Why Is My Deep Sleep So Low? Wearable Accuracy Runs 50.5% to 79.5%

Your ring said 42 minutes of deep sleep last night. Three nights ago it said 91. You did not do anything differently.

Before you change your training over that, know what the number is. Deep sleep is the least accurate figure your wearable reports. In a 2024 Brigham and Women’s Hospital validation against polysomnography, deep-sleep detection sensitivity ran from 79.5% on the Oura Ring Gen3 down to 50.5% on the Apple Watch Series 8 — and the Apple Watch undercounted deep sleep by 43 minutes a night against the lab reference.¹

The same three devices separated sleep from wake at ≥95% sensitivity.¹

So the numbers on your sleep screen are not equally real. They are ranked in almost exactly the reverse order of the attention people give them — and the two your device measures best are the two carrying the hard outcome data.

Question	Short answer
Why is my deep sleep so low?	Usually the algorithm. Device-versus-lab agreement on deep-sleep minutes had ICCs of 0.13–0.36 — “poor” on the study’s own scale¹
How accurate is deep-sleep tracking?	50.5%–79.5% detection sensitivity by device¹; across 11 trackers, macro F1 for staging ran 0.26–0.69²
How accurate is sleep vs. wake?	≥95% sensitivity across the devices tested¹
Is there a clean ground truth?	Not a crisp one — trained technicians scoring the same records agreed 82.6% of the time, and only 67.4% on deep sleep³
How much deep sleep do I need?	No validated personal target exists; the share of the night spent in slow-wave sleep declines with age⁴
Does deep-sleep percentage predict recovery?	Unestablished — the largest athlete dataset, 35,140 nights, built its performance model on duration and consistency, not stage composition⁵
What does predict outcomes?	Regularity. In 60,977 UK Biobank adults it out-predicted duration for all-cause mortality⁶
What should I track instead?	Bedtime and wake-time consistency plus total sleep opportunity, on a 14-night trend
Is there a cost to tracking?	Yes — orthosomnia was identified in 3.0%–14.0% of a 523-person sample, depending on threshold⁷

Why Is My Deep Sleep So Low? Usually It’s the Algorithm, Not You

In most cases, a low deep-sleep reading is a measurement artifact rather than a change in your sleep — deep-sleep minute counts agree with polysomnography at intraclass correlations of just 0.13 to 0.36, a band the study’s own authors label “poor.”¹

Deep sleep is the hardest stage for a wrist or finger sensor to identify, and the minute count is the least reproducible number on the screen. Broken out by device, when Robbins and colleagues compared each device’s deep-sleep minutes against simultaneous polysomnography, the intraclass correlations were 0.32 for the Oura Ring Gen3, 0.36 for the Fitbit Sense 2, and 0.13 for the Apple Watch Series 8 — all below the 0.40 threshold the authors themselves label “poor” concordance.¹

An intraclass correlation is a scale-of-agreement number, not a percentage of correctness. Think of it as asking whether the device and the lab rank the same night the same way. Below 0.40, they largely don’t.

Here are four causes of a low deep-sleep number, ordered by what the measurement data suggests you should suspect first.

Classification error. The device assigned epochs to light sleep that the lab would have called deep. The Apple Watch’s 45-minute overestimate of light sleep is the mirror image of its 43-minute deep-sleep shortfall — the minutes did not vanish, they got relabeled.¹
Normal night-to-night biology. Slow-wave sleep is front-loaded and pressure-dependent. It moves on its own.
Age. The proportion of the night spent in slow-wave sleep declines across adulthood, which is a slow drift rather than a Tuesday-to-Friday swing.⁴
An actual disruptor. Alcohol, illness, a late hard session, a new mattress, a warm room.

