Predicting Treatment Interruption Among People Living With HIV in Nigeria: Machine Learning Approach

Area of Science:

• Public Health
• Health Informatics
• Machine Learning in Medicine

Background:

• Antiretroviral therapy (ART) has shifted HIV management to a chronic disease model.
• High rates of treatment interruptions necessitate effective adherence and re-engagement strategies.
• Current interventions for ART adherence and re-engagement are often resource-intensive and may not be sustainable.

Purpose of the Study:

• To develop and integrate a machine learning (ML) model for predicting 30-day interruption in treatment (IIT) among new ART enrollees in Nigeria.
• To assess health workers' perceptions and utilization of the ML model's outputs for patient case management.

Main Methods:

• Trained and tested ML models (boosting tree, Extreme Gradient Boosting) using routine program data (2005-2021).
• Utilized an 80/20 data split for training and testing, with preselected variables associated with IIT.
• Defined IIT as failure to refill ART within 28 days of the scheduled follow-up date.

Main Results:

• Analysis included 136,747 clients; IIT rates decreased from 58.6% pre-2017 to 14.2% post-October 2019.
• Factors associated with IIT included illness severity at enrollment, pregnancy, breastfeeding, and facility characteristics.
• The selected ML model achieved 81% sensitivity, 88% specificity, 83% PPV, and 87% NPV, and was integrated into the electronic medical records system.

Conclusions:

• High-performing ML models for predicting HIV IIT can be developed from routine data and integrated into health management information systems.
• ML enhances intervention targeting through differentiated care models, improving cost-effectiveness and patient outcomes before treatment interruption occurs.