What is the gold standard for measuring heart rate? - ECG explained

TL;DR: The clinical reference standard, the electrocardiogram (ECG), measures the heart’s electrical activity directly, which is why it is more precise than any wearable.

Every heartbeat begins as an electrical event. The sinoatrial node, a small cluster of cells in the right atrium, fires an impulse that travels across the cardiac muscle and triggers contraction. An ECG records this activity by placing electrodes on the skin, which detect the tiny voltage changes produced as the impulse moves through the heart. The signals are small, on the order of millivolts, but consistent enough to resolve into a characteristic waveform: the P wave as the atria depolarize, the QRS complex as the ventricles depolarize and contract, the T wave during recovery. Heart rate and HRV are calculated from the R-R intervals, the time between consecutive R peaks. Because ECG captures the electrical cause of the heartbeat rather than its downstream mechanical effect, it is more precise than PPG and remains the clinical reference standard.

How do wearables measure cardiac metrics? - PPG explained

TL;DR: Your wearable does not measure electricity; it measures light, using photoplethysmography (PPG) to read the pulse of blood moving under your skin.

I have written about this before, but the questions keep coming, so here is how PPG works, what it records, and how to read it. Photoplethysmography (PPG) is an optical technique that detects changes in blood volume at the skin surface using light. The sensor shines LEDs into the skin and measures how much is reflected back by the underlying tissue. Consumer wearables typically use two LED colors: green light, which penetrates only as far as the surface capillaries and is absorbed well by the blood moving through them, and red or infrared light, which reaches deeper into tissue. Green is the workhorse for heart rate and HRV during everyday wear because it handles motion better. The red and infrared pair makes blood oxygen measurements possible, since oxygenated and deoxygenated hemoglobin absorb those wavelengths differently. With each heartbeat, blood volume in the capillaries changes slightly. More blood means more light absorbed and less reflected back to the photodetector. The sensor records these tiny fluctuations as a continuous waveform that rises and falls with each arriving pulse. That waveform is the raw output from which you can extract heart rate (time between peaks), HRV (how much that timing varies beat to beat), respiration rate (a slower oscillation riding on top of the main signal), and blood oxygen saturation.

How is the light translated to a number you see on your device?

TL;DR: The watch turns a noisy stream of reflected light into a single heart rate by cleaning the signal, finding each pulse, and averaging the time between beats.

When you put the watch on, the LEDs on the underside activate and begin sampling your skin. The photodetector captures the reflected light continuously, building a raw waveform that fluctuates with each pulse. But that signal is noisy: motion, ambient light, skin tone, and how firmly the device presses against your wrist all affect what the sensor picks up. Onboard algorithms filter out that noise and identify the peaks in the cleaned waveform, each one corresponding to a single heartbeat. The time between consecutive peaks is the inter-beat interval, measured in milliseconds. Heart rate is derived from those intervals: 60,000 divided by the average inter-beat interval gives you beats per minute. What appears on the watch face is that number, smoothed over a rolling window of several seconds to reduce beat-to-beat variation.

How do you check whether a wearable got the number right?

TL;DR: Validation papers describe the gap between device and ECG in a handful of standard ways.

Every study below compares a device reading against a simultaneous ECG reading and describes the gap using the same small vocabulary.

Mean bias is the average difference between device and reference. It tells you whether a device runs high or low and hides a great deal else: a device that reads 5 bpm too high on half the nights and 5 bpm too low on the rest has a mean bias of zero.

Mean absolute error (MAE) drops the direction and reports the average size of the miss, in beats per minute or milliseconds. Mean absolute percentage error (MAPE) expresses that miss as a share of the reference value, which makes errors comparable across metrics measured in different units.

Limits of agreement mark the range you would expect about 95% of individual differences to fall inside. One night in twenty falls outside them, with nothing capping how far. Look at this one if you care about tonight’s reading rather than your average across a month.

Concordance (CCC) and intraclass correlation (ICC) are close cousins, and both improve on plain Pearson correlation (r) by penalizing a device that is consistently offset. Correlation on its own is the least useful number here: it can approach 1 while every reading is 10 bpm too low, or while a device squashes your range, because it only asks whether the two measurements move together. Concordance asks whether they land on the same number. Both reward a varied sample, so the same device scores better in a group whose resting heart rates span 45 to 75 than in one where everyone is near 55, which is why these numbers travel badly between studies.

Which validation studies to look at?

TL;DR: Three studies published since 2022 put twelve consumer devices against an ECG reference, and each is built differently enough that the disagreements between them are informative.

Miller et al. (2022) brought 53 healthy young adults, 26 women and 27 men, averaging 25.4 years, into a sleep lab for a single night [1]. Six devices ran simultaneously against what the paper calls a II-lead ECG: Apple Watch S6, Garmin Forerunner 245 Music, Polar Vantage V, Oura Ring Generation 2, WHOOP 3.0 and Somfit. Funded by the Australian Institute of Sport, with the authors disclosing that their research group at Central Queensland University receives research support from WHOOP Inc., which had no role in the design, conduct or reporting. Two methods details are important in this study. WHOOP was the only device that supplied raw R-R intervals rather than an app-exported summary, and the devices were not all measuring the same thing: Apple, Oura, WHOOP and Somfit reported across the whole sleep period, Polar sampled a four-hour block, and Garmin’s HRV came from a three-minute test taken 30 to 60 minutes before lights out, while the participant was still awake.

Sarhaddi et al. (2022) took the opposite approach, following 28 adults, 14 male and 14 female, through 24 hours of ordinary life wearing a Samsung Gear Sport alongside a Shimmer3 chest ECG [2]. Because the monitoring ran around the clock, the authors could report sleep and wake separately.

Dial et al. (2025) is the most recent and, for a reader who wants to know about their own ring, the most relevant [3]. Thirteen adults, six of them women, averaging 33.2 years, slept at home wearing a Polar H10 single-lead ECG alongside five devices: Garmin Fenix 6, Oura Generation 3, Oura Generation 4, Polar Grit X Pro and WHOOP 4.0. Every participant contributed at least ten nights, 536 in total. Thirteen people cannot represent a population, and the nights are treated as independent even though each person contributed roughly forty, so the precision on offer is closer to thirteen people’s worth than to 536. What the design buys instead is a good picture of how each device behaves across many nights in one person.

How accurate is your overnight resting heart rate?

TL;DR: On current hardware, overnight resting heart rate lands within about one to two beats per minute of a clinical ECG, comfortably inside the margin at which a change would mean anything.

At home, across 536 nights, Oura Generation 3 missed the ECG value by an average of 0.98 bpm and Generation 4 by 1.08 bpm [3]. Polar Grit X Pro and WHOOP 4.0 came in at 1.72 and 1.78 bpm. The authors put those numbers against the clinical picture themselves: a meaningful deviation in resting heart rate is somewhere between 5 and 7 bpm from your baseline, so every device tested lands well inside the margin that would matter.

Table 1. Overnight resting heart rate agreement with ECG across 536 home nights, from Dial et al. (2025). Garmin Fenix 6 was excluded from this analysis.

Garmin is missing for a reason that will be familiar to anyone who read my post on how providers define RHR. The Fenix 6 reports the lowest 30-minute average in a 24-hour period, and neither the watch nor the app says when that half hour occurred, so there was no way to line the ECG up against it [3]. A metric that cannot be located in time cannot be validated. Polar carries an asterisk of its own. It reports RHR and HRV from only the first four hours of sleep, so the researchers recomputed the ECG reference over those same four hours for that device alone, while every other device was compared against the whole night [3].

The lab study points the same way with different hardware. Apple Watch S6 and Polar Vantage V both averaged 1.5 bpm of absolute error against ECG, Oura Gen 2 reached 1.8, and WHOOP 3.0 came in at 0.7 bpm with an ICC of 0.99 [1]. Garmin Forerunner 245 was the outlier at 5.4 bpm, with 95% limits of agreement of ±25.0 bpm.

How accurate is your overnight HRV?

TL;DR: HRV error is several times larger than heart rate error and five of six devices tested in the lab compressed the range, reading your high nights low and your low nights high.

Measuring HRV in wearables is a harder problem than measuring heart rate. ECG-derived HRV comes from R-R intervals, the gap between electrical triggers. PPG-derived HRV comes from the gap between pulses arriving at your wrist, which researchers call pulse rate variability and treat as a substitute for HRV rather than an equivalent of it [2]. In one direct comparison of the two methods, heart rate differed by about 0.01% while PPG-derived RMSSD differed by 3 to 6% [4]. Those were five-minute fingertip recordings in young adults, so the figures do not transfer to a night on the wrist, but the overnight studies show the same asymmetry.

In the lab, every one of the six devices underestimated HRV, from 4.5 ms on WHOOP 3.0 to 22.4 ms on the Garmin [1].

Table 2. Overnight HRV (RMSSD) agreement with ECG in a single lab night, from Miller et al. (2022).

Two rows need reading carefully. WHOOP 3.0 was the only device that supplied raw R-R intervals and the only one with near-perfect agreement, which is a plausible mechanical explanation rather than a coincidence. Garmin’s 0.24 is not a measurement of overnight HRV at all: its three-minute sample was taken while the participant was awake, which the authors offer as the likely reason.

The pattern underneath the table is directional. Across five of the six devices, WHOOP excepted, HRV was overestimated at the low end of the ECG range and underestimated at the high end [1]. These devices compress the range they report. Read across your own nights, that would mean a restorative night coming out lower than it was and a bad one higher, though the study measured one night each in 53 people rather than many nights in one.

Three years later, at home, on newer hardware, the picture is better.

Table 3. Overnight HRV agreement with ECG across 536 home nights, from Dial et al. (2025).

The two studies used different samples, settings and agreement statistics, so this is not a clean before-and-after and I would not read a specific improvement out of it. But an average miss of roughly 4 ms on the two Oura rings and WHOOP 4.0 is a different quality of measurement from the 18.8 to 33.1 ms recorded by five of the six devices in 2022. The bottom of that table is worth naming too: the paper treats a MAPE above 10% as unacceptable, which puts Garmin at 10.52% and Polar at 16.32% on the wrong side of the authors’ own line.

Why does accuracy fall apart when you are awake?

TL;DR: The same watch on the same wrist produced acceptable HRV during sleep and high-error HRV during waking hours, in the same people on the same day.

The 24-hour study is the cleanest demonstration available, because sleep and wake come from the same participants, the same device and the same day [2]. During sleep, the Samsung watch produced acceptable heart rate and acceptable time-domain HRV against the chest ECG. During waking hours, only heart rate and average beat interval stayed satisfactory, while every other HRV parameter showed high errors and their correlations dropped from significantly high to low. The paper does not say why, though the authors note that PPG is highly susceptible to motion artifacts and environmental noise, and that collecting HRV during daily activities needs noise cancellation or signal-quality checks to be usable [2].

Motion does not degrade the signal equally for everyone, either. In a cycling protocol comparing a Fitbit Charge 5 against a Polar H10 chest strap, resting heart rate error averaged 2.8 bpm and was similar across three skin tone groups, while above 60% of heart rate reserve the mean error reached 16.5 bpm in the dark skin tone group against 3.5 bpm in the light [5]. Exercise heart rate deserves its own post and will get one for sure so stay tuned.

Lessons Learned: The overnight number is good, and the trend is harder to measure than I assumed

On Sunday I cycled 160 kilometers. That night my Oura reported HRV down and RHR up, both outside my usual range. None of it surprised me; I could feel the ride in my legs long before I opened the app. What the ring added was a meaningful, explainable number for something I already knew, and a record of how many days it took to come back. That is the job these numbers do, and the validation literature says they can do it well with Oura standing out with the best results.

What I found genuinely interesting reading the studies was that the precision differs between the two metrics. Your ring produces both from one sensor, one waveform, one night: same LEDs, same reflected pulses, same algorithm picking out the peaks. Across the four devices in the home study that report both, resting heart rate landed between 1.67% and 3.00% off the ECG value while HRV landed between 5.96% and 16.32% off [3]. Not one device escaped the pattern.

One more thing about Dial et al. study, because it is a useful lesson in how to read this literature. Two WHOOP researchers published a comment arguing the comparison was unfair to WHOOP, which weights its HRV toward slow-wave sleep and was compared against an all-night ECG average [6]. The reply is the interesting part: the authors say they had already run the stage-matched comparison in house and WHOOP came out worse that way, so the published analysis was the more generous of the two [7].

The same reply says the clearest thing in this whole literature about transparency. Dial and colleagues validate the number shown in your app rather than asking companies for raw intervals, because the app number is the one you see and every processing step behind it is undocumented [7].

And as usual, some tips for later:

• Overnight resting heart rate is accurate to roughly 1 to 2 bpm against a 5 to 7 bpm threshold for a meaningful change, so a sustained rise across several nights is a real signal.