That order is a reading of the agreement data above, not a measured frequency. When device and lab disagree that badly on a single night, causes one and two plausibly swamp three and four, and a 30-minute drop tells you very little about your body. Across 11 consumer sleep trackers — wearables, bedside sensors, and phone-based apps — benchmarked against polysomnography, macro F1 for stage classification ranged from 0.26 to 0.69, a spread wide enough that the same night on two devices produces two different stories.²

How Accurate Is Deep Sleep Tracking? The Per-Device Numbers

Deep-sleep detection sensitivity ranged from 79.5% on the Oura Ring Gen3 to 50.5% on the Apple Watch Series 8, while the same devices separated sleep from wake at ≥95% sensitivity.¹

That gap is the whole story. Sleep/wake detection is genuinely good and it is the part of the product that works; stage classification is a different engineering problem. The 2024 validation behind both figures enrolled 35 healthy adults aged 20 to 50 for one inpatient night, all three devices worn simultaneously against polysomnography in the journal Sensors.¹ The analyzable sample differed by device — 35 for Oura, 33 for Fitbit, 29 for the Apple Watch.¹

Device (generation studied)	Sleep vs. wake sensitivity	Deep-sleep detection sensitivity	Deep-sleep bias vs. PSG	Source
Oura Ring Gen3	≥95%	79.5%	No significant difference (95 vs. 95 min)	Robbins 2024, n=35, 1 night¹
Fitbit Sense 2	≥95%	61.7%	−15 min (p < 0.001)	Robbins 2024, n=33, 1 night¹
Apple Watch Series 8	≥95%	50.5%	−43 min (p < 0.001)	Robbins 2024, n=29, 1 night¹
WHOOP strap (hardware generation not stated)	95%	68% (slow-wave sleep)	Not reported	Miller 2020, n=12, 86 nights⁸

These are protocol- and generation-specific results, not permanent brand grades. A Gen3 ring number does not validate a Gen4 ring, a Series 8 number does not indict a Series 11, and firmware ships new staging algorithms without telling you. The same generation-specific pattern shows up in wearable HRV validation studies, where a strong result for one model gets quoted for years as if it applied to the whole brand.

Worth knowing about that table: the study’s first author sits on Oura’s medical advisory board, a disclosure the paper makes plainly. It is a well-run study — hold the best-performing device’s result a little more loosely than the worst.

Is Oura Deep Sleep Accurate?

Oura Ring Gen3 detected deep sleep at 79.5% sensitivity — the best result in the comparison, and still its own weakest metric.¹

The stronger finding is at the summary level: Oura’s estimates of wake, light sleep, deep sleep, and REM sleep were not statistically different from polysomnography, the only device in that study to clear all four.¹ On average, across 35 people, it landed on the right total.

But the per-night agreement told a different story. Oura’s deep-sleep ICC was 0.32, still in the “poor” band.¹ Getting the group average right and getting your Tuesday right are separate achievements, and the ring managed the first without the second.

A separate multicenter validation including the Oura Ring 3 found overall stage-classification performance varied with body-mass index, sleep efficiency, and apnea-hypopnea index — the number you get is partly a function of who you are.²

Apple Watch Deep Sleep Accuracy

Apple Watch Series 8 detected deep sleep at 50.5% sensitivity and underestimated it by 43 minutes per night versus polysomnography.¹

Coin-flip is the right instinct and the wrong conclusion to generalize from. That 50.5% is the floor of Apple’s stage range; its best stage reached 86.1% sensitivity, and its deep-sleep precision was the highest of any device at 87.8%.¹ Together those describe a conservative classifier: when the Series 8 calls something deep sleep it is usually right, and it misses a lot of the real thing.

The mechanical consequence is the 45-minute light-sleep overestimate reported in the same paper.¹ Nearly all the deep sleep the watch failed to catch was filed as light.

And the headline stayed strong where it counts: sleep versus wake at ≥95% sensitivity, with total sleep duration similar to the lab across all three devices.¹ Ask an Apple Watch how long you slept and you are on solid ground. Ask it how much deep sleep you got and you are reading a stage it never resolved well.

Are WHOOP Sleep Stages Accurate?

