The behavior where the same individual wearing multiple devices (either on the same or different wrists, or a chest strap) sometimes sees different heart rate results is a very well-known phenomenon.
There can be many reasons for this and the simple truth is that the system we are measuring is subject to an incredible number of variables which will always result in some differences manifesting when comparing multiple devices on the same person, especially when comparing instantaneous observations.
We have conducted a large number of tests using the wearable devices alongside a gold-standard ECG reference device during both exercise tests (40-minute protocol containing walking, running, cycling, and rowing), and more importantly, also long duration (“24-hour” protocol free-living) tests, and the results we see are spot-on with what we would expect. We have done similar tests with many devices available commercially; Garmin, TomTom, Fitbit, Mio, Samsung etc, and what we see on the LifeQ-Enabled and LifeQ-Connected devices matches or exceeds the level of accuracy we would consider as representative of state-of-the-art in this field.
To understand where these differences between devices would manifest, consider that the heart rate algorithm has two information streams as input; PPG signal and accelerometer. The PPG signal is the signal measured by the light sensor on the device; light is shone into the skin where it interacts with the person’s blood and is then received again by a photo-sensor in the device. The quality and integrity of this signal is the main factor in the ultimate accuracy of any heart rate algorithm, and the most common issues which would affect this signal are:
- device fit being loose vs tight
- skin tone variations, tattoos, hair
- sub-surface physiological differences such as muscles/tendons/blood vessels which can vary considerably depending on the placement of the device
- pressure affecting the blood flow to the device, i.e. laying on your arm
- external disturbances causing movement of the device (e.g. intense exercise / tapping the device)
- composition of the tissue being measured: muscle, fat, etc
The accelerometer signal, on the other hand, is used by the heart rate algorithm to identify and filter out motion-induced disturbances on the PPG signal; any disturbance/change of this signal will also affect the heart rate output.
These are typically:
- if one arm is stationary vs the other, the signals get distorted differently and, in some cases, can cause extreme differences in heart rate results
- tapping, bouncing, or typing (keyboard or phone) causes different types of movement distortions which also cause similar differences in heart rate results
Considering the above, in general, we find that placement higher up the wrist generally improves the heart rate for the following main reasons;
- the device is worn more snugly (the strap can only extend so much)
- the perfusion effects from cold extremities improve which improves signal measurements
- there are more blood vessels, more tissue, and more softness in general (including muscle)
We also expect to see slightly better results on the non-dominant wrist since it is subject to fewer motion disturbances.
Another important factor to account for when doing an instantaneous comparison between devices (as opposed to plotting the logged data afterward), is that there is always a delay between the user’s actual heart rate, and when the value finally appears on screen or on an App. This delay will vary between devices and is as a result of the nature of the algorithms; buffers; transmission delays etc. It may, therefore, appear that there is a difference in heart rate, but in fact the one output might just be delayed by a few seconds. When we do an analysis of the accuracy, we always align the data to account for this.
When interpreting the heart rate accuracy between two devices, the following general rules of thumb apply:
- Under good signal conditions, the accuracy is reported to be within ±5bpm of the reference and is considered good. This means that one device can read 90bpm and the other 100bpm, where the reference is 95bpm.
- Under bad signal conditions, the accuracy is reported to be within ±15bpm or more is considered poor but still expected to occur a few times during a normal day. In this case, one device can read 80bpm and the other 110bpm, where the reference is 95bpm.
- When the signal quality becomes extremely poor the heart rate algorithm has no signal to track and the errors can become very large. In these cases, however, LifeQ makes use of the heart rate confidence metric to discard those values from downstream analysis.
Our analysis of heart rate accuracy across long durations of time (24-hour + tests) shows that the mean absolute error and the mean absolute deviation of LifeQ-Connected heart rate compared to a gold standard reference meets the accuracy required for all our downstream analytics.
Finally, it is important to note that the metrics calculated in the Cloud are derivatives or summaries of the instantaneous measurements taken on the wearable device; therefore, when we look at longer durations of data, the typically instantaneous anomalies one would observe become negligible. Knowing that the outputs from the device are near identical across larger timescales gives us confidence that the downstream analytics is unaffected; especially since all the cloud algorithms have been trained using data of this nature.
Was this article helpful?
Thank you for your feedback
Sorry! We couldn't be helpful
Thank you for your feedback
We appreciate your effort and will try to fix the article