Dr Ciro della Monica

Study Objectives: Portable electroencephalography (EEG) devices offer the potential for accurate quantification of sleep at home but have not been evaluated in relevant populations. We assessed the Dreem headband (DHB) and its automated sleep staging algorithm in 62 older adults [age (mean±SD) 70.5±6.7 years; 12 Alzheimer’s]. Methods: The accuracy of sleep measures, epoch-by-epoch staging, and the quality of EEG signals for quantitative EEG (qEEG) analysis were compared to standard polysomnography (PSG) in a sleep laboratory. Results: The DHB algorithm accurately estimated total sleep time (TST) and sleep efficiency (SEFF) with a Symmetric Mean Absolute Percentage Error (SMAPE)

Background: Contactless sleep technologies (CSTs) hold promise for longitudinal, unobtrusive sleep monitoring in the community and at scale. They may be particularly useful in older populations wherein sleep disturbance, which may be indicative of the deterioration of physical and mental health, is highly prevalent. However, few CSTs have been evaluated in older people. Objective: This study evaluated the performance of 3 CSTs compared to polysomnography (PSG) and actigraphy in an older population. Methods: Overall, 35 older men and women (age: mean 70.8, SD 4.9 y; women: n=14, 40%), several of whom had comorbidities, including sleep apnea, participated in the study. Sleep was recorded simultaneously using a bedside radar (Somnofy [Vital Things]: n=17), 2 undermattress devices (Withings sleep analyzer [WSA; Withings Inc]: n=35; Emfit-QS [Emfit; Emfit Ltd]: n=17), PSG (n=35), and actigraphy (Actiwatch Spectrum [Philips Respironics]: n=18) during the first night in a 10-hour time-in-bed protocol conducted in a sleep laboratory. The devices were evaluated through performance metrics for summary measures and epoch-by-epoch classification. PSG served as the gold standard. Results: The protocol induced mild sleep disturbance with a mean sleep efficiency (SEFF) of 70.9% (SD 10.4%; range 52.27%-92.60%). All 3 CSTs overestimated the total sleep time (TST; bias: >90 min) and SEFF (bias: >13%) and underestimated wake after sleep onset (bias: >50 min). Sleep onset latency was accurately detected by the bedside radar (bias: 16 min). CSTs did not perform as well as actigraphy in estimating the all-night sleep summary measures. In an epoch-by-epoch concordance analysis, the bedside radar performed better in discriminating sleep versus wake (Matthew correlation coefficient [MCC]: mean 0.63, SD 0.12, 95% CI 0.57-0.69) than the undermattress devices (MCC of WSA: mean 0.41, SD 0.15, 95% CI 0.36-0.46; MCC of Emfit: mean 0.35, SD 0.16, 95% CI 0.26-0.43). The accuracy of identifying rapid eye movement and light sleep was poor across all CSTs, whereas deep sleep (ie, slow wave sleep) was predicted with moderate accuracy (MCC: >0.45) by both Somnofy and WSA. The deep sleep duration estimates of Somnofy correlated (r2=0.60; P

