Wearable accuracy: which numbers you can trust and which are educated guesses
A ring or watch shows every metric with the same confident font. The research sorts them into three groups: measured well, estimated with known error, and proprietary scores that cannot be checked.
Wearables report a dozen numbers each morning. Some are close to what a clinic would measure. Some are estimates with an error band wider than the day-to-day changes people watch. And some are scores built by each company from the others, which cannot be validated at all because the formula is private. Knowing which is which is most of what it takes to use the data well.
The scorecard
| Metric | Category | Evidence |
|---|---|---|
| Resting heart rate | Measured well | All seven devices accurate at rest vs ECG (Stone 2021); overnight bias within about 1 bpm for four of six devices (Miller 2022) |
| Heart rate during exercise | Measured, with more error | Absolute error 30% higher during activity than rest (Bent 2020); Apple Watch and Garmin most accurate in a 158-study review (Fuller 2020) |
| Steps | Measured well in the lab | Fitbit, Apple Watch and Samsung measured steps accurately in laboratory settings (Fuller 2020) |
| Sleep vs wake, total sleep | Estimated, good | Sleep-wake agreement 86 to 89% for six devices (Miller 2022); sensitivity for sleep 0.93 or higher, specificity for wake 0.18 to 0.54 (Chinoy 2021) |
| Overnight HRV (RMSSD) | Estimated, reasonable on average | Oura error 6.84% at rest (Stone 2021); overnight RMSSD bias -4.5 to -22.4 ms across six devices (Miller 2022) |
| Respiratory rate | Estimated, good at night | Mean absolute error 0.46 breaths/min for one HRV-derived method (Natarajan 2021) |
| Skin temperature | Relative only | Reported as deviation from baseline; tracks illness and cycle-related shifts (Mason 2022, Maijala 2019) |
| Sleep stages | Estimated, moderate | Four-stage agreement 50 to 65% (Miller 2022); stage results "inconsistent" (Chinoy 2021) |
| SpO2 | Estimated, wellness only | Wrist device Arms 5.82% below 88% saturation vs 3.52% for a clinical fingertip unit (Jiang 2026) |
| VO2 max | Estimated, wide individual error | Limits of agreement about 10 mL/kg/min either side for exercise-based algorithms (Molina-Garcia 2022) |
| Calories burned | Poor | No device under 20% error (Shcherbina 2017); no brand accurate (Fuller 2020) |
| Readiness, Recovery, Body Battery, sleep scores | Proprietary composite | Inputs published by each company; weights and formulas are not, so they cannot be independently validated |
The key studies
| Study | Year, journal | Who and how many | Design | Main result |
|---|---|---|---|---|
| Stone et al. | 2021, Frontiers in Sports and Active Living | 5 healthy adults, 148 trials, 7 devices | Resting heart rate and RMSSD vs multi-lead ECG | Heart rate accurate on all devices; RMSSD error from 4.10% (HRV4Training with chest strap) and 6.84% (Oura) to 112% (a camera app) |
| Miller et al. | 2022, Sensors | 53 adults, one lab night | Apple, Garmin, Polar, Oura, WHOOP, Somfit vs PSG and ECG | All valid for sleep timing and duration; all need improvement for stages; heart rate and HRV agreement ranged from poor to excellent by device |
| Chinoy et al. | 2021, Sleep | 34 adults, three nights | 7 consumer sleep devices and actigraphy vs PSG | Most matched or beat research actigraphy on sleep-wake; Garmin models performed worse; all worse on disrupted nights |
| Fuller et al. | 2020, JMIR mHealth and uHealth | 158 publications, 9 brands | Systematic review | Steps and heart rate reasonably accurate in the lab; energy expenditure not accurate for any brand |
| Bent et al. | 2020, npj Digital Medicine | Consumer and research devices across skin tones | Heart rate vs ECG at rest and during activity | No significant difference by skin tone; significant differences between devices and activity types |
| Shcherbina et al. | 2017, Journal of Personalized Medicine | 60 adults, 7 wrist devices | Heart rate and energy expenditure vs ECG and calorimetry | Heart rate error lowest in cycling; error higher with darker skin, higher BMI, in men and when walking; energy expenditure error above 20% for all |
Two of those studies disagree on skin tone. The 2017 Stanford study found higher heart rate error with darker skin; the 2020 Duke study, which set out to test the full range of skin tones, found no significant difference. Pulse oximetry, which depends more heavily on light absorption, has a clearer problem, as the FDA has recognised. See blood oxygen.
