Cardiac monitoring and implantable cardiovascular electronic devices can increase the detection of atrial fibrillation (AF), but the devices have limitations. New smartphone tools aim to record an electrocardiogram (ECG) strip and make an automated diagnosis to overcome these limitations. Within the most extensive study to date, in the Canadian Journal of Cardiology, published by Elsevier, the use of these devices is challenging in patients with abnormal ECGs. Investigators say that improved algorithms and machine learning may help these tools provide more accurate diagnoses.
The study involved 734 consecutive hospitalized patients who underwent a 12-lead ECG, followed by a 30-second Apple Watch recording. The smartwatch classified automated single-lead ECG AF detections as “no signs of atrial fibrillation,” “atrial fibrillation,” or “inconclusive reading.” Electrophysiologists then looked at the smartwatch recordings and conducted a blind interpretation, assigning each tracing a diagnosis of “AF,” “absence of AF,” or “diagnosis unclear.” A second electrophysiologist blindly interpreted 100 randomly selected traces to determine the extent to which the observers agreed.
In 20% of patients, the smartwatch ECG failed to produce an automatic diagnosis.
The smartphone app correctly identified 78% of the patients who were in AF and 81% who were not in AF. The electrophysiologists identified 97% of the patients who were in AF and 89% who were not.
Smartwatch automated detection algorithms are based solely on cycle variability. Any algorithm limited to the analysis of cycle variability will have poor accuracy in detecting AT/AFL. Machine learning approaches may increase smartwatch AF detection accuracy.
