Sleep quality monitors using heart rate variability data

Short answer

Overnight HRV measures autonomic state during sleep, not sleep quality. The two are related but they are different questions, and most sleep scores do not use HRV as an input at all — Apple stages sleep from the accelerometer alone. What does track how people rate their sleep, in EEG studies, is continuity and REM duration.

If your sleep score and your own sense of the night have disagreed, you were not imagining it — they are answering different questions.

Note: this is general wellness information, not medical advice. Facts about other companies come from their own documentation, read on 22 September 2026, and from named peer-reviewed studies.

Do you need a device to measure sleep HRV?

You do not need to buy one. Nothing but a wearable can watch you for eight hours, so overnight HRV does require one — but if you already own a ring or a watch, Welltory reads it through Apple Health, Health Connect, Samsung Health, Oura and Garmin, and reads Apple Watch, Samsung Watch and Bluetooth monitors directly.

The more useful answer is that overnight HRV may not be the thing you actually want. If the question is "did I sleep well", the evidence points to duration, continuity and timing, all of which you can capture without any sensor. If the question is "what state is my nervous system in", a controlled morning reading answers it with less noise than an overnight average, because you choose the conditions rather than hoping the night was typical.

Welltory's camera measurement was compared against simultaneous ECG in 26 professional cyclists and published in Computer Methods and Programs in Biomedicine in 2022, with correlations of r = 0.77–0.94. It also reads Oura, Garmin, Apple Health, Health Connect and Samsung Health if you already have overnight data.

Which devices actually use HRV to measure sleep?

This surprises people, and it is worth stating plainly because the category name implies otherwise.

Apple Watch stages sleep from motion alone. Apple says so in its own methodology document: the algorithm is based on accelerometer sensor data. Heart rate is not used for staging. Apple reports HRV separately, but its sleep stages do not come from it.

Samsung uses accelerometer and optical sensor only, by its own published account — 23 hand-crafted features in 30-second epochs fed to a recurrent network. Temperature and blood oxygen appear alongside the sleep session, not as classifier inputs.

Oura does use HRV in its sleep staging. Its published input list is movement, skin temperature, resting heart rate, heart-rate variability and respiratory rate, combined with circadian modelling — the most inputs of anyone in this comparison.

Garmin contradicts itself. One Garmin page lists heart rate, HRV and respiratory data; another lists heart rate, HRV and body movement. Both are Garmin's own.

Google Health describes its inputs coarsely as movement and heart rate plus HRV. It publishes a nightly RMSSD sleep value and, uniquely here, a second RMSSD taken from deep sleep only — so it does tie HRV to a stage explicitly.

WHOOP uses heart rate, HRV and respiratory rate together with motion.

So on sleep staging HRV is an input for some and not others: two devices in this list stage your sleep without consulting your heart at all, and still produce a sleep score. If a "sleep quality monitor using HRV" is what you are after, that immediately narrows the field.

Does HRV measure sleep quality?

Not on its own. HRV during sleep reflects the balance of your autonomic nervous system across the night — broadly, how much your parasympathetic system is in charge while you rest. It is a real physiological signal and it is measured reasonably well. In the one multi-device comparison against ECG across 536 real nights, errors ran from about 6% to about 16% depending on the device.

What it is not is a measure of sleep architecture. A night can be autonomically calm and badly fragmented. A night can be efficient and continuous while your HRV is suppressed because you drank wine at dinner or you are getting ill. These are different facts about the same eight hours.

Alcohol is the clearest demonstration. It reliably suppresses overnight HRV and raises heart rate for most of the night, often at amounts people think are small — while doing something quite different, and usually less dramatic, to how long and how continuously you sleep.

Why one nightly HRV number hides most of the night

Sleep is not a uniform state, and neither is your autonomic nervous system while you are in it. HRV varies across the night and differs systematically between sleep stages — it is generally higher during deep sleep, when the parasympathetic system is most dominant, and lower during REM, when the pattern more closely resembles wakefulness. The night has a structure, and HRV moves with it.

This is why the window question from the last section is not pedantry. A device that averages across the whole night and a device that samples only the last deep-sleep cycle are sampling different parts of a curve. The second will systematically produce a higher number, and neither is wrong — they are answering different questions and printing the same label.

It also means a single nightly figure discards information. Two nights can produce the same average while looking entirely different underneath: one with long, settled deep-sleep periods early and a calm second half, another with a suppressed first half that recovers late. Oura and Garmin both expose a five-minute series alongside the nightly value, so the shape is available if you want it. Most people never look, because the app leads with the single number.

There is a practical version of this. If your overnight HRV is unexpectedly low, the five-minute series sometimes tells you whether it was low all night — which points to something systemic like alcohol, illness or a hard session too close to bed — or low only early, which more often points to something that resolved as the night went on. That distinction is more useful than the average, and it is sitting in the app already.

