Objective To develop an AutoML-based interpretable prediction model for blood transfusion requirements in severe traumatic brain injury (sTBI) patients, optimizing blood resource management through clinical-translational tools. Methods In this retrospective cohort study (January 2020–January 2025), 638 sTBI patients were enrolled. Random stratified sampling divided data into training ( n = 447) and testing ( n = 191) sets (7:3 ratio). We constructed an Automated Machine Learning (AutoML) framework using the Improved Hannibal Barca Optimizer (IHBO), which synchronously integrated LASSO feature selection verification and Shapley Additive exPlanations (SHAP) interpretability analysis. Model evaluation covered discriminative ability (AUC/PR-AUC), calibration performance (Brier score), and clinical utility (Decision Curve Analysis). Results The AutoML model demonstrated exceptional performance in the independent testing set, with ROC-AUC and PR-AUC values reflecting high predictive accuracy. It consistently outperformed comparator models across all metrics, including F1-score (0.8387), while DCA confirmed superior net benefit across clinically relevant thresholds. SHAP analysis identified nine key predictors hierarchically influencing transfusion risk: treatment type, GCS score, INR, K + , Ca2 + , Hct, age, hemorrhagic shock, and skull fracture. Conclusion This explainable AutoML framework successfully deciphers multidimensional determinants of sTBI transfusion needs. The clinically deployable interactive system eliminates technical barriers through intuitive nine-feature input, establishing new paradigm for trauma care decision-support and blood resource optimization.
Gong et al. (Wed,) studied this question.