Sleep Trackers: Which Metrics to Ignore

Your wrist says you logged 42 minutes of deep sleep last night. It probably doesn't know. Consumer sleep trackers estimate sleep from movement and heart rate, and validation studies show they reliably overestimate total sleep time while guessing at sleep stages with, at best, moderate accuracy. That matters because roughly one in five US adults regularly wears a smartwatch or fitness tracker, according to Pew Research Center, and CDC surveillance data show about one in three American adults sleeps less than seven hours a night. Widespread anxiety about sleep, meeting widespread trust in unvalidated numbers — that is the problem this article addresses.

Why the numbers look so precise

Lab-based polysomnography (PSG) defines sleep stages by measuring brain waves, eye movements and muscle tone. That is the gold standard, and it is how "deep sleep" and "REM" are actually defined.

Consumer wearables measure none of that. They combine an accelerometer with photoplethysmography — an optical sensor that estimates heart rate and heart-rate variability — and run the output through proprietary algorithms to produce an estimate of what your brain was probably doing.

The gap shows up clearly in validation research. In a 2021 study by Chinoy and colleagues, working with the Walter Reed Army Institute of Research sleep team, four popular wearables were compared against polysomnography in people sleeping at home. Most devices detected sleep with sensitivity above 90% — they were good at confirming you slept. But specificity for detecting wakefulness was poor, in some cases between roughly 20% and 50%, meaning quiet wakefulness was routinely logged as sleep. Agreement on individual sleep stages was moderate at best. Systematic reviews led by sleep researcher Massimiliano de Zambotti reach the same conclusion: total sleep time is inflated, wake after sleep onset is underestimated, and stage-level data should be read with caution.

Why it matters

The stakes are not abstract. In 2017, researchers at Northwestern University published a case series in the Journal of Clinical Sleep Medicine describing "orthosomnia" — patients whose pursuit of perfect sleep-tracker scores was itself disrupting sleep. Some spent months adjusting routines over stage data of unknown accuracy; one declined a clinician's recommendations because her tracker indicated she slept fine.

There are quieter costs too. People skip evening exercise because a readiness score dipped — even though that score can be dragged down by dehydration, alcohol the night before, or a late meal. Others worry over a "low REM" night the device may have misclassified by 30 to 60 minutes. And a small number of watches now carry FDA-authorized features that flag signs suggestive of sleep apnea. Useful as a prompt to see a doctor — but screening is not diagnosis, and a clean night on a wrist sensor does not rule out a breathing disorder.

The metrics to ignore

1. Exact sleep-stage percentages

The minutes of "deep" and "REM" sleep are the least validated numbers on the device. They are inferred, not measured, the algorithms are trade secrets, and two trackers worn on the same wrist can disagree by more than half an hour. Treat them as a rough sketch, never a lab result.

2. Proprietary "sleep scores"

A single figure like 78/100 blends duration, stages, heart rate and movement in proportions the manufacturer does not publish. No independent body has validated these composites, and scores are not comparable across brands. A 92 tonight and an 84 tomorrow may mean nothing at all.

3. HRV-based "readiness" or "recovery" ratings

Heart-rate variability genuinely responds to stress, alcohol, illness and training load — which is exactly why it is noisy. No consumer readiness score has been shown to predict health outcomes, and using one to cancel plans or push through fatigue means acting on an unvalidated signal.

4. Minute-precise totals and wake times

Because stillness is read as sleep, devices systematically overestimate total sleep time and underestimate night-time waking — worst for people with insomnia, who lie still and awake. If you remember being awake at 3 a.m., trust your memory over the graph.

What to watch instead

Strip the dashboard down to three things. First, consistency: a 2024 analysis of more than 60,000 UK Biobank participants by Windred and colleagues found that sleep regularity predicted mortality risk more strongly than sleep duration did — and bedtime consistency is the metric your tracker measures best. Second, the weekly trend in duration: a rolling average over two to four weeks is far more meaningful than any single night, and public-health guidance for adults centres on seven or more hours, not on stage percentages. Third, how you function: daytime alertness, mood and reliance on caffeine are outcome measures no wearable captures, and they should anchor your judgement.

The honest counterpoint

The strongest objection to this argument is that calling tracker data unreliable makes the devices sound useless, when the same validation studies show high sensitivity for sleep detection — wearables are genuinely decent at confirming you slept, roughly how long, and whether your schedule is drifting. Self-monitoring is also a proven behavioural tool: people who can see a pattern are more likely to correct it.

That objection is fair, and it is why the answer is not "throw the tracker away" but "change which numbers you read." The failure mode is not the sensor; it is treating an estimate with the confidence of a measurement. Our view: the device is a mirror for habits, not a laboratory on your wrist. If checking the score makes sleep more anxious rather than less, that is the clearest signal of all — a reason to stop opening the app, and a reason to talk to a clinician about insomnia rather than buy another gadget.

Sleep disorders are diagnosed in sleep labs, not apps. If you snore loudly, gasp during the night, or feel exhausted despite adequate time in bed, the right next step is a professional evaluation — not a better score.