Abstract:

Objective: Based on the clinical data of patients with retreatment pulmonary tuberculosis, this study systematically compared the performance of nine machine learning models in treatment outcome classification to construct the optimal prediction model and provide a predictive tool for optimizing the management of retreatment pulmonary tuberculosis patients. Methods: A total of 1396 patients with retreatment pulmonary tuberculosis registered and managed by the Kashgar in Xinjiang Uyghur Autonomous Region from January 1 to December 31, 2022 were enrolled. They were randomly divided into a training set (978 cases) and a test set (418 cases) at a 7∶3 ratio. Feature selection was performed using Random Forest and Cramér’s V coefficient. Nine machine learning models were constructed, including Logistic Regression, Support Vector Machine, Decision Tree, Random Forest, XGBoost, LightGBM, Gradient Boosting Tree, Multilayer Perceptron, and CatBoost. Model performance was evaluated using accuracy, precision, recall, F1-score, area under the curve (AUC), and average precision (AP). The optimal model was selected and interpreted using SHapley Additive exPlanations (SHAP) analysis. Results: The CatBoost model exhibited the best overall performance on the test set, with an accuracy of 89.2%, precision of 88.5%, recall of 89.2%, F1-score of 0.885, AUC of 0.829, and AP of 0.941. SHAP analysis revealed that 2-month sputum smear examination (absolute MeanSHAP value=0.595), treatment regimen (absolute MeanSHAP value=0.367), and treatment mode (absolute MeanSHAP value=0.290) were the three features with the highest contribution to predicting treatment outcomes in retreatment pulmonary tuberculosis. Conclusion: The CatBoost model performed robustly and excellently in predicting adverse treatment outcomes in patients with retreatment pulmonary tuberculosis. Combined with SHAP interpretation, it can effectively identify key predictors and provide an effective tool for risk assessment of adverse treatment outcomes in retreatment pulmonary tuberculosis patients.

Key words: Tuberculosis, pulmonary, Retreatment, Treatment outcome, Models, statistical, Forecasting