A machine learning model trained on 150 patients with acute myocardial infarction showed high discrimination for new-onset atrial fibrillation (accuracy 97.0%, ROC-AUC 0.991), with echocardiographic, oxidative stress, NT-proBNP and revascularization variables as the top predictors, pending validation in larger multicenter cohorts.
The machine learning model achieved very high predictive performance for new-onset atrial fibrillation (AF) after acute myocardial infarction in the independent test set.
SHAP and permutation importance analyses identified a set of echocardiographic, biochemical, and procedural variables as the most influential predictors of new-onset AF.
The study combined clinical, laboratory, electrocardiographic, and echocardiographic data in a prospective cohort to build the prediction model.
The authors conclude that machine learning with SHAP explainability shows high potential for predicting new-onset AF after acute myocardial infarction, but requires validation in larger multicenter cohorts.
This study looked at whether a computer-based prediction method (machine learning) could identify which heart attack patients would go on to develop atrial fibrillation, an irregular heart rhythm that is linked to higher risk of death, heart failure, and further cardiovascular events. The researchers followed 150 patients who arrived at hospital within 24 hours of heart attack symptoms and collected clinical, blood, ECG, and heart ultrasound data. The model predicted new atrial fibrillation very accurately on a separate test set (97% accuracy, ROC-AUC 0.991). The factors that mattered most included measures of heart stiffness and structure on ultrasound (E/e' ratio, left ventricular mass, ejection fraction, left atrial volume), markers of oxidative stress in the blood, NT-proBNP (a heart strain marker), and whether blood flow was completely restored to the heart. This research suggests such models could eventually help doctors identify higher-risk patients early and tailor monitoring. However, the study was small (150 patients) and the authors themselves note the findings need to be validated in larger, multicenter groups before the approach can be relied on in practice.
Chiriliuc N, Corlateanu A, Arnaut O, Esanu I, David L, Bursacovschi D. (2026). Prediction of New-Onset Atrial Fibrillation in Patients with Acute Myocardial Infarction: A Machine Learning Approach.. Medical sciences (Basel, Switzerland). https://doi.org/10.3390/medsci14060626