Arrhythmias & electrophysiology
Machine learning to risk stratify individuals for undiagnosed atrial fibrillation at scale using population-wide electronic health records
Authors' conclusion
The FIND-AF machine learning algorithm was scalable in routine EHR data with good discrimination for incident AF.
Study context
- Population
- Individuals aged ≥30 years without diagnosed atrial fibrillation (AF), Clalit Health Services (Israel) electronic health records (EHR), January 1 2019 to June 30, 20194,275 developed AF
- Participants
- 2,166,795 individuals
- Exposure
- FIND-AF vs CHA2DS2-VASc vs C2HEST for predicting undiagnosed AF
- Outcome
- Diagnosis of AF or atrial flutter
- Follow-up
- Within 6 months
Main finding
As reported in the abstract
AF or atrial flutter within 6 months · 6 months
| Group | AUROC | Calibration slope |
|---|---|---|
| FIND-AF | 0.871 (95% CI 0.864-0.877) | 0.73 (95% CI 0.67-0.79) |
| CHA2DS2-VASc | 0.838 (95% CI 0.831-0.845) | 0.63 (95% CI 0.60-0.67) |
| C2HEST | 0.834 (95% CI 0.823-0.844) | 0.62 (95% CI 0.58-0.65) |
AUROC
FIND-AF0.871 (95% CI 0.864-0.877)
CHA2DS2-VASc0.838 (95% CI 0.831-0.845)
C2HEST0.834 (95% CI 0.823-0.844)
Calibration slope
FIND-AF0.73 (95% CI 0.67-0.79)
CHA2DS2-VASc0.63 (95% CI 0.60-0.67)
C2HEST0.62 (95% CI 0.58-0.65)
Observational data: associations, not effects.
Further findings
AUROC
| Outcome | FIND-AF | CHA2DS2-VASc | C2HEST |
|---|---|---|---|
| AUROC in women (subgroup) | 0.883 (95% CI 0.876–0.889) | 0.865 (95% CI 0.858–0.872) | 0.853 (95% CI 0.846–0.861) |
| AUROC in men (subgroup) | 0.857 (95% CI 0.850–0.864) | 0.835 (95% CI 0.828–0.843) | 0.814 (95% CI 0.806–0.822) |
Lift analysis: AF diagnoses identified by screening top 15% FIND-AF risk vs screening by age
- Top 15% FIND-AF risk
- 72%
- Screening by age
- 63%
Original publication
Machine learning to risk stratify individuals for undiagnosed atrial fibrillation at scale using population-wide electronic health records. Eur J Prev Cardiol. Published 4 March 2026. doi:10.
Summarised from the published abstract; every number is checked against it