[Construction of a prediction model for in-hospital major adverse cardiovascular events after emergency percutaneous coronary intervention in patients with ST-elevation myocardial infarction].

In-hospital MACE occurred in 524 of 5829 STEMI patients (about 9.0%) who underwent emergency PCI at a single center between 2015 and 2025.

• Patients were treated at the Emergency Critical Care Center of Beijing Anzhen Hospital, Capital Medical University, from January 1, 2015, to July 1, 2025.
• The MACE group had 524 cases and the non-MACE group had 5,305 cases.
• MACE comprised cardiogenic shock, severe heart failure, severe arrhythmia, acute stent thrombosis or recurrent acute MI, non-fatal stroke, and cardiogenic death.
• The percentage of MACE (about 9.0%) is calculated from the reported counts, not stated in the abstract.

The study cohort was middle-aged and predominantly male.

• Mean age was (58.1±12.3) years.
• 4,875 patients (83.63%) were male.

Univariate analysis and Lasso regression selected 14 variables for model construction.

• Selected variables: B-type natriuretic peptide, myoglobin, left ventricular ejection fraction, HDL cholesterol, high-sensitivity troponin I, triglycerides, triglyceride-glucose index, LDL cholesterol, glycated hemoglobin, creatine kinase-MB, smoking history, heart rate, Killip classification, and age.
• Candidate variables included demographics, lifestyle factors, past history, admission vital signs, laboratory results within 24 hours of admission, echocardiographic parameters, Killip class, and coronary interventional data.
• Patients were split 7:3 into a training set (4,080) and a test set (1,749).

Among five machine-learning algorithms, XGBoost had the best predictive performance for in-hospital MACE.

• The XGBoost model reached an AUC-ROC of 0.992 and an AUPRC of 0.947.
• Algorithms compared were SVM, logistic regression, random forest, naive Bayes, and XGBoost.
• The optimal model was chosen on AUC-ROC and AUPRC.

The XGBoost model showed favorable probability calibration and clinical net benefit.

• Calibration was assessed with calibration curves.
• Clinical net benefit was assessed with decision curve analysis.
• The model was internally validated using the test set.

SHAP analysis identified B-type natriuretic peptide, creatine kinase-MB, myoglobin, left ventricular ejection fraction, and triglyceride-glucose index as the five largest contributors to model predictions.

• The ranking, in descending order of contribution, was BNP, CK-MB, myoglobin, LVEF, and TyG index.
• SHAP was used to interpret the contribution of each feature.

The authors conclude the XGBoost model may serve as a reference for early in-hospital risk stratification and clinical decision-making in STEMI patients undergoing emergency PCI.

• The model was described as successfully constructed and internally validated.
• It was reported to have good discrimination, calibration, and clinical net benefit.