Heart failure & cardiomyopathiesDiagnostic or prognostic study
Pediatric ECG-Based Deep Learning to Predict Left Ventricular Dysfunction and Remodeling
- Diagnostic accuracy study
- external validation
- deep learning (convolutional neural network)
Authors' conclusion
This externally validated algorithm shows promise to inexpensively screen for LV dysfunction and remodeling in children
Study context
- Population
- Patients ≤18 years without major congenital heart disease with paired ECG-echocardiograms (≤2 days apart)training 46,261 patients
- Participants
- 12,631 / 2,830 / 5,088
- Comparison
- ECG-based deep learning model for greater than mild left ventricular (LV) dysfunction, hypertrophy and dilation
- Internal testingn = 12,631
- Emergency departmentn = 2,830
- External validationn = 5,088
- Outcome
- AUROC and AUPRC for expert-classified greater than mild LV dysfunction, hypertrophy and dilation (individually and as a composite outcome)
Performance
As reported in the abstract
Model performance with quantitative outcome cutoffs
| Test or model | Composite AUROC | Composite area under the precision-recall curve (AUPRC) | LV dysfunction AUROC | LV dysfunction AUPRC | LV hypertrophy AUROC | LV hypertrophy AUPRC | LV dilation AUROC | LV dilation AUPRC | Composite NPV |
|---|---|---|---|---|---|---|---|---|---|
| Internal testing | 0.88 | 0.43 | 0.92 | 0.23 | 0.88 | 0.28 | 0.91 | 0.47 | 99.0% |
| External validation | 0.86 | 0.39 | 0.94 | 0.32 | 0.84 | 0.25 | 0.87 | 0.33 | 99.2% |
Further findings
LV hypertrophy vs pediatric cardiologist expert benchmark
- Model predictions of LV hypertrophy outperformed the pediatric cardiologist expert benchmark
Adding age and sex to the model
- No benefit to model performance
Original publication
Pediatric ECG-Based Deep Learning to Predict Left Ventricular Dysfunction and Remodeling. Circulation. Published 5 February 2024. doi:10.1161/circulationaha.123.067750
Summarised from the published abstract; every number is checked against it