The pediatric-specific feature-enhanced ECG-age model reduced mean absolute error in age prediction to 7.878 months compared to 48.814 years with the adult-derived model in children under 84 months.
Observational (n=58,672)
No
Does a dedicated pediatric-specific ECG-age model improve age prediction accuracy compared to an adult-derived model, and can the ECG-age gap signal pathological conditions in children?
A dedicated pediatric ECG-age model accurately tracks cardiac maturation, and its deviation from chronological age can serve as a digital biomarker for pediatric diseases and postoperative complications.
Effect estimate: MAE 7.878 months vs. 48.814 years (approx. 585.77 months)
Absolute Event Rate: 7.878% vs 585.77%
ECG-age, derived from ECG signals using deep neural networks (DNNs), correlates with health status but has been predominantly studied in adults, neglecting the unique development trajectories of pediatric hearts. This study evaluates and proposes a pediatric-specific ECG-age model for clinical application. Validated on 58,672 ECGs from healthy children, three approaches were assessed: a pre-trained adult model, the same model retrained on pediatric data, and feature-enhanced models which was used to enhance the interpretability of the model. Results demonstrate limited applicability of adult-derived models, while dedicated pediatric models achieved accurate age assessment (MAE: 7.878 ± 6.658 months, R²=0.821), further confirmed on external data. Analysis of 209,644 ECGs across 21 pediatric diseases and 102,441 ECGs with 20 abnormalities revealed that an large ECG-age gap between chronological age indicates stronger associations with diseases or abnormalities. In a specific scenario study, ECG-age significantly correlated with postoperative complications in tetralogy of Fallot (TOF) surgery children. These findings indicate that pediatric-specific ECG-age models can effectively measure children's cardiac development and signal pathological conditions, supporting their potential as a widely applicable clinical digital biomarker in pediatrics.
Cao et al. (Fri,) conducted a observational in Children under 84 months with normal ECGs and no clinical diagnoses (n=58,672). Pediatric-specific ECG-age model (feature-enhanced via Recursive Feature Elimination) vs. Pre-trained adult-derived ECG-age model was evaluated on Accuracy of ECG-age prediction compared to chronological age (MAE 7.878 months vs. 48.814 years (approx. 585.77 months)). The pediatric-specific feature-enhanced ECG-age model reduced mean absolute error in age prediction to 7.878 months compared to 48.814 years with the adult-derived model in children under 84 months.