Overnight HRV: Comparison of PPG to ECG

Problem:  Short-duration morning HRV measurements may be susceptible to exactly when you wake, your HR, breathing rate, thinking about work etc in that moment. Different apps report widely different values of HRV.  Perhaps more importantly, morning snapshot measurements do not record HRV over a physiologically meaningful interval.  The best case would be to record HRV 24-7, but this is not feasible with current wearable technology.  However, wearables can and do report night time HRV.  At least this is measuring vagal balance over a physiologically meaningful interval, namely, the sleep recovery timeframe.

Methods:  I recorded overnight HRV using Polar H7 chest strap connected to the iOS app HRV Logger, for 12 nights over more than a month.  HRV Logger stopped recording on 2 nights, perhaps due to loss of connection with the H7 or other errors, so I have 10 nights of complete data.  The Polar H7 chest strap directly measures electrical activity (ECG) and is considered the gold standard for at-home HRV measurements.  

I also wore a Fitbit Inspire 2 each night and downloaded the data file from Fitbit.com at the end of the month.  Fitbit uses a green light PPB (photoplethysmography) sensor to measure blood flow changes at the wrist and infer heart rate and HRV.  

HRV Logger was set to 5 minute recording windows and RR-interval correction was set at 25%.  Fitbit also reports HRV in 5 minute recording windows.  However, the 5-minute interval windows are only approximately aligned as they are based on offset starting times.  HRV is reported as rMMSD.  

Results:  At this point, I am reporting preliminary data from 3 representative nights.  

Night of 9/19

Summary

	Fitbit	Polar
mean	38.5	41.48
std dev	6.75	8.44

Night of 9/21

 

Summary

	Fitbit	Polar
mean	39.82	44.32
std dev	6.94	9.23

Night of 9/28

 

Summary

	FitbitInspire2	PolarH7
mean	35.06	39.57
Std dev	5.31	7.80

Conclusion

An examination of the time course data shows that Fitbit was able to pick up the broad changes in HRV during the night.  It appears Fitbit misses spikes in HRV that the Polar strap is able to detect.  Although it is also possible that the Polar strap is biased by movement artifacts, it seems most likely that either Fitbit is not sensitive enough to detect short-duration changes in HRV, or a Fitbit error-correcting algorithm is too aggressive at removing short changes in HRV.

In addition to being less sensitive to variations in HRV (as measured by standard deviations), Fitbit also consistently records a lower average nightly HRV than the Polar strap.

The overall R squared value is only 0.38, however this statistic probably underestimates the correlation of time series data due to the fact that the 5-minute windows are not perfectly synchronized.  Even so, R squared of 0.38 is a higher correlation that I was able to achieve using the Polar H7 with 2 other apps.