Background: Sleep disorders are common among the aging population and people with neurodegenerative diseases. Sleep disorders have a strong bidirectional relationship with neurodegenerative diseases, where they accelerate and worsen one another. Although one-to-one individual cognitive behavioral interventions (conducted in-person or on the internet) have shown promise for significant improvements in sleep efficiency among adults, many may experience difficulties accessing interventions with sleep specialists, psychiatrists, or psychologists. Therefore, delivering sleep intervention through an automated chatbot platform may be an effective strategy to increase the accessibility and reach of sleep disorder intervention among the aging population and people with neurodegenerative diseases. Objective: This work aims to (1) determine the feasibility and usability of an automated chatbot (named MotivSleep) that conducts sleep interviews to encourage the aging population to report behaviors that may affect their sleep, followed by providing personalized recommendations for better sleep based on participants' self-reported behaviors; (2) assess the self-reported sleep assessment changes before, during, and after using our automated sleep disturbance intervention chatbot; (3) assess the changes in objective sleep assessment recorded by a sleep tracking device before, during, and after using the automated chatbot MotivSleep.Methods: We will recruit 30 older adult participants from West London for this pilot study. Each participant will have a sleep analyzer installed under their mattress. This contactless sleep monitoring device passively records movements, heart rate, and breathing rate while participants are in bed. In addition, each participant will use our proposed chatbot MotivSleep, accessible on WhatsApp, to describe their sleep and behaviors related to their sleep and receive personalized recommendations for better sleep tailored to their specific reasons for disrupted sleep. We will analyze questionnaire responses before and after the study to assess their perception of our proposed chatbot; questionnaire responses before, during, and after the study to assess their subjective sleep quality changes; and sleep parameters recorded by the sleep analyzer throughout the study to assess their objective sleep quality changes.Results: Recruitment will begin in May 2023 through UK Dementia Research Institute Care Research and Technology Centre organized community outreach. Data collection will run from May 2023 until December 2023. We hypothesize that participants will perceive our proposed chatbot as intelligent and trustworthy; we also hypothesize that our proposed chatbot can help improve participants' subjective and objective sleep assessment throughout the study.Conclusions: The MotivSleep automated chatbot has the potential to provide additional care to older adults who wish to improve their sleep in more accessible and less costly ways than conventional face-to-face therapy.

To compare the 24-hour sleep assessment capabilities of two contactless sleep technologies (CSTs) to actigraphy in community-dwelling older adults. We collected 7 to 14 days of data at home from 35 older adults (age: 65-83), some with medical conditions, using Withings Sleep Analyser (WSA, n=29), Emfit-QS (Emfit, n=17), a standard actigraphy device (Actiwatch Spectrum [AWS, n=34]) and a sleep diary. We compared nocturnal and daytime sleep measures estimated by the CSTs and actigraphy without sleep diary information (AWS-A) against sleep diary assisted actigraphy (AWS|SD). Compared to sleep diary, both CSTs accurately determined the timing of nocturnal sleep (ICC: going to bed, getting out of bed, time in bed > 0.75) whereas the accuracy of AWS-A was much lower. Compared to AWS|SD, the CSTs overestimated nocturnal total sleep time (WSA: +92.71±81.16 min; Emfit: +101.47±75.95 min) as did AWS-A (+46.95±67.26 min). The CSTs overestimated sleep efficiency (WSA: +9.19±14.26 %; Emfit: +9.41±11.05 %) whereas AWS-A estimate (-2.38±10.06 %) was accurate. About 65% (n=23) of participants reported daytime naps either in-bed or elsewhere. About 90% in-bed nap periods were accurately determined by WSA while Emfit was less accurate. All three devices estimated 24-h sleep duration with an error of ≈10% compared to the sleep diary. CSTs accurately capture the timing of in-bed nocturnal sleep periods without the need for sleep diary information. However, improvements are needed in assessing parameters such as total sleep time, sleep efficiency and naps before these CSTs can be fully utilized in field settings.

