Machine learning approach based on echocardiographic data to improve prediction of cardiovascular events in hypertrophic cardiomyopathy

Abstract

Introduction

Structural changes and myocardial fibrosis quantification by cardiac imaging have become increasingly important to predict cardiovascular events in hypertrophic cardiomyopathy patients. In this setting, it is likely that a supervised approach, using machine learning, may improve their risk assessment.

Method

We retrospectively included patients with confirmed HCM (n = 265, 52 ± 17 years) through clinical and echocardiographic. A supervised machine learning prognosis algorithm, based on echocardiographic data, was obtained to predict cardiovascular (CV) outcomes, and subsequently investigated for their association with myocardial fibrosis (n = 185) assessed by CMR imaging.

Results

At follow-up at 57 months, 13 (4.9%) of patients died and 114 (43%) had been hospitalized for CV events. Patient with CV events had higher indexed LV mass, worse diastolic dysfunction, and more severe LV obstruction. HCM-patients with myocardial fibrosis have more severe LV hypertrophy (OR: 3.1; P = 0.003) and longitudinal myocardial deformation (OR: 0.8; P = 0.008). Prognosis algorithm established using machine learning identified left atrium area (> 24 cm²), mechanical dispersion (> 49 ms), posterior wall thickness (> 1.8 cm), and TAPSE (27 mm) as the four most relevant variables to correctly predict cardiovascular events.

Conclusion

Our findings suggest that a simple algorithm based on four key variables (posterior wall thickness, mechanical dispersion, LA area and TAPSE) may help risk stratification and decision-making in patients with HCM. Using new treatments to target these parameters might improve outcomes in HCM-patients (Fig. 1).