Prediction and Feature Importance Analysis for Severity of COVID-19 in South Korea Using Artificial Intelligence:...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Epidemiology

Background:

• COVID-19 has caused a significant increase in global mortality.
• Severe COVID-19 cases requiring invasive ventilation have a higher fatality rate.

Purpose of the Study:

• To identify factors associated with COVID-19 severity.
• To develop an artificial intelligence (AI) model for early prediction of COVID-19 severity.

Main Methods:

• An AI model was developed using data from 5601 COVID-19 patients in South Korea.
• A 5-layer deep neural network (DNN) was trained on 37 variables, selecting the 20 most important features.
• Feature importance was analyzed using AdaBoost, random forest, and XGBoost.

Main Results:

• Age was the most significant predictor of COVID-19 severity, followed by lymphocyte count, platelet count, and dyspnea.
• The AI model achieved high performance metrics: 90.4% accuracy, 90.2% sensitivity, 90.4% specificity, and 0.96 AUC.
• The model effectively categorizes patients into low and high severity groups.

Conclusions:

• The developed AI model accurately predicts COVID-19 severity.
• A web application is available for public access to the AI model.
• Sharing the AI model can facilitate validation and performance enhancement.