Artificial intelligence-enabled electrocardiography to predict left ventricular ejection fraction decline in patients with obstructive hypertrophic cardiomyopathy receiving mavacamten
J Choi, M J Yoon, S G Kwak, Y J Cho, J H Kim, J J Park, H K KimAbstract
Background
Mavacamten improves symptoms and reduces left ventricular (LV) outflow tract gradients in patients with obstructive hypertrophic cardiomyopathy (oHCM). However, a subset of patients experience a decline in left ventricular ejection fraction (LVEF) during therapy. Early identification of individuals at risk for systolic dysfunction remains an unmet clinical need. Artificial intelligence–enabled electrocardiography (AI-ECG) provides quantitative indices reflecting LV systolic function.
Purpose
To evaluate whether AI-ECG–derived indices can predict subsequent LVEF decline during mavacamten therapy in patients with oHCM.
Methods
We retrospectively analyzed 134 patients with oHCM treated with mavacamten at two tertiary hospitals between 2023 and 2025. Baseline AI-ECG parameters, including AI-derived LVEF (AI-LVEF), AI-derived LV dysfunction score, and AI-derived LV global longitudinal strain, were assessed prior to mavacamten initiation. Patients were managed according to the standard mavacamten dosing algorithm. The primary endpoint was a decline in LVEF to <55%, defined by the lowest LVEF recorded during follow-up.
Results
During follow-up, 24 patients (17.9%) developed an LVEF <55%. Baseline echocardiographic LVEF did not significantly differ between patients with and without subsequent LVEF decline (64.7 ± 5.0% vs. 66.6 ± 4.4%, P = 0.07). In contrast, patients who experienced LVEF decline had significantly lower baseline AI-LVEF (55.4 ± 10.8% vs. 61.1 ± 7.5%; P = 0.020). AI-LVEF demonstrated moderate discriminatory performance for predicting LVEF decline (AUC 0.73; 95% CI 0.62-0.82). Using an optimal cutoff of AI-LVEF < 64%, patients below this threshold had a six-fold higher risk of subsequent LVEF decline (HR 6.08, 95% CI 1.81-20.5). In Cox proportional hazards analysis, each 5% decrease in AI-LVEF was associated with a 27% higher risk of LVEF decline (HR 1.27, 95% CI 1.07-1.50; P = 0.0067).
Conclusions
Among patients with oHCM treated with mavacamten, baseline AI-ECG identifies individuals at increased risk for subsequent LVEF decline. These findings highlight the potential role of AI-ECG as a noninvasive tool for early safety surveillance and risk stratification during mavacamten therapy.LVEF decline by AI-LVEFFor image description, please refer to the figure legend and surrounding text.