Eighteen young adults and 18 cognitively unimpaired older adults were recruited between March 2023 and January 2025. All participants were recruited from the University of Texas at Austin and the surrounding Austin community using mailings, flyer distribution, and phone calls. Interested individuals completed an initial REDCap (Research Electronic Data Capture) intake form for eligibility screening before trained research staff scheduled enrollment by phone. Participants were enrolled if they self-reported that they were fluent in English, had normal or corrected-to-normal vision, had no history of sleep disorders, and were free from untreated depression or anxiety, psychiatric or neurological disorders, or diabetes. Each participant had the opportunity to use the sleep monitoring system for up to 7 contiguous nights. Of the 18 young adults and 18 older adults recruited, 7 were excluded from the young group and 6 from the older group for failing to have at least 1 full night of usable EEG data. Thus, the analytic sample for sleep macrostructure and microstructure analysis included the remaining 11 young adults (8 women, aged 18‐28 y; mean education 15.1, SD 2.7 y; 4 Non-Hispanic Asian, 1 Hispanic Black, 4 Hispanic White, 2 Non-Hispanic White) and 12 older adults (6 women, aged 64‐79 y; mean education 17, SD 3 y; 1 Non-Hispanic Asian, 1 Non-Hispanic Black, 2 Hispanic White, 7 Non-Hispanic White, 1 did not report). Reasons for data loss are described in Table 1. Older adults completed the California Verbal Learning Test 3 for long-term verbal memory [30] and the Digit Span (Wechsler Adult Intelligence Scale-Fourth Edition) for executive functioning [31] to screen for clinically significant cognitive impairment. To this end, participants scoring 2 SDs below the across-participant mean on these tests would be excluded. No participants were excluded based on their cognitive assessments.

Full details of device fabrication, mechanical reliability, and reusability testing can be found in Kown et al [24]. In the previous version of this system, which was validated in a controlled laboratory setting, an additional chin patch was used for monitoring EMG with a separate reference electrode from the one used to monitor sleep EEG in the forehead patch. To simplify the system for participants applying it at home, we excluded the chin patch. This forehead system is consistent with the forehead-only systems used in most wearable sleep monitoring devices. Our team has developed a portable, wireless, skin-conforming sleep monitoring system that overcomes the limitations of existing home polysomnography and wearable sleep monitoring devices. A composite of elastic fabric and ultrasoft silicone elastomer provides strain distribution, omnidirectional elasticity, strong yet nonsticky adhesion to the skin, and a substrate for the integration of gold, laser-cut mesh electrodes, and copper interconnectors linking the electrodes with electronics (rechargeable battery, amplifier, Bluetooth microcontroller). The electrodes and the integrated circuit are encapsulated on the silicone side of the device, while the circuit—including a Bluetooth wireless module and a rechargeable 150-mAh Li-polymer battery with an average battery life of approximately 10 hours based on repeated laboratory measurements—is located on the fabric side of the device. The fabric provides a dry surface that allows for easy handling by the wearer while simultaneously protecting the electronics against mechanical deformations during wear and handling. The system requires only 1 unobtrusive patch placed on the forehead, with a reference electrode positioned on the nasal bridge to enable single-patch self-application. The patch contains 6 embedded electrodes: 3 electrodes embedded in the forehead region (2 for measuring EEG and 1 to serve as the ground); 1 embedded in the patch placed below the left eye and 1 above the right eye for measuring electrooculography (EOG); and 1 in the patch on the bridge of the nose to serve as the reference. Data were collected at 250 Hz, and real-time signals were transmitted via Bluetooth to a nearby mobile device or tablet. Participants can easily charge the patches, clean them with alcohol spray, and upload data to the provided tablet.

An Android app was developed in Kotlin to manage continuous EEG data streaming, time alignment, user-defined event labeling, and local file storage. The app uses the native Bluetooth Low Energy application programming interface to discover and connect to the microcontroller device based on its Universally Unique Identifier and subscribes to characteristic notifications for EEG data packets. In the current implementation, the app supports connection to a single device at a time. Upon receiving each packet, the app parses the raw hexadecimal data into floating buffers in memory to allow real-time visualization of waveforms and concurrent storage on-device. To accommodate 8-hour or longer sleep recordings, data are stored in a structured large CSV format. A periodic file rotation scheme for every 4 buffer sets (<0.05 s) is implemented to reduce the risk of file corruption during battery depletion or app shutdown. An image of the app interface with EEG signals is shown in Figure 2.

