Why Sleep Trackers Measure Your Breathing Better Than Your Sleep

Every sleep wearable reports two broad things: how you slept, and how you breathed. People read the first and ignore the second.

The validation literature says that is exactly backwards.

What a proper validation looks like

A 2026 external validation study tested a fingertip wearable against full polysomnography at the Duke Sleep Disorders Center — 198 participants, with a diagnostic cohort of 147 and a split-night titration cohort of 29. External validation matters: it means the device was tested on a population it was not developed on, which is the step most consumer sleep tech never publishes.

The results split cleanly by what was being measured.

Breathing events — strong agreement. The device's respiratory event index correlated with the lab's apnea-hypopnea index at 0.879 under one scoring rule and 0.910 under the other. It also tracked changes over the night well enough to follow pressure titration, with a median correlation of around 0.88.

Sleep staging — moderate agreement. Epoch-by-epoch agreement on wake, REM and non-REM gave a Cohen's kappa of 0.567. Total sleep time correlated at 0.786 and REM duration at 0.655, with the device underestimating both by roughly nine and a half minutes on average. Its best stage was non-REM, at a sensitivity of 0.824.

A kappa of 0.567 is conventionally read as moderate. For context, that means that on a decent proportion of 30-second windows the device and the laboratory disagreed about what stage you were in.

Why the gap exists

It is not a quality problem, it is a physics problem.

Breathing produces strong, direct, peripheral signals. Airflow, effort, oxygen saturation and pulse-rate variation all change markedly when you stop breathing. A desaturation and the arousal that follows are large events that a wrist, finger or chest sensor can genuinely detect.

Sleep stages are defined by brain electrical activity. The scoring rules are built on EEG, plus eye movements and chin muscle tone. No wrist, ring or fingertip device measures any of those. They infer stages from movement, heart rate and heart rate variability — which correlate with sleep stages without being them.

So when your ring tells you that you got 48 minutes of deep sleep, it is producing an estimate from proxies, by a proprietary algorithm, against a definition it cannot observe.

Evidence tier 1 that validated wearables agree closely with polysomnography on respiratory event counts. Tier 2 at best for consumer sleep staging, and tier 3 for any specific stage duration quoted to the minute.

The thing most people get wrong

Use the breathing number, question the sleep number. If your device reports a breathing-disturbance metric, oxygen variation or a snoring summary, that is the part with the better measurement basis and the part worth acting on.

Treat stage durations as relative, not absolute. Comparing your deep sleep across a month on the same device is a reasonable use. Concluding you have a deep sleep deficit from the absolute number is not.

Note the difference between device classes. The device in that study is a clinical Type IV recorder used in a sleep centre. A consumer ring or watch is not that, and its respiratory metrics are generally wellness features rather than validated diagnostics — which is why only a few consumer devices carry regulatory clearance for apnea notification at all. Which consumer devices are actually cleared, and what the Apple Watch does and does not do.

And a wearable is not a diagnosis. Even strong AHI agreement in a validation cohort does not make a consumer device a sleep study, and nothing it reports rules apnea in or out. A flagged breathing disturbance is a reason to get tested. What a home sleep apnea test involves.

If what you actually want is to know whether and how you snore, a phone recording outperforms an expensive wrist device at that specific job, because sound is the thing being measured rather than inferred. What snoring apps can and cannot do, and how to record properly.