• An HRV reading taken while you are awake and moving is less trustworthy than the one recorded during sleep.

• Do not compare HRV across devices, or across generations of the same device. Two generations of the same ring did not perform identically against ECG in the same study.

• After your next hard session, compare that night against an easy one. Both HRV and RHR numbers should move, and you can trust the direction of each.

If this was worth your time, subscribe for more. Each post explores the science behind wearable health data: how it’s measured, what it means, and what the research says.

Subscribe now

This post is part of The Science Behind Wearables, a series explaining the health metrics your devices track, built around the health scores we’re developing with the Open Wearables team. Open Wearables is an open-source platform for standardised access to health data from consumer wearables, supported and maintained by Momentum.

Know someone obsessed with wearables? Forward this their way.

Share The Science Behind Wearables

Sources

[1] Miller, D. J., Sargent, C., & Roach, G. D. (2022). A Validation of Six Wearable Devices for Estimating Sleep, Heart Rate and Heart Rate Variability in Healthy Adults. Sensors, 22(16), 6317. doi:https://doi.org/10.3390/s22166317

[2] Sarhaddi, F., Kazemi, K., Azimi, I., Cao, R., Niela-Vilen, H., Axelin, A., Liljeberg, P., & Rahmani, A. M. (2022). A comprehensive accuracy assessment of Samsung smartwatch heart rate and heart rate variability. PLOS ONE, 17(12), e0268361. doi:https://doi.org/10.1371/journal.pone.0268361

[3] Dial, M. B., Hollander, M. E., Vatne, E. A., Emerson, A. M., Edwards, N. A., & Hagen, J. A. (2025). Validation of nocturnal resting heart rate and heart rate variability in consumer wearables. Physiological Reports, 13(16), e70527. doi:https://doi.org/10.14814/phy2.70527

[4] Burma, J. S., Griffiths, J. K., Lapointe, A. P., Oni, I. K., Soroush, A., Carere, J., Smirl, J. D., & Dunn, J. F. (2024). Heart Rate Variability and Pulse Rate Variability: Do Anatomical Location and Sampling Rate Matter? Sensors, 24(7), 2048. doi:https://doi.org/10.3390/s24072048

[5] Hung, S. H., Serwa, K., Rosenthal, G., & Eng, J. J. (2025). Validity of heart rate measurements in wrist-based monitors across skin tones during exercise. PLOS ONE, 20(2), e0318724. doi:https://doi.org/10.1371/journal.pone.0318724

[6] Grosicki, G. J., & Presby, D. M. (2025). Accurate comparison of wearables requires contextual equivalence. Physiological Reports, 13(23), e70710. doi:https://doi.org/10.14814/phy2.70710

[7] Dial, M. B., Hollander, M. E., Vatne, E. A., Emerson, A. M., Edwards, N. A., & Hagen, J. A. (2025). Contextual equivalence for accurate comparison of wearables requires transparency. Physiological Reports, 13(23), e70706. doi:https://doi.org/10.14814/phy2.70706

What did this study actually measure?

TL;DR: The researchers drew on the UK Biobank, a cohort of roughly half a million adults aged 37 to 84, and mapped self-reported sleep against 23 biological ageing clocks spanning 17 organs and three data types.

The 23 clocks came from three different windows onto the body: 11 from plasma proteomics, 5 from plasma metabolomics, and 7 from MRI scans. Together they cover 17 organs, so ageing is measured organ by organ rather than as one number for the whole body. Each clock produces a biological age gap (BAG). The algorithm is trained on people without disease to guess a person’s age from one organ’s measurements, and the BAG is how far that guess lands from their real age. A positive gap means the organ looks older than the calendar says it should. Sleep duration came from a UK Biobank questionnaire, and the team used generalized additive models, which let the data reveal whatever shape the relationship takes rather than forcing a straight line through it [1].

Is there a sweet spot for how long you sleep?

TL;DR: Nine of the 23 clocks showed a clear U-shape, with the youngest-looking organs clustering between 6.4 and 7.8 hours of sleep and both shorter and longer nights linked to accelerated ageing.

Across the nine clocks that crossed the significance threshold, the bottom of the curve, the sleep duration linked to the youngest-looking organs, sat between 6.4 and 7.8 hours depending on the organ and on sex [1]. On that basis the team defined short sleep as under 6 hours, long sleep as over 8, and normal as the band in between. The pattern at the two ends was not subtle. Both short and long sleepers died at a higher rate during follow-up: 50% higher for short sleep (hazard ratio 1.50) and 40% higher for long sleep (hazard ratio 1.40) [1]. Short and long sleep together accounted for 153 significant links to disease outcomes out of 726 the team tested, with short sleep behind most of them and hazard ratios reaching 6.7 [1].

Why is the whole-body view the real breakthrough?

TL;DR: The U-shape appeared in nine clocks across brain and body, and short and long sleep left distinct fingerprints.

This is the part I find most exciting. Sleep and the brain have been studied for decades. Seeing the same signature in the lungs, liver, immune system and skin, read from blood proteins, and again in MRI clocks for the brain, adipose tissue and pancreas, tells us something the brain-only work could not. Sleep is a whole-body input, and these organs are not ageing in isolation. The same sleep pattern showed up in several organ clocks at once.

Short and long sleep also turned out to be genetically different. Short sleep correlated broadly with disease across the body, from ischaemic heart disease and type 2 diabetes to depression, anxiety and back pain. Long sleep had a narrower profile, concentrated in brain-related conditions like major depressive disorder, schizophrenia and bipolar disorder [1].

That difference showed up again when the team traced how each connects to late-life depression. Long sleep’s link ran mostly through accelerated organ ageing, with the brain clock alone accounting for 62% of the effect. Short sleep’s link was far more direct, showing up as depression largely without any of the seven MRI clocks in between with only ageing in fat tissue playing a mediating role. So the two extremes seem to reach the same destination by different roads.

Does the right amount of sleep depend on who you are?

TL;DR: The ideal sleep duration shifted by organ and by sex, from about 6.4 hours for some tissues to 7.8 hours for others, with women’s optima sitting slightly higher than men’s on several clocks.

There is no single magic number that suits every tissue. The brain’s protein-based clock put its lowest point close to 7.8 hours. The endocrine clock, built from blood metabolites, sat nearer 6 to 6.7 hours. The brain’s MRI clock came in around 6.4 hours [1]. Each of those figures is the low point of a fitted curve for a whole group, not a personal target, and they sit well over an hour apart.

The picture also differed between women and men. For the endocrine clock, women’s lowest point was 6.67 hours against 6.06 for men, a gap of well over half an hour, and this was the one clock where the sex-by-sleep interaction was statistically strong. On the brain protein clock, women’s optimum was 7.82 hours and men’s 7.70. Across the nine significant clocks, women’s ideal window ran from 6.5 to 7.8 hours and men’s from 6.4 to 7.7. Although the differences are modest, they are consistent. I find the second sex difference particularly interesting. Men carried more brain ageing on the MRI clock, while women carried more on the brain protein clock. The authors read this as the two clocks measuring different layers of biology, since brain MRI reflects lifelong structural and hormonal trajectories, whereas brain proteins in the blood respond to inflammation, endocrine signalling and how freely those proteins cross out of the brain.

Can you actually change your sleep, and does causation matter?

TL;DR: The pattern looks environmental rather than inherited, and although the study cannot prove that sleep drives the ageing it maps, the case for protecting your sleep does not rest on that proof.

The team ran a genome-wide scan across more than 300,000 people and found only eight genetic locations tied to their short and long sleep groups, each compared against normal sleep. That thin signal is part of why they read the U-shape as environmental. The clearer test came next: when they replaced the measured organ clocks with genetic predisposition to organ ageing, the curve mostly flattened, which points away from inherited susceptibility as the thing producing it [1]. The authors are careful to frame this as a hypothesis rather than a finding, that the U-shape between sleep and biological age is primarily environmentally driven and therefore modifiable. Of everything people try in order to age well, sleep is one of the few that is both consistently linked to health outcomes and reachable without an elaborate routine.

That modifiability only counts for something if the arrow points the way you hope, so it is worth being clear about what the study can and cannot show. It is observational, so the association could in principle run in either direction. The authors went further than most and ran genetic causal tests (Mendelian randomization), which found little evidence that disease drives abnormal sleep. That leaves room for sleep acting as a risk factor rather than a symptom, and they say plainly that bidirectional effects cannot be ruled out. I would gently push back on treating any of this as a reason to wait for cleaner proof. Better sleep already carries a long, well-established list of benefits on its own, from mood to metabolic health, and none of that evidence depends on this paper. You do not need a settled causal arrow to justify protecting your sleep tonight, and for most people the cost of trying is small.

What could wearables add?

TL;DR: Sleep here was self-reported, so an objective, wearable-based follow-up at this scale could enhance the findings.

Sleep duration came from a single questionnaire item, and self-report correlates only moderately with what actigraphy or polysomnography actually record [1]. A questionnaire cannot see fragmentation, time spent awake after first falling asleep, or the gap between time in bed and time genuinely asleep.

This is precisely the space wearables were built for. A follow-up using objective, device-measured sleep across a cohort this size would be extraordinary. It would put real numbers on the sleep continuity measures and open up the time spent in each sleep stage. The biggest gain would be cleaner data underneath all of it.

Lessons Learned: Protect your sleep window before you optimise anything else

TL;DR: This study is one more reason to treat sleep as a habit worth adjusting, and a wearable takes the guesswork out of knowing where you stand.

The part I keep coming back to is that sleep is one of the few things in this picture you can actually adjust. The authors suspect the link between sleep and biological ageing is environmental rather than inherited, which means the window they found is somewhere you can move toward, and that seems worth a bit of attention.

Knowing where you fall used to mean keeping a sleep diary, which almost nobody sustains for long. A wearable does the same job quietly in the background, so you can see your real numbers and get a sense of whether a change is helping. I also found it oddly reassuring that one habit showed up in the lungs, liver, immune system and brain at once. It suggests the body works as a connected system, and that the changes worth making tend to be the ones that support the whole of it.

• The link between sleep and biological ageing looks environmental rather than inherited, so the window they found is somewhere you can move toward.

• A wearable can do the tracking for you, so you can see your real numbers and notice whether a change is helping, without the effort of a sleep diary.

• A range of roughly six to eight hours is a more useful target than one perfect figure, since the lowest point shifted from one organ clock to the next.

• The findings are whole-body rather than brain-only, so changes that support the whole system will probably serve you better than ones aimed at a single organ.

If this was worth your time, subscribe for more. Each post explores the science behind wearable health data: how it’s measured, what it means, and what the research says.

Subscribe now

This post is part of The Science Behind Wearables, a series explaining the health metrics your devices track, built around the health scores we’re developing with the Open Wearables team. Open Wearables is an open-source platform for standardised access to health data from consumer wearables, supported and maintained by Momentum.

Know someone obsessed with wearables? Forward this their way.

Share The Science Behind Wearables

Sources:

[1] MULTI Consortium, O’Toole, C.K., Song, Z., Anagnostakis, F., Yang, Z., Tian, Y.E., Duggan, M.R., Zou, C., Leng, Y., Cai, Y. and Bai, W., 2026. Sleep chart of biological ageing clocks in middle and late life. Nature, pp.1-11.

What is resting heart rate and how to interpret it?

TL;DR: Resting heart rate is how many times your heart beats per minute at rest, normally 60 to 100 in adults, with a lower number generally pointing to a fitter cardiovascular system.

Resting heart rate is the number of times your heart beats per minute while you are at rest. In healthy adults that usually falls between 60 and 100 beats [1]. A lower resting heart rate generally reflects a more efficient cardiovascular system, while an elevated one can signal stress, illness or poor sleep. Several factors push the number up or down, and you can see them in the table below.

Table 1. Factors which influence resting heart rate.

RHR is associated with a number of factors that sit outside of our control like age, genetics and sex [2]. Women’s RHR tends to be higher than men’s, simply due to anatomy: we have smaller hearts, which beat faster to move the same amount of blood [3]. Women also see more fluctuation tied to hormonal changes, RHR typically rises during the second half of the menstrual cycle [4]. Independently of sex, a sustained rise in RHR can mean you’re fighting off an illness, or it can be a sign of chronic stress. Finally, some medications affect RHR too and these changes are best to be consulted with your doctor.

That said, many of the factors driving your RHR are within your control. For example, the biggest difference lies in physical activity. While RHR spikes after an intense training session, regular exercise is what brings your baseline down over time. Bedtime matters more than people think: one study following 255,736 nights of Fitbit data from 557 college students found that going to bed just 30 minutes later than usual raised RHR significantly for that night’s sleep [5]. Stimulants including caffeine and nicotine all push RHR up. Same goes with alcohol: a single standard drink raises heart rate by about 5 beats per minute on average, and it climbs further with higher doses [6]. Dehydration raises RHR too, giving you one more reason to hit your daily water intake goal [7].

