How much deep sleep do you get on average?

https://www.sleepfoundation.org/stages-of-sleep/d…

"What Is Deep Sleep?

Deep sleep, also called slow-wave sleep5, occurs in the third stage of non-rapid eye movement (NREM) sleep. During deep sleep, electrical activity in the brain appears in long, slow waves called delta waves6. These waves have a frequency of 0.5 to 2 Hertz7 and they must make up at least 6 seconds of a 30-second window for that window to count as deep sleep.

Typically you descend into deep sleep within an hour of falling asleep, and experience progressively shorter periods of deep sleep as the night wears on. During this stage, automatic body functions like breathing and heart rate are also very slow and your muscles are relaxed. It can be difficult for someone to wake you up, and waking up out of deep sleep may make you feel mentally foggy for up to an hour.

Why Is Deep Sleep Important?

While all stages of sleep are necessary for good health, deep sleep offers specific physical and mental benefits. During deep sleep, your body releases growth hormone and works to build and repair muscles, bones, and tissue, and immune system functioning. Additionally, slow-wave sleep may be important for regulating glucose metabolism8. Elite athletes value slow-wave sleep as it helps replenish energy stores9.

Deep sleep is important for cognitive function and memory10, and it is thought to play a role in language learning, 11motor skills12, and the developing brain13.

Throughout the day, you receive information inputs14 that strengthen the synapses, or points of communication15, in the networks of your brain. However, your brain cannot take on information forever without rest.

Researchers suggest that deep sleep plays a role in preparing your synapses for the next day. Put another way, your brain evaluates new memories and then preserves and consolidates16 only the ones that are most relevant to avoid saturating memory pathways. Although this theory is still under investigation, evidence shows that people obtain a higher proportion of deep sleep after learning a new task, and show higher concentrations of slow waves in brain areas related to the task.

The results of this process are evident: after a night of good sleep, you are better prepared to take on new information and adapt to new environments."

Comments (11)

5mo 
Isaiah_53_5 💎🙌💎🙌💎, what's your opinion? Comment below:

If you wear a smart watch it should tell you the breakdown under sleep metrics.

"If you always put limits on everything you do, physical or anything else, it will spread into your work and into your life. There are no limits. There are only plateaus, and you must not stay there, you must go beyond them." - Bruce Lee

5mo 
EnergyIBK, what's your opinion? Comment below:

I wonder how accurate they are though? I wear an Apple Watch using the AutoSleep app as well as a Whoop strap. Get fairly different readings on both but feel the Apple Watch is better.

Have noticed that the biggest consistently bad influences on my deep sleep (both devices) have been: alcohol (especially before bed), followed by cardio within a couple hours before bed, then too much caffeine and then eating right before bed.

  • 1
5mo 
Isaiah_53_5 💎🙌💎🙌💎, what's your opinion? Comment below:

EnergyIBK

I wonder how accurate they are though? I wear an Apple Watch using the AutoSleep app as well as a Whoop strap. Get fairly different readings on both but feel the Apple Watch is better.

Have noticed that the biggest consistently bad influences on my deep sleep (both devices) have been: alcohol (especially before bed), followed by cardio within a couple hours before bed, then too much caffeine and then eating right before bed.

I have a Garmin and it touts 69.7% accuracy. Yeah, I am trying to monitor my habits to maximize deep sleep as it seems so important. Here is the Garmin Sleep study:

https://www.garmin.com/en-US/blog/health/garmin-health-announces-sleep-study-results/

Garmin Health Announces Sleep Study Results

June 7, 2019

Garmin announced Advanced Sleep Monitoring, an enhanced capability to more accurately identify sleep stages, last June. Developed against truth data generated by a clinical device, the feature is the result of a sleep study by Garmin Health conducted under the supervision of Dr. Suzanne Stevens, Director of the University of Kansas Medical Center (KUMC) Sleep Medicine Clinic with certifications by the American Board of Psychiatry and Neurology and the American Board of Sleep Medicine. 

Dr. Stevens presented the details of the study on May 7th at the Annual Meeting of the American Academy of Neurology. The poster presentation covered study design, results, and the conclusion that Garmin wearables that support Advanced Sleep Monitoring present a valid method to estimate sleep stages. For consumers, this results in invaluable insights regarding their sleep hygiene and how their sleep patterns may be affecting their overall health. For researchers, the results indicate Garmin sleep tracking wearables may be suitable for longitudinal studies where monitoring participant activity and sleep patterns are desired endpoints.

