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

Machine learning–based risk stratification identifies heart failure with preserved ejection fraction as an independent predictor of adverse outcomes in hypertrophic cardiomyopathy - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Sci Rep . 2026 Apr 16;16:12885. doi: 10.1038/s41598-026-46573-z Search in PMC Search in PubMed View in NLM Catalog Add to search Machine learning–based risk stratification identifies heart failure with preserved ejection fraction as an independent predictor of adverse outcomes in hypertrophic cardiomyopathy Weijie Zhang Weijie Zhang 1 Heart Center of Henan Provincial People’s Hospital, Central China Fuwai Hospital, Central China Fuwai Hospital of Zhengzhou University, Zhengzhou, 450003 Henan China Find articles by Weijie Zhang 1, # , Huan Zhao Huan Zhao 2 Department of Cardiac Function, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China Find articles by Huan Zhao 2, # , Zhuchang Tian Zhuchang Tian 3 Department of Cardiology, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China Find articles by Zhuchang Tian 3, # , Wei Fu Wei Fu 3 Department of Cardiology, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China Find articles by Wei Fu 3 , Zongyang Li Zongyang Li 3 Department of Cardiology, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China Find articles by Zongyang Li 3 , Zhouxu Geng Zhouxu Geng 3 Department of Cardiology, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China Find articles by Zhouxu Geng 3 , Yuhan He Yuhan He 3 Department of Cardiology, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China Find articles by Yuhan He 3 , Honghou He Honghou He 3 Department of Cardiology, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China Find articles by Honghou He 3 , Peihong Wu Peihong Wu 3 Department of Cardiology, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China Find articles by Peihong Wu 3 , Shengsong Zhu Shengsong Zhu 4 Department of Cardiology, State Key Laboratory of Cardiovascular Disease, Fuwai Hospital, National Center for Cardiovascular Diseases, Chinese Academy of Medical Sciences, Peking Union Medical College, Beijing, 100037 China Find articles by Shengsong Zhu 4 , Min Yang Min Yang 4 Department of Cardiology, State Key Laboratory of Cardiovascular Disease, Fuwai Hospital, National Center for Cardiovascular Diseases, Chinese Academy of Medical Sciences, Peking Union Medical College, Beijing, 100037 China Find articles by Min Yang 4 , Jing Chen Jing Chen 4 Department of Cardiology, State Key Laboratory of Cardiovascular Disease, Fuwai Hospital, National Center for Cardiovascular Diseases, Chinese Academy of Medical Sciences, Peking Union Medical College, Beijing, 100037 China Find articles by Jing Chen 4 , Min Lin Min Lin 5 Division of Cardiac Arrhythmia, Cardiac and Vascular Center, The University of Hong Kong-Shenzhen Hospital, Shenzhen, 518053 Guangdong China Find articles by Min Lin 5 , Zhiyuan Zhang Zhiyuan Zhang 6 Department of Plastic and Burn Surgery, National Key Clinical Construction Specialty, The Affiliated Hospital of Southwest Medical University, Luzhou, 646000 Sichuan China Find articles by Zhiyuan Zhang 6 , Mengshen Wang Mengshen Wang 7 The Department of Computer Science and Engineering, The Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong China Find articles by Mengshen Wang 7 , Zijia Zhu Zijia Zhu 7 The Department of Computer Science and Engineering, The Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong China Find articles by Zijia Zhu 7 , Yanli Cui Yanli Cui 8 School of Foreign Languages and Literature, Beijing Normal University, Beijing, 100875 China 9 Faculty of Arts and Sciences, Beijing Normal University, Beijing, 100875 China Find articles by Yanli Cui 8, 9, ✉ , Fushi Piao Fushi Piao 10 Department of Cardiology, Peking University Shenzhen Hospital, Shenzhen, 518000 Guangdong China Find articles by Fushi Piao 10, ✉ , Mingqi Zheng Mingqi Zheng 3 Department of Cardiology, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China Find articles by Mingqi Zheng 3, ✉ Author information Article notes Copyright and License information 1 Heart Center of Henan Provincial People’s Hospital, Central China Fuwai Hospital, Central China Fuwai Hospital of Zhengzhou University, Zhengzhou, 450003 Henan China 2 Department of Cardiac Function, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China 3 Department of Cardiology, The First Hospital of Hebei Medical University, Shijiazhuang, 050000 Hebei China 4 Department of Cardiology, State Key Laboratory of Cardiovascular Disease, Fuwai Hospital, National Center for Cardiovascular Diseases, Chinese Academy of Medical Sciences, Peking Union Medical College, Beijing, 100037 China 5 Division of Cardiac Arrhythmia, Cardiac and Vascular Center, The University of Hong Kong-Shenzhen Hospital, Shenzhen, 518053 Guangdong China 6 Department of Plastic and Burn Surgery, National Key Clinical Construction Specialty, The Affiliated Hospital of Southwest Medical University, Luzhou, 646000 Sichuan China 7 The Department of Computer Science and Engineering, The Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong China 8 School of Foreign Languages and Literature, Beijing Normal University, Beijing, 100875 China 9 Faculty of Arts and Sciences, Beijing Normal University, Beijing, 100875 China 10 Department of Cardiology, Peking University Shenzhen Hospital, Shenzhen, 518000 Guangdong China ✉ Corresponding author. # Contributed equally. Received 2026 Jan 14; Accepted 2026 Mar 26; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13096635  PMID: 41991628 Abstract 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. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-026-46573-z. Keywords: Hypertrophic cardiomyopathy, Heart failure with preserved ejection fraction, Machine learning, Personalized risk stratification, BNP, Atrial fibrillation Subject terms: Biomarkers, Cardiology, Diseases, Medical research, Risk factors Introduction Hypertrophic cardiomyopathy (HCM) is a common inherited cardiac disorder characterized by left ventricular hypertrophy, diastolic dysfunction, and a highly heterogeneous clinical course 1 . Although advances in imaging, risk stratification, and disease-modifying therapies have improved the management of HCM, heart failure remains a leading cause of morbidity and adverse outcomes in this population 2 . Notably, a substantial proportion of patients with HCM develop heart failure with preserved ejection fraction (HFpEF), reflecting the dominant role of diastolic dysfunction, myocardial stiffness, and atrial remodeling in disease progression 3 . HFpEF represents a complex clinical syndrome with diverse pathophysiological mechanisms and variable prognosis 4 . In the general population, HFpEF is associated with advanced age, atrial fibrillation, chronic kidney disease, and elevated natriuretic peptide levels, and is increasingly recognized as a major contributor to cardiovascular morbidity 5 . Although HCM and other cardiomyopathies have historically been excluded from major HFpEF clinical trials to ensure population homogeneity, recent expert consensus documents—including the 2023 ACC Expert Consensus Decision Pathway on Management of