Atrial Fibrillation Detection by a Smartwatch
How well does a neural network detect atrial fibrillation (AF) using smartwatch data?
Heart rate data (from the photoplethysmography [PPG] sensor) and step count data (from the accelerometer) were input into a neural network via the Cardiogram application. The algorithm was based on a heuristic method that had previously been shown to detect AF using a smartphone, primarily using R-R intervals. The neural network was then validated in a cohort of patients undergoing electrical or chemical cardioversion of persistent AF. The performance of the algorithm was also assessed in ambulatory patients who self-reported persistent AF.
In the cardioversion validation group, the sensitivity, specificity, and c-statistic were 98%, 90%, and 0.97, respectively. The corresponding data in the group that self-reported AF were 68%, 68%, and 0.72, respectively. The positive predictive value was 8%.
The authors concluded that PPG data along with machine-based learning has the potential for AF detection using a smartwatch.
AF is a common arrhythmia, but may escape diagnosis because it may be intermittent and/or may not be associated with symptoms. In patients at risk for thromboembolism, its diagnosis may call for long-term oral anticoagulation. Therefore, ‘passive’ detection (i.e., that which does not require user intervention) of AF is attractive and can potentially bring at-risk patients to a physician’s attention for management. Although the algorithm performed reasonably well within the ‘clean’ context of patients undergoing cardioversion, its discrimination was lacking in the ambulatory cohort. Despite multiple studies utilizing a host of sensors in the quest to identify AF, the clinician ultimately has to confirm the diagnosis with an electrocardiographic tracing, as opposed to probabilistic results of machine learning, before prescribing therapy with potential hazard (e.g., long-term oral anticoagulation).
Clinical Topics: Anticoagulation Management, Arrhythmias and Clinical EP, Prevention, Anticoagulation Management and Atrial Fibrillation, Implantable Devices, EP Basic Science, SCD/Ventricular Arrhythmias, Atrial Fibrillation/Supraventricular Arrhythmias
Keywords: Accelerometry, Anticoagulants, Arrhythmias, Cardiac, Atrial Fibrillation, Electric Countershock, Electrocardiography, Heart Rate, Photoplethysmography, Risk Management, Secondary Prevention, Thromboembolism
< Back to Listings