A pragmatic clinical risk score for adverse outcomes in hypertrophic cardiomyopathy using routine laboratory and bedside variables Authors: Zi-Yi Chen, Jian-Hai Chen Published: 2026 ABSTRACT AbstractBackgroundEarly risk prediction in hypertrophic cardiomyopathy (HCM) remains challenging, particularly for short-term adverse outcomes. We aimed to develop a pragmatic risk score for predicting 1-year composite adverse outcomes using routinely available clinical, laboratory, arrhythmic, and echocardiographic variables.MethodsThis single-center retrospective cohort study included 1,200 Chinese adults with HCM evaluated between 2015 and 2024. The primary endpoint was a 1-year composite of sudden cardiac death (SCD) or SCD-equivalent events, heart failure hospitalization, or all-cause mortality. Candidate predictors were selected based on clinical relevance, routine availability, and prior HCM risk evidence, followed by univariate screening and multivariable logistic regression with stepwise selection. A point-based score was developed and internally validated using 1,000 bootstrap resamples and stratified 5-fold cross-validation.ResultsDuring follow-up, 120 patients experienced the composite endpoint, including heart failure hospitalization in 78 patients, SCD or SCD-equivalent events in 24, and all-cause mortality in 18. Ten predictors were included in the HCM Pragmatic Risk Score: age ≥ 60 years, nonsustained ventricular tachycardia, unexplained syncope, maximal left ventricular wall thickness ≥ 20 mm, left atrial diameter ≥ 45 mm, left ventricular outflow tract gradient ≥ 30 mmHg, left ventricular ejection fraction < 50%, NT-proBNP > 1,000 pg/mL, elevated troponin, and creatinine > 1.2 mg/dL. The score showed moderate discrimination (AUC = 0.75) and good calibration. Event rates increased across risk strata: 3.8%, 16.7%, and 77.8%. Bootstrap validation yielded an optimism-corrected AUC of 0.74.ConclusionThe HCM Pragmatic Risk Score may assist short-term risk stratification and follow-up planning. Because the endpoint was mainly driven by heart failure hospitalization, it should not be interpreted as a dedicated SCD prediction model or stand-alone ICD decision tool. External validation is required.Keywords: Hypertrophic cardiomyopathy, Risk prediction, Hheart failure, Sudden cardiac death, Biomarkers, Echocardiography, Logistic regression, Clinical score Introduction Hypertrophic cardiomyopathy (HCM) is a common inherited cardiac disease (prevalence ~ 1 in 500) characterized by unexplained left ventricular hypertrophy and a heterogeneous clinical course [1]. Many patients remain asymptomatic or mildly symptomatic, but a subset develop severe complications including arrhythmic sudden cardiac death (SCD) and progressive heart failure leading to hospitalization or death [2]. Contemporary management of HCM therefore places great importance on risk stratification to identify patients at high risk for adverse outcomes who may benefit from interventions such as implantable cardioverter-defibrillators (ICDs) or advanced heart failure therapies [3]. Current guideline-recommended risk stratification for SCD in HCM is based on a combination of established clinical markers [4]. The 2014 ESC HCM Risk-SCD model and recent guidelines incorporate factors such as maximal wall thickness, family history of SCD, unexplained syncope, non-sustained ventricular tachycardia on Holter monitoring, left atrial size, left ventricular outflow tract (LVOT) obstruction, and age [5]. These models provide a 5-year SCD risk estimate and guide ICD decision-making. However, they rely predominantly on imaging and clinical history data, and notably do not include potentially informative biomarkers or certain hemodynamic and ECG variables [6]. For example, the presence of extensive myocardial fibrosis on cardiac MRI or an apical aneurysm has been associated with adverse outcomes, but such features have not been integrated into standard risk scores due to limited and inconsistent data [7]. Similarly, circulating cardiac biomarkers are readily available in routine practice but are not part of current risk algorithms. Emerging evidence suggests that routine laboratory biomarkers could enhance risk prediction in HCM [8]. High-sensitivity cardiac troponin (hs-cTn) and N-terminal pro-B-type natriuretic peptide (NT-proBNP) are frequently elevated in HCM patients and correlate with disease severity (e.g. degree of hypertrophy and fibrosis) [9]. Notably, elevated troponin and NT-proBNP levels have been linked to a higher incidence of adverse cardiovascular events in HCM [10]. For instance, a recent cohort study of obstructive HCM found that an age- and sex-adjusted NT-proBNP ratio was independently associated with increased risk of death, transplant, or appropriate ICD shock [11]. These biomarkers reflect myocardial injury and hemodynamic stress, respectively, and may provide additional prognostic information beyond static anatomical risk factors. Likewise, other routine measures—such as renal function or anemia status—may capture comorbidity and end-organ perfusion factors affecting outcomes. Therefore, there is a strong rationale for developing a pragmatic risk stratification tool for HCM that leverages routine bedside variables: easily obtained clinical, laboratory, ECG, and standard echocardiographic data. Such a tool could be more broadly applicable in general practice (including resource-limited settings) and provide timely risk assessment without the need for advanced imaging or genetic testing. We aimed to derive and internally validate a clinical risk score to predict short-term (1-year) adverse outcomes in HCM using these readily available variables. We hypothesized that combining conventional risk markers with routine laboratory and physiologic variables would yield a robust yet practical risk score for use at the point of care. Methods Study design and population We performed a single-center retrospective cohort study of patients with HCM evaluated at a tertiary hospital in China between 2015 and 2024. All participants were recruited from a tertiary hospital in China, and all enrolled participants were Chinese adults. Baseline was defined as the date of the index clinical evaluation after meeting diagnostic criteria for HCM. Clinical and laboratory data were extracted from institutional HCM registries and medical records. Baseline clinical and laboratory data were extracted retrospectively from institutional HCM registries and medical records, and 1-year outcomes were retrospectively ascertained from medical records, documented