Machine learning methods to predict mechanical ventilation and mortality in patients with COVID-19

Area of Science:

• Artificial Intelligence in Medicine
• Computational Biology
• Epidemiology

Background:

• The COVID-19 pandemic caused a global health crisis, straining medical resources.
• High mortality rates associated with COVID-19 necessitate predictive tools for patient management.

Purpose of the Study:

• Develop and validate machine learning models to predict mechanical ventilation (MV) in emergency room (ER) patients.
• Predict in-hospital mortality for admitted COVID-19 patients.

Main Methods:

• Two cohorts were used: 1980 patients for MV prediction (XGBoost) and 3491 patients for mortality prediction (CatBoost).
• Data included demographics, medical history, vital signs, lab values, and treatments.
• Models were trained and validated using distinct patient groups.

Main Results:

• Older age, higher temperature, increased respiratory rate (RR), and lower oxygen saturation (SpO2) predicted MV need (86.2% accuracy).
• For mortality prediction, higher RR, BMI, and longer hospital stay were key indicators (80% accuracy).
• Both models demonstrated high accuracy and negative predictive values.

Conclusions:

• Machine learning models, specifically XGBoost and CatBoost, can accurately predict MV requirements and mortality in COVID-19 patients.
• These predictive capabilities can assist healthcare providers in resource allocation and patient care strategies.