How Accurate Are Under-Mattress Sleep Trackers Compared With Wearables?
It caught 95% of sleep but only 37% of waking time. In a separate home study of 117 healthy adults, wake detection was 73%.
Evidence rating for the proposition “Consumer wearables measure sleep parameters with acceptable agreement against polysomnography”: Uncertain · KnowYourPrime Evidence Registry, Finding KYP-ER-0007
Under-mattress sensors overstate sleep because they miss waking time. In a sleep lab, the Withings Sleep Analyzer overestimated total sleep time by 48 minutes on average (SD 81). It caught 95% of sleep epochs and only 37% of wake epochs. Wrist trackers erred in the same direction. A meta-analysis of 24 studies found total sleep time differed from polysomnography by about 17 minutes. Neither device type scored sleep stages well.
How does an under-mattress sensor compare with a sleep lab?
It overcounts sleep. A 2025 study tested the Withings Sleep Analyzer against polysomnography in 183 participants across more than 400 nights. Some had sleep disorders, and some slept in the day. Two of its authors had held an investigator-initiated research grant supported by Withings.
Accuracy for sleep versus wake was 83%. Sensitivity for sleep was 95%. Specificity, the share of true wake it caught, was 37%.
On average it overestimated total sleep time by 48 minutes (SD 81), sleep efficiency by 9% and sleep-onset latency by 6 minutes. It underestimated time awake after sleep onset by 54 minutes.
Results were better in healthy people at night. Accuracy was 89% for healthy people against 81% for people with sleep disorders. Daytime sleep dropped specificity to 26%.
1 study
- In 183 participants over more than 400 nights, the Withings Sleep Analyzer had sleep-wake accuracy of 83%, sensitivity of 95% and specificity of 37%. It overestimated total sleep time by 48 minutes (SD 81) and underestimated wake after sleep onset by 54 minutes.
Does it do better at home?
It caught more waking time in one home study. A 2026 study put the same device under 117 healthy adults during home polysomnography, with no limits on caffeine, alcohol or schedule. Withings funded that study, and several of its authors work for Withings.
Sensitivity for sleep was 93% and for wake 73%. Overall sleep-wake accuracy was 87%. Performance was consistent across age, body mass index, mattress and bed-partner groups.
Stages were harder. Accuracy across stages was 63% (kappa 0.49), and light versus deep sleep was where it struggled most. It overestimated light sleep by 1 hour 21 minutes and underestimated deep sleep by 46 minutes and REM by 15 minutes.
The two studies differ in setting and people, so the 73% and 37% wake figures cannot be compared cleanly.
1 study
- In 117 healthy adults during home polysomnography, the Withings Sleep Analyzer had sensitivity of 93% for sleep and 73% for wake, sleep-wake accuracy of 87%, stage accuracy of 63% (kappa 0.49), and overestimated light sleep by 1 hour 21 minutes while underestimating deep sleep by 46 minutes.
How accurate are heart rate and breathing from under the mattress?
Closer than sleep-wake scoring. In a sleep lab, 35 adults aged 65 to 83 slept with two under-mattress trackers, the Withings Sleep Analyzer and the Emfit QS, and a bedside radar, Somnofy, measuring alongside the lab equipment for one night.
Heart rate from both under-mattress trackers was off by under 2.12 beats per minute on average against the lab ECG, under 5 percent, scored minute by minute. Breathing rate from all three devices was off by at most 1.6 breaths per minute, under 12 percent. The Withings snoring and apnea-hypopnea estimates also tracked the lab (r squared 0.76 and 0.59).
This was not a healthy-volunteer group: 57 percent had conditions such as type 2 diabetes, hypertension or obesity, and 49 percent had moderate to severe sleep apnea, so the result describes an older group living with common conditions.
HRV is a different question, and that study did not test it. No study cited here tests HRV from a consumer under-mattress tracker. The nearest is a bed sensor placed under the pillow of 22 healthy adults lying awake for 20 minutes: its RMSSD correlated closely with ECG (r 0.92) but carried a consistent offset, so its HRV figure was not interchangeable with an ECG reading (Heart and Vessels, 2025).
2 studies
- In 35 adults aged 65 to 83 over one lab night against polysomnography, the Withings Sleep Analyzer and Emfit QS estimated heart rate with a mean absolute error under 2.12 beats per minute (under 5%) and, with the Somnofy radar, breathing rate within 1.6 breaths per minute (under 12%) at 1-minute resolution; Withings snoring and apnea-hypopnea estimates correlated with polysomnography (r squared 0.76 and 0.59).
- In 22 healthy adults lying awake for 20 minutes, a ballistocardiography sensor under the pillow measured beat intervals in near-perfect agreement with ECG (r 0.999); RMSSD correlated at r 0.92 but with a fixed bias, and frequency-domain HRV agreed less well (r about 0.7).
How do wrist trackers err?
In a similar direction, by smaller amounts. A meta-analysis of 24 studies and 798 people covered Fitbit, Garmin, Whoop, Apple Watch and others. Total sleep time differed from polysomnography by about 17 minutes. Sleep efficiency, sleep latency and time awake also differed. Fitbit did not differ on time awake.
Fitbit Charge 2 and Alta HR were tested in 65 adults at a sleep clinic. Most had obstructive sleep apnea. The devices overestimated total sleep time and underestimated time awake after sleep onset and sleep-onset latency. The authors concluded sensitivity was acceptable and specificity poor.
A review of eight studies of recent devices found Whoop closest on total sleep time (-1.4 minutes) and deep sleep (-9.3 minutes) and furthest on REM (21.0 minutes). Fitbit Charge 4 was closest on REM (4.0 minutes). The review says research on these wearables remains limited. For more, see how accurate Fitbit, Garmin and Whoop sleep tracking is.
3 studies
- In 24 studies with 798 participants, consumer wrist devices differed significantly from polysomnography on total sleep time (mean difference -16.9 minutes), sleep efficiency, sleep latency and wake after sleep onset. Fitbit did not differ on wake after sleep onset.
- In 65 adults at a sleep clinic, Fitbit Charge 2 and Alta HR overestimated total sleep time and underestimated wake after sleep onset and sleep-onset latency against polysomnography.
- In 8 studies, Whoop differed from polysomnography by -1.4 minutes for total sleep time, -9.3 for deep sleep and 21.0 for REM. Fitbit Charge 4 differed by 4.0 minutes for REM, with sensitivity of 75% for deep sleep and 86.5% for REM.
What about rings, and telling sleep stages apart?
Rings varied. A study compared Oura, SleepOn and Circul rings against polysomnography in a university sleep lab with patients who had many kinds of conditions. Sleep versus wake accuracy was about 85% for Oura and SleepOn and about 65% for Circul. Stage accuracy across four stages was 53.18% for Oura, 50.48% for SleepOn and 35.06% for Circul.
Group averages looked better than individuals. Total sleep time differed by under 12 minutes for the Oura ring on average, yet individual differences often stayed large. The authors say this rules out clinical use.
A rapid review of 29 studies of consumer sleep devices in ambulatory settings agreed. Accuracy was moderate for total sleep time and time in bed. It was lower for sleep efficiency, wake after sleep onset and stages. REM and deep sleep estimates were particularly unreliable. For how stages are estimated, see how wearables track sleep stages.
What about smart mattresses like Eight Sleep and Sleep Number?
Smart mattresses put the same kind of sensor inside the bed instead of under it, and they have less published testing behind them. A review of nine smart mattresses judged them comfortable and passive but costlier than wearables, with limited clinical validation.
Sleep Number's bed is the most tested. In one night in a sleep lab with 45 adults aged 22 to 64, its overnight heart rate and breathing rate tracked the lab closely: mean overnight heart rate correlated at r = 0.94 with a bias of 0.15 beats per minute. It identified 94% of sleep but only 48% of time awake, the same pattern as the under-mattress sensors above.
A second Sleep Number study, in 104 adults, flagged moderate to severe sleep apnea (15 or more breathing events an hour) with 76% sensitivity and 85% specificity. Both studies were run by Sleep Number Labs, the company's own research group, and the apnea algorithm was developed by the authors who tested it. Neither has an independent replication.
Eight Sleep's tracking has no published validation. Its heart rate and breathing accuracy figures come from its own blog, and the independent trials of the Pod tested its cooling, not its sensors. The Eight Sleep page covers what those trials found.
3 studies
- In 45 adults over one night against polysomnography, the Sleep Number smart bed's mean overnight heart rate correlated at r = 0.94 (bias 0.15 beats per minute) and breathing rate at r = 0.96; sleep-versus-wake sensitivity was 0.94 and specificity 0.48.
- In 104 adults, an algorithm on the smart bed's ballistocardiography identified moderate to severe sleep apnea (AHI 15 or more) with 83.3% accuracy, 76% sensitivity and 85% specificity against polysomnography.
