Machine learning model including left ventricular strain analysis for sudden cardiac death prediction in hypertrophic cardiomyopathy

Abstract

Abstract Funding Acknowledgements Type of funding sources: None. Background The excess mortality in hypertrophic cardiomyopathy (HCM) patients is mainly attributed to the occurrence of sudden cardiac death (SCD). The prediction of ventricular arrhythmias remains challenging and could be improved. Purpose This study evaluated the added predictive value of a machine learning-based model combining clinical and conventional imaging parameters with information from left ventricular strain analysis to predict SCD in patients with HCM. Methods A total of 434 HCM patients (65% men, mean age 56 years) were retrospectively included from two referral centers from two different countries and followed longitudinally (mean duration 6 years). Strain parameters were automatically extracted from the left ventricle longitudinal strain segmental curves of each patient and included in a Ridge Regression model alongside conventional clinical and imaging data. The composite endpoint included sustained ventricular tachycardia, appropriate implantable cardioverter defibrillator therapy, aborted cardiac arrest, or sudden cardiac death. Results 34 patients (7.8%) met the endpoint with an incidence of ventricular arrhythmias of 0.9%/years. Among the 18 most discriminating parameters, 7 were derived from left ventricle longitudinal strain segmental curves analysis (figure 1). After n=200 rounds of cross-validation, the final model showed superior predictive performance compared to conventional models with a mean area under the curve (AUC) of 0.83 ± 0.8 compared with an AUC of 0.56 and 0.61 for the 2014 ESC risk score and the 2020 AHA/ACC model, respectively. Conclusion A machine learning model including automatically extracted left ventricular strain-derived parameters was superior in the prediction of sustained ventricular arrhythmias and SCD in patients with HCM compared to existing models. A machine learning model including left ventricle longitudinal strain analysis could improve SCD risk stratification in HCM patients.