Congenital heart diseaseDiagnostic or prognostic study
Electrocardiogram-based deep learning to predict mortality in paediatric and adult congenital heart disease
- Prognostic model development and validation study
- single-centre
- convolutional neural network
- 50% training / 50% test split
- temporal validation
Authors' conclusion
This temporally validated model shows promise to inexpensively risk-stratify individuals with CHD across the lifespan
Study context
- Population
- Paediatric and adult patients with congenital heart disease, electrocardiogram (ECGs) from cardiology clinic at Boston Children's Hospital (ECG age range 0–92 years)39,784 patients per cohort
- Comparison
- AI-enhanced ECG vs age at ECG, QRS duration and LVEF
- Outcome
- 5-year mortality prediction (area under ROC and precision-recall curves)
- Follow-up
- 5 years
Performance
As reported in the abstract
5-year mortality
| Test or model | AUC (ROC) | AUC (precision-recall) |
|---|---|---|
| Model | 0.79 (95% CI 0.77-0.81) | 0.17 (95% CI 0.15-0.19) |
Further findings
Prediction by congenital heart disease (CHD) lesion (subgroup)
- AI-ECG outperformed LVEF across a wide range of CHD lesions
Longer-term mortality (Kaplan-Meier)
- Predictive value for longer-term mortality in overall cohort and lesion subgroups
Salient ECG features
- Precordial lead QRS complexes most salient; high-risk features wide and low-amplitude QRS; QRS fragmentation in tetralogy of Fallot
Original publication
Electrocardiogram-based deep learning to predict mortality in paediatric and adult congenital heart disease. Eur Heart J. Published 10 October 2024. doi:10.1093/eurheartj/ehae651
Summarised from the published abstract; every number is checked against it