How accurate are sleep trackers? What the validation studies show

How sleep trackers compare with a sleep lab

On total sleep time, some devices differed significantly from a sleep lab and others didn’t^{3, 5, 7}. In one single-night study of young adults, most devices tracked heart rate during sleep closely². Several of the lab studies had industry ties^{2, 4, 6}. The step-count link comes from observational cohorts, not trials of wearables^{8, 14}.

Polysomnography (PSG), the overnight sleep-lab recording, is the reference standard that consumer trackers are judged against¹. Wearables infer sleep from other signals. In one of the FDA’s own examples, a wrist-worn wellness product measures sleep with an accelerometer and pulse rate with a photoplethysmogram (PPG) sensor⁹. Validation studies then go through the night epoch by epoch. They compare the device’s call of sleep or wake, and of light, deep or REM sleep, with a PSG scorer’s¹.

Detecting sleep is the easy part. Spotting wake is harder. A US lab study followed 34 healthy young adults (mean age 28) over three nights. The devices tested caught at least 93% of sleep epochs but correctly identified only 18% to 54% of wake epochs. They also tended to do worse on nights with poorer or disrupted sleep¹.

A 2025 Belgian study found the same pattern for six wrist devices. It enrolled 62 adults (mean age 46, mostly men), both healthy participants and patients with suspected sleep apnea. The devices detected more than 90% of sleep but only 29% to 52% of wake. Most differed significantly from PSG on total sleep time, sleep efficiency and time awake after falling asleep³.

An earlier meta-analysis of Fitbit models found that those without sleep staging tended to overestimate total sleep time and underestimate wake. Models with sleep staging didn’t differ significantly from PSG on total sleep time, sleep efficiency or wake after sleep onset. Even so, they identified only 58% to 69% of wake epochs⁷.

Sleep stages are where devices diverge. The 2025 study measured stage agreement with PSG using Cohen’s kappa. Scores ranged from 0.21 for a Garmin Vivosmart 4 to 0.53 for an Apple Watch Series 8, which the authors describe as fair to moderate³. In a study of 53 healthy adults (mean age 25), six devices agreed with PSG on 50% to 65% of epochs once stages were included. Their kappa ran from 0.20 to 0.52².

A Korean study of 75 adults with sleep complaints tested 11 products: five wearables, three bedside devices and three phone apps. Their stage agreement scores (macro F1, where 1.0 is perfect) ranged from 0.26 to 0.69. Among the wearables, Fitbit Sense 2 and Galaxy Watch 5 each scored 0.58. Oura Ring 3 scored 0.52 and Apple Watch 8 scored 0.49⁴. The top-scoring product overall was a phone app developed by Asleep, a company most of the authors were affiliated with and hold stock or stock options in⁴.

A larger ring study followed 96 generally healthy Japanese adults aged 20 to 70 for up to three nights of ambulatory PSG. It tested the Oura Ring Gen3 with its 2.0 staging algorithm⁵. Total sleep time, time awake after sleep onset, and light and deep sleep didn’t differ significantly from PSG. REM sleep was underestimated by about 4 to 6 minutes⁵.

The ring correctly identified wake 73% to 75% of the time. Per-stage accuracy ranged from 75.5% for light sleep to 90.6% for REM⁵. Oura says it partly funded the study but took no role in data analysis or writing⁶.

How accurate are heart rate and HRV readings?

Heart rate during sleep is where most devices did best. The 2022 lab study of 53 healthy young adults measured agreement with ECG as an intraclass correlation, where 1.0 is perfect. Four of six devices scored 0.85 to 0.99. One scored 0.65, and a Garmin Forerunner 245 Music scored 0.41². The Garmin’s readings came from a 3-minute sample taken before lights out, not during sleep².

Heart rate variability (HRV) was harder to get right. HRV is the beat-to-beat variation in heart rate, and one ring maker’s readiness score draws on it¹³. The WHOOP 3.0 strap reached 0.99. Four devices fell between 0.63 and 0.69, and the Garmin scored 0.24².

These were models available in 2022, tested on a single night. The Australian Institute of Sport funded the study. The authors’ research group receives support from WHOOP, which they say had no role in the study². The FDA lists HRV among the parameters a wellness product may estimate. One condition is that it doesn’t show values that mimic clinical ones unless those values are validated⁹.

Step counts and mortality

Of the metrics covered here, step counts have the clearest link to long-term health outcomes. That evidence comes from observational cohorts, though, not trials of wearables. A meta-analysis pooled 15 international cohorts, covering 47,471 adults followed for a median of 7.1 years⁸.

