Predictive Models of Mortality for Hospitalized Patients With COVID-19: Retrospective Cohort Study

Area of Science:

• Medical Informatics
• Epidemiology
• Machine Learning

Background:

• The COVID-19 pandemic necessitated accurate patient risk stratification for effective resource allocation.
• Predicting mortality in hospitalized COVID-19 patients is crucial for managing healthcare resources.

Purpose of the Study:

• To develop and validate machine learning models for predicting COVID-19 patient mortality.
• Utilize basic demographics and easily obtainable laboratory data for mortality prediction.

Main Methods:

• Retrospective analysis of 375 hospitalized COVID-19 patients.
• Trained logistic regression and support vector machine models on derivation and validation cohorts.
• Models utilized laboratory data from the entire hospital stay or within 12 hours of admission, externally validated on 542 patients.

Main Results:

• Models achieved up to 97% accuracy in predicting mortality.
• Models using early laboratory data (within 12 hours) predicted outcomes 11.5 days in advance with 93% accuracy.
• Key predictors included lactate dehydrogenase, C-reactive protein, lymphocyte percentage, and patient age.

Conclusions:

• Machine learning models effectively predict COVID-19 patient mortality with high accuracy.
• Early prediction of outcomes aids in hospital resource allocation and patient management.
• Age is a significant factor in COVID-19 mortality prediction.