WHOOP’s stage classification carries the same limitation as the category. In a 10-day laboratory protocol with 86 sleeps scored against polysomnography, the WHOOP strap reached 95% sensitivity to sleep in two-stage classification but only 68% sensitivity to slow-wave sleep in four-stage classification, with overall four-stage agreement of 64% and Cohen’s kappa of 0.47.⁸

Two caveats travel with that number. The sample was 12 young healthy adults, and the paper does not specify a hardware generation — WHOOP has shipped multiple platforms since. No newer independent deep-sleep sensitivity figure for a current WHOOP model appeared in the validation literature we could verify, so treat 68% as the last checkable data point rather than today’s spec.

The class-level answer is the safer one. Across 11 consumer sleep trackers — wearables, bedside sensors, and phone-based apps — and 349,114 scored epochs, macro F1 for stage classification ran from 0.26 to 0.69.² The five wrist and finger wearables in that set sat well short of lab-grade staging.

Why There’s No Clean Ground Truth for a Sleep Stage

Sleep stages are a scoring convention, not a physical quantity like body mass.

When trained technicians score the same polysomnography records, they agree about 83% of the time — and deep sleep is among the stages they agree on least.

In the American Academy of Sleep Medicine’s inter-scorer reliability program, more than 2,500 experienced scorers made over 3.2 million scoring decisions on 1,800 shared epochs. Overall sleep-stage agreement averaged 82.6%. Agreement on stage N3, deep sleep, was 67.4%. The authors singled out discrimination between N2 and N3 as “particularly difficult.”³

Sit with that. The rubric your wearable is being graded against is one that human experts apply inconsistently, precisely at the boundary your ring is trying to find.

Rebecca Robbins, PhD, of the Division of Sleep and Circadian Disorders at Brigham and Women’s Hospital and the Division of Sleep Medicine at Harvard Medical School, led the three-device comparison at the center of this article.¹ Her team’s framing is notably measured: in adults with healthy sleep, all the devices were similar to polysomnography in estimating sleep duration, with moderate-to-substantial agreement on stages.¹ That is a real endorsement of one capability and a bounded one of the other.

The measurement problem also degrades in exactly the conditions that make you want to check. Across seven consumer devices tested on three consecutive nights including a deliberately disrupted one, epoch-by-epoch sensitivity stayed high at ≥0.93, specificity was low-to-medium at 0.18–0.54, stage assessments were inconsistent, and devices “tended to perform worse on nights with poorer/disrupted sleep.”⁹

The pooled picture is the same. A meta-analysis of 24 studies and 798 participants across Fitbit, WHOOP, Garmin, Apple Watch and others found consumer wrist devices underestimated total sleep time by about 16.9 minutes, underestimated sleep efficiency by about 4.7 points, and overestimated wake after sleep onset by about 13.3 minutes.¹⁰ Directionally consistent, modest at the summary level, and not remotely precise enough to prescribe a stage target.

How Much Deep Sleep Do You Actually Need?

There is no validated per-night deep-sleep target for an individual, and no consumer device measures it precisely enough to hit one if there were.

What the normative literature establishes is a direction, not a number you should chase. Ohayon and colleagues pooled 65 studies using overnight polysomnography or actigraphy, covering 3,577 healthy people aged 5 to 102. In adults, total sleep time, sleep efficiency, percentage of slow-wave sleep, percentage of REM sleep, and REM latency all decreased significantly with age, while stage 1, stage 2, and wake after sleep onset increased.⁴

Deep sleep is a share of the night that trends down across the decades. A 45-year-old comparing their number to a 25-year-old’s is comparing two different normal ranges.

Notice what that meta-analysis does not provide: a per-person minimum. It describes population averages on samples whose screening for sleep disorders varied — the authors note that screening quality substantially changed the effect sizes.