Participants were enrolled in a 7-day study that included 3 laboratory visits. They were asked to wear a sleep-monitoring patch for 7 nights (see Figure 3A). Collecting data over seven nights helps protect against potential data loss due to participant errors (eg, failure to charge or apply patches correctly) or device malfunctions. At the first laboratory visit, participants were provided with a sleep-monitoring patch, a Samsung tablet, a Philips Actiwatch 2 accelerometer, a sleep log, and an instruction packet for reference in case they forgot any of the procedures explained during the visit. All data measured by the patch were transmitted via Bluetooth and stored on the Samsung tablet. Participants were instructed to wear the accelerometer watch continuously throughout the 7-day study period and to complete a daily sleep log documenting the times the patch was applied and removed, as well as their bedtime and wake time. Sleep log and actigraphy data were used to identify any lapses or artifacts in the sleep patch data (eg, device malfunctions or failure to apply the device). Participants were instructed to clean their skin with an alcohol prep pad before applying the patch each night at bedtime. They were shown how to apply the patch, gently press it to their skin, start the app on the tablet, toggle the switch to the ON position on the patch, and connect the patch to the app. They were also instructed on how to reconnect the patch to the app if they moved away from the tablet (eg, to use the restroom). Each morning upon waking, participants were shown how to rinse the patch with alcohol spray, set it to dry, and connect it to the magnetic charger to recharge the battery. To increase adherence, a researcher texted participants daily to check in and remind them to apply the patch. Participants returned to the laboratory 3 days later to complete the encoding portion of a memory task (data not presented here). During this second visit, researchers downloaded the data from the tablets and checked it to ensure proper patch functioning and participant compliance. Participants were asked if they had experienced any difficulties using the patch system and were provided with additional instructions if needed. Three days later, participants returned to the laboratory for their third and final visit to complete the retrieval portion of the memory task (data not presented here) and to fill out a debriefing questionnaire with several questions about their experience using the sleep-monitoring system. During the third visit, researchers downloaded all data from the patches before dismantling them to recycle the components.

Before the analysis of sleep stages and microarchitecture, it was important to first ensure sufficient data quality for accurate sleep scoring. First, we computed multitaper spectrograms for each dataset to visualize data quality for an entire night, noting any movements, electrode adhesion issues, or disconnections. Visualizing the spectrograms also allowed us to easily observe, in a single graphic, the frequency power indicative of NREM and REM sleep stages and the repetition of these stages over the course of the night’s recording. To this end, we applied the parameters recommended by Prerau and colleagues [32]. We bandpass-filtered the entire time series for each channel, referenced to the nose, between 0.3 and 35 Hz. We also applied a notch filter to remove any power line noise (59‐61 Hz). Second, data files containing less than 5 hours of recording time were removed from further analysis unless the accelerometer also showed a total sleep duration of less than 5 hours. Given the novelty of this device, we chose to be conservative with the data retained for analysis. We assumed that short-duration data files might have been subject to data loss due to loss of connectivity to the tablet and included these files in this loss category in the Results section. Third, a registered polysomnography technologist, naïve to the source of the data (eg, young, old, estimated sleep duration), visually inspected each data file to make a summary judgment of whether the tracings were of sufficient quality to allow manual scoring without difficulty. Data passing these criteria were included for further analyses described below.

This study was approved by the University of Texas at Austin Institutional Review Board (STUDY00001995; Biosensing). Informed consent was obtained from all participants prior to their participation. Study data were deidentified, and all personal identifiers are stored in the Health Insurance Portability and Accountability Act (HIPAA)-compliant REDCap database, separate from the study data. Participants were compensated US $150 for their participation. All data were collected prospectively for this project.

We asked study participants to answer several questions about their experiences with our sleep monitoring system. These data are presented in Table 2. As can be seen in the table, young and older adults had similar responses.

