Biohacking Kompakt

Tip · Tracking

HRV & sleep stages with a ring/band

Rings and bands detect sleep and wakefulness well, but the individual sleep stages unreliably – and those are exactly what the advertising features.

Rings and bands measure movement and pulse during sleep and use these to estimate sleep stages, heart rate variability and a daily score. The tip is here because this class of device has been tested against the gold standard, polysomnography — and because the test results hit precisely the functions that feature in the advertising.

In short

The devices are very good at detecting whether you are asleep. Across seven consumer devices, epoch-by-epoch sensitivity was ≥ 0.93, while specificity for wakefulness was only 0.18 to 0.54. In a laboratory study with six current devices, all detected over 90 % of sleep epochs, with specificities between 29.39 % and 52.15 % and Cohen’s kappa between 0.21 and 0.53. Stage classification is the weakest part: pooled accuracy for finger-worn devices 0.65 for light sleep, 0.81 for deep sleep, 0.74 for REM. Readiness and recovery scores are proprietary computed values without published validation.

What is behind it

The gold standard for sleep is polysomnography: EEG, eye movements and muscle tone define what light sleep, deep sleep and REM are. Rings and bands have none of these. They measure movement via an accelerometer and pulse via photoplethysmography, and from these derive heart rate, heart rate variability, respiratory rate and in some cases skin temperature.

Estimating stages is possible at all only because the autonomic nervous system changes along with sleep: in deep sleep the parasympathetic system dominates, while in REM sleep heart rate and its variability rise again. The systematic error follows from the method: anyone lying still with a slow pulse is classified as asleep, even when awake.

What testing against polysomnography found

In the cleanest openly accessible comparison study, 34 healthy young adults spent three consecutive nights in the sleep laboratory, including deliberately one night with fragmented sleep, with polysomnography, a research actigraph and seven consumer devices in parallel. Sensitivity was consistently ≥ 0.93, specificity 0.18 to 0.54, and all devices performed worse in the disrupted night.

A more recent laboratory study with 62 adults and six current devices confirms the pattern: over 90 % of sleep epochs detected, specificity 29.39 % to 52.15 %, Cohen’s kappa 0.21 to 0.53 — the best values for the Apple Watch Series 8 (0.53), Fitbit Sense (0.42) and Fitbit Charge 5 (0.41). Usable for sustained and marked changes, not for a single night.

How large the deviations are in everyday life

A meta-analysis of 24 studies with 798 people quantifies the offset: compared with polysomnography, wrist-worn devices on average report 16.854 minutes less total sleep time, 4.691 percentage points lower sleep efficiency, 2.574 minutes longer sleep onset latency and 13.255 minutes more wake time after sleep onset.

The error is also proportional. In a laboratory study with 53 young adults and five devices, all devices overestimated short wake periods and underestimated long ones. In practice this means: precisely in the nights you want to know something about — the bad ones — the error is greatest.

What is well supported

Validation against polysomnography is the strongest part of this page. Sleep–wake separation: sensitivity ≥ 0.93 across seven devices, pooled accuracy 87 % (95 % CI 86–89 %) across 11 finger-worn devices from 28 papers. Wake detection: specificity 0.18 to 0.54 or 29.39 % to 52.15 %. Stages: 0.65 light sleep, 0.81 deep sleep, 0.74 REM for finger-worn devices, Cohen’s kappa 0.21 to 0.53 for wrist-worn devices. Systematic offset across 24 studies and 798 people: −16.854 minutes total sleep time, −4.691 percentage points sleep efficiency, +13.255 minutes wake time. For the Oura ring, a meta-analysis of 6 studies with n = 388 found no significant differences from the reference — the wide confidence intervals for the stages show, however, that this does not mean precise.

What the studies show

Chinoy 2021, Sleep

34 healthy young adults (22 women, 28.1 ± 3.9 years), three consecutive nights in the sleep laboratory including one night with deliberately disrupted sleep, with polysomnography, a research actigraph and seven consumer devices simultaneously. Sensitivity ≥ 0.93, specificity 0.18 to 0.54. The authors’ conclusion: promising for detecting sleep, not for sleep architecture.

Schyvens 2025, Sleep Advances

62 adults (52 men, 10 women, 46.0 ± 12.6 years), one laboratory night with polysomnography and two to four devices each in parallel, including Fitbit Charge 5, Fitbit Sense, Whoop 4.0 and Apple Watch Series 8. All detected over 90 % of sleep epochs, specificity 29.39 % to 52.15 %, Cohen’s kappa 0.21 to 0.53.

Lee 2025, J Clin Sleep Med

Meta-analysis of 24 studies with data from 798 people. Compared with polysomnography: total sleep time −16.854 minutes (95 % CI −26.332 to −7.375), sleep efficiency −4.691 percentage points (−7.079 to −2.302), sleep onset latency +2.574 minutes, wake time after sleep onset +13.255 minutes (4.522–21.988).