Once you know what’s driving the number, most of it comes down to habits you already have some control over.

Now let’s take a look at how wearables actually measure these values and how the method compares to the clinical standard.

How do wearables measure heart rate, and how does that compare to the clinical standard?

TL;DR: Wearables estimate heart rate optically through photoplethysmography (PPG), reading pulses of blood volume under the skin, while the clinical standard, the electrocardiogram (ECG), measures the heart’s electrical signal directly, which is why it remains more precise.

Every heartbeat starts as an electrical event, triggered by the sinoatrial node and picked up by an ECG through electrodes on the skin that detect the millivolt-level voltage changes as the impulse crosses the heart. That gives clinicians a direct read on the electrical cause of each beat, the P wave, QRS complex, and T wave, with heart rate and HRV calculated from the R-R intervals between peaks. Your wearable does not have that access. Instead it uses PPG: LEDs shine light into the skin, usually green for everyday heart rate tracking and red or infrared for blood oxygen, and a photodetector measures how much bounces back as blood volume rises and falls with each pulse [8]. That reflected light forms a waveform, and onboard algorithms find the peaks, calculate the time between them (the inter-beat interval, in milliseconds), and convert that into beats per minute; 60,000 divided by the average interval, smoothed over a few seconds before it reaches your screen. Because PPG reads a downstream mechanical effect of the heartbeat rather than its electrical cause, and because motion, skin tone, and fit all interfere with the light signal, it is inherently less precise than ECG, which is why ECG remains the clinical reference standard.

Fig. ECG and PPG traces.

How accurate are wearables at measuring resting heart rate?

TL;DR: For resting heart rate specifically, wearables are accurate; the real weakness is consistency.

Wearables measure resting heart rate well. At rest there is little motion to corrupt the optical signal, and validation studies against ECG repeatedly show wrist and ring devices tracking closely under these conditions [9]. Performance does vary by brand, some devices hold up better than others, but the broader picture is consistent: at rest, PPG-based heart rate is a solved problem. It gets more complicated once you add motion, however, I will cover the accuracy of wearables-derived cardiac metric in my next article. So if accuracy is largely resolved, where does the problem lie? The answer is consistency. Not only across devices, but within the same device from one day to the next.

How do wearable providers calculate resting heart rate?

TL;DR: Every brand defines resting heart rate differently, and some are not even consistent with themselves.

Ask five wearables what your resting heart rate is, and you can get five different numbers, and the reason has nothing to do with which sensor is better. There is no unified way to define resting heart rate across consumer wearables, and sometimes providers are not even consistent within their daily measures.

Apple and Fitbit sample RHR at moments while you’re resting, not sleeping, and that window moves from day to day. You might have been sitting during Monday’s reading and lying down during Friday’s, and the app gives you no way to tell which. Random measures during the day are also prone to confounding factors. Cardiovascular variables respond to stressors like exercise or a meal, and to read RHR or HRV as a signal of your physiology rather than your reaction to whatever you just did, you need a reading taken well outside those stressor windows which this approach cannot guarantee.

Night-based measurements avoid these confounders, but they trade one problem for another. A single lowest reading is more prone to artifact, so averaging across the night is the sounder approach, which is what Oura gives you: both an average nightly RHR and a lowest value, and the average is the one I actually track on my own ring. Whoop does something similar but leans heavily on your deepest sleep stage specifically, so how much deep sleep you got, and when you got it, becomes part of what’s driving the number, and that varies night to night on its own. Garmin averages your lowest 30 minutes across a full 24-hour window, which usually lands at night but isn’t guaranteed, and when it doesn’t, the consistency is affected. Whether a provider is averaging your lowest readings or taking a single lowest value, that window shifts from night to night, and so does your number.

Morning measurement sidesteps both problems. You’re far enough from yesterday’s stressors, and you avoid the night-to-night drift that comes from anchoring to sleep stages. Polar recommends this to its users for that reason, and it’s the closest thing to capturing your actual physiological baseline.

Table 2. RHR calculations from common wearable providers.

Trend matters more than any single reading, a point I keep coming back to in this newsletter, but a trend is only as good as the consistency behind it.

Lessons Learned: Why is such a simple metric so complicated

When you only ever wear one device, you rarely question the number. Working on Open Wearables changed that for me. Pull resting heart rate from several providers and line them up, and you see that each one means something different by it, gaps that are hard to reconcile.

For personal use what you want from the metric is accuracy and day-to-day consistency. But when a single wearable keeps shifting when it measures you, your own trend line gets just as hard to read as comparing two different brands.

What I keep coming back to is how wearables entering clinical space changes how we should think about this data. Wouldn’t it be simpler if we could always treat RHR as the same measurement, taken at the same consistent time, across every provider? Wouldn’t that be easier for users and clinicians alike to interpret? I come from experimental neuroscience, where I spent years running lab experiments, and the whole point was consistency: keep the conditions identical so results can be compared across cohorts. Maybe it’s time we held wearables to the same standard, now that they’re moving into that same space.

If this was worth your time, subscribe for more. Each post explores the science behind wearable health data: how it’s measured, what it means, and what the research says.

Subscribe now

This post is part of The Science Behind Wearables, a series explaining the health metrics your devices track, built around the health scores we’re developing with the Open Wearables team. Open Wearables is an open-source platform for standardised access to health data from consumer wearables, supported and maintained by Momentum.

Know someone obsessed with wearables? Forward this their way.

Share The Science Behind Wearables

Sources:

[1] American Heart Association https://www.heart.org/en/healthy-living/exercise-and-physical-activity/fitness-basics/target-heart-rates

[2] Van De Vegte, Y. J., Eppinga, R. N., Van Der Ende, M. Y., Hagemeijer, Y. P., Mahendran, Y., Salfati, E., ... & Erdmann, J. (2023). Genetic insights into resting heart rate and its role in cardiovascular disease. Nature communications, 14(1), 4646.

[3] Prabhavathi, K., Selvi, K., Poornima, K. N., & Sarvanan, A. (2014). Role of biological sex in normal cardiac function and in its disease outcome–a review. Journal of clinical and diagnostic research: JCDR, 8(8), BE01.

[4] Moran, V. H., Leathard, H. L., & Coley, J. (2000). Cardiovascular functioning during the menstrual cycle. Clinical physiology, 20(6), 496-504.

[5] Faust, L., Feldman, K., Mattingly, S. M., Hachen, D., & V. Chawla, N. (2020). Deviations from normal bedtimes are associated with short-term increases in resting heart rate. NPJ digital medicine, 3(1), 39.

[6] Tasnim, S., Tang, C., Musini, V. M., & Wright, J. M. (2020). Effect of alcohol on blood pressure. Cochrane Database of Systematic Reviews, (7).

[7] Charkoudian, N., Halliwill, J. R., Morgan, B. J., Eisenach, J. H., & Joyner, M. J. (2003). Influences of hydration on post‐exercise cardiovascular control in humans. The Journal of Physiology, 552(2), 635-644.

[8] Tamura, T., Maeda, Y., Sekine, M., & Yoshida, M. (2014). Wearable photoplethysmographic sensors—past and present. Electronics, 3(2), 282-302.

[9] Nelson, B. W., Low, C. A., Jacobson, N., Areán, P., Torous, J., & Allen, N. B. (2020). Guidelines for wrist-worn consumer wearable assessment of heart rate in biobehavioral research. NPJ digital medicine, 3(1), 90.

Two back-to-back studies recently published by the science team at WHOOP offer an insight into this question. The first used continuous wearable data from nearly 21,000 WHOOP users to document what alcohol does to physiological variables a device can track: resting heart rate, HRV, sleep, and next-day activity [1]. The second followed 30,000 users across 72 weeks to ask whether exposure to that feedback correlates with behavioral change, specifically, drinking less over time [2].

What does alcohol do to your body?

TL;DR: Alcohol raises resting heart rate, suppresses HRV, and disrupts sleep in ways that are dose-dependent and measurable, which is why your recovery score drops the morning after drinking.

The physiology study starts with something most people experience but rarely quantify. Alcohol feels sedating, but what follows in your body is more disruptive than it appears. Using data from 20,968 participants, Grosicki et al. took a within-person approach: rather than comparing drinkers to non-drinkers, they compared each participant’s nights with above-average drinking to their own baseline [1]. One drink above a personal average was associated with a resting heart rate increase of 2.4 bpm in men and 2.8 bpm in women. Heart rate variability (HRV) fell by 3.3 ms in men and 3.8 ms in women for that same one-drink difference, reaching 5.1 ms and 5.6 ms respectively at five drinks above usual. Women showed more pronounced responses than men, a finding the authors attribute to lower first-pass metabolism and smaller alcohol distribution volume. Younger adults also showed larger disruptions, with the effect declining gradually across age groups. Timing mattered too: drinking 60 minutes earlier than usual was associated with meaningfully attenuated cardiovascular disruption [1].

Elevated resting heart rate, suppressed HRV, disrupted sleep, and reduced next-day activity are all within a consumer wearable’s measurement range. In practice, a night of drinking produces a recognizable signature in your data: higher than usual resting heart rate, lower HRV, a reduced recovery. Rather than identifying that you drank, the device is measuring the physiological consequences accurately enough that the output reflects real disruption.

Table: Acute physiological effects of alcohol across wearable-tracked metrics.

If you wear a tracker, you do not need to take the researchers’ word for it. Compare two consecutive nights in your own data: one after drinking, one without. The HRV drop and RHR elevation described at population scale in this study appear with consistency at the individual level too. The within-person design is actually the study’s most practically useful feature: the reference point is your own baseline, not a population average. What counts as disruption is calibrated to you.

What does the data show about behavior change?

TL;DR: Across 30,000 WHOOP users tracked for 72 weeks, the daily probability of drinking fell from 23.0% to 17.2%, and drinking volume fell by over one drink per week, with reductions sustained across age groups and both sexes.

A 2026 observational cohort study followed 30,000 adults from their first week of WHOOP membership through 72 weeks, capturing daily self-reported alcohol use via the platform’s morning journal prompts [2]. Participants were enrolled throughout 2023, contributed an average of 59 weeks of data, and were evenly split by sex.

At baseline (weeks 1 to 12), the adjusted daily probability of alcohol use was 23.0% (95% CI 22.7 to 23.3). By weeks 61 to 72, it had fallen to 17.2% (95% CI 16.9 to 17.4), an absolute reduction of 5.8 percentage points. In a sensitivity analysis restricted to participants who reported full drink counts, drinking volume decreased by 1.11 drinks per week. The frequency decline was not offset by higher consumption on drinking occasions.

The pattern held across subgroups. Women showed a greater relative decline than men from baseline to week 72. Older adults showed progressively larger reductions than younger cohorts. The declines appeared in both people who set explicit alcohol-related goals in the app and those who did not, with the relative reduction larger among non-goal-setters [2]. The authors propose three possible mechanisms: increased health awareness from using the platform, heightened sensitivity to physiological feedback, and broader lifestyle changes following wearable enrollment [2].

The sex difference is worth pausing on. The physiology study found that women experience greater cardiovascular disruption per drink. The behavioral study found that women also reduced their drinking more than men over 72 weeks. Whether those two findings are causally connected, greater physiological disruption producing a stronger behavioral signal, is not something either study can confirm. But the convergence across both datasets is interesting and suggests a research direction worth pursuing.

The study has no control group, which means the decline could reflect secular trends, selection effects among health-motivated subscribers, or natural regression following an initial period of higher-than-usual drinking around enrollment. The authors are explicit about this: the design establishes correlation, not causation.

What does this study actually capture?

TL;DR: The 30,000 participants are not a representative sample of people who drink: they are likely a self-selected group of health-motivated wearable users, and that distinction shapes what the findings can and cannot tell us.

As someone who spends a lot of time analyzing digital health data, I always read these kinds of results with a critical eye. There is a well-documented problem in observational health research called healthy user bias: the tendency for interventions to be studied in, and to benefit, people who were already more health-motivated to begin with. When we observe a sustained drop in alcohol consumption over 72 weeks, we have to ask the hard scientific question: is the wearable driving this change, or are we watching a group of intrinsically motivated users do what they were already inclined to do?

The demographics of wearable ownership make this tension concrete. Recent demographic studies consistently show that device ownership heavily skews affluent and educated. For instance, national survey data reveals that individuals with advanced degrees or household incomes over $200,000 have more than double the odds of owning a wearable compared to the general population [3]. A person who purchases a WHOOP subscription, logs their alcohol intake daily, and maintains that habit for over a year is not the person at greatest risk from their drinking. Because this is an observational study, it cannot definitively prove that the device caused the behavioral shift. I would argue that dismissing these findings on those grounds is too cynical. In behavioral science, sustaining any lifestyle change for 72 weeks is genuinely difficult, even for the highly motivated. The 5.8 percentage point decline held across all age groups and both sexes, and the non-goal-setters showed it as clearly as those who set explicit targets. That consistency is harder to explain away as a pure selection artifact.

