INTRODUCTION
Both short and long sleep durations are consistently linked to higher mortality and adverse health outcomes, such as an increased risk of stroke, diabetes, heart disease, and psychiatric disorders [1,2]. Shift work is also known to be linked to poor health outcomes, including a heightened risk of diabetes, obesity, hypertension, atrial fibrillation, coronary heart disease, stroke, and cancer [3,4]. This is important, as approximately 29% of U.S. workers have work schedules outside of the traditional workday [5], with 4% working exclusively on nights. Expanding on this knowledge, irregular sleep-wake schedules, marked by significant day-to-day variability in sleep timing, have recently been associated with increased mortality risk [1], cardiovascular disease (CVD), and adverse metabolic profiles, such as elevated blood pressure and insulin resistance [6]. However, the prevalence of such irregular sleep-wake schedules has only been studied in young populations regarding school times and academic performance. To our knowledge, there are no objective studies in a large population tracking more than 2–3 weeks per individual [2]. This study aimed to characterize sleep-wake schedules in a large U.S. sample.
Consensus guidelines published by the American Academy of Sleep Medicine and Sleep Research Society [7] recommend that adults get at least 7 hours of sleep per night to support optimal health and lower the risk of adverse outcomes. Similarly, the National Sleep Foundation [8] recommends a minimum of 7 hours of sleep per night but specifies a maximum of 9 hours of sleep per night. Unfortunately, no consensus recommendation considers sleep timing or other circadian factors.
Emerging evidence indicates that individuals with greater variability (i.e., less regularity) in night-to-night sleep duration also exhibit poorer cardiometabolic health, characterized by greater adiposity, inflammation, higher resting blood pressure, and greater likelihood of metabolic syndrome, compared with those with more consistent sleep patterns [9-11]. Additionally, among older adults without CVD at baseline, those with the highest variations in sleep duration had more than twice the odds of developing CVD over approximately 5 years of follow-up, compared with those with the lowest variations [6]. This also holds in the younger population, in whom irregular sleep duration was found to be associated with poorer microvascular function and poorer academic performance [12,13].
Despite its importance, the prevalence of irregular sleep-wake schedules in a large population objectively measured over multiple nights has not been studied. Therefore, this study aimed to characterize sleep-wake schedules in a large U.S. sample.
METHODS
Participants
Data were recorded using a validated commercially available home-based sleep-monitoring device (Sleeptracker-AI monitor; Fullpower Technologies, Santa Cruz, CA, USA) [14]. The participants were selected from a population of Sleeptracker-AI Monitor users with sleep recordings from 2019, totaling 51,215 individuals with 9,360,523 total sleep recordings. To avoid seasonal confounders, all Sleeptracker-AI Monitor users with at least 300 days of recordings between January 2019 and December 2019, totaling 12,507 individuals, were included in this analysis. More recent data were excluded to avoid the effects of the coronavirus disease pandemic. Days with multiple sleep recordings were excluded from the study. The population was divided into 6 age groups, as we aimed to include only a few years in each age group to detect subtle differences between these groups: Groups 1 (20–29), 2 (30–39), 3 (40–49), 4 (50–59), 5 (60–69), and 6 (70–79). The number of individuals included in these groups 1–6 was 689, 2,807, 3,062, 2,917, 1,845, and 713, respectively (with the remaining 474 individuals outside this age range or not reporting their age).
The protocol for reviewing and analyzing these de-identified sleep studies and datasets was submitted to Stanford University’s Institutional Review Board (IRB) Research Compliance Office, which concluded that IRB approval was not required as the protocol did not involve human participants. Furthermore, users who onboard into the Sleeptracker-AI monitor Fullpower system agreed to the following clause (by checking a box): “From time to time, we will share anonymized information with our scientific research partners, and they may publish from time to time, the results of the research in scientific journals.”
Sleeptracker-AI monitor recordings
Sleeptracker-AI monitor users purchased or obtained the commercially available device bundled with a bed mattress and foundation or as a standalone device. The users were instructed to place the two sensors in parallel on both sides of their beds between the mattress and foundation beneath the pillow region. The sensor measured 14.7 cm in length, 7.7 cm in width, and 1.5 cm in thickness. Each sensor captured the physical forces exerted by the sleeper through the mattress, including 1) body movements, 2) respiratory efforts through chest and abdominal forces on the mattress, 3) heartbeats as a ballistocardiographic signal superimposed on the previous signals, and 4) snoring vibrations transferred through the mattress. These signals were processed by an automated algorithm that uses signal processing, machine and deep learning models, and statistical inference techniques to separate the effects and produced the following estimates: 1) bed occupancy, 2) sleep vs. wake when the bed is occupied, 3) light (N1+N2) vs. deep (N3) vs. rapid eye movement (REM) sleep, and 4) apnea or hypopnea. Although respiratory efforts and ballistocardiographic forces were included as input signals to the models, the respiratory and heart rates (RR and HR, respectively) were not included as direct inputs. Two parallel sensors (one on each side of the bed) were connected to a Sleeptracker-AI processor, which produced sleep recordings using the above estimations.
