Abstract 4364782: Machine Learning–Based Risk Stratification in Hypertrophic Cardiomyopathy: The Added Prognostic Value of Combined Cardiopulmonary Exercise Testing and Stress Echocardiography

Abstract

Introduction: Hypertrophic cardiomyopathy (HCM) is a genetically mediated myocardial disorder with variable clinical expression and reduced functional capacity. Cardiopulmonary exercise testing when integrated with stress echocardiography (CPET-TTE), offers a comprehensive evaluation of cardiovascular performance. In this study, we aimed to assess the added value of CPET-TTE for risk stratification in HCM using a machine learning–based approach to identify individuals at higher risk for major adverse events Methods: We retrospectively analyzed 413 HCM patients (46% obstructive; 63.1% male; mean age 48.3 years) who underwent CPET with rest and stress echocardiography, 24-hour ECG Holter, cardiac MRI, and genetic testing. Several machine learning models were developed to predict a composite outcome (SCD, aborted SCD, heart transplantation, stroke, myocardial infarction), with Gradient Boosting using Cox proportional hazards loss achieving the best performance. A semi-supervised clustering approach applying k-means to out-of-fold risk scores was used to stratify patients into high- and low-risk groups. Feature importance was assessed using the Kruskal-Wallis test and ranked by -log10(p-value); effect sizes were quantified using eta-squared. Results: Gradient Boosting achieved the best predictive performance (C-index: 0.722). Survival analysis showed a clear separation between high-risk (n = 56, 30.4% events) and low-risk (n = 357, 10.1% events) groups (p < 0.000001). High-risk patients were older (56.0 vs 49.0 years) and had significantly reduced exercise capacity (VO2max%: 49.8% vs 68.0%; AT%: 41.3% vs 55.5%; pVO2: 15.1 vs 19.7 ml/kg/min), greater ventilatory inefficiency (VE/VCO2: 29.0 vs 26.5), lower watt ( 80.0 vs 100.0 W), and oxygen pulse (HR/VO2: 9.7 vs 11.4).No significant differences were observed in rest or peak LVOT gradient (14.5 vs 11.0 mmHg; 31.0 vs 30.0 mmHg) or E/e′ ratio at rest (11.7 vs 11.2) and stress (11.4 vs 10.3). Key features contributing to risk prediction included VO2max%, AT%, pVO2 and VE/VCO2 slope (all p < 0.001). Conclusion: The integration of CPET-TTE parameters with advanced machine learning techniques allowed for effective and clinically relevant risk stratification in patients with hypertrophic cardiomyopathy.. These findings highlight the central prognostic value of exercise capacity and ventilatory efficiency in HCM, supporting the routine use of CPET-TTE as a non-invasive, functional tool for personalized risk assessment.