The WHOOP platform offers personalized insights, goal-setting prompts, and contextual explanations alongside raw metrics. Research on digital health tools more broadly shows that passive data exposure rarely produces behavior change on its own, and active engagement features appear to drive most of the effect [4]. The study cannot isolate which elements mattered.

There is also a measurement question worth holding alongside the findings. All behavioral data comes from voluntary morning journal entries on a wellness platform. People engaged enough to log their alcohol intake daily may be precisely the population for whom self-monitoring itself is the active ingredient, a behavior change technique with decades of evidence behind it, independent of whatever the device’s sensors record [4, 5]. The wearable may be the delivery mechanism for something that would work with or without the HRV data. The daily logging and the physiological feedback are intertwined in this design, and separating them is not possible from the data alone.

What would it take to extend these findings?

TL;DR: The study provides a strong rationale for testing wearable-based alcohol feedback with broader, higher-risk populations, but getting there requires addressing access, cost, and the design of the intervention itself.

The public health relevance of wearable-assisted behavior change depends on whether the mechanism generalizes beyond health-motivated, higher-income users. A 2025 randomized controlled trial of at-risk drinking young adults found that a wearable-supported behavioral program produced clinically meaningful reductions in alcohol consumption and improved sleep-related outcomes [6]. That design gets closer to causal inference, and the structured intervention model is more representative of how wearable programs would need to be deployed at scale. Deployment in federally qualified health centers and community settings has begun to appear in the literature. A 2022 study surveying over 1,000 adult patients across six FQHCs found genuine interest in wearable devices, but concluded that widespread adoption would require educational support and financial investment in devices [7]. Cost and device access remain practical barriers. Program design matters separately from access: a wearable worn without support for interpreting its output is likely to produce different results than one embedded in a structured intervention.

The WHOOP study is a compelling illustration of what sustained physiological feedback looks like at scale in a motivated population. Testing the same mechanism where the need is greatest is where the harder, more consequential work lies.

Lessons Learned: Reading behavior-change research in digital health

TL;DR: These two studies establish that wearables can detect alcohol's physiological effects and that users of health tracking platforms reduce their drinking over time. The population gap between research participants and at-risk communities is the central open question.

Together, the two studies trace a coherent arc: here is what alcohol does to your body at the physiological level, and here is evidence that people who can see that data over time drink less. Both findings hold up on their own terms. The limitations of the behavioral study are limitations of design and population, not of the underlying premise.

A few things to keep in mind when reading this kind of research:

• Within-person designs, as used in the physiology study, are a methodological strength. Comparing each participant to their own baseline removes between-person noise and makes effect sizes more interpretable. It also means the numbers translate directly to what you can observe in your own data which I encourage you to do.

• Observational designs cannot establish causality. The 5.8 percentage point reduction in daily drinking probability is meaningful, but the study cannot confirm whether the platform drove it. Independent replication with control groups is the next evidentiary step.

• The population question is not peripheral. Alcohol-related harm is disproportionately concentrated in lower-income, less health-engaged populations. A finding in WHOOP subscribers does not automatically transfer, and the equity gap in wearable ownership is large enough that assuming it does would be a significant leap.

• Self-monitoring is itself a behavior change technique. Whether the effect here comes from physiological feedback, daily journaling, or both is something this design cannot resolve.

• Conflict of interest disclosures are relevant context. Both WHOOP studies were authored by WHOOP employees. That does not invalidate the data, but it is worth holding alongside the findings, particularly where conclusions favor the product.

If this was worth your time, subscribe for more. Each post explores the science behind wearable health data: how it’s measured, what it means, and what the research says.

Subscribe now

This post is part of The Science Behind Wearables, a series explaining the health metrics your devices track, built around the health scores we’re developing with the Open Wearables team. Open Wearables is an open-source platform for standardised access to health data from consumer wearables, supported and maintained by Momentum.

Know someone obsessed with wearables? Forward this their way.

Share The Science Behind Wearables

Sources

[1] Grosicki, G. J., Robinson, A. T., Joyner, M. J., Carter, J. R., von Hippel, W., Presby, D. M., Fielding, F., Bigalke, J. A., Kim, J., Chapman, C., & Holmes, K. E. (2026). Real-world effects of alcohol on heart rate, sleep, and physical activity by age and sex. PLOS digital health, 5(3), e0001284. https://doi.org/10.1371/journal.pdig.0001284

[2] Grosicki, G. J., Hippel, W. V., Fielding, F., Chapman, C. J., Presby, D. M., Leota, J., & Holmes, K. E. (2026). Alcohol Use Trajectories During the First 72 Weeks of WHOOP Wearable Platform Membership: Observational Cohort Study. JMIR mHealth and uHealth, 14, e91288. https://doi.org/10.2196/91288

[3] Nagappan, A., Krasniansky, A., & Knowles, M. (2024). Patterns of Ownership and Usage of Wearable Devices in the United States, 2020-2022: Survey Study. Journal of medical Internet research, 26, e56504. https://doi.org/10.2196/56504

[4] Milne-Ives, M., Homer, S. R., Andrade, J., & Meinert, E. (2023). Potential associations between behavior change techniques and engagement with mobile health apps: a systematic review. Frontiers in psychology, 14, 1227443. https://doi.org/10.3389/fpsyg.2023.1227443

[5] Korotitsch, W. J., & Nelson-Gray, R. O. (1999). An overview of self-monitoring research in assessment and treatment. Psychological Assessment, 11(4), 415–425. https://doi.org/10.1037/1040-3590.11.4.415

[6] Fucito LM, Ash GI, Wu R, et al. (2025). Wearable Intervention for Alcohol Use Risk and Sleep in Young Adults: A Randomized Clinical Trial. JAMA Network Open, 8(5), e2513167. https://doi.org/10.1001/jamanetworkopen.2025.13167

[7] Holko M, Litwin TR, Munoz F, et al. (2022). Wearable fitness tracker use in federally qualified health center patients: strategies to improve the health of all of us using digital health devices. npj Digital Medicine, 5, 53. https://doi.org/10.1038/s41746-022-00593-x

How does your wearable track sleep?

TL;DR: Clinical sleep studies measure brain activity directly; wearables use movement and heart rate as proxies, and the accuracy limits of wearables follow directly from that difference.

Understanding the measurement gap between the clinical settings and wearables matters before you read a single number from your device.

The gold standard: polysomnography (PSG)

In clinical settings, sleep is measured using polysomnography (PSG). A PSG session connects you to sensors that record four independent physiological signals simultaneously:

• Electroencephalography (EEG): measures your brain waves and provides a direct window into the sleep stages

• Electrooculography (EOG): identifies the eye movements that define REM sleep

• Electromyography (EMG): detects the muscle paralysis characteristic of REM

• Heart rate and respiration: monitors autonomic function and screens for breathing disorders

A trained technician scores the recording in 30-second windows called epochs, assigning each to Wake, N1, N2, N3, or REM according to AASM guidelines [1]. An 8-hour night produces roughly 960 epochs. Even among expert scorers, agreement on sleep staging ranges from 78.9% to 82.6% [2]. That variability sets a practical ceiling on how accurately any device can be validated against PSG, regardless of sensor quality.

The wearable approach: actigraphy and PPG

Your wearable has no access to EEG, EOG, or EMG signals. Instead it relies on two sensors:

• Actigraphy (via accelerometer): measures movement, interpreting stillness as sleep and movement as wakefulness

• Photoplethysmography (PPG): measures heart rate, HRV, and respiratory rate by shining light into the skin and detecting changes in blood volume

What converts these raw signals into a sleep stage label are algorithms trained on datasets where sensor data was collected simultaneously with PSG. The algorithm learns which patterns of movement and cardiac signals correlate with which PSG-scored stages, then applies that learned mapping to your data each night [2].

Fig. Polysomnography vs. Wearable sleep detection.

How is sleep accuracy measured?

TL;DR: A device that labels every epoch as sleep scores above 90% sensitivity without detecting a single awakening, which is why wake sensitivity, not sleep sensitivity, is the more informative measure of device quality.

Sleep sensitivity

In plain English: Of all the epochs where PSG said “sleep,” what percentage did the device also correctly label as “sleep”? For sleep overall, this number is almost always above 90%. A typical 8-hour night contains roughly 870 sleep epochs and 90 wake epochs. A device biased toward labelling everything as sleep still scores ~91% sensitivity without detecting a single awakening. That is why overall sleep sensitivity, while commonly reported, tells you less about device quality and more about the structure of the night.

Sleep specificity (Wake sensitivity)

In plain English: Of all the epochs where PSG said “wake,” what percentage did the device also correctly label as “wake”? This is where wearables genuinely struggle. The main signal most devices use to detect wakefulness is movement, supplemented by heart rate changes. But consider what wake after sleep onset (WASO) looks like mid-night: you are lying still, eyes closed, heart rate low. From an accelerometer’s perspective, this is indistinguishable from sleep. The device misses these brief wake periods and scores them as sleep, producing false positives that accumulate in your reported sleep time.

Per-stage sensitivity

Per-stage sensitivity follows the same logic as overall sleep sensitivity, applied separately to light sleep, deep sleep, REM, and wake. The overall agreement across all four stages is often expressed as Cohen’s kappa (κ), which adjusts for the agreement you would expect by chance given how frequently each stage appears in a typical night. The scale runs from 0 (no better than chance) to 1 (perfect agreement), with 0.21-0.40 considered fair, 0.41-0.60 moderate, and 0.61-0.80 substantial.

Which validation studies to look at?

TL;DR: Three recently published PSG validation studies test overlapping devices together covering ten consumer wearables.

To assess sleep tracking accuracy across devices, I focused on three recently published validation studies, all using more up-to-date hardware-generation hardware and each with a meaningfully different design.

Robbins et al. (2024), recruited 35 adults across a single lab night, each wearing three devices simultaneously: Oura Ring Gen3, Apple Watch Series 8, and Fitbit Sense 2 [3]. The study applied an extensive list of exclusion criteria, screening out any sleep disorders, mental health conditions, and a range of general health markers, the kind of controls that produce a clean sample but also a narrow one. The study was funded by Oura Ring Inc., with the lead author serving as a member of the Oura Medical Advisory Board, as disclosed in the paper.

Schyvens et al. (2025) tested six wrist-worn devices in a single study: Apple Watch Series 8, Fitbit Sense, Fitbit Charge 5, WHOOP 4.0, Withings Scanwatch, and Garmin Vivosmart 4 [4]. Sixty-two adults participated, each wearing two to four devices simultaneously. The sample is worth noting: 84% male, 52 of 62 participants. If you are a woman reading these numbers, they come primarily from men’s sleep. Unlike Robbins, suspected sleep apnea was not an exclusion criterion, making this cohort closer to a real-world population in that respect.

Lee et al. (2023) is the largest and most varied of the three: 75 participants across two independent medical institutions in South Korea, with a more balanced sex split (52% male) [5]. Eleven devices were tested across three categories: wearables, nearables (pad-type devices and motion sensor devices), and airables. Wearables, the category of devices I will focus on, included Fitbit Sense 2, Galaxy Watch 5, Google Pixel Watch, Oura Ring 3, and Apple Watch Series 8. They overlap with both earlier studies, which is what makes cross-study comparison possible.

How well do wearables track sleep and wake?

TL;DR: Wearables are good at detecting sleep but poor at detecting wake, and most devices tested overestimate how much sleep you actually got.

Sleep sensitivity sits between 91% and 98% across all devices and all three studies. As the epoch maths above shows, that number is less informative than it looks. What the data actually reveals is in the wake column.

Table: Sleep sensitivity and specificity for devices tested in Robbins et al. (2024).

In Robbins et al., Apple Watch S8 reaches 97% sleep sensitivity and 52.4% wake sensitivity within the same device, a 45 percentage point gap. Two out of every five wake periods go undetected. Oura Ring Gen3 and Fitbit Sense 2 do better at 68.6% and 67.7%, but still miss roughly one in three awakenings. Schyvens et al. reproduces the same pattern across six devices: Withings Scanwatch and Garmin Vivosmart 4 detect fewer than a third of wake epochs, the weakest results in the set.

Table: Sleep sensitivity and specificity for devices tested in Schyvens et al. (2025).

In Lee et al., no wearable reaches 50% wake sensitivity, and Google Pixel Watch sits at 23%.

Table: Sleep sensitivity and specificity for devices tested in Lee et al. (2023).

