With the prevalence of personal ECG devices already on the rise and the Apple Watch Series 4 recently hitting the market, the number of daily ECG recordings in the developed world is poised to explode. While some individuals will undoubtedly benefit from enhanced diagnosis of cardiac arrhythmias, the consequences of false alarms are likely to be detrimental to patients and clinicians alike. The published specificity of the Alivecor AF algorithm is 76%. If this is representative of other devices including the Apple Watch, which is anticipated to sell in the millions, the consequence is likely to be hundreds of thousands of unnecessary consultations each year.
The small number of patients within our NHS trust who have purchased AliveCor devices at their own discretion are already causing logistical difficulties. Primary care physicians, cardiac physiologists and cardiologists are all understandably reluctant to take responsibility for ECG recordings that they did not order and over whose quantity they have no control. The current default position, therefore, is for patients presenting to primary care with potentially abnormal recordings from personal ECG devices to be referred for routine review by a cardiologist. If the acquisition rate of personal ECG devices follows the anticipated trajectory, our waiting lists will soon become too long for this arrangement to remain untenable.
“Expert-level” arrhythmia detection using a deep learning approach offers a potential solution here, by virtue of reducing the false positive rate of arrhythmia detection. However, the opacity of deep learning algorithms makes it more difficult to anticipate their failures than with rule-based algorithms, which is concerning from a clinician standpoint. We argue that the optimal management of personal ECG device recordings represents an urgent research question.