Measured, cleared and wellness
A few wearable features have gone through FDA review as medical devices. In September 2018 the FDA granted Apple's ECG app a De Novo classification as an over-the-counter device that determines the presence of atrial fibrillation or sinus rhythm from a single-lead ECG. Most other metrics, including HRV, sleep stages, readiness scores and wrist SpO2 on many devices, are offered as general wellness features. Companies say so: Oura states its ring is not a medical device, and Apple states Blood Oxygen readings are not for medical use.
Things that make any device less accurate
- Fit. A loose band lets light leak and the sensor move.
- Movement. Optical heart rate is least reliable during rapid changes in intensity.
- Disrupted nights. Devices missed more wake on the disrupted-sleep night in the Chinoy study.
- Software updates. A validation of one model and algorithm version does not carry over automatically to the next.
- Industry funding. Several studies are funded by or written with the device maker. That is declared in each paper, and it is worth checking.
The practical rule
Trust measured numbers, read estimated numbers as trends, and treat composite scores as one company's opinion. The reading wearable trends page shows how to separate a real change from noise.
Logging it
Peptide AI imports wearable data through wearable integrations and builds a baseline from your own history before it flags changes. Wearable data in the app is not diagnostic, and the app does not diagnose. It puts the numbers next to the dose log and lab results, which are measured, so you can weigh one against the other.
Questions people ask
Which wearable metrics are most accurate?
Resting heart rate, step counts in the lab, and total sleep time. In a 2021 study all seven devices tested measured resting heart rate accurately against ECG.
Which wearable metrics are least accurate?
Calories burned. In a Stanford study of seven wrist devices none had energy expenditure error below 20 percent, and a 2020 review of 158 publications found no brand accurate.
Are sleep stages from wearables reliable?
Moderately. Four-stage agreement with polysomnography was 50 to 65 percent across six devices in a 2022 study, compared with 86 to 89 percent for sleep versus wake.
Are any wearable features FDA cleared?
Some. In September 2018 the FDA granted Apple's ECG app a De Novo classification for detecting atrial fibrillation or sinus rhythm. Most other metrics are general wellness features.
Does skin tone affect wearable heart rate?
Studies disagree. A 2017 study found higher error with darker skin; a 2020 study found no significant difference. Pulse oximetry has a clearer documented skin pigmentation effect.
Sources
- Assessing the accuracy of popular commercial technologies that measure resting heart rate and heart rate variability (Stone et al.), Frontiers in Sports and Active Living, 2021
- A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults (Miller et al.), Sensors, 2022
- Performance of seven consumer sleep-tracking devices compared with polysomnography (Chinoy et al.), Sleep, 2021
- Reliability and validity of commercially available wearable devices for measuring steps, energy expenditure, and heart rate: systematic review (Fuller et al.), JMIR mHealth and uHealth, 2020
- Investigating sources of inaccuracy in wearable optical heart rate sensors (Bent et al.), npj Digital Medicine, 2020
- Accuracy in wrist-worn, sensor-based measurements of heart rate and energy expenditure in a diverse cohort (Shcherbina et al.), Journal of Personalized Medicine, 2017
- Measurement of respiratory rate using wearable devices and applications to COVID-19 detection (Natarajan et al.), npj Digital Medicine, 2021
- Validity of estimating the maximal oxygen consumption by consumer wearables: a systematic review with meta-analysis and expert statement of the INTERLIVE network (Molina-Garcia et al.), Sports Medicine, 2022
- Performance of wearable pulse oximetry during controlled hypoxia induction: instrument validation study (Jiang et al.), JMIR Formative Research, 2026
- De Novo classification request DEN180044: ECG App, U.S. Food and Drug Administration, 11 September 2018
- Blood Oxygen Sensing (SpO2), Oura Help Center
- How to use the Blood Oxygen app on Apple Watch, Apple Support
