Apple Watch sleep apnea detection: accuracy and what to do

Yes, within limits worth knowing. The Sleep Apnea Notification Feature watches your breathing for 30 nights at a time and warns you when it sees consistent signs of moderate to severe sleep apnea. In Apple's prospective validation, submitted to the United States Food and Drug Administration (FDA), it notified 89.1% of participants with severe apnea and 43.4% of those with moderate apnea, a weighted sensitivity of 66.3%¹². It notified nobody whose home sleep test was normal and about 9% of people with mild apnea, a weighted specificity of 98.5%¹. In plain terms: a notification almost always points at real apnea, but silence misses about one moderate case in two.

The FDA cleared the feature on 13 September 2024 as an over-the-counter device to assess risk of sleep apnea; its labeling states that it is not intended to diagnose, treat or aid in the management of the condition². A notification is an invitation to get tested, not a diagnosis, and the labeling adds that the absence of a notification is not intended to indicate the absence of sleep apnea².

Everything below describes the feature as it stands in October 2026. watchOS 26, watchOS 27, Apple Watch Series 11 and Series 12 brought no documented change to it, Apple's validation paper has not been updated since September 2024, and we found no independent peer-reviewed study of Apple's feature in PubMed or Europe PMC through October 2026¹⁶. The accuracy figures are Apple's own, reviewed by a regulator; where independent data exist for a comparable feature, we show them.

Sleep apnea notifications work on Apple Watch Series 9 or later, Apple Watch Ultra 2 or later, and Apple Watch SE 3, with the latest version of watchOS and an iPhone on the latest iOS³⁶. The feature launched in September 2024 on Series 9, Series 10 and Ultra 2⁴; Apple Watch SE 3 gained it when it shipped in September 2025⁵; Series 12 and Ultra 4, announced in September 2026, inherit it unchanged⁶. Apple Watch SE (2nd generation), Series 8 and every earlier model are not on Apple's list.

Four other conditions apply. Sleep must be set up with Track Sleep with Apple Watch turned on; you must wear the watch to sleep for at least 10 nights in a 30-day period; the feature is intended for people 18 or older who have not been diagnosed with sleep apnea; and your region must be among the more than 150 countries and regions where Apple offers it³⁴. You turn it on in the Health app on iPhone, confirming your date of birth and whether you have ever been diagnosed³.

Apple Watch does not use its blood oxygen sensor for this. The same accelerometer that feeds the sleep stage estimates picks up the small wrist movements that come with breathing, and a deep learning model trained on nights recorded alongside polysomnography or home sleep apnea tests turns them into a nightly Breathing Disturbances value in events per hour¹². The FDA summary defines breathing disturbances as transient changes in breathing patterns, such as temporary breathing interruptions², and Apple is explicit that the value is not equivalent to the apnea-hypopnea index (AHI) a sleep laboratory reports¹.

Each night's value is filed in the Health app as Elevated or Not Elevated, viewable over the past month, six months or year³. A few disturbances a night are normal, and alcohol, a respiratory infection, some medications and sleep position can push a night into the Elevated range without any apnea³⁴.

The notification sits on top of those nightly values and follows a published rule. The algorithm looks at fixed, non-rolling 30-day windows, starting 30 days after setup. If a window contains at least 10 nights with a value, it checks whether at least 50% of them were Elevated; if so, you get a notification, otherwise it stays silent and tries again in the next window¹. With fewer than 10 nights it also stays silent¹. So the first possible notification comes about a month after setup, and one bad night, or even a bad week, cannot trigger it.

Apple built the algorithm on 3,936 nights from 2,160 people and tested it on a sequestered set of 7,220 nights from 2,542 people the model had never seen; on that test set the notification reached 66.6% sensitivity and 95.9% specificity¹. Both sets were scored with the strictest clinical convention, which requires a 4% oxygen desaturation to count a hypopnea¹.

The validation that went to regulators was a separate prospective study at several US sites. It enrolled 1,499 adults, 1,448 of whom completed it and 1,278 of whom contributed to the notification analysis; participants wore the watch for up to 30 nights and did at least two nights with a Nox T3s home sleep apnea test, scored by two blinded technologists, with the higher of the two nights as ground truth¹². The cohort was enriched on purpose: 559 normal, 362 mild, 216 moderate and 201 severe participants by AHI¹².