HFpEF and the 2024 AHA/ACC Guideline for the Management of HCM—increasingly recognize that patients with HCM can develop HFpEF as a clinically relevant and distinct phenotype, driven by diastolic dysfunction, elevated filling pressures, and atrial remodeling. In patients with HCM, HFpEF may represent a particularly high-risk phenotype, given the coexistence of myocardial hypertrophy, left ventricular outflow tract obstruction, microvascular dysfunction, and elevated filling pressures 6 – 8 . However, the prognostic implications of HFpEF in HCM remain incompletely characterized. Critically, the pathophysiological substrate of HFpEF in HCM is mechanistically distinct from that of HFpEF in the general population. In non-HCM patients, HFpEF is predominantly driven by systemic comorbidities such as obesity, hypertension, diabetes, and aging-related vascular stiffening, which collectively promote concentric remodeling and diastolic dysfunction. In contrast, HCM creates a unique milieu for HFpEF development through disease-specific mechanisms: myocardial disarray and interstitial fibrosis cause intrinsic myocardial stiffness; microvascular dysfunction and impaired coronary flow reserve produce demand ischemia even in the absence of epicardial coronary disease; dynamic left ventricular outflow tract obstruction generates phasic hemodynamic stress and mitral regurgitation; and progressive left atrial myopathy secondary to chronically elevated filling pressures leads to atrial fibrillation and further hemodynamic compromise. These interacting pathways create a self-reinforcing cycle of diastolic failure, neurohormonal activation, and exercise intolerance that is fundamentally different from the comorbidity-driven phenotype of general HFpEF. Consequently, studying HFpEF specifically within the HCM context provides unique insights into diastolic heart failure mechanisms arising from primary myocardial disease, rather than from systemic comorbidity accumulation. Prior studies have established that heart failure, including HFpEF, is associated with adverse prognosis in HCM; however, these studies have primarily focused on heart failure as a binary clinical diagnosis or have concentrated on advanced or end-stage disease 9 . The extent to which HFpEF status independently stratifies risk among patients with HCM, after accounting for baseline clinical characteristics, remains uncertain. Moreover, HFpEF is increasingly recognized as a spectrum rather than a dichotomous entity 10 , yet few studies have evaluated whether the degree of HFpEF burden conveys graded prognostic information in the HCM population. In parallel, conventional statistical models may be limited in their ability to capture complex, multidimensional risk patterns inherent to HCM and HFpEF. Machine learning approaches have shown promise in improving risk prediction across a range of cardiovascular diseases 11 , yet their application in HCM—particularly for the prediction of HF-related adverse outcomes—remains limited. Importantly, concerns regarding model interpretability have hindered the clinical adoption of machine learning–based risk models 12 . Accordingly, in this multicenter cohort study, we sought to comprehensively evaluate the prognostic significance of HFpEF in patients with HCM. Specifically, we aimed to (I) compare event-free survival between patients with and without HFpEF using propensity score–matched analyses; (II) assess whether H₂FPEF score–based risk subgrouping confers graded risk stratification; (III) characterize the association between BNP levels and adverse outcomes, including potential non-linear relationships; and (IV) develop and interpret machine learning models to improve individualized risk prediction. By integrating traditional survival analyses, spline modeling, and interpretable machine learning approaches, this study aims to refine risk stratification and enhance clinical decision-making in patients with HCM and HFpEF. Methods Study population This multicenter retrospective cohort study consecutively enrolled patients diagnosed with hypertrophic cardiomyopathy (HCM) from three tertiary referral centers in China: The First Hospital of Hebei Medical University, Henan Provincial People’s Hospital, and Fuwai Hospital, National Center for Cardiovascular Diseases. HCM was defined according to contemporary guideline criteria, including a maximal left ventricular wall thickness ≥ 15 mm in the absence of other cardiac or systemic conditions capable of producing myocardial hypertrophy 13 . Between October 2009 and December 2024, a total of 2,802 consecutive patients with HCM were screened for eligibility. Patients were excluded if they met any of the following criteria: (I) left ventricular systolic dysfunction or end-stage HCM, defined by progressive LV wall thinning, cavity dilation, or declining systolic function irrespective of the absolute LVEF value; (II) secondary causes of left ventricular hypertrophy, including hypertensive heart disease, Fabry disease, cardiac amyloidosis, and other infiltrative or metabolic cardiomyopathies (phenocopies); (III) loss to follow-up; or (IV) incomplete baseline clinical data. Patients with LVEF ≥ 50% who exhibited signs or symptoms of heart failure without features of end-stage HCM were eligible for HFpEF classification, thereby ensuring a clear diagnostic separation between the HFpEF phenotype and the burned-out stage of HCM. After applying these criteria, 2,651 patients were included in the final analytical cohort. This retrospective study was approved by the Ethics Committees of the participating institutions (The First Hospital of Hebei Medical University, Henan Provincial People’s Hospital, and Fuwai Hospital, National Center for Cardiovascular Diseases). The requirement for written informed consent was waived due to the retrospective nature of the study and the use of anonymized clinical data, in accordance with the Declaration of Helsinki. Definition of HFpEF and clinical variables Heart failure with preserved ejection fraction was diagnosed based on established criteria, including the presence of heart failure symptoms and/or signs, a left ventricular ejection fraction ≥ 50%, and objective evidence of structural heart disease or diastolic dysfunction 14 . Importantly, HFpEF classification was based on a comprehensive assessment at baseline evaluation, incorporating objective echocardiographic parameters (including E/e’ ratio and left atrial dimensions) rather than symptoms alone, thereby ensuring diagnostic stability independent of transient symptom fluctuations such as those associated with atrial fibrillation episodes or left ventricular outflow tract obstruction. Baseline demographic characteristics, echocardiographic parameters, and medication use were collected from electronic medical records. Atrial fibrillation was defined by documented electrocardiography or medical history. Chronic kidney disease was defined according to estimated glomerular filtration rate criteria. B-type natriuretic peptide (BNP) levels were measured at baseline using standardized assays at each participating center. H₂FPEF score–based risk subgrouping HFpEF risk subgrouping was further performed using the H 2 FPEF score. Receiver operating characteristic (ROC) analysis was performed to determine the optimal cutoff value for predicting adverse outcomes. The optimal threshold was identified using the Youden index. Based on this cutoff, patients were stratified into HFpEF-High and HFpEF-Low subgroups for subsequent analyses. It should be noted that this use of the H₂FPEF score for prognostic risk subgrouping extends beyond its originally validated diagnostic application; the H₂FPEF score was developed to diagnose HFpEF rather than to grade its severity, and the present stratification approach should therefore be considered exploratory. Outcome definition and follow-up The primary endpoint of the study was event-free survival, defined as the time from baseline evaluation to the first occurrence of a predefined adverse clinical event. Adverse events included all-cause mortality and heart failure–related hospitalization, whichever occurred first.The index date was defined as the date of baseline clinical assessment at study enrollment. Patients were followed through outpatient visits and review of electronic medical records until the occurrence of an endpoint event or the date of last available follow-up, whichever came first. Follow-up duration was calculated in months. Restricted cubic spline analysis To explore potential non-linear associations between BNP levels and adverse outcomes, restricted cubic spline (RCS) analysis was performed within the Cox regression framework. BNP was modeled as a continuous variable, with knots placed at predefined percentiles of its distribution. A BNP level of 810 pg/mL was used as the reference value. Models were adjusted for the same covariates included in the multivariable Cox regression analyses. P values for overall association and non-linearity were calculated to assess the shape of the relationship between BNP and risk of adverse outcomes. The distribution of BNP values was displayed alongside spline curves to illustrate data density. Machine learning model development, evaluation, and interpretability Four supervised machine learning models—k-nearest neighbors, logistic regression, support vector machine, and random forest—were developed to predict adverse outcomes using key clinical variables. The dataset was randomly divided into a training set (70%) and an independent testing set (30%) using stratified sampling to ensure comparable event rates between the two subsets. Model performance was evaluated primarily by the area under the receiver operating characteristic curve (AUC). Additional performance metrics included sensitivity, specificity, and overall accuracy.Clinical utility was assessed using decision curve analysis (DCA), and model calibration was evaluated by calibration plots comparing predicted and observed event risks. To enhance model interpretability, Shapley Additive Explanations (SHAP) analysis was applied to the random forest model. Feature importance was quantified using mean absolute SHAP values, and SHAP summary and dependence plots were generated to visualize the direction and relative magnitude of feature contributions, facilitating clinical interpretability and comparison with results from Cox proportional hazards analyses. Statistical analysis Continuous variables are presented as mean ± standard deviation as appropriate, and were compared using Student’s t-test or the Mann-Whitney U test. Categorical variables are presented as counts and percentages and were compared using the χ² test or Fisher’s exact test. Univariable Cox proportional hazards regression analyses were first performed to identify clinical variables associated with adverse outcomes. Variables with clinical relevance or statistical significance in univariable analyses were subsequently included in multivariable Cox regression models. Results are presented as hazard ratios (HRs) with 95% confidence intervals (CIs). Event-free survival was estimated using the Kaplan-Meier method and compared between groups using the log-rank test. PSM was employed to reduce bias due to differences in observed variables between the two groups in the retrospective observational study, and the analysis was conducted using R version 4.3.0. A P-value less than 0.05 was considered statistically significant. Results Study population and cohort construction A total of 2802 consecutive patients with HCM were screened from three tertiary centers in China (Fig. 1 ). Among these, 1339 patients (47.8%) were classified as having HFpEF, while 1463 patients (52.2%) were classified as Non-HFpEF. After excluding patients with systolic dysfunction or end-stage HCM, secondary causes of left ventricular hypertrophy, loss to follow-up, or incomplete baseline data, 1272 HFpEF patients and 1379 Non-HFpEF patients remained eligible for analysis. Fig. 1. Open in a new tab Flow diagram of patient selection and PSM cohort construction in patients with HCM. To reduce baseline imbalances between groups, PSM was performed at a 1:1 ratio. This yielded a matched cohort of 2304 patients, comprising 1152 patients in the HFpEF group and 1152 patients in the Non-HFpEF group (Fig. 1 ). BNP and NYHA functional class were not included in the propensity score model because they reflect HFpEF burden and may lie on the causal pathway between HFpEF status and adverse outcomes. Distributions of propensity scores before and after matching were summarized in Supplementary Fig. 1 . Baseline characteristics before and after propensity score matching Baseline characteristics of patients before and after PSM are summarized in Table 1 . Before matching, patients with HFpEF were significantly older than those without HFpEF (median age: 56.8 vs. 51.3 years, P < 0.001) and exhibited a more advanced clinical profile. HFpEF patients had markedly higher BNP levels (median: 2410.05 vs. 410.15, P < 0.001) and a substantially higher prevalence of NYHA functional class III–IV symptoms (18.7% vs. 6.3%, P < 0.001). A history of syncope was also more frequent in the HFpEF group (20.1% vs. 17.9%, P = 0.043). Table 1. Baseline clinical characteristics of patients with and without HFpEF before and after propensity score matching (PSM). Variables Before PSM After PSM HFpEF Group ( n = 1272) Non-HFpEF Group ( n = 1379) P value HFpEF Group ( n = 1152) Non-HFpEF Group ( n = 1152) P value Demographics Age (years) 56.8 ± 13.60 51.3 ± 12.72 < 0.001 53.89 ± 13.90 53.27 ± 12.30 0.231 Male (n, %) 824 (64.8%) 916 (66.4%) 0.061 751 (65.2%) 760 (66.0%) 0.780 BMI (kg/m 2 ) 25.12 ± 5.62 24.68 ± 4.98 0.125 25.09 ± 4.15 24.95 ± 5.03 0.237 Hypertension disease (n, %) 509 (40.0%) 455 (33.0%) 0.403 444 (38.5%) 438 (38.0%) 0.801 Smoking (n, %) 566 (44.5%) 626 (45.4%) 0.509 517 (44.9%) 521 (45.2%) 0.912 BNP (pg/mL) 2410.05 ± 1820.42 410.15 ± 220.52 < 0.001 981.63 ± 561.16 565.46 ± 103.07 < 0.001 Chronic Kidney Disease (n, %) 286 (22.5%) 295 (21.4%) 0.481 252 (21.9%) 246 (21.4%) 0.722 Atrial Fibrillation (n, %) 504 (39.6%) 535 (38.8%) 0.602 453 (39.3%) 448 (38.9%) 0.145 NYHA III-IV (n, %) 238 (18.7%) 87 (6.3%) < 0.001 184 (16.0%) 85 (7.4%) < 0.001 Unexplained Syncope (n, %) 256 (20.1%) 247 (17.9%) 0.043 221 (19.2%) 217 (18.8%) 0.213 Echocardiography LV Septal Thickness (mm) 18.25 ± 5.71 18.18 ± 5.42 0.258 18.21 ± 5.86 18.02 ± 5.65 0.621 LVEDD (mm) 44.27 ± 6.20 44.75 ± 5.19 0.745 44.30 ± 6.01 44.69 ± 5.87 0.556 LVOT Gradient (mmHg) 41.84 ± 13.62 40.55 ± 13.85 0.564 40.97 ± 11.35 40.72 ± 12.08 0.177 Obstructive HCM (LVOT ≥ 30 mmHg), n (%) 1034 (81.3%) 1063 (77.1%) 0.009 972 (84.4%) 942 (81.8%) 0.107 LVEF (%) 64.09 ± 6.30 65.51 ± 6.22 0.126 64.54 ± 6.15 64.74 ± 5.88 0.690 E/e’ ratio 15.4 ± 5.8 10.2 ± 3.9 < 0.001 12.5 ± 4.7 11.0 ± 4.2 < 0.001 LA Diameter (mm) 44.3 ± 7.1 39.5 ± 6.4 < 0.001 41.8 ± 6.7 40.3 ± 6.5 < 0.001 Medication Use β-blockers (n, %) 784 (61.6%) 798 (57.9%) 0.211 692 (60.1%) 676 (58.7%) 0.466 Calcium Channel Blockers (n, %) 309 (24.3%) 379 (27.5%) 0.024 295 (25.6%) 302 (26.2%) 0.061 ACEI/ARB (n, %) 239 (18.8%) 230 (16.7%) 0.058 207 (18.0%) 201 (17.5%) 0.141 Diuretics (n, %) 207 (16.3%) 203 (14.7%) 0.075 187 (16.2%) 181 (15.7%) 0.366 Open in a new tab Abbreviations: HFpEF, heart failure with preserved ejection fraction; BMI, body mass index; BNP, B-type natriuretic peptide; CKD, chronic kidney disease; NYHA, New York Heart Association; LV, left ventricular; LVEDD, left ventricular end-diastolic diameter; LVOT, left ventricular outflow tract; LVEF, left ventricular ejection fraction; E/e’, ratio of peak early diastolic transmitral flow velocity to tissue Doppler early diastolic mitral annular velocity; LA, left atrial; ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin receptor blocker; PSM, propensity score matching. After propensity score matching, most baseline demographic, clinical, echocardiographic, and treatment variables were well balanced between the two groups. There were no significant differences in age, sex distribution, body mass index, prevalence of atrial fibrillation, chronic kidney disease, left ventricular septal thickness, or use of guideline-directed medical therapy. Despite matching, BNP levels remained significantly higher in the HFpEF group compared with the Non-HFpEF group (median: 981.63 vs. 565.46 pg/mL, P < 0.001). Similarly, NYHA functional class III–IV symptoms remained more prevalent among HFpEF patients (16.0% vs. 7.4%, P < 0.001), indicating persistent differences in hemodynamic burden between groups even after adjustment (Table 1 ). H₂FPEF score–based risk subgrouping Receiver operating characteristic analysis identified an optimal H₂FPEF score cutoff of 5.4 for discriminating adverse outcomes, yielding an AUC of 0.826, with a sensitivity of 79.6% and a specificity of 89.3% ( P < 0.001) (Fig. 2 ). Fig. 2. Open in a new tab ROC curve analysis of the H₂FPEF score for predicting adverse outcomes in patients with hypertrophic cardiomyopathy. Based on this cutoff, patients were further stratified into HFpEF-High ( n = 534) and HFpEF-Low ( n = 1770) groups. Compared with the HFpEF-Low group, HFpEF-High patients exhibited substantially higher circulating BNP levels (median: 994.63 vs. 624.40 pg/mL, P < 0.001) and a markedly higher prevalence of New York Heart Association (NYHA) functional class III–IV symptoms (17.2% vs. 7.9%, P < 0.001). Other baseline characteristics, including age, sex, comorbidities, and echocardiographic parameters, were largely comparable between the two subgroups (Table 2 ). Table 2. Clinicopathological characteristics in the HFpEF-Low group and the HFpEF-High group. Variables HFpEF- High group ( n = 534) HFpEF- Low group ( n = 1770) t/χ 2 P value Demographics Age (years) 53.97 ± 13.61 53.52 ± 12.80 0.682 0.496 Male (n, %) 347 (65.0%) 1176 (66.4%) 0.411 0.521 BMI (kg/m 2 ) 25.12 ± 4.95 25.05 ± 5.99 0.263 0.793 Hypertension disease (n, %) 213 (39.8%) 659 (37.2%) 1.247 0.264 Smoking (n, %) 244 (45.6%) 793 (44.8%) 0.105 0.746 BNP (pg/mL) 994.63 ± 589.55 624.40 ± 165.23 14.212 < 0.001 Chronic Kidney Disease (n, %) 120 (22.5%) 365 (20.6%) 0.965 0.326 Atrial Fibrillation (n, %) 206 (38.5%) 701 (39.6%) 0.231 0.631 NYHA III-IV (n, %) 92 (17.2%) 140 (7.9%) 37.405 < 0.001 Unexplained Syncope (n, %) 103 (19.2%) 336 (19.0%) 0.014 0.906 Echocardiography LV Septal Thickness (mm) 18.43 ± 5.56 17.96 ± 5.02 1.781 0.075 LVEDD (mm) 44.57 ± 5.74 44.01 ± 5.27 1.890 0.059 LVOT Gradient (mmHg) 41.14 ± 11.07 40.97 ± 11.57 0.305 0.760 Obstructive HCM (LVOT ≥ 30 mmHg), n (%) 457 (85.6%) 1483 (83.8%) 0.863 0.353 LVEF (%) 64.01 ± 6.15 64.54 ± 6.02 −1.646 0.176 E/e’ ratio 12.5 ± 4.7 11.2 ± 4.3 5.990 < 0.001 LA Diameter (mm) 41.8 ± 6.7 40.6 ± 6.5 3.712 < 0.001 Medication Use β-blockers (n, %) 327 (61.2%) 1064 (60.1%) 0.215 0.643 Calcium Channel Blockers (n, %) 138 (25.8%) 460 (26.0%) 0.015 0.908 ACEI/ARB (n, %) 98 (18.4%) 322 (18.2%) 0.012 0.913 Diuretics (n, %) 85 (16.0%) 274 (15.5%) 0.095 0.758 Open in a new tab Abbreviations: HFpEF, heart failure with preserved ejection fraction; BMI, body mass index; BNP, B-type natriuretic peptide; CKD, chronic kidney disease; NYHA, New York Heart Association; LV, left ventricular; LVEDD, left ventricular end-diastolic diameter; LVOT, left ventricular outflow tract; LVEF, left ventricular ejection fraction; E/e’, ratio of peak early diastolic transmitral flow velocity to tissue Doppler early diastolic mitral annular velocity; LA, left atrial; ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin receptor blocker. Association between HFpEF status and event-free survival During follow-up, Kaplan-Meier analysis demonstrated a significantly lower event-free survival in patients with HFpEF compared with propensity score–matched Non-HFpEF patients (log-rank P < 0.001) (Fig. 3 A). At approximately 5 years of follow-up, event-free survival was substantially lower in the HFpEF group, with an absolute difference of nearly 20% points compared with the Non-HFpEF group. This separation persisted and further widened over time, indicating a sustained divergence in long-term outcomes. Fig. 3. Open in a new tab Kaplan–Meier curves for event-free survival according to HFpEF status and H₂FPEF score–based risk subgroups in patients with HCM. Similarly, when patients were stratified according to H₂FPEF score–based risk subgroups, those classified as HFpEF-High exhibited significantly worse event-free survival than those in the HFpEF-Low group (log-rank P < 0.001) (Fig. 3 B). Event-free survival was consistently lower in the HFpEF-High group across the follow-up period, supporting a graded relationship between H₂FPEF score–based risk subgroups and adverse outcomes. Predictors of adverse outcomes in Cox proportional hazards analyses In univariable Cox regression analyses, atrial fibrillation, BNP level (per 100 pg/mL increment), and HFpEF status were each significantly associated with adverse events (Table 3 ). Compared with Non-HFpEF patients, those with HFpEF exhibited a markedly increased risk of adverse outcomes (HR = 2.545; 95% CI, 1.155–4.480; P < 0.001). Table 3. Univariate Cox regression analyses of risk factors. Variables HR (95%CI) P value Age (years) 1.029 (0.986–1.086) 0.214 BMI (kg/m 2 ) 1.051 (0.945–1.147) 0.209 Sex (male vs. female) 0.895 (0.722–1.023) 0.156 Atrial Fibrillation (yes vs. no) 2.016 (1.053–3.140) < 0.001 BNP (per 100 pg/mL) 1.257 (1.123–1.399) < 0.001 Syncope History (yes vs. no) 1.552 (0.962–2.208) 0.344 LV Septal Thickness (mm) 1.008 (0.915–1.121) 0.212 LVEDD (mm) 1.019 (0.976–1.064) 0.387 LVOT Gradient (per 10mmHg) 1.094 (0.986–1.231) 0.402 Hypertension disease (n, %) 1.530 (0.699–2.375) 0.451 HFpEF vs. Non-HFpEF 2.545 (1.155–4.480) < 0.001 Open