Positive experiences with the system were reported. Specifically, participants noted that the sleep device was easy to use, did not interfere with their sleep, and that they would be likely to wear it again for a research or clinical study. Mann-Whitney U test showed that young and older adults did not differ in their ratings of the experience (all Ps>.05).

As can be seen in Table 1, the number of usable nights varied across participants in both age groups. Seven young and 6 older adults provided 0 usable nights of data (Table 1, Panel A). Among these participants, the primary reason for data loss was poor reference electrode adhesion (young: 78%, 38/49; old: 62%, 26/42), with the remaining losses attributable to participant noncompliance, resulting in no data being produced (young: 22%, 11/49; old: 38%, 16/42). Thus, sleep macrostructure and microstructure analyses were conducted on the 11 young and 12 older adults who contributed at least 1 usable night of data. In Table 1, Panel B, “usable nights” is reported as a percentage of nights in which the device was worn, whereas “not worn/connected,” “poor reference electrode adhesion,” and “Bluetooth issues” are each reported as percentages of all scheduled nights. Among participants with greater than or equal to 1 usable night (Table 1, Panel B), the primary source of unusable recordings was poor reference electrode adhesion (young: 48.1%; old: 36.9%). Only a small number of recordings were lost due to inconsistent Bluetooth connectivity. For data that were retained, we computed SNR for each participant separately using power in the delta band before and after sleep onset. As can be seen in Table 1, the average SNR was similar between groups, with little change over the nights of use. These SNR estimates were similar to those obtained in our prior laboratory-based validation study (22.7 dB) [24], suggesting that home use did not diminish signal quality or stability.

EEG signals can be observed for each of the sleep stages in a representative young and older adult dataset in Figure 3B. Representative spectrograms and associated hypnograms for an entire night from 1 young and 1 older adult are shown in Figure 3C. As is evident in the figure, cycles of sleep stages are visible, with a notable reduction in SWS in the older adult.

Average values for sleep stage and microarchitecture features are shown for each age group in Table 3. Mann-Whitney U tests were performed to compare the sleep variables between groups. As can be seen in the table, older adults exhibited lower sleep efficiency (total sleep time divided by time in bed), less time in SWS (N3) and REM, but more time in N2. Regarding the microarchitecture features, spindle activity in both N2 and N3 stages was significantly reduced with age. SWA and SO density were not significantly reduced with age.

The goal of this pilot study was to assess the acceptance and feasibility of our newly developed, ultrathin, wearable epidermal electronics system for at-home sleep monitoring in healthy young and older adults. To our knowledge, this is the first report of the home deployment of a forehead, fabric-based electronics, wireless, wearable system for sleep assessment in young and older people. Our results show that (1) participants across young and older age groups rated the system as easy to apply and comfortable to wear during sleep; (2) nights of data were lost for each age group, primarily due to poor adhesion of the reference electrode; (3) signal power was stable over multiple nights of wear at home for retained data; (4) independent (nonauthors) trained sleep technicians verified that they would be able to score retained datasets with little difficulty; and (5) typical age-related differences in both macroarchitecture and microarchitecture features were observed. We discuss these results and their implications below.

Across both young and older adult groups, participants rated the device and tablet application as easy to use, highly comfortable to wear, nonintrusive to their sleep, and something they would use again for both clinical and research purposes. These ratings were as high as or higher than those assessing participant comfort and usability for various wearables, including dry electrode patches for EEG [34], supporting the acceptance of this novel technology for home sleep monitoring in adults across the lifespan. The nose bridge placement for the reference channel is commonly used in many traditional wake EEG studies and in our prior laboratory-based validation study of this sleep monitoring system [24], without similar data loss. Consequently, we used the same electrode placement in this home deployment pilot study. Despite providing thorough instructions to participants—including cleaning the skin with alcohol swabs, washing the patch after every use, and gently pressing the electrodes to the skin—there was a significant amount of data loss, primarily due to poor adhesion of the reference electrode. We believe that a combination of factors, including participants rubbing their nose while sleeping, individual differences in the fit of the electrode over the nose bridge, sweating or oil secretions, and inconsistent adherence to our instructions, contributed to the data loss. In our future work, we will relocate the reference channel to the forehead patch, similar to the location used in other forehead-only EEG systems [16,17], to mitigate these issues.