Introduction Longitudinal monitoring of vital signs provides a method for identifying changes to general health in an individual and particularly so in older adults. The nocturnal sleep period provides a convenient opportunity to assess vital signs. Contactless technologies that can be embedded into the bedroom environment are unintrusive and burdenless and have the potential to enable seamless monitoring of vital signs. To realise this potential, these technologies need to be evaluated against gold standard measures and in relevant populations. Methods We evaluated the accuracy of heart rate and breathing rate measurements of three contactless technologies (two under-mattress trackers: Withings sleep analyser (WSA) and Emfit QS (Emfit) and a bedside radar: Somnofy) in a sleep laboratory environment and assessed their potential to capture vital signs (heart rate and breathing rate) in a real-world setting. Data were collected in 35 community dwelling older adults aged between 65 and 83 years (mean ± SD: 70.8 ± 4.9; 21 men) during a one-night clinical polysomnography (PSG) in a sleep laboratory, preceded by 7 to 14 days of data collection at-home. Several of the participants had health conditions including type-2 diabetes, hypertension, obesity, and arthritis and ≈49% (n = 17) had moderate to severe sleep apnea while ≈29% (n = 10) had periodic leg movement disorder. The under-mattress trackers provided estimates of both heart rate and breathing rate while the bedside radar provided only breathing rate. The accuracy of the heart rate and breathing rate estimated by the devices was compared to PSG electrocardiogram (ECG) derived heart rate (beats per minute, bpm) and respiratory inductance plethysmography thorax (RIP thorax) derived breathing rate (cycles per minute, cpm). We also evaluated breathing disturbance indices of snoring and the apnea-hypopnea index (AHI) available from the WSA. Results All three contactless technologies provided acceptable accuracy in estimating heart rate [mean absolute error (MAE) < 2.2 bpm and mean absolute percentage error (MAPE) < 5%] and breathing rate (MAE ≤ 1.6 cpm and MAPE < 12%) at 1 minute resolution. All three contactless technologies were able to capture changes in heart rate and breathing rate across the sleep period. The WSA snoring and breathing disturbance estimates were also accurate compared to PSG estimates (R-squared: WSA Snore: 0.76, p < 0.001; WSA AHI: 0.59, p < 0.001). Conclusion Contactless technologies offer an unintrusive alternative to conventional wearable technologies for reliable monitoring of heart rate, breathing rate, and sleep apnea in community dwelling older adults at scale. They enable assessment of night-to-night variation in these vital signs, which may allow the identification of acute changes in health, and longitudinal monitoring which may provide insight into health trajectories.

Sleep timing varies between individuals and can be altered in mental and physical health conditions. Sleep and circadian sleep phenotypes, including circadian rhythm sleep-wake disorders, may be driven by endogenous physiological processes, exogeneous environmental light exposure, social constraints or behavioural factors. Identifying the relative contributions of these driving factors to different phenotypes is essential for the design of personalised sleep interventions. The timing of the human sleep-wake cycle has been modelled as an interaction between a relaxation oscillator (the sleep homeostat), a stable limit cycle oscillator with a near 24-hour period (the circadian process), man-made light exposure, and the natural light-dark cycle generated by the Earth's rotation. However, these models have rarely been used to quantitatively describe sleep at the individual level. Here, we present a new Homeostatic-Circadian-Light model (HCL) which is simpler, more transparent, and more computationally efficient than other available models and is designed to run using longitudinal sleep and light exposure data from wearable sensors. We carry out a systematic sensitivity analysis for all model parameters and discuss parameter identifiability. We demonstrate we can describe individual sleep phenotypes in each of 34 older participants (65-83y) by feeding individual participant light exposure patterns into the model and fitting two parameters that capture individual average sleep duration and timing. The fitted parameters describe endogenous drivers of sleep phenotypes. We then quantify exogenous drivers using a novel metric which encodes the circadian phase dependence of the response to light. Combining endogenous and exogeneous drivers better explains individual mean mid-sleep (adjusted R-squared 0.64) than either driver on its own (adjusted R-squared 0.08 and 0.17 respectively). Critically, our model and analysis highlights different people exhibiting the same sleep phenotype may have different driving factors and opens the door to personalised interventions to regularize sleep-wake timing that are readily implementable with current digital health technology.Competing Interest StatementDJD and ACS are consultants to F. Hoffmann-La Roche, Ltd.Footnotes* This version of the manuscript has been revised to: - take account of referees comments - updated to correct minor errors, including typographical errors. The central messages and mathematical modelling remain unchanged.* https://github.com/anneskeldon/Homeostatic_circadian_light_model-factors_driving_sleep_phenotypes