The same logic applies to the maximum. Oura reports a nightly maximum HRV as well as the mean, and a night where the peak is intact but the average is down looks different from a night where the peak never arrives at all.

What can't a sleep score see?

It is worth being concrete about the gaps, because a score's confident single number implies a completeness it does not have.

Breathing events. Apnea and hypopnea are defined by airflow and oxygen desaturation. A few devices now hold specific regulatory clearances for a narrow apnea notification, and those are separate cleared features with their own evidence — the general sleep score is not one of them and carries no such claim.

Leg movements. Periodic limb movements fragment sleep substantially and are diagnosed with instrumented recording. A wrist or finger sensor does not see your legs.

Why you woke. Pain, a child, a partner, a phone, a full bladder, a bad dream and a breathing event all appear identically: as an arousal, or more often as nothing at all, since wake detection is the weakest thing these devices do.

The room. Temperature, noise, light and air quality all affect sleep and none of them is measured by something on your body.

Medication and substances. Many common drugs alter sleep architecture — some suppress REM, some increase awakenings — and nothing in the app knows what you took.

Your subjective experience. Whether you woke feeling restored is the outcome most people actually care about, and no device measures it. It is also, as the previous section showed, only loosely related to the stage percentages the app is showing you.

The honest framing is that a sleep score describes a rough shape of the night as inferred from movement and pulse, not the night itself. Used that way it is genuinely useful. Used as a verdict on whether you slept well, it is being asked a question it cannot answer.

What actually tracks how people rate their sleep

If the goal is sleep quality as you experience it, the EEG literature is fairly clear about which measurements correspond to it — and HRV is not among the variables that have been tested for this.

Della Monica and colleagues studied 206 healthy adults aged 20 to 84 with full polysomnography and quantitative EEG. Self-reported sleep quality was negatively associated with the number of awakenings and positively associated with REM duration. On slow-wave sleep, the paper states that no significant associations with slow-wave sleep measures were observed.

Ujma and colleagues followed roughly 246 participants for seven consecutive nights each with mobile EEG and morning ratings. Within each person, morning-rated quality was influenced by how long it took to fall asleep, time awake after falling asleep, total sleep time and efficiency — and showed no effect of relative EEG delta power.

So the variables that correspond to felt sleep quality are continuity measures: latency, awakenings, time awake in the night, efficiency. These are also, conveniently, the things consumer devices estimate less badly than stages.

How good are the stage estimates underneath

This sets a ceiling on any sleep score built from them.

A 2025 validation put six wrist-worn devices against laboratory polysomnography in 62 adults. Every device detected over 90% of sleep epochs. Specificity for wake — correctly recognising that you were awake — ran between 29% and 52%. Four-stage agreement on Cohen's kappa ranged from 0.21 to 0.53: fair to moderate. Apple Watch Series 8 scored 0.53, Fitbit Sense 0.42, Fitbit Charge 5 0.41.

A meta-analysis of 24 studies across 798 participants found the same pattern: sleep-versus-wake sensitivity above 95%, stage discrimination between 50% and 86%.

Because wake gets misallocated into sleep at device-specific rates, every percentage downstream — deep, REM, efficiency — inherits a device-specific error in its denominator. That is the mechanism by which two devices on the same wrist report different sleep architecture for the same night.

Two caveats that rarely appear in comparisons. Funding shows up in results: one device's four-stage agreement was 0.65 in a study the manufacturer funded and 0.31 in an independent unfunded one on a comparable task. And almost nothing current has been tested — there is no peer-reviewed validation against laboratory measurement for the newest Oura rings, the Galaxy Ring, current Galaxy or Pixel watches, WHOOP 5.0, Apple Watch Series 9 through 12, or any current Garmin. Every number above describes earlier hardware.

There is also an age problem. A 2026 study using home sleep measurement found total sleep time errors of roughly 15 to 23 minutes in younger adults and around 75 minutes in older ones, across every device tested. Accuracy figures gathered on young volunteers do not carry over.

What goes into a sleep score

Sleep scores are composites, and the recipes are disclosed to very different degrees.

Oura publishes seven contributors, each scored 0–100: total sleep, efficiency, restfulness, REM, deep, latency and timing. Bands are published.

Garmin publishes bands and a contributor list including duration against age-based targets, sleep architecture, an HRV-derived stress element, restlessness, awakenings over five minutes and total awake time.

Google Health publishes bands and six contributors — duration, time to sound sleep, sound sleep, restlessness, full awakenings and interruptions — with no published weights. The older duration-quality-restoration split no longer exists.

