Detection of Postoperative Atrial Fibrillation with a Smart Watch: Preliminary Results of a Clinical Investigation

Objectives: The long-term incidence of atrial fibrillation (AF) after cardiac surgery is still not sufficiently studied, mainly due to the lack of cheap, noninvasive devices for the detection of AF. Recently, wrist worn fitness trackers have gained attention in the detection of AF. Here, we compared the reliability of heart rates (HR) gathered from the Fitbit Charge 2 fitness tracker under sinus rhythm (SR) and AF.

Methods: HRs were collected using the Fitbit Charge 2 device from patients undergoing elective cardiac surgery of any kind. The HRs were compared to those of a regular six-channel electrocardiogram (ECG) that was registered by the patient monitoring system. Postoperative HRs were recorded after referring the patient to the intensive care unit for a maximum of 24 hours, excluding HRs under pacing. The Fitbit device was measured four times each minute, and data were either collected as the mean HR (Fitbit HR mean) or the maximum or minimum HR (Fitbit HR max or min). Data were analyzed using SPSS.

Results: The number of patients included into this study was 30. Overall, a total of 29,398 HRs were recorded, which were further divided into HRs under SR (N = 22,092), normofrequent AF (N = 2,094), and tachycardic AF (N = 5,212). The median absolute percent difference between the Fitbit HR mean and the ECG HR was 1.3% for SR, 5.1% for normofrequent AF, and 11.3% for tachycardic AF (p = 0,000). Bland Altmann analyses showed that only 49% of Fitbit HR mean under tachycardic AF were within 0 to 5% of difference to the ECG HR. In correlation analyses, a strong correlation of HR between Fitbit HR mean and ECG was found during SR (r = 0.940), but not during normofrequent AF (r = 0.403) or tachycardic AF (r = 0.001). However, the variability in Fitbit HR, expressed as the median difference between Fitbit HR max and min, was significantly higher under tachycardic AF compared to SR (5.0 vs. 2.0) and under normofrequent AF (3.0). In power analyses, a mean difference of Fitbit HR max–min of 0.30 would be enough to detect normofrequent AF with a power of 80% (alpha 5%), while a mean difference of 0.31 would be enough to detect tachycardic AF.

Conclusion: Heart rates detected by Fitbit Charge 2 are reliable under SR, but not under normofrequent and tachycardic atrial fibrillation. Interestingly, we found that the fluctuations in detected Fitbit HRs vary significantly between SR and AF. These findings could be useful to develop algorithms for the detection of AF by wearable fitness trackers.