The comparison group was the lowest quarter, at a median of 3,553 steps a day. People averaging 5,801 steps had a hazard ratio for death of 0.60. At 7,842 steps it was 0.55, and at 10,901 steps it was 0.47⁸.

In adults 60 and older, risk kept falling until about 6,000 to 8,000 steps a day. In younger adults, it kept falling until about 8,000 to 10,000. The authors noted little evidence for the popular 10,000-step target⁸.

A 2025 dose-response meta-analysis drew on 57 studies from 35 cohorts. It compared 7,000 steps a day with 2,000¹⁴. The higher count was associated with 47% lower all-cause mortality (hazard ratio 0.53, 14 studies). It was also associated with 25% lower incidence of cardiovascular disease and 38% lower dementia risk. The risk of depressive symptoms was 22% lower¹⁴.

All four results were rated moderate certainty. A 28% lower risk of falls was rated very low certainty, and cancer incidence didn’t differ significantly¹⁴. The authors concluded that 7,000 steps a day is associated with clinically meaningful improvements and may be a more realistic target for some people. They added that 10,000 can still be a viable target for more active people¹⁴.

These are observational associations. Both teams describe more daily steps as associated with lower risk^{8, 14}. The 2025 review lists residual confounding and the small number of studies for most outcomes among its limitations¹⁴. Neither was designed to show that buying a tracker, or adding steps, causes a longer life.

What the evidence does not show

Readiness and recovery scores are proprietary composites. One ring maker describes its readiness score as a number from 0 to 100. It’s built from overnight metrics: the lowest resting heart rate and its timing, body temperature, sleep and the previous day’s movement. It also draws on longer-term balances of HRV, sleep and activity, comparing your recent two weeks with your average over the past two months¹³. The company says scores below 70 suggest you may benefit from prioritizing rest and recovery¹³.

FDA policy lets wellness devices put their readings in the context of sleep, activity, stress or recovery⁹. But none of the validation studies reviewed here tested whether such a score predicts performance, illness or any health outcome^1–5. The accuracy studies also have limits that headline claims tend to skip.

• They’re small and short: 34 to 96 people, usually over one to three nights^1–5
• Three of the five studies enrolled only healthy or generally healthy adults^1–5
• Devices tended to do worse on disrupted nights¹, so results from healthy sleepers may not carry over to people whose sleep is often disrupted
• Results apply to the model and algorithm version tested, such as one ring study’s staging-algorithm release⁵ or the Apple Watch Series 6, Oura Ring Gen2 and WHOOP 3.0 in the 2022 study²
• Studies use different agreement measures (Cohen’s kappa, macro F1, per-stage accuracy), so their accuracy figures can’t be compared directly^3–5
• Several studies had industry ties: partial manufacturer funding⁶, support for the research group², or authors affiliated with, and holding stock in, a company whose own app was tested⁴
• None of these studies tested whether wearing a tracker improves sleep or health
• The step research shows an association with lower risk, not an effect of owning the device^{8, 14}

Safety and FDA oversight

Wearables sold for wellness rather than medical use fall under the FDA’s general wellness policy. Under it, the agency doesn’t intend to examine low-risk general wellness products to decide whether they’re medical devices⁹. Its revised guidance, issued January 6, 2026, addresses products that use non-invasive sensing, such as optical sensing, to estimate blood pressure, oxygen saturation, blood glucose or HRV. It says they may fall under the policy if their outputs are intended solely for wellness and they meet several other conditions⁹.

To qualify, the product must not be intended to diagnose, cure, mitigate, prevent or treat a disease, or to substitute for an FDA-authorized device. It must not prompt or guide specific clinical action. And it must not show values that mimic clinical ones unless those values are validated⁹.

Products that claim clinical accuracy, clinical equivalence, or medical or clinical grade fall outside the policy. Falling under the policy doesn’t establish that a product has been shown to be safe or effective⁹. One of the guidance’s own examples is a wrist-worn activity and recovery product that reports sleep, pulse rate and blood pressure. It’s treated as low-risk wellness only if its blood pressure values are validated and its claims don’t imply medical or clinical use⁹.

The rules were tested in 2025. On July 14, 2025, the FDA sent WHOOP a warning letter about its Blood Pressure Insights feature, which gave daily systolic and diastolic estimates. The letter said the feature was a medical device, not a wellness feature, because blood pressure estimation is inherently associated with diagnosing low and high blood pressure. It also said an erroneously low or high reading could lead to delayed or missed treatment¹⁰.