in a new tab Abbreviations: HR, hazard ratio; CI, confidence interval; BMI, body mass index; BNP, B-type natriuretic peptide; LV, left ventricular; LVEDD, left ventricular end-diastolic diameter; LVOT, left ventricular outflow tract; HFpEF, heart failure with preserved ejection fraction. In multivariable Cox regression analyses adjusting for established clinical risk factors, including age, atrial fibrillation, syncope history, chronic kidney disease, left ventricular septal thickness, and BNP level, HFpEF remained an independent predictor of adverse outcomes (HR = 2.463; 95% CI, 1.179–4.170; P < 0.001) (Table 4 , Model 2). Table 4. Multivariate Cox regression analyses of risk factors. Variables Model 1 Model 2 Model 3 HR (95%CI) P value HR (95%CI) P value HR (95%CI) P value LV Septal Thickness (mm) 1.015 (0.941–1.130) 0.403 1.010 (0.918–1.129) 0.195 1.012 (0.923–1.115) 0.201 BNP (per 100 pg/mL) 1.237 (1.131–1.342) < 0.001 1.268 (1.128–1.438) < 0.001 1.250 (1.109–1.388) < 0.001 Chronic Kidney Disease (yes vs. no) 1.121 (0.890–1.412) 0.343 1.235 (0.912–1.663) 0.216 1.209 (0.921–1.351) 0.295 Age (years) 1.018 (0.992–1.054) 0.284 1.006 (0.963–1.065) 0.125 1.010 (0.958–1.084) 0.203 Syncope History (yes vs. no) 1.365 (0.770–2.419) 0.287 1.327 (0.825–1.609) 0.235 1.295 (0.860–1.702) 0.247 Atrial Fibrillation (yes vs. no) 2.706 (1.379–5.308) 0.004 2.028 (1.129–3.411) < 0.001 2.309 (1.203–3.509) < 0.001 HFpEF vs. Non-HFpEF 2.463 (1.179–4.170) < 0.001 HFpEF-High vs. HFpEF-Low 2.925 (2.210–3.701) < 0.001 Open in a new tab Abbreviations: HR, hazard ratio; CI, confidence interval; BNP, B-type natriuretic peptide; CKD, chronic kidney disease; HFpEF, heart failure with preserved ejection fraction. Model 1 adjusted for LV septal thickness, BNP (per 100 pg/mL), chronic kidney disease, age, syncope history, and atrial fibrillation. Model 2 additionally included HFpEF status (HFpEF vs. non-HFpEF). Model 3 replaced HFpEF status with H₂FPEF score–based risk subgroup (HFpEF-High vs. HFpEF-Low) as an alternative specification. When H₂FPEF score–based risk subgroups were incorporated into the model, patients classified as HFpEF-High had a substantially higher risk of adverse events compared with those in the HFpEF-Low group (HR = 2.925; 95% CI, 2.210–3.701; P < 0.001) (Table 4 , Model 3). Across all multivariable models, BNP level and atrial fibrillation consistently emerged as independent predictors of adverse outcomes. These associations are visually summarized in the multivariable Cox regression forest plot (Fig. 4 ). Fig. 4. Open in a new tab Forest plot of multivariable Cox proportional hazards regression for predictors of adverse outcomes in patients with hypertrophic cardiomyopathy. Non-linear association between BNP levels and adverse outcomes BNP was modeled as a continuous variable in the Cox regression to provide clinically interpretable hazard ratios, while restricted cubic spline analysis was used to assess potential non-linear associations across the full BNP spectrum. Restricted cubic spline analysis demonstrated a significant non-linear association between BNP levels and the risk of adverse outcomes (P for overall association < 0.001; P for non-linearity = 0.042), after adjustment for established clinical covariates. Using a BNP level of 810 pg/mL as the reference value, the hazard ratio increased progressively with higher BNP concentrations, with a more pronounced rise observed at higher BNP ranges. The distribution of BNP values within the study population is displayed alongside the spline curve to illustrate data density across BNP levels (Fig. 5 ). Fig. 5. Open in a new tab Restricted cubic spline analysis depicting the non-linear association between BNP levels and the risk of adverse outcomes. Performance of machine learning models for outcome prediction Four machine learning models—k-nearest neighbors, logistic regression, support vector machine, and random forest—were developed to predict adverse outcomes using key clinical variables. In the test sets, the random forest model achieved the highest discriminative performance among the evaluated models, with an AUC of 0.856, compared with AUC of 0.803 for logistic regression, 0.760 for support vector machine, and 0.792 for k-nearest neighbors (Table 5 ). Decision curve analysis demonstrated that the random forest model provided a higher net clinical benefit than alternative models across a broad range of clinically relevant threshold probabilities, particularly between approximately 0.10 and 0.70. Calibration analysis showed good agreement between predicted and observed event risks for the random forest model, with calibration curves closely aligned with the ideal reference line and a calibration slope near unity (Fig. 6 ). Consistent performance was observed in the training dataset, indicating stable model discrimination. (Supplementary Tables 1 and Supplementary Fig. 2 ) Table 5. .Performance of machine learning models for outcome prediction in the test set. Model AUC 95%CI Sensitivity Specificity Accuracy KNN 0.792 0.756–0.828 0.612 0.897 0.672 LR 0.803 0.768–0.838 0.695 0.808 0.719 SVM 0.760 0.719–0.801 0.666 0.787 0.691 RF 0.856 0.828–0.885 0.719 0.869 0.751 Open in a new tab Abbreviations: AUC, area under the receiver operating characteristic curve; CI, confidence interval. Fig. 6. Open in a new tab ROC curves, decision curve analyses, and calibration plots of four machine learning models evaluated in the test set. Model interpretability using SHAP analysis SHAP analysis was performed to assess feature contributions within the random forest model. Based on mean absolute SHAP values, HFpEF status and BNP level emerged as the most influential predictors of adverse outcomes, followed by atrial fibrillation, age, chronic kidney disease, syncope history, and left ventricular septal thickness (Fig. 7 ). Fig. 7. Open in a new tab SHAP-based interpretation of the random forest model showing feature importance and directionality of effects on outcome prediction. SHAP dependence and summary plots demonstrated that higher BNP levels and the presence of HFpEF were associated with increased predicted risk, whereas the directionality of other clinical features was consistent with their effects observed in Cox proportional hazards analyses. These findings support the clinical interpretability and internal consistency of the machine learning model. Additional SHAP dependence analysis suggested that the contribution of BNP to predicted risk varied according to atrial fibrillation status, providing further insight into risk heterogeneity within the model (Supplementary Fig. 3 ). Discussion In this multicenter cohort study of patients with hypertrophic cardiomyopathy, we provide a detailed evaluation of the prognostic significance of HFpEF using complementary statistical and machine learning approaches, building on established evidence linking HFpEF to adverse outcomes in HCM. The principal findings are as follows. First, HFpEF was highly prevalent among patients with HCM and was independently associated with significantly worse event-free survival, even after rigorous propensity score matching. Second, H₂FPEF score–based risk subgrouping conferred graded prognostic information, with patients classified as HFpEF-High