Rapid Assessment of COVID-19 Mortality Risk with GASS Classifiers.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Epidemiology

Background:

• Effective triage of COVID-19 patients is crucial for resource allocation.
• Existing risk prediction models often lack accuracy, are costly, or are difficult to interpret.
• There is a need for accessible and reliable COVID-19 mortality risk prediction tools.

Purpose of the Study:

• To develop and validate novel classification methods for predicting COVID-19 mortality risk.
• To assess the performance of machine learning (SVM22-GASS) and clinical expertise-based (Clinical-GASS) models.
• To demonstrate the utility of routine clinical variables for accurate risk prediction.

Main Methods:

• Developed and validated two classifiers: SVM22-GASS (Support Vector Machine) and Clinical-GASS (General Assessment of SARS-CoV-2 Severity score).
• Utilized a derivation cohort (499 patients) and an independent validation cohort (250 patients).
• Employed routine clinical variables available early in hospital admission.

Main Results:

• SVM22-GASS achieved an AUC of 0.87 and accuracy of 0.88 in the validation cohort.
• Clinical-GASS achieved an AUC of 0.77 and accuracy of 0.78 in the validation cohort.
• Both models demonstrated high accuracy and interpretability compared to existing methods.

Conclusions:

• Accurate COVID-19 mortality risk prediction is feasible using readily available routine clinical variables.
• SVM22-GASS and Clinical-GASS offer promising tools for improved patient triage and management.
• The findings support the integration of these models into early hospital admission protocols.