Plasma biomarkers of dementia, including phosphorylated tau (p-tau217), offer promise as tools for diagnosis, stratification for clinical trials, monitoring disease progression, and assessing the success of interventions in those living with Alzheimer's disease. However, currently, it is unknown whether these dementia biomarker levels vary with the time of day, which could have implications for their clinical value. In two protocols, we studied 38 participants (70.8 ± 7.6 years; mean ± SD) in a 27-h laboratory protocol with either two samples taken 12 h apart or 3-hourly blood sampling for 24 h in the presence of a sleep-wake cycle. The study population comprised people living with mild Alzheimer's disease (PLWA, n = 8), partners/caregivers of PLWA (n = 6) and cognitively intact older adults (n = 24). Single-molecule array technology was used to measure phosphorylated tau (p-tau217) (ALZpath), amyloid-beta 40 (Aβ40), amyloid-beta 42 (Aβ42), glial fibrillary acidic protein, and neurofilament light (NfL) (Neuro 4-Plex E). Analysis with a linear mixed model (SAS, PROC MIXED) revealed a significant effect of time of day for p-tau217, Aβ40, Aβ42, and NfL, and a significant effect of participant group for p-tau217. For p-tau217, the lowest levels were observed in the morning upon waking and the highest values in the afternoon/early evening. The magnitude of the diurnal variation for p-tau217 was similar to the reported increase in p-tau217 over one year in amyloid-β-positive mild cognitively impaired people. Currently, the factors driving this diurnal variation are unknown and could be related to sleep, circadian mechanisms, activity, posture, or meals. Overall, this work implies that the time of day of sample collection may be relevant in the implementation and interpretation of plasma biomarkers in dementia research and care.

Abstract Background Increased daytime napping and excessive sleepiness are associated with cognitive decline in older adults, especially in people living with dementia (PLWD) [1]. Subjective assessments of naps are burdensome and maybe unreliable in PLWD and hence there is a need for technologies that provide objective longitudinal assessment of the incidence and duration of naps. Here we compare two contactless sleep technologies (CSTs) against sleep diary and actigraphy for monitoring daytime napping in community dwelling non‐demented older adults. Method Two under‐mattress CSTs (Withings Sleep Analyser [WSA] and Emfit QS [Emfit]) along with actigraphy (Actiwatch Spectrum [AWS]) were deployed in the home of 17 older adults for a period of 14 days ( = 65 years; Mean Age ± SD = 72 ± 4.49; 6 Women). The ground truth nap information was collected using an extended consensus sleep diary that included additional information about naps (timing, duration, and location). We analyzed the agreement of the napping events and duration estimated by WSA, Emfit and AWS against the sleep diary reported events. Result The CSTs only detected in‐bed naps whilst the AWS detected both in‐bed and not‐in‐bed naps. Although all the compared devices detected spurious naps unreported in the sleep diary, it was highest in AWS (81% of total naps detected) followed by Emfit (63%) and WSA (16%) as shown in Figure 1. Among the CSTs, WSA accurately detected more in bed naps while registering less spurious naps compared to Emfit but had lower duration agreement to sleep diary (Figure 2). Further, when the contribution of daytime naps to 24‐h total sleep time was computed, the WSA estimate (12.8±6.1%) was closest to the sleep diary estimate (12.9±9.1%) followed by AWS (15.6±9.9%) and Emfit (17.2±11.1%). Conclusion CSTs, with their ability to provide both contextual location information and objective measures of napping, such as timing and duration, offer a reliable and unobtrusive alternative to traditional methods such as sleep diary and actigraphy for long‐term round‐the‐clock monitoring of sleep in older adults. References: [1] Li, P, Gao, L, Yu, L, et al. Daytime napping, and Alzheimer’s dementia: A potential bidirectional relationship. Alzheimer’s Dement. 2023; 19: 158– 168. https://doi.org/10.1002/alz.12636

