Smart Ring Sleep Stage Accuracy: What Sleep-Lab Studies Show
How accurate are smart-ring sleep stages? Four sleep-lab studies stage by stage, what kappa and specificity mean, and why no fixed correction works.

What does 'sleep stage accuracy' actually mean?
The gold standard for measuring sleep is a polysomnogram (a clinical overnight study that combines EEG brain-wave recording with eye movement, muscle tone and respiratory sensors). PSG scores sleep in 30-second windows and assigns each window to one of four stages: wake, light (combining N1 and N2), deep (N3 or slow-wave sleep) and REM. Any consumer device that claims to report sleep stages is making an estimate of what the PSG would have scored - using a far smaller set of sensors than the lab does.
Smart rings sit at the wrist-and-finger end of the consumer-wearables spectrum. They do not measure brain waves. Instead they use a finger-worn photoplethysmography sensor (an optical heart-rate and heart-rate-variability reader, often abbreviated PPG), a skin-temperature sensor and a 3-axis accelerometer for movement. The ring's app runs a proprietary model that takes those inputs and outputs a sleep-stage hypnogram. The honest question is how often that hypnogram agrees with the PSG it is trying to predict.
Almost all of the peer-reviewed sleep-lab evidence on rings concerns one brand. The Oura Ring is a Finnish smart ring that has been studied against polysomnography since its second generation, so this guide leans on Oura data and says so. Where a claim rests on a single study or on a manufacturer-funded one, we flag it.
Why do accuracy claims use so many different numbers?
A headline like '96 percent accurate' can be true and still mislead, because sleep-tracker validations report several different statistics and each one answers a different question. Knowing which is which is the fastest way to judge any accuracy claim you read, including a ring maker's own.
Sensitivity (for sleep) - The share of epochs the lab scored as sleep that the ring also called sleep. Rings score very high here, because most of the night is sleep and a still body looks asleep.
Specificity (for wake) - The share of epochs the lab scored as wake that the ring also called wake. This is the hard part: lying still but awake looks like sleep to a motion-and-pulse sensor. The 2019 Oura validation found 96 percent sensitivity but only 48 percent wake specificity.
Epoch-by-epoch agreement - The percentage of 30-second windows where the ring's stage matches the lab's stage. It flatters devices on long, uniform stages and hides misses on short ones.
Cohen's kappa - Agreement corrected for the matches you would get by chance. Cohen's kappa is a score from 0 to 1 where roughly 0.4-0.6 is moderate and 0.6-0.8 substantial. PABAK is a variant that also adjusts for the fact that sleep epochs vastly outnumber wake epochs.
Minute bias - The average difference in minutes between the ring and the lab for a whole-night total, such as deep-sleep time. A ring can have decent epoch agreement and still carry a 30-minute bias in one stage.
For context, two trained human scorers reading the same polysomnogram do not agree perfectly either. Oura's own summary of its 2024 study puts inter-scorer agreement at about 83 percent, which is the practical ceiling any consumer device is measured against.
What did the four peer-reviewed sleep-lab studies find?
Four studies carry most of the weight. Read them as a sequence: each tested a different generation of hardware or algorithm, on a different population, which is why their numbers differ.
- de Zambotti and colleagues, Behavioral Sleep Medicine, 2019: 41 adolescents and young adults aged 14-22, an early ring running the first version of Oura's algorithm, one lab night. 96 percent sleep sensitivity, 48 percent wake specificity, and stage agreement of 65 percent (light), 51 percent (deep) and 61 percent (REM). The ring under-counted deep sleep by about 20 minutes and over-counted REM by about 17.
- Chee and colleagues, Nature and Science of Sleep, 2021: 53 adolescents aged 15-19, second-generation ring, several lab nights. Sleep/wake accuracy of about 89 percent, kappa around 0.45 (moderate). Total sleep time came in 33-47 minutes short, light sleep 51-81 minutes short, deep sleep 31-47 minutes long. The authors concluded staging 'needs improvement'.
- Svensson and colleagues, Sleep Medicine, 2024: 96 adults aged 20-70, third-generation ring running Oura's OSSA 2.0 staging algorithm, up to three nights each and 421,045 epochs. Sensitivity 94.4-94.5 percent, wake specificity 73.0-74.6 percent, PABAK 0.83-0.84, and stage accuracy from 75.5 percent (light) to 90.6 percent (REM). Light and deep sleep time did not differ significantly from the lab; REM was under-counted by 4-6 minutes. The study was partly funded by Oura.
- Robbins and colleagues, Sensors, 2024: 35 healthy adults, one lab night, Oura Ring against Apple Watch and Fitbit Sense. Apple Watch is Apple's wrist-worn smartwatch, and Fitbit Sense is Fitbit's health-focused smartwatch. All three detected sleep with at least 95 percent sensitivity; Oura's stage sensitivity was 76.0-79.5 percent with no significant bias in any stage, while the Apple Watch under-counted deep sleep by 43 minutes. The study was funded by Oura and run by researchers at Brigham and Women's Hospital.
