Introduction:
AF screening can detect undiagnosed AF, but concerns exist regarding false positives and downstream consequences. AF screening using single-lead handheld electrocardiograms (1L ECGs) is feasible, but diagnostic performance in a pragmatic screening setting is not well-understood.
Methods:
VITAL-AF is a cluster-randomized trial (NCT035115057) of population-based AF screening with 30-second 1L ECGs. Individuals aged ≥65 years receiving primary care in a study practice were eligible, with practices randomized to screening (n=8) or usual care (n=8). Screening was performed with the AliveCor Kardia 1L ECG (AliveCor US) administered by medical assistants. In addition to the AliveCor interpretation, all 1L ECGs were overread by cardiologists. We assessed the distribution of automated interpretations and predictive values using cardiologist 1L ECG overread as the gold standard.
Results:
Screening generated 38,190 tracings representing 16,496 individuals read by 13 cardiologists (median time to read: 10h, quartile 1: 5, quartile 3: 19). We removed 1,554 tracings overread as uninterpretable, resulting in 36,636 tracings in the analysis. A total of 27,251 (74.4%) tracings had an AliveCor interpretation of normal, 6,223 (17.0%) unclassified, 2,420 (6.6%) possible AF, and 742 (2.0%) no interpretation. The positive predictive value (PPV) of possible AF by AliveCor was 76.2% (95% CI 74.5-77.9,
Figure
). The negative predictive values (NPVs) were 99.7% (95% CI 99.6-99.8) for normal, 92.3% (95% CI 91.6-92.9) for unclassified, and 87.1% (95% CI 84.4-89.4) for no interpretation. Of 2,502 tracings overread as AF, 482 (19.3%) were unclassified by AliveCor.
Conclusions:
In an elderly primary care population, the PPV of possible AF by AliveCor algorithm was modest, whereas the NPV of a normal interpretation was very high. Unclassified tracings were common and enriched for AF. Abnormal findings, including unclassified 1L ECG tracings, merit follow-up testing.