CardioGraph: GCN-Inspired Short-Term Mortality Predictor in First-Time STEMI Patients

Cardiovascular diseases (CVDs) are a leading cause of death globally, with ST Elevation Myocardial Infarction (STEMI) being a major contributor to both death and cardiac complications. The 30-day post-STEMI period is critical to patient outcomes and necessitates accurate risk prediction to enable timely interventions and personalized treatment plans. Traditional statistical risk assessment tools, while widely used, are predominantly based on Caucasian population and do not address the specific risk profiles of other demographics, such as the Indian population. We propose CardioGraph, a Graph Convolutional Network (GCN) based cardiovascular risk predictor, designed specifically for Indian patients, to predict the 30-day mortality risk following STEMI. Results show that our proposed system outperforms in comparison to traditional Machine Learning based classifiers, such as Random Forest, Multi-Layer Perceptron, and Extreme Gradient Boosting, achieving a Macro F1 score of 93.5%. SHAP interpretability analysis of the proposed model identified some key predictors of 30-day mortality, such as mitral regurgitation, tobacco use, low ejection fraction, diabetes, and hypertension. Our proposed model effectively predicts 30-day mortality post-STEMI with high accuracy, which suggests a significant advancement in cardiovascular risk prediction.