Pathophysiological stratification of left ventricular hypertrophy using artificial intelligence-based electrocardiogram analysis

Abstract Background Left ventricular hypertrophy (LVH) is often caused by abnormal loading conditions (e.g., hypertension, valvular disease). However, in 20–30% of cases, LVH remains unexplained and may result from hypertrophic cardiomyopathy (HCM), cardiac amyloidosis (CA), or Anderson-Fabry disease (AFD). Early identification of these etiologies is crucial for risk stratification, genetic counseling, and timely treatment. Purpose We developed and validated an artificial intelligence (AI) algorithm using electrocardiogram (ECG) data to detect LVH unrelated to abnormal loading conditions. Methods We retrospectively analyzed ECG-transthoracic echocardiogram (TTE) pairs from patients with TTE-confirmed LVH at the University Medical Center Utrecht. ECGs were obtained within 28 days of TTE. The primary endpoint was a composite of CA, AFD, or HCM, identified through text-mining of clinical notes with manual validation. Patients with a left ventricular wall thickness ≥ 15 mm, without hypertension-mediated organ damage, were also classified as HCM. Patients were split (9:1) into training and test sets. A convolutional neural network was trained to predict the presence of the primary endpoint. Results The model was trained on 8,602 ECGs from 1,899 patients and tested on 890 patients (median age 66 [IQR 51–75]; 59.9% male). In the test set, 89 patients (10.0%) had LVH without abnormal loading (40 HCM, 48 CA, 1 AFD). The algorithm achieved an area under the receiver operating characteristic curve of 0.74 (95% CI: 0.68–0.79), a sensitivity of 0.73 (95% CI: 0.63–0.82), and a specificity of 0.64 (95% CI: 0.60–0.68). The positive predictive value was 0.23 (95% CI: 0.18–0.29), while the negative predictive value was 0.94 (95% CI: 0.92–0.96). Conclusion Our algorithm predicts LVH unrelated to abnormal loading conditions using ECG data. As emerging therapies for HCM, CA, and AFD become available, early and accurate detection is increasingly important for timely intervention. With further validation, this approach could improve diagnostic accuracy and support clinical decision-making for patients with unexplained LVH.