How Accurate Is Your Smartwatch’s ‘Deep Sleep’ Score? Should a Low Score Worry You?

Originally written in Korean and translated for international readers.

Imagine waking up and checking your smartwatch. The display shows “deep sleep 12%” or “sleep score 68.” Those numbers can shape how you feel about the day.

First, these numbers are for reference, not diagnosis. A smartwatch does not measure deep sleep directly. Its estimates can differ substantially from test results^[1–2] [1] Chee MWL, et al. World Sleep Society recommendations for the use of wearable consumer health trackers that monitor sleep. Sleep Med. 2025;131:106506.[2] Lee T, et al. Accuracy of 11 Wearable, Nearable, and Airable Consumer Sleep Trackers: Prospective Multicenter Validation Study. JMIR Mhealth Uhealth. 2023;11:e50983.. The World Sleep Society (WSS) recommends looking at weekly averages and trends rather than dwelling on a single night’s reading^[1].

There are five points to understand.

Question	Brief answer
What does deep sleep duration mean?	Time estimated from movement and heart rate, not measured directly
How accurate is it?	In the cited study, the highest deep sleep F1 score was about 0.6, with substantial variation across devices
Does a low score indicate a disorder?	There is no established threshold separating normal from abnormal
Will increasing deep sleep improve health?	Firm evidence that increasing it changes health outcomes is still lacking
Which values should I look at?	Weekly averages and trends recorded with the same device

The “deep sleep” on your screen is not a direct measurement

During polysomnography (PSG), electrodes attached to the scalp record brain waves directly. The clinic also measures eye movements and chin muscle tone, then uses these records to score sleep stages. This test currently serves as the reference standard.

Wrist-worn devices cannot measure brain waves, so they rely on other signals. An algorithm uses arm movements and small changes in heart rate to infer that you are probably in deep sleep. It is like trying to work out what is happening inside a room by listening through the wall without being able to see inside. That is why the “deep sleep” shown in an app is not a direct measurement. The value also depends on the sensors and algorithms used^[1].

The WSS explains that the proportions of sleep stages reported by a device can differ substantially from test results. These values should therefore be read as reference information rather than a basis for making a judgment^[1].

How well do devices identify deep sleep? Accuracy is limited

A Korean study recruited 75 participants with subjective sleep complaints through a hospital and a sleep clinic, and compared 11 consumer sleep trackers with simultaneous polysomnography (PSG) in a sleep laboratory. These included five wrist-worn or ring devices, as well as bedside sensors and smartphone apps. To avoid interference between devices, the researchers divided them into two sets; each participant used eight devices simultaneously. In this study, the five wrist-worn or ring devices had deep sleep F1 scores ranging from about 0.3 to 0.6. Even the best result was only around 0.6. These results do not establish that the devices would perform the same way in people without sleep complaints or when used at home. The accuracy measure used here, F1, is strictly the harmonic mean of precision and recall. Put more simply, it combines two proportions into one score: how much of the time a device labeled as deep sleep really was deep sleep, and how much of the actual deep sleep the device managed to catch. Missed time and wrongly labeled time both count against it, so the score rises only when both are good — doing well on just one does not lift it. It takes a moment to get used to, but in the end a score of 1 means nothing was missed and nothing was mislabeled. A score of about two-thirds means that if you lay the laboratory scoring and the device's reading side by side, only half of the combined time overlaps, while a score around 0.5 means the mistaken time runs to roughly twice the time it got right^[1–2] [1] Chee MWL, et al. World Sleep Society recommendations for the use of wearable consumer health trackers that monitor sleep. Sleep Med. 2025;131:106506.[2] Lee T, et al. Accuracy of 11 Wearable, Nearable, and Airable Consumer Sleep Trackers: Prospective Multicenter Validation Study. JMIR Mhealth Uhealth. 2023;11:e50983..

Would choosing the right device solve the problem? Even among the five devices with the best overall performance, the average F1 for deep sleep was 0.528^[2]. In other words, even the best of them sit at around half marks.

Consider the pooled research as well. A 2025 analysis pooled data from 798 participants across 24 studies. Device estimates of total sleep time and sleep efficiency differed clearly from laboratory measurements. Differences in the time taken to fall asleep and time spent awake during sleep became less clear after adjustment for publication bias. Variation between studies was also substantial. The authors reached a measured conclusion: although wrist-worn devices cannot be considered as reliable as a clinical test, they can be used to follow changes in overall sleep patterns^[3].

