Trackers - recommendations

[mccoy]

Very good, Ron, now we have a clear picture of the accuracy of the Dreem headband, which in some cases is not very high because its reference, the PSG, is inaccurate. So, taking it as a reference may generate misclassification in the assigned reliability of wrist devices.

In the Arnal et al. article you posted, the main points I saw are as follows:

• The Polisomnigraph (PSG) is the golden standard but is subject to inaccuracies, especially in the estimate of wake stage and N1 (deep) sleep.
• The Wake band exhibits good accuracy if compared to the PSG.

So, it is legitimate to use the Dreem band as a reference, recalling that this reference has a significant probability to misclassify deep sleep and a nontrivial probability to misclassify wake stages.

With the above premise, the comparison Dreem headband versus Versa3 watch could be extended to all fitbit devices since they all use the Google algorithm for sleep analysis.

In the Kuosmanen et al. study, I can have some idea of how the Fitbit measurements stand compared to the Dreem. For example, REM sleep is underestimated by Fitbit (as I suspected) whereas light sleep is overestimated. Deep sleep is about the same, but it is not reliable even in the PSG, so it's a value that constitutes a very rough estimate. The wake time in the headband has a strange distribution, a very low median, also the PSG is not very reliable in that, so the comparison here may not be valid. Total sleep duration is similar in the Fitbit estimate and in the headband estimate.

All in all, the Google sleep analyzer seems to be a decent estimator of sleep, with all its inherent limitations, excluding deep sleep which is not reliably analyzed even by polysomnography.

 

 

Edited July 23, 2023 by mccoy

[mccoy]

One more pro of the Inspire3: this morning I measured the HR while training in zone 2 on the treadmill. The values measured by the Inspire3 were very similar to those measured by the polar H10. My hands were free and not on the lateral rests.

If the watch HR is reliable training with the treadmill free-handed, much more so it should be reliable when training on a stationary bike.

This is good news since it would be possible to train in zone 2 only using the watch (if kept with the display always ON).

 

[mccoy]

This is a very useful plot displaying the waking episodes in red. Tonight I had a really agitated sleep and woke up many, many times, the cause being unknown to myself. The Fitbit accurately recorded this, suggesting 15 wake episodes before the two main ones which constitute the definitive wake-up interrupted by a brief sleep interval. 

 

[manuel5]

What do you think about the Garmin VivoSmart 5? seems to be a reliable alternative...

[mccoy]

On 7/23/2023 at 10:08 PM, manuel5 said:

What do you think about the Garmin VivoSmart 5? seems to be a reliable alternative...

Manuel5, the Garmin products have a good overall reputation, for the specific product you should search the reviews available on the web and then use some discrimination

[mccoy]

HRV on the inspire3 and Fitbit products. I just discovered that such values are often not comparable to other values like those provided by the EliteHRV app. The latter calculates a score based on ln(RMSSD), the former I don't know.

The ECG-based methods like the Polar strap measure the R-R intervals, the light sensors like those on wristwatches measure the P-P peaks, apparently.

So, probably the right way to evaluate the Fitbit HRV quantity is in relative, rather than absolute, terms, that is, determine one own's baseline and watch the trends. Last night I was sick, and this was obvious in a dip in the nightly HRV Fitbit score.

 

Edited July 28, 2023 by mccoy

[mccoy]

I did further measurements and it turns out that the EliteHRV app also provides values in RMSSD units. 

So, ms in RMSSD units are pretty comparable to the values reported in the Fitbit app which , according to the info, are probably in the same units.

Also, measurements on-demand (apps like EliteHRV) are often not directly comparable to whole-night measurements like those of Fitbit, for various reasons.

An in depth analysis of HRV values can be found in the site by Marco Altini, with a focus on training but very interesting reports on techniques used and so on (for example, it is possible in Apple watches to elicit an on-demand HRV measurements by starting a breathing measurements).

 

https://www.marcoaltini.com/

Edited August 3, 2023 by mccoy

[corybroo]

The title says it all.

Fitbits perform well in capturing circadian rhythms but not sleep time

"A widely available commercial product that many people own can be used to measure rest-activity cycles with precision accuracy," said UT psychology professor and department chair David Schnyer, one of the study's authors. "This was true in both younger and older adult groups. The latter is important because research has shown that changes in circadian rhythms associated with aging can be sensitive to both cognitive decline as well as dementia vulnerability."