Ding et al. [14] performed validation of Sleeptracker-AI recordings against polysomnography (PSG), including evaluating agreement of Sleeptracker-AI estimated sleep architecture with PSG in 102 sleep studies, and finding “high accuracy, sensitivity, and specificity in estimating sleep continuity measures and sleep architecture.” More specifically, in terms of sleep architecture parameters, our interest was in total sleep time (TST) and sleep onset (SO) and their standard deviations (SDs); therefore, agreement with PSG in TST and SO latency is essential, and wake after sleep onset (WASO) and per-30-s-epoch sleep/ wake are also relevant indications. In the validation study by Ding et al. [14], TST showed a bias of +6.3 minutes and a correlation of 0.96, WASO showed a bias of -10.2 minutes and correlation of 0.88, and SO latency showed a bias of +4.0 minutes and correlation of 0.34, and each correlation was statistically significant. Per the 30-s epoch, sleep vs. wake agreement with PSG was 93.3% (92.4% vs. 94.1%, respectively), with sensitivity of 96.8% and 71.3% for sleep and wake, respectively.
Statistical analysis
First, descriptive analyses were performed detailing the population statistics. To analyze sleep-wake schedules and sleep regularity, the TST SD and SO SD were included as parameters. Users were categorized into six age groups, and weekly summaries of sleep parameters were collected for each participant. The TST and SO were considered for each participant and night of sleep, and for each week in which the participant had a recording for each night of the week; each sleep parameter SD was computed, including TST SD and SO SD. The means for each sleep parameter SD were then calculated for each participant across weeks. We defined irregular sleep as a SD of >1 hour in sleep duration and onset. This threshold was reported by Huang et al. [6], who demonstrated that SD >1 hour of sleep duration or onset significantly increased the risk of CVD. This criterion enabled us to analyze sleep irregularities with a meaningful cut-off based on prior research linking sleep variability with adverse health outcomes, particularly CVD risk.
Second, to understand the extent to which the weekday-weekend split influenced the sleep-wake schedule and duration variability, the percentage variance explained by the weekday-weekend split for TST SD and SO SD was assessed using analysis of variance (ANOVA).
Third, inferential analyses were performed to understand the effects of the sleep-wake schedule and duration variability on five secondary sleep parameters: TST, sleep efficiency (SE%), WASO, mean HR, and mean RR. Specifically, unpaired t-tests were performed to compare secondary sleep parameters in participants with generally high and low regularity parameters (TST SD and SO SD). Paired t-tests were performed to compare secondary sleep parameters per participant for weeks with high- and low-regularity parameters (TST SD and SO SD).
Statistical analyses were performed using Python (Python Software Foundation, version 3.8.3; https://www.python.org/).
RESULTS
Participants
Data from 12,507 users (51.5% male, 46.9% female, 1.6% no sex specified; mean age, 48.5±13.4 years) who met the inclusion criteria of the study were analyzed. The demographic characteristics of the study population are shown in Table 1.
Sleep parameters
Data from a total of 4,175,260 recorded nights were included in the analysis. The overall estimated TST SD across participants’ means was 66.1 (18.7) minutes, and SO SD was 55.6 (20.5) minutes; the SD of the mean sleep parameter SD is shown in parentheses. ANOVA was used for each participant and week to determine the percent variance in sleep duration over the week explained by weekday-weekend splits. Aggregating across weeks, only 25.0% (10.9%) of the variance (mean, SD across weeks) was explained by the difference between weekends and weekdays, and for SO SD, this value was only 26.7% (11.3%). In other words, substantial variation remained even when considering only weekdays.
The estimated TST SD in age groups 1, 2, 3, 4, 5, and 6 were as follows: 70.7 (20.0), 67.2 (18.0), 66.8 (18.5), 66.1 (18.4), 63.4 (18.2), and 60.5 (18.9) minutes (Fig. 1A). The differences between neighboring age groups were significant when assessed using Welch’s t-test at p<0.001 for the comparisons of groups 1 vs. 2, 2 vs. 3, and 5 vs. 6, with each showing a lower value for the older group. No other differences between neighboring age groups were significant, even at p<0.05. The estimated TST SD in age groups 1, 2, 3, 4, 5, and 6, when considering only weekdays, was as follows: 60.6 (18.7), 58.1 (16.2), 57.0 (16.6), 57.1 (16.8), 57.0 (16.7), and 57.6 (16.9) minutes (Fig. 1B). In this case, the differences between neighboring age groups were significant at p<0.05 (but not at p<0.001) only for the comparisons of groups 1 vs. 2 and 2 vs. 3. The estimated SO SD in each age group were 62.1 (21.1), 57.4 (19.4), 57.0 (20.2), 55.7 (20.0), 51.6 (20.3), and 46.7 (20.8) minutes (Fig. 1C). In this case, the differences between neighboring age groups were significant at p<0.001 for the comparisons of groups 1 vs. 2, 4 vs. 5, and 5 vs. 6, and additionally at p<0.05 for the comparison of group 3 vs. 4.
Regular vs. irregular sleep duration
When categorically divided into two groups with regular or irregular sleep duration (with mean TST SD <60 minutes and ≥60 minutes, respectively), we found that 67%, 61%, 60%, