Predicting immune reconstitution after antiretroviral therapy in HIV/AIDS using ensemble machine learning: a...

Area of Science:

• Immunology
• Computational Biology
• Data Science

Background:

• Accurate prediction of long-term CD4+ T-cell recovery in people living with HIV (PLWH) on antiretroviral therapy (ART) is essential for personalized monitoring.
• Traditional statistical models struggle with the complex, non-linear longitudinal data common in HIV clinical studies.

Purpose of the Study:

• To develop a robust framework for predicting longitudinal CD4+ T-cell count, CD8+ T-cell count, and CD4/CD8 ratio in PLWH on ART.
• To create a tool that aids in personalized monitoring and treatment optimization for HIV care.

Main Methods:

• Developed a heterogeneous stacking ensemble framework integrating XGBoost, LightGBM, Random Forest, and Gradient Boosting with a Ridge regression meta-learner.
• Trained and tested the model on a retrospective cohort of 5,436 patients, excluding baseline CD4+/CD8+ counts to prevent data leakage.
• Utilized only demographic and clinical features for model training and prediction.

Main Results:

• The ensemble model achieved high predictive accuracy on an independent test set: R² of 0.768 for CD4+ count and 0.636 for CD8+ count.
• Demonstrated significant relative improvements in prediction accuracy compared to a baseline Robust Transformer model (66.4% for CD4+, 128.6% for CD8+).
• The model accurately replicated outcome distributions and showed stable learning dynamics without overfitting.

Conclusions:

• The developed ensemble learning framework is a robust and clinically applicable tool for forecasting immune reconstitution in HIV care.
• This approach offers a foundation for data-driven clinical decision support, enabling personalized long-term treatment monitoring.
• The model's ability to predict immune markers without baseline immunological data enhances its utility for personalized HIV management.