outpatient follow-up, and telephone contact when necessary. All outcomes were adjudicated by independent cardiologists who were blinded to baseline risk-score information. The study protocol was approved by the Medical Ethics Committee of Jiujiang No.1 People’s Hospital, which served as the central approving ethics committee for the study, and the requirement for written informed consent was waived because of the retrospective study design and the use of anonymized clinical data. Inclusion criteria were adults aged ≥ 18 years with a diagnosis of HCM, defined by unexplained left ventricular wall thickness ≥ 15 mm on echocardiography or cardiac MRI, or ≥ 13 mm in individuals with a confirmed pathogenic sarcomere gene mutation, in the absence of alternative loading conditions such as systemic hypertension or valvular stenosis. Both symptomatic and asymptomatic patients were eligible. Exclusion criteria included phenocopies or alternative cardiomyopathies (e.g., cardiac amyloidosis, Fabry disease, RASopathies, hypertensive heart disease, or athlete’s heart); significant coexisting cardiac conditions (such as more than mild valvular disease or prior myocardial infarction); prior septal reduction therapy before baseline (surgical myectomy or alcohol septal ablation); and presence of an implantable cardioverter-defibrillator (ICD) at baseline, because ICD therapy may modify the natural occurrence and clinical consequences of malignant ventricular arrhythmias and complicate interpretation of SCD-related endpoints. Patients with missing baseline data for candidate predictors or insufficient follow-up (< 1 year) were excluded from the analytic cohort unless a predefined endpoint occurred earlier. The study followed a standardized protocol for HCM diagnosis and data collection to ensure consistency of case identification and variable acquisition. The predefined composite endpoint consisted of SCD, heart-failure hospitalization, or all-cause mortality, as adjudicated during the 1-year follow-up. Baseline variables and measurements We collected four categories of baseline variables: clinical characteristics, laboratory biomarkers, arrhythmia monitoring data, and echocardiographic measurements. All data were recorded at the time of the index evaluation prior to outcome follow-up. Clinical data collected included age, sex, body mass index (BMI), resting blood pressure, heart rate, New York Heart Association (NYHA) class, and relevant history, such as unexplained syncope and family history of SCD. Clinical data collected included age, sex, body mass index (BMI), resting blood pressure, heart rate, New York Heart Association (NYHA) class, and relevant history, such as unexplained syncope and family history of SCD. Additional baseline clinical variables included hypertension, diabetes mellitus, and atrial fibrillation (AF), as documented in the medical record. The presence of AF was determined from clinical history and confirmed on baseline electrocardiogram (ECG). Baseline cardiovascular medications included beta-blockers and anticoagulants. Arrhythmia data were obtained via standardized 48-hour Holter monitoring systems (e.g., GE Healthcare). Non-sustained ventricular tachycardia (NSVT) was defined as ≥ 3 consecutive ventricular beats at ≥ 120 bpm lasting < 30 s, consistent with guideline-based definitions. Laboratory biomarkers were measured from fasting blood samples obtained at baseline. High-sensitivity cardiac troponin T or I (hs-cTnT/I) was quantified, with values exceeding the 99th percentile considered elevated. N-terminal pro-B-type natriuretic peptide (NT-proBNP) was measured as a marker of hemodynamic stress; values were log-transformed for regression analyses and dichotomized at > 1,000 pg/mL for risk-score assignment. Serum creatinine was measured to assess renal function, with values > 1.2 mg/dL considered indicative of renal dysfunction. Estimated glomerular filtration rate (eGFR) was also calculated using the CKD-EPI formula. Transthoracic echocardiography was performed by certified sonographers according to American Society of Echocardiography (ASE) guidelines. Measured parameters included maximal left ventricular (LV) wall thickness (MWT, mm), left atrial anteroposterior diameter and/or volume index, left ventricular ejection fraction (LVEF) using the biplane Simpson’s method, and resting LV outflow tract (LVOT) gradient. Obstructive HCM was defined as a resting LVOT gradient ≥ 30 mmHg, consistent with definitions used in the analytic model. The presence of an apical aneurysm was documented when identified on imaging. Outcome definition The primary outcome was a composite of adverse clinical events occurring within 1 year of baseline and included SCD, heart failure hospitalization, and all-cause mortality. SCD was defined as unexpected death due to a cardiac cause, or aborted cardiac arrest requiring external defibrillation or successful resuscitation, consistent with guideline-based definitions. For patients who received an implantable cardioverter-defibrillator (ICD) during follow-up, appropriate ICD therapy for ventricular fibrillation or sustained ventricular tachycardia (VT) was counted as an SCD-equivalent event. heart failure hospitalization was defined as admission for decompensated heart failure requiring intravenous diuretics or vasoactive therapy, after excluding non-cardiac causes of volume overload or dyspnea. All-cause mortality included death from any cause during the follow-up period. These definitions were harmonized with contemporary ESC and AHA/ACC recommendations. For each patient, the 1-year outcome window was reconstructed from the baseline date to the first qualifying endpoint or to 1 year if no endpoint occurred. All outcomes were adjudicated independently by two cardiologists blinded to baseline characteristics; discrepancies were resolved by consensus. Outcome information was retrospectively reconstructed from medical record review, documented outpatient follow-up records, and telephone contact when necessary. A 1-year observation window was chosen a priori to evaluate short-term adverse outcomes after the index evaluation. This fixed time horizon was intended to support early risk stratification, follow-up planning, and reassessment of clinical status, while limiting the influence of subsequent treatment changes, disease progression, and heterogeneous follow-up duration across the long inclusion period. Patients who underwent septal myectomy, heart transplantation, or other major cardiac interventions during follow-up were censored at the time