The practical consequence lands in your total sleep time figure. In Schyvens et al., five of the six devices significantly overestimate TST relative to PSG: Withings by ~39 minutes, Garmin by ~38 minutes, WHOOP by ~24 minutes, and Apple Watch by ~19 minutes. Your app is counting the time you lay awake as sleep. The two exceptions are both Fitbit devices: Fitbit Sense overestimates by just ~6 minutes without reaching statistical significance, and Fitbit Charge 5 by ~11 minutes, also non-significantly. Fitbit Sense 2 showed the same non-significant result in Robbins et al., making Fitbit the only device family to consistently land close to PSG on total sleep time across independent studies.

Which device is best at tracking sleep stages?

TL;DR: Deep sleep is the stage with the highest variance across devices and the one where individual weaknesses are most pronounced, a pattern visible in all three studies regardless of the metric used.

No device leads on all stages simultaneously, and the pattern that holds most consistently across all three studies is that deep sleep is where the spread is widest.

Apple Watch S8 is the clearest example. Its deep sleep sensitivity sits at 50.5% in Robbins, 50.7% in Schyvens et al., and 41.3% in Lee et al: the same weakness replicated across three independent cohorts. On light sleep and REM the picture is different: Apple Watch leads in both Robbins et al. (light 86.1%, REM 82.6%) and Schyvens et al. (light 83.3%). The pattern suggests a stage-specific problem rather than a general accuracy issue.

Fig. Sleep stage detection for devices tested in Robbins et al. (2024). Adapted from the original paper.

WHOOP 4.0, tested only in Schyvens et al., shows the opposite profile. Its deep sleep sensitivity of 69.6% is the highest for that stage across all six devices in that study, while light sleep sits at 62.0%, placing it in the lower half of the field. A device that detects deep sleep well but underperforms on light and wake produces a different stage distribution from what PSG would score, skewed toward deep sleep at the expense of the rest of the night.

Fig. Sleep stage detection for devices tested in Schyvens et al. (2025). Adapted from the original paper.

Fitbit is the most consistent performer across studies. Light, deep, and REM sensitivity stay within a relatively contained range across device variants and cohorts: Fitbit Sense 2 in Robbins et al. scores light 78%, deep 62%, REM 68%; Fitbit Sense in Schyvens scores light 73%, deep 51%, REM 61%; Fitbit Sense 2 in Lee et al. scores light 77%, deep 67%, REM 68%. Deep sleep dips in Schyvens et al. but the profile does not show the sharp stage imbalances visible in Apple Watch or WHOOP.

Fig. Sleep stage detection for devices tested in Lee et al. (2023). Adapted from the original paper.

Oura Ring is the most complicated case. Its deep sleep sensitivity of 79.5% in Robbins et al. is the highest single figure for that stage across all three studies, and it is the only device in this study where no stage duration differed significantly from PSG. In Lee et al., deep sleep sensitivity remains high at 77.8%, but the deep sleep specificity is 80%, the lowest of any wearable tested. Roughly one in five non-deep-sleep epochs gets labelled as deep sleep. High sensitivity with low specificity means over-labelling: the device catches most real deep sleep but also pulls in epochs that PSG would score as something else.

The kappa values reinforce the same pattern from a different angle. No device reaches substantial four-stage agreement consistently across studies. Oura Ring Gen3 scores highest in Robbins et al. at κ = 0.65, the only value in the substantial range across any study, but drops to κ = 0.35 in Lee et al.

Table: Four-stage agreement for devices tested in Robbins et al. (2024).

Apple Watch follows the same trajectory: κ = 0.60 in Robbins et al., 0.53 in Schyvens et al., 0.30 in Lee et al. Fitbit lands in moderate agreement across all three studies, the most stable cross-study result in the set.

Table: Four-stage agreement for devices tested in Schyvens et al. (2025).

At the bottom, Withings Scanwatch and Garmin Vivosmart 4 score κ = 0.22 and 0.21 in Schyvens et al., just above the threshold for fair agreement.

Table: Four-stage agreement for devices tested in Lee et al. (2023).

The consistently higher kappa values in Robbins et al. across all three devices are worth reading alongside the study's design: a tightly controlled sample of healthy adults with an extensive list of exclusion criteria. Cleaner, more homogeneous sleep might be easier to classify correctly, which may partly explain why all three devices perform better in that study.

How should you interpret these studies?

TL;DR: No single validation study tells the full story: sample composition, hardware generation, study design, and funding source all shape what the numbers mean for you.

The three studies covered here are a starting point, not an exhaustive review. Dozens of PSG validation papers exist across different devices, populations, and lab settings, and they do not always agree. Robbins et al. and Schyvens et al. tested overlapping devices in the same year and found meaningfully different results for some stages. Before drawing conclusions from any study, a few things are worth checking.

Who was in the study?

Most validation studies recruit healthy adults without sleep disorders. If your sleep is fragmented, you have sleep apnea, or you work shifts, device performance in your case may differ from what the papers report. Population characteristics matter: sample composition can limit how far findings generalise across age groups, fitness levels, or sex. Schyvens et al., for example, enrolled 62 adults but 84% were male, a skew worth keeping in mind when reading those numbers.

How new is the hardware?

Wearable algorithms improve with each product generation, and manufacturers push firmware updates that can change sleep staging behavior without new hardware. Validation studies take time: recruitment, data collection, peer review, and publication can span years. By the time a paper is published, the device it tested may already have a newer model. Always check the device generation listed in the methods section, then compare it to what is currently on your wrist.

Who paid for it?

Industry-funded studies are not automatically unreliable, but they are worth flagging. Check the funding and conflicts of interest section of any paper (found at the end). A study funded by a device manufacturer, or where authors hold equity in the company, does not invalidate the findings, but it is a reasonable prompt to look for a second independent study before accepting the results.

Was it one night in a lab?

All three studies used single-night inpatient designs. Participants slept in an unfamiliar environment, knew they were being monitored, and in PSG studies had electrodes attached to their scalp. A single lab night may not reflect how accurately a device tracks your sleep at home across weeks or months.

What you can take from the data?

The direction of error is consistent across devices and cohorts. Deep sleep is under-detected or over-labelled. Total sleep time runs long. WASO is underestimated. These are structural patterns across studies, not noise.

Lessons Learned: Reading your sleep data with the right expectations

TL;DR: Wearable sleep data is most useful as a relative indicator across nights; the specific numbers, particularly deep sleep duration and time awake, carry more uncertainty than the apps suggest.

The studies above do not say wearable sleep tracking is useless. They say it is imprecise in specific, predictable ways. Knowing which direction the error runs helps you read the numbers more accurately.

• Your total sleep time is probably an overestimate. Read it as an approximation, not a precise count.

• Your time awake during the night is probably an underestimate. If your app shows low time awake but you don’t feel rested, trust your body and prioritise rest.

• Deep sleep duration is the least reliable stage number. It is the metric with the highest variance across devices and the most inconsistency across studies.

• Track trends, not single nights. A device that consistently over- or under-reports by a fixed amount is still useful for detecting relative changes: a poor night versus a good one, baseline versus recovery after travel or illness.

• Do not compare your numbers across devices. A Garmin and an Oura will classify the same night differently. Comparing your deep sleep minutes to someone else’s on a different device is not informative.

If this was worth your time, subscribe for more. Each post explores the science behind wearable health data: how it’s measured, what it means, and what the research says.

Subscribe now

This post is part of The Science Behind Wearables, a series explaining the health metrics your devices track, built around the health scores we’re developing with the Open Wearables team. Open Wearables is an open-source platform for standardised access to health data from consumer wearables, supported and maintained by Momentum.

Know someone obsessed with wearables? Forward this their way.

Share The Science Behind Wearables

Sources:

[1] Berry, R.B., 2014. The AASM manual for the scoring of sleep and associated events: rules, terminology and technical specifications. version 2.1. Darien Illinois: American Academy of Sleep Medicine.

[2] Birrer, V., Elgendi, M., Lambercy, O. and Menon, C., 2024. Evaluating reliability in wearable devices for sleep staging. NPJ Digital Medicine, 7(1), p.74.

[3] Robbins, R., Weaver, M.D., Sullivan, J.P., Quan, S.F., Gilmore, K., Shaw, S., Benz, A., Qadri, S., Barger, L.K., Czeisler, C.A. and Duffy, J.F., 2024. Accuracy of three commercial wearable devices for sleep tracking in healthy adults. Sensors, 24(20), p.6532.

[4] Schyvens, A.M., Peters, B., Van Oost, N.C., Aerts, J.M., Masci, F., Neven, A., Dirix, H., Wets, G., Ross, V. and Verbraecken, J., 2025. A performance validation of six commercial wrist-worn wearable sleep-tracking devices for sleep stage scoring compared to polysomnography. Sleep Advances, 6(2), p.zpaf021.

[5] Lee, T., Cho, Y., Cha, K.S., Jung, J., Cho, J., Kim, H., Kim, D., Hong, J., Lee, D., Keum, M. and Kushida, C.A., 2023. Accuracy of 11 wearable, nearable, and airable consumer sleep trackers: prospective multicenter validation study. JMIR mHealth and uHealth, 11(1), p.e50983.

TL;DR: At rest, Apple Watch heart rate is within roughly 7 beats per minute of a clinical reference, with a near-zero average bias.

The review reports a mean bias of -0.27 bpm when the watch is compared against ECG or chest-strap references in resting conditions [1]. That is effectively indistinguishable from the reference. The catch lives in the limits of agreement: -7.19 to 6.64 bpm. Those limits mean that 95% of your individual readings will fall within that range. A resting reading of 62 bpm on your wrist could reflect anything from 55 to 69 bpm at the source.

For most non-clinical uses, that window is fine. If you are watching your resting heart rate (RHR) trend across weeks, the average smooths out the noise. If you are comparing today’s number to yesterday’s and making a training call, the noise matters more than you think. For many people, the 7-bpm spread is wider than their actual day-to-day resting HR variation.

The other variable is the photoplethysmography (PPG) signal itself. Your Apple Watch shines green light into your skin and measures how much is reflected back with each pulse. Any motion, any loose strap, any change in skin contact pressure changes the signal. At rest, those factors stabilise. In almost any other context, they don’t.

What happens to heart rate accuracy when you move?

TL;DR: Accuracy degrades progressively with exercise intensity, and the error can become large enough to mislead training decisions.

The review shows a clear trend in the included studies: at moderate activity, heart rate bias widens, and at high-intensity exercise, it widens further [1]. The mechanism is straightforward. Motion displaces the sensor, sweat changes skin optics, and blood pressure changes the waveform shape. The algorithm behind the watch does a lot of work to correct for all three. It does not always succeed. This is why validation studies matter more than marketing claims. A number you can trust at rest is not the same as a number you can trust during a tempo run.

There is a real sign of progress in the overall heart rate data. The third-generation optical sensor, used in every Apple Watch from Series 6 onward, narrowed the pooled limits of agreement across all conditions to -3.68 to +2.59 bpm - about half the width of the pooled estimate across a decade of hardware.

Note: If you use your wrist heart rate as a training signal during intervals, cross-check it against a chest strap at least once. The difference tells you whether your wrist reading is usable for your training pattern or whether it is smoothing out precisely the spikes you are trying to see.

Can Apple Watch detect atrial fibrillation?

TL;DR: Apple Watch AFib detection has 91% specificity and 79% sensitivity, which is good enough for screening and not good enough for diagnosis.

The review pooled AFib detection studies into two numbers: specificity at 91%, and sensitivity at 79% [1]. Specificity means that among people without AFib, the watch correctly says “no AFib” 91% of the time; sensitivity means that among people with AFib, the watch correctly catches it 79% of the time. One caveat: 15 to 25% of ECG tracings across several studies were inconclusive, a failure mode the headline numbers do not capture. Sensitivity and specificity both improved substantially when inconclusive tracings were excluded from the analysis,

Those are still screening numbers, not diagnostic ones. If the watch flags an irregular rhythm, the correct next step is a medical evaluation. If the watch says your rhythm is fine, that is reassuring but not definitive - especially for paroxysmal AFib that comes and goes in windows the watch happens to miss.

What do your SpO2, sleep, and step numbers actually tell you?

TL;DR: SpO2 has near-zero average error but a wide confidence band per reading, sleep duration is trustworthy while stage classification is not, and step counts are reliable in daily life but degrade on uneven terrain.

The SpO2 numbers in the review show a mean bias of -0.04% with limits of agreement from -4.01% to 3.94% [1]. Practically: your wrist reading of 94% could correspond to a true value between 90% and 98%. On most days, that spread is irrelevant. For anyone watching their oxygen saturation for clinical reasons, an eight-point error window is the difference between ‘fine’ and ‘call someone’.

