Predicting Antituberculosis Drug–Induced Liver Injury Using an Interpretable Machine Learning Method: Model Development and Validation Study

Background Tuberculosis (TB) is a pandemic, being one of the top 10 causes death and main cause from single source infection. Drug-induced liver injury (DILI) most common serious side effect during treatment TB. Objective We aim to predict status in patients with TB at clinical stage. Methods designed an interpretable prediction model based on XGBoost algorithm identified robust meaningful predictors risk TB-DILI basis data extracted Hospital Information System Shenzhen Nanshan Center for Chronic Disease Control 2014 2019. Results In total, 757 were included, 287 (38%) had developed TB-DILI. Based values relative importance area under receiver operating characteristic curve, machine learning tools selected patients’ recent alanine transaminase levels, average rate change last 2 measures cumulative dose pyrazinamide, ethambutol as best assessing validation set, precision 90%, recall 74%, classification accuracy 76%, balanced error 77% predicting cases The curve score upon 10-fold cross-validation was 0.912 (95% CI 0.890-0.935). addition, provided warnings high advance DILI onset median 15 (IQR 7.3-27.5) days. Conclusions Our shows interpretability TB-DILI, which can provide useful information clinicians adjust medication regimen avoid more patients.