An interpretable machine learning approach to predict left ventricular aneurysm formation following primary PCI in STEMI: A retrospective study with independent external validation.
BACKGROUND Left ventricular aneurysm (LVA) remains a clinically important structural complication after primary percutaneous coronary intervention (pPCI) in patients with ST-segment elevation myocardial infarction (STEMI). This study aimed to develop and externally validate an interpretable model for predicting LVA after pPCI. METHODS We retrospectively included 1507 patients from the development center and 535 patients from an external validation center. Least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm were used for feature selection. Eight routinely available predictors were retained: albumin (ALB), N-terminal pro-B-type natriuretic peptide (NT-proBNP), C-reactive protein (CRP), Killip class ≥2, left ventricular ejection fraction (LVEF), Gensini score, lactate dehydrogenase (LDH), and infarct-related artery involving the left anterior descending artery (IRA-LAD). These variables were incorporated into eight machine learning algorithms. Model performance was evaluated using discrimination, calibration, decision curve analysis (DCA), and classification metrics. SHapley Additive exPlanations (SHAP) was used for model interpretation, and a web-based calculator was developed. RESULTS Logistic regression showed the most favorable balance between performance and interpretability. The model achieved an area under the receiver operating characteristic curve (AUC) of 0.948 (95% confidence interval [CI], 0.897-0.997) in internal validation and 0.950 (95% CI, 0.908-0.992) in external validation. Calibration and DCA showed acceptable agreement and clinical net benefit. SHAP analysis identified LVEF, NT-proBNP, Killip class ≥2, and CRP as major predictors. CONCLUSIONS An interpretable model using routine clinical variables was developed and externally validated for predicting LVA after pPCI in STEMI patients. This model may support individualized risk assessment and follow-up planning.