In a study published in Smart Health, the authors compared two popular Fitbit devices against the Actiwatch 2.0—a known, reliable method for measuring activity patterns outside of a laboratory

Participants wore an Actiwatch and one of the Fitbit options on opposite wrists for 10-14 days while going about their normal activities. They also kept daily sleep diaries. Fitbit accuracy was comparable to the Actiwatch for rest-activity cycles but tended to underestimate both total sleep time and time it took wearers to fall asleep.

 

My weekly average definitely looks like an underestimate:

[mccoy]

14 hours ago, corybroo said:

The title says it all.

Fitbits perform well in capturing circadian rhythms but not sleep time

I wonder how we can extract circadian signals from the Fitbit, also in the article they compared a device to another device, which may be more accurate but certainly not absolutely accurate.

14 hours ago, corybroo said:

My weekly average definitely looks like an underestimate:

I cannot find on my screen the parameter 'restful sleep'. Maybe it means effective sleep, which si time spent in bed while sleeping as opposed to time spent in bed awake. 

2h 13m average is definitely very little, what's your average sleep score?

I find that the Fitbit in my case tends to overestimate slightly time spent awake, but not of huge amounts, like it may indicate half an hour more but not hours.

Also, I dream a lot of structured dreams=REM state and sometimes I am awakened abruptly by my son, so I can check later what the watch reports at that time; by such observation, I can say that most probably the Fitbit/google algorithm in my case underestimates REM sleep by a wide margin. It probably also underestimates deep sleep, but I cannot judge it.

From the article linked by Ron, even the polysomnography, the golden standard, errs by a nontrivial amount when determining wake/sleep states and likewise errs when determining deep sleep. So we are dealing with intrinsically uncertain quantities.

So, I carry out my own critic judgment of sleep data from fitbit everyday.

My opinion is that it is a useful tool anyway, but a big drawback in the Fitbit watch is the lack of possibility to input naps. I'm a regular napper and that would definitely increase my daily sleep score.

 

[mccoy]

Tonight's example of Fitbit/Google sleep capabilities. Sleep duration graph. It was pretty accurate.

Sleeping time is right. Then I had my usual fragmented sleep until about 2 or 3 AM. Then my son started to wake up and walk around home, trying to pull me and my wife out of the bed. I got up a couple of times. Also, when diving back to slumber, that was actually a borderline state, half sleep half awake, which was captured by the thickness of the red=awake lines. The algorithm construed borderline as awake and there is not a big error in that. 7hr20min sleeping, 1hr 1 min awake. 

Same night, sleep stages. By a triangulation with the Aktiia blood pressure sensor, I probably had some deep sleep before midnight, which was not detected by the Fitbit. I also remember I had various periods of REM (sleep with vivid, structured dreams) in between awakenings, but these were not detected, maybe too short? But the impression is that in me Fitbit underestimates REM stages.

The following is probably an accurate plot of HR. Last night the HR was on average higher than usual, with about 60% of the time > resting HR, which is 59 bpm. Also, often other nights I hit a trough of 50 bpms, whereas last night my heart rested less. The reason is unknown.

Edited August 9, 2023 by mccoy

[corybroo]

8 hours ago, mccoy said:

2h 13m average is definitely very little, what's your average sleep score?

I'm happy when it gets as high as 70 (out of 100).  To be fair, due to a spinal injury some time ago, I have phantom pains in my foot and the twitching is probably interpreted as wakefulness.

Here's last night's breakdown.

I wish you pleasant dreams - it looks like you're getting a lot more dream time than I am.

[mccoy]

19 hours ago, corybroo said:

I wish you pleasant dreams - it looks like you're getting a lot more dream time than I am.

Yes, but I would bet some money that Fitbit underestimates REM sleep in my case, probably in yours as well. I remember distinctly dreams especially after a waking episode. Those REM, structured dreams are very frequent late at night and early morning and often not detected by Fitbit. It is also true that our subjective perception dilates time during dreams and they seem longer, but, nonexisting?

Your sleep would appear terrible, do you really perceive it like that?

[mccoy]

I have bought my wife a Samsung Galaxy watch5. It seemed to be the logical choice since she already has a galaxy phone, and the quality-to-cost ratio is good. No subscriptions, which is a good thing.

The tracking suite includes a sleep analyzer, HR, spO2, ECG on demand and HRV on demand (in this case the result is a stress index, the underlying units are not disclosed strangely). It also carries blood pressure detection capabilities, but I am skeptic about that, blood pressure must be calibrated by traditional methods, I'm curious about the results.

Last, it has a body impedenziometer which measures lean body mass, fat % and so on. The figures look reasonable at first, maybe a little high, I could not check it on my own not to change body parameters.

The advantage of a large dial is very easy reading of the time, plus the capability to answer to phone calls, read notifications and answer to them and so on. I may switch to this kind of watch, still undecided about the model.

The one suggested by Ron, Garmin venu 2 plus, has good reviews and is pretty complete, although a little higher in price compared to seemingly equivalent samsung galaxy watches.

[Ron Put]

Samsung is a very decent choice if your wife has a Samsung phone. Tons of metrics, most pretty accurate, including sleep duration and quality.

For me, the main issue with the Samsung, Apple, or Pixel is the battery life. If they can double it, it will make all of them usable for my purposes.

[mccoy]

23 hours ago, Ron Put said:

For me, the main issue with the Samsung, Apple, or Pixel is the battery life. If they can double it, it will make all of them usable for my purposes.

Of course, if you travel a lot, a short battery life means a watch out of charge.

Another aspect I discovered is the aesthetic one. For example, after having seen my wife's Samsung  watch in blue, I was more attracted to it, an interesting blue suited to my choice to dress only in blue hues.

But I still have a Huawei phone which is working well, so I'm going to wait and see (and watch reviews).

[IgorF]

I already mentioned somewhere here that there is a malpractice in the modern devices to report "algorithmical" values instead of more "instrumental" (in the sense of - something that represent a fact from the physical world) ones, here is a good example on my fitbit where the heart rate is represented wrongly due to improper application of rolling average (or maybe with spline interpolation over it or whatever) - a case of inflammatory response on a common cold. The first sympthoms started 31.12 4pm and 01.01 the body instrumented the response detectable by such devices - temperature increased 2C and the heart rate added 20bpm. Instead of report it as a "shelf", e.g. -53-68-71-65-65-53- (I cross-checked the values with other devices, including a spo2 ring with very good sampling rate for constant overnight measuring) it shifted the data forward and mangled it to some usless stuff completely. Interesting that skin temperature anomaly, increased breating rate and hrv drop is more or less adequately reported.

 

Br,

Igor

[mccoy]

Good example. There is a moving average to real-time HR as well, which in this case I find adequate though.

[Ron Put]

Igor, if I am understanding this correctly, you are looking at the resting HR value, which is obviously a different calculation than just presenting a sampling of actual HR. I haven't used Fitbit in a long while and back when I did I found it less accurate than Garmin for HR under load, and the RHR values appeared to be lower on Fitbit (and the estimated VO2Max was significantly higher on the Fitbit).

I've actually moved on to the Garmin Epix, because I travel and often hike in new locations, so the topo maps are invaluable to me (and also for general city navigation, so that I don't drain my phone's battery).

[IgorF]

Yes, my post was about rhr and fitbit's will to report something that its algo fails to do properly. Actually it tweaks the values with a delay and without inflammation as a huge change it was like a 1-2 days delay to show something that other devices show immediately (with some 5-10% difference in accuracy between the devices). I personally don't trust devices with such approach, just because they do report something their creators think their customers are expecting to see, this is bad practice similar to snafu principle. But that off course my imho, I understand that there will be no market for devices I would like or they will cost two orders of magnitude more(

 

[Saul]

I use a device that was chosen by the NHI for their "AllOfUs" study.  The device is the Fitbit Versa 4.  It gives results for HR (as you experience it); the values that if gives seem plaudible to me.  (I'll see my nephrologist sometime in the next few months -- he checks my bloodwork, etc. [Note: I  have excellent kidney function; I was fortunate to get regular appointments with a nephrologist for bloodwork.  I'll ask him about the accuracy of the fitbit's HR readings -- they go to my cellphone, and can be easily viewed].  Unfortunately, the fitbit doesn't track BP -- not a surprise -- that should be hard to do on a wrist device (but not impossible -- of course, wrist BP will vary greatly, according to the position of your wrist).

  --  Saul

 

[IgorF]

A set of my rhr measures last night:

miband 6 - 53

fitbit luxe - 53

wellue o2ring - 50

beautyrest sleep tracker - 50

aktiia - 50/47/46 (in 10 measures it did during the night)

In the past when I had no wellue ring I also used to measure randomly the heart rate manually and with a strip (sigma), the values were the same as for beautyrest and aktiia. Kardia tool's measure during the day were the same as with strip and manual observations.

With fitbit it seems constantly reports 5-6% higher hr value for the overnight measuring, miband is rather 4-5% overreporting. But when the things starts to move (e.g. late supper 1000+kkals or going CR for a few weeks) all the tools reflect the change immediately the next morning except fitbit that tries to "guess and elongate the trend".

Maybe it is just luxe model or my particular is prone to such behavior but intuitively I think it is not in the tool but in the app.

Br,

Igor

 

Edited January 14 by IgorF