Assay of the dreem device on sleep metrics and an exploration of sleep staging in chronic short sleepers during time in bed extension

Despite clear and plentiful research that sleeping less than seven hours per night has a wide array of health consequences, a large portion of American adults report sleeping less than seven hours per night and thus receive chronic insufficient sleep. Many studies exploring the consequences of insufficient sleep are restricted to small sample sizes and short recording times due to a significant cost to gold-standard polysomnography in terms of expense, time, and reliance on trained sleep technicians to prepare and monitor subjects. Additionally, most studies adopt a design of interventional sleep restriction on otherwise healthy sleepers, which excludes people who receive long term insufficient sleep over months to years. Here, we attempt to explore possible solutions to these issues through the use of a sleep extension study using the Dreem headband, a wireless dry electrode consumer electroencephalography (EEG) device, to measure overall sleep metrics and EEG data. When compared to wrist-mounted actigraphy, the Dreem indicates little systemic skew for data over 75% quality (as assigned by Dreem), but reports significant random error with limits of agreement starting approximately 70 minutes off of actigraphy baseline. Exploration of sleep metrics in baseline insufficient sleep vs interventional sleep extension revealed an increase in total sleep time; increase in all recorded sleep stages; and no significant changes in sleep onset latency, wakefulness after sleep onset, or sleep efficiency. Although several limitations of producing high quality data were identified, the Dreem headband shows promise as a home environment sleep research device. With an improvement in data iii quality the Dreem, or another wireless consumer sleep device, has the potential to help advance the sleep field in ways that have traditionally proven inaccessible.

ABSTRACT Despite clear and plentiful research that sleeping less than seven hours per night has a wide array of health consequences, a large portion of American adults report sleeping less than seven hours per night and thus receive chronic insufficient sleep. Many studies exploring the consequences of insufficient sleep are restricted to small sample sizes and short recording times due to a significant cost to gold-standard polysomnography in terms of expense, time, and reliance on trained sleep technicians to prepare and monitor subjects. Additionally, most studies adopt a design of interventional sleep restriction on otherwise healthy sleepers, which excludes people who receive long term insufficient sleep over months to years. Here, we attempt to explore possible solutions to these issues through the use of a sleep extension study using the Dreem headband, a wireless dry electrode consumer electroencephalography (EEG) device, to measure overall sleep metrics and EEG data. When compared to wrist-mounted actigraphy, the Dreem indicates little systemic skew for data over 75% quality (as assigned by Dreem), but reports significant random error with limits of agreement starting approximately 70 minutes off of actigraphy baseline. Exploration of sleep metrics in baseline insufficient sleep vs interventional sleep extension revealed an increase in total sleep time; increase in all recorded sleep stages; and no significant changes in sleep onset latency, wakefulness after sleep onset, or sleep efficiency. Although several limitations of producing high quality data were identified, the Dreem headband shows promise as a home environment sleep research device. With an improvement in data ii quality the Dreem, or another wireless consumer sleep device, has the potential to help advance the sleep field in ways that have traditionally proven inaccessible. iii TABLE OF CONTENTS ABSTRACT ii INTRODUCTION 1 METHODS 4 RESULTS 8 DISCUSSION 13 CONCLUSION 16 REFERENCES 17 iv INTRODUCTION Approximately 35% of American adults report sleeping less than 7 hours per night, the minimum recommended by the American Academy of Sleep Medicine [1, 10]. Insufficient sleep is a risk factor for diverse health issues including diabetes, obesity, poor mental health, and impaired cognitive performance [1-4]. Typically, sleep research in the laboratory is based on the