experiencing progressively poorer outcomes. Third, BNP demonstrated a significant non-linear association with adverse outcomes, with risk accelerating at higher BNP ranges. Finally, machine learning models—particularly the random forest model—achieved robust discriminative performance and, through SHAP analysis, identified HFpEF status and BNP as dominant contributors to risk prediction. Together, these findings underscore the HFpEF-defined phenotype as a clinically meaningful, heterogeneous, and distinct risk state in HCM—one that extends beyond the well-established observation that heart failure symptoms predict adverse outcomes—and highlight opportunities for refined, phenotype-driven risk stratification. Heart failure represents a major determinant of morbidity and mortality in HCM, yet its prognostic implications have traditionally been evaluated in a binary manner or confined to advanced disease stages 15 . Importantly, our findings extend beyond the well-known association between heart failure symptoms and poor outcomes in HCM. By applying formal guideline-based HFpEF diagnostic criteria and quantitative scoring frameworks, our study reframes HFpEF in HCM as a distinct phenotypic risk state rather than a mere symptomatic marker, enabling more precise risk stratification. The integrated HFpEF phenotype—encompassing symptoms, preserved systolic function, and objective structural and diastolic abnormalities—carries independent prognostic significance (HR = 2.612 after PSM) that persisted after balancing baseline clinical characteristics, supporting the notion that this HFpEF-defined phenotype captures the net pathophysiological burden of disease rather than the accumulation of conventional risk factors. From a pathophysiological perspective, HFpEF in HCM is mechanistically distinct from HFpEF in the general population and likely represents the integrated effects of disease-specific processes including myocardial disarray, interstitial fibrosis, microvascular ischemia due to impaired coronary flow reserve, progressive atrial myopathy, and neurohormonal activation 16 – 18 . These processes culminate in elevated filling pressures and impaired functional reserve, predisposing patients to recurrent decompensation and adverse outcomes 19 . Unlike comorbidity-driven HFpEF in the general population, HFpEF in HCM reflects the progression of primary myocardial disease through a self-reinforcing cycle of diastolic failure, dynamic obstruction, demand ischemia, and atrial dysfunction. Our results reinforce the clinical relevance of recognizing HFpEF as a distinct and prognostically important manifestation of HCM, one that captures the net pathophysiological burden of disease rather than merely the presence of heart failure symptoms. It is important to emphasize that our study does not merely confirm the established observation that heart failure symptoms predict adverse outcomes in HCM. Rather, we provide several distinct and novel contributions. First, by applying formal, guideline-based HFpEF diagnostic criteria to a large multicenter HCM cohort, we demonstrate that the integrated HFpEF phenotype—defined by the co-occurrence of symptoms, preserved systolic function, and objective structural/diastolic abnormalities—carries independent prognostic significance beyond individual risk factors, as evidenced by the persistent hazard ratio of 2.612 even after rigorous propensity score matching. This phenotypic characterization differs fundamentally from a simple symptom-based observation: it reflects the net pathophysiological burden of disease, including myocardial fibrosis, diastolic failure, and neurohormonal activation, rather than transient symptomatic episodes that may fluctuate with dynamic triggers such as atrial fibrillation or left ventricular outflow tract obstruction. Second, the demonstration that H₂FPEF score–based stratification confers graded prognostic information within HCM is novel, moving beyond the binary “heart failure yes/no” paradigm toward a quantitative, spectrum-based approach to risk assessment. Third, the identification of a non-linear BNP–outcome relationship via restricted cubic spline analysis has practical implications for BNP-based risk thresholds in HCM. Fourth, the application of SHAP-based interpretable machine learning provides transparent, patient-level risk decomposition that reveals feature interactions not captured by traditional Cox models. Beyond a dichotomous classification, our study demonstrates that H₂FPEF score–based risk subgrouping provides incremental prognostic information. Patients classified in the higher-risk subgroup experienced markedly worse event-free survival, indicating a graded relationship between HFpEF burden and adverse outcomes. This finding aligns with the contemporary view of HFpEF as a spectrum rather than a uniform entity and emphasizes the limitations of binary diagnostic frameworks 20 , 21 . In the context of HCM, risk subgrouping based on H₂FPEF scores may be particularly valuable, as patients often exhibit heterogeneous combinations of obstruction, fibrosis, atrial dysfunction, and comorbidities 21 , 22 . Incorporating H₂FPEF score–based risk subgrouping may therefore facilitate more nuanced risk stratification and support individualized management strategies. BNP is a cornerstone biomarker in the diagnosis and management of heart failure and reflects myocardial wall stress and neurohormonal activation 23 . While elevated BNP levels have been associated with adverse outcomes in HCM, most prior studies have assumed a linear relationship 24 . By applying restricted cubic spline modeling, we demonstrate a significant non-linear association between BNP and risk, with a steeper increase in adverse outcomes at higher BNP levels. This non-linear pattern has important clinical implications. It suggests that modest BNP elevations may carry limited incremental risk, whereas marked BNP elevations identify a subgroup of patients with disproportionately high vulnerability. These findings support the use of BNP not only as a diagnostic marker but also as a tool for refined risk stratification, particularly when interpreted across its full distribution rather than relying on arbitrary thresholds. Traditional regression models remain the cornerstone of prognostic research; however, their ability to capture complex, multidimensional relationships may be limited 25 , 26 . In this study, machine learning models—especially the random forest model—demonstrated superior discriminative performance and favorable clinical utility across a wide range of threshold probabilities. Importantly, the application of SHAP analysis enhanced the interpretability of the machine learning model, revealing that HFpEF status and BNP were the most influential predictors of adverse outcomes. The consistency between SHAP-derived feature importance and findings from Cox regression analyses supports the internal validity of the model and mitigates concerns regarding the “black-box” nature of machine learning 27 , 28 . These results suggest that interpretable machine learning approaches may complement conventional statistical methods and offer clinically meaningful insights in HCM risk assessment. Our findings have several potential clinical