New Echocardiographic Risk Score for HCM Patients Follow-up

Marion Taconne1, Valentina Corino1, Annamaria del Franco2, Eleonora Insinna3, Pietro Cerveri1, iacopo olivotto3, Luca Mainardi1
1Politecnico di Milano, 2Careggi University Hospital, 3University of Florence


Abstract

Context: Hypertrophic cardiomyopathy (HCM) is a complex myocardial disease associated with an increased risk of adverse cardiovascular events. Existing risk score do not fully exploit the richness of echocardiographic data and its potentials for dynamic patient monitoring. In this study, we propose a novel, echocardiographic-based machine learning risk score for predicting composite cardiovascular events in HCM patients.

Methods: We retrospectively analyzed data from 1186 HCM patients included in a monocentric part of SHARE registry. This database comprises systematically collected echocardiographic and follow-up data. We only used conventional echocardiographic parameters and trained four supervised machine learning models (logistic regression, support vector machine, random forest, gradient boosting) to predict the occurrence of a composite outcome at 5 years, including heart failure events (cardiac transplant, implantation, left ventricular ejection fraction <35%, new occurrence of NYHA) and ventricular arrhythmic events (SCD, cardiac arrest, appropriate implantable cardioverter therapy). The models were evaluated with a 5 nested cross validation (CV).

Results: The random forest model had the best performance, with a balanced accuracy of 0.688±0.014 (specificity of 0.671±0.048, and sensitivity of 0.705±0.053), highlighting the complexity of outcome prediction in HCM patients. Beyond baseline stratification, we analyzed the longitudinal evolution of the risk score in patients with serial echocardiographic data. In 77% of patients who experienced an event on the first CV test fold, the score progressively increased in the years preceding the event, revealing a dynamic risk profile.

Conclusions: Although the model's discriminative performance remains moderate, the proposed machine-learning based score evidenced potential for longitudinal tracking of patient progression. It offers additional insight into disease evolution and may support personalized follow-up and therapeutic decision-making in the management of HCM patients.