Automated machine learning model to predict anti-tuberculosis drug-induced liver injury in patients with tuberculous...

Tuberculous meningitis (TBM) is a severe central nervous system infection with high disability and mortality rates. However, during TBM treatment, anti-tuberculosis drug-induced liver injury (ATB-DILI) often precipitates treatment interruption and contributes to poor clinical outcomes. This study aims to develop an automatic machine learning (AutoML) model for predicting the risk of ATB-DILI in TBM patients.

A retrospective cohort study was conducted in adult TBM patients. After feature selection via least absolute shrinkage and selection operator (LASSO) regression, AutoML was employed to construct predictive models. The bootstrap resampling method was used for internal validation. Furthermore, feature importance, partial dependence plots, and SHapley Additive exPlanations (SHAP) analysis were utilized for model interpretation.

A total of 253 TBM patients were included in this study, of whom 54 (21.34%) developed ATB-DILI. LASSO regression analysis identified four characteristic factors, including cerebrospinal fluid (CSF) chloride, blood platelet, hypertension, and total bilirubin. Among the candidate models generated by AutoML, the optimal gradient boosting machine (GBM) model demonstrated superior performance. It achieved an optimism-corrected area under the receiver operating characteristic curve (AUC) of 0.828 (95% CI: 0.794-0.861) and an optimism-corrected area under the precision-recall curve (PR-AUC) of 0.711 (95% CI: 0.628-0.782). In addition, interpretability analysis revealed that CSF chloride was the most important variable for the optimal GBM model.

The ATB-DILI prediction model, developed using AutoML technology, demonstrated high predictive ability and interpretability. It can assist clinicians in identifying TBM patients at risk of ATB-DILI, thereby optimizing patient management and facilitating the formulation of personalized medication regimens.