Prediction of variant angia using Artificial intelligence enhanced electrocardiography

Abstract Introduction Variant angina, also known as Prinzmetal's angina, is essential to recognize as it results from coronary artery spasms rather than typical atherosclerotic plaque. Variant angina can lead to significant myocardial ischemia and may result in life-threatening arrhythmias or sudden cardiac arrest. Invasive coronary angiography is required for definitive diagnosis. Recently, AI-based ECG analysis has been studied for various cardiovascular diseases, suggesting its potential utility in diagnosing variant angina. Hypothesis We aimed to develop and validate a predictive model using ECG data to identify variant angina. Methods This study included all patients aged 18 years and older who had undergone at least one ECG and a coronary spasm test. Only cases with an electrocardiogram performed within one month of the coronary spasm test were selected. After dividing the dataset into training, test, validation sets, we developed an end-to-end deep neural network to predict the presence of variant angina. Model performance was evaluated using metrics including AUROC, AUPRC, sensitivity, specificity, and F1 score. Results The dataset comprised 1,153 ECGs from 1,153 patients. The AUROC for distinguishing variant angina in the validation set was 0.79, with sensitivity, specificity, and F1 score at 0.87, 0.62, and 0.75, respectively. These findings were consistent across subanalyses based on patient age, gender, comorbidities, and heart rate. Conclusion Our deep learning model demonstrates potential in predicting variant angina from ECG data. Further studies are underway to clarify the relative importance of individual ECG features in the model’s predictions.