paradigm of restricting sleep in healthy young adults to 4-5 hours of sleep per night for 1-14 nights. This is followed by a period (~2-7 days) of recovery sleep to relieve the sleep restriction. This paradigm allows precise analyses of the health consequences of experimental sleep restriction in otherwise healthy adults. However, this type of sleep research is restricted in scope, both in study duration and in participant number, due to the cost and burden of labor-intensive sleep restriction where participants live in the laboratory 24 hours per day. Thus, it is not feasible or ethical to conduct months-long studies of experimental sleep restriction in the laboratory. As such, the development of a noninvasive method for large scale at home sleep research has large potential to advance the entire field of sleep research. Not only is this a more ethical approach for long-term research, studying sleep in the home-environment is more ecologically valid compared to artificially manipulating sleep in the constructed laboratory setting. Polysomnography (PSG) remains the gold standard of in-laboratory and at home sleep, due to its superior reliability and validity. PSG requires the application of electroencephalography (EEG) electrodes to the scalp, electromyography (EMG) electrodes to the facial muscles and limbs, electrooculography (EOG), electrocardiography (EKG), pulse oximetry, thoracic and abdominal belts, and nasal (or 1 oro-nasal) flow sensors to measure breathing patterns [18]. EEG data of brain activity is used to determine the stage of sleep that the participant is in. Stages are broadly represented as stage one (light sleep), stage two (moderate sleep), stage three (deep sleep), and REM (rapid eye movement) sleep. PSG is often used to determine homeostatic sleep pressure using delta power, which is a measure of slow wave activity (neural pulses between 0.5-4.5 hertz). Slow wave activity is thought to represent sleep pressure due to observations of its increase during restricted sleep, and its decrease during recovery sleep in laboratory settings. Despite its strengths, PSG is expensive, requires technical experience to apply and monitor, and involves the nightly dermal application of 20-50 electrodes with individual wires. All of these factors make PSG unfeasible for an at home environment without specialized technicians onsite, especially over time frames longer than a week. Ambulatory dry EEG devices offer a possible solution to this issue, combining some aspects of PSG with a more convenient and user-friendly design. This combination has the potential to allow for high quality EEG data to be gathered without the financial and technical limitations of PSG, and has the potential to significantly advance sleep research. The Dreem 2 headband is one such dry-electrode consumer EEG device. Preliminary data show the Dreem 2 headband provides sufficient quality sleep recordings and that the algorithm provided by Dreem is as accurate as certified Sleep Technicians scoring in-laboratory PSG data. It should be noted however that most of this data is derived from studies funded and conducted by the Dreem company themselves [11-13]. Additionally, these studies were conducted in a controlled environment with sleep technicians present, not with free-living participants in a home environment where the 2 participant has full control over the Dreem headband. The goal of our pilot study is to explore the quality of data recorded by the Dreem 2 headband device in a hands-off, at home environment, to give initial independent assessment into the efficacy of such a device in its intended use case. As we are broadly interested in the health consequences of short sleep duration, we focus our analyses on participants who voluntarily choose to sleep less than 6.5 hours a night, allowing us to sample a population that is rarely studied in sleep research, but is significantly impacted by insufficient sleep. Our analyses are derived from a larger study that involves two weeks of baseline assessments followed by a 4-week sleep extension intervention. Therefore, we will analyze Dreem 2 data from both the baseline and sleep extension study segments. 