Temporal Knowledge Discovery in Drug-Resistant Tuberculosis: A Decade-Long Machine Learning Analysis From Egyptian...

Area of Science:

• Medical Informatics
• Computational Biology
• Epidemiology

Background:

• Tuberculosis (TB) remains a global health challenge, exacerbated by multidrug-resistant TB (MDR-TB).
• Traditional diagnostic methods for TB are often slow and lack precision.
• There is a critical need for advanced predictive tools to combat TB and MDR-TB.

Purpose of the Study:

• To develop and validate machine learning (ML) models for identifying risk factors of drug-resistant tuberculosis (DR-TB).
• To analyze clinical and longitudinal treatment-history data from Egyptian patients (2012-2022).
• To differentiate DR-TB from drug-susceptible TB (DS-TB) using AI-driven insights.

Main Methods:

• A retrospective case-control study involving 1,462 patients across three Egyptian respiratory hospitals.
• Data included demographics, clinical history, lifestyle factors, and drug resistance patterns (pDST, Xpert MTB/RIF).
• Four ML models (Random Forest, XGBoost, KNN, Neural Networks) were trained and validated using SHAP for feature selection and cross-validation.

Main Results:

• The Random Forest model achieved the highest performance with 94.54% accuracy and 94.46% ROC-AUC.
• XGBoost also demonstrated strong performance with 93.17% accuracy.
• Key predictors included prior first-line drug use, patient category, treatment history, geographic region, and radiological findings.

Conclusions:

• Machine learning models demonstrate high efficacy in distinguishing DR-TB from DS-TB using longitudinal clinical data.
• Integrating treatment history and geographic risk factors with AI provides a robust framework for early DR-TB prediction.
• This AI-driven approach can optimize treatment initiation and resource allocation in TB high-burden settings.