A defensible target looks like this instead:

A sleep opportunity, not a stage quota. Adults should sleep seven or more hours per night on a regular basis to promote optimal health, per the joint AASM and Sleep Research Society consensus statement.¹¹ That is a number your device can actually measure.
A window, not a night. Judge deep sleep on a two-week average if you judge it at all, because single-night agreement with the lab is poor.¹
A comparison to yourself, on the same hardware. Firmware changes and device generations both move the baseline.¹⁹
A performance check. How the warm-up went is a better readout of recovery than a stage estimate whose ICC is 0.13.¹

Does Deep Sleep Percentage Matter for Recovery?

Across 35,140 nights from 389 professional golfers — the largest athlete sleep dataset published to date — sleep duration and sleep consistency were associated with tournament scoring. Stage composition was not in the model at all.⁵

Grosicki and colleagues drew that data from 521 events across multiple professional seasons, all male athletes. The metrics they put in the model were sleep duration (7.2 ± 0.7 h), sleep consistency (69.1% ± 6.9%), resting heart rate, heart-rate variability, and a composite recovery score — nothing about stages.⁵

That omission is the finding. When a team with tens of thousands of nights and objective performance outcomes builds its analysis, duration and regularity are what they reach for.

Here is what they found, converted into plain language:

One extra hour of sleep was associated with a 0.522-stroke lower score, between athletes.⁵
A 10-percentage-point gain in sleep consistency was associated with 0.382 strokes lower, between athletes.⁵
Within the same athlete, improving sleep consistency by 10 points tracked with 0.193 strokes lower — the same direction, at roughly half the size.⁵
A 1-beat-per-minute lower resting heart rate was associated with 0.038 strokes lower.⁵

These are observational associations in elite male golfers, and the authors are employees of WHOOP, which supplied the data. That does not invalidate the analysis; it does mean the effect sizes should be read as associations rather than a dose you can prescribe.

The intervention literature points the same way. A systematic review of 25 studies of sleep interventions in athletes concluded that sleep extension and naps were the most effective strategies for improving sleep and subsequent performance, while sleep-hygiene education and removing electronic devices at night showed no effect on performance outcomes.¹² What worked was more sleep. Not better-distributed sleep.

This is why SensAI compares planned-versus-performed session data against your sleep context rather than against a stage score. What shows up in your training — sets completed, load moved, how the session actually went — is a firmer signal than a deep-sleep number that moved 20 minutes overnight. If you want the mechanics of the composite scores themselves, we’ve broken down how each brand actually calculates its readiness score.

The Two Numbers Your Wearable Gets Right

Timing and total sleep are the two things a consumer wearable measures accurately, and they are also the two with the strongest outcome data behind them.

Across 60,977 UK Biobank participants with more than 10 million hours of accelerometer data, the Sleep Regularity Index predicted all-cause mortality more strongly than sleep duration did. Participants in the top four regularity quintiles had a 20% to 48% lower risk of all-cause mortality than the least regular quintile, with 16% to 39% lower cancer mortality and 22% to 57% lower cardiometabolic mortality, adjusted for age, sex, ethnicity, and sociodemographic, lifestyle, and health factors.⁶

The Sleep Regularity Index is simpler than it sounds: it’s the probability that you are in the same state — asleep or awake — at any two moments exactly 24 hours apart, scaled from 0 to 100. Median in that cohort was 81.⁶ Going to bed and getting up at roughly the same times is the whole mechanism.

This was a prospective observational cohort, so regularity was associated with survival, not proven to cause it.

Andrew J. K. Phillips, PhD, and first author Daniel P. Windred, both of the Turner Institute for Brain and Mental Health at Monash University, are careful about that distinction throughout, and their strongest claim is predictive rather than causal.⁶

The replication is broad. A 2025 systematic review of 59 primary studies found consistent, moderate-certainty evidence linking irregular sleep timing to depressive and anxiety symptoms, elevated BMI, insulin resistance, hypertension, and incident cardiovascular events — and five low-bias cohorts showing 20% to 88% higher all-cause mortality among the least regular sleepers, independent of sleep duration and quality.¹³ Proposed mechanisms — circadian misalignment, autonomic imbalance, inflammation, HPA-axis disruption, gut dysbiosis — remain hypotheses rather than demonstrated pathways.¹⁴

Keep the three concepts separate:

Regularity — the consistency of your sleep and wake times. Measured well by any wearable that knows when you fell asleep. Strongest outcome evidence.⁶¹³
Duration — total time asleep. Measured well; consumer devices matched polysomnography on sleep duration in the three-device validation.¹ Directly recommended at seven or more hours.¹¹
Stage composition — the deep/REM/light split. Measured poorly.¹² No consumer-level outcome evidence attached.

SensAI reads sleep duration and quality through Apple HealthKit — Apple Watch directly, and Garmin, Oura, and WHOOP when those services write to HealthKit — and feeds that into programming alongside HRV and resting heart rate, working from the multi-night trend rather than reacting to one night’s stage estimate. If you’ve been short for a week and are wondering how long the hole takes to fill, we’ve covered how sleep debt accumulates and how long it takes to clear.

Deep Sleep vs REM: Which Is More Important for Recovery?

Neither, at the level your device can measure them — deep-sleep detection sensitivity runs 50.5% to 79.5% across devices and macro F1 for staging runs 0.26 to 0.69, so no consumer wearable classifies either stage accurately enough to justify a stage-specific training decision.¹²

Both stages have real physiological roles. But those classification numbers, plus the largest athlete performance model built on duration and consistency rather than stage composition,⁵ all point the same way.

The comparison also assumes the two numbers are independent measurements. They aren’t. Misclassification moves minutes between stages — the Apple Watch’s missing 43 minutes of deep sleep reappeared as 45 extra minutes of light sleep in the same dataset.¹ Your deep-versus-REM ratio is partly an artifact of where the classifier drew its boundary.

If the question you’re really asking is whether a rough night should change today’s session, that’s a different question with a better answer, and we’ve worked through what sleep quality does and doesn’t do to workout performance separately.

How to Read Your Sleep Screen: A 14-Night Framework

Read sleep on a 14-night trend, not a morning. Single-night deep-sleep changes are dominated by measurement noise; two weeks of bedtimes and wake times is a signal your device can actually resolve.

What the 14-night trend shows	What it likely means	What to do tomorrow
Bedtime or wake time varies by more than an hour night to night	The strongest modifiable signal in the data⁶¹³	Anchor the wake time first — it stabilizes everything else
Sleep opportunity under 7 h on 5+ of 14 nights	A real, cumulative deficit	Extend sleep before you touch training load¹²¹¹
Deep sleep down 20 min vs. your average, everything else stable	Almost certainly noise — deep-sleep ICC 0.13–0.36¹	Nothing. Train as planned
Deep sleep down and duration down and timing drifting	A genuine bad-sleep block	Keep the session, cut the intensity¹⁵
Deep sleep low but you feel fine and your lifts are on target	Performance beats the estimate	Train as planned
Short sleep plus snoring or daytime sleepiness	Possible sleep-disordered breathing — and tracker stage accuracy varies by apnea-hypopnea index²; irregular sleepers also carry higher odds of OSA¹⁶	See a clinician; this is not a firmware problem

Two of those rows deserve expansion.

The intensity row comes from season-long monitoring rather than a lab. Merrigan and colleagues tracked 25 NCAA Division I women’s ice hockey players across a championship season and found that as internal workload rose during training or competition, that night’s heart-rate variability fell, heart rate rose, and sleep duration and readiness scores dropped.¹⁵

The study only looked one way — higher internal workload was followed by worse sleep and lower nocturnal HRV that same night. It’s an observational season, not a causal test. The practical inversion, that a stack of short nights is a reason to spend the session on quality rather than volume, is a judgment call the data doesn’t make for you.

The clinical row matters because tracker error is not evenly distributed — stage performance varied with body-mass index, sleep efficiency, and apnea-hypopnea index across the 11-tracker validation.² Separately, in 602 middle-aged adults from the Raine Study, severely irregular sleepers had roughly twice the odds of obstructive sleep apnea, and OSA plus severely irregular sleep carried the cohort’s highest odds of hypertension at OR 2.34 (95% CI 1.07–5.12).¹⁶ Observational — and a strong argument for treating snoring plus daytime sleepiness as a medical question.

Constraints are the part most tools ignore. Tell the SensAI coach that your Tuesdays run short because of a standing early meeting and that constraint persists across sessions, so the plan stops treating every week as identical. And when two devices give you different verdicts on the same night, we’ve written about when your devices disagree with each other.

Orthosomnia: The Cost of Optimizing a Guess

There is a documented failure mode for all of this. Orthosomnia — anxiety about achieving perfect tracked sleep — was identified in 3.0% to 14.0% of a 523-person sample depending on the threshold used, and identified cases had more severe insomnia symptoms than non-cases.⁷

The identification was algorithmic, not diagnostic. Researchers required four things at once: ownership of a sleep-tracking wearable, insomnia symptoms, sleep-related preoccupation, and the absence of severe generalized anxiety. Tightening the threshold moved the prevalence from 14.0% to 8.6% to 3.0%, and about 36% of the sample were regular tracker users.⁷

That is a single cross-sectional sample and cannot establish that tracking caused the anxiety. It is still the right number to hold in mind before you start optimizing a metric with an ICC of 0.13.

Practical de-escalations:

Check the app after breakfast, not before. A number you read at 6:04 a.m. sets the mood for a day it cannot predict.
Delete the daily score from your attention and keep the weekly one.
Set one behavioral goal, not a metric goal — a fixed wake time beats a deep-sleep target you can’t influence directly.
Take a tracking break if the number is driving the anxiety rather than describing it.

Because SensAI is a conversation, “my ring says 22 minutes of deep sleep but I feel fine and my warm-up sets moved well” is something you can just say, and it gets weighed alongside the data instead of being overridden by it.

How SensAI Uses Sleep Data

SensAI receives connected wearable context through Apple HealthKit. Apple Watch connects directly; Garmin, Oura, and WHOOP data can reach SensAI through HealthKit when those services write the relevant fields there.

The LLM coaching layer combines aggregated recovery metrics — sleep duration and quality, HRV trends, resting heart rate, workout summaries, planned-versus-performed history — with your goals, schedule, equipment, and constraints, and regenerates the weekly program from actual data rather than a template.

Sleep does not independently rewrite a session. There is no rule that converts a stage estimate into an exercise substitution, because the measurement does not support one. What sleep context does is inform the conversation and the weekly plan, alongside how your sessions actually went.

Raw HealthKit data stays on your device. Aggregated recovery metrics, sleep quality, and workout summaries used for coaching are sent server-side, and SensAI does not sell your data to third parties.

If you are still deciding which device to wear in the first place, our full device-by-device comparison covers what each platform measures well.

Frequently Asked Questions

Why is my deep sleep suddenly so low?

Most often the classifier moved minutes into a neighboring stage rather than your physiology changing. In the three-device validation, the Apple Watch Series 8’s 43-minute deep-sleep shortfall was accompanied by a 45-minute light-sleep overestimate in the same nights — the same sleep, differently labeled.¹

Is 30 minutes of deep sleep enough?

No validated per-night threshold exists for an individual, so “enough” isn’t answerable from that number. Slow-wave sleep as a percentage of the night declines with age in healthy adults,⁴ and the device reporting your 30 minutes agrees with the lab poorly enough that the figure itself carries wide uncertainty.¹

Can I increase my deep sleep?

The reliable lever is total sleep, not the stage directly. A review of 25 athlete intervention studies found sleep extension and napping were the most effective strategies for improving sleep and performance, while sleep-hygiene education alone showed no performance effect.¹² Give yourself more sleep opportunity and you’re pulling the lever that actually has evidence behind it.

Is sleep consistency more important than sleep duration?