3 METHODS Protocol The parent study is a longitudinal, single-arm experimental protocol designed to examine metabolic changes in response to sleep extension in adults aged 18-35 years who report sleeping less than 6.5 hours per night. The goal of sleep extension was to achieve ≥8 hours of time in bed per night for 4 weeks. Participants were screened for physical and mental illness, as well as sleep disorders by an overnight clinical PSG assessment at the University of Utah Sleep Wake Center. Fifteen subjects were recruited for the Dreem 2 portion of the study. One participant was removed from our analyses due to an inability to wear the Dreem headband. Sleep and activity were monitored throughout the study by wrist-actigraphy (Actiwatch Spectrum Plus; Philips Inc.). A week of baseline EEG data was recorded using the Dreem 2 headband (study days 7-14) in the participants home environment. During this baseline segment participants were instructed to maintain their habitual short sleep schedules. After baseline, participants were instructed to extend their time in bed to ≥8 hours per night in their home environment for the next 4 weeks. Sleep duration was logged by electronic sleep diary and wrist-actigraphy. Weekly communication from the study team helped participants maintain adherence to the sleep extension schedule. One week of EEG data was collected similarly to the baseline segment with the Dreem 2 headband recording days 35-42 in the participants home environment. Day 43 involved a possible overnight sleep visit in the hospital depending on availability and current COVID-19 restrictions. 4 Dreem 2 Headband Participants were given initial instruction, both verbal and written, on how to wear the Dreem headband most effectively, and were sent a follow-up text if initial data had low quality as calculated by the Dreem algorithm. Participants were instructed to wear the headband nightly unless it proved prohibitively disruptive to sleep. The Dreem 2 headband is composed of a mix of flexible plastic panels and soft fabric headband, with an elastic strap mounted in the back. Battery life is roughly 12 hours. It records data through 6 EEG electrodes, with 4 dry silicone rubber electrodes along the forehead and 2 proprietary electrodes positioned at the base of the back of the skull. This data, combined with an accelerometer and a pulse sensor, are initially assigned a quality from 0-100 by the onboard Dreem computer. The computer then takes the two highest quality EEG channels and uses a Dreem developed deep learning algorithm to automatically assign a sleep stage to each 30s epoch. Each night and each EEG channel within each night, are assigned a quality score at the end of the recording based on the estimated percent of the recording that is scorable. Quality over 85% is listed as green, quality between 85%-70% is listed as yellow, and quality below 70% is listed as red. Data is uploaded to Amazon A3 servers owned by Dreem, where it is stored and can be accessed. Bland-Altman Plots To compare nightly total sleep time measured by the Dreem 2 versus wristactigraphy we used Bland-Altman plots. Dreem and actigraphy data was processed through R using a data package (BAplot.R, R version 4.2.1) for the production of BlandAltman plots [14]. We assigned the Actiwatch data as the reference data and the Dreem 2 5 as experimental data. Data was processed using bootstrap to account for the low n value, and data was log transformed to minimize possible impacts of proportional bias of sleep duration, consistent with current recommendations. Power Spectral analysis Following our determination that Dreem quality over 70% was sufficient for further data analysis, we attempted to compare nightly delta power in baseline vs sleep extension conditions. Due to the individual nature of delta power, we determined that a minimum of 4 nights of 70% or higher quality data for both conditions was necessary. As only three participants satisfied this metric, we determined that a case study of those three participants would be the most useful in determining our ability to effectively perform a delta power analysis. Recorded EEG data for three case study participants was processed by Dreem algorithms, and an automatic hypnogram was generated. To ensure spatial continuity of delta power, we restricted the assay to one channel, either F7-O1 or F8-O2. This had the consequence of reducing our available channels from 4 to one, which had a significant impact on the overall quality of our data. We used the BrainVision