Sleep disturbances such as insomnia, fragmented sleep, daytime sleepiness, and sleep-disordered breathing are common features of Alzheimer disease (AD) and Lewy body disease (LBD) and often appear early in the disease course and before clinical diagnosis [1,2]. Short and fragmented sleep are associated with processes linked to neurodegeneration, including reduced glymphatic clearance of waste [3], increased amyloid beta (Aβ) and tau burden [4], increased neuroinflammation [5], and impaired cardiovascular health [6]. Several small studies of sleep apnea treatment have demonstrated improvement in cognitive outcomes and blood biomarkers of Aβ and tau, suggesting that sleep interventions could improve prognosis in mild cognitive impairment (MCI) and dementia [7-9]. Poor quality or insufficient sleep in midlife increases the risk of all-cause dementia and MCI [10,11], indicating that improving sleep may also protect against dementia. Large-scale clinical studies are needed to enhance our mechanistic understanding of sleep, confirm the most promising therapeutic targets, and monitor the effectiveness and safety profile of sleep interventions [12,13].

However, selecting sleep assessment tools is challenging, particularly in populations with cognitive impairment. Self-report is convenient, inexpensive, and scalable, and has often been used to examine sleep in people with dementia and MCI [14]; however, it is potentially prone to recall bias due to memory deficits or anosognosia. Self-report also correlates poorly with objective sleep measures, particularly in participants with MCI or dementia [15,16] and those with subjectively poor sleep [17], and cannot inform on key components of sleep such as sleep staging or microarchitecture. Polysomnography (PSG) is typically considered the gold standard for sleep measurement, as it provides rich objective sleep data. However, PSG typically requires expert setup, analysis, and a controlled clinic environment, meaning it is expensive, not easily scalable, and therefore typically used for 1 or a few nights [18,19]. PSG setup, especially in an artificial environment, may also not reflect natural sleep [19] as it disrupts usual routines. Longitudinal data collection in sleep research would be beneficial for monitoring clinical trials and disease progression and could account for variation in sleep from external factors such as acute illness or stress, as well as natural intraindividual sleep variation [19].

Wearable devices, smartphone apps, and telemedicine, apart from actigraphy devices, have rarely been used in MCI or dementia research [20], but offer an opportunity to collect objective and subjective data longitudinally in a natural setting [21], often at relatively low cost. Detailed and accurate sleep analysis can now be achieved through wireless technology, including electroencephalography (EEG) headbands and overnight pulse oximetry [21,22]. While actigraphy has been used in dementia research, EEG headbands and pulse oximetry are less well tested, especially in the earlier stages of impairment [23-25]. There is also increasing interest in remote cognitive testing and digital biomarkers for diagnosis and monitoring progression [26]. Despite this, the adoption of digital health technologies into neurology clinical trials, particularly in older people, has been slow [27].

Remote sleep and cognitive data collection has the potential to decentralize clinical trials, making research more convenient for participants and reducing the costs, participant burden, and carbon footprint associated with study visits, while enabling real-time longitudinal data collection of treatment effects. Improved digital access and literacy among older adults [28] and the use of technology among patients living with MCI or dementia to support independent living and for recreation indicate increasing acceptance of technology [29]. However, not all older adults are comfortable using technology, and changes in cognition, sensory processing, and communication might negatively impact the usability and acceptability of novel devices for research purposes in people with cognitive impairment [29,30]. The few studies that have tested wearable devices and digital health technologies for sleep and dementia research in the home have predominantly collected feasibility data for a single device across only a few nights [31] and required support from a study partner or care home staff [32-34]. Before trials invest in digital health technologies and remote study designs, it is important to know whether participants can and are willing to engage and provide data in such studies over a longer period and how much support might be required.

This study aimed to establish the feasibility of predominantly technology-based longitudinal sleep and cognitive assessments from home in older adults with and without MCI or dementia.

The Remote Evaluation of Sleep to Enhance Understanding of Early Dementia (RESTED) study was a prospective longitudinal observational cohort study involving remote sleep and cognitive monitoring of participants with MCI or dementia and age-matched cognitively healthy controls. The full study protocol has been previously published [35]. The study has been reported in line with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines [36] (Multimedia Appendix 1).