of intervention. Deaths occurring during or immediately after transplant were classified as all-cause mortality, per prespecified protocol, whereas patients who remained event-free prior to intervention were censored to avoid misclassification. Sample size and statistical analysis Baseline characteristics were summarized as mean ± standard deviation (SD) or median [interquartile range, IQR] for continuous variables—evaluating distributional assumptions using visual inspection of histograms and Q–Q plots—and as frequency (%) for categorical variables. Between-group comparisons (patients with vs. without 1-year events) used Student’s t-test or Mann–Whitney U test for continuous data and the chi-square test (or Fisher’s exact test, when appropriate) for categorical data. Missing data were handled using complete-case analysis. Before model development, the completeness of baseline candidate predictors and 1-year outcome ascertainment was reviewed. Patients with missing data for variables required for model development or without adequate outcome ascertainment were excluded from the analytic cohort, unless a predefined endpoint occurred before 1 year. No multiple imputation was performed. The final analytic cohort therefore consisted of patients with complete baseline candidate predictor data and complete 1-year outcome ascertainment or an earlier qualifying event. For model development, candidate predictors were selected a priori based on clinical relevance, routine availability in bedside or laboratory practice, and prior evidence in HCM risk stratification. These variables included established HCM risk markers, such as age, unexplained syncope, NSVT, maximal LV wall thickness, left atrial size, LVOT obstruction, and LVEF, as well as routinely available biomarkers reflecting myocardial injury, hemodynamic stress, and renal function. Univariate logistic regression was then performed as an initial screening step. Variables with p < 0.10 in univariate analysis were considered for multivariable logistic regression using stepwise selection (entry p < 0.10, stay p < 0.05). Final model development was guided by both statistical significance and clinical plausibility, with the aim of retaining predictors that were routinely available, clinically interpretable, and representative of complementary pathophysiological domains. Multicollinearity was assessed using variance inflation factors (VIF), with VIF > 5 indicating potential concern. The final model retained predictors with p < 0.05. Consistent with this complete-case approach, the modeling dataset included only patients with complete baseline candidate predictor data and complete 1-year outcome ascertainment or an earlier qualifying event. Data partitioning and internal validation strategy. The model was derived using the full eligible cohort (n = 1,200) for model development. No independent external validation cohort or separate prespecified validation dataset was available; therefore, only internal validation was performed. Internal validation was performed using (i) repeated resampling (1,000 bootstrap resamples) to estimate optimism-corrected discrimination and calibration, and (ii) stratified 5-fold cross-validation. For cross-validation, the dataset was randomly partitioned into five approximately equal folds while preserving the event proportion; in each iteration, four folds (approximately 80% of patients) served as the training subset to refit the model, and the remaining fold (approximately 20%) served as the validation subset. Thus, each patient served in the validation subset once and in the training subset four times across the five iterations, and performance metrics were averaged across folds. From the final logistic model, we derived a point-based risk score. Model coefficients were rescaled to a common denominator and rounded to the nearest integer for point assignment. Continuous predictors were first examined for potential nonlinearity using restricted cubic splines. For the point-based score, continuous variables were categorized to improve bedside usability and clinical interpretability. Thresholds were selected according to established clinical conventions, guideline-based definitions, prior HCM literature, and the observed distribution of values in the present cohort. Age ≥ 60 years was used to represent older age; maximal LV wall thickness ≥ 20 mm, left atrial diameter ≥ 45 mm, LVOT gradient ≥ 30 mmHg, and LVEF < 50% reflected structural, obstructive, and systolic abnormalities; NT-proBNP > 1,000 pg/mL represented elevated hemodynamic stress; and creatinine > 1.2 mg/dL represented renal dysfunction. Each patient’s total score was the sum of points across present risk factors. Given the multi-center design, cluster-robust standard errors at the center-level were evaluated as a supplementary sensitivity analysis, which did not materially alter effect estimates. Model performance was assessed by discrimination and calibration. Discrimination was quantified using the C-statistic (area under the ROC curve, AUC) with 95% confidence intervals derived via bootstrap resampling. Calibration was evaluated using the calibration slope, a smooth loess calibration curve, and comparison of observed versus predicted event rates across tenths of predicted risk; the Hosmer–Lemeshow test (10 groups) was also reported. As part of the internal validation framework, bootstrap resampling was used to estimate optimism and generate an optimism-corrected AUC and calibration slope, while stratified 5-fold cross-validation provided an additional assessment of model stability. In addition to internal validation, we performed direct same-cohort comparisons between the proposed HCM Pragmatic Risk Score and two comparator models. First, the ESC HCM Risk-SCD model was used as an established guideline-related comparator. This model was originally developed by O’Mahony et al. to estimate individualized 5-year risk of sudden cardiac death in patients with HCM and incorporates age, maximal left ventricular wall thickness, left atrial diameter, maximal left ventricular outflow tract gradient, family history of SCD, non-sustained ventricular tachycardia, and unexplained syncope. In the present study, the ESC HCM Risk-SCD estimate was calculated using available baseline variables according to the published model formula. Second, we constructed a biomarker-excluded reduced model by refitting the clinical and echocardiographic predictors after excluding NT-proBNP and troponin variables. This reduced model was used as an internal nested comparator to assess