Objective: Sleep monitoring has extensively utilized electroencephalogram (EEG) data collected from the scalp, yielding very large data repositories and well-trained analysis models. Yet, this wealth of data is lacking for emerging, less intrusive modalities, such as ear-EEG.Methods and procedures: The current study seeks to harness the abundance of open-source scalp EEG datasets by applying models pre-trained on data, either directly or with minimal fine-tuning; this is achieved in the context of effective sleep analysis from ear-EEG data that was recorded using a single in-ear electrode, referenced to the ipsilateral mastoid, and developed in-house as described in our previous work. Unlike previous studies, our research uniquely focuses on an older cohort (17 subjects aged 65-83, mean age 71.8 years, some with health conditions), and employs LightGBM for transfer learning, diverging from previous deep learning approaches. Results: Results show that the initial accuracy of the pre-trained model on ear-EEG was 70.1%, but fine-tuning the model with ear-EEG data improved its classification accuracy to 73.7%. The fine-tuned model exhibited a statistically significant improvement (p < 0.05, dependent t-test) for 10 out of the 13 participants, as reflected by an enhanced average Cohen's kappa score (a statistical measure of inter-rater agreement for categorical items) of 0.639, indicating a stronger agreement between automated and expert classifications of sleep stages. Comparative SHAP value analysis revealed a shift in feature importance for the N3 sleep stage, underscoring the effectiveness of the fine-tuning process.Conclusion: Our findings underscore the potential of fine-tuning pre-trained scalp EEG models on ear-EEG data to enhance classification accuracy, particularly within an older population and using feature-based methods for transfer learning. This approach presents a promising avenue for ear-EEG analysis in sleep studies, offering new insights into the applicability of transfer learning across different populations and computational techniques.Clinical impact: An enhanced ear-EEG method could be pivotal in remote monitoring settings, allowing for continuous, non-invasive sleep quality assessment in elderly patients with conditions like dementia or sleep apnea.

Nocturnal disturbance is frequently observed in dementia and is a major contributor to institutionalisation. Unobtrusive technology that can quantify sleep/wake and determine bed occupancy during the major nocturnal sleep episode may be beneficial for long-term clinical monitoring and the carer. Such technologies have, however, not been validated in older people. Here we assessed the performance of the Withings Sleep Mattress (WSM) in a heterogenous older population to ensure external validity.

Quantification of sleep is important for the diagnosis of sleep disorders and sleep research. However, the only widely accepted method to obtain sleep staging is by visual analysis of polysomnography (PSG), which is expensive and time consuming. Here, we investigate automated sleep scoring based on a low‐cost, mobile electroencephalogram (EEG) platform consisting of a lightweight EEG amplifier combined with flex‐printed cEEGrid electrodes placed around the ear, which can be implemented as a fully self‐applicable sleep system. However, cEEGrid signals have different amplitude characteristics to normal scalp PSG signals, which might be challenging for visual scoring. Therefore, this study evaluates the potential of automatic scoring of cEEGrid signals using a machine learning classifier (“random forests”) and compares its performance with manual scoring of standard PSG. In addition, the automatic scoring of cEEGrid signals is compared with manual annotation of the cEEGrid recording and with simultaneous actigraphy. Acceptable recordings were obtained in 15 healthy volunteers (aged 35 ± 14.3 years) during an extended nocturnal sleep opportunity, which induced disrupted sleep with a large inter‐individual variation in sleep parameters. The results demonstrate that machine‐learning‐based scoring of around‐the‐ear EEG outperforms actigraphy with respect to sleep onset and total sleep time assessments. The automated scoring outperforms human scoring of cEEGrid by standard criteria. The accuracy of machine‐learning‐based automated scoring of cEEGrid sleep recordings compared with manual scoring of standard PSG was satisfactory. The findings show that cEEGrid recordings combined with machine‐learning‐based scoring holds promise for large‐scale sleep studies.