This study was approved by the Health Research Authority (Yorkshire and the Humber-Bradford Leeds Research Ethics Committee, reference 21/YH/0177). This study was conducted in accordance with the principles of the Declaration of Helsinki. All participants provided written informed consent before engaging in any study activities and were reminded of their right to withdrawal without giving a reason and that withdrawal would not affect their health care. Following consent, data were pseudonymized with participants allocated a participant identifier code to maintain privacy. Participants were not compensated for their involvement in the study; however, travel expenses for research visits outside of usual clinic visits were reimbursed. All researchers engaging with participants were trained in good clinical practice.

The RESTED study recruited community-dwelling adults aged ≥50 years with internet access at home. Participants were recruited to 1 of 3 participant subgroups according to clinical diagnosis meeting standardized diagnostic criteria [37-40]. Participants with MCI or mild dementia due to probable AD were recruited to the AD group, participants with MCI or mild dementia due to probable LBD or Parkinson disease (PD) were recruited to the LBD group, and sex- and age-matched individuals with no known neurodegenerative conditions or cognitive impairment were recruited as controls. Exclusion criteria included advanced dementia, acute or terminal illness, and significant unrelated comorbidities that might interfere with sleep, except for obstructive sleep apnea (OSA) if participants were already undergoing treatment.

Participants were recruited from the city of Bristol, United Kingdom, and the surrounding areas via cognitive and movement disorders clinics and research volunteer databases, including Join Dementia Research. Our original target sample size was 75 participants [41]. Due to the COVID-19 pandemic, the study was delayed, and the budget was partially reallocated to studies that required no patient contact [12,13,42]. Therefore, we revised the target sample size to 40 participants and recruited from a single site (North Bristol NHS Trust).

We did not recruit or require participants to have a study partner. Participants were welcome to invite someone to support them, and study support was recorded.

Unless otherwise stated, descriptive statistics are provided as mean (SD) for continuous variables, and frequency and percentage for categorical variables, and provided for the full cohort as well as by participant subgroup (AD vs LBD vs controls). Data analysis was performed using R (version 4.3.1; R Foundation for Statistical Computing) and R Studio software (version 2023.6.0.421; R Foundation for Statistical Computing). Actigraphy data were processed using the open-source AX3/AX6 OMGUI application and analyzed in R using the GGIR package [54]. Where available, sleep timing was adjusted using participants’ sleep diaries; otherwise, GGIR uses the heuristic algorithm looking at distribution of change in z-angle [55] to estimate sleep timing. All files were also visually inspected to check for the accuracy of sleep timing. The quality of the Dreem 2 recordings was assessed by inspecting Dreem’s automated record quality index in the sleep report for each night, which indicates the percentage of the recording that is of scorable quality for sleep analysis. Associations between participant characteristics and adherence were calculated using Spearman rank correlation with α set to P<.05. Reasons for nonparticipation and missing data are provided where known and mapped to the Capability, Opportunity, Motivation model of behavior change [56]. Where additional data were collected on any outcome (eg, participants completed an additional night of EEG than instructed), additional data points were removed before analysis to avoid biasing feasibility metrics.

Patient and public involvement (PPI) was sought from individuals with lived experience of MCI and dementia before and throughout the study. PPI contributors reviewed and improved study documents and advised on the acceptability of adding blood biomarker testing. The PPI group strongly endorsed our additional recruitment materials (including an advertisement poster and a plain English participant information sheet) that we introduced partway through the study, after we received feedback from a prospective participant that standard participant information sheets are too long and use inaccessible language for people with cognitive impairment. We also introduced a reduced study protocol, which involved use of a paper (rather than digital) sleep diary, actigraphy, and EEG, to encourage recruitment of participants who may be more comfortable with less frequent use of technology; however, nobody chose the reduced protocol, they either declined altogether or participated in the full study.

In this study, we show that it is feasible to remotely measure sleep and cognition longitudinally in community-dwelling older adults with MCI and dementia due to AD and LBD and healthy older adults. Eligible participants were interested in, enrolled in, and remained in the study, despite being asked to complete a high volume of remote and novel study tasks across an 8-week period while continuing their usual routines. Only 1 participant withdrew from the study, and this was before remote data collection started. Most participants were receptive to multimodal home-based research using technology and alternatives to in-laboratory sleep assessment. Across the cohort, data completeness rate was high and ranged from 79.8% to 97.1%. Our findings support the use of remote, technology-supported research methods to study natural sleep in future trials and indicate some areas where further improvements and refinements would be helpful. With just under two-thirds of our sample (25/40, 63%) having sleep apnea, and only 3 (8%) being aware of this before joining the study, our results also highlight the importance of sleep apnea screening in older adults, as a risk factor and potentially reversible contributor to cognitive impairment and poor sleep quality [59].

