Improving Risk Stratification in Hypertrophic Cardiomyopathy: A Novel Score Combining Echocardiography, Clinical, and Medication Data

Marion Taconné; Valentina D. A. Corino; Annamaria Del Franco; Sara Giovani; Iacopo Olivotto; Adrien Al Wazzan; Erwan Donal; Pietro Cerveri

cs.LGMar 27, 2026

Improving Risk Stratification in Hypertrophic Cardiomyopathy: A Novel Score Combining Echocardiography, Clinical, and Medication Data

Authors: Marion Taconné, Valentina D. A. Corino, Annamaria Del Franco, Sara Giovani, Iacopo Olivotto, Adrien Al Wazzan, Erwan Donal, Pietro Cerveri,

Organizations: Department of Electronics, Information and Bioengineering (DEIB), Politecnico di Milano, 20133 Milano, Italy · CardioTech Lab, IRCCS Centro Cardiologico Monzino, 20138 Milano, Italy · Cardiomyopathy Unit, Careggi University Hospital, 50134 Florence, Italy · Univ Rennes, CHU Rennes, Inserm, LTSI – UMR 1099, 35000 Rennes, France ·

Abstract

Hypertrophic cardiomyopathy (HCM) requires accurate risk stratification to inform decisions regarding ICD therapy and follow-up management. Current established models, such as the European Society of Cardiology (ESC) score, exhibit moderate discriminative performance. This study develops a robust, explainable machine learning (ML) risk score leveraging routinely collected echocardiographic, clinical, and medication data, typically contained within Electronic Health Records (EHRs), to predict a 5-year composite cardiovascular outcome in HCM patients. The model was trained and internally validated using a large cohort (N=1,201) from the SHARE registry (Florence Hospital) and externally validated on an independent cohort (N=382) from Rennes Hospital. The final Random Forest ensemble model achieved a high internal Area Under the Curve (AUC) of 0.85 +- 0.02, significantly outperforming the ESC score (0.56 +- 0.03). Critically, survival curve analysis on the external validation set showed superior risk separation for the ML score (Log-rank p = 8.62 x 10^(-4) compared to the ESC score (p = 0.0559). Furthermore, longitudinal analyses demonstrate that the proposed risk score remains stable over time in event-free patients. The model high interpretability and its capacity for longitudinal risk monitoring represent promising tools for the personalized clinical management of HCM.

Figures & tables

Study	Measures	Output	Clinical flaws
[ 8 ]	ECG-derived electrical phenotypes	ECG pattern subgroups in HCM (unsupervised phenotyping)	No direct risk stratification output; no time-to-event modeling; no longitudinal assessment of ECG evolution
[ 4 ]	ECG-derived arrhythmic risk markers	Associations between ECG repolarization markers and myocardial disarray/fibrosis	Mechanistic insight rather than patient-level risk; no risk stratification; No ICD decision support
[ 15 ]	Conventional echocardiography measurements; basic clinical variables	Predicted probability or class of composite adverse cardiac events	Static baseline risk estimation; no longitudinal risk evolution; limited generalizability
[ 9 ]	LGE extent and location; CMR feature-tracking strain; clinical and echocardiographic variables	Static risk score for long-term major adverse cardiovascular events; stratification into low/medium/high risk groups	Dependence on CMR imaging; static risk estimation without temporal updating
[ 10 ]	Myocardial fibrosis and scar patterns from LGE-CMR images; CMR-derived structural measures; Echo-derived features	Static probability of arrhythmic sudden cardiac death	Computationally intensive; reliance on advanced CMR imaging; static risk estimation

TABLE I: Summary of prior machine-learning approaches for risk stratification in hypertrophic cardiomyopathy (HCM), highlighting quantified measures, model outputs, and clinical limitations.

Fig. 1: Methodological steps separated in training-test (top) and validation phase (bottom) with the survival analysis and longitudinal analysis. The five training models are gathered to form a final ensemble model.

Fig. 2: Mean ROC curves of the nested cross-validation of the LR, GB, RF and SVM models compared against the ESC risk score.

Fig. 3: Average SHAP summary plot of the global feature importance and directional impact for the five folds of the RF model. Echocardiographic features are written in blue. NYHA: New-York Heart Association index, LVESV or LVESDV: Left Ventricle End Systolic/Diastolic Volume, LVEF: Left Ventricle Ejection Fraction, PWT: Posterior Wall Thickness, sep E/E’: ratio of the E wave with septal E’, E’:peak E’ velocity, LA diameter: Left Atrium diameter, LVOT, LV Outflow Tract Obstruction, LVIDd/s: LV Internal Dimension at end-diastole/systole, MWT: Maximum Wall Thickness, lat E’: lateral E’, dias/sys BP: diastolic/systolic Blood Pressure, FH HCM: family history of HCM, NSVT: non sustained ventricular tachycardia, RAAS : Renin-angiotensin-aldosterone system medication.

Model	Sensitivity	Specificity	Balanced Accuracy	F1-Score	AUC
RF	$0.738\pm 0.042$	$0.823\pm 0.033$	$0.780\pm 0.019$	$0.656\pm 0.027$	$0.852\pm 0.017$
GB	$0.725\pm 0.060$	$0.818\pm 0.012$	$0.771\pm 0.029$	$0.644\pm 0.036$	$0.850\pm 0.027$
LR	$0.767\pm 0.056$	$0.763\pm 0.030$	$0.765\pm 0.015$	$0.626\pm 0.014$	$0.837\pm 0.020$
SVM	$0.731\pm 0.043$	$0.813\pm 0.040$	$0.772\pm 0.020$	$0.645\pm 0.028$	$0.830\pm 0.015$
Ensemble RF	0.351	0.877	0.614	0.342	0.723

TABLE II: Sensitivity, specificity, accuracy F1 score and AUC of the 4 types of model on nested-CV (SHARE dataset): Gradient Boosting (GB), Support Vector Machine (SVM), Logistic Regression (LR) and Random Forest (RF). The last row correspond to the ensemble RF on the external validation cohort (Rennes).

