A machine learning-based penalized Cox proportional hazards model comprising routinely available clinical and cardiac MRI variables showed strong performance (C index, 0.75) for major adverse cardiac event prediction in patients with hypertrophic cardiomyopathy.
The CoxNet machine learning model demonstrated favorable performance for MACE prediction in HCM patients.
The CoxNet model performance was similar to, though numerically higher than, the 2014 European Society of Cardiology sudden cardiac death risk model.
Key predictors of MACE in the CoxNet model included apical aneurysm, left ventricular end-systolic volume indexed to body surface area, extensive late gadolinium enhancement, native T1 z score, and male sex.
The study cohort consisted of 604 HCM patients who underwent cardiac MRI evaluation over a period from September 2015 to December 2022.
MACEs were defined as a composite outcome of cardiovascular death, resuscitated sudden cardiac death, or heart failure hospitalization.
Cardiac MRI features were highlighted as important contributors to risk stratification in HCM.
This research suggests that a machine learning approach can accurately predict which patients with hypertrophic cardiomyopathy (HCM) — a condition where the heart muscle becomes abnormally thick — are at high risk of serious cardiac events such as cardiovascular death, cardiac arrest, or hospitalization for heart failure. Researchers developed a model using 33 variables drawn from routine clinical assessments, genetic testing, echocardiography (ultrasound of the heart), and cardiac MRI scans in 604 patients followed over a median of three years. The model performed well, with a discrimination score (C index) of 0.75, meaning it correctly ranked higher-risk patients above lower-risk patients about 75% of the time. The most important predictors identified by the model were features detectable on cardiac MRI: the presence of an apical aneurysm (a bulge at the tip of the heart), extensive late gadolinium enhancement (a marker of scar tissue in the heart muscle), and native T1 z score (a measure of heart tissue composition). The size of the heart's left ventricle at its smallest point and male sex were also key predictors. The model performed similarly to the widely used European Society of Cardiology risk calculator, though no statistically significant difference was found between them. This research suggests that incorporating cardiac MRI data into machine learning models could improve how clinicians identify HCM patients most in need of close monitoring or preventive treatments, such as implantable defibrillators. The fact that all variables used were already routinely collected in clinical practice makes this type of model potentially practical to implement without requiring additional testing.
Geyer T, McIntosh C, Sood V, Matos J, McKee H, Hanneman K. (2026). Machine Learning-based Model for Major Adverse Cardiac Event Prediction in Patients with Hypertrophic Cardiomyopathy.. Radiology. Cardiothoracic imaging. https://doi.org/10.1148/ryct.250433