- Across nine smart mattresses, the review found passive, comfortable tracking at higher cost than wearables, with limited clinical validation.
What does this not settle?
None of these studies supports using a consumer device instead of polysomnography for diagnosis. Most validation was in healthy adults or a single sleep center. The rapid review found few studies in people with insomnia, depression or older adults.
Each result describes one device, one firmware and one population. An under-mattress result does not describe a ring or a watch, and the reverse.
The lab study enrolled 183 participants and the home study 117 healthy adults, both on one device, and both have ties to Withings, the home study through direct funding. The ring study used clinic patients from one university sleep lab. None establishes accuracy for a specific person's night, and a rapid review found few studies in older adults or people with insomnia or depression.
Common questions
How accurate is an under-mattress sleep sensor?
The Withings Sleep Analyzer had sleep-wake accuracy of 83% in a lab study of 183 participants, with 95% sensitivity for sleep and 37% specificity for wake. It overestimated total sleep time by 48 minutes on average.
Does an under-mattress sensor detect waking time?
Poorly in the lab study, with 37% specificity. In a home study of 117 healthy adults, wake sensitivity was 73%. The studies differ in setting and participants.
Are wrist trackers more accurate than under-mattress sensors?
Not clearly. A meta-analysis of 24 studies found wrist devices differed significantly from polysomnography on total sleep time, sleep efficiency, sleep latency and time awake, though by varying amounts by brand.
Can these devices score sleep stages?
Not well. The under-mattress device had 63% stage accuracy at home and misjudged deep sleep by 46 minutes. A rapid review of 29 studies called REM and deep sleep estimates particularly unreliable.
Are sleep rings more accurate?
It varies. In sleep clinic patients, Oura and SleepOn had about 85% sleep-wake accuracy and Circul about 65%. Individual-level differences from polysomnography often stayed large.
Are smart mattresses like Eight Sleep or Sleep Number accurate for sleep tracking?
Sleep Number is the only one with published sleep-lab testing, and the company ran it: in 45 adults its overnight heart rate and breathing rate tracked the lab closely, but it identified only 48% of time awake. Eight Sleep's sleep and heart tracking has no published validation.
How accurate is the Withings Sleep Analyzer?
In a sleep lab with 183 participants it told sleep from wake correctly 83% of the time, catching 95% of sleep but only 37% of waking time, and overstated total sleep by 48 minutes on average. At home in 117 healthy adults it caught 73% of waking time. Its heart rate and breathing rate came within about 2 beats and 1.6 breaths per minute of the lab in 35 older adults.
KnowYourPrime Evidence Registry: Finding KYP-ER-0008: Under-Mattress Sleep Sensor Accuracy · Finding KYP-ER-0003: Wearable HRV Accuracy Against ECG · Finding KYP-ER-0016: Wearable Sleep Stage Accuracy
Sources
- Performance evaluation of an under-mattress sleep sensor versus polysomnography in >400 nights with healthy and clinical participants. [conflict statement: two authors had a Withings-supported research grant]
- Evaluation of a Contactless Sleep Monitoring Device for Sleep Stage Detection at Home in a Healthy Population: Prospective Study in Free-Living Conditions [author affiliations include Withings]
- Performance of consumer wrist-worn sleep tracking devices compared to polysomnography: a meta-analysis.
- Validation of Fitbit Charge 2 and Fitbit Alta HR Against Polysomnography for Assessing Sleep in Adults With Obstructive Sleep Apnea.
- Accuracy of Fitbit Charge 4, Garmin Vivosmart 4, and WHOOP Versus Polysomnography: Systematic Review.
- Performance of wearable finger ring trackers for diagnostic sleep measurement in the clinical context.
- Real-World Use of Consumer Sleep Devices
- Performance Evaluation of a Smart Bed Technology against Polysomnography. [author affiliations include Sleep Number]
- Validating a smart bed against polysomnography for sleep apnea detection. [author affiliations include Sleep Number]
- A Comprehensive Review of Home Sleep Monitoring Technologies: Smartphone Apps, Smartwatches, and Smart Mattresses.
- Reliable Contactless Monitoring of Heart Rate, Breathing Rate, and Breathing Disturbance During Sleep in Aging: Digital Health Technology Evaluation Study
- Comparative analysis of heart rate variability indices from ballistocardiogram and electrocardiogram: a study on measurement agreement