Machine learning algorithms to predict treatment success for patients with pulmonary tuberculosis

Area of Science:

• Infectious Diseases
• Medical Informatics
• Artificial Intelligence

Background:

• Tuberculosis (TB) remains a global health threat despite treatment advances.
• Accurate prediction of treatment success in pulmonary TB (PTB) is crucial for patient outcomes.
• Machine learning (ML) offers potential for analyzing complex patient data to predict treatment response.

Purpose of the Study:

• To identify and compare effective ML algorithms for predicting treatment success in PTB patients.
• To evaluate the predictive performance of various ML models for sputum culture conversion.

Main Methods:

• A retrospective analysis of 1236 PTB patients from a clinical trial.
• Development and validation of multiple ML models including Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and Naïve Bayes (NB).
• Performance evaluation using metrics like accuracy, Area Under the Curve (AUC), precision, recall, and F1-score.

Main Results:

• All validated ML models achieved an AUC > 80%, indicating high performance.
• The Decision Tree (DT) model demonstrated superior performance with 92.72% accuracy and an AUC of 0.909.
• DT model achieved high precision (95.90%), recall (95.60%), and F1-score (95.75%).

Conclusions:

• The Decision Tree (DT) model is a highly effective algorithm for predicting treatment success in PTB patients.
• This ML-driven methodology can aid in precise classification and management of TB patients during treatment.
• Further application of this approach can improve patient monitoring and treatment strategies for tuberculosis.