Work

Accuracy of Apple Watch for Detection of Atrial Fibrillation

Year: 2020

Type: article

Abstract: atrial fibrillation I n the Apple Heart Study, 34% of individuals who received a notification of arrhythmia were later found to have atrial fibrillation (AF), and the positive predictive value in participants notified of an irregular pulse was 0.84. 1 That study relied on optical sensors in Apple Watch Series 1 through 3 to detect an irregular pulse and a proprietary algorithm to indicate arrhythmia. 1 In contrast, the Apple Watch Series 4 (AW4) uses electrodes to generate a single-lead ECG and provides 2 mechanisms for rhythm assessment: rhythm notification (AF, sinus rhythm [SR], or inconclusive) and single-lead ECG downloaded as a PDF for interpretation.2 This study assessed the accuracy of the AW4 for AF detection in patients who had undergone cardiac surgery, a group who frequently cycle between AF and SR. METHODSWe recruited 50 patients who had undergone cardiac surgery who were on telemetry.Exclusion criteria included cardiac pacemaker, surgical use of a radial artery, chronic heart rhythm abnormalities, and wrist tattoos.The protocol was approved by the institutional review board and registered at ClinicalTrials.gov(NCT03798613).Subjects provided written informed consent.Rhythm assessments were performed 3 times per day (morning, early/late afternoon, and early/late evening) over 2 days, resulting in 6 assessments per patient.At each assessment, the subject was randomly assigned by computer program 1 of 5 AW4 watches that was placed on a wrist (randomly assigned); 5 watches were purchased to mitigate the impact of possible hardware malfunctions and the potential impact of a single device.When instructed, subjects depressed the crown for 30 seconds, and the watch reading (SR, AF, inconclusive) was recorded.Simultaneously, a PDF of the waveform on the Apple Health App and a rhythm strip from telemetry were saved and adjudicated by a cardiologist blinded to the patient's telemetry rhythm.ECGs were generated and viewed according to instructions provided by Apple. 2 The Cohen κ statistic was used to measure agreement between the AW4 notification and telemetry and between the AW4 ECG interpretation and telemetry.The sensitivity and specificity in detecting AF were calculated from generalized estimating equations with an exchangeable correlation matrix accounting for repeated measurements.3 Statistical analyses were performed with R and SAS statistical software (SAS version 9.4; SAS Institute Inc, Cary, NC). RESULTSOf 300 possible rhythm assessments, 292 were obtained; 2 patients were discharged before study completion, and 1 patient withdrew consent.On telemetry, AF was observed in 90 instances, and SR was seen in 202; 25 of 50 patients had ≥1 episodes of AF.The AW4 notification correctly identified AF in 34 of 90 instances, yielding a sensitivity of 41% (Table ).Of 25 patients with at least 1 episode of AF, AF was

Source: Circulation

DOI: 10.1161/circulationaha.119.044126