In contrast to similar studies, the Garmin Health and KUMC study was conducted outside a sleep clinic in real-world conditions typical for most users. Study participants used a clinical reference device at home in their own beds, providing data representative of their normal sleeping patterns. Only participants using CPAP machines or with tattoos on their wrist were excluded, meaning data was collected from participants with a variety of known and unknown sleeping conditions, as would generally be the case for users of consumer wearables. 

Outside of the study, algorithm performance from 3,200 nights of sleep in a production environment collected during beta testing was cross-checked with clinical reference data regarding the average distribution of sleep stages by age group. The results aligned with expectations, including a decrease of deep sleep and total sleep duration as a function of age.

Garmin Health provides enterprise solutions that leverage Garmin wearables and the high-quality sensor data they produce for use in the corporate wellness, population health, and patient monitoring markets.  Contact us to request more information about how to build your own innovative experience or research study.


Commercially Available Wearable
Provides Valid Estimate of Sleep Stages

Scott Burgett1,Robert Blair1, Darrell Lightfoot2, Catherine Siengsukon3, Adam Reetz1, Suzanne Stevens2
1Garmin International, 2Neurology, University of Kansas Health Systems, 3University of Kansas Medical Center

Introduction

In the last several years there has been an explosion of smart wearable devices that are capable of measuring steps, activity, heart rate, and other biometric data about the wearer. An estimated 117 million smart wearables will be sold in 2019, with that number rising to 233 million in 2022 1.  Actigraphy has been the gold standard for many years to estimate sleep stages using wearables. Actigraphy uses accelerometer information to estimate wake/sleep periods of the user 2. Many modern wearables also measure heart rate and heart rate variability by using optical photoplethysmography (PPG). PPG uses light emitted into the skin to measure pulse rate by observing small changes in the intensity of the reflected light due to capillary blood volume changes as the pulse pressure wave transits the sensor field of view. The use of actigraphy, heart rate, and heart rate variability together provides the opportunity to estimate more than just wake/sleep, such as light (N1 and N2), deep, and REM stages of sleep.

Although many wearables purport to accurately distinguish between light and deep sleep 3, little information is available on the validity of using wearables in this manner. A previous study by Fitbit described a non-production method to estimate sleep stages in a healthy adult population using PPG and actigraphy in a wearable device 4. This study describes the method and results of using optical PPG and actigraphy in a wearable device (a vivosmart 3 manufactured by Garmin International) to estimate sleep stages in a population of adults.

Method

This study involved recruiting subjects to wear a Garmin vívosmart 3 and a reference device so that the accuracy of the Garmin device in estimating sleep stages could be assessed. The most accurate method to determine sleep stages involves the use of in-laboratory polysomnography (PSG), which includes the use of EEG, EOG, and EMG sensors. However, the use of an in-laboratory PSG is cumbersome for the subjects, and many times does not yield the same quality and quantity of sleep as a subject sleeping in familiar surroundings in their own bed. EEG systems that are take home devices such as the Sleep Profiler have been developed and yield comparable sleep architecture estimates to PSG 5. In this study, the Sleep Profiler was used as the reference device, reducing burden on study participants and allowing for data collection in an environment identical to production use of the Garmin wearable.  The accuracy of the Sleep Profiler was deemed to be comparable to a laboratory PSG.

Individuals in the study were trained on use of the Sleep Profiler device and directed to wear the Garmin device on the nondominant wrist. The Sleep Profiler device recorded three channel EEG for sleep staging (as well as other measurements such as heart rate). The Garmin device logged data simultaneously with the Sleep Profiler. Accelerometer xy, and z channels were recorded, along with the wide bandwidth PPG signal. Only the green channel was used in this study. Data was time aligned to within a second by setting the internal clocks of both devices to the same time, and then comparing the heart rate output of the Garmin device to the heart rate output of the Sleep Profiler, as these two measurements were very similar, highly correlated, and yielded time alignments of better than one second.

The Sleep Profiler output data, including the automated sleep staging output, was then reviewed by a Registered Polysomnographic Technologist using AASM guidelines [6] to provide the reference sleep stages. Scoring was done at the 30-second epoch level. In this case, N1 and N2 sleep stages were combined into a light sleep category. Wake, light, deep (N3) and REM stages were then scored.

The scored data was then used to train a neural network to perform the sleep stage classification. The neural network architecture was chosen such that it could be realized in a production system that included the wearable device, with somewhat limited computing resources, and the Garmin cloud, which has extensive computing resources. In the end, a two-stage approach was used, where data was pre-processed on the wearable to segment and compress the wide bandwidth sensor data into a manageable payload for syncing off the device via Bluetooth low energy (BLE) and the Garmin Connect Mobile app to the Garmin cloud for further processing.

The neural network used in the cloud was an adaptation and simplification of the architecture of Zhao 7. Over 60 features were computed on the device. Seven more features were computed on the cloud over the entire night's data.

The network was trained using a k-fold cross validation re-sampling procedure with a k value of five 8. In this method, the data was randomly sampled into five equally sized groups. The network was trained on four of the groups (80%), and scored on remaining group (20%). The training set was randomly subdivided into a training set and a validation set used to determine when to stop training the network. The stopping criteria used was a variation of the criteria given in 9. This method was repeated five times such that each subject was used four times in the training set, and one time in the test set.   

Results

55 individuals were recruited to participate in the study. The only exclusion criterion was if there was a tattoo on the dorsum of wrist that would interfere with the optical heart rate reading. Of these 55 individuals, 14 indicated that they have a sleep disorder or take medications that may impact sleep architecture. These individuals were noted, however their data was still included in the study to represent the general population and a percentage of real-world users with sleep disorders, whether known or unknown. Of the 55 participants, 31% (17) were female and 69% (38) were male. Age distribution was 35.5 +/- 8.7 years, BMI distribution was 26.3 +/-6.4.

The overall accuracy of the algorithm against this real-world data was 69.7%, with the sensitivity in detecting sleep 95.8% and the specificity in detecting awake 73.4% (Table 1). The Cohen's kappa (k) value is 0.54 +/- 0.12. These results should be assessed in light of the fact that the inter-scorer agreement between two trained human scorers using PSG is approximately 83% 10 with a Cohen's kappa of approximately 0.78 4.

Table 1. Sensitivity, Specificity, Accuracy, and
Cohen's Kappa of Wearable Sleep Stage Classifier

Sensitivity     95.8%
Specificity73.4%
Overall accuracy (correctly classifying sleep stage)69.7%
Cohen's kappa0.54 +/- 0.12

The full confusion matrix for the sleep stage classifier is given in Table 2. The matrix shows that that the classifier predicts deep, light, and REM sleep stages at roughly the same 69% accuracy rate. Wake is slightly more accurate at 73%. The most common mis-classifications are classifying true deep sleep as light sleep and classifying true REM sleep as light sleep.

Table 2. Confusion Matrix for All Participants
True Stage vs. Predicted Stage

 True DeepTrue LightTrue REMTrue Awake
Predicted Deep68.9%12.0%1.4%1.9%
Predicted Light29.1%68.6%26.4%14.9%
Predicted REM0.9%13.7%69.8%9.7%
Predicted Awake1.1%5.7%2.3%73.4%

A comparison between the reference device and the wearable device of the overall sleep architecture for all 55 participants is given in Table 3.

Table 3. Sleep Architecture for All Participants

 % Deep% Light% REM% Awake
Garmin
Wearable
15.5%46.0%21.6%16.9%
Sleep
Profiler
12.7%50.7%18.3%18.3%

It is important to note that the algorithm described in this paper was deployed into production on many Garmin wearable devices, including the vívosmart 3. During beta testing of the production implementation of the wearable sleep stage classifier, an additional 3,200 nights of data were analyzed and reviewed to ensure the results aligned with those of the 55-person study. There was no accompanying reference device data with these 3200 nights – only the production implementation using actigraphy and optical heart rate was used. On this larger cohort, the distribution of sleep stages as a function of age was used to assess algorithm performance. REM sleep distribution remained relatively constant as a function of age, deep sleep decreased markedly as a function of age, and total sleep time generally decreased with age – as expected 11.

