Abstract 18722: Development and External Validation of an ECG-Based Deep Learning Algorithm for Screening Hypertrophic Cardiomyopathy in Elderly Patients with Unexplained Left Ventricular Hypertrophy: A Retrospective Multicenter Study
Abstract
Introduction: Hypertrophic cardiomyopathy (HCM) is an underrecognized cause of heart failure, arrhythmias, and sudden cardiac death in older adults, where distinguishing HCM from age-related or hypertensive left ventricular hypertrophy (LVH) remains challenging. Early identification is critical but often delayed due to overlapping phenotypes and limited access to advanced imaging. We developed and externally validated a deep learning algorithm using standard 12-lead electrocardiograms (ECGs) to screen for HCM among elderly patients with unexplained LVH across multiple centers. Methods: This retrospective multicenter study included elderly patients (≥65 years) evaluated between January 2021 and December 2025 for unexplained left ventricular hypertrophy. Standard 12-lead ECGs and imaging data were collected from participating centers. A deep learning convolutional neural network was developed and internally trained, followed by external validation in independent cohorts. Model performance was assessed using area under the receiver-operating characteristic curve, sensitivity, specificity, calibration, and predictive values. Results: A total of 9,842 elderly patients with unexplained left ventricular hypertrophy were included, of whom 1,476 (15.0%) had imaging-confirmed hypertrophic cardiomyopathy (HCM). In the external validation cohort, the deep learning ECG algorithm achieved an area under the receiver-operating characteristic curve of 0.93 (95% CI, 0.92-0.94), with sensitivity of 88.4% (95% CI, 86.3-90.2), specificity of 85.7% (95% CI, 84.8-86.6), positive predictive value of 53.8% (95% CI, 51.4-56.2), and negative predictive value of 97.6% (95% CI, 97.1-98.0). The algorithm demonstrated an overall accuracy of 86.1% (95% CI, 85.4-86.8) and F1-score of 0.67 (95% CI, 0.64-0.70). Patients categorized as high risk had significantly greater odds of confirmed HCM (OR 7.8, 95% CI, 6.9-8.9; p<0.001). Subgroup analyses showed consistent performance in patients aged ≥75 years (AUC 0.91, 95% CI, 0.89-0.93), women (AUC 0.92, 95% CI, 0.90-0.94), and those with hypertension (AUC 0.90, 95% CI, 0.88-0.92). Use of the algorithm could reduce advanced imaging referrals by 41.2% (95% CI, 37.5-44.8) while maintaining high diagnostic sensitivity. Conclusion: In this multicenter study, an ECG-based deep learning algorithm accurately identified hypertrophic cardiomyopathy among elderly patients with unexplained left ventricular hypertrophy. The model demonstrated robust external validity, high negative predictive value, and consistent performance across subgroups, supporting its potential as a scalable, cost-effective screening tool to facilitate earlier diagnosis.