Sleep is a similar story. Total sleep duration tracks polysomnography closely enough to be useful as a trend metric. Sleep stage classification shows the largest divergence from clinical staging. If you are reading your nightly breakdown to make lifestyle decisions, the total time is real, the stage percentages are approximations. Sleep apnea detection sits in the same place as AFib: Apple’s own validation reports 98.5% specificity against 66.3% sensitivity - good at ruling it out, less good at catching it, and not a substitute for a proper sleep study if you have symptoms.

Step counts do well on flat, predictable surfaces. They do worse on hills, stairs, and during activities where arm motion and stride don’t track together. For a daily step goal, the watch is honest enough. For a precise physiological measure of ambulation, it is not.

Why does the Apple Watch struggle so much with calories?

TL;DR: Energy expenditure combines three noisy inputs, which compounds error, and mean errors of 20 to 30% during exercise are common.

Calorie estimation is the weakest metric in the review. The included studies show mean errors ‘exceeding 20-30% during exercise’ in multiple cases, with inconsistent directional bias [1]. Some watches overestimate, some underestimate, and the same watch can do both during the same session.

The reason is structural. Energy expenditure on an Apple Watch is computed from heart rate, accelerometer data, and demographic inputs (age, sex, weight, height). Each of these has its own error. Heart rate during exercise is already noisy, as we covered. Accelerometer signals vary with motion type. Demographic inputs rely on what you typed into the Health app, which may or may not reflect your actual metabolism.

When you multiply three noisy signals and pass them through a generalised physiological model, you get a number that works as a relative activity indicator and fails as a precise calorie counter.

What can change Apple Watch accuracy?

TL;DR: The validation studies skewed male and physically active, and skin tone was too inconsistently reported to analyse - leaving known PPG biases unquantified rather than resolved.

Across the 82 included studies, 57% of participants were male (reaching a higher percentage depending on the metric analysed), and the cohorts were dominated by physically active adults [1]. That last bias is not unique to this review but rather it mirrors the early-adopter profile of users themselves. The tolerances in the headline numbers were established on a population that does not represent everyone who wears the device. Skin tone is a separate issue the review could not resolve. The authors were not able to run a subgroup analysis as skin tone was too infrequently reported across the included studies [1]. However, the underlying physics is well established elsewhere. The green light the photoplethysmography (PPG) sensor uses is absorbed more strongly by melanin. Less reflected light means a weaker signal, more noise, and more room for the algorithm to guess wrong. This is not unique to the Apple Watch - all consumer wrist-based PPG devices share it. If you are a woman reading your watch, or if your skin is darker than the average study participant, you are looking at a number validated mostly on people who do not look like you. That does not make the number useless. It means the tolerance around it is wider than the review headline suggests - and how much wider is, for now, an open question.

Lessons Learned: How to read your Apple Watch like the instrument it actually is

TL;DR: Treat every number on your wrist as a signal with a known tolerance, and use trends rather than single readings for anything you actually care about.

The review is not a verdict on whether the Apple Watch works. It works. It works well at some things and poorly at others, and the ratio between ‘well’ and ‘poorly’ is not what the marketing materials suggest. Once you know which metrics have tight tolerances and which have wide ones, your data becomes more useful, not less.

• Use resting heart rate for trend tracking across weeks, not for comparing one morning to the next

• Take AFib alerts seriously and book the cardiology appointment, and do not interpret the absence of an alert as proof of cardiac health

• Read SpO2 as a rough range, not a precise figure, unless you are following a specific clinical instruction

• Trust total sleep duration, treat sleep stage percentages as indicative only

• Ignore absolute calorie numbers and use the ‘active minutes’ or ‘exercise ring’ framing, which handles activity intensity more honestly than the calorie figure

• Know your own biases: if you are in a demographic under-represented in validation studies, assume the tolerance band is wider than published averages suggest

The value of the watch is that it shows you something you had no access to five years ago. The risk of the watch is that it shows it to you with a certainty the measurement does not actually have.

If this was worth your time, subscribe for more. Each post explores the science behind wearable health data: how it’s measured, what it means, and what the research says.

Subscribe now

This post is part of The Science Behind Wearables, a series explaining the health metrics your devices track, built around the health scores we’re developing with the Open Wearables team. Open Wearables is an open-source platform for standardised access to health data from consumer wearables, supported and maintained by Momentum.

Know someone obsessed with wearables? Forward this their way.

Share The Science Behind Wearables

Sources:

[1] Lambe, R., Baldwin, M., O’Grady, B., Schumann, M., Caulfield, B. and Doherty, C. (2026). The accuracy of Apple Watch measurements: a living systematic review and meta-analysis. npj Digital Medicine, [online] 9(63). doi:https://doi.org/10.1038/s41746-025-02238-1.

TL;DR: Far from being a state of “nothingness,” sleep is a highly active, neurobiological process of “cellular housekeeping” that restores the brain and the body.

I used to be the person who dreamt of functioning perfectly on five or six hours of sleep. I saw sleep as a tax I had to pay on my productivity. Years of neuroscience training changed that; I am now someone who protects an eight-hour window, knowing it is the single most effective thing I can do for my health.

We spend about one-third of our lives asleep, yet we are only beginning to unlock its complexity. For a long time, we viewed sleep as a passive state - a “light switch” that simply turned the brain off. In reality, sleep is a masterpiece of biological engineering. While you drift off, your brain flushes out metabolic waste through the glymphatic system, repairs damaged tissues, and processes memories and emotional data. From a scientific point of view, sleep is defined as a bio-behavioral state observable through changes in brain electrical activity, altered consciousness, reduced sensory responsiveness, and decreased muscle tone [1].

What does your brain do when you sleep?

TL;DR: Sleep is orchestrated by a specialized network of “switches” in the hypothalamus and brainstem, coordinated by a master clock that syncs our biology with the sun.

For neuroscience aficionados like myself, understanding sleep requires a look at the “hardware” behind the behavior. This hardware is your brain’s anatomy and it determines how you sleep.

The Hypothalamus: The Master Clock

The hypothalamus is a peanut-sized structure that acts as the brain’s primary command center for homeostasis. Within this structure sits a tiny cluster of neurons called the Suprachiasmatic Nucleus (SCN). Think of the SCN as your body’s master clock. It receives light information directly from the retina in your eyes. When the sun goes down and light fades, the SCN signals the rest of the brain to shift from “active mode” to “recovery mode.” It is the SCN that dictates your circadian rhythm - the 24-hour cycle that tells you when it’s time to sleep.

The Pineal Gland: The Starting Pistol

While the SCN is the clock, the Pineal Gland is the messenger. Once it receives the signal cascading from the SCN, the pineal gland begins producing melatonin. It’s a common misconception that melatonin is a sedative that knocks you out. In reality, it acts more like a biological starting pistol, signaling to the rest of your systems that the race for sleep has officially begun [2].

The Brainstem: Quieting the Brain

In our last article, we talked about the Autonomic Nervous System and its two branches: the Accelerator (Sympathetic) and the Brake (Parasympathetic). As you prepare for sleep, the brainstem withdraws the ‘Accelerator’ signals - the wake-promoting neurotransmitters that kept you alert and responsive. In parallel, the hypothalamus releases GABA, the brain’s primary inhibitory neurotransmitter, actively quieting the brain’s arousal centres. Together, this essentially ‘muffles’ sensory input from the outside world, allowing you to drift off. During REM sleep, the brainstem sends a temporary “disconnect” signal to your muscles, ensuring you don’t physically act out your dreams - a vital safety feature of our biological architecture.

What does a night of sleep look like?

TL;DR: Sleep is not a uniform state but a series of 90-minute cycles composed of REM and non-REM stages, each serving a distinct purpose for physical and mental recovery.

After learning how your brain regulates sleep, let’s talk about what sleep is composed of. When you look at your wearable data, you’ll see your night broken down into a “hypnogram” - a map of your sleep stages. These stages are defined by specific patterns of brain wave activity and are broadly divided into REM (Rapid Eye Movement) and Non-REM (NREM) sleep.

Fig. What a typical night of sleep might look like.

Think of a night of sleep like a healthy meal: you need every micronutrient, to feel “full” the next morning.

Non-REM Stage 1 and 2: Light sleep (~50% of total sleep)

• Stage 1: This is the “dozing off” phase. Your brain waves begin to slow, but you’re still easily awoken.

• Stage 2: This makes up the bulk of your night. Your heart rate drops, and your brain produces “sleep spindles” - short bursts of activity that are thought to be essential for memory consolidation.

Non-REM Stage 3: Deep Sleep (13-23% of total sleep)

Often called “Slow Wave Sleep,” this is the most restorative stage for the body. During this phase, your brain produces large, rhythmic waves, blood flow is diverted from the brain to the muscles to repair tissue, the immune system is boosted, and growth hormones are released. If you wake up feeling physically “heavy” or groggy, you were likely pulled out of this deep state.

REM Sleep (20-25% of total sleep)

While NREM is for the body, REM is for the mind. During REM, your brain activity looks remarkably similar to when you are awake. Your eyes move rapidly behind your lids, and your body enters a state of temporary paralysis to prevent you from acting out your dreams. This is where we process emotions and integrate complex memories. A lack of REM sleep often leaves us feeling “foggy” on the next day.

One thing worth knowing is that these stages are not evenly distributed across the night. Deep sleep dominates the first half - your brain prioritises physical restoration early on. REM, on the other hand, becomes increasingly concentrated in the second half, particularly in the final one to two hours before you wake. This is why sleep timing matters as much as sleep duration. An early alarm doesn’t just shorten your night - it disproportionately cuts your REM.

Changes in physiological variables during different sleep stages.

How much should you sleep?

TL;DR: While the expert consensus for healthy adults is 7–9 hours, the ideal duration is a dynamic target that shifts across the lifespan.

In 2015, the National Sleep Foundation (NSF) conducted a comprehensive review to establish guidelines for sleep duration [3]. The most significant finding is that our sleep requirements are not “one size fits all” - they shift significantly as our brain and body age. While newborns require nearly double the sleep of an adult to support rapid development, the biological requirement for adults stabilizes between 7 and 8-9 hours.

For the majority of adults, 7 hours is the biological “floor.” When we consistently get less than this, we begin to see a measurable impact on our health data. Specifically, short sleep duration is a primary driver of sympathetic dominance - meaning if you don’t hit these duration targets, the “Accelerator” stays active, your heart rate remains elevated, and your HRV baseline will likely drop over time.

What defines sleep quality?

TL;DR: Sleep quality is measured by “continuity” - how efficiently you stay asleep and how quickly you return to sleep if interrupted.

The duration of sleep, however, doesn’t tell you the whole story. In 2017, an expert panel from the NSF performed a systematic review of 277 studies combined with voting to define exactly what “good sleep quality” looks like across life-span [4]. They concluded that quality is best indicated by sleep continuity variables. These are the metrics that determine whether your sleep is consolidated or fragmented. According to the NSF panel, for a healthy adult, “good quality” sleep typically meets the following criteria:

• Sleep Latency: You should be falling asleep within 30 minutes of getting into bed. If you’re lying there for an hour running through tomorrow’s to-do list, that’s a signal worth paying attention to.

• Awakening: Waking once in the night is normal. Waking two, three, or four times, for more than 5min, is a sign of fragmented sleep that adds up over time.

• Waking After Sleep Onset (WASO): If you do wake up after falling asleep, staying awake for more than 20 minutes overall will also affect your sleep.

• Sleep Efficiency: This is the ratio of time actually asleep versus time spent in bed. The target is 85% or above. If you’re in bed for eight hours but only sleeping six, your efficiency is 75% - and your body knows the difference, even if you don’t.

How does your wearable track sleep?

TL;DR: While clinical sleep studies measure brain activity directly, wearables use movement and heart rate as “proxies” to estimate your sleep stages.

It is important to understand that how sleep is measured differs significantly between clinical laboratory conditions and the wearable world. Distinguishing between these two methods helps explain why your watch might miss a wake-up event or miscalculate a specific stage.

The Gold Standard: Polysomnography (PSG)

In clinical settings, sleep is measured using Polysomnography. If you were to participate in a sleep study, you would be hooked up to a variety of sensors that track:

• Brain Waves (EEG): The only direct way to see which sleep stage the brain is in.

• Eye Movements (EOG): Essential for identifying the “Rapid Eye Movement” in REM sleep.

• Muscle Tone (EMG): Used to detect the physical paralysis that occurs during REM.

• Heart Rate and Respiration: To monitor autonomic health and breathing disorders.

The Wearable Approach: Actigraphy and Photoplethysmography (PPG)

Because we cannot easily measure brain waves at home, wearables rely on two primary sensors to infer when and how we sleep:

• Actigraphy (Accelerometers): This measures your movement. The logic is simple: if you are moving, you are likely awake; if you are perfectly still, you are likely asleep.

