Proposing an Intelligent Monitoring System for Early Prediction of Need for Intubation among COVID-19 Hospitalized Patients

  • Mohammad Reza Afrash Department of Medical Informatics, Student Research Committee, School of Allied Medical Sciences, Shahid Beheshti University of Medical Sciences, Tehran, Iran
  • Hadi Kazemi-Arpanahi Department of Health Information Technology, Abadan University of Medical Sciences, Abadan, Iran.
  • Raoof Nopour Department of Health Information Management, Student Research Committee, School of Health Management and Information Sciences Branch, Iran University of Medical Sciences, Tehran, Iran.
  • Elmira Sadat Tabatabaei Department of Genetics, Islamic Azad University, Tehran Medical Branch, Tehran, Iran.
  • Mostafa Shanbehzadeh Department of Health Information Technology, School of Paramedical, Ilam University of Medical Sciences, Ilam, Iran
Keywords: COVID-19, Coronavirus, Artificial Intelligence, Machine Learning, Intubation, Prognosis.

Abstract

Introduction: Predicting acute respiratory insufficiency due to coronavirus disease 2019 (COVID-19) can diminish the severe complications and mortality associated with the disease. This study aimed to develop an intelligent system based on machine learning (ML) models for frontline clinicians to effectively triage high-risk patients and prioritize who needs mechanical intubation (MI).

Materials and Methods: In this retrospective-design study, the data regarding 482 COVID-19 hospitalized patients from February 9, 2020, to July 20, 2021, was analyzed by six ML classifiers. The most critical clinical variables were identified by a minimal-redundancy-maximal-relevance (mRMR) feature selection technique. In the next step, the models' performance was assessed using confusion matrix criteria and, finally, the best model was adopted.

Results: Proposed models were implemented using 23 confirmed variables. Results of comparing six selected ML algorithms indicated the extreme gradient boosting (XGBoost) classifier with 84.7% accuracy, 76.5 % specificity, 90.7% sensitivity, 85.1% f-measure, 87.4% Kappa statistic, and 85.3% for receiver operating characteristic (ROC) had the best performance in the intubation prediction.

Conclusion: It is found that ML enables a satisfactory accuracy level in calculating intubation risk in COVID-19 patients. Therefore, using the ML-based intelligent models, notably the XGBoost algorithm, actually enables recognizing high-risk cases and advising correct therapeutic and supportive care by the clinicians.

Published
2022-09-20
Section
Articles