WHOOP had argued the feature wasn’t a device because it wasn’t intended to diagnose or treat disease¹⁰. In a June 17, 2026 close-out letter citing the updated guidance, the FDA said it doesn’t intend to enforce device requirements for the feature as WHOOP modified it. It added that this applies only to the modified feature^{11, 15}.

To count as low-risk wellness products, these devices must be non-invasive and not implanted⁹. So the risk the FDA has flagged is less about the sensor than about the number. An inaccurate blood pressure estimate could lead to delayed or missed treatment¹⁰.

Your data and privacy

HIPAA governs the privacy and security of online health records for most hospitals, doctors’ offices and insurance companies¹². Health apps and connected devices outside HIPAA can fall under the FTC’s Health Breach Notification Rule instead. The FTC’s own example of a business that’s probably covered is a health app that collects information from consumers and can sync with their fitness tracker. The FTC also says its July 2024 amendments make clear that makers of health apps, connected devices and similar products must comply¹².

Under the rule, a breach isn’t limited to cybersecurity intrusions. Disclosing covered health information without the person’s authorization can count, for example sending it to a social media platform¹². Companies must notify affected people without unreasonable delay, and within 60 calendar days after discovering a breach. Violators may face civil penalties of up to $53,088 per violation¹². The rule is about notice after a breach, so check the company’s privacy policy to see what it shares with your permission.

What to look for

The evidence fits using a sleep or recovery wearable to follow sleep patterns and longer-term changes. It doesn’t fit treating one as a sleep lab or a medical monitor^{3, 9}.

• Published validation against polysomnography or ECG for the current model and algorithm, ideally by researchers independent of the maker^{2, 5, 6}
• Study participants who resemble you, since most studies tested adults over one to three nights and accuracy dropped on disrupted nights^{1, 5}
• More weight on detected sleep and heart rate during sleep than on time awake at night or minutes of deep or REM sleep, which devices measured less accurately^1–3
• A plain explanation of what feeds any readiness or recovery score and what baseline it compares you with¹³
• Whether a blood pressure, oxygen or glucose feature is FDA-authorized or wellness only: claims of clinical accuracy or medical or clinical grade put a product outside the wellness policy⁹, and the FDA’s 2025 warning letter to WHOOP quoted the company’s “medical-grade” marketing¹⁰
• A privacy policy that says what’s shared, sold or used for research, how to export and delete your data, and how the company will tell you about a breach¹²

Sources

1. Performance of seven consumer sleep-tracking devices compared with polysomnography Sleep (Chinoy et al.), 2021
2. A Validation of Six Wearable Devices for Estimating Sleep, Heart Rate and Heart Rate Variability in Healthy Adults Sensors (Miller, Sargent and Roach), 2022
3. A performance validation of six commercial wrist-worn wearable sleep-tracking devices for sleep stage scoring compared to polysomnography SLEEP Advances (Schyvens et al.), 2025
4. Accuracy of 11 Wearable, Nearable, and Airable Consumer Sleep Trackers: Prospective Multicenter Validation Study JMIR mHealth and uHealth (Lee et al.), 2023
5. Validity and reliability of the Oura Ring Generation 3 (Gen3) with Oura sleep staging algorithm 2.0 (OSSA 2.0) when compared to multi-night ambulatory polysomnography: A validation study of 96 participants and 421,045 epochs Sleep Medicine (Svensson et al.), 2024
6. Oura Accuracy Further Validated by New Study from the University of Tokyo Oura, 2024 · manufacturer
7. Accuracy of Wristband Fitbit Models in Assessing Sleep: Systematic Review and Meta-Analysis Journal of Medical Internet Research (Haghayegh et al.), 2019 · meta-analysis
8. Daily steps and all-cause mortality: a meta-analysis of 15 international cohorts The Lancet Public Health (Paluch et al.), 2022 · meta-analysis
9. General Wellness: Policy for Low Risk Devices (guidance issued January 6, 2026) US Food and Drug Administration, 2026 · regulator
10. Warning letter: WHOOP, Inc. (MARCS-CMS 709755), July 14, 2025 US Food and Drug Administration, 2025 · regulator
11. Close-out letter: WHOOP, Inc. (MARCS-CMS 709755), June 17, 2026 US Food and Drug Administration, 2026 · regulator
12. Complying with FTC's Health Breach Notification Rule US Federal Trade Commission, 2024 · regulator
13. Readiness Score Oura · manufacturer
14. Daily steps and health outcomes in adults: a systematic review and dose-response meta-analysis The Lancet Public Health (Ding et al.), 2025 · meta-analysis
15. FDA closes warning letter to Whoop for blood pressure insights feature MobiHealthNews, 2026 · news