Abstract 4144431: Machine Learning to Differentiate Myocardial Infarction with Obstructive versus Non-Obstructive Coronary Artery Disease

Background: Myocardial infarction (MI) can occur with or without coronary artery disease, classified as myocardial infarction with non-obstructive coronary artery disease (MINOCA) and MI with obstructive coronary artery disease (MICAD), respectively. Differentiating between these conditions is important as they require potentially different diagnostic and therapeutic approaches. No pre-test probability score currently exists to aid in this differentiation. This study aims to develop and validate machine learning (ML) models to predict whether a patient with suspected MI has MINOCA or MICAD. Methods: Data from the Alberta Provincial Project for Outcome Assessment in Coronary Heart Disease (APPROACH) database, including patients who underwent their first cardiac catheterization for chest pain or an anginal equivalent from 2002 to 2017, was used. We analyzed a cohort of 53,348 patients suspected of having myocardial infarction. For machine learning model development, 20 demographic and clinical features were selected using the Boruta algorithm. The dataset was divided into training (80%) and testing (20%) sets. Model performance was evaluated using 5-fold cross-validation, and the area under the receiver operating characteristic curve (AUROC) was the primary metric for model evaluation. Results: Among the five machine learning models trained, the XGBoost model showed superior performance, achieving an AUROC of 0.76 [95% CI: 0.76–0.77] and an accuracy of 0.86 during internal validation. Logistic regression achieved an AUROC of 0.74 [95% CI: 0.74–0.74]. Calibration of the XGBoost model was assessed via a calibration plot, which showed good agreement between predicted probabilities and observed event probabilities, with a Brier score of 0.10. Conclusions: The developed ML models demonstrated moderate accuracy for predicting MICAD versus MINOCA using clinically relevant features. Implementing such models in clinical practice could guide alternative diagnostic approaches for patients with suspected MINOCA.