Data presented here are part of a larger clinical trial to examine the efficacy of the SSM to (1) accurately measure naturalistic sleep at home in people with chronic insomnia as compared to healthy-sleeper controls and wrist-worn actigraphy and (2) test the ability of the SSM’s integrated sleep coaching program to improve sleep in people with chronic insomnia compared to an active control group using the current standard of care, online cognitive behavioral therapy for insomnia, and passive control groups of individuals with chronic insomnia and healthy-sleeper controls. All participants collected 8 weeks of continuous nightly recordings using the SSM, with 1 week of wrist actigraphy recordings preceding and 1 week following the 8-week monitoring phase. We planned to enroll 90 carefully screened adults with chronic insomnia and 30 carefully screened healthy adult controls. The chronic insomnia group was randomized to 1 of 3 groups: SSM alone (passive control), SSM with online cognitive behavioral therapy for insomnia (active control), and SSM with a coaching program (test intervention), relative to healthy controls using the SSM alone. Here, we present results from aim 1, which was to evaluate the ability of the SSM to accurately measure sleep at home in people with chronic insomnia as compared to healthy-sleeper controls.

Our preliminary data from a prior study with 17 chronic sleep-onset insomniacs and 29 healthy controls show a large effect size (Cohen d=1.99) for between-group differences. Using this effect size in our power calculations, we determined that 30 participants must be enrolled per group (90 total) in the proposed study to achieve 80% statistical power to address each of our aims at a type I error threshold of .05.

We enrolled adults with chronic insomnia as well as healthy-sleeper controls. Participants were recruited via print and online advertisements. Recruitment took place between November 2019 and November 2021. Six participants were screened in the laboratory and completed data collection before the COVID-19 pandemic. All other participants were screened remotely—through videoconference meetings and secure administration of online questionnaires—to adhere to COVID-19 social distancing requirements.

Prior to the start of the at-home sleep monitoring study, participants were screened via an interview with trained staff to determine whether they met the eligibility criteria. All participants were aged 18 to 65 years. Participants in the chronic insomnia group met criteria for chronic insomnia as defined by the International Classification of Sleep Disorders, 3rd edition, with no other clinically relevant condition contributing to their reported sleep disturbance [3]. Specifically, individuals with insomnia had difficulty sleeping at least 3 times per week for a period of at least 3 months. In addition, those with insomnia reported having at least 1 of the following next-day consequences: fatigue or malaise; attention, concentration, or memory impairment; impaired social, family, occupational, or academic performance; mood disturbances or irritability; daytime sleepiness; behavioral problems (eg, hyperactivity, impulsivity, and aggression); reduced motivation, energy, or initiative; proneness to errors or accidents; or concerns about or dissatisfaction with sleep. They had no other diagnosed sleep disorder other than chronic insomnia, as assessed by questionnaire. The reported sleep or wake concerns could not be explained by inadequate opportunity (ie, enough time allotted for sleep) or inadequate circumstances (ie, an environment that was safe, dark, quiet, and comfortable) for sleep.

Individuals in the healthy good-sleeper control group reported no current clinically relevant history of medical disorders or other illnesses and were free of suspected sleep disorders, as determined by the STOP-BANG (snoring, tiredness, observed apnea, pressure [high blood pressure], BMI, age, neck circumference, and gender) questionnaire [29] (score no greater than low risk for OSA) and the Pittsburgh Sleep Quality Index (score<5) [30]. Participants were required to have daily access to an iPhone (Apple Inc) to run the smartphone app associated with the SSM. All participants reported that they were not currently engaged in shift work.

All participants provided written informed consent, and the study was approved by the institutional review board of Washington State University (protocol 17379). All sleep recording data were collected via participant IDs rather than direct identifiers. Upon completion of the study, participants were compensated with a US $20 gift card and were allowed to keep their SSM device.

For data collection, participants were instructed to place the SSM within arm’s length on their bedside table or nightstand and to collect 8 consecutive weeks of nightly sleep recordings. Participants manually initiated and ended each nightly recording through the smartphone app associated with the SSM.

The SSM recorded sleep data in 30-second epochs classified as “wake,” “light,” “deep,” “rapid eye movement,” and “absence.” “Absence” denotes that the subject was not present in the bed. The analysis window begins at the first epoch following the last absence preceding sleep onset and ends at the first absence after the final sleep epoch or at the end of the recording, whichever occurs first. Data were measured by the SSM and processed by the associated smartphone app. Time in bed and sleep parameters were estimated based on ultra-low energy radar measurements and were automatically uploaded to the cloud and later downloaded by the researchers. These data were linked to and coded for each participant using a unique study ID.

The following SSM sleep parameters were assessed for each of the recorded nights:

Data from the SSM were tabulated day by day, from which weekly means and SDs were calculated. Sleep variables were analyzed using maximum likelihood estimation in a nonlinear mixed-effects model called PROC NL Mixed (SAS, version 9.4; SAS Institute Inc) [31] to establish group means and within-subject SDs (as an index of night-to-night variability), while also controlling for age and sex. A random effect on the intercept for subjects was included to account for idiosyncratic, systematic interindividual differences. Results compared the chronic insomnia group to the good-sleeper controls by means of planned contrasts determined prior to data analysis. No imputation of missing data was performed for any of the analyses, as they were not contingent upon having complete datasets. In addition, 95% CIs for fixed effects and derived contrasts were computed from the model-estimated SEs. Data are plotted for female participants (who predominated in the sample) and the grand average age (36.1, SD 12.14 years).

As shown in the CONSORT (Consolidated Standards of Reporting Trials) diagram (Figure 1), 127 participants were enrolled in the study: (n=96, 76%) adults with chronic insomnia and (n=31, 24%) healthy-sleeper controls; 112 completed the study. Table 1 shows that the final sample included 83 (74%) adults with chronic insomnia, aged 38.3 (SD 12.0) years, of which 58 (69.9%) were female. The average duration of insomnia was 8.5 years, and insomnia was of moderate severity as measured by the Insomnia Severity Index [31]. The controls were 29 (26%) healthy adults with self-reported good sleep, aged 29.7 (SD 8.8) years, of which 21 (72.4%) were female. The composition of our sample is consistent with higher prevalence rates of insomnia observed in women [32,33]. Baseline characteristics of the sample show that the insomnia group had significantly higher levels of subjective daytime sleepiness, as measured by the Epworth Sleepiness Scale [34], and sleep disruption, as measured by the Pittsburgh Sleep Quality Index [30]. They also endorsed significantly higher levels of anxiety and depression symptoms, as measured by the Generalized Anxiety Disorder 7-item scale [35] and the Patient Health Questionnaire–9 [36]. SSM data capture was generally high throughout, with an initial 97.3% (n=123) participation rate, a 91.7% (n=116) average weekly participation rate, with 83.0% (n=105) of participants using their device at least 2 nights per week across each of the 8 weeks of data collection. Across the entire 8-week period, overall use was higher in the insomnia group (75.9%) than in the healthy controls (71.8%). Overall weekly use in the insomnia group was more than 80% in weeks 1 to 2 before peaking in week 3 (83.8%), and then lowest at week 8 (63.2%). For the control group, overall use was highest in week 1 (80.8%) and lowest 69.5% in week 8 (69.5%).

Figure 2 shows the daily measurements made with the SSM, averaged over the duration of the study (weeks 1-8) for participants with chronic insomnia as compared to the good sleepers. Table 2 shows the comparisons between study conditions for means and SDs. Compared to the good-sleeper controls, the insomnia group had lower mean sleep efficiency (P=.001) with higher night-to-night variability (P<.001) in SDs. For sleep latency, participants with insomnia had a higher mean sleep latency (P=.001) and greater night-to-night variability in SDs (P<.001) compared to good-sleeper controls. Additionally, participants with insomnia exhibited more intermittent wakefulness both in terms of daily means (P=.001) and SDs (P<.001) as compared to the good-sleeper controls. Across groups, no significant differences were observed in mean time in bed, which averaged 7.65 (SD 0.30) hours per night (P=.11), or in night-to-night variability for SDs (P=.34). The groups also did not differ in mean sleep duration, which averaged 6.59 (SD 0.16) hours per night (P=.91), or in night-to-night variability for SDs (P=.12).

To provide further context, on average, sleep efficiency was 5.34% lower, sleep latency was 10.15 minutes longer, and intermittent wakefulness was 18.45 minutes higher in the individuals with chronic insomnia than in the good-sleeper controls. Night-to-night variability was greater for sleep efficiency by 1.77%, for sleep latency by 8.80 minutes, and for intermittent wakefulness by 8.60 minutes among individuals with insomnia compared to good-sleeper controls.

In this longitudinal naturalistic study, we observed that the SSM detected objective differences in the sleep of individuals with primary chronic insomnia compared to healthy, good-sleeper controls. Even though average time in bed and sleep duration were similar, individuals with chronic insomnia demonstrated significantly less sleep efficiency, longer sleep latency, and more intermittent wakefulness than the good-sleeper controls. Importantly, we report a novel longitudinal, objective finding of significantly greater night-to-night variability in sleep efficiency, sleep latency, and intermittent wakefulness in participants with chronic insomnia compared with good-sleeper controls.