For mortality risk, the objective data says yes. Sleep Regularity Index scores out-predicted sleep duration for all-cause mortality across 60,977 UK Biobank participants,⁶ and five low-bias cohorts in a 2025 systematic review found 20% to 88% higher all-cause mortality in the least regular sleepers, independent of duration.¹³ Both are observational, so treat regularity as the stronger predictor rather than a proven cause.

Do sleep trackers work if you have sleep apnea?

Less well, and unevenly. Stage-classification accuracy across 11 consumer trackers varied with apnea-hypopnea index and body-mass index,² and devices generally performed worse on nights with disrupted sleep in a seven-device comparison.⁹ Suspected apnea is a reason to see a clinician, not to buy a better ring.

Should I stop tracking my sleep?

Not necessarily — the sleep/wake and duration data are accurate and genuinely useful.¹ But if checking the app is producing anxiety rather than information, that pattern has a name and a measured prevalence of 3.0% to 14.0% depending on the definition used.⁷ Switching to a weekly review is usually enough.

The Bottom Line

Your wearable is very good at telling you when you slept and for how long, mediocre at telling you which stage you were in, and worst of all at deep sleep specifically.

That hierarchy matches the outcome literature almost exactly. Regularity and duration are where the survival data,⁶¹³ the athlete performance associations,⁵ and the intervention evidence¹² all live. Stage composition is where the measurement error lives.

Track the two numbers your device measures well, judge them over fourteen nights, and let the deep-sleep figure be a curiosity rather than a verdict.