That is a reasonable way to understand these devices today: use them to follow trends rather than taking the displayed numbers at face value.

Deep sleep percentages cannot establish a diagnosis

Is there a threshold such as “less than 15% deep sleep is dangerous”? No such criterion exists. The WSS cautions against becoming overly concerned about the duration and proportion of deep and REM sleep. The quality of sleep-stage values varies by device, and these values are for reference rather than a basis for judgment^[1]. Clear thresholds are still lacking: even while recommending regular sleep, the WSS notes that there is no numerical criterion for defining that regularity^[1]. This is why an app’s deep sleep percentage cannot distinguish normal from abnormal^[1]. Keep in mind, too, that the dementia studies discussed next examined sleep measured by polysomnography, not smartwatch scores.

Dementia may be the greatest concern. Yet the relevant evidence does not all point in the same direction. A study that followed 346 people aged 60 or older for 17 years found that those whose deep sleep declined faster each year had a higher subsequent risk of dementia^[4]. Meanwhile, an analysis of approximately 4,650 people aged 60 or older from five cohorts found no clear association between dementia and the proportions of deep and REM sleep measured at a single time point^[5]. The studies measured different things: the first examined the rate of decline, and the second examined proportions at a particular time. It is therefore difficult to conclude that one was wrong. These findings do not yet establish that a low proportion of deep sleep means dementia.

Aspect	Rate of decline	Proportion at a particular time
What was measured	Annual decline between two tests	Proportions of deep and REM sleep measured at a single time point
Study size	346 participants, 17 years of follow-up (52 dementia cases)	Approximately 4,650 participants in five cohorts (998 dementia cases)
Observation	Faster decline was associated with greater dementia risk	No clear association with dementia
Evidence	[4]	[5]

Both studies analyzed records obtained through polysomnography. Their findings do not justify using a smartwatch’s deep sleep percentage as a decision threshold^[1].

Evidence that increasing deep sleep changes health outcomes is still lacking

The WSS’s review finds that observational results on the relationship between deep sleep and cognitive decline or dementia are inconsistent. Firm evidence that artificially increasing deep sleep changes health outcomes is also still lacking^[1].

Researchers have tested playing sounds during sleep to try to increase deep sleep. The overall analysis of 177 participants across 10 studies found a small improvement in memory that did not reach statistical significance (effect size Hedges' g 0.25, p=0.07). However, a subgroup of studies using phase-locked stimulation, which times sounds to a particular point in the brain-wave cycle, showed a significant effect (g 0.36, p=0.047), as did the subset of those studies involving young adults (g 0.44, p=0.01). The subgroup of all studies involving young adults did not reach significance (g 0.31, p=0.051). The authors concluded that the evidence was insufficient to recommend commercial devices^[6]. A low smartwatch score is not a reason to make increasing deep sleep itself a goal.

What should you pay attention to instead?

Health is linked to habitual sleep, rather than a record of last night alone. A synthesis of studies that followed approximately 1.38 million people found that habitual short sleepers had about 1.12 times the risk of death. The risk was about 1.3 times as high among long sleepers, so this does not mean that sleeping longer is always safer. The authors note, though, that no mechanism has been shown by which long sleep itself would cause harm, and that the longer sleep may be the result of an illness the person already has. This is not a reason to cut sleep short — it means that if you have started sleeping much longer than before, it is worth checking whether something else is going on. These figures did not come from a single night’s score, either. Researchers asked about usual sleep duration and then followed participants for years^[7].

Focus on basic habits rather than the score. Keep consistent bedtimes and wake times, and allow enough time for 7–8 hours of sleep. Regular exercise and limiting alcohol before bed are also part of these habits^[8].

Do not base your judgment on a single night’s score.

Aspect	Avoid this approach	Use this approach
Values to review	One night’s deep sleep percentage	Weekly averages from the same device
Purpose	Deciding what is normal or abnormal; raising the score	Following trends in sleep duration, bedtime, and regularity
What to consider alongside the data	Focusing only on the score	Daytime sleepiness and fatigue
What to compare	Scores from different manufacturers’ devices	Changes over time recorded with the same device

If the score stays on your mind in bed and makes you worry about sleep, you can pause tracking and discuss it with a healthcare professional^[10].

This article provides general information, not individualized medical advice. If sleep problems persist or you experience excessive daytime sleepiness and fatigue, consult a healthcare professional regardless of your device’s score^[9].