Electrocardiogram-Based Deep Learning to Predict Mortality in Repaired Tetralogy of Fallot
Authors' conclusion
This externally validated AI-ECG model may complement imaging biomarkers to improve risk stratification in patients with rTOF.
Study context
- Population
- Patients with repaired tetralogy of Fallot (rTOF)
- Participants
- 1,054 vs 335
- Comparison
- Internal testing vs External validation
- Internal testingn = 1,054
- External validationn = 335
- Outcome
- 5-year mortality prediction (AUROC, AUPRC)
- Follow-up
- 5 years
Methodological detailsdevelopment and external validation, multicentre registry (INDICATOR)
- Development and external validation
- Multicentre registry (INDICATOR)
Performance
As reported in the abstract
AI-ECG prediction of 5-year mortality
| Test or model | AUROC | AUPRC |
|---|---|---|
| Internal testing | 0.83 | 0.18 |
| External validation | 0.81 | 0.21 |
AUROC
Internal testing0.83
External validation0.81
AUPRC
Internal testing0.18
External validation0.21
Further findings
Cohort mortality
Comparison with other predictors
- AI-ECG performed similarly to biventricular global function index and outperformed QRS duration
- AI-ECG, but not QRS duration, was a significant independent predictor of shorter time-to-death in Cox model with biventricular global function index, in internal and external cohorts
Saliency mapping high-risk ECG features
- QRS fragmentation, wide and low-amplitude QRS complexes, flattened T waves
Original publication
Electrocardiogram-Based Deep Learning to Predict Mortality in Repaired Tetralogy of Fallot. JACC Clin Electrophysiol. Published 18 September 2024. doi:10.
Summarised from the published abstract; every number is checked against it