whether routinely available biomarkers provided additional discriminatory information beyond conventional clinical and structural predictors. Because the ESC HCM Risk-SCD model was designed for 5-year SCD prediction, whereas the present study evaluated a 1-year composite endpoint, these comparisons were considered exploratory same-cohort benchmarks rather than endpoint-matched external validation. Discriminatory performance was compared using the C-statistic (AUC), and between-model differences were assessed using DeLong’s test. For clinical interpretability, patients were stratified into low, intermediate, and high-risk groups according to the total score, and observed event rates were compared across strata (Cochran–Armitage trend test). All analyses were conducted using SPSS 27 (IBM) and R 4.2.2 (key packages: rms, pROC, and boot). A study flow diagram is provided in Fig. 1. Flowchart depicting patient screening, exclusion criteria, final cohort composition, and follow-up completeness for model derivation Results Patient characteristics After excluding patients with missing baseline candidate predictor data or insufficient follow-up, 1,200 patients were included in the final complete-case analytic cohort. The baseline characteristics of patients with and without 1-year events are presented in Table 1. Among 1,200 patients with HCM, the mean age was 55 ± 15 years, and 65% were male. Obstructive physiology (resting LVOT ≥ 30 mmHg) was present in 40%. Most patients were mildly symptomatic (NYHA I–II: 70%), whereas 30% were NYHA III. Mean maximal wall thickness was 19 ± 4 mm, apical hypertrophy occurred in 8%, and apical aneurysm in 5%. Mean LVEF was 64% ± 7%. Left atrial diameter averaged 44 ± 7 mm; atrial fibrillation was present in 20%, and nonsustained ventricular tachycardia (NSVT) in 15%. Hypertension was present in 28.0%, diabetes mellitus in 12.0%, beta-blocker use in 65.0%, and anticoagulant use in 18.0%. Median NT-proBNP was 600 [200–1,200] pg/mL, and elevated high-sensitivity troponin occurred in 25%. Mean serum creatinine was 1.0 ± 0.3 mg/dL, and mean estimated glomerular filtration rate (eGFR) was 70 ± 20 mL/min. Compared with patients without events (n = 1,080), those with 1-year adverse events (n = 120) were older (60 vs. 54 years, p < 0.001), more frequently obstructive (55% vs. 40%, p = 0.006), and exhibited larger left atria (48 ± 8 vs. 43 ± 6 mm, p < 0.001) with more frequent left atrial enlargement (LA ≥ 45 mm: 60% vs. 38%, p < 0.001) and lower LVEF (62% vs. 65%, p = 0.01). They had greater hypertrophy (21 ± 5 vs. 18 ± 4 mm, p < 0.001), more NSVT (30% vs. 13%, p < 0.001), and more unexplained syncope (20% vs. 5%, p < 0.001). They also had higher prevalences of hypertension (37.5% vs. 26.9%, p = 0.015), diabetes mellitus (18.3% vs. 11.3%, p = 0.024), and anticoagulant use (25.0% vs. 17.2%, p = 0.035), whereas beta-blocker use did not differ significantly (58.3% vs. 65.7%, p = 0.107). Biomarkers were substantially higher (NT-proBNP 1,500 vs. 500 pg/mL; troponin elevation 50% vs. 22%, both p < 0.001). Renal function was worse in patients with events, with higher serum creatinine (1.2 ± 0.4 vs. 1.0 ± 0.3 mg/dL, p < 0.001) and lower eGFR (55 ± 18 vs. 75 ± 20 mL/min, p < 0.001). Blood pressure and heart rate did not differ significantly. Table 1Baseline characteristics stratified by 1-year outcomesVariableOverall (n = 1200)1-year event (+) (n = 120)1-year event (–) (n = 1080)p valueDemographics Age, years55 ± 1560 ± 1454 ± 15< 0.001 Male sex, %6555660.02Clinical subtype and symptoms Obstructive HCM (LVOT ≥ 30 mm Hg), %4055400.006 NYHA class I-II, %7065710.15 NYHA class III, %3035280.15Structural parameters Maximal LV wall thickness, mm19 ± 421 ± 518 ± 4< 0.001 Apical hypertrophy, %81070.30 Apical aneurysm, %5840.12 LVEF, %64 ± 762 ± 865 ± 70.01 LA ≥ 45 mm, %406038< 0.001 Left atrial diameter, mm44 ± 748 ± 843 ± 6< 0.001Arrhythmia and syncope Atrial fibrillation, %2028190.06 NSVT on 48 h Holter, %153013< 0.001 History of unexplained syncope, %9205< 0.001Comorbidities and baseline cardiovascular treatments Hypertension, %28.037.526.90.015 Diabetes mellitus, %12.018.311.30.024 Beta-blocker use, %65.058.365.70.107 Anticoagulant use, %18.025.017.20.035Laboratory parameters NT-proBNP, pg/mL (median [IQR])600 [200-1,200]1,500 [700-3,000]500 [200-1,000]< 0.001 High-sensitivity troponin (ng/mL, median [IQR])0.012 [0.005–0.030]0.025 [0.010–0.050]0.010 [0.005–0.025]< 0.001 Elevated troponin, %255022< 0.001 Serum creatinine, mg/dL1.0 ± 0.31.2 ± 0.41.0 ± 0.3< 0.001 eGFR, mL/min70 ± 2055 ± 1875 ± 20< 0.001 Hemoglobin, g/dL13.8 ± 1.613.2 ± 1.713.9 ± 1.50.03 Mild anemia, %1825170.04Hemodynamics Systolic BP, mm Hg126 ± 15127 ± 16126 ± 150.42 Heart rate, bpm72 ± 1073 ± 1172 ± 100.61Values are presented as mean ± standard deviation, median (interquartile range), or percentage as appropriate. p values compare patients with and without 1-year composite adverse outcomes (sudden cardiac death, heart failure hospitalization, or all-cause mortality)LVOT Left ventricular outflow tract, LV Left ventricle, LVEF Left ventricular ejection fraction, MWT Maximal wall thickness, LA Left atrium, NSVT Non-sustained ventricular tachycardia, NT-proBNP N-terminal pro–B-type natriuretic peptide, eGFR Estimated glomerular filtration rate, BP Blood pressure, bpm Beats per minute, HCM Hypertrophic cardiomyopathy Baseline characteristics stratified by 1-year outcomes Values are presented as mean ± standard deviation, median (interquartile range), or percentage as appropriate. p values compare patients with and without 1-year composite adverse outcomes (sudden cardiac death, heart failure hospitalization, or all-cause mortality) LVOT Left ventricular outflow tract, LV Left ventricle, LVEF Left ventricular ejection fraction, MWT Maximal wall thickness, LA Left atrium, NSVT Non-sustained ventricular tachycardia, NT-proBNP N-terminal pro–B-type natriuretic peptide, eGFR Estimated glomerular filtration rate, BP Blood pressure, bpm Beats per minute, HCM Hypertrophic cardiomyopathy Components of the 1-year composite endpoint During the 1-year follow-up, 120 patients experienced the composite endpoint (Table 2). Among these first qualifying events, heart failure hospitalization occurred in 78 patients, SCD or SCD-equivalent events occurred in 24 patients, and all-cause mortality occurred in 18 patients. Thus, the composite endpoint included both arrhythmic and non-arrhythmic adverse outcomes, but was mainly driven by heart failure hospitalization rather than SCD events. Table 2Components of the 1-year composite