COVID-19 ICU mortality prediction: a machine learning approach using SuperLearner algorithm.

Area of Science:

• Critical Care Medicine
• Machine Learning in Healthcare
• Infectious Diseases Epidemiology

Background:

• The early stages of the COVID-19 pandemic highlighted the need for predictive models due to uncertainties in diagnosis, treatment, and prognosis.
• Developing reliable tools to forecast patient outcomes is crucial for effective resource allocation and clinical decision-making in intensive care units (ICUs).

Purpose of the Study:

• To develop and validate a machine learning model for predicting ICU mortality in COVID-19 patients.
• To identify key clinical parameters that significantly influence mortality risk in critically ill COVID-19 patients.

Main Methods:

• An observational multicenter cohort study enrolled adult COVID-19 patients admitted to 25 ICUs within the VENETO ICU network.
• A SuperLearner machine learning algorithm was employed for model development, utilizing clinical variables such as age, comorbidities, and organ support.
• Internal validation used a training set (n=1293), and external validation was performed on two independent test sets (n=124 and n=199).

Main Results:

• Three distinct predictive models were developed, demonstrating comparable predictive performance with training balanced accuracy ranging from 0.72 to 0.90.
• Cross-validation performance varied between 0.75 and 0.85, indicating robust model generalizability.
• Age emerged as the most influential predictor of ICU mortality across all developed models.

Conclusions:

• The study successfully developed a reliable machine learning tool for predicting ICU mortality in COVID-19 patients.
• Age is identified as the primary clinical variable impacting mortality risk, underscoring its importance in risk stratification.