Analyzer 2 software to process EEG data, with a raw data analysis using filters for gradient (50μV/ms max), amplitude (minimum of -200 μV and a maximum of 200 μV), and low activity (minimum activity max-min of 0.5 μV). Only the first three NREM episodes of each night were analyzed to ensure the same number of sleep cycles were included in each night of data, consistent with other studies in the field. The beginning of a NREM episode was defined as 2 consecutive 30 second epochs of S1 sleep, or one epoch of S2 sleep. A NREM episode ended upon one epoch of REM sleep, and a second NREM 6 episode was not considered until at least 15 minutes had elapsed between REM episodes. We used Analyzer 2’s raw data inspection tool to remove artifacts from the data, removing an average of 25.4% of the total data. We then performed a Fast Fourier Transformation of average delta power (0.5-4.5 hertz) to assess power spectra (slow wave activity). Statistical analysis All statistical processing was performed in R (ver 4.2.1) using Lmer4 and Multcomp packages. Models were made using linear mixed effects models with subject as a random variable. 7 RESULTS Fourteen participants, 8 male and 6 female, completed the Dreem headband analysis portion of our study. Their age was 20.6±2.5y (mean±SD), with an average BMI of 22.6 (at study enrollment medical screening visit). Data was collected between 8/5/2021 and 11/12/2022. Dreem 2 quality metrics describe overall skew, but have little bearing on random error Before delving into possible sleep metrics, an initial landmark of relative Dreem data quality needed to be established. Gold-standard PSG would have been the optimal choice for determining the accuracy of the Dreem 2 Device, but PSG was unavailable for this home-based study. However, wrist mounted actigraphy is widely accepted in sleep research as an accurate measure of total sleep time in the ambulatory environment, so we compared Dreem derived total sleep time (TST) to actiwatch TST. Here we compare all nights of data from the Dreem 2 versus wrist-actigraphy for the baseline and sleep extension segments of the study. When sorting data by Dreem assigned quality metrics, low quality data (reported as less than 70% scorable) showed significantly more systemic error against wrist-actigraphy, with negative relative skew increasing with total sleep time (fig 1). It should be noted that there are few data points beyond a total sleep time of 480 minutes, and that several data points show significant deviation from the trend. When analyses were limited to only high-quality data (more than 85% scorable), there was minimal systemic error between the Dreem 2 data and wrist-actigraphy data. However, 8 Analysis of Sleep metrics Next, we analyzed relative sleep metrics for the baseline and sleep extension segments. Based on the generated Bland-Altman plots we determined that using data of 75% quality or higher allowed for a balance of including enough data while still eliminating systemic skew due to low quality data. We performed a linear mixed effects regression (LMER) to compare total sleep time for baseline and sleep extension conditions, using subject ID as a random variable to account for individual subject effects. The intercept for our baseline condition was 308±14.7 minutes (Estimate ± Standard error) of TST, with an increased TST by 101 minutes (standard error 14.5 minutes, p value of <0.001) during sleep extension. In other words, these results show participants achieved sleep extension with 5:08 hours of TST at baseline and 6:49 hours of TST during sleep extension. We then moved on to analysis of Dreem reported sleep staging across the night (Table 1). Sleep Stage Baseline (minutes Sleep extension Sleep extension p value ± SEM) segment (minutes ± SEM) N1 17.9 ± 1.8 22.9 ± 1.7 0.003 N2 134.6 ± 11.9 192.7 ± 10.6 <0.001 N3 71.9 ± 6.0 91.1 ± 4.4 <0.001 REM 82.5 ± 7.6 100.4 ± 6.7 0.007 Table 1: LMER of individual sleep stages as reported by the Dreem scoring algorithm. The Dreem reported significant increases in all sleep stages across the night in sleep extension versus baseline conditions, with stage one sleep increasing the least and stage two increasing the most. In the context of the roles of stage one and two sleep, this 10 aligns with expectations. However, stage three sleep (also called deep sleep) is reported as increasing