Artificial intelligence for predicting mortality in hospitalized COVID-19 patients

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Epidemiology

Background:

• COVID-19 pandemic significantly altered global demographics.
• Early identification of high-risk patients is crucial for effective management.
• Clinical and laboratory data within 72 hours of admission are key indicators.

Purpose of the Study:

• To develop a predictive model for COVID-19 associated mortality.
• To utilize machine learning for risk stratification of hospitalized COVID-19 patients.
• To provide a decision support tool for clinicians.

Main Methods:

• Analysis of 3024 PCR-confirmed COVID-19 patients admitted between May 2020 and August 2021.
• Application of five machine learning models and Boruta-SHAP for feature selection.
• Validation of model performance using Receiver Operating Characteristic Area Under the Curve (ROC AUC).

Main Results:

• Six point two five percent (6.25%) of patients experienced a fatal outcome.
• All machine learning models demonstrated high efficacy with ROC AUC > 80%.
• The random forest model with Boruta-SHAP features achieved a 93.1% ROC AUC in validation.

Conclusions:

• Machine learning models show high efficacy in predicting COVID-19 mortality.
• The developed model can serve as a valuable decision support system in clinical practice.
• Early data-driven risk assessment can improve patient outcomes and healthcare resource allocation.