Retrospective analysis of the Tempus ECG-AF algorithm for predicting new-onset atrial fibrillation in a large health system

Background
Early identification of patients at increased risk (IR) for new onset or undiagnosed atrial fibrillation (AF) enables timely intervention and improved outcomes. The Tempus ECG-AF device offers a novel approach to risk stratification using standard electrocardiograms (ECGs). We retrospectively evaluated the device's potential clinical impact at a new clinical center.

Methods
A retrospective analysis was conducted on 50,956 ECGs from Apr to Sep 2024 at the hospitals of Bon Secours Mercy Health in VA. Inclusion criteria were defined as patients ≥65 years old, no prior AF, and no pacemaker/ICD. The primary outcome was the clinical documentation of new diagnoses of AF within 12 months.

Results
Of 50,956 ECGs and 48,938 distinct patients with an ECG, 9,323 (19%) met the inclusion criteria. Of these, 1,791 (19%, 3.6% of the entire ECG population) were flagged as IR for AF. The 1-year incidence of AF was 37.1% in the IR group vs 13.9% in the no-IR group (Hazard Ratio 3.44, 95% CI 3.03-3.91, p<0.005). Of the patients flagged as IR, 11% did not have a subsequent ECG, Echo, or Clinician note to evaluate presence of AF, and an additional 763 were managed by non-cardiologists and do not have diagnosis of AF.

Conclusion
The Tempus ECG-AF algorithm effectively identified a substantial subset of older adults at IR of AF. The 11% of increased-risk patients with no documented cardiac follow-up also identifies a population to target with this technology.