• Photoplethysmography (PPG): It measures your heart rate, HRV and respiratory rate.

Accuracy and Validation

Researchers use validation studies to compare wearable data against Polysomnography. These studies show that while wearables are good at tracking total sleep duration and “wake vs. sleep” transitions, they are less precise at distinguishing between specific stages like light vs. REM. I will discuss wearable accuracy in sleep in my next article, stay tuned!

Why is sleep so important?

TL;DR: Sleep disturbances are clinical predictors for cardiovascular diseases, metabolic dysfunction, mental health disorders and long-term cognitive decline.

When we think of sleep, we often associate it with productivity and performance. In reality, it is much more than this. Sleep disturbances are linked to poor cardiovascular health, diabetes, mental health, and even dementia.

• Cardiovascular Health: A meta-analysis of over 400,000 participants found that both short and long sleep durations were associated with an increased risk of stroke and coronary heart disease [5]. The evidence is so compelling that in 2022, the American Heart Association officially added sleep duration to its “Life’s Essential 8”- the checklist of the most important predictors of cardiovascular health - placing it alongside blood pressure, cholesterol, and blood sugar [6].

• Metabolic Regulation: A 2015 meta-analysis described a “U-shaped” risk profile for diabetes: the risk is lowest at 7–8 hours, while both shorter and longer sleep durations significantly increase the likelihood of developing Type 2 Diabetes [7].

• Dementia: A study published in Nature Communications found that consistent sleep of 6 hours or less in middle age is associated with a 30% increase in dementia risk, independent of sociodemographic, cardiometabolic, or mental health factors [8].

• Mental Health: I recently attended back-to-back lectures by Prof. Terrie Moffitt and Prof. Avshalom Caspi, two of the world’s leading researchers in longevity and human development. What struck me most was their discussion of a recent publication in Nature Medicine where researchers identified sleep disturbances in adolescents as a robust predictor of psychiatric illness - surpassing even adverse childhood experiences and family mental health history as a predictive tool [9].

These are just a few examples, and for many of them, we are still uncovering the exact biological mechanisms. However, they demonstrate how powerful sleep is for healthy development, living, and aging.

Lessons learned: Healthy sleep is the best health outcome

Although we are still uncovering the exact biological mechanisms behind all these results, one thing is clear: sleep emerges as a promising therapeutic and preventative approach for health across different domains. This is fundamentally good news. Why? Because for many people, sleep is a target we can actively support. Sleep therapies, such as Cognitive Behavioral Therapy for Insomnia (CBT-I), already exist and have been proven effective in helping people navigate and improve their sleep patterns [10]. This is particularly exciting for the field of mental health, where traditional therapies often fail.

I strongly believe that wearables will play a vital role in this shift. They give us a window into our patterns over a longer period of time - something that was never achievable on this scale before. This doesn’t mean you should obsess over a single sleep score recorded by your watch or ring. It means you can observe how your sleep looks over weeks and months, adjust your environment where possible, and, most importantly, have the longitudinal data to seek professional help if you see persistent disturbances. Your wearable isn’t just a tracker; it’s a tool for advocacy and early intervention.

If this was worth your time, subscribe for more — each post explores the science behind wearable health data: how it’s measured, what it means, and what the research says.

Subscribe now

This post is part of The Science Behind Wearables - a series explaining the health metrics your devices track, built around the health scores we’re developing with the Open Wearables team. Open Wearables is an open-source platform for standardized access to health data from consumer wearables, supported and maintained by Momentum.

Know someone obsessed with wearables? Forward this their way.

Share The Science Behind Wearables

Sources:

[1] Luyster, F.S., Strollo, P.J., Zee, P.C. and Walsh, J.K. (2012). Sleep: A Health Imperative. Sleep, 35(6), pp.727–734. doi:https://doi.org/10.5665/sleep.1846.

[2] Arendt, J. and Skene, D.J. (2005). Melatonin as a chronobiotic. Sleep Medicine Reviews, 9(1), pp.25–39. doi:https://doi.org/10.1016/j.smrv.2004.05.002.

[3] Hirshkowitz, M., Whiton, K., Albert, S.M., Alessi, C., Bruni, O., DonCarlos, L., Hazen, N., Herman, J., Adams Hillard, P.J., Katz, E.S., Kheirandish-Gozal, L., Neubauer, D.N., O’Donnell, A.E., Ohayon, M., Peever, J., Rawding, R., Sachdeva, R.C., Setters, B., Vitiello, M.V. and Ware, J.C. (2015). National Sleep Foundation’s updated sleep duration recommendations: final report. Sleep Health, 1(4), pp.233–243. doi:https://doi.org/10.1016/j.sleh.2015.10.004

[4] Ohayon, M., Wickwire, E.M., Hirshkowitz, M., Albert, S.M., Avidan, A., Daly, F.J., Dauvilliers, Y., Ferri, R., Fung, C., Gozal, D., Hazen, N., Krystal, A., Lichstein, K., Mallampalli, M., Plazzi, G., Rawding, R., Scheer, F.A., Somers, V. and Vitiello, M.V. (2017). National Sleep Foundation’s sleep quality recommendations: first report. Sleep Health, 3(1), pp.6–19. doi:https://doi.org/10.1016/j.sleh.2016.11.006.

[5] Cappuccio, F.P., Cooper, D., D’Elia, L., Strazzullo, P. and Miller, M.A. (2011). Sleep duration predicts cardiovascular outcomes: a systematic review and meta-analysis of prospective studies. European Heart Journal, 32(12), pp.1484–1492. doi:https://doi.org/10.1093/eurheartj/ehr007.

[6] Lloyd-Jones, D.M., Allen, N.B., Anderson, C.A.M., Black, T., Brewer, L.C., Foraker, R.E., Grandner, M.A., Lavretsky, H., Perak, A.M., Sharma, G. and Rosamond, W. (2022). Life’s Essential 8: Updating and Enhancing the American Heart Association’s Construct of Cardiovascular Health: a Presidential Advisory from the American Heart Association. Circulation, 146(5). doi:https://doi.org/10.1161/cir.0000000000001078.

[7] Shan, Z., Ma, H., Xie, M., Yan, P., Guo, Y., Bao, W., Rong, Y., Jackson, C.L., Hu, F.B. and Liu, L. (2015). Sleep Duration and Risk of Type 2 Diabetes: A Meta-analysis of Prospective Studies. Diabetes Care, 38(3), pp.529–537. doi:https://doi.org/10.2337/dc14-2073.

[8] Sabia, S., Fayosse, A., Dumurgier, J., van Hees, V.T., Paquet, C., Sommerlad, A., Kivimäki, M., Dugravot, A. and Singh-Manoux, A. (2021). Association of sleep duration in middle and old age with incidence of dementia. Nature Communications, 12(1), p.2289. doi:https://doi.org/10.1038/s41467-021-22354-2.

[9] Hill, E.D., Kashyap, P., Raffanello, E., Wang, Y., Moffitt, T.E., Caspi, A., Engelhard, M. and Posner, J. (2025). Prediction of mental health risk in adolescents. Nature Medicine, 31, pp.1–7. doi:https://doi.org/10.1038/s41591-025-03560-7.

[10] Mei, Z., Cai, C., Luo, S., Zhang, Y., Lam, C. and Luo, S. (2024). The efficacy of cognitive behavioral therapy for insomnia in adolescents: a systematic review and meta-analysis of randomized controlled trials. Frontiers in Public Health, 12. doi:https://doi.org/10.3389/fpubh.2024.1413694.

TL;DR: Wearables use light, not electricity, to detect your heartbeat, which has real consequences for how accurately they measure HRV.

The wearable on your wrist doesn’t read your heart the way a hospital ECG does. Instead of capturing the electrical signal your heart generates, it uses a method called photoplethysmography (PPG): a green LED shines into your skin, and a photodetector measures how much of that light bounces back. Blood absorbs light. As your heart pumps, the volume of blood flowing through your skin’s capillaries rises and falls with each beat, creating tiny, detectable changes in that reflected signal. The device infers the timing of your heartbeats, and the variation between them, from those optical changes.

This distinction matters more than it might initially seem. ECG captures the electrical moment of cardiac depolarization directly. PPG captures the mechanical consequence of that event, filtered through your skin, capillaries, and the contact your device is making with your wrist. For most health metrics, this approximation is close enough. For HRV measurement at the millisecond level, the gap between a direct electrical measurement and an optical proxy becomes meaningful.

What is SDNN, and how does it differ from RMSSD?

TL;DR: SDNN and RMSSD are both HRV metrics, but they measure different things, and most consumer wearables report RMSSD, not SDNN.

When a wearable has a stream of time intervals between heartbeats (called NN intervals), it needs to summarize that information into a single number. The two most common ways to do this are:

• RMSSD (Root Mean Square of Successive Differences): captures the variation between consecutive heartbeat pairs. It’s primarily sensitive to parasympathetic activity, the recovery and relaxation branch of your nervous system, and works well in short recording windows. Most consumer wearables report RMSSD for this reason [3].

• SDNN (Standard Deviation of NN Intervals): captures the overall spread of all NN intervals across a recording. This includes both short-term beat-to-beat variation and slower patterns tied to breathing, circadian rhythms, and systemic regulatory processes.

Because the two metrics use different math and are designed for different recording lengths, they are not interchangeable. Clinical guidelines most commonly reference SDNN from 24-hour Holter recordings, where values below 50ms are suggested to be associated with elevated cardiovascular risk [4]. Your wearable’s 5-minute RMSSD reading cannot be benchmarked against that threshold. The comparison is a category error, not just a rough approximation.

Note: For the same reason, never compare your HRV score across different apps or devices. The value is in the trend within a single ecosystem, not the raw number.

Why does your reading change when nothing else has?

TL;DR: Contact pressure between the wrist sensor and your skin is one of the least-discussed variables in wrist HRV accuracy, and small shifts away from optimal can multiply error by 12x.

If you’ve noticed your HRV drop on a night where nothing obvious changed, the measurement itself may be part of the explanation.

A study published in Nature Scientific Data tested 27 participants using custom equipment designed to precisely control wrist contact pressure, while simultaneously recording ECG and fingertip PPG as reference signals [1]. The researchers found that wrist PPG sensors produce one of five distinct waveform types depending on that pressure:

• Type 1 (insufficient pressure): A single peak. The sensor lacks enough contact to resolve the cardiac pulse cleanly. HRV measurement is unreliable.

• Type 2E (slightly suboptimal): Two peaks of similar amplitude, with the systolic peak not yet clearly dominant. Accuracy is reduced.

• Type 2L (optimal): Two peaks with a clearly dominant systolic component and a visible dicrotic notch. This is the waveform that produces reliable readings.

• Type 1L (excessive pressure): Over-compression flattens the diastolic component. Accuracy degrades in the opposite direction.

• Type 3 (poor contact): Low signal-to-noise ratio. Unusable.

At optimal pressure, wrist PPG showed a heart rate error of 0.37 bpm and an HRV error of 0.89ms. At suboptimal pressure, HRV error rose to 10.95ms. A 12-fold increase from a single variable. Same sensor, same wrist, same algorithm, just a different contact state.

The transition from Type 1 through Type 2L to Type 3 follows a predictable pattern as pressure increases. The exact pressure thresholds vary between individuals, meaning there is no universal “correct” tightness for a wrist-worn device.

Are fingertip sensors more accurate?

TL;DR: Yes, substantially, and the advantage comes from anatomy rather than design.

The same study used fingertip PPG as a reference condition. Fingertip sensors showed a heart rate error of 0.08 bpm and an RMSSD error of 0.15ms, compared to 0.89ms on the wrist at optimal conditions. When wrist contact was suboptimal, the gap grew wider. Fingertip accuracy remained stable throughout. The full dataset is publicly available for those building or validating PPG-based algorithms [2].

Two anatomical factors explain this. Capillary density in the fingertip is substantially higher than at the wrist, giving the sensor a stronger, cleaner optical signal. And the geometry of pressing a fingertip against a sensor creates inherently more stable contact than a wristband worn through hours of sleep. You place a finger deliberately and hold it; a strap shifts and loosens throughout the night without you noticing.

Fingertip PPG devices appear in clinical pulse oximeters and research-grade monitors. The reason wrist wearables dominate is convenience, not capability. For continuous daily tracking, wrist sensors are the practical choice. For situations where HRV precision matters, diagnostics, clinical monitoring, research data collection, wrist PPG has a genuine accuracy ceiling that a better sensor placement can bypasses.

Lessons Learned: How To Interpret Data You Know Is Noisy

TL;DR: Knowing where the measurement error comes from lets you work with wearable HRV data more intelligently, rather than either over-trusting or dismissing it.

