Under-Mattress Sleep Trackers vs Wearables: What the Validation Data Actually Shows

This article covers validation research comparing under-mattress sleep sensors and wrist-worn wearables against polysomnography (PSG). It does not cover ring trackers in depth or make purchase recommendations; it reports what accuracy studies found and where they stop.

Under-mattress sensors and wrist-worn wearables fail in different, fairly predictable ways when checked against polysomnography. One large under-mattress sensor study found it caught actual sleep well but struggled to correctly flag wakefulness, while wrist-worn devices across multiple studies tend to overestimate total sleep time and underestimate the time spent awake after falling asleep. Neither pattern means one device category is simply 'more accurate' than the other across the board; the errors just show up in different places.

The question people are actually asking

If you have ever compared notes between a mattress sensor and a wrist tracker on the same night and gotten two different sleep scores, you already know the frustration here. Which number, if either, reflects what actually happened while you were asleep? The devices are measuring the same night and disagreeing, sometimes by a lot, and there is no obvious way to know which one is closer to the truth without a sleep lab in your bedroom.

The tension is not really about which brand wins. It is about whether contactless, under-mattress sensing is fundamentally more or less trustworthy than the accelerometer-and-heart-rate approach used on the wrist, or whether both are just approximating sleep in their own flawed ways.

What the under-mattress data specifically shows

The largest under-mattress validation study in this evidence set tested the Withings Sleep Analyser across more than 400 nights, deliberately including people with and without sleep disorders and both day and night sleep opportunities. The pattern that emerged was consistent: the sensor was very good at recognizing sleep when it happened (95% sensitivity), but much weaker at correctly flagging wakefulness (37% specificity). That asymmetry is basically the whole explanation for why these devices tend to overestimate total sleep time, sleep efficiency, and sleep-onset latency compared to a lab-based reading.

A separate prospective study of the same category of device, this one in a home setting with 117 healthy volunteers under free-living conditions, evaluated sleep-wake distinction and stage identification without imposing a fixed sleep schedule or restricting substances beforehand. That real-world framing matters, it's testing the device the way people actually use it, not in a controlled lab. But the population studied was healthy adults, not people with diagnosed sleep disorders, which limits how far the home-setting findings can be extended.

Where wrist-worn wearables drift, and why it looks different

Wrist-worn devices don't show the same wake-detection weakness in isolation, they show a broader pattern across several sleep measures at once. The meta-analysis pooling 24 studies and nearly 800 participants using devices including Fitbit, Garmin, WHOOP, and several others found statistically significant differences from polysomnography in total sleep time, sleep efficiency, sleep latency, and wake after sleep onset, with wearables tending to overestimate sleep time and efficiency while underestimating wake time. It's a similar direction of error to what shows up in the under-mattress data, but arrived at through a different sensing approach: accelerometer motion plus heart rate rather than pressure and movement sensing from below.

A more specific validation looked at Fitbit Charge 2 and Alta HR in adults with obstructive sleep apnea and found significant differences from polysomnography on every sleep outcome except REM sleep, with devices overestimating total sleep time and underestimating both wake after sleep onset and sleep onset latency, even after a correction was applied for sleep-start detection bias. That's a clinical population, and it matters. A systematic review focused specifically on newer-generation devices (Fitbit Charge 4, Garmin Vivosmart 4, WHOOP) also set out to appraise their sleep-stage and sleep-parameter accuracy against polysomnography or ambulatory EEG, underscoring that even within the wrist-worn category, performance gets evaluated model by model rather than assumed to be uniform. For a closer look at how these specific brands stack up against each other, that comparison lives in a dedicated look at Fitbit, Garmin, and WHOOP accuracy.

The harder problem: sleep stages, not just sleep-or-awake

Distinguishing sleep from wake is one problem. Telling REM from deep from light sleep is a much harder one, and the evidence here is more cautious across both device categories. A rapid review pulling together 29 studies of consumer sleep technologies in ambulatory settings found moderate accuracy for total sleep time and time in bed, but lower precision for sleep efficiency, wake after sleep onset, and stage classification, specifically flagging REM and deep sleep estimates as particularly unreliable. It also noted that multi-sensor devices modestly improved basic sleep-wake detection and that contactless devices reduced user burden but came with limited portability, and found that few studies had actually included clinical populations like insomnia, depression, or older adults.

There is active work trying to close that stage-detection gap. One approach applied a neural network incorporating topological features of wearable data alongside model-driven circadian timing estimates to Apple Watch motion and heart rate data, and found meaningful improvement in classifying wake, REM, and NREM sleep compared to a version of the same network using raw data alone, an improvement attributed mainly to the heart rate-derived features. It's a promising signal for where algorithms might be headed. Not proof that current consumer devices on shelves now perform at that level. For more on how the underlying sensors attempt this in the first place, how wearables actually track sleep stages covers the mechanics in more depth.

What none of this settles

None of the studies in this evidence base position either device category as a diagnostic substitute for polysomnography, and the clinical scoring rules that define things like apnea and hypopnea events remain the domain of formally trained sleep technicians reading lab-grade signals, not consumer sensors. Ring trackers, a growing third category, show their own pattern. One clinical-population study of Oura, SleepOn, and Circul found average group-level total sleep time differences under 12 minutes for at least one of the rings, even while individual-level discrepancies were more pronounced, a reminder that group averages can look reassuring while still masking meaningful night-to-night or person-to-person error.

What the research does consistently support is that both under-mattress sensors and wrist wearables are picking up something real. Sleep and wake states correlate with movement, heart rate, and pressure changes for a reason. But the size and direction of their errors are specific enough that treating either device's nightly readout as clinically precise outruns what the validation studies actually found.

Common questions