IFN-γ and IL-6 as Key Predicting Biomarkers for Active TB Among PLWH: Results from Four Machine Learning Methods.

Area of Science:

• Infectious Diseases
• Immunology
• Biomedical Data Science

Background:

• Tuberculosis (TB) is a leading cause of death in people living with HIV (PLWH).
• Early and accurate diagnosis of active TB in PLWH is critical but remains a significant challenge.
• Biomarker discovery and advanced computational methods are needed to improve diagnostic capabilities.

Purpose of the Study:

• To identify novel combinations of biomarkers for predicting active TB in PLWH.
• To develop and evaluate machine learning (ML) models for active TB prediction.
• To assess model performance on both randomly and chronologically split datasets.

Main Methods:

• Enrolled 760 PLWH with pulmonary symptoms, collecting demographic, clinical, and cytokine data.
• Developed four ML models using 10-fold cross-validation, feature selection, and hyperparameter optimization.
• Evaluated models using ROC-AUC, sensitivity, specificity, and variable importance on random and chronological data splits.

Main Results:

• Interferon-gamma (IFN-γ) and Interleukin-6 (IL-6) levels were significantly elevated in active TB patients.
• A gradient boosting machine (GBM) model utilizing IFN-γ and IL-6 achieved high AUC on random datasets (0.96 training, 0.73 test).
• The GBM model showed limited generalizability, with lower AUC (0.66) on a chronologically ordered test set.

Conclusions:

• Elevated IFN-γ and IL-6 may indicate TB activation in PLWH.
• Current ML models, while effective on random splits, face limitations in predicting active TB in chronologically ordered, real-world data.
• Further research is needed to enhance the generalizability of ML-based diagnostic tools for TB in PLWH.