Background: With smartwatch-enabled electrocardiograms (ECGs) becoming more common, there is an opportunity to develop novel preventative strategies at scale. However, accuracy of wearable ECGs has not been fully investigated.
Aim: To validate 3 characteristics of wearable ECGs: (1) morphology, (2) cardiac intervals (heart rate, HR, and QT interval), and (3) beat-to-beat ultra-short HR and QT variability (HRV, QTV). Methods: Wearable (Withings Scanwatch, WS, and Fitbit Sense, FS) and medical-grade (Bittium Faros) ECGs were simultaneously recorded in 20 participants (median age=23.5, interquartile range=(20, 36.5) years, 7 females). All recordings were single-lead (Lead I) and lasted 30 seconds. Recordings were synchronised using timestamps and a method based on signals' cross-correlation. A signal-averaged ECG waveform was computed by aligning and averaging all heartbeats. R-R and RT (a robust surrogate of QT) intervals were automatically measured using algorithms developed in-house and not manually corrected. Standard measures of ultra-short HRV and QTV included the root mean square of successive differences (RMSSD), the short-term variability of RT (STVRT), and the RT variability index (RTVI). Comparisons were based on bias and limits of agreement by Bland Altman analysis, Pearson's correlation coefficients (cc), and absolute percentage errors (MAPE). Morphological similarity was measured as the cc between signal-averaged ECGs (cc-SA).
Results: All participants were in sinus rhythm except one who was in atrial fibrillation. Compared to medical-grade ECGs, wearable ECGs showed high morphological similarity (cc-SA=0.91 (0.86, 0.95) and 0.89 (0.84, 0.93) for WS and FS, respectively), and almost identical mean HR and HRV. Mean RT was slightly underestimated, with high correlation (cc~0.90), MAPE<5% but relatively large limits of agreement (-40, +17ms). Agreement for repolarisation variability was poorer (cc~0.35, MAPE>40%). Conclusions: Compared to medical-grade ECGs, ECGs simultaneously recorded using consumer-grade smartwatches show very similar morphology and almost identical HR and HRV, however, agreement is poorer for beat-to-beat repolarisation variability.