Many older adults with MCI and dementia routinely leverage technology for cognitive stimulation, performing activities of daily living (such as shopping and banking), entertainment, and socializing, and use assistive-technology solutions, such as reminders and navigation to support independence [60]. However, cognitive impairment might impact ability to understand or remember to complete remote or technology-based study tasks, resulting in missing data, and may trigger anxiety if patients feel unable to perform tasks correctly [61]. Although a few previous studies had examined the use of sleep wearables in adults with AD, this was typically done over a brief period, examined feasibility of using only 1 digital health technology, or required input from a carer or study partner [31-34,62]. Our study demonstrated that older adults with MCI and dementia can successfully complete novel multimodal sleep and cognitive assessments, including longitudinal concurrent use of wrist actigraphy, web-based cognitive tests, digital sleep diaries, and wireless EEG headbands. However, more effort will be required to recruit samples which are representative of the older adult population and those with MCI or dementia, as our study predominantly recruited White male individuals who were familiar with smart technology, albeit not the devices used in RESTED.

Data completeness rates, reflecting both the ability to complete study tasks and produce data of analyzable quality, were high across all study tasks and participants. Remote data collection comes with a risk of missing data or nonadherence; however, we were able to mitigate this risk by using devices which upload data to servers at regular intervals (eg, Dreem and Dignio) and, less conveniently, by downloading data manually from devices to check compliance and data quality and offer support where needed (eg, actigraphy). Passive monitoring devices and devices that upload to servers automatically can simultaneously reduce the burden on participants and researchers while minimizing data loss by allowing researchers to monitor and respond quickly to any user or technical issues.

Reasons for missing data were usually known and typically related to infrequent but intentional decisions to remove a device (eg, due to a social event or illness) or, in most cases, technical issues with devices or software beyond the participants’ control, which would not introduce bias. We observed a weak correlation between EEG record quality and baseline cognitive impairment, which could be explained by cognitive impairment impacting ability to correctly wear the EEG headband for maximal impedance or reflect that patients were moving more during the night. Even subtle movements can shift the headband and reduce signal quality and getting optimal signal quality across entire or several recordings is a recognized challenge for dry EEG [52]. More comprehensive cognitive assessment than the MoCA, particularly a deeper assessment of executive dysfunction and long-term memory, might reveal a stronger relationship between cognitive impairment and adherence. However, recent studies using remote monitoring technologies found that, although patients with more advanced neurodegenerative disease or cognitive impairment reported more problems and were less compliant, the study remained feasible [63,64]. Larger studies may also want to consider regression analyses to identify key predictors of adherence to identify where support might be most indicated by the research team or caregiver support. However, importantly, overall record quality remained high across the cohort. Having MCI or mild dementia did not preclude study participation, adherence to remote supervised and unsupervised study tasks, or obtaining good-quality data.

Use of remote sleep and cognitive monitoring can help to partly or fully decentralize research, as participants can use their own devices or be provided with technology via post where needed [65], which is more scalable for clinical trials [66]. Frequent or lengthy clinic visits, particularly those that require overnight stays for sleep analysis, can be burdensome for participants, and may reduce inclusivity by requiring participants to live near a study site or travel long distances. More frequent but briefer remote assessments may also help to mitigate against the risk of large amounts of missing data resulting from a missed clinic appointment, can act as “digital biomarkers” that may detect changes more sensitively than an annual follow-up [67], and are more convenient for participants [68]. Although not the focus of our work, research could also explore the use of digital tools to assess fluctuations in cognition over the day in LBD (where fluctuation in cognition is a key diagnostic criteria) and AD (where a sundowning effect of heightened distress is often observed during the evening hours), both of which are difficult to assess in clinical settings [69,70].