Finally, results show the age group differences in both macroarchitecture and microarchitecture features typically observed in laboratory-based polysomnography studies [26-29]. Specifically, SWS (N3) was reduced, while N2 was proportionally elevated in older compared to younger adults. Our device was also able to detect NREM microarchitecture features, including SOs and spindles, with spindles particularly attenuated with age. Multiple studies have found age-related attenuation of NREM sleep features, including SWS and spindles, which numerous studies have shown play an important role in the sleep-dependent consolidation of episodic memories [35]. These age-associated reductions are concomitant with age-related reductions in sleep-dependent consolidation, indicated by memory retention, with individual differences in the amount of SWS, as well as the density of spindles, predicting those in memory performance across the adult lifespan [36]. Importantly, low levels of these and other sleep EEG features may predict conversion to mild cognitive impairment or dementia over 3 to 6 years [37] or subtle general cognitive decline in cognitively unimpaired older adults [38]. The present findings highlight the potential for using our sleep patch system to monitor sleep from the comfort of one’s own home to track changes over time in the levels of sleep features that may forecast future cognitive decline.

Our sleep monitoring wearable system is novel in many respects. While wearable sleep EEG monitoring electronic systems exist, such as forehead headbands containing one or several dry electrodes, they are hard and bulky with rigid form factors [13]. They additionally lack electrodes placed to monitor eye movements, which facilitate the scoring of REM sleep. Other wearable systems with smaller form factors, such as those worn around or in the ear, either have to be customized for an individual for optimal fit [39] or contain wires [21,39], which could contribute to relatively low signal quality due to mechanical noise [40]. Our wearable system, with its wireless nanomembrane electrodes embedded in soft adhesive fabric, is designed to maximize comfort, ease of application, and obtain reliable signals over multiple nights. The Bluetooth-based wireless data transmission to a mobile tablet with a simple user interface offers additional portability and accessibility for our system. Finally, the more costly circuit components can be recycled, reducing production costs and negative environmental impacts. As discussed in our prior work, our manufacturing process is highly scalable, making it possible to produce and ship devices to a large number of participants.

There are a few limitations to note. A primary limitation of the current home-deployment prototype relates to data loss, which occurred mostly due to the instability of the reference electrode placed on the bridge of the nose. While this placement was effective and not problematic in our prior laboratory-based validation study, home-based recordings introduced additional variability due to interindividual differences in facial anatomy and sleep-related movements. In particular, contact with bedding during sleep and variations in nasal shape made stable adhesion difficult for some participants, resulting in unusable recordings when adhesion was insufficient. These findings highlight a limitation specific to unsupervised home use that could not have been fully anticipated from laboratory testing alone. To address this issue, we are developing updated designs in which all electrodes, including the reference, are integrated into a single forehead patch, relocating the reference electrode to a mechanically more stable position to improve robustness and data retention during future home deployments. Given that greater attenuation of certain sleep features, including SWS, may predict later cognitive decline, it would be of great interest to identify individuals at the greatest risk of decline. Future studies recruiting larger samples of participants and tracking their sleep and cognitive changes over time would allow for such investigations.

Our system overcomes the limitations of home polysomnography and wearable sleep monitoring devices with regard to comfort, sustained signal reliability, and accurate sleep analysis. These results help lay the groundwork for studies in participants at risk for AD to determine whether particular sleep EEG patterns, measured accurately from one’s own home, can help identify individuals most likely to exhibit cognitive decline. Hesitancy to participate in research is well known among historically minoritized individuals (eg, [41]), and furthermore, individuals with poor sleep may be hesitant to participate in a laboratory-based sleep research study. Our sleep monitoring patch, which is manufactured using laser micromachining and printing, is highly scalable and could be shipped to large numbers of participants for future cognitive and clinical studies. Sleep architecture results could potentially differ as a function of age and/or race or ethnicity and could serve as targets for future personalized brain stimulation, cognitive behavioral therapy, or pharmacological interventions.