A machine Learning-Based risk stratification using CMR volumetric and strain parameters in patients with hypertrophic cardiomyopathy.

Hypertrophic cardiomyopathy (HCM) is an inherited myocardial disorder associated with sudden cardiac death, atrial fibrillation (AF), and stroke, with prognosis varying widely among patients. Accurate long-term risk prediction requires comprehensive assessment of cardiac function, including left atrial performance, but conventional clinical and echocardiographic markers may not fully reflect the complexity of disease progression. This study aimed to develop a machine learning-based model that integrates clinical, echocardiographic, and CMR-derived parameter to improve prognostic risk stratification in patients with HCM. Between June 2008 and January 2016, 223 patients with complete demographic, clinical, echocardiographic, and CMR data were enrolled in the analysis. The composite of clinical outcome included of all-cause mortality, implantable cardioverter defibrillator shock, heart failure-related hospitalization, new-onset AF, and new-onset ischemic stroke. Five machine learning methodologies were evaluated. Models incorporating CMR-derived parameters outperformed those based solely on clinical and echocardiographic data. Among all models, penalized logistic regression integrating clinical, echocardiographic, and left atrial (LA) strain parameters demonstrated the highest predictive performance (AUC = 0.840, C-index = 0.795), and effectively stratified the long-term risk of outcome (low-risk group: 0%, intermediate-risk group: 27.3%, high-risk group: 77.8%, p = 0.002). Additionally, the LA strain model showed robust predictive performance across individual outcome components. Shapley additive explanations (SHAP) value analysis identified LA strains related to conduit function as the most significant predictor. CMR-derived parameters provide supplementary predictive impact in patients with HCM, with LA strain consistently outperforming other variables. Machine learning-based methodologies can effectively incorporate multiple variables and offer an effective approach for long-term risk stratification.