Machine learning models to predict the maximum severity of COVID-19 based on initial hospitalization record

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Public Health

Background:

• The prolonged global spread of coronavirus disease 2019 (COVID-19) necessitates early prediction of maximum patient severity.
• Effective treatment strategies for COVID-19 depend on accurate prognostication of disease progression.

Purpose of the Study:

• To develop and validate artificial intelligence (AI) and machine learning (ML) models for predicting the maximum severity of hospitalized COVID-19 patients.
• To create user-friendly clinical tools based on these predictive models.

Main Methods:

• Utilized medical records of 2,263 COVID-19 patients from 10 hospitals in Daegu, Korea.
• Developed predictive models using initial hospitalization data, employing both 4-group and binary classification approaches.
• Evaluated model performance using AUC values, with three binary classification models showing high predictive power.

Main Results:

• The moderate severity group constituted the largest patient cohort (49.5%).
• Three binary classification models (Mild vs. Above Moderate, Below Moderate vs. Above Severe, Below Severe vs. Critical) demonstrated high predictive performance with AUCs of 0.883, 0.879, and 0.887, respectively.
• Statistically significant predictors for severity exacerbation were identified.

Conclusions:

• Successfully developed and deployed web-based nomograms for predicting maximum COVID-19 severity.
• These nomograms are expected to significantly assist clinicians in tailoring effective treatment plans for individual patients.