Machine Learning Predictions for HIV Patient Loss to Follow-Up in Haiti, 2018-2024

From 2018 to 2024, this dataset contains analysis for predicting loss to follow-up among 115,822 people living with HIV on antiretroviral therapy across 167 health facilities in Haiti. It was authored by Jiaqin Wu and uses national electronic medical record data to train machine learning models. The study assessed model performance over four quarters using metrics like F2-score and ROC-AUC.

Use Cases

• Training predictive models for patient retention based on demographic, clinical, and institutional predictors mentioned in the description.
• Identifying key risk factors for loss to follow-up based on features like prior visit frequency and viral load test patterns.
• Evaluating machine learning model performance for time-series clinical forecasting across multiple quarters.
• Supporting the development of tiered intervention strategies in HIV care systems using individual risk scores.

Strengths

• Contains data from 115,822 patient records, providing substantial scale for analysis.
• Covers a 6-year time range from 2018 to 2024 across 167 health facilities.
• Uses multiple performance metrics (F2-score, recall, precision, ROC-AUC, PR-AUC, calibration) for robust model evaluation.
• Includes model interpretation via SHAP values to explain key predictors.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying dataset is unknown, which may limit suitability assessment.
• The primary file is a 16.0 KB DOCX document, suggesting the dataset itself may be a summary or analysis table rather than the raw patient data.

Provenance

Source

Haiti’s national electronic medical record database.

Collection Method

Analysis of routine electronic medical record data for people living with HIV receiving antiretroviral therapy.

Time Range

2018 to 2024

Freshness

Last updated 2026-04-28 05:01:17; freshness should be verified.

Geography

Haiti