more than REM sleep. This result is somewhat unexpected, as stage three sleep occurs early in the night, while REM sleep primarily occurs later in the sleep cycle and therefore we hypothesized there would be a greater increase in REM sleep. We did not detect significant changes in sleep onset latency (SOL) (p=0.559), sleep efficiency (SE) (p=0.283), or wake after sleep onset (WASO) (p=0.373). Delta Power Analysis Finally, we performed a proof of concept case study to determine how granularly we could process the Dreem 2 data. We selected three subjects with the highest overall data quality for baseline and sleep extension, and used only nights of greater than 70% total quality. Due to spatial restriction of delta power requiring a consistent use of only two of the Dreem channels, we found that overall data quality was not always reflective of individual channel quality, with only 23% of nights providing a channel with greater than 70% quality in either the F7-O1 or F8-O2 channels. We processed all nights through a series of filters and a fast Fourier Transformation to establish delta power [Fig 2]. Note that our additional data processing for this step flagged over 25% of the data as artifact on average, with a maximum removal of 56.5% and a minimum removal of 5.5%. With this level of quality, it is difficult to draw any meaningful conclusion from calculated delta power. Two participants showed delta power increase of approximately one μV^2, and the third showed a decrease by one μV^2. Despite the poor-quality data, we established that delta power can be extracted from the Dreem headband, but additional steps would 11 need to be taken in either a hardware or software domain to ensure that data quality was acceptable. Fig 2: An example FFT produced by BrainVision Analyzer 2 for a 30s epoch of data, in μV^2/Hz. This particular FFT was derived from the F7-O1 channel. 12 DISCUSSION A major driving force for this research was to act as a pilot study for the field of sleep research as a whole in order to determine the current obstacles to using wireless consumer EEG devices. Participants were given initial instruction, both verbal and written, on how to wear the Dreem headband most effectively. We monitored nightly data quality and when a night had less than 70% quality we followed up with a text message to the participant explaining the recording had low quality and we provided tips and tricks to try and improve overall quality. Despite this, quality of recordings was highly variable. 58.5% of all nights were above 70% quality (marked as yellow or green), and 30.7% were above 85% quality (marked as green). Only 140 nights of data were recorded, 56 less than expected for two weeks of 14 participants. Thus, there was 29% missing data prior to any data processing. Nights were lost primarily due to user error by participants, either through an inability to log into the necessary companion app to initiate recording or failure to begin recording during a night. Bland-Atman Plots A possible factor in the systemic skew, and the widening LOA, seen at higher recording times could be an interplay between the metrics used for determining sleep in the actiwatch and Dreem headband. The actiwatch uses primarily motion to assign sleep values, supplemented by user created waypoints that indicate when the participant is attempting to sleep and wake-up. However, the Dreem headband makes these determinations based on EEG activity paired with heart rate and motion data. It is 13 possible that in nights with higher sleep opportunities (i.e., time in bed), periods of quiet wakefulness before or after sleep were logged as sleep by the actiwatch and as wakefulness by the Dreem headband. This would align with consistent underreporting of the Dreem relative to the actiwatch at higher TST. Pairing the Dreem headband with gold-standard PSG would be useful in exploring to what degree the Dreem or actiwatch truly become less accurate with extended sleep times. Sleep Metrics Our observation of sleep metrics raised a few interesting questions. First was the observation that stage three sleep and REM sleep seemed to increase roughly the same amount on average in the sleep extension condition (19.2 vs 17.9 min, or 26.7% vs 21.7% respectively). This is an unexpected result as typically in sleep restriction studies, the greatest deficit