endpointEndpoint componentNo. of patients% of total cohort% of composite eventsAny composite endpoint12010.0100.0Heart failure hospitalization786.565.0SCD or SCD-equivalent event242.020.0All-cause mortality as the first qualifying event181.515.0Patients were counted once for the primary composite endpoint according to the first qualifying event within 1 year. Individual components are shown descriptively to clarify the composition of the composite outcomeSCD Sudden cardiac death Components of the 1-year composite endpoint Patients were counted once for the primary composite endpoint according to the first qualifying event within 1 year. Individual components are shown descriptively to clarify the composition of the composite outcome SCD Sudden cardiac death Risk model derivation In univariate analyses, several clinically prespecified routine clinical, structural, arrhythmic, and biomarker variables were associated with the 1-year composite outcome. Variables meeting the prespecified threshold of p < 0.10 were subsequently entered into multivariable logistic regression using stepwise selection. A total of ten variables remained independently associated with adverse outcomes (all p < 0.05), representing complementary domains including clinical status, electrical instability, structural remodeling, hemodynamic load, myocardial injury, and renal function. Older age (≥ 60 years) was associated with an approximately two-fold increase in risk (OR 2.0, 95% CI 1.3–3.1). Two arrhythmic factors—NSVT (OR 2.5, 95% CI 1.5–4.2) and a history of unexplained syncope (OR 3.3, 95% CI 1.8–6.2)—were among the strongest clinical predictors, underscoring the importance of electrical instability in near-term risk. Structural markers of disease severity, including maximal LV wall thickness ≥ 20 mm (OR 2.0, 95% CI 1.2–3.3), left atrial enlargement ≥ 45 mm (OR 1.9, 95% CI 1.1–3.1), and resting LVOT obstruction ≥ 30 mmHg (OR 1.7, 95% CI 1.1–2.8), also independently contributed to risk, consistent with the prognostic impact of myocardial hypertrophy and chronic outflow obstruction. Left ventricular systolic dysfunction emerged as the strongest individual predictor. Patients with LVEF < 50% had more than a four-fold higher risk of 1-year adverse events (OR 4.5, 95% CI 2.3–8.7), reflecting a transition toward early pump failure and heightened vulnerability. NT-proBNP > 1,000 pg/mL (OR 2.4) and elevated troponin (OR 1.8) provided further prognostic information, while mild renal dysfunction (creatinine > 1.2 mg/dL; OR 2.0, 95% CI 1.2–3.3) contributed additional risk, supporting the relevance of extracardiac stress markers in short-term deterioration. A graphical summary of adjusted associations is shown in Fig. 2. To facilitate bedside application and clinical interpretability, we developed a pragmatic point-based score using the prespecified categorical thresholds described in the Methods, by scaling and rounding regression coefficients. LVEF < 50% received the highest weight (3 points), syncope received 2 points, and all other predictors 1 point each (Table 3). Total scores ranged from 0 to 13, with increasing values corresponding to higher estimated 1-year risk. Forest plot of adjusted ORs and 95% CIs for independent predictors of 1-year adverse events in HCM. ORs were obtained from multivariable logistic regression. The dashed line denotes OR = 1.0; the x-axis uses a logarithmic scale Table 3Derivation of the pragmatic HCM risk score: multivariable associations and point assignmentRisk Factor (Unit or Category)Multivariable OR (95% CI)Β-coefficientPoints in ScoreAge ≥ 60 years2.0 (1.3–3.1)0.701 pointNSVT on 48-hour Holter2.5 (1.5–4.2)0.921 pointUnexplained syncope (history)3.3 (1.8–6.2)1.202 pointsMaximal LV wall thickness ≥ 20 mm2.0 (1.2–3.3)0.691 pointLeft atrial diameter ≥ 45 mm1.9 (1.1–3.1)0.641 pointLVEF < 50%4.5 (2.3–8.7)1.503 pointsLVOT gradient ≥ 30 mm Hg1.7 (1.1–2.8)0.531 pointNT-proBNP > 1,000 pg/mL2.4 (1.5–3.8)0.871 pointElevated troponin (hs-cTnT/I)1.8 (1.1–3.1)0.591 pointCreatinine > 1.2 mg/dL2.0 (1.2–3.3)0.691 pointTotal Score Range--0–13 pointsRisk strata in this study were defined empirically based on the observed score distribution and corresponding 1-year event rates: 0–3 points (low risk, ~ 3.8%), 4–6 points (intermediate risk, ~ 16.7%), and ≥ 7 points (high risk, ~ 77.8%). These cutoffs were used for all subsequent analyses (Table 4; Fig. 3). Cutoffs for continuous variables were selected to balance clinical interpretability and bedside usability, based on established clinical conventions, guideline-related definitions, prior HCM literature, and the observed distribution of the study cohortβ logistic regression coefficient, OR Odds ratio, CI Confidence interval Derivation of the pragmatic HCM risk score: multivariable associations and point assignment Risk strata in this study were defined empirically based on the observed score distribution and corresponding 1-year event rates: 0–3 points (low risk, ~ 3.8%), 4–6 points (intermediate risk, ~ 16.7%), and ≥ 7 points (high risk, ~ 77.8%). These cutoffs were used for all subsequent analyses (Table 4; Fig. 3). Cutoffs for continuous variables were selected to balance clinical interpretability and bedside usability, based on established clinical conventions, guideline-related definitions, prior HCM literature, and the observed distribution of the study cohort β logistic regression coefficient, OR Odds ratio, CI Confidence interval Distribution of risk scores and risk stratification The distribution of the HCM Pragmatic Risk Score across the cohort is shown in Fig. 3. Scores ranged from 0 to 9, with a right-skewed pattern typical of clinical risk models. Most patients fell within the lower range of 0–3 points (approximately 65% of the cohort), whereas intermediate (4–6 points) and high-risk scores (≥ 7 points) were progressively less common. Reflecting this distribution and supported by clinical interpretability, the cohort was stratified into three risk groups: low (0–3 points), intermediate (4–6 points), and high risk (≥ 7 points). Observed 1-year adverse event rates increased stepwise across strata (Table 4), from 3.8% in the low-risk group to 16.7% in the intermediate-risk group, and 77.8% in the high-risk group. This monotonic gradient indicates that the score captured clinically meaningful differences in observed short-term risk. The high-risk category, although representing only a small proportion of the cohort, contained a disproportionate number of early adverse events. These findings support the descriptive value of the score for short-term risk