Related Experiment Video
Updated: Aug 20, 2025
12:18
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
7.6K
Risk Stratification of COVID-19 Using Routine Laboratory Tests: A Machine Learning Approach
Farai Mlambo1, Cyril Chironda1, Jaya George2,3
1School of Statistics and Actuarial Science, University of the Witwatersrand, 1 Jan Smuts Ave, Braamfontein, Johannesburg 2000, South Africa.
Infectious Disease Reports
|November 22, 2022
Summary
Machine learning models can predict severe COVID-19 cases using routine lab tests. Random Forest and Self-Normalising Neural Network showed the best performance for early patient risk stratification.
Area of Science:
- Medical Informatics
- Machine Learning in Healthcare
- Epidemiology
Background:
- The COVID-19 pandemic strained healthcare systems globally.
- Diagnostic delays, particularly with RT-PCR testing, complicated patient management.
- Effective risk stratification is crucial for prioritizing resources and patient care.
Purpose of the Study:
- To develop and evaluate machine learning models for classifying COVID-19 patients as severe or not severe.
- To utilize routine laboratory test results for predicting patient outcomes.
- To assess the feasibility of integrating ML into laboratory systems for early risk stratification.
Main Methods:
- Extracted data from 15,437 patients tested for SARS-CoV-2 between March and July 2020.
- Trained six machine learning models: Logistic Regression, Decision Trees, Random Forest, XGBoost, CNN, and SNN.
- Employed a 70:30 train-test split with 10-fold cross-validation; defined severe disease by ICU or high care admission.
Main Results:
- Random Forest (RF) achieved the highest sensitivity (75%), while Convolutional Neural Network (CNN) had the highest accuracy (75%).
- The Area Under the Curve (AUC) ranged from 57% (CNN) to 75% (RF).
- RF and SNN demonstrated superior performance in risk stratification compared to other models.
Conclusions:
- Machine learning models, particularly RF and SNN, show significant promise for early identification and risk stratification of COVID-19 patients.
- Integrating ML into laboratory information systems can enhance patient management, especially in resource-limited settings.
- Routine laboratory tests are valuable predictors for COVID-19 severity assessment.