Table 3 of the white paper is the part worth remembering. The notification fired for 164 of 184 people with severe apnea (89.1%) and 89 of 205 with moderate apnea (43.4%); it fired for none of the 543 people with a normal home test and for 31 of 346 with mild apnea, a specificity of 100% and 91.0% respectively¹. Apple weighted these to approximate the real population, moderate and severe one to one, normal and mild five to one, which gives the headline 66.3% sensitivity and 98.5% specificity¹. Both were comparable across age, sex, race, ethnicity and body mass index¹.

The trade-off was a choice. Apple writes that the operating point was intentionally chosen to favor specificity, because a feature that re-checks you every month would otherwise accumulate false alarms¹². The price is sensitivity: more than half of moderate cases go unflagged in any given window, although the algorithm runs again in every later 30-day window; Apple has published no data on how often a later window catches a case the first one missed. A secondary analysis found 91.4% of participants' nightly values inside Apple's own tolerance funnel, a tolerance that widens as the true AHI rises, so it is not a conventional agreement statistic¹.

As of October 2026, no. Searching PubMed and Europe PMC for Apple Watch with sleep apnea, breathing disturbances or notification from September 2024 onward returns no peer-reviewed test of Apple's feature against polysomnography or a home sleep test. The only Apple Watch apnea paper in that window is a 2025 pilot in Sleep and Breathing: 61 patients at one clinic in Nagoya wore an Apple Watch SE while the authors trained their own model on the raw motion data¹⁴. It correlated with PSG-derived AHI at r = 0.93 and picked out AHI ≥ 15 with 100% sensitivity and 90% specificity in a 20-person test group, but it was validated only internally, the cohort had a 50% pretest probability of moderate to severe apnea, and Sony funded the work¹⁴. Blogs that cite it as proof that Apple Watch detects apnea almost perfectly are describing a different algorithm.

The nearest thing to independent evidence comes from the other FDA-authorized smartwatch feature. Samsung's Sleep Apnea Feature on Galaxy Watch, granted De Novo authorization on 6 February 2024, uses blood oxygen data over two chosen nights rather than motion data over 30, and its FDA summary reported 82.7% sensitivity and 87.7% specificity²¹⁷. In 2026 a Stanford team tested it against two nights of in-lab polysomnography in 147 adults who already had, or were likely to have, moderate to severe apnea: the area under the curve was 0.94, and at the default threshold the watch caught 94.1% of cases but cleared only 66.7% of the rest; a higher threshold flipped that to 82.4% and 94.9%¹¹. The commentary published with it applies to any wrist-worn screener: an enriched cohort inflates predictive values, false positives would be substantially higher at population prevalence, 38.8% of watch recordings were unusable during attended polysomnography, and confirmatory testing remains necessary before treatment¹². Samsung's second version was cleared on 20 July 2026, so the FDA has now issued three authorizations for such features, to two companies¹⁷¹⁸.

For scale, a 2025 meta-analysis of 59 studies pooled wearable devices of all kinds at 0.87 sensitivity and 0.86 specificity for AHI above 15 and concluded that their role is to screen and complement polysomnography, not replace it¹³. Apple's feature sits at one end of that trade-off, lower on sensitivity and higher on specificity: the right place for an alert that arrives unprompted, the wrong place for anyone who takes silence as reassurance. Our comparison of watches, rings and bedside sensors covers what other form factors can measure.

If you received a notification. In Apple's validation, every participant the algorithm flagged had at least mild sleep apnea, so the alert is a strong signal that something is there¹⁴. What it cannot tell you is the severity, whether the apnea is obstructive or central, or anything about your symptoms; the American Academy of Sleep Medicine (AASM) notes that most consumer devices cannot separate obstructive sleep apnea from other breathing problems such as central apnea, and that none can diagnose sleep apnea or confirm its absence⁷. A positive result from a self-screening device does not establish a diagnosis; further confirmatory evaluation is needed¹⁰.