stratification, but they should not be interpreted as direct evidence of clinical utility or treatment benefit. Distribution of HCM pragmatic risk scores. Histogram showing the distribution of total risk scores (0–15 possible points) in the study cohort. The score exhibited a right-skewed distribution, with most patients clustering between 0 and 3 points, and progressively fewer patients in higher score ranges. Risk strata used for subsequent analyses were defined as low (0–3), intermediate (4–6), and high risk (≥ 7 points) Table 4Risk strata and observed 1-year event rates (based on score distribution)Risk StratumScore RangeN (estimated from distribution)EventsEvent Rate (%)Low risk0–3780303.8Intermediate risk4–63305516.7High risk≥ 7453577.8Data are observed rates for the composite endpoint (sudden cardiac death, heart failure hospitalization, or all-cause mortality). Risk strata were defined according to the pragmatic HCM risk score cutoffs Risk strata and observed 1-year event rates (based on score distribution) Data are observed rates for the composite endpoint (sudden cardiac death, heart failure hospitalization, or all-cause mortality). Risk strata were defined according to the pragmatic HCM risk score cutoffs Model performance and validation Model performance was assessed using internal validation only; no independent external validation cohort was available or used. In 5-fold cross-validation, the model was refit in each iteration using approximately 80% of patients (training subset) and evaluated in the remaining 20% (validation subset), with each fold serving as the validation subset once; thus, each patient contributed to the validation subset once and to the training subset four times across the five iterations. The final model demonstrated moderate discriminative ability, with an AUC of 0.75 for 1-year outcome prediction (Fig. 4). This level of discrimination indicates that the score captures clinically meaningful variation in short-term risk. The inclusion of troponin and NT-proBNP was associated with improved discrimination: a reduced model excluding biomarker terms yielded an AUC of 0.70, suggesting that myocardial stress and injury markers provided additional discriminatory information beyond clinical and structural predictors. The comparator models were applied as exploratory same-cohort benchmarks: the ESC HCM Risk-SCD model represented an established SCD-focused risk tool, whereas the biomarker-excluded reduced model represented an internally derived nested comparator. In direct same-cohort comparisons, the proposed HCM Pragmatic Risk Score achieved an AUC of 0.75, compared with 0.66 for the ESC HCM Risk-SCD model and 0.70 for the biomarker-excluded reduced model. The proposed model showed significantly higher discrimination than the ESC model (DeLong p = 0.003) and modestly but significantly higher discrimination than the reduced model (DeLong p = 0.041). These findings indicate improved discriminatory performance of the proposed model in this cohort. The model was well calibrated (Fig. 5). Predicted probabilities closely matched observed event rates across deciles, with only mild overestimation at higher predicted probabilities. The calibration slope was 0.95, and the Hosmer–Lemeshow test was non-significant (χ² = 8.5, p = 0.38), indicating no major deviation from perfect calibration. Observed and predicted risks were numerically similar in both low- and intermediate-risk ranges, with greater—but expected—variability in the highest decile due to limited sample size. Together, these findings support the adequacy of the model’s probabilistic estimates across the full spectrum of risk. Internal validation via 1,000-bootstrap resampling showed minimal optimism, yielding an optimism-corrected AUC of 0.74. Similarly, 5-fold cross-validation produced stable discrimination (AUC range 0.73–0.77), confirming the robustness of the model under internal validation of the model and indicating low risk of overfitting. Taken together, the discrimination, calibration, and internal validation results suggest that the score may provide a simple tool for short-term risk stratification in HCM, although its clinical utility requires further external and prospective evaluation. Receiver operating characteristic (ROC) curve for 1-year adverse outcome prediction. The model demonstrated moderate discriminative ability, with an area under the curve (AUC = 0.75). The diagonal dashed line represents the reference line for a non-informative classifier (AUC = 0.50) Calibration plot comparing predicted versus observed 1-year event rates across deciles of predicted risk. The dashed line denotes perfect calibration (predicted = observed). The observed event rates closely approximated the predicted probabilities in the low- and intermediate-risk ranges, with mild overestimation at higher predicted risk levels, consistent with acceptable overall calibration Discussion In this study, we derived and internally validated a novel risk score for 1-year adverse outcomes in HCM using ten routine clinical and investigational variables. The HCM Pragmatic Risk Score showed moderate discriminative ability (AUC = 0.75) and was well calibrated for predicting the composite endpoint of sudden cardiac death, heart failure hospitalization, or all-cause mortality. In direct same-cohort benchmarking, it showed significantly higher discrimination than the ESC HCM Risk-SCD model (AUC 0.75 vs. 0.66; DeLong p = 0.003) and modestly but significantly higher discrimination than the biomarker-excluded reduced model (AUC 0.75 vs. 0.70; DeLong p = 0.041), suggesting that routinely available biomarkers may provide additional discriminatory information for short-term risk prediction [12]. Importantly, this simple point-based tool stratified patients into distinct risk tiers, identifying a small high-risk subgroup (score ≥ 7) with a dramatically elevated 1-year event rate (~ 78%). These findings suggest that combining traditional risk markers with readily available biomarkers may improve short-term risk stratification in HCM, although additional analyses and external validation are needed to determine clinical utility [13, 14]. Our results reinforce many established risk factors while also extending prior risk stratification paradigms. Several predictors in our score—including age ≥ 60, nonsustained VT, unexplained syncope, maximal wall thickness, left atrial enlargement, LVOT obstruction, and reduced LVEF—have long been recognized as markers of high risk in HCM [15, 16]. These factors overlap with those used in current