ECG-Age for Cardiovascular Outcome Prediction: Evaluation in an Adult Congenital Heart Disease Cohort

Introduction: Electrocardiogram (ECG)-based estimated age has been proposed as a marker of cardiac well-being, beyond chronological age (CA). Significant gaps between CA and ECG-based age estimates (ECG-Age) have been associated with adverse cardiovascular outcomes [1]. Existing ECG-Age models are dominantly based on deep learning (DL) approaches, which limit their interpretability. Furthermore, their physiological interpretation and generalization across different cohorts remains unclear–whether they reflect intrinsic aging processes or serve as a surrogate for aggregate cardiovascular risk. Subjects with adult congenital heart disease (ACHD) can potentially benefit from the notion of ECG-Age as a measure of their lifelong cardiovascular disease progression and severity. Congenital heart disease affects approximately 1% of newborns worldwide and can range from minor conditions requiring no treatment to severe conditions that may need cardiac surgery after birth, during childhood, or later in adulthood [2]. Patients with CHD require lifelong follow-up to reduce the risk of progressive complications. Electrocardiography is a cost-effective and widely accessible method to assess cardiovascular health. However, the utility of the electrocardiogram (ECG) for predicting adverse cardiac outcomes in the adult congenital heart disease (ACHD) population remained unexplored [3]. Approach: In this study, machine learning (ML) is used to address two research questions. The first aim is to develop a model to estimate ECG-Age from interpretable ECG features. The model was trained and validated on the CODE15 dataset (345 779 ECGs, 233 770 subjects) using 1,316 explainable features of waveform morphology and rhythm ECG features. The trained model was then applied to an adult congenital heart disease (ACHD) dataset (62 330 ECGs, 10 974 subjects) with different ventricular loading conditions. The second ML pipeline focuses on evaluating the developed ECG-Age estimation model utility for outcome prediction. We built a model to predict adverse cardiac outcome–cardiac hospitalization, surgery, intervention, or death–within 3 to 24 months from ECG recording in ACHD using 1,316 ECG features, ECG-Age estimates, and demographics. Main Results: Developed ECG-Age estimation model performance was benchmarked against state-of-the-art deep learning (DL) models [2]. The ML model achieved a correlation coefficient of 0.83 between CA and the estimated ECG-Age on the CODE15 test dataset. ECG-Age results on the ACHD dataset show that the ECG-Age of different ACHD ventricular loading groups is significantly higher than their CA. In outcome prediction using four input feature set scenarios: (1) sex and CA, (2) sex, CA, and ECG-Age, (3) sex, CA, and ECG features, and (4) sex, CA, ECG-Age, and ECG features, models achieved AUROCs of 0.53, 0.59, 0.67, and 0.67, respectively. Conclusion: This study highlights the feasibility of using ECGs in the ACHD population. Although ECG-Age provides incremental predictive value beyond sex and CA alone, it does not provide additional predictive value when ECG features are included in the feature set for predicting adverse cardiac outcomes in the ACHD population. In addition, the observed variability between models suggests that ECG-Age is not uniquely identifiable from the ECG and instead reflects model-dependent regression of age-correlated features. Therefore, deviations of ECG-Age from CA should not be interpreted as a biomarker for cardiovascular risk. Ref : [1] Lima EM, Ribeiro AH, Paixão GMM, Ribeiro MH, Pinto-Filho MM, Gomes PR, Oliveira DM, Sabino EC, Duncan BB, Giatti L, Barreto SM, Meira Jr W, Schon TB, Ribeiro ALP. “Deep neural network-estimated electrocardiographic age as a mortality predictor”. Nature Communications. 2021;12(1). [2] Centers for Disease Control and Prevention, Data and Statistics: Congenital Heart Defects (CHDs) (Sept.02.2025) [3] Mousavi SS, Raskind-Hood CL, Haffner A, Robichaux C, Ivey LC, Sameni R, Book WM. “Electrocardiogram Characteristics and Outcomes in an Adult Congenital Heart Disease Population”, Annals of Noninvasive Electrocardiology, July 2026.