Jin 2026, J Transl Med

11 finger-worn devices from 28 papers, all against polysomnography. Pooled accuracy for sleep versus wake: 87 % (95 % CI 86–89 %). Pooled accuracy for the individual stages: light sleep 0.65, deep sleep 0.81, REM 0.74. For detecting sleep apnea, all devices but one were more accurate at an apnea-hypopnea index of 30 than at 5 or 15.

What the measurement does not show

There is no published validation for the readiness and recovery scores. The number on the screen is a proprietary computed value; a review of 107 smart ring studies with around 100,000 participants notes that 89 % were based on proprietary algorithms and 65 % had a moderate to high risk of bias. No study exists for the common threshold of readiness below 70. The target of sleep efficiency above 90 % is also based on a measure that the devices on average report 4.691 percentage points too low. And the rule of thumb of wearing a wearable for at least 14 days has no evidence behind it: the only robust finding is that at least 5 of 7 nights must be available for the nightly HRV coefficient of variation.

How to do it

Use the rough metrics, not the fine ones: time of falling asleep, time of waking and approximate total sleep time are robust; the stage minutes are not. Calculate with a fixed offset rather than absolute numbers — on average 16.854 minutes less sleep, 4.691 percentage points less sleep efficiency and 13.255 minutes more wake time compared with polysomnography. Your own change over the weeks is the usable information. For the HRV trend: at least five of seven nights, the only robust figure on the required data density. Wear the device consistently: in smart ring studies, adherence fell from 80 % after 3 months to 43 % after 12 months. One function beyond sleep is well tested: in 1,155 cycles from 964 participants, the physiology method detected 1,113 (96.4 %) of ovulations with a mean error of 1.26 days, the calendar method with 3.44 days.

Safety

Rings and bands are physically harmless. The real risk is psychological and diagnostic. Orthosomnia describes an excessive preoccupation with perfect sleep, triggered by tracker data; the first description comes from a sleep clinic where patients sought help because the tracker data seemed more convincing to them than polysomnography. In a population sample of 523 people, 176 (35.8 %) regularly used a sleep tracker; depending on the threshold, the proportion with orthosomnia was 3.0 %, 8.6 % or 14.0 %. The picture is not one-sided: in a survey of 1,200 Canadians, just under 45 % of users perceived a positive effect on sleep and stress levels. With loud snoring, breathing pauses or severe daytime sleepiness, the workup belongs in a physician’s hands.

BK-Score Thin human evidence

Human evidence4
Mechanism6
Safety data10
Hype gap3
Track record of use9

Validation studies against polysomnography show: the devices detect sleep and wakefulness fairly well, but the individual sleep stages unreliably. Yet precisely these stages are at the center of the marketing.

The score rates the state of knowledge, not the effect. “Safety data 9” means well studied – not harmless.
Subjective assessment by Biohacking Kompakt based on published scoring rules – not a scientific rating and not a medical recommendation. Rules and all ratings (German)

Frequently asked questions about HRV & sleep stages with a ring/band

How accurately does my ring really measure sleep stages?

For the four-way distinction between wake, light sleep, deep sleep and REM, the pooled accuracies for finger-worn devices are 0.65, 0.81 and 0.74; for wrist-worn devices, overall agreement with polysomnography in Cohen’s kappa is between 0.21 and 0.53. The separation of sleep from wake, by contrast, is good, with around 87 percent pooled accuracy for rings. The device knows fairly well when you slept, but not so much how you slept.

Why does my device show sleep even though I was lying awake in bed?

Because it looks at stillness and a low pulse. That is exactly the best-known error of these devices: across all models studied, sensitivity for sleep was over 93 percent, but specificity for wake only between 18 and 54 percent. Anyone lying still and breathing calmly is counted as asleep.

What actually is the readiness or recovery score?

A company-specific computed value from sleep duration, resting heart rate, HRV and in some cases skin temperature. The formula is not disclosed, and there is no published validation against an external standard; in a review of 107 smart ring studies, 89 percent were based on proprietary algorithms. You can use the score as a rough summary of your own values, but not as a measured quantity.

When is a trend meaningful?

For the nightly HRV course there is one robust figure: at least five of seven nights must be available for the weekly value to be reliable. No comparable study exists for sleep metrics. In practice this means: one night says nothing, a week with gaps says little, a month of consistent use shows a direction.

Can I use it to rule out sleep apnea?

No. In the meta-analysis, finger-worn devices achieved usable accuracy only for severe apnea, that is, from an apnea-hypopnea index of 30, not for mild and moderate forms. The authors explicitly recommend them at most for pre-screening. With snoring with breathing pauses or severe daytime sleepiness, the workup belongs in the sleep laboratory.

Can tracking make my sleep worse?

There is a described pattern for this: orthosomnia, that is, an excessive preoccupation with the perfect sleep score. In a population sample, between 3.0 and 14.0 percent of respondents met the criteria, depending on the definition. If you notice that you sleep worse because of a bad score or plan your day around it, that is the moment to switch off the notifications.

Related

Sources

Open in the database – with all tips, filters and BK-Score (German app)

Information only, not medical advice and not a usage recommendation. If you have pre-existing conditions or before major changes, consult a physician. Last updated: 2026-09-13.