Interpretable Machine Learning in Predicting Drug-Induced Liver Injury among Tuberculosis Patients: Model Development and Validation Study

Abstract Background: This study aimed to develop and validate an interpretable prediction model for Drug-Induced Liver Injury during tuberculosis treatment. Methods: Using a dataset of TB patients from Ningbo City, the models were developed using eXtreme Gradient Boosting, random forest, and logistic regression algorithms. Features were selected using the Least Absolute Shrinkage and Selection Operator method. The model's performance was assessed through various metrics, including receiver operating characteristic and precision-recall curves. Calibration and clinical utility were also evaluated. Variable contributions were interpreted using SHapley Additive exPlanations and Partial Dependence plots. Results: Of 7,071 TB patients (median age: 47 years; 68.0% male), 16.3% developed DILI. Calibration showed minimal brier score differences among algorithms (0.003 to 0.004). XGBoost had the highest recall at 0.742, while random forest and logistic regression posted 0.675 and 0.649, respectively. All models demonstrated enhanced clinical utility in the validation set. SHAP analysis for XGBoost highlighted prior DILI instances as a significant risk. Elevated alanine aminotransferase ratios were linked to DILI in both XGBoost and random forest models. Conclusion: In conclusion, this study introduces an interpretable prediction model for assessing DILI risk among TB patients. The model's interpretability shed light on the significance of patients' disease history and ALT levels. This model holds potential for advancing personalized risk assessment and enhancing patient care in the context of TB treatment.