implications. First, systematic recognition of HFpEF in patients with HCM may facilitate earlier identification of high-risk individuals who warrant closer surveillance and more aggressive management. Second, H₂FPEF score–based risk subgrouping of HFpEF may support individualized therapeutic decision-making and patient counseling. Third, integrating BNP across its non-linear risk spectrum may improve prognostic accuracy beyond conventional cutoff-based approaches. Finally, machine learning models with built-in interpretability may serve as adjunctive tools for personalized risk prediction, bridging the gap between advanced analytics and clinical applicability. Limitations Several limitations should be acknowledged. First, the retrospective observational design precludes causal inference, and residual confounding may persist despite propensity score matching and multivariable adjustment. Second, HFpEF classification was based on a single time-point assessment at baseline, and therefore does not capture the dynamic and episodic nature of heart failure symptoms in HCM. Heart failure in HCM may be triggered by transient events such as atrial fibrillation onset or left ventricular outflow tract obstruction, and symptom status may fluctuate over time. Future prospective studies with serial assessments are needed to characterize the temporal evolution of HFpEF in HCM, identify precipitating factors, and evaluate whether the clinical context of HFpEF onset influences long-term prognosis. Third, BNP measurements were obtained at baseline and may not capture temporal variability or treatment-related changes. Fourth, the H₂FPEF score was originally developed as a diagnostic tool for HFpEF rather than a severity or prognostic index; its application for risk subgrouping in this study, while data-driven, extends beyond its validated use and requires further prospective validation. Fifth, although the machine learning models demonstrated robust internal performance through stratified train-test splitting and calibration analysis, external validation in independent cohorts is essential before clinical implementation and remains the gold standard for establishing model generalizability. Our research group has initiated formal collaborations with two additional cardiovascular centers—Peking University Third Hospital and Guangdong Provincial People’s Hospital—to assemble an independent external validation cohort comprising HCM patients with standardized HFpEF phenotyping. We further plan to explore collaboration with international HCM registries to enable cross-population validation across different ethnic and healthcare system contexts. Until external validation is completed, the proposed models should be considered hypothesis-generating rather than clinically implementable. Sixth, genetic testing data were not uniformly available across all participating centers, as sarcomeric gene mutation testing is not yet part of routine clinical practice at all institutions in China; future studies incorporating genetic data would enable more comprehensive phenotyping. Finally, the study population was derived from tertiary referral centers, which may limit generalizability to broader HCM populations. Conclusion In this multicenter cohort of patients with hypertrophic cardiomyopathy, heart failure with preserved ejection fraction was prevalent and independently associated with adverse outcomes. H₂FPEF score–based risk subgrouping conferred graded prognostic information, and BNP exhibited a non-linear relationship with risk. Integrating survival analyses with spline modeling and interpretable machine learning approaches enhanced individualized risk stratification and may, pending external validation, support refined clinical decision-making in HCM. Supplementary Information Below is the link to the electronic supplementary material. Supplementary material 1 (JPG 206.1 kb) (206.1KB, jpg) Supplementary material 2 (JPG 226.6 kb) (226.6KB, jpg) Supplementary material 3 (JPG 216.0 kb) (216KB, jpg) 41598_2026_46573_MOESM4_ESM.tif (8MB, tif) Supplementary material 4 (TIF 8188.5 kb) 41598_2026_46573_MOESM5_ESM.tif (9.1MB, tif) Supplementary material 5 (TIF 9303.7 kb) 41598_2026_46573_MOESM6_ESM.tif (13.7MB, tif) Supplementary material 6 (TIF 14058.4 kb) Acknowledgements The authors thank all clinicians and staff involved in patient care and data collection at the participating centers. We also acknowledge the patients whose clinical data made this study possible. Abbreviations HFpEF Heart failure with preserved ejection fraction HCM Hypertrophic cardiomyopathy BNP B-type natriuretic peptide LVEF Left ventricular ejection fraction PSM Propensity score matching HR Hazard ratio CI Confidence interval ROC Receiver operating characteristic AUC Area under the curve DCA Decision curve analysis KNN K-nearest neighbors LR Logistic regression SVM Support vector machine RF Random forest SHAP Shapley Additive exPlanations RCS Restricted cubic spline KM Kaplan–Meier NYHA New York Heart Association Author contributions Conceptualization: Weijie Zhang, Mingqi ZhengData curation: Weijie Zhang, Huan Zhao, Zhuchang Tian, Wei Fu, Zongyang Li, Zhouxu Geng, Yuhang He, Honghou He, Peihong Wu, Shengsong Zhu, Min Yang, Jing Chen, Mengshen Wang, Zijia ZhuMethodology/formal analysis/validation: Min Lin, Zhiyuan Zhang, Yanli CuiProject administration: Yanli Cui, Fushi Piao, Mingqi ZhengResources: Min Lin, Yanli Cui, Fushi Piao, Mingqi ZhengSupervision: Fushi Piao, Mingqi ZhengFunding acquisition: Weijie Zhang, Min Lin, Fushi Piao, Mingqi ZhengWriting – original draft: Weijie Zhang, Huan Zhao, Zhuchang TianWriting – review & editing: Fushi Piao, Mingqi Zheng. Funding This work was funded by Hebei Province Higher Education Science and Technology Research Project (CXZX2025030), Hebei Provincial Government Funded Clinical Talent Project (ZF2025062), Guangdong Medical Association Clinical Research Fund (2024QC-B1018). Data availability The datasets used and analyzed during the current study are available from the corresponding author on reasonable request. Declarations Competing interests The authors declare no competing interests. Ethics approval and consent to participate This retrospective study was approved by the Ethics Committees of the participating institutions (The First Hospital of Hebei Medical University, Henan Provincial People’s Hospital, and Fuwai Hospital, National Center for Cardiovascular Diseases). The requirement for written informed consent was waived due to the retrospective nature of the study and the use of anonymized clinical data, in accordance with the Declaration of Helsinki. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Weijie Zhang, Huan Zhao and Zhuchang Tian contributed equally to this work. Contributor Information Yanli Cui, Email: [email protected]. Fushi Piao, Email: [email protected]. Mingqi Zheng, Email: [email protected]. References 1. Ommen, S. R. et al. 2020 AHA/ACC guideline for the diagnosis and treatment of patients with hypertrophic cardiomyopathy: A report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. J. Thorac. Cardiovasc. Surg. 162 , e23–e106. 10.1016/j.jtcvs.2021.04.001 (2021). [ DOI ] [ PubMed ] [ Google Scholar ] 2. Amano, M. et al. Validation of guideline recommendation on sudden cardiac death prevention in hypertrophic cardiomyopathy. JACC Heart Fail 13 , 1014–1026. 