If you never received one. Silence is weak evidence. The feature missed 57% of moderate cases in the validation, Apple states that not all people with sleep apnea receive a notification, and the FDA labeling says the absence of a notification is not intended to indicate the absence of sleep apnea¹²³. The algorithm also ignores symptoms¹. If you snore loudly, someone has seen you stop breathing, you fall asleep during the day, or you wake gasping or choking, those are the signs clinicians act on, with or without a watch⁹. Risk is higher in men, after menopause, between 40 and 70, with a higher body mass index, and with anatomy that narrows the upper airway⁹.

Sleep physicians writing in the Journal of Clinical Sleep Medicine in 2025 named the two ways these alerts go wrong: a symptomatic person reassured by a false negative, and an asymptomatic person who tests falsely positive and goes through anxiety, testing and costs¹⁰. Apple's threshold makes the second rare and the first common, so the symptom list above matters more than your notification history.

The first four steps follow Apple's instructions and the AASM's guidance; the fifth is our suggestion. In order:

1. Export the report. In the Health app, open Breathing Disturbances, scroll down and tap Export PDF. The PDF shows when sleep apnea may have occurred and three months of breathing disturbance data, and it is what Apple intends you to bring to a doctor³⁴.

2. See a clinician, not a store. The AASM's advice for anyone who thinks they may have sleep apnea is to talk to a health care professional⁷. Expect a comprehensive sleep evaluation: history, symptoms, risk factors, medications and an examination⁸¹⁰.

3. Expect a real test. Diagnosis requires polysomnography or a technically adequate home sleep apnea test; questionnaires and prediction algorithms must not be used to diagnose obstructive sleep apnea without one, and a watch alert is no exception⁸. If a single home test is negative or inconclusive and suspicion remains, the guideline calls for polysomnography⁸.

4. Do not self-treat. Positive airway pressure and mandibular advancement devices improve AHI, sleepiness and blood pressure in people who have been evaluated and referred⁹; the test tells you whether, and how severely, you need them.

5. Keep wearing the watch unchanged until the appointment, so the three-month report is complete. A diagnosis ends the feature's job: Apple and the FDA state that it is neither intended nor cleared for people already diagnosed¹².

Sensitivity is the share of people with moderate to severe apnea (AHI of 15 or more) who were flagged; specificity is the share of people below that line who were not. Reference tests and populations differ from row to row, so the rows are not directly comparable.

Independent accuracy. Nobody outside Apple has published a test of the shipped feature, so the 66.3% and 98.5% rest on one large company-run study¹. The FDA reviewed the data and we found no correction or retraction, but replication in an ordinary population is missing, and population matters: moderate to severe obstructive sleep apnea affects an estimated 13% of men and 6% of women aged 30 to 70 in the US, and 425 million adults aged 30 to 69 worldwide⁹¹⁵.

Newer watches. The white paper predates Apple Watch SE 3 and still says the feature excludes Apple Watch SE; Apple has published no model-specific accuracy data for SE 3, Series 11 or Series 12¹⁵⁶.

Central apnea, other conditions and outcomes. The AASM notes that most consumer devices cannot distinguish obstructive from central apnea⁷; the independent smartwatch study cited here excluded people with significant heart or lung disease, movement disorders and other sleep disorders¹²; and the USPSTF found insufficient evidence that screening adults without symptoms improves health, a gap monthly notifications have not yet filled⁹. The often-quoted claim that up to 80% of cases are undiagnosed, which Apple repeats, is usually traced to a 1997 Wisconsin cohort in which 82% of men and 93% of women with moderate to severe apnea had never been diagnosed¹¹⁶.

Moon has no sleep apnea detection. It does not read Breathing Disturbances, and nothing in a Moon report is a verdict on your breathing. What Moon does with the data your Apple Watch writes to Apple Health is show sleep duration, stages, continuity and regularity as trends over weeks, in the spirit of how to stop over-analyzing Apple Watch sleep data.

One pattern is still worth carrying to the same appointment. Nights that are persistently fragmented, with many short awakenings and poor continuity over weeks rather than days, are the kind of history a sleep clinician asks about. It is not an apnea measure, and the stage estimates behind it are the weakest part of what the Watch reports. Bring it as context for a conversation that the notification, or your symptoms, should already have started.