Predictive Analytics Using Machine Learning to Identify ART Clients at Health System Level at Greatest Risk of...

Area of Science:

• Public Health
• Machine Learning
• Artificial Intelligence

Background:

• Effective HIV/AIDS programming requires high patient retention in treatment to improve health outcomes and reduce transmission.
• Machine learning (ML) and artificial intelligence (AI) offer advanced capabilities to analyze complex datasets for predicting patient adherence.
• Identifying clients at risk of loss to follow-up (LTFU) enables targeted support interventions.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the risk of LTFU among antiretroviral therapy clients.
• To assess the performance of different ML classifiers in identifying high-risk patients for HIV/AIDS treatment.
• To determine the generalizability of predictive models across different demographic groups.

Main Methods:

• Binary classification techniques with temporal cross-validation were employed.
• Predictive models included logistic regression, neural networks, and tree-based models (Random Forest, boosted trees).
• Analysis utilized diverse data sources including health facility records, geospatial data, satellite imagery, and de-identified electronic medical records.

Main Results:

• Machine-learned models demonstrated strong predictive power for LTFU in both Nigeria and Mozambique.
• In Mozambique, a Random Forest model achieved an area under the precision-recall curve of 0.65 (LTFU rate 23%).
• In Nigeria, a boosted tree model achieved an area under the precision-recall curve of 0.52 (LTFU rate 27%).

Conclusions:

• ML models significantly outperformed traditional classification methods in predicting LTFU.
• These models can optimize the allocation of health worker resources to patients most in need.
• The models exhibited consistent performance across different sex and age groups, indicating broad applicability.