Related Experiment Video
Updated: Aug 5, 2026
19:57
An Affordable HIV-1 Drug Resistance Monitoring Method for Resource Limited Settings
Published on: March 30, 2014
Predicting HIV viral non-suppression in Uganda: development and validation of machine learning and risk
Maria Magdalene Namaganda1,2,3,4, Stathis Gennatas2, Laura Merson3
1Department of Immunology and Molecular Biology, School of Biomedical Sciences, College of Health Sciences, Makerere University, Kampala, Uganda.
Frontiers in Artificial Intelligence
|July 31, 2026
Summary
Machine learning models using electronic medical records can predict viral non-suppression in HIV patients in Uganda. This proactive approach identifies at-risk individuals for timely intervention, improving HIV care outcomes.
Area of Science:
- Machine learning in public health
- HIV/AIDS research
- Predictive modeling in healthcare
Background:
- Viral non-suppression is a critical risk in HIV care, often detected late.
- Electronic medical records (EMR) in resource-limited settings are underutilized for risk stratification.
- Proactive identification of viral non-suppression is needed for timely intervention.
Purpose of the Study:
- Develop and validate machine learning models for predicting viral non-suppression.
- Utilize routine EMR data for risk stratification in HIV care.
- Enhance proactive HIV management in resource-limited settings.
Main Methods:
- Developed and validated four models: logistic regression, elastic net, random forest, and extreme gradient boosting.
- Used 20 EMR variables from a large Ugandan HIV cohort (n=33,384) from 2014-2024.
- Evaluated models on an 80:20 test set using PR-AUC, ROC-AUC, Brier score, and decision curve analysis.
Main Results:
- Random forest and elastic net models showed comparable performance (PR-AUC ~0.24).
- Models achieved significant risk enrichment (approx. 5-fold) at a capacity-first threshold.
- Key predictors included ART class, adherence, TB history, and male sex.
Conclusions:
- Routine EMR data can effectively predict viral non-suppression risk in Ugandan HIV programs.
- Machine learning models demonstrate clinical utility for proactive HIV care.
- External validation and workflow integration are necessary for deployment.