10.1016/j.jchf.2024.12.006 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 3. Ommen, S. R. & Semsarian, C. Hypertrophic cardiomyopathy: A practical approach to guideline directed management. Lancet 398 , 2102–2108. 10.1016/s0140-6736(21)01205-8 (2021). [ DOI ] [ PubMed ] [ Google Scholar ] 4. Borlaug, B. A., Sharma, K., Shah, S. J. & Ho, J. E. Heart failure with preserved ejection fraction: JACC scientific statement. J. Am. Coll. Cardiol. 81 , 1810–1834. 10.1016/j.jacc.2023.01.049 (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 5. Brady, P. F. et al. Interactions between atrial fibrillation and natriuretic peptide in predicting heart failure hospitalization or cardiovascular death. J. Am. Heart Assoc. 11 , e022833. 10.1161/jaha.121.022833 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Ommen, S. R. et al. AHA/ACC Guideline for the Diagnosis and Treatment of Patients With Hypertrophic Cardiomyopathy: A Report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. J Am Coll Cardiol 76, e159-e240 (2020). (2020). 10.1016/j.jacc.2020.08.045 [ DOI ] [ PubMed ] 7. Cecchi, F. et al. Coronary microvascular dysfunction and prognosis in hypertrophic cardiomyopathy. N. Engl. J. Med. 349 , 1027–1035. 10.1056/NEJMoa025050 (2003). [ DOI ] [ PubMed ] [ Google Scholar ] 8. Garcia Brás, P. et al. Microvascular dysfunction is associated with impaired myocardial work in obstructive and nonobstructive hypertrophic cardiomyopathy: A multimodality study. J. Am. Heart Assoc. 12 , e028857. 10.1161/jaha.122.028857 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Olivotto, I., Cecchi, F., Poggesi, C. & Yacoub, M. H. Patterns of disease progression in hypertrophic cardiomyopathy: An individualized approach to clinical staging. Circ. Heart Fail. 5 , 535–546. 10.1161/circheartfailure.112.967026 (2012). [ DOI ] [ PubMed ] [ Google Scholar ] 10. Shah, S. J., Katz, D. H. & Deo, R. C. Phenotypic spectrum of heart failure with preserved ejection fraction.. Heart Fail. Clin. 10 , 407–418. 10.1016/j.hfc.2014.04.008 (2014). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Krittanawong, C. et al. Machine learning prediction in cardiovascular diseases: A meta-analysis.. Sci. Rep. 10 , 16057. 10.1038/s41598-020-72685-1 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Petch, J., Di, S. & Nelson, W. Opening the black box: The promise and limitations of explainable machine learning in cardiology. Can. J. Cardiol. 38 , 204–213. 10.1016/j.cjca.2021.09.004 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 13. Ommen, S. R. et al. 2024 AHA/ACC/AMSSM/HRS/PACES/SCMR Guideline for the Management of Hypertrophic Cardiomyopathy: A Report of the American Heart Association/American College of Cardiology Joint Committee on Clinical Practice Guidelines. Circulation 149 , e1239–e1311. 10.1161/cir.0000000000001250 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] 14. Kittleson, M. M. et al. 2023 ACC Expert Consensus Decision Pathway on Management of Heart Failure With Preserved Ejection Fraction: A Report of the American College of Cardiology Solution Set Oversight Committee. J Am Coll Cardiol 81 , 1835–1878. 10.1016/j.jacc.2023.03.393 (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 15. Liang, L. W. et al. Advanced heart failure therapies for hypertrophic cardiomyopathy: State-of-the-art review and an updated analysis from UNOS.. JACC Heart Fail. 11 , 1473–1480. 10.1016/j.jchf.2023.07.004 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Zhang, Y. et al. Prognostic Value of Coronary Angiography-Derived Index of Microvascular Resistance in Patients With Hypertrophic Cardiomyopathy. MedComm 6 , e70289. 10.1002/mco2.70289 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Harrington, J. et al. Heart Failure With Preserved Ejection Fraction Across the Spectrum of Body Mass Index: Hemodynamic and Echocardiographic Characteristics and Outcomes. JACC Heart Fail. 102768 10.1016/j.jchf.2025.102768 (2025). [ DOI ] [ PubMed ] 18. Piazzai, C. et al. Atrial remodelling and dysfunction in hypertrophic cardiomyopathy: Prognostic role and therapeutic target.. Front. Cardiovasc. Med. 12 , 1620313. 10.3389/fcvm.2025.1620313 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Ismail, T. F. et al. Coronary microvascular ischemia in hypertrophic cardiomyopathy - A pixel-wise quantitative cardiovascular magnetic resonance perfusion study.. J. Cardiovasc. Magn. Reson. 16 , 49. 10.1186/s12968-014-0049-1 (2014). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Borlaug, B. A. The pathophysiology of heart failure with preserved ejection fraction.. Nat. Rev. Cardiol. 11 , 507–515. 10.1038/nrcardio.2014.83 (2014). [ DOI ] [ PubMed ] [ Google Scholar ] 21. Triposkiadis, F. et al. The continuous heart failure spectrum: Moving beyond an ejection fraction classification.. Eur. Heart J. 40 , 2155–2163. 10.1093/eurheartj/ehz158 (2019). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Tower-Rader, A. et al. Prognostic value of global longitudinal strain in hypertrophic cardiomyopathy: A systematic review of existing literature. JACC Cardiovasc. Imaging 12 , 1930–1942 (2019). [ DOI ] [ PubMed ] [ Google Scholar ] 23. Castiglione, V. et al. Biomarkers for the diagnosis and management of heart failure. Heart Fail. Rev. 27 , 625–643 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Matthia, E. L. et al. Circulating biomarkers in hypertrophic cardiomyopathy. J. Am. Heart Assoc. 11 , e027618 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Jiang, N., Wu, Y. & Li, C. Limitations of using COX proportional hazards model in cardiovascular research. Cardiovasc. Diabetol. 23 , 219 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. de Groot, L. M., Twisk, J. W. R., Kok, A. A. L. & Heymans, M. W. Incorporating longitudinal variability in prediction models: A comparison of machine learning and logistic regression in a cohort study with long follow-up. Ann. Epidemiol. 110 , 51–65 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 27. Ponce-Bobadilla, A. V., Schmitt, V., Maier, C. S., Mensing, S. & Stodtmann, S. Practical guide to SHAP analysis: Explaining supervised machine learning model predictions in drug development. Clin. Transl. Sci. 17 , e70056 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. van Royen, F. S. et al. In humble defense of unexplainable black box prediction models in healthcare. J. Clin. Epidemiol. 189 , 112013. 10.1016/j.jclinepi.2025.112013 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary material 1 (JPG 206.1 kb) (206.1KB, jpg) Supplementary material 2 (JPG 226.6 kb) (226.6KB, jpg) Supplementary material 3 (JPG 216.0 kb) (216KB, jpg) 41598_2026_46573_MOESM4_ESM.tif (8MB, tif) Supplementary material 4 (TIF 8188.5 kb) 41598_2026_46573_MOESM5_ESM.tif (9.1MB, tif) Supplementary material 5 (TIF 9303.7 kb) 41598_2026_46573_MOESM6_ESM.tif (13.7MB, tif) Supplementary material 6 (TIF 14058.4 kb) Data Availability Statement The datasets used and analyzed during the current study are available from the corresponding author on reasonable request. Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (2.6 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top