Abstract 4369790: AI-Driven Electrocardiographic Risk Stratification for Ventricular Arrhythmias in Hypertrophic Cardiomyopathy: A Systematic Review and Meta-Analysis

Background: Hypertrophic cardiomyopathy (HCM) is a common cause of sudden cardiac death (SCD) in young people, sometimes accompanied by malignant ventricular arrhythmias. Recent advances in artificial intelligence (AI) applied to electrocardiographic (ECG) data have enabled unique risk prediction systems. However, their predictive accuracy and clinical significance across trials are still dubious. Objective: To systematically review and meta-analyze the diagnostic performance of AI-based ECG models for predicting ventricular arrhythmias and SCD risk in patients with HCM. Methods: We followed PRISMA criteria and searched PubMed, Embase, and Scopus from 2010 to 2025 for research that used AI/ML models to ECG data for risk prediction in HCM. Eligible studies provided sensitivity, specificity, AUC, and hazard ratios for ventricular arrhythmias or SCD outcomes. The meta-analysis was conducted using a random-effects model with pooled diagnostic accuracy ratings and heterogeneity (I square). The risk of bias was assessed using PROBAST. Results: Twelve studies (n = 5,732 HCM patients) satisfied the inclusion criteria. AI-based ECG models had a pooled sensitivity of 84 percent(95 percent CI: 77-90 percent), specificity of 81 percent (95 percent CI: 74-86 percent), and pooled AUC of 0.88 (95 percent CI: 0.85-0.91) for predicting ventricular arrhythmias. Subgroup analysis revealed that deep learning models (CNN, LSTM) outperformed classical machine learning (AUC 0.90 vs. 0.82, p = 0.03). There was modest heterogeneity (I square = 28%), and no significant publication bias was observed. Conclusion: AI-based ECG models are very accurate in diagnosing ventricular arrhythmias in HCM patients, with deep learning models outperforming them. This meta-analysis supports incorporating AI-enhanced ECG interpretation into HCM-specific risk assessment techniques. Prospective validation and incorporation into clinical decision processes are critical.