Sleep and its sub-states are assumed to be important for brain function across the lifespan but which aspects of sleep associate with various aspects of cognition, mood and self-reported sleep quality has not yet been established in detail. Sleep was quantified by polysomnography, quantitative Electroencephalogram (EEG) analysis and self-report in 206 healthy men and women, aged 20–84 years, without sleep complaints. Waking brain function was quantified by five assessments scheduled across the day covering objectively assessed performance across cognitive domains including sustained attention and arousal, decision and response time, motor and sequence control, working memory, and executive function as well as self-reports of alertness, mood and affect. Controlled for age and sex, self-reported sleep quality was negatively associated with number of awakenings and positively associated with the duration of Rapid Eye Movement (REM) sleep, but no significant associations with Slow Wave Sleep (SWS) measures were observed. Controlling only for age showed that associations between objective and subjective sleep quality were much stronger in women than in men. Analysis of 51 performance measures demonstrated that, after controlling for age and sex, fewer awakenings and more REM sleep were associated significantly with better performance on the Goal Neglect task, which is a test of executive function. Factor analysis of the individual performance measures identified four latent variables labeled Mood/Arousal, Response Time, Accuracy, and Visual Perceptual Sensitivity. Whereas Mood/Arousal improved with age, Response Times became slower, while Accuracy and Visual perceptual sensitivity showed little change with age. After controlling for sex and age, nominally significant association between sleep and factor scores were observed such that Response Times were faster with more SWS, and Accuracy was reduced where individuals woke more often or had less REM sleep. These data identify a positive contribution of SWS to processing speed and in particular highlight the importance of sleep continuity and REM sleep for subjective sleep quality and performance accuracy across the adult lifespan. These findings warrant further investigation of the contribution of sleep continuity and REM sleep to brain function.

Sleep and circadian rhythm disturbance are predictors of poor physical and mental health, including dementia. Long-term digital technology-enabled monitoring of sleep and circadian rhythms in the community has great potential for early diagnosis, monitoring of disease progression, and assessing the effectiveness of interventions. Before novel digital technology-based monitoring can be implemented at scale, its performance and acceptability need to be evaluated and compared to gold-standard methodology in relevant populations. Here, we describe our protocol for the evaluation of novel sleep and circadian technology which we have applied in cognitively intact older adults and are currently using in people living with dementia (PLWD). In this protocol, we test a range of technologies simultaneously at home (7-14 days) and subsequently in a clinical research facility in which gold standard methodology for assessing sleep and circadian physiology is implemented. We emphasize the importance of assessing both nocturnal and diurnal sleep (naps), valid markers of circadian physiology, and that evaluation of technology is best achieved in protocols in which sleep is mildly disturbed and in populations that are relevant to the intended use-case. We provide details on the design, implementation, challenges, and advantages of this protocol, along with examples of datasets.

[This corrects the article DOI: 10.2196/46338.].

