Development and validation of a prognostic prediction model for patients with traumatic multiple fractures and hemorrhagic shock using an Automated Machine Learning framework: a retrospective cohort study

Objective To develop and validate a prognostic prediction model for patients with traumatic multiple fractures and hemorrhagic shock using an Automated Machine Learning (AutoML) framework, evaluating its predictive performance and clinical utility. Methods A total of 1,028 patients with traumatic multiple fractures and hemorrhagic shock admitted to the Emergency Departments and Intensive Care Units of seven public hospitals between January 2020 and December 2025 were retrospectively enrolled, with data from 4 hospitals designated as the training set ( n = 720) and data from 3 hospitals serving as the test set ( n = 308). Multidimensional data—including demographic characteristics, trauma/injury features, admission vitals/perfusion indices, and laboratory parameters—were extracted. The improved beaver behavior optimizer (IBBO) algorithm synchronously optimized feature subsets, base learners, and hyperparameter combinations. Clinical rationality of features was verified using LASSO regression and SHAP interpretability analysis. Results The IBBO algorithm demonstrated superior stability and outperformed the original BBO and comparative algorithms in most test functions. The AutoML model achieved best performance. The test set further confirmed its robustness, yielding a ROC-AUC of 0.9357 and PR-AUC of 0.9270. Decision curve analysis demonstrated that the AutoML model’s clinical net benefit surpassed that of traditional methods across a threshold range of 1–96%; The calibration curve likewise indicated high consistency between predicted probabilities and actual outcomes, with a Brier score as low as 0.110. SHAP analysis identified key predictors in descending order of importance: GCS score, ISS score, time from injury to ER admission, lactate, and fibrinogen. Conclusion The IBBO-based AutoML prognostic model provides an efficient, accurate tool for in-hospital mortality prediction in traumatic multiple fractures with hemorrhagic shock. The model identified that core predictors—including GCS score, ISS score, time to ER admission, lactate, and fibrinogen—critically influence in-hospital mortality outcomes. Clinical decision support software derived from this model offers visual, intelligent guidance for stratified care, promising utility in trauma emergency practice.