Friday, July 28, 2017

How Valid is Polar V800 HR to Measure HRV Metrics?

According to a 2015 paper by Giles, the Polar V800 "improves" over previous generation Polar devices with narrower intra-class correlation coefficients, limits of agreement and effect sizes when compared to R-R readings obtained from a 3 lead Biopac ECG.

One of the key aspects of the paper is that it highlights the importance of understanding the effect of different softwares and their error correction routine on the R-R time series.  The paper references some earlier studies which flagged the concern that each system's individual HRV processing capability can potentially alter the HRV metrics in their own way making cross-comparability a big issue.

The paper discusses chiefly 7 errors that can be associated with the Polar V800 HRM signal. Potential factors leading to those errors are also stated. I'd like to borrow the learning and place them here :

Type of ErrorDescription of ErrorPotential Cause
T1Single interval of discrepancyNot stated
T2Long interval and short intervalNot stated
T3Short interval and long intervalNot stated
T4Too few intervals detectedDecrease or loss in contact between electrode-skin and resulting decrease in amplitude of the R wave
T5Too many intervals detectedNot stated
T6-aInterval(s) missed entirely, undetectableSoftware error due to time asynchronicity in the HRM and/or Loss of, or decrease in contact between electrode-skin.
T6-bInterval(s) missed entirely, detectableSame as T6-a

Although the causes of T1-T5 errors are not addressed in the paper, the authors state that all of them may be recognized and corrected without the use of a simultaneous ECG recording. However, they highlight the fact that the most appropriate technique for correction of R-R time series is still pending proper research and agreement.

The combined error rate of Polar V800 in standing and supine positions was 0.086% which is a marked improvement over previous research findings from the Polar S810 (0.40%) and Polar S810 (6.93%). The paper concludes that the device is a "valid" tool for the detection of RR intervals "at rest" and improves on previous HRM models with regard to comparability against ECG. 

The meat of the paper is it's tabulated summary of means, bias (LoA), ICC (95% CI) and effect size, which I'd like to reproduce here. These numbers are the justification for stating the validity of the V800. An interesting area to note is the level of discrepancy in HF power between V800 and ECG in the supine position. It is an order of magnitude different, whereas much more agreeable in the standing position. Is this a typo or the actual finding?

No comments: