The Apple Watch sleep-apnea alert finds severe cases well and misses most moderate ones

The Apple Watch's sleep-apnea notification, cleared by the FDA in September 2024, does not measure how many times a sleeper stops breathing and does not diagnose sleep apnea. It uses the watch's accelerometer to track small movements linked to disrupted breathing, aggregates them into a "breathing disturbances" metric, and — over a month of data — flags a wearer whose pattern is consistent with moderate-to-severe obstructive sleep apnea [s1]. In Apple's own validation, how well it worked depended heavily on how severe the apnea was.

How the feature works

The FDA cleared the Sleep Apnea Notification Feature as a Class II over-the-counter mobile medical application, intended "to identify patterns of breathing disturbances suggestive of moderate-to-severe sleep apnea and provide a notification" [s2]. The clearance is for adults 18 and older who have not previously been diagnosed, and the agency states the feature "is not intended to diagnose, treat, or aid in the management of sleep apnea," adding that "the absence of a notification is not intended to indicate the absence of sleep apnea" [s2].

The mechanics matter for reading the result. The watch classifies each night's breathing disturbances as elevated or not, and the notification algorithm evaluates the data over discrete 30-day windows — non-rolling, and requiring at least 10 nights of recording — before deciding whether to alert [s1]. A wearer is not being told about last night. They are being told about a month.

What the validation found

Apple validated the feature against reference sleep testing in a study that enrolled 1,499 participants, of whom 1,448 completed it [s1]. The overall sensitivity for moderate-to-severe sleep apnea was 66.3% (95% CI, 62.2 to 70.3), with a specificity of 98.5% (95% CI, 98.0 to 99.0) [s1].

The average conceals a wide spread. For severe sleep apnea, sensitivity was 89.1% (95% CI, 83.7 to 93.2); for moderate sleep apnea, it was 43.4% (95% CI, 36.5 to 50.5) [s1]. The feature is, in effect, good at catching the cases that are hardest to miss clinically and unreliable at the borderline ones. It also does not target mild sleep apnea at all — by design, the algorithm is tuned to signs of moderate-to-severe disease [s1].

The very high specificity is the flip side of the same design. False alarms are rare, so a notification is a reasonably meaningful prompt. But the modest overall sensitivity means a silent watch is weak reassurance, exactly as the FDA's labeling warns: many people with real, even moderate, sleep apnea will never be flagged.

What a notification is, and is not

The correct response to an alert is a diagnostic sleep study, not a conclusion. Home sleep apnea tests and in-laboratory polysomnography measure the apnea-hypopnea index directly; the watch only infers a risk pattern from movement. Nothing about the notification establishes severity, the type of apnea, or whether treatment is warranted — questions that carry real weight, since untreated moderate-to-severe obstructive sleep apnea is linked to cardiovascular and metabolic harm.

How it compares

Apple's is not the only wrist device making this claim. An independent validation of a Samsung Galaxy Watch against two nights of polysomnography reached a high area under the curve for moderate-to-severe apnea but showed the same tension between sensitivity and the threshold the device ships with — a recurring pattern in this category. The broader problem is that consumer sleep trackers went mainstream faster than the validating evidence, and the metrics people trust most are often the ones devices estimate least reliably.

For a wearer, the honest framing is narrow. The Apple Watch can notice, over a month, that a sleeper's breathing looks disturbed in a way that severe apnea tends to produce — and prompt a visit that might otherwise never happen, given how many cases go undiagnosed for years. It cannot tell a moderate case from a normal night with any consistency, and it was not cleared to try.