Fig. 4: Longitudinal analysis visualization: the model’s prediction value (blue-to-red scatters) at each exam and the linear regression colored by their slope value (red-to-green).Only patients of the first test set fold who experienced endpoint with at least 5 exams are plotted (P1-P45). Four patients (P4, P17, P32 and P41) were zoomed to show the dynamic of the slopes.

Fig. 5: Distribution of RF ensemble mean prediction on the external validation population (Rennes) separated depending on the true label (event 0/1), compared with the ESC score ( right ). Kernel Density Estimations (KDE) were plotted over both histograms.

Fig. 6: Survival curves of the external validation population (Rennes) separated depending on the ensemble RF model prediction (mean probability) at 5 years (predicted 0/1) and ESC score (threshold to separate the population computed on the SHARE dataset).

Clinical Outcome Prediction

Clinical outcome prediction uses patient records, imaging, signals, and biomarkers to estimate risks, disease trajectories, treatment response, and survival.

See all 261 papers

Pre-AF 13: An Interpretable Atrial Fibrillation Risk Score Mined from Discharge Reports
Jun 9, 2026Olga Shakhmatova, Dmitrii Kriukov, Daniil Larionov +10Clinical Risk Prediction Cardiovascular Risk Prediction
National Medical Research Center of Cardiology named after Academician E.I. Chazov, Moscow, Russia · Skolkovo Institute of Science and Technology (Skoltech), Moscow, Russia · Artificial Intelligence Research Institute (AIRI), Moscow, Russia +8
Detecting Structural Heart Disease from Electrocardiograms via a Generalized Additive Model of Interpretable Foundation-Model Predictors
Mar 3, 2026Ya Zhou, Zhaohong Sun, Tianxiang Hao +1Electrocardiogram Classification Electrocardiography
Department of Information Center, Fuwai Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100037, China · The Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen, 518055, China · National Clinical Research Center for Cardiovascular Diseases, Fuwai Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, National Center for Cardiovascular Diseases, Beijing, 100037, China
A Multimodal and Explainable Machine Learning Approach to Diagnosing Multi-Class Ejection Fraction from Electrocardiograms
Apr 17, 2026Catherine Ning, Yu Ma, Cindy Beini Wang +4Left Ventricular Ejection Fraction Estimation Multimodal Classification
Operations Research Center, Massachusetts Institute of Technology, Cambridge, MA, USA · Wisconsin School of Business, University of Wisconsin–Madison, Madison, WI, USA · Heart and Vascular Institute, Hartford Hospital, Hartford, CT, USA

Jun 9, 2026J/K move · Enter open · S save

Pre-AF 13: An Interpretable Atrial Fibrillation Risk Score Mined from Discharge Reports

Olga Shakhmatova, Dmitrii Kriukov, Daniil Larionov, Nikita Khromov+9

National Medical Research Center of Cardiology named after Academician E.I. Chazov, Moscow, Russia · Skolkovo Institute of Science and Technology (Skoltech), Moscow, Russia · Artificial Intelligence Research Institute (AIRI), Moscow, Russia+8

Clinical Risk Prediction Cardiovascular Risk Prediction

Open paper AskPDF arXiv

Background. Atrial fibrillation (AF) is the most prevalent cardiac arrhythmia and a major determinant of prognosis. Established AF risk scores rely on factors (older age, hypertension) nearly ubiquitous among patients with cardiovascular disease (CVD), offering limited stratification in this high-risk group. Most target long-term (5-10 year) rather than medium-term prediction. We developed interpretable ML models predicting AF risk over a 24-month and entire follow-up horizon in CVD patients using routinely collected hospital data. Methods. Single-center retrospective study of electronic health records from the National Research Cardiology Center (Russia) for patients aged >=18 with CVD but without pre-existing AF, hospitalized more than once between January 2012 and May 2019. A custom NLP pipeline transformed unstructured discharge reports into 73 structured features, combining a rule-based parser with transformer-based NER. Using LightAutoML we built a full model (73 features), a simple model (reduced subset), and a linear model for a bedside risk score. Performance was assessed by ROC AUC, compared with CHARGE-AF, C2HEST, MHS, and HAVOC, and interpreted via SHAP. Results. Of 80,576 records from 45,000 patients, 17,562 met inclusion criteria; 1,438 (8.19%) developed AF. The full model reached ROC AUC 0.735 (24-month) and 0.696 (entire follow-up); the simple model was nearly identical (0.725, 0.696). All non-linear models outperformed the four clinical risk scores (ROC AUC 0.53-0.64). The simple model uses 13 features and is named Pre-AF 13. SHAP identified age and left atrial volume as dominant predictors. A linear risk score (Pre-AF 9) stratified observed 24-month AF incidence from ~7% to 36%. Conclusion. Interpretable ML models built from routinely collected EHR data identify high-AF-risk CVD patients, outperforming established clinical risk scores.

Abstract

Clinical Outcome Prediction

Explore similar work

Pre-AF 13: An Interpretable Atrial Fibrillation Risk Score Mined from Discharge Reports

Detecting Structural Heart Disease from Electrocardiograms via a Generalized Additive Model of Interpretable Foundation-Model Predictors

A Multimodal and Explainable Machine Learning Approach to Diagnosing Multi-Class Ejection Fraction from Electrocardiograms