Predicting the Need for Intubation among COVID-19 Patients Using Machine Learning Algorithms: A Single-Center Study.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Critical Care Medicine

Background:

• Shortage of mechanical ventilators (MV) necessitates accurate prognosis for critical COVID-19 patients.
• Objective prediction tools are vital for effective patient triage and resource management.
• Machine learning (ML) offers potential for developing such predictive models.

Purpose of the Study:

• To construct and evaluate machine learning models for predicting the need for mechanical ventilation in COVID-19 patients.
• To identify key clinical variables predictive of mechanical ventilator requirement.
• To assist frontline clinicians in prioritizing patients for MV.

Main Methods:

• Retrospective analysis of 482 COVID-19 patients.
• Application of multiple machine learning algorithms: multi-layer perception (MLP), logistic regression (LR), J-48 decision tree, and Naïve Bayes (NB).
• Identification of significant clinical features using Chi-square test (P < 0.01) and model performance evaluation using metrics like F-Score and AUC.

Main Results:

• Fifteen clinical variables were identified as significant predictors, including symptoms (cough, dyspnea), laboratory values (lymphocyte count, blood glucose), and comorbidities.
• The J-48 decision tree algorithm demonstrated the highest predictive accuracy, achieving an F-score of 0.868 and an AUC of 0.892.
• The model effectively predicted the requirement for mechanical intubation.

Conclusions:

• Machine learning algorithms can significantly enhance traditional clinical criteria for predicting intubation needs in hospitalized COVID-19 patients.
• ML-based prediction models can optimize intubation timing and improve the allocation of critical care resources.
• These tools support physicians in managing mechanical ventilator resources and personnel more effectively.