in the restricted sleep time is REM sleep, while stage three sleep remains relatively unchanged due to its typical occurrence earlier in the sleep episode (15-17). Therefore, the result that REM sleep increased less than stage three sleep is puzzling, and merits further study in a sleep extension model. If sleep extension and sleep restriction models do not agree, particularly for chromic short sleepers, this may point to a difference of sleep staging or sleep power for people who chose routine short sleep. In other words, there are likely different sleep needs between individuals and understanding individual sleep need could help support better public health recommendations for achieving optimal sleep. Another result of note is the lack of significance for SOL, SE, or WASO. In the sleep extension condition, a significant drop to SE or a significant increase in SOL or 14 WASO would indicate that extended sleep resulted in a lowered sleep pressure to the degree that sleep quality began to deteriorate. As we do not see this, we can continue to operate under the assumption that the extended sleep phase is producing longer sleep of similar quality. More research is needed to determine if this indicates that extended sleep is beneficial for chronic short sleepers, and in what regards. For example, it is unclear at this time if extending sleep in otherwise healthy young adults who get habitual short sleep duration can help reduce their lifelong risk for chronic disease like diabetes, cardiovascular disease, or Alzheimer's Disease. Delta Power Our exploration of possible delta power analysis using the data collected by the Dreem headband revealed several significant issues. Dreem calculates total recording quality for a night based on all channels, meaning that only a single channel needed to be of high quality for the entire recording to register as high quality. This became challenging when trying to restrict channel use to a single spatial area, as delta power is well documented to vary by cranial region. Thus, if the only good quality channel was outside our spatial restriction, a night of initially over 85% quality could yield usable channels of significantly below 70% quality. We observed a connection between channel quality and percent data excluded by the BrainVision Analyzer 2 software, indicating again that Dreem assigned quality has some bearing on actual data usability. 15 CONCLUSION We explored the feasibility of the Dreem Headband for use in at home sleep extension studies. Initially we compared the Dreem headband to actigraphy using a Bland-Altman analysis, where we found significant systemic skew in data with a Dreem assigned quality of lower than 75%. We also established limits of agreement that are narrowest with shortest sleeping times, beginning at roughly 70 minutes above or below actigraphy reported times and widening with increasing sleep duration. We then analyzed sleep metrics for the baseline vs sleep extension conditions, where we observed a significant increase in total sleep time and all sleep phases, notably with a proportionally larger increase in stage three sleep vs REM sleep. We also explored a case study of three individuals with high quality data to determine the feasibility of performing delta power analysis using the Dreem headband, and found no direct software or hardware obstacles. However, data quality declined with the increasing specificity, becoming prohibitively poor when restricting to a single channel. In all, our analyses show the Dreem headband is promising as an at-home sleep monitoring device for sleep research. The primary limiting factor is data quality and this appears at least somewhat dependent on the individual user and highly variable across individuals. With continued work to improve data quality the Dreem has high potential to help advance the sleep field on the whole. 16 REFERENCES 1. Watson NF, Badr MS, Belenky G, et al. 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. 