An increased focus on night-to-night variability in studies of insomnia has become prevalent in recent literature [5,13,17-19,37,38], but technological and methodological challenges—such as smaller sample sizes [5,13,17], lack of objective measures [37], shorter observation periods [5,13,17,19,37], and a focus on older adult populations [5,17,18]—have hampered efforts to consistently demonstrate night-to-night variability. While work by Buysse et al [5] has shown greater variability in intermittent wakefulness and sleep efficiency using 2 weeks of actigraphy, this was observed in a group of older adults (mean age 71.4 years) with insomnia, and they did not find significant differences in variability in sleep latency or in mean values of any sleep measure, as found in this study.

Previous work by Wohlgemuth et al [39] argued that 1 week of recordings with polysomnography or sleep logs was sufficient to assess temporal stability in sleep parameters in elderly adults with insomnia but noted that up to 3 weeks of recordings were needed to demonstrate stability in metrics such as WASO and sleep latency. This may explain why previous research has been unable to detect the differences observed in this study and provides support for longer observational periods across wider age ranges.

We show that a validated, contactless, longitudinal CST can detect differences in sleep patterns between individuals with insomnia and healthy good-sleeper controls. Thus, the SSM can be considered a valid method to accurately assess the sleep of those with insomnia and, as such, has the potential for clinical utility in the screening of this condition and may be considered for use in insomnia medication clinical trials.

Using a noncontact, CST-enabled, automated, relatively unobtrusive recording of sleep over extended periods (8 consecutive weeks). We found evidence of substantially increased night-to-night variability in measures of sleep onset and sleep continuity that has not previously been observed. The heightened variability from day to day in these sleep characteristics and the resulting uncertainty in the sleep experience of people with chronic insomnia may contribute to the characteristic concerns and burden of this disorder. We posit that night-to-night variability is a hallmark of the chronic insomnia phenotype and that objectively and longitudinally capturing this variability in an ecologically valid manner is key to investigating and understanding the disorder.

This study focuses on the characterization of night-to-night variability through the examination of overall means and SDs in those with insomnia vs good-sleeper controls. The SSM also has the capability to perform sleep staging; however, as this was not the focus of this study, we are unable to comment on its accuracy.

Six of the 112 study participants completed data collection prior to the COVID-19 pandemic, and 106 were studied during the pandemic. We do not know to what degree the pandemic may have contributed to symptomology in our sample of individuals with chronic insomnia, and whether and how the pandemic may have shaped the observations in this study. Although the differences we found between individuals with chronic insomnia and good-sleeper controls are likely to be robust qualitatively, they may not generalize quantitatively to postpandemic circumstances. It is worth noting that sleep duration for our study participants averaged about 30 minutes less than the consensus recommendation of 7 or more hours per night on a regular basis [40], regardless of group or condition. However, the average sleep duration of our predominantly female sample was consistent with wearable-based findings for average sleep duration in US adults before the COVID-19 pandemic [41]. Our statistical analysis controlled for 2 specific factors (ie, age and sex) and combined all participants with insomnia into a single group. Future studies should consider including additional factors that may influence sleep patterns, as well as examining the potential effects of group assignment.

Our sample included 83 individuals with chronic insomnia and 29 healthy good-sleeper controls. While our included age range was broad (18 to 65 years), participants in our sample were relatively young overall (mean age 36.1, SD 12.1 years), which may indicate selection bias related to online recruitment efforts as well as the study requirement to have a smartphone (iPhone). Combined, these factors may have hindered access to a more representative age sample as well as those that were less economically affluent. Although the disparity in group size and age might have biased our results, it is more likely that this would have tempered, as opposed to enhanced, the observed differences between individuals with chronic insomnia and good-sleeper controls.

Capturing objective, longitudinal, contactless, within-person variability of sleep and mean differences in common measures of insomnia in the home setting adds an important dimension to our understanding of poor sleep and provides a more comprehensive, ecologically valid characterization of sleep problems as experienced in daily life. This study demonstrates the potential clinical and research utility of CSTs in the characterization and potential management of insomnia and our understanding of the disease.

In this 2-month, naturalistic sleep monitoring study, we documented the novel finding of persistently elevated night-to-night variability in measures of sleep onset and sleep continuity in individuals with chronic insomnia compared to healthy good-sleeper controls. This important finding was made possible by the use of a CST device, which enabled long-term, objective, naturalistic sleep recordings in a large sample at a relatively low financial and logistical cost. Night-to-night variability may be a critical aspect of the chronic insomnia phenotype, and long-term naturalistic sleep monitoring, as demonstrated in this study, is essential to reliably document this phenomenon.

It remains to be determined whether recognition of night-to-night sleep variability as a hallmark of chronic insomnia can shed new light on some enigmatic features of the disorder, such as the discrepancy between subjective and objective daytime impairments [11,42], the heightened sensitivity to performance impairment during sleep deprivation [43], or the dampened responsiveness to stressors [44,45]. In this context, the use of contactless CST for long-term naturalistic sleep monitoring in insomnia may constitute an important methodological advancement in both research and clinical settings.