A 12-fold error increase at suboptimal contact doesn’t make your wearable’s readings useless. It means they contain physiological signal and measurement noise together, and the two aren’t always easy to separate. On a stable night with a well-fitted strap and normal skin temperature, your reading is a reasonable reflection of your autonomic state. On a night with movement, a loosened strap, or elevated body temperature, the noise contribution grows.

A few things worth keeping in mind when you work with your data:

• Use rolling windows of at least two weeks. Individual readings carry too much environmental noise to be reliable on their own. Trends wash it out.

• Know which metric your device reports, RMSSD or SDNN, before applying any benchmark you’ve read in research. The recording length matters just as much as the metric name.

• Strap fit is a real variable. Consistent placement from night to night removes one source of noise you can actually control.

• If your device offers signal quality or confidence indicators, treat low-confidence readings with more scepticism. They encode information about recording reliability that the headline number doesn’t.

If this was worth your time, subscribe for more — each post explores the science behind wearable health data: how it’s measured, what it means, and what the research says.

Subscribe now

This post is part of The Science Behind Wearables — a series explaining the health metrics your devices track, built around the health scores we’re developing with the Open Wearables team. Open Wearables is an open-source platform for standardized access to health data from consumer wearables, supported and maintained by Momentum.

Know someone obsessed with wearables? Forward this their way.

Share The Science Behind Wearables

Sources

[1] Ho, M.Y., Pham, H.M., Saeed, A. and Ma, D. (2025). WF-PPG: A Wrist-finger Dual-Channel Dataset for Studying the Impact of Contact Pressure on PPG Morphology. Scientific Data, 12(1). doi:https://doi.org/10.1038/s41597-025-04453-7.

[2] Ho, M., Saeed, A., MA, D., & PHAM, M. (2025). WF-PPG: A Wrist-finger Dual-Channel Dataset for Studying the Impact of Contact Pressure on PPG Morphology (Version 1). figshare. https://doi.org/10.6084/m9.figshare.27011998

[3] Shaffer, F. and Ginsberg, J.P. (2017). An Overview of Heart Rate Variability Metrics and Norms. Frontiers in Public Health, [online] 5(258). doi:https://doi.org/10.3389/fpubh.2017.00258.

[4] Electrophysiology, T.F. of the E.S. (1996). Heart Rate Variability. Circulation, [online] 93(5), pp.1043–1065. doi:https://doi.org/10.1161/01.cir.93.5.1043.

TL;DR: Heart Rate Variability, or HRV for short, measures the variation of individual beats of your heart.

If you were to listen to your heart right now, you would probably expect to hear a steady, rhythmic thump-thump, thump-thump. We have been taught that “steady” means “healthy.” In the health science world, we know the opposite is true. In the same way a mountain bike’s suspension system adjusts to every rock and root on the trail, a healthy heart is constantly micro-adjusting. It speeds up slightly when you inhale and slows down when you exhale. This subtle “chaos” in the timings between heart beats is the signature of a body that is ready for challenges. Heart Rate Variability, or HRV for short, measures the variation of individual beats of your heart.

What shapes your HRV?

TL;DR: HRV is shaped by your autonomic nervous system, with parasympathetic branch slowing down the heart and increasing the HRV and sympathetic speeding up the heart and thus decreasing HRV.

At the core of every heart beat is a tiny cluster of cells in your heart called the Sinoatrial (SA) node. It is considered your heart’s internal pacemaker and produces a steady, rhythmic rate of about 60 to 100 beats per minute. But your heart doesn’t operate in a vacuum. It’s constantly taking orders from your Autonomic Nervous System (ANS) with its sympathetic and parasympathetic branches. The variability we measure is the result of a constant, silent conversation between the SA node and these two branches. To visualize how they interact, think of them as the Accelerator and the Brake:

• The Parasympathetic Branch (The Brake): Led by the vagus nerve, this system tells the SA node to slow down and relax. Naturally, when the beat is slower, there is more room for variations and that is when you would see higher HRV.

• The Sympathetic Branch (The Accelerator): When you’re under stress-physical or emotional-this branch kicks in, telling the SA node to floor it. As the heart rate increases, the sympathetic branch takes control, supporting a perfectly rhythmic pulse.

Because this conversation between your sympathetic and parasympathetic branches is constant, we don’t look at HRV as a single ‘snapshot’. We look at the trend-the story of how the branches are balancing each other out over days, weeks and months.

How can HRV be calculated?

TL;DR: HRV is calculated by measuring the millisecond-level variations between consecutive heartbeats, which are then processed into metrics like RMSSD or SDNN.

So how do we calculate HRV? When you look at your ECG (electrocardiogram) trace, you see distinct “spikes.” These are called R-peaks, and they represent your heart’s contractions. The time between these spikes is the R-R interval, measured in milliseconds (ms).

Fig. HRV visualised as the variation in duration between successive R-R intervals.

In order to assign meaning to these numbers, we can use different metrics to represent HRV. The two most common ones are RMSSD (root mean square of successive differences between heartbeats) and SDNN (standard deviation of normal-to-normal intervals) [1]:

Note: Because these metrics use different math and are often recorded at different times, never compare your HRV score from one app to another. The value is in the trend within a single ecosystem, not the raw number itself.

What influences your HRV?

TL;DR: While factors like age and genetics set your biological baseline, your HRV is primarily a reflection of your daily choices-driven by a balance of consistent exercise, quality sleep and stress management.

After learning how the nervous system regulates your heart, it might seem like you are merely a passenger in the process. However, this is far from the truth. While certain elements of your HRV are “hardwired” into your biology, many others act as “dials” you can adjust daily. Distinguishing between the two is essential for interpreting your data without discouragement. The factors beyond our control include genetics, sex, and age, as well as cardiovascular, respiratory, or psychiatric conditions [2-7].

Fortunately, there are numerous levers you can pull to actively improve your autonomic resilience. The most effective approach is regular, moderate and high intensity exercise [8]. When monitoring your values, it is important to recognize the “acute-versus-chronic” relationship: while a strenuous workout acts as a temporary stressor that drops your HRV in the short term, consistent cardiovascular training is what raises your baseline over the months. Beyond physical exercise, proactive stress management through techniques like breathwork or meditation can directly reduce sympathetic activity, thus increasing your HRV [9]. The quality and consistency of your sleep serve as the primary engine for recovery and long-term HRV improvement [10]. Finally, your immediate daily choices play a large role in your autonomic health [11-12]. In wearable data, alcohol consumption and late-night meals are the most frequent “HRV stressors,” as they force the body to remain in an active, metabolic state when it should be transitioning into deep recovery. By understanding these influences, you move from simply observing your numbers to actively managing your physiological state.

Why is HRV such an important variable?

TL;DR: Beyond simple fitness tracking, HRV is a vital sign of biological resilience and systemic health; a higher, more adaptable HRV is a key indicator of longevity and your body’s ability to maintain homeostasis.

Knowing what shapes your HRV is one thing, but asking why you should care is the more important question. Most people encounter HRV as a tool for athletes to measure fitness or recovery, but that’s not the end of the story. In reality, HRV is a powerful indicator of your overall biological resilience. It isn’t just a snapshot of yesterday’s workout; it is a “dashboard light” for your health that often flickers long before you actually feel symptoms of stress or illness. This is because your HRV provides a direct window into the state of your nervous system. If you spend too much time in a “fight or flight” state, that chronic sympathetic dominance is reflected in a lower HRV, which over time is a known risk factor for cardiovascular disease. Conversely, actively strengthening your parasympathetic response builds the resilience needed to handle life’s challenges without taking a permanent toll on your health.

Perhaps most importantly, HRV is a recognized marker of longevity. Because it measures how well your autonomic nervous system can adapt to challenges, a higher age-adjusted HRV is consistently associated with a longer, healthier life as well as a lower risk of chronic metabolic and cardiovascular decline . By supporting your body’s ability to remain adaptable, you aren’t just optimizing for the next 24 hours-you are preserving a physiological resilience that is fundamental to a healthier lifespan.

Lessons Learned: Finding Your Personal Balance

TL;DR: The goal of tracking HRV isn’t to achieve a specific “high score,” but to understand your personal trends; because your baseline is unique to your biology, the real value lies in monitoring how your autonomic balance shifts in response to your specific environment and lifestyle.

When interpreting your own HRV, the most important takeaway is that balance is the key. While the sympathetic and parasympathetic branches have opposite roles, they are not competitors; they are complementary systems that work in tandem to maintain your internal homeostasis. As we have learned, HRV is a highly sensitive metric influenced by a complex set of variables-ranging from genetics and age to hormonal cycles, psychological state, and lifestyle choices. This sensitivity is precisely why population-wide averages are rarely useful for the individual. Instead of comparing your values to a general population, the true value of this data lies in understanding your own trends. By tracking your personal baseline over time, you move away from chasing a specific score and toward a deeper understanding of how your body responds to the world around you.

This post is part of The Science Behind Wearables - a series explaining the health metrics your devices track, built around the health scores we're developing with the Open Wearables team. Open Wearables is an open-source platform for standardized access to health data from consumer wearables, supported and maintained by Momentum.

Sources:

[1] Shaffer, F., & Ginsberg, J. P. (2017). An Overview of Heart Rate Variability Metrics and Norms. Frontiers in Public Health, 5(258). https://doi.org/10.3389/fpubh.2017.00258

[2] Golosheykin, S., Grant, J. D., Novak, O. V., Heath, A. C., & Anokhin, A. P. (2017). Genetic influences on heart rate variability. International Journal of Psychophysiology, 115, 65–73. https://doi.org/10.1016/j.ijpsycho.2016.04.008

[3] De Meersman, R. E., & Stein, P. K. (2007). Vagal modulation and aging. Biological Psychology, 74(2), 165–173. https://doi.org/10.1016/j.biopsycho.2006.04.008

[4] Koenig, J. and Thayer, J.F. (2016). Sex differences in healthy human heart rate variability: A meta-analysis. Neuroscience & Biobehavioral Reviews, 64, pp.288–310. doi:https://doi.org/10.1016/j.neubiorev.2016.03.007.

[5] Wang, B.X., Brennand, E., Le Page, P. and Mitchell, A.R.J. (2026). Heart rate variability in cardiovascular disease diagnosis, prognosis and management. Frontiers in Cardiovascular Medicine, 12. doi:https://doi.org/10.3389/fcvm.2025.1680783.

[6] Alqahtani, J.S., Aldhahir, A.M., Alghamdi, S.M., Ghamdi, A., AlDraiwiesh, I.A., Alsulayyim, A.S., Alqahtani, A.S., Alobaidi, N.Y., Lamia Al Saikhan, AlRabeeah, S.M., Alzahrani, E.M., Heubel, A.D., Mendes, R.G., Alqarni, A.A., Alanazi, A.M. and Tope Oyelade (2023). A systematic review and meta-analysis of heart rate variability in COPD. Frontiers in Cardiovascular Medicine, 10. doi:https://doi.org/10.3389/fcvm.2023.1070327.

[7] Wang, Z., Zou, Y., Liu, J., Peng, W., Li, M. and Zou, Z. (2025). Heart rate variability in mental disorders: an umbrella review of meta-analyses. Translational psychiatry, [online] 15(1), p.104. doi:https://doi.org/10.1038/s41398-025-03339-x.

[8] Grässler, B., Thielmann, B., Böckelmann, I. and Hökelmann, A. (2021). Effects of Different Training Interventions on Heart Rate Variability and Cardiovascular Health and Risk Factors in Young and Middle-Aged Adults: A Systematic Review. Frontiers in Physiology, 12. doi:https://doi.org/10.3389/fphys.2021.657274.

[9] Natarajan, A. (2023). Heart rate variability during mindful breathing meditation. Frontiers in Physiology, 13. doi:https://doi.org/10.3389/fphys.2022.1017350.

[10] Zhang, S., Niu, X., Ma, J., Wei, X., Zhang, J. and Du, W. (2025). Effects of sleep deprivation on heart rate variability: a systematic review and meta-analysis. Frontiers in neurology, [online] 16, p.1556784. doi:https://doi.org/10.3389/fneur.2025.1556784.

[11] Santa-Rosa, F.A., Shimojo, G.L., Dias, D.S., Viana, A., Lanza, F.C., Irigoyen, M.C. and De Angelis, K. (2020). Impact of an active lifestyle on heart rate variability and oxidative stress markers in offspring of hypertensives. Scientific Reports, [online] 10(1), p.12439. doi:https://doi.org/10.1038/s41598-020-69104-w.

[12] Narjisse Damoun, Youssra Amekran, Taiek, N. and Abdelkader (2024). Heart rate variability measurement and influencing factors: Towards the standardization of methodology. Global Cardiology Science and Practice, [online] 2024(4). doi:https://doi.org/10.21542/gcsp.2024.35.