patients at the initial presentation. Methods: This retrospective study was conducted on 1225 laboratory-confirmed COVID-19 hospitalized patients during January 9, 2020-January 20, 2021. The top clinical parameters contributing to COVID-19 ICU admission were identified based on a correlation coefficient at P-value < 0.05. Next, the predictive models were constructed using five ML algorithms. Finally, to evaluate the perfor-mances of models, the metrics derived from the confusion matrix, classification error, and receiver operating characteristic were calculated. Results: Following feature selection, a total of 11 parameters were selected as the top predictors to build the prediction models. The results showed that the best performance belonged to the random forest (RF) algorithm with the mean accuracy of 99.5, mean specificity of 99.7, mean sensitivity of 99.4, Kappa metric of 95.7, and root mean squared error of 0.015. Conclusions: The ML algorithms, particularly RF, enable a reasonable level of accuracy and certainty in predicting disease progres-sion and ICU admission for COVID-19 patients. The proposed models have the potential to inform frontline clinicians and health authorities with quantitative tools to assess illness severity and optimize resource allocation under time-sensitive and resource-constrained situations. © 2022, Author(s).

Item Type: Article
Creators:
CreatorsEmail
Shanbehzadeh, M.UNSPECIFIED
Haghiri, H.UNSPECIFIED
Afrash, M. R.UNSPECIFIED
Amraei, M.UNSPECIFIED
Erfannia, L.UNSPECIFIED
Kazemi-Arpanahi, H.UNSPECIFIED
Keywords: Coronavirus COVID-19 Decision Support Systems Intensive Care Unit Machine Learning alanine aminotransferase albumin alkaline phosphatase aspartate aminotransferase bilirubin C reactive protein calcium creatinine glucose hemoglobin lactate dehydrogenase magnesium phosphorus potassium sodium troponin absolute lymphocyte count absolute neutrophil count activated partial thromboplastin time adult aged ageusia alcohol consumption anosmia Article artificial neural network cardiovascular disease chill controlled study coronavirus disease 2019 correlation coefficient coughing cross validation cross-sectional study decision tree diabetes mellitus diagnostic accuracy diagnostic test accuracy study dyspnea erythrocyte count erythrocyte sedimentation rate female fever gastrointestinal disease headache hematocrit hospital admission hospitalization human hypertension k nearest neighbor leukocyte count lung contusion major clinical study male myalgia nausea and vomiting oxygen therapy platelet count pneumonia predictive model prothrombin time random forest receiver operating characteristic retrospective study reverse transcription polymerase chain reaction rhinorrhea sensitivity and specificity smoking sore throat support vector machine urea nitrogen blood level
Divisions:
Journal or Publication Title: Shiraz E Medical Journal
Journal Index: Scopus
Volume: 23
Number: 5
Identification Number: https://doi.org/10.5812/semj.117849
ISSN: 17351391 (ISSN)
Depositing User: مهندس مهدی شریفی
URI: http://eprints.medilam.ac.ir/id/eprint/4124

Actions (login required)

View Item View Item