Background Nocturnal disturbance is frequently observed in dementia and is a major contributor to institutionalisation. Unobtrusive technology that can quantify sleep/wake and determine bed occupancy during the major nocturnal sleep episode may be beneficial for long-term clinical monitoring and the carer. Such technologies have, however, not been validated in older people. Here we assessed the performance of the Withings Sleep Mattress (WSM) in a heterogenous older population to ensure external validity. Method Eighteen participants (65 – 80 years, 10M:8F) completed 7-12 days of sleep/wake monitoring at home prior to an overnight laboratory session. WSM performance was compared to gold-standard (laboratory polysomnography [PSG] with video) and silver standard (actiwatch [AWS] and sleep diary at home). WSM data were downloaded from a third party API and the minute-to-minute sleep/wake timeseries extracted and time-ordered to create a sleep profile. Discontinuities in the timeseries were labelled as ‘missing data’ events. Results Participants contributed 107 nights with WSM and PSG or AWS data. In the laboratory, the overall epoch to epoch agreement (accuracy) of sleep/wake detection of WSM compared to PSG was 0.71 (sensitivity 0.8; specificity 0.45) and to AWS was 0.74 (sensitivity 0.77; specificity 0.53). Visual inspection of video recordings demonstrated that 20 of 21 ‘missing data’ events were true ‘out of bed’ events. These events were always associated with an increase in activity (AWS). At home, all 97 WSM ‘missing data’ events that occurred within the major nocturnal sleep episode defined by sleep diary data, were associated with an increase in activity levels in the AWS data and 36 of these events were also associated with an increase in light levels, indicating that the participant had left the bed. In several participants, data recorded by the WSM during daytime coincided with reported naps in the sleep diary. Conclusion Although WSM cannot reliably distinguish between sleep and wake, the presence/absence of data in WSM seem to be an accurate representation of whether older people are in or out of bed (bed occupancy). Thus, in dementia, this contactless, low-burden technology may be able to provide information about nocturnal disturbances and daytime naps in bed.

The effects of orexinergic peptides are diverse and are mediated by orexin-1 and orexin-2 receptors. Antagonists that target both receptors have been shown to promote sleep initiation and maintenance. Here, we investigated the role of the orexin-2 receptor in sleep regulation in a randomised, double-blind, placebo-controlled, three-period crossover clinical trial using two doses (20 and 50 mg) of a highly selective orexin-2 receptor antagonist (2-SORA) (JNJ-48816274). We used a phase advance model of sleep disruption where sleep initiation is scheduled in the circadian wake maintenance zone. We assessed objective and subjective sleep parameters, pharmacokinetic profiles and residual effects on cognitive performance in 18 healthy male participants without sleep disorders. The phase advance model alone (placebo condition) resulted in disruption of sleep at the beginning of the sleep period compared to baseline sleep (scheduled at habitual time). Compared to placebo, both doses of JNJ-48816274 significantly increased total sleep time, REM sleep duration and sleep efficiency, and reduced latency to persistent sleep, sleep onset latency, and REM latency. All night EEG spectral power density for both NREM and REM sleep were unaffected by either dose. Participants reported significantly better quality of sleep and feeling more refreshed upon awakening following JNJ-48816274 compared to placebo. No significant residual effects on objective performance measures were observed and the compound was well tolerated. In conclusion, the selective orexin-2 receptor antagonist JNJ-48816274 rapidly induced sleep when sleep was scheduled earlier in the circadian cycle and improved self-reported sleep quality without impact on waking performance.

Background Sleep disturbances are both risk factors for and symptoms of dementia. Current methods for assessing sleep disturbances are largely based on either polysomnography (PSG) which is costly and inconvenient, or self‐ or care‐giver reports which are prone to measurement error. Low‐cost methods to monitor sleep disturbances longitudinally and at scale can be useful for assessing symptom development. Here, we develop deep learning models that use multimodal variables (accelerometers and temperature) recorded by the AX3 to accurately identify sleep and wake epochs and derive sleep parameters. Method Eighteen men and women (65‐80y) participated in a sleep laboratory‐based study in which multiple devices for sleep monitoring were evaluated. PSGs were recorded over a 10‐h period and scored according to established criteria per 30 sec epochs. Tri‐axial accelerometers and temperature signals were captured with an Axivity AX3, at 100Hz and 1Hz, respectively, throughout a 19‐h period, including 10‐h concurrent PSG recording and 9‐h of wakefulness. We developed and evaluated a supervised deep learning algorithm to detect sleep and wake epochs and determine sleep parameters from the multimodal AX3 raw data. We validated our results with gold standard PSG measurements and compared our algorithm to the Biobank accelerometer analysis toolbox. Single modality (accelerometer or temperature) and multimodality (both signals) approaches were evaluated using the 3‐fold cross‐validation. Result The proposed deep learning model outperformed baseline models such as the Biobank accelerometer analysis toolbox and conventional machine learning classifiers (Random Forest and Support Vector Machine) by up to 25%. Using multimodal data improved sleep and wake classification performance (up to 18% higher) compared with the single modality. In terms of the sleep parameters, our approach boosted the accuracy of estimations by 11% on average compared to the Biobank accelerometer analysis toolbox. Conclusion In older adults without dementia, combining multimodal data from AX3 with deep learning methods allows satisfactory quantification of sleep and wakefulness. This approach holds promise for monitoring sleep behaviour and deriving accurate sleep parameters objectively and longitudinally from a low‐cost wearable sensor. A limitation of our current study is that the participants were healthy older adults: future work will focus on people living with dementia.

