Scholarlink

<p>Heart failure with preserved ejection fraction (HFpEF) is increasingly recognized in hypertrophic cardiomyopathy (HCM); however, its prognostic significance, phenotypic heterogeneity, and optimal risk stratification strategies remain incompletely defined. In this multicenter retrospective cohort study, 2802 patients with HCM were enrolled from three tertiary centers. HFpEF was diagnosed using established criteria, and H₂FPEF score–based risk subgrouping was performed to further stratify patients. Propensity score matching was applied to balance baseline characteristics between HFpEF and Non-HFpEF patients. Event-free survival was assessed using Kaplan–Meier and multivariable Cox analyses. Restricted cubic spline modeling evaluated non-linear associations between B-type natriuretic peptide (BNP) levels and outcomes. Four machine learning models were developed for individualized risk prediction, with model interpretability assessed using SHAP analysis. HFpEF was present in 47.8% of patients with HCM and was independently associated with worse event-free survival after propensity score matching (HR = 2.612, 95% CI 2.188–3.118, <i>P</i> &lt; 0.001). Higher H₂FPEF scores conferred graded risk, with HFpEF-High patients exhibiting substantially poorer outcomes (HR 2.925, 95% CI 2.210–3.701; <i>P</i> &lt; 0.001). BNP demonstrated a significant non-linear relationship with adverse events, with risk accelerating at higher concentrations. Among machine learning models, the random forest achieved the best discrimination (AUC = 0.856), with SHAP analysis identifying HFpEF status and BNP as dominant contributors to risk prediction. HFpEF represents a prevalent, heterogeneous, and high-risk phenotype in HCM. Integrating H₂FPEF score–based risk subgrouping, non-linear biomarker modeling, and interpretable machine learning enhances personalized risk stratification and may, pending external validation, inform precision management strategies in HCM.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning–based risk stratification identifies heart failure with preserved ejection fraction as an independent predictor of adverse outcomes in hypertrophic cardiomyopathy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月16日

查看全文
  • Weijie Zhang,
  • Huan Zhao,
  • Zhuchang Tian,
  • Wei Fu,
  • Zongyang Li,
  • Zhouxu Geng,
  • Yuhan He,
  • Honghou He,
  • Peihong Wu,
  • Shengsong Zhu,
  • Min Yang,
  • Jing Chen,
  • Min Lin,
  • Zhiyuan Zhang,
  • Mengshen Wang,
  • Zijia Zhu,
  • Yanli Cui,
  • Fushi Piao,
  • Mingqi Zheng

摘要

Heart failure with preserved ejection fraction (HFpEF) is increasingly recognized in hypertrophic cardiomyopathy (HCM); however, its prognostic significance, phenotypic heterogeneity, and optimal risk stratification strategies remain incompletely defined. In this multicenter retrospective cohort study, 2802 patients with HCM were enrolled from three tertiary centers. HFpEF was diagnosed using established criteria, and H₂FPEF score–based risk subgrouping was performed to further stratify patients. Propensity score matching was applied to balance baseline characteristics between HFpEF and Non-HFpEF patients. Event-free survival was assessed using Kaplan–Meier and multivariable Cox analyses. Restricted cubic spline modeling evaluated non-linear associations between B-type natriuretic peptide (BNP) levels and outcomes. Four machine learning models were developed for individualized risk prediction, with model interpretability assessed using SHAP analysis. HFpEF was present in 47.8% of patients with HCM and was independently associated with worse event-free survival after propensity score matching (HR = 2.612, 95% CI 2.188–3.118, P < 0.001). Higher H₂FPEF scores conferred graded risk, with HFpEF-High patients exhibiting substantially poorer outcomes (HR 2.925, 95% CI 2.210–3.701; P < 0.001). BNP demonstrated a significant non-linear relationship with adverse events, with risk accelerating at higher concentrations. Among machine learning models, the random forest achieved the best discrimination (AUC = 0.856), with SHAP analysis identifying HFpEF status and BNP as dominant contributors to risk prediction. HFpEF represents a prevalent, heterogeneous, and high-risk phenotype in HCM. Integrating H₂FPEF score–based risk subgrouping, non-linear biomarker modeling, and interpretable machine learning enhances personalized risk stratification and may, pending external validation, inform precision management strategies in HCM.