Sleep tracking by age group

Figure 1. Sleep Duration – Average by Age Group in Production System

Finally, some representative results from the cohort of 55 subjects are shown below. In the following figures, wake is represented as stage 3, REM as stage 2, light (N1 and N2) as stage 1, and deep as stage 0. Figure 2 shows the median result (27th of the 55 data sets when rank ordered by accuracy), with an accuracy of 71.5%.

Sleep tracking

Figure 3 shows the best result of the cohort (first of 55), with an accuracy of 80.1%

Sleep tracking

Figure 3. Best Result, Accuracy 80.1%, Cohen's Kappa 0.70

Figure 4 shows the worst result of the cohort (55 of 55), with an accuracy of 49.9%. Note that this user self-reported having a sleep disorder.


Figure 4.
Worst Result, Accuracy 49.9%, Cohen's Kappa 0.18 (Self-Reported Sleep Disorder)

Discussion

The work provides evidence that a wearable device using only actigraphy and optical heart rate sensors can estimate sleep stages of users to nearly 70% accuracy with a good Cohen's kappa of 0.54 +/- .12. This compares closely to the results in 4, despite the increased challenges of production systems and real-world data collection. It also compares favorably, but does not reach the same level of accuracy, as EEG based sleep stage measurement devices, which reach an inter-scorer accuracy level of 82%.

The major benefit to using a wrist wearable device to estimate sleep stages is the very low level of user friction involved – a user can simply wear the device on the wrist during sleep without cumbersome, special equipment. Use of a wearable for estimating sleep stages accurately allows for monitoring of sleep for virtually an unlimited number of nights, assuming the device is kept charged during the day, providing a user with useful data about sleep hygiene and sleep architecture.

References

  1. P Lamkin, Smart Wearables Market to Double by 2022: $27 Billion Industry Forecast," Forbes, October 23, 2018. online. Available:  https://www.forbes.com/sites/paullamkin/2018/10/23/smart-wearables-market-to-double-by-2022-27-billion-industry-forecast/#419523e92656  [Accessed April 25, 2019]
  2. L de Souza, AA Benedito-Silva, ML Pires, D Poyares, S Tufik, HM Calil, "Further validation of actigraphy for sleep studies," Sleep. 2003 Feb 1;26(1):81-5.
  3. J Mantua, N Gravel, R M. C. Spencer, "Reliability of Sleep Measures from Four Personal Health Monitoring Devices Compared to Research-Based Actigraphy and Polysomnography," Sensors (Basel). 2016 May; 16(5): 646.
  4. Z Beattie, Y Oyang, A Statan, A Ghoreyshi, A Pantelopoulos, A Russell, C Heneghan, "Estimation of sleep stages in a healthy adult population from optical plethysmography and accelerometer signals," Physiol. Meas. 38 (2017) 1968–1979.
  5. P Finan, J Richards, C Gamaldo, D Han, JM Leoutsakos, R Salas, M Irwin, M Smith, Validation of a Wireless, Self-Application, Ambulatory Electroencephalographic Sleep Monitoring Device in Healthy VolunteersJ. Clin. Sleep Med. 12 1443–50.
  6. Berry R B, Brooks R, Gamaldo C E, Harding S M, Lloyd R M, Marcus S L, Vaughn B V 2015 The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications, version 2.5 (Darien, IL: American Academy of Sleep Medicine).
  7. M Zhao, S Yue, D Katabi, T Jaakkola, M Bianchi, "Learning Sleep Stages from Radio Signals: A Conditional Adversarial Architecture," International Conference on Machine Learning (ICML'17).
  8. J. Brownlee, "A Gentle Introduction to k-fold Cross Validation", Statistical Methods, May 23, 2018. online. Available: https://machinelearningmastery.com/k-fold-cross-validation/. [Accessed April 25, 2019].
  9. L. Prechelt, "Early Stopping – but when", online. Available: https://page.mi.fu-berlin.de/prechelt/Biblio/stop_tricks1997.pdf. [Accessed April 25, 2019].
  10. R RosenbergS Van Hout, "The American Academy of Sleep Medicine Inter-scorer Reliability Program: Sleep Stage Scoring," J Clin Sleep Med. 2013 Jan 15; 9(1): 81–87.
  11. M Ohayon, M Carskadon, C Guilleminault, M Vitiello, "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):1255-73.