While high adherence has been previously demonstrated in feasibility studies of remote monitoring technologies in participants with cognitive impairment or neurodegenerative conditions, most studies have required intensive study partner support throughout the study period as an eligibility criterion [63,71]. It has been argued that study partners are essential for participant safety and well-being in dementia research, even at the preclinical stage where cognitively normal individuals are enrolled [72]. For some participants, partners or relatives provided essential support for study participation. However, many participants provided good-quality data with minimal to no input from others. Crucially, several participants would not have been able to participate if there had been a study partner requirement. Strict eligibility criteria for trials are a recognized barrier to research participation [73]. Requiring a study partner for all studies may unnecessarily increase the burden for relatives already providing care or support, reducing the participant pool, and may disproportionately affect different groups in society who are underserved by research [74,75]. Our findings encourage researchers to consider inclusive and flexible research designs, including options to formally recruit and collect data from study partners without excluding participants who do not have someone available to support them in research studies.

There are an increasing number of options when considering how to measure sleep and cognition from home, including both research-grade and consumer sleep trackers [76,77]. We opted to use existing technologies that were noninvasive, affordable, required minimal training or supervision, were commercially available research-grade or consumer-grade devices that met necessary data privacy regulations, and provided data in a format that could be analyzed using open-source software. Existing technologies can be implemented immediately, reducing time and cost to setup research studies, have a more mature user interface, and are less error-prone than a newly developed solution, and crucially, may be easier to compare among studies for meta-analysis. Several study tasks required ongoing technical support or services from the manufacturers or developers. As a customer rather than a collaborator, we could not always identify or respond to technical problems, and with some providers, we were not informed in advance of several significant changes, which impacted the study or its participants resulting in data loss. Collaborating with industry, ideally at an early stage of the research process, might ensure longevity and continued support throughout the life course of the study and increased flexibility to adapt technology to better fit research (eg, data validation for digital patient-reported outcomes).

Women, adults aged >80 years, and minoritized ethnic groups were underrepresented in our study, which is often observed in dementia research [78]. Although some barriers (eg, mistrust and accessibility of research) and motivators (eg, altruism) have been identified, further work is needed to identify how to improve research access, inclusion, and participation in dementia research [79]. There may also have been a self-selection bias, whereby individuals who were likely to be more competent and enthusiastic about using technology volunteered to participate in the study. Most of our participants were regular smartphone users, and several prospective participants declined to participate due to concerns regarding their confidence or interest in using the technologies. Although the proportion of older adults in the United Kingdom using the internet is increasing, a significant proportion of older adults do not regularly use the internet or lack fundamental skills such as being able to turn on a device and enter login details [28]. Older adults with MCI and dementia who regularly use devices report that smartphones and tablets can be useful to support activities of daily living, as well as for communication, entertainment, and recreation, but also list concerns, including cybersecurity and vulnerability to fraud [80]. Education around sleep and brain health, and basic digital skills training on how devices can support individuals living with cognitive impairment, may improve perceptions of capability and motivation to participate in similar studies [56,80]. Exploring the feasibility of more passive monitoring technologies, such as mattress sensors or smartphone-based passive sensing, may help to increase participation and confidence from those with less experience or interest in technology. Although these techniques may invite additional concerns around data privacy, passive monitoring technologies are already being used to support older adults aging in place [81,82]. Increasing the pool of prospective participants through recruiting via multiple sites and meaningful engagement with community groups, and considering digital inclusivity is likely to enhance recruitment in future studies. Finally, we did not set a priori feasibility cutoffs for study tasks. Despite these limitations, the high completion and retention rate from the RESTED study suggests that remote sleep and cognitive monitoring could offer detailed sleep profiling suitable for tracking change in sleep over disease progression or for monitoring change in sleep clinical trials in older adults with or at risk of dementia.

Practical, detailed, and scalable assessment of sleep is essential for understanding how sleep disturbances affect neurodegeneration over time and for developing and evaluating effective interventions. Our results suggest that older adults with MCI and dementia, as well as healthy older adults, can and do engage in multimodal remote sleep and cognitive research, including using wireless EEG, actigraphy, and mobile apps or web applications. Remote monitoring technologies and research designs offer the opportunity to study natural sleep and its relationship with cognition and dementia over extended periods, are scalable, and should be considered when designing future clinical trials in sleep and dementia in these populations.