Machine Learning Using Clinical and Cardiac MRI Features to Predict Long-term Outcomes in Acute STEMI

Background Current risk stratification models fail to effectively integrate a broad range of parameters to predict major adverse cardiovascular events (MACE) in patients with ST-segment elevation myocardial infarction (STEMI). Purpose To develop and externally test a machine learning (ML) model integrating comprehensive clinical data and cardiac MRI parameters for long-term MACE prediction in patients with STEMI. Materials and Methods This retrospective study included data from patients with STEMI who underwent clinically indicated cardiac MRI within 7 days after percutaneous coronary intervention, with data from one center composing the training set (September 2015 to September 2023) and data from another center composing the external test set (January 2015 to July 2023). The primary end point was MACE, defined as a composite of cardiovascular death, recurrent myocardial infarction, unplanned coronary revascularization, stroke, and rehospitalization for heart failure or arrhythmia. Sixty-seven variables were initially evaluated to inform the ML model. The final model included established clinical predictors combined with features selected using recursive feature elimination. Model performance was assessed using integrated area under the receiver operating characteristic curve (AUC). Results A total of 1066 patients with STEMI (mean age, 58.15 years ± 11.40 [SD]; 904 male patients) were included: 682 in the training set and 384 in the external test set. During a median follow-up of 40 months (IQR, 22-55 months), 142 patients in the training set and 81 in the external test set experienced MACE. In the external test set, the ML model achieved an integrated AUC for MACE prediction of 0.91, compared with an integrated AUC of 0.86 for a clinical model (P < .001), 0.86 and 0.89 (P = .005) for Cox regression models, 0.66 for Global Registry of Acute Coronary Events score (P < .001), and 0.62 for Thrombolysis in Myocardial Infarction score. The model effectively stratified patients into distinct risk groups (log-rank P < .001). Conclusion An ML model integrating cardiac MRI and clinical data demonstrated excellent long-term prognostic performance compared with traditional models and aided individualized risk stratification in patients with STEMI. © RSNA, 2026 Supplemental material is available for this article. See also the editorial by Garot and Duhamel in this issue.