Arrhythmias & electrophysiologyDiagnostic or prognostic study
Deep Learning of Electrocardiograms in Sinus Rhythm From US Veterans to Predict Atrial Fibrillation
- Prognostic study
- multicentre
- external validation
Authors' conclusion
Deep learning of outpatient sinus rhythm ECGs predicted AF within 31 days in populations with diverse demographics and comorbidities.
Study context
- Population
- Outpatients with 12-lead electrocardiograms (ECGs) in sinus rhythm, 6 US Veterans Affairs (VA) hospital networks and 1 non-VA academic medical center, 1987–2022
- Comparison
- VA sites vs Non-VA site
- VA sitesn = 907,858 ECGs
- Non-VA siten = 72,483 ECGs
- Outcome
- Prediction of atrial fibrillation (AF) by a deep learning model (convolutional neural network) applied to sinus rhythm ECGs
- Follow-up
- Within 31 days
Performance
As reported in the abstract
AF within 31 days of a sinus rhythm ECG · 31 days
| Test or model | AUROC | Accuracy | F1 score |
|---|---|---|---|
| VA sites | 0.86 (95% CI, 0.85-0.86) | 0.78 (95% CI, 0.77-0.78) | 0.30 (95% CI, 0.30-0.31) |
| Non-VA site | 0.93 (95% CI, 0.93-0.94) | 0.87 (95% CI, 0.86-0.88) | 0.46 (95% CI, 0.44-0.48) |
- Well calibrated, Brier score across all sites
- 0.02
Further findings
Number needed to screen among individuals deemed high risk
- Testing sensitivity 25%
- 2.47
- Testing sensitivity 75%
- 11.48
Performance in subgroups (Black, female, age <65 years, CHA2DS2-VASc ≥2)
- Model performance similar in patients who were Black, female, younger than 65 years or with CHA2DS2-VASc scores of 2 or greater
Original publication
Deep Learning of Electrocardiograms in Sinus Rhythm From US Veterans to Predict Atrial Fibrillation. JAMA Cardiol. Published 1 December 2023. doi:10.1001/jamacardio.2023.3701
Summarised from the published abstract; every number is checked against it