Clinical prediction model for the risk of bleeding during hospitalization in patients with acute myocardial infarction: a retrospective cohort study from the MIMIC-IV database

Background Bleeding is a serious and potentially life-threatening complication in patients with acute myocardial infarction (AMI), significantly affecting patient outcomes. Early identification of high-risk patients is critical for reducing complications, improving outcomes, and guiding clinical decision-making. Objective This study aims to develop and validate a machine learning (ML)-based model to predict the risk of in-hospital bleeding events in AMI patients, identify key risk factors, and assess the clinical applicability of the model for risk stratification and decision support. Methods This study retrieved data from 2,646 AMI patients in the MIMIC-IV database, with missing data imputed using KNN. LASSO regression was used to screen the imputed data, and eight machine learning models were constructed based on the selected clinical features. Model performance was evaluated using ROC curves, calibration curves, and SHAP interpretability analysis, while clinical net benefit was assessed using decision curves. Results Among the 2,646 patients with acute myocardial infarction (AMI), the bleeding group had a higher proportion of males (73.0% vs. 64.6%), coagulation disorders (PT 1.6 vs. 1.2, APTT 77.7 vs. 77.0), diabetes (43.1% vs. 32.6%), and CABG treatment rate (84.7% vs. 10.3%) were significantly higher in the bleeding group than in the non-bleeding group (all P < 0.001); age, CABG treatment, IABP use, white blood cell count (WBC), coagulation function (PT, APTT), direct bilirubin (DBIL), platelets (Plt), ALT, and clopidogrel use. Among the eight machine learning models constructed, XGBoost demonstrated the best overall performance, with an AUC of 0.925 (accuracy 0.925, F1 score 0.802). Its decision curve showed the highest net benefit in the threshold range of 0.1–0.2, and the calibration curve indicated that the predicted risk was highly consistent with the actual risk (R² = 0.817). SHAP analysis revealed that CABG treatment and IABP use were the strongest risk factors, while elevated platelet counts were associated with reduced risk. The XGBoost model based on multidimensional features can accurately predict the risk of bleeding during hospitalization for AMI (sensitivity 0.849, specificity 0.884). Its high accuracy and interpretability provide a reliable tool for clinical dynamic decision-making, particularly for personalized optimization of treatment strategies. Conclusion The ML-based XGBoost model provides a reliable and clinically applicable tool for predicting bleeding events during hospitalization in AMI patients. This model combines high accuracy and interpretability, providing a quantitative tool for clinical dynamic assessment of anticoagulation therapy risks. Future multi-center validation can further optimize its application value in personalized treatment decision-making.