What Your Sleep Tracker Can and Cannot Tell You

I like sleep trackers, which occasionally surprises patients, since sleep physicians are expected either to worship every new biometric or to confiscate the ring at the office door. The more defensible position sits between those. A wearable can reveal patterns that are genuinely difficult to see from memory: bedtime drift, short nights, irregular schedules, changes across travel or illness, and whether a new routine is moving sleep consistently in one direction. What it cannot do is convert a consumer sensor into an EEG, diagnose a sleep disorder from a proprietary score, or establish that your brain obtained precisely forty-seven minutes of deep sleep because the application drew a convincing purple bar.

The American Academy of Sleep Medicine has stated the distinction explicitly: consumer sleep technologies may be useful for patient engagement, but they should not be used to diagnose or treat sleep disorders without proper validation and clinical context.[1] A device can be useful without being the final authority, and most of the trouble I see comes from collapsing those two roles into one.

Most consumer wearables infer sleep from movement, heart rate, heart-rate variability, temperature, and sometimes oxygen-related measures. A laboratory polysomnogram measures a different set of signals altogether, including brain electrical activity, eye movements, muscle tone, respiratory effort, airflow, oxygen saturation, cardiac rhythm, and often leg movements; sleep stages are formally scored from physiologic signals that consumer devices generally do not record at all. That does not make the wearable fraudulent. It means the wearable is solving a different problem, and solving it by inference rather than by measurement.

A good algorithm can make a reasonable estimate of whether somebody is asleep from patterns in movement and cardiovascular physiology, and newer devices are markedly better than the pedometers with bedtime ambitions we had a decade ago. Inference, however, has predictable failure modes. Quiet wakefulness may be labeled sleep. Restless sleep may be labeled wake. Stages may be misclassified because several physiologic states produce overlapping peripheral signals, and different manufacturers can take similar raw data and generate different labels, since the scoring algorithms are proprietary and revised over time. This creates the strange situation in which your biology remains exactly the same while a software update improves your deep sleep. A tracker is therefore best understood as a model of your sleep rather than a recording of it.

Devices tend to perform better at detecting sleep than at detecting wake. In a study comparing seven consumer devices with polysomnography, sleep sensitivity was generally high, meaning the devices usually recognized epochs that truly were sleep, while wake specificity was considerably more variable and substantially lower, meaning quiet wake was frequently mistaken for sleep.[2] That asymmetry matters most in insomnia. Someone who lies motionless and entirely conscious for forty-five minutes may be congratulated by a movement-based algorithm on forty minutes of sleep, and then finds themselves arguing with a wristwatch about whether they were awake. I generally side with the conscious human on that particular question.

Stage estimates require more humility still. Some newer platforms have demonstrated real improvement against ambulatory polysomnography; a multi-night validation of the Oura Ring Generation 3 found reasonably strong performance for several sleep-wake and staging measures in generally healthy adults,[3] a study supported by Oura, which belongs in the weighing rather than in a footnote nobody reads. Even good validation does not mean every stage estimate is correct for every person on every night, since performance can shift with disrupted sleep, unusual movement, arrhythmia, skin contact, illness, and the population being measured. The right response is not to ignore stage data entirely. It is to stop treating it as laboratory-grade fact.

The most useful information from a wearable is usually boring, which I intend as a compliment. If bedtime has drifted ninety minutes later over a month, that is probably real and probably actionable. If you average six hours on work nights and seven and a half on weekends, that is worth knowing. If resting heart rate rises during an illness and returns to baseline in recovery, the trend helps contextualize how you feel. What matters much less is whether Tuesday scored 71 and Wednesday scored 76. Proprietary sleep scores combine multiple inputs through formulas the company designed, and the clinical meaning of a five-point difference is generally not established. The score looks precise because software is very good at printing integers; precision of display and precision of measurement are unrelated properties.

Multi-night trends also insulate you from the ordinary variability of human sleep. A late dinner, exercise, menstrual symptoms, alcohol, stress, travel, a barking dog, or simple chance will move the numbers, and one unusual night should rarely drive a decision. If you are testing a change, compare enough nights on either side that the signal has a chance to separate itself from the noise, which is closer to how we evaluate anything else in medicine.

A wearable may indicate that sleep looks fragmented. It cannot tell you why. Obstructive sleep apnea, restless legs syndrome, periodic limb movements, chronic insomnia, perimenopausal hot flashes, pain, alcohol, reflux, circadian misalignment, medication effects, and environmental disruption all produce fragmented sleep, and all require different treatment. A device may also miss clinically important disease entirely, which is why the AASM position statement warns specifically against using consumer sleep technology for diagnosis.[1] If there is loud snoring, witnessed apnea, gasping, excessive daytime sleepiness, drowsy driving, persistent insomnia, morning headache, or unexplained fatigue, a reassuring score does not close the case; conversely, an alarming score in an otherwise well-rested person does not open one. The watch has not met you. It does not know that your child was sick, that you started an antidepressant last month, or that your partner has been lying awake watching you stop breathing.

There is also a point at which useful curiosity becomes performance anxiety. Researchers have described orthosomnia, a pattern in which people become preoccupied with achieving perfect wearable sleep data and grow more anxious about sleep in the process.[4] I see versions of it regularly. A patient wakes feeling reasonably well, checks the application, learns that deep sleep was allegedly poor, and retroactively decides they feel terrible. Another stays in bed an extra hour attempting to improve a recovery score. A third changes supplements every three nights because a graph moved in the wrong direction. This is the opposite of what the technology is for, and people with insomnia are particularly vulnerable, since the disorder already involves excessive attention and effort directed at sleep. More data are not automatically more therapeutic. If the tracker leaves you better informed and less anxious, keep it; if it has turned bedtime into an examination you can fail, take it off for a week or hide the score.

Sleep trackers are genuinely useful when asked modest questions. When am I sleeping, how regular is my schedule, is there a sustained trend, did something change across several nights. They become unreliable when asked to settle questions they were never built to settle: whether you have sleep apnea, whether you obtained enough REM, whether your insomnia has resolved, whether last night was medically good because a number said 84. If the device and your experience disagree occasionally, that is normal. If they disagree persistently and you have significant symptoms, the answer is not to buy a second tracker so they can vote. It is to have the human being evaluated.

[1] Khosla S, Deak MC, Gault D, et al. Consumer sleep technology: an American Academy of Sleep Medicine position statement. J Clin Sleep Med. 2018;14(5):877-880. doi:10.5664/jcsm.7128. PMID: 29734997.

[2] Chinoy ED, Cuellar JA, Huwa KE, et al. Performance of seven consumer sleep-tracking devices compared with polysomnography. Sleep. 2021;44(5). doi:10.1093/sleep/zsaa291. PMID: 33378539.

[3] Svensson T, Madhawa K, Nt H, Chung UI, Svensson AK. Validity and reliability of the Oura Ring Generation 3 with Oura sleep staging algorithm 2.0 when compared to multi-night ambulatory polysomnography: a validation study of 96 participants and 421,045 epochs. Sleep Med. 2024;115:251-263. doi:10.1016/j.sleep.2024.01.020. PMID: 38382312.

[4] Baron KG, Abbott S, Jao N, Manalo N, Mullen R. Orthosomnia: are some patients taking the quantified self too far? J Clin Sleep Med. 2017;13(2):351-354. doi:10.5664/jcsm.6472. PMID: 27855740.