2. Van Cauter, E., Spiegel, K., Tasali, E., & Leproult, R. (2008). Metabolic consequences of sleep and sleep loss. Sleep medicine, 9 Suppl 1(0 1), S23–S28. https://doi.org/10.1016/S1389-9457(08)70013-3 3. Tomaso, C. C., Johnson, A. B., & Nelson, T. D. (2021). The effect of sleep deprivation and restriction on mood, emotion, and emotion regulation: three metaanalyses in one. Sleep, 44(6), zsaa289. https://doi.org/10.1093/sleep/zsaa289 4. Kaliyaperumal, D., Elango, Y., Alagesan, M., & Santhanakrishanan, I. (2017). Effects of Sleep Deprivation on the Cognitive Performance of Nurses Working in Shift. Journal of clinical and diagnostic research : JCDR, 11(8), CC01–CC03. https://doi.org/10.7860/JCDR/2017/26029.10324 5. Gottlieb, D. J., Ellenbogen, J. M., Bianchi, M. T., & Czeisler, C. A. (2018). Sleep deficiency and motor vehicle crash risk in the general population: a prospective cohort study. BMC medicine, 16(1), 44. https://doi.org/10.1186/s12916-018-1025-7 6. Cohen, D. A., Wang, W., Wyatt, J. K., Kronauer, R. E., Dijk, D. J., Czeisler, C. A., & Klerman, E. B. (2010). Uncovering residual effects of chronic sleep loss on human performance. Science translational medicine, 2(14), 14ra3. https://doi.org/10.1126/scitranslmed.3000458 7. Van Dongen HP, Maislin G, Mullington JM, Dinges DF. The cumulative cost of additional wakefulness: dose-response effects on neurobehavioral functions and sleep 17 physiology from chronic sleep restriction and total sleep deprivation. Sleep. 2003 Mar 15;26(2):117-26. doi: 10.1093/sleep/26.2.117. Erratum in: Sleep. 2004 Jun 15;27(4):600. PMID: 12683469. 8. Belenky G, Wesensten NJ, Thorne DR, Thomas ML, Sing HC, Redmond DP, Russo MB, Balkin TJ. Patterns of performance degradation and restoration during sleep restriction and subsequent recovery: a sleep dose-response study. J Sleep Res. 2003 Mar;12(1):1-12. doi: 10.1046/j.1365-2869.2003.00337.x. PMID: 12603781. 9. Patricia Tassi, Anne Bonnefond, Ophélie Engasser, Alain Hoeft, Roland Eschenlauer, Alain Muzet. EEG spectral power and cognitive performance during sleep inertia: The effect of normal sleep duration and partial sleep deprivation. Physiology & Behavior, Volume 87, Issue 1, 2006, Pages 177-184, ISSN 0031-9384 https://doi.org/10.1016/j.physbeh.2005.09.017 10. Centers for Disease Control and Prevention. (2014). Short Sleep Duration Among US Adults. https://www.cdc.gov/sleep/data_statistics.html. 11. Pierrick J Arnal, Valentin Thorey, Eden Debellemaniere, Michael E Ballard, Albert Bou Hernandez, Antoine Guillot, Hugo Jourde, Mason Harris, Mathias Guillard, Pascal Van Beers, Mounir Chennaoui, Fabien Sauvet, The Dreem Headband compared to polysomnography for electroencephalographic signal acquisition and sleep staging, Sleep, Volume 43, Issue 11, November 2020, zsaa097, https://doi.org/10.1093/sleep/zsaa097 12. V Thorey, A Guillot, K El Kanbi, M Harris, P J Arnal, 1211 Assessing the Accuracy of a Dry-EEG Headband for Measuring Brain Activity, Heart Rate, Breathing 18 and Automatic Sleep Staging, Sleep, Volume 43, Issue Supplement_1, April 2020, Page A463, https://doi.org/10.1093/sleep/zsaa056.1205 13. Debellemaniere E, Chambon S, Pinaud C, Thorey V, Dehaene D, Léger D, Chennaoui M, Arnal PJ and Galtier MN (2018) Performance of an Ambulatory Dry-EEG Device for Auditory Closed-Loop Stimulation of Sleep Slow Oscillations in the Home Environment. Front. Hum. Neurosci. 12:88. doi: 10.3389/fnhum.2018.00088 14. Menghini, Luca, et al. “A Standardized Framework for Testing the Performance of Sleep-Tracking Technology: Step-by-Step Guidelines and Open-Source Code.” Sleep, Oxford Academic, 10 July 2020, https://academic.oup.com/sleep/article/44/2/zsaa170/5901094. 15. Skorucak J, Arbon EL, Dijk DJ, Achermann P. Response to chronic sleep restriction, extension, and subsequent total sleep deprivation in humans: adaptation or preserved sleep homeostasis? Sleep. 2018 Jul 1;41(7). doi: 10.1093/sleep/zsy078. PMID: 29722893. 16. Carskadon, M.A. and Dement, W.C. (1981), Cumulative Effects of Sleep Restriction on Daytime Sleepiness. Psychophysiology, 18: 107-113. https://doi.org/10.1111/j.1469-8986.1981.tb02921.x 17. Guilleminault, Christian, et al. “Preliminary Observations on the Effects of Sleep Time in a Sleep Restriction Paradigm.” Sleep Medicine, vol. 4, no. 3, May 2003, pp. 177–84. ScienceDirect, https://doi.org/10.1016/S1389-9457(03)00061-3. 18. Depner CM, Cheng PC, Devine JK, Khosla S, de Zambotti M, Robillard R, Vakulin A, Drummond SPA. Wearable technologies for developing sleep and circadian 19 biomarkers: a summary of workshop discussions. Sleep. 2020 Feb 13;43(2):zsz254. doi: 10.1093/sleep/zsz254. PMID: 31641776; PMCID: PMC7368340. 20 Name of Candidate: Zachary C Mallender Date of Submission: Dec 16, 2022 21