REF-1698 · Agostinho et al., 2026 · Behavioral Sleep Medicine, · Paper · PubMed

Wearable sleep trackers read more sleep and less night-time wake than a laboratory sleep test (polysomnography) in healthy adults, and no device was consistently best

A meta-analysis of 16 studies, 10 of them pooled; the abstract gives no sizes for these differences

Main caveat: The authors': variable device performance, methodological heterogeneity and risk-of-bias concerns limit use as a diagnostic tool or for sleep stages.
moderate evidenceMeasured on non-athletes to trained athletesRead from the abstract only

Agostinho and colleagues asked how consumer sleep-tracking wearables compare with polysomnography, the laboratory reference, in healthy adults.

They searched five databases, included 16 studies in which the wearable and polysomnography recorded the same nights, and pooled 10 of them for total sleep time, time to fall asleep, sleep efficiency, wake after sleep onset and the time spent in light, deep and REM sleep, appraising each study with QUADAS-2.

Across devices, wearables over-estimated total sleep time and sleep efficiency and under-estimated wake after sleep onset, and devices differed widely; none was consistently better than the others. In single studies, the third-generation Oura Ring came closest to polysomnography for sleep latency and sleep efficiency, and some Fitbit models for deep and REM sleep.

The authors conclude that wearables give reasonable estimates of overall sleep measures and can support population monitoring and long-term self-tracking, but are not stand-alone diagnostic tools and are not suited to detailed sleep-stage description, given variable performance, heterogeneous methods and risk of bias.

Participants

16 studies (10 pooled) of healthy adults with simultaneous wearable and polysomnography recordings.
Number of participants, age, sex, number of nights per person, devices and their generations: not in the abstract.

Protocol

Search: PubMed, Scopus, SciELO, Web of Science, Cochrane Library, following PRISMA; search dates not in the abstract.
Inclusion: healthy adults, wearable and polysomnography recorded simultaneously.
Outcomes: mean differences from polysomnography in sleep efficiency, wake after sleep onset, total sleep time and time to fall asleep, and in minutes of light (N1+N2), deep (N3) and REM sleep.
Analysis: fixed or random effects depending on heterogeneity; p < .01 as the significance level.
Appraisal: QUADAS-2 for risk of bias and applicability; its findings are not reported in the abstract.

Key Results

Measure (wearable vs polysomnography)Pooled directionSize
Total sleep timeover-estimatednot in the abstract
Sleep efficiencyover-estimatednot in the abstract
Wake after sleep onsetunder-estimatednot in the abstract
Sleep latencynot stated in the abstractnot in the abstract
Light, deep and REM sleep timenot stated in the abstractnot in the abstract
Devices: substantial variability between devices; no device consistently superior.
Closest single-study agreement: Oura Ring Gen 3 for sleep latency and sleep efficiency; selected Fitbit models for deep and REM sleep (study-level observations, not pooled comparisons).
Heterogeneity, confidence intervals and the QUADAS-2 results: not in the abstract.

Practical Application

Read a tracker's total sleep as probably generous and its night-time wake as probably too low; use it for the trend on one device, not as an exact number. In healthy adults in the laboratory, wearables as a group read more sleep and less wake than polysomnography, with wide differences between devices; the abstract does not say by how much.
This agrees with [SYNTH-91]: sleep duration and timing as a trend on one device, never an exact total or the wake count, the same direction as the pooled 24-study analysis in REF-1252; it adds a second pooled analysis but no numbers; how many of its studies are also in REF-1252 is not known from the abstract.
Stage minutes stay context. The authors judge wearables unfit for detailed sleep-stage description, even though some Fitbit models came closest for deep and REM sleep in single studies; this is not a reason to treat any device's "deep sleep" as a target (SYNTH-91, SYNTH-07).
No brand wins across the board. The Oura Ring Gen 3 result (latency, efficiency) comes from single studies; inference, not tested here: a newer generation or another brand is not covered by it, so a new device starts a new baseline (SYNTH-91).
When the tracker and the athlete disagree about a night, believe the athlete (SYNTH-91's practice); a wearable is not a sleep-disorder test, and suspected insomnia or sleep apnoea goes to a physician.

Limitations

The authors': variable device performance, methodological heterogeneity and risk-of-bias concerns limit use as a diagnostic tool or for sleep stages.
Only 10 of 16 studies could be pooled; healthy adults in a laboratory, so home sleep, athletes and sleep disorders are not covered.
From the abstract alone, unknown: the size of every difference and its interval, heterogeneity, the number of participants and nights, the devices and firmware tested, the search dates, the QUADAS-2 judgements, and funding.
The device-specific statements (Oura Gen 3, Fitbit) come from individual studies, not from a pooled comparison between devices.
Written from the abstract and the PubMed record; full text not read.
How this entry was checked and changed
Why this level

risk of bias not read (abstract only): the abstract names QUADAS-2 but reports its result only as "risk-of-bias considerations" limiting use, and gives no pooled values, intervals, heterogeneity or participant numbers