Comparing machine learning algorithms for predicting ICU admission and mortality in COVID-19

Area of Science:

• Medical Informatics
• Computational Biology
• Epidemiology

Background:

• Predicting the clinical course of COVID-19 is complex.
• Machine learning (ML) offers potential for identifying high-risk patients.
• Accurate risk stratification is crucial for resource allocation and patient management.

Purpose of the Study:

• To compare the performance of 18 ML algorithms for predicting ICU admission and mortality in COVID-19 patients.
• To identify key clinical variables for predicting these outcomes.
• To evaluate the utility of ML models in a real-world healthcare setting.

Main Methods:

• Utilized COVID-19 patient data from Mass General Brigham (MGB) Healthcare database.
• Developed and internally validated ML models on ED patients from March-April 2020 (n=3597).
• Externally validated models on ED patients from May-August 2020 (n=1711).

Main Results:

• Ensemble-based ML models demonstrated superior performance in predicting 5-day ICU admission and 28-day mortality.
• Key predictors for ICU admission included C-reactive protein (CRP), lactate dehydrogenase (LDH), and oxygen saturation.
• Important variables for mortality prediction were estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m², and neutrophil and lymphocyte percentages.

Conclusions:

• Ensemble ML models show promise for accurately predicting COVID-19 patient trajectories.
• These models can assist clinicians in identifying high-risk individuals, optimizing care.
• The findings support the implementation of ML tools for managing future infectious disease outbreaks.