Consumer wearable technology has recently become an integral part of
behavioral research. In addition to common motion sensors, such as
accelerometers and gyroscopes, used to compute steps and activity intensity
values, newer devices have also begun to include heart rate sensors. These
typically take the form of optical sensors for photoplethysmogram (PPG)
measures that compute heart rate. While many wearable devices are
wrist-based sensors, there are a growing number of companies that are
integrating multiple sensor suites into smart garments.
This study, conducted as part of a larger ongoing study aiming to
understand predictors of productivity in a high-stress work environment,
compares two commercial wearable systems that measure heart rate obtained
using different methods. The Fitbit Charge 2 is a wrist-worn device that
incorporates PPG sensing for 24/7 measurement of heart rate. The OMsignal
smart garment (bra or compression shirt) includes a variety of sensors,
including printed electrodes for ECG measurement.
A total of 192 adult participants (71.9% female) wore the Fitbit Charge 2
and an OMsignal garment during working hours over a 10-week period in early
2018. Fitbit heart rate data was captured at one minute epochs (mean heart
rate over one minute) and OMsignal data was captured at five minute epochs
(mean heart rate over five minutes). Fitbit data was resampled to reflect
the available five minute epochs from OMsignal. Additionally, OMsignal
values associated with poor RR coverage were removed, per manufacturer
analysis guidelines. Comparison between the two devices was measured by
mean bias, mean absolute percent error (MAPE), concordance correlation
coefficients (CCC), and Bland-Altman plots with 95% limits of agreements
(LOA). The OMsignal was treated as the reference device in this analysis
due to its use of ECG measurement.
Across all participants, 3,738 days of observation were included in this
analysis. The mean number of paired Fitbit-OMsignal epochs per participant
was 1,277.96 (SD = 964.21). The mean bias was 1.69bpm, and MAPE was 4.98%.
The overall CCC was 0.892 (95% CI = 0.891 - 0.893). Bland-Altman analysis
indicated 95% of values fall between -10.06bpm and 13.43bpm.
We found that in a long-term free-living study the Fitbit Charge 2
performed similarly to the OMsignal garment for measuring heart rate. When
compared to prior validation studies comparing Fitbit heart rate to gold
standards in free-living and laboratory settings, these results indicate
slightly higher correlation, a narrower LOA, and smaller mean bias. This
study is unique due to the large amount of free-living data spanning
real-world activities of daily living and the inclusion of a new device,
the OMsignal smart garment. Further research should explore additional
features extracted from ECG data beyond heart rate and how they compare
against laboratory studies.