Abstract 15027: Predicting Recurrence of Myocardial Infarction in Post-PCI Patients Using Machine Learning

Abstract

Introduction: While prevention of recurrent myocardial infarction (MI) is important in the management of patients after percutaneous coronary intervention (PCI), risk stratification of MI remains a significant challenge with increasingly diverse and complex patient backgrounds. This retrospective observational study aimed to explore whether machine learning (ML) methods can predict recurrent MI after PCI using a variety of features extracted from electronic medical records (EMRs). Methods: Seven machine learning algorithms were retrospectively applied to predict recurrent MI within two years after PCI (124 cases, 4.2%) from a total of 2984 PCI cases. The imbalance in the proportion of positive and negative data was resolved by introducing misclassification costs. Basic patient information (e.g., age, sex, risk factors) and laboratory test results were used for the analysis. Feature importance was calculated to identify key factors for MI recurrence. Five-fold cross-validation was used to calculate accuracy, sensitivity, specificity, positive predictive value, and negative predictive value (NPV) for each model. Results: As shown in Figure 1a, several ML models showed excellent diagnostic accuracies in the prediction of recurrent MI, demonstrating high NPV. Although the weighting of feature importance for MI recurrence varied among ML models, infarct size (peak CK levels), increased inflammatory status, acute coronary syndrome, and traditional coronary risk factors were identified as important features for predicting recurrent MI within two years after PCI (Figure 1b). Conclusions: It is suggested that ML using EMR data is promising for predicting recurrent MI within two years after PCI. Further studies are warranted to address whether ML-based risk stratification could lead to improved clinical decision-making after PCI.