guideline algorithms for sudden death prevention, underscoring that our model’s core elements are grounded in well-validated clinical risk indicators [17]. Notably, we found LVEF < 50% to be the strongest single predictor (OR ~ 4.5), in line with emerging guideline updates that now include systolic dysfunction as a criterion for risk assessment [18]. Unlike existing 5-year SCD risk scores, however, our model targets short-term (1-year) outcomes and a broader composite endpoint [19]. The interpretation of the score should consider the composition of the composite endpoint. Because the endpoint included SCD, heart failure hospitalization, and all-cause mortality, the score should be interpreted as a short-term composite adverse-outcome score rather than a dedicated SCD prediction model. This distinction may also help explain why older age was associated with higher risk in the present model. Although SCD risk in HCM is often emphasized in younger patients, older age may capture vulnerability to heart failure progression, comorbidity burden, renal dysfunction, atrial arrhythmias, and other non-arrhythmic adverse outcomes. Therefore, the proposed score should not be used as a stand-alone basis for ICD decisions, which should continue to rely on guideline-recommended SCD risk assessment and individualized clinical judgment. This distinction addresses an important gap: traditional tools focus on longer-term arrhythmic risk and may not adequately capture near-term heart failure decompensation or mortality [20, 21]. The comparator analyses should be interpreted according to the original purpose of each reference model. The ESC HCM Risk-SCD model was developed to estimate individualized 5-year SCD risk using established clinical and echocardiographic variables, including age, maximal left ventricular wall thickness, left atrial diameter, maximal LVOT gradient, family history of SCD, NSVT, and unexplained syncope [22]. In contrast, our model was designed to predict a broader 1-year composite endpoint including SCD, heart failure hospitalization, and all-cause mortality. Therefore, the lower AUC observed for the ESC model in the present cohort should not be interpreted as poor performance for its intended purpose, but rather as reflecting differences in endpoint definition and prediction horizon. The biomarker-excluded reduced model was an internally derived comparator used to assess whether NT-proBNP and troponin provided additional discriminatory information beyond conventional clinical and structural predictors. A key novelty of our work is the integration of routine biomarkers and non-imaging variables into HCM risk prediction. Prior risk models for HCM have largely excluded laboratory biomarkers, but we demonstrated that including high-sensitivity troponin and NT-proBNP significantly improved risk discrimination [23, 24]. This is biologically plausible: troponin elevation in HCM patients signals active myocardial injury or microinfarction, and has been associated with greater arrhythmic burden and appropriate ICD shocks in recent studies [25]. Similarly, NT-proBNP is a marker of hemodynamic stress that correlates with filling pressures and heart failure severity; it has proven prognostic value for mortality in cardiomyopathies and heart failure populations [26]. In our cohort, patients with adverse outcomes had substantially higher NT-proBNP and more frequent troponin elevation, and the final multivariable model retained both biomarkers as independent predictors. The inclusion of these factors likely enabled our score to capture incipient heart failure and subtler aspects of disease severity that static anatomical metrics alone might miss. Additionally, mild renal dysfunction (elevated creatinine > 1.2 mg/dL) emerged as an independent risk contributor. This finding aligns with the understanding that reduced renal function often reflects systemic comorbidity and poorer reserve, portending worse cardiovascular outcomes in diverse populations [27]. By accounting for extracardiac factors like renal function, our score may better stratify patients who are medically vulnerable despite traditional HCM risk markers. From a clinical perspective, the proposed score may serve as a pragmatic aid for short-term risk stratification. It relies exclusively on information that is routinely obtained in the care of HCM patients, making it potentially feasible for preliminary risk assessment in routine clinical settings. The ability to identify a high-risk subset (e.g. those with scores ≥ 7) at the time of initial evaluation may help clinicians recognize patients who require closer clinical attention. These patients—who in our study had a markedly elevated short-term event rate—may warrant closer monitoring, reassessment of heart failure status, and careful review of established HCM risk markers. Conversely, low-risk patients (scores in the 0–3 range) may be considered for standard follow-up pathways, although management should remain individualized according to clinical context. Notably, heart failure complications are a major component of morbidity in HCM, particularly among those with obstructive physiology [28]. By incorporating heart failure hospitalization into the endpoint and using markers like NT-proBNP and LVEF, our model may capture aspects of short-term clinical deterioration that are not fully reflected by SCD-focused algorithms. This broader prognostic may support discussions regarding short-term risk and the need for closer surveillance. However, the score should not be interpreted as a stand-alone decision-making tool for ICD implantation or other invasive interventions. Such decisions should continue to rely on guideline-recommended SCD risk assessment, established clinical indications, and individualized clinical judgment. Therefore, the main clinical implication of the present model is to provide an accessible framework for identifying patients who may require closer surveillance and further clinical reassessment, rather than to directly determine specific therapeutic strategies. Despite its strengths, our study has several limitations that warrant consideration. First, this was a retrospective study from tertiary centers; therefore, selection and referral bias are possible, and our cohort may not fully represent the broader HCM population in community settings. Notably, outcome follow-up was reconstructed retrospectively from medical records, documented outpatient follow-up, and telephone contact, which may have introduced ascertainment bias despite blinded outcome adjudication. In