Data-Driven Prediction for COVID-19 Severity in Hospitalized Patients

Area of Science:

• * Medical Informatics
• * Machine Learning in Healthcare
• * Infectious Disease Epidemiology

Background:

• * Clinicians require stable tools for predicting COVID-19 severity to optimize hospital resource allocation.
• * Evolving COVID-19 guidelines and limitations in early prediction models hinder effective clinical decision-making.
• * Existing models often lack generalizability and clinical validation in real-world hospital settings.

Purpose of the Study:

• * To develop and validate a real-time COVID-19 severity prediction tool for hospitalized patients.
• * To assist clinicians in resource management and patient care during COVID-19 surges.
• * To provide a data-driven framework for predicting COVID-19 severity at admission using comprehensive clinical data.

Main Methods:

• * Evaluation of four machine learning models using a large dataset (1386 patients, March 2020-April 2021).
• * Utilization of comprehensive patient-level clinical data from electronic medical records, vital sign monitors, and PCR tests.
• * Development of a multi-class, data-driven framework by clinical and data experts.

Main Results:

• * The random forest model achieved high discrimination in concurrent validation (AUC 0.83-0.87).
• * Prospective validation demonstrated promising performance with recall ranging from 78.4-90.0% and precision from 75.0-97.8% across severity classes.
• * The developed framework effectively predicts COVID-19 severity at the time of hospital admission.

Conclusions:

• * The proposed machine learning framework offers a reliable tool for predicting COVID-19 severity in hospitalized patients.
• * This tool can significantly improve the management of healthcare resources during pandemic surges.
• * The study highlights the potential of leveraging comprehensive EMR data for real-time clinical decision support.