Sleep timing varies between individuals and can be altered in mental and physical health conditions. Sleep and circadian sleep phenotypes, including circadian rhythm sleep-wake disorders, may be driven by endogenous physiological processes, exogeneous environmental light exposure along with social constraints and behavioural factors. Identifying the relative contributions of these driving factors to different phenotypes is essential for the design of personalised interventions. The timing of the human sleep-wake cycle has been modelled as an interaction of a relaxation oscillator (the sleep homeostat), a stable limit cycle oscillator with a near 24-hour period (the circadian process), man-made light exposure and the natural light-dark cycle generated by the Earth's rotation. However, these models have rarely been used to quantitatively describe sleep at the individual level. Here, we present a new Homeostatic-Circadian-Light model (HCL) which is simpler, more transparent and more computationally efficient than other available models and is designed to run using longitudinal sleep and light exposure data from wearable sensors. We carry out a systematic sensitivity analysis for all model parameters and discuss parameter identifiability. We demonstrate that individual sleep phenotypes in each of 34 older participants (65-83y) can be described by feeding individual participant light exposure patterns into the model and fitting two parameters that capture individual average sleep duration and timing. The fitted parameters describe endogenous drivers of sleep phenotypes. We then quantify exogenous drivers using a novel metric which encodes the circadian phase dependence of the response to light. Combining endogenous and exogeneous drivers better explains individual mean mid-sleep (adjusted R-squared 0.64) than either driver on its own (adjusted R-squared 0.08 and 0.17 respectively). Critically, our model and analysis highlights that different people exhibiting the same sleep phenotype may have different driving factors and opens the door to personalised interventions to regularize sleep-wake timing that are readily implementable with current digital health technology.

Electroencephalography (EEG) recordings represent a vital component of the assessment of sleep physiology, but the methodology presently used is costly, intrusive to participants, and laborious in application. There is a recognized need to develop more easily applicable yet reliable EEG systems that allow unobtrusive long-term recording of sleep-wake EEG ideally away from the laboratory setting. cEEGrid is a recently developed flex-printed around-the-ear electrode array, which holds great potential for sleep-wake monitoring research. It is comfortable to wear, simple to apply, and minimally intrusive during sleep. Moreover, it can be combined with a smartphone-controlled miniaturized amplifier and is fully portable. Evaluation of cEEGrid as a motion-tolerant device is ongoing, but initial findings clearly indicate that it is very well suited for cognitive research. The present study aimed to explore the suitability of cEEGrid for sleep research, by testing whether cEEGrid data affords the signal quality and characteristics necessary for sleep stage scoring. In an accredited sleep laboratory, sleep data from cEEGrid and a standard PSG system were acquired simultaneously. Twenty participants were recorded for one extended nocturnal sleep opportunity. Fifteen data sets were scored manually. Sleep parameters relating to sleep maintenance and sleep architecture were then extracted and statistically assessed for signal quality and concordance. The findings suggest that the cEEGrid system is a viable and robust recording tool to capture sleep and wake EEG. Further research is needed to fully determine the suitability of cEEGrid for basic and applied research as well as sleep medicine.