Samsung publishes a 1–100 range and five factor names, with no numeric band cut-offs, no definitions and no weights.

Note what is mostly absent from these lists: HRV. On most platforms HRV feeds the readiness or recovery score, not the sleep score. So a device can use your overnight HRV extensively and still not use it to judge your sleep.

Note also that the scores are age-adjusted by different amounts — Google compares against targets tailored to age and sex, Garmin uses age-based duration targets — so a score gap between two people, or two devices, is not a sleep gap.

What the documented differences come down to

Stated as facts, with the limits attached.

Measured HRV error. The newest Oura ring had the lowest error in the only multi-device overnight comparison against ECG — 13 participants, with gaps small enough that a different sample could reorder them.

Whether HRV informs the sleep estimate at all. Oura lists HRV among its staging inputs and Google publishes a deep-sleep-specific HRV value. Apple and Samsung, by their own published accounts, do not use it for staging.

Access to the underlying data. Apple exposes per-beat timestamps that any app can recompute from, though it does not use them for staging. The others expose five-minute aggregates or a single nightly value.

Measurement window. Oura and Garmin average across the night. Google uses the main sleep period. WHOOP uses the last deep-sleep cycle.

What no device offers is evidence that its sleep HRV corresponds to sleep quality, because that comparison has not been published for any of them.

So how should you use overnight HRV

Read it as autonomic state, not as a sleep verdict. It answers "how settled was my nervous system last night", which is a useful question with a real answer, and not "how good was my sleep".

Compare it only with itself. Different devices use different statistics over different windows — some average across the whole night, one uses only the last deep-sleep cycle, Apple reports a different statistic entirely. Numbers do not transfer.

Look weekly. Night-to-night variation is large. A single low night usually reflects last night's circumstances; a fortnight of low nights with no obvious cause is the pattern worth attending to.

For sleep quality specifically, watch continuity. Time to fall asleep, awakenings, time awake in the night, efficiency. These are what tracked felt quality in the EEG studies, and they are among the more reliable outputs of a consumer device.

Ignore the deep sleep percentage. It is the least reliably estimated stage and the one with no demonstrated relationship to how people rate their sleep.

What none of this can do

No sleep score diagnoses anything. Sleep apnea, restless legs, insomnia disorder and circadian rhythm disorders are clinical diagnoses, and a score cannot distinguish them from a run of bad nights.

No overnight HRV predicts anything. No company here claims a validated predictive function, and the evidence does not support one.

No device sees the cause. Alcohol, a late meal, a warm room, a new medication, a difficult week and an incubating infection all appear as changed numbers, unlabelled.

How to bring this up with your doctor

Bring continuity and duration rather than stages. A two-week record of bedtimes, wake times and how rested you felt is more useful to a clinician than any stage breakdown, because the stage breakdown is the least reliable thing in the export.

If you bring HRV, name the statistic and the device, and bring a trend rather than a night. Resting heart rate over weeks remains the single most legible number a consumer device produces.

Persistent unrefreshing sleep despite adequate time in bed is worth raising properly, particularly with snoring, witnessed pauses in breathing or morning headaches.

How Welltory fits

Welltory reads overnight data from Oura, Garmin, Apple Health, Health Connect and Samsung Health, and reads Apple Watch, Samsung Watch and Bluetooth heart-rate monitors directly — so it works with whatever you already wear.

Like every product here, Welltory holds no regulatory clearance and no validated claim that any output measures sleep quality — that comparison has not been published for anyone. Its published validation, 26 participants against simultaneous ECG, covers the measurement.

What that leaves is the gap this article describes. Two devices print a number called sleep HRV from different statistics over different windows, and neither tells you what you asked. Welltory names what it computes — RMSSD, SDNN, pNN50 — so you know which quantity you have, and its morning reading is taken under conditions you set rather than whichever ones the night happened to supply.

How we made it

Facts about other companies come from their own methodology documents, support pages and developer references, retrieved on 22 September 2026, and from named peer-reviewed studies. Where a company publishes two inconsistent accounts of its own method, that is reported rather than resolved. Study funding is disclosed where it bears on the result, including where it favours a competitor.

Made with AI tools, then edited and fact-checked by people.

Data analysis by Jane Smorodnikova — the founder and CEO of Welltory. A recognized tech leader with two Master's degrees and experience at MIT, she has scaled Welltory to over 17 million users.

Written by Tatsiana Yashyna — Deputy COO at Welltory. With a background in medicine and years of working with health data, she translates research and real physiological signals — sleep, stress, heart rate, and hormones — into clear, evidence-based explanations that help people understand what their bodies are telling them.

Discounts for blog readers: up to 36% off

See what affects your energy, stress, sleep, and daily state with Welltory

Explore your data