| de Zambotti 2019 | Chee 2021 | Svensson 2024 | Robbins 2024 | |
|---|---|---|---|---|
| Who was tested | 41 people aged 14-22, early ring | 53 people aged 15-19, Gen 2 ring | 96 adults aged 20-70, Gen 3 ring | 35 adults aged 20-50 |
| Wake detection | 48% specificity | 89% overall sleep/wake accuracy | 73-75% specificity | Sleep sensitivity 95% or more |
| Stage result | 51-65% per stage | Kappa about 0.45 | 75.5-90.6% per stage | 76-79.5% stage sensitivity |
| Biggest bias | Deep about 20 min short | Deep 31-47 min long | REM 4-6 min short | None significant |
How accurate is each sleep stage?
Wake - the weakest link historically. Lying awake and still is the classic miss, which is why older studies found wake after sleep onset over- or under-counted by half an hour. Newer algorithms have narrowed this but not closed it.
Light sleep - the lowest per-stage accuracy in the 2024 multi-night study (75.5 percent) and badly under-counted in the adolescent study. Light sleep is the stage the ring uses as a catch-all when the signal is ambiguous.
Deep sleep - the most contested number. It was under-counted by about 20 minutes in one study, over-counted by 31-47 minutes in another, and unbiased in the two 2024 studies. Deep sleep is the stage people most want to optimise and the one where older evidence disagrees most.
REM sleep - over-counted by about 17 minutes in 2019, under-counted by 4-6 minutes in 2024. The newer algorithm scored REM with the highest accuracy of any stage (90.6 percent), partly because REM has a distinctive heart-rate and heart-rate-variability signature.
Why have the numbers improved since 2019?
The hardware changed less than the software. The 2019 and 2021 studies tested early rings with Oura's original staging model; the 2024 studies tested newer rings running a machine-learning model trained on large sets of lab-scored nights. Rings infer stages from heart rate, heart-rate variability, skin temperature, movement and the time of night, and a better model makes more of the same signals. Two cautions apply: both 2024 studies had Oura funding, and both recruited healthy sleepers. Independent replication on people with insomnia, apnoea or shift-work schedules is still thin.
What makes your own stage readings less accurate?
Validation studies test healthy volunteers in controlled conditions. Your bedroom adds variables the studies screened out, and each one tends to push the stage breakdown further from what a lab would score.
A loose ring - the optical sensor needs steady skin contact. A ring that spins or slides gives a noisier pulse signal, and the staging model is only as good as that signal.
Lying awake still - reading in bed, or quiet wakefulness in the night, is the scenario every study shows is hardest to catch.
Alcohol and late meals - both raise overnight heart rate and depress heart-rate variability, which the model also uses as stage cues.
Medication that alters heart rate - beta-blockers and similar drugs blunt the heart-rate changes the algorithm reads, and none of the validation studies tested people taking them.
A sleep disorder - all four studies screened for healthy sleepers. Fragmented sleep from apnoea or insomnia is exactly where consumer staging is least tested.
Being outside the study population - the evidence covers teenagers and adults up to 70; most participants in the 2024 studies were East Asian or under 40. Results may not carry over evenly.
What about smart rings other than Oura?
Peer-reviewed polysomnography studies of comparable size are scarce for most other rings, so their stage accuracy rests largely on manufacturer statements and independent YouTube testing such as the head-to-head comparisons published by Quantified Scientist, a reviewer who tests wearables against a home EEG device. Those comparisons are useful but are not peer-reviewed and can shift with every firmware update. When a brand quotes an accuracy figure, ask four questions: who funded the study, how many people and nights it covered, whether they were healthy sleepers, and whether the figure is sensitivity, agreement or kappa.
For our ranked picks see the best smart ring for sleep tracking, and for per-ring detail the Oura Ring 4 review, Ultrahuman Ring Air review, RingConn Gen 2 review and Samsung Galaxy Ring review. The health metrics explainer covers the sensors themselves.
Where does smart-ring sleep tracking fall short?
Absolute stage minutes - treat nightly deep and REM totals as relative numbers to compare against your own baseline, not absolute targets to compare against a friend or a population average.
Sleep-onset latency - the minutes between lying down still and actually falling asleep are easy to misread as light sleep.
Short naps - the algorithms need a sustained low-movement, lowered-heart-rate window, so naps under about 20 minutes are often missed.
Sleep disorders - a ring is not a diagnostic tool. Our sleep apnoea guide covers what the newer breathing features can and cannot flag.
Frequently asked questions
How accurate are smart rings at measuring sleep stages?
Do smart rings over-estimate deep sleep?
What is a good kappa score for a sleep tracker?
Why does my ring say I was asleep when I was lying awake?
Are the Oura validation studies independent?

Best Smart Ring for Sleep Tracking

How Accurate Is the Readiness Score?

HRV Explained: What Your Smart Ring HRV Number Really Means