"If you always put limits on everything you do, physical or anything else, it will spread into your work and into your life. There are no limits. There are only plateaus, and you must not stay there, you must go beyond them." - Bruce Lee

  • 5
  • 2
5mo 
Harvard_GS_HBS_KKR, what's your opinion? Comment below:

Anyone consistently get poor sleep yet still perform extremely well and have very great cognition in daily life? I suspect I have sleep issues but have never stumbled in academics or at work enough to feel compelled to see a doctor, but am worried that issues could arise in the future.

Learn More

300+ video lessons across 6 modeling courses taught by elite practitioners at the top investment banks and private equity funds -- Excel Modeling -- Financial Statement Modeling -- M&A Modeling -- LBO Modeling -- DCF and Valuation Modeling -- ALL INCLUDED + 2 Huge Bonuses.

Learn more
5mo 
ThreeStripes, what's your opinion? Comment below:

On average? Maybe 4-5 hours. I try to sleep for 7-8, don't know what the ratio's supposed to be.

  • 1
Most Helpful
5mo 
Isaiah_53_5 💎🙌💎🙌💎, what's your opinion? Comment below:
ThreeStripes

On average? Maybe 4-5 hours. I try to sleep for 7-8, don't know what the ratio's supposed to be.

4-5hrs of deep sleep is insane! I want to see a screenshot of that lol.

https://www.sleepfoundation.org/stages-of-sleep/deep-sleep

"To calculate how much deep sleep you need, first determine how much sleep you need overall. Most adults should aim for seven to nine hours17 of sleep each night. Between 13% and 23%18 of that time should be spent in deep sleep. If you get seven hours of sleep each night, then you spend approximately 55 to 97 minutes each night in deep sleep.

To a certain extent, the body self-regulates amounts of deep sleep. For example, you might spend more time in deep sleep if you are recovering from sleep deprivation or if you regularly experience short sleep, such as over the course of a work week19."

"If you always put limits on everything you do, physical or anything else, it will spread into your work and into your life. There are no limits. There are only plateaus, and you must not stay there, you must go beyond them." - Bruce Lee

  • 3
4mo 
CRE Finance, what's your opinion? Comment below:

Christ, Isaiah. I think you just broke the record for the longest comment ever. 

“Bestow pardon for many things; seek pardon for none.”
  • 1
5mo 
Kairo, what's your opinion? Comment below:

6 or 7. Longest I've gone without sleep is 3 days

  • 1
4mo 
ImALawyer, what's your opinion? Comment below:

Voluptatem consequuntur incidunt voluptas rerum consequatur. Libero at deserunt numquam quia voluptas. Consequatur error sint recusandae molestiae at natus sit. Nulla est sed veniam perspiciatis ipsa et. Praesentium laudantium itaque reprehenderit tempora. Fuga numquam iure earum deleniti unde beatae alias.

Voluptatem sed quisquam dolorem sapiente ducimus ab voluptatem quis. Hic eum ratione culpa nobis. Animi ut in adipisci error exercitationem. Iusto ut est esse deleniti. Quia eius illum qui voluptas alias occaecati.

Start Discussion

Career Advancement Opportunities

February 2023 Investment Banking

  • Lazard Freres (+ +) 99.5%
  • Jefferies & Company (▽01) 99.1%
  • Lincoln International (▽01) 98.6%
  • Financial Technology Partners (▽01) 98.1%
  • William Blair (▲08) 97.7%

Overall Employee Satisfaction

February 2023 Investment Banking

  • Canaccord Genuity (▲04) 99.5%
  • William Blair (▲04) 99.1%
  • Lincoln International (▲09) 98.6%
  • Jefferies & Company (▲06) 98.1%
  • Financial Technology Partners (▲09) 97.6%

Professional Growth Opportunities

February 2023 Investment Banking

  • Lazard Freres (▲15) 99.5%
  • Financial Technology Partners (▲09) 99.1%
  • Lincoln International (= =) 98.6%
  • Jefferies & Company (▽03) 98.1%
  • William Blair (▲01) 97.7%

Total Avg Compensation

February 2023 Investment Banking

  • Director/MD (6) $592
  • Vice President (25) $422
  • Associates (137) $262
  • 3rd+ Year Analyst (9) $194
  • 2nd Year Analyst (80) $172
  • 1st Year Analyst (258) $171
  • Intern/Summer Associate (42) $166
  • Intern/Summer Analyst (186) $91