addition, the model was evaluated using internal validation only, without an independent external validation cohort. Although bootstrap resampling and stratified 5-fold cross-validation suggested reasonable stability, external validation in independent, prospectively enrolled cohorts is required to confirm the model’s generalizability and calibration. Moreover, although the proposed score showed improved discrimination compared with the reference models, discrimination metrics alone do not establish incremental clinical utility. We did not perform formal reclassification analyses, such as NRI or IDI, or decision curve analysis in the present study. Therefore, the comparative findings should be interpreted primarily as evidence of improved discriminatory performance rather than confirmed clinical benefit. Future studies should evaluate reclassification performance, net benefit, and the impact of score-guided management on clinical outcomes. Predictor selection involved univariate screening followed by stepwise multivariable regression. Although this was combined with clinical judgment and internal validation, this approach may introduce model instability and overfitting. External validation and alternative modeling strategies, such as penalized regression or prespecified full-model approaches, are needed to confirm the robustness of predictor selection. Besides, missing data were handled using complete-case analysis rather than multiple imputation. Although this approach ensured that all variables used for model development were directly observed, it may have introduced selection bias and reduced statistical power if data were not missing completely at random. Future studies with prospective data collection and prespecified missing-data protocols are needed to assess the robustness of the findings. Second, the 1-year composite outcome, while clinically relevant, had a relatively low incidence (10%), resulting in small numbers of events in the highest-risk category. This may limit the precision of risk estimates at the extremes and suggests that our proposed risk cutoffs (low, intermediate, high) should be further refined with larger samples. Third, we deliberately restricted candidate variables to those available from routine evaluation—a design choice that enhances practicality but also excludes potentially powerful predictors. In addition, several continuous predictors were categorized to create a simple point-based score. Although this approach improves bedside usability and interpretability, it may lead to loss of information, reduced statistical efficiency, and threshold-dependent risk estimates. Future studies should evaluate whether retaining continuous predictors or using flexible modeling approaches, such as restricted cubic splines or penalized regression, improves model performance and calibration. Finally, our score is intended for short-term risk stratification; its utility for long-term prognostication remains untested. The long inclusion period may have introduced temporal heterogeneity in diagnostic work-up, follow-up patterns, and treatment strategies. Although a fixed 1-year endpoint was used to focus on early risk and reduce heterogeneity from unequal follow-up duration, the long-term prognostic value of the score remains to be validated. A logical next step will be prospective studies to evaluate how use of this risk score in clinical practice might influence decision-making and patient outcomes—for instance, whether it helps triage interventions more effectively or improves follow-up strategies. Addressing these questions will determine the ultimate clinical impact of our proposed tool. Conclusion We developed and internally validated a pragmatic, point-based score to predict 1-year adverse outcomes in hypertrophic cardiomyopathy using routine clinical, laboratory, and echocardiographic data. The score demonstrated good discrimination and calibration, and effectively stratified patients by short-term risk. By incorporating conventional and biomarker-based variables, the score may provide additional discriminatory information for short-term risk stratification. However, its incremental clinical utility, generalizability, and impact on management decisions require confirmation in external validation cohorts and prospective clinical-impact studies. Acknowledgements The authors would like to thank all clinicians and staff at Jiujiang No.1 People’s Hospital for their assistance in data collection and patient management. Abbreviations Atrial fibrillation Area under the receiver operating characteristic curve Body mass index Blood pressure Confidence interval Electrocardiogram Estimated glomerular filtration rate European Society of Cardiology Hypertrophic cardiomyopathy High-sensitivity cardiac troponin T or I Implantable cardioverter-defibrillator Interquartile range Left atrium Left ventricle Left ventricular ejection fraction Left ventricular outflow tract Maximal wall thickness Non-sustained ventricular tachycardia N-terminal pro-B-type natriuretic peptide New York Heart Association Odds ratio Receiver operating characteristic Sudden cardiac death Standard deviation Ventricular tachycardia Authors’ contributions Jianhai Chen. designed the study, collected and analyzed the data, and drafted the initial manuscript. Ziyi Chen. contributed to the study conception and design, supervised the research process, and critically revised the manuscript for important intellectual content. Both authors read and approved the final manuscript. Funding This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Data availability The datasets generated and/or analyzed during the current study are not publicly available due to patient privacy and institutional restrictions, but are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate This single-center retrospective study was approved by the Medical Ethics Committee of Jiujiang No.1 People’s Hospital, which served as the central approving ethics committee for the study (Approval No. JJSDYRMY-YXLL-2025110102). The study was conducted in accordance with the Declaration of Helsinki and relevant institutional guidelines. Given the retrospective study design and the use of anonymized clinical data, the requirement for written informed consent was waived by the ethics committee. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The datasets generated and/or analyzed during the current study are not publicly available due to patient privacy and institutional restrictions, but are available from the corresponding author upon reasonable request.