References

Robbins R, Weaver MD, Sullivan JP, Quan SF, Gilmore K, Shaw S, Benz A, Qadri S, Barger LK, Czeisler CA, Duffy JF. “Accuracy of Three Commercial Wearable Devices for Sleep Tracking in Healthy Adults.” Sensors (Basel), 2024;24(20):6532. https://pubmed.ncbi.nlm.nih.gov/39460013/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴ ↩¹⁵ ↩¹⁶ ↩¹⁷ ↩¹⁸ ↩¹⁹ ↩²⁰ ↩²¹ ↩²² ↩²³ ↩²⁴ ↩²⁵ ↩²⁶ ↩²⁷ ↩²⁸ ↩²⁹ ↩³⁰ ↩³¹ ↩³² ↩³³ ↩³⁴
Lee T, Cho Y, Cha KS, Jung J, Cho J, Kim H, Kim D, Hong J, Lee D, Keum M, Kushida CA, Yoon IY, Kim JW. “Accuracy of 11 Wearable, Nearable, and Airable Consumer Sleep Trackers: Prospective Multicenter Validation Study.” JMIR mHealth and uHealth, 2023;11:e50983. https://pubmed.ncbi.nlm.nih.gov/37917155/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹
Rosenberg RS, Van Hout S. “The American Academy of Sleep Medicine inter-scorer reliability program: sleep stage scoring.” Journal of Clinical Sleep Medicine, 2013;9(1):81-87. https://pubmed.ncbi.nlm.nih.gov/23319910/ ↩ ↩²
Ohayon MM, Carskadon MA, Guilleminault C, Vitiello MV. “Meta-analysis of quantitative sleep parameters from childhood to old age in healthy individuals: developing normative sleep values across the human lifespan.” Sleep, 2004;27(7):1255-1273. https://pubmed.ncbi.nlm.nih.gov/15586779/ ↩ ↩² ↩³ ↩⁴
Grosicki GJ, von Hippel W, Fielding F, Kim J, Chapman C, Holmes KE. “Wearable-Derived Sleep and Physiological Metrics Are Associated With Performance in Professional Golfers.” International Journal of Sports Physiology and Performance, 2025;21(2):180-191. https://pubmed.ncbi.nlm.nih.gov/41265431/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹
Windred DP, Burns AC, Lane JM, Saxena R, Rutter MK, Cain SW, Phillips AJK. “Sleep regularity is a stronger predictor of mortality risk than sleep duration: A prospective cohort study.” Sleep, 2024;47(1):zsad253. https://pubmed.ncbi.nlm.nih.gov/37738616/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸
Jahrami H, Trabelsi K, Husain W, Ammar A, BaHammam AS, Pandi-Perumal SR, Saif Z, Vitiello MV. “Prevalence of Orthosomnia in a General Population Sample: A Cross-Sectional Study.” Brain Sciences, 2024;14(11):1123. https://pubmed.ncbi.nlm.nih.gov/39595886/ ↩ ↩² ↩³ ↩⁴
Miller DJ, Lastella M, Scanlan AT, Bellenger C, Halson SL, Roach GD, Sargent C. “A validation study of the WHOOP strap against polysomnography to assess sleep.” Journal of Sports Sciences, 2020;38(22):2631-2636. https://pubmed.ncbi.nlm.nih.gov/32713257/ ↩ ↩²
Chinoy ED, Cuellar JA, Huwa KE, Jameson JT, Watson CH, Bessman SC, Hirsch DA, Cooper AD, Drummond SPA, Markwald RR. “Performance of seven consumer sleep-tracking devices compared with polysomnography.” Sleep, 2021;44(5):zsaa291. https://pubmed.ncbi.nlm.nih.gov/33378539/ ↩ ↩² ↩³
Lee YJ, Lee JY, Cho JH, Kang YJ, Choi JH. “Performance of consumer wrist-worn sleep tracking devices compared to polysomnography: a meta-analysis.” Journal of Clinical Sleep Medicine, 2025;21(3):573-582. https://pubmed.ncbi.nlm.nih.gov/39484805/ ↩
Watson NF, Badr MS, Belenky G, Bliwise DL, Buxton OM, Buysse D, Dinges DF, Gangwisch J, Grandner MA, Kushida C, Malhotra RK, Martin JL, Patel SR, Quan SF, Tasali E. “Recommended Amount of Sleep for a Healthy Adult: A Joint Consensus Statement of the American Academy of Sleep Medicine and Sleep Research Society.” Sleep, 2015;38(6):843-844. https://pubmed.ncbi.nlm.nih.gov/26039963/ ↩ ↩² ↩³
Cunha LA, Costa JA, Marques EA, Brito J, Lastella M, Figueiredo P. “The Impact of Sleep Interventions on Athletic Performance: A Systematic Review.” Sports Medicine - Open, 2023;9(1):58. https://pubmed.ncbi.nlm.nih.gov/37462808/ ↩ ↩² ↩³ ↩⁴
Kalkanis A, Lenkens D, Steiropoulos P, Testelmans D. “Sleep regularity as an important component of sleep hygiene: a systematic review.” Sleep Medicine Reviews, 2025;84:102203. https://pubmed.ncbi.nlm.nih.gov/41259946/ ↩ ↩² ↩³ ↩⁴ ↩⁵
Zhang C, Qin G. “Irregular sleep and cardiometabolic risk: Clinical evidence and mechanisms.” Frontiers in Cardiovascular Medicine, 2023;10:1059257. https://pubmed.ncbi.nlm.nih.gov/36873401/ ↩
Merrigan JJ, Stone JD, Kraemer WJ, Friend C, Lennon K, Vatne EA, Hagen JA. “Analysis of Sleep, Nocturnal Physiology, and Physical Demands of NCAA Women’s Ice Hockey Across a Championship Season.” Journal of Strength and Conditioning Research, 2024;38(4):694-703. https://pubmed.ncbi.nlm.nih.gov/38513177/ ↩ ↩²
Sansom K, Reynolds A, Windred D, Phillips A, Dhaliwal SS, Walsh J, Maddison K, Singh B, Eastwood P, McArdle N. “The interrelationships between sleep regularity, obstructive sleep apnea, and hypertension in a middle-aged community population.” Sleep, 2024;47(3):zsae001. https://pubmed.ncbi.nlm.nih.gov/38180870/ ↩ ↩²