Evaluation and application of electrocardiographic age model for children

Area of Science:

• Biomedical Engineering
• Pediatric Cardiology
• Artificial Intelligence in Medicine

Background:

• Electrocardiogram-age (ECG-age), derived from deep neural networks (DNNs), is a health indicator primarily studied in adults.
• Pediatric heart development has unique trajectories, necessitating specialized models distinct from adult-derived ones.

Purpose of the Study:

• To evaluate and propose a pediatric-specific ECG-age model for clinical use.
• To assess the accuracy of adult vs. pediatric ECG-age models in children.
• To explore the association between ECG-age discrepancies and pediatric diseases/abnormalities.

Main Methods:

• Developed and validated pediatric-specific ECG-age models using 58,672 ECGs from healthy children.
• Compared performance against a pre-trained adult model and a retrained adult model.
• Analyzed ECG-age in large cohorts of pediatric diseases (209,644 ECGs) and abnormalities (102,441 ECGs).
• Investigated ECG-age correlation with postoperative outcomes in tetralogy of Fallot (TOF) surgery.

Main Results:

• Adult-derived ECG-age models showed limited applicability in children.
• Dedicated pediatric models achieved high accuracy in age assessment (MAE: 7.878±6.658 months, R²=0.821).
• A significant ECG-age gap correlated with increased disease presence and abnormalities.
• ECG-age showed significant correlation with postoperative complications in TOF surgery patients.

Conclusions:

• Pediatric-specific ECG-age models accurately reflect cardiac development in children.
• ECG-age serves as a potential digital biomarker for detecting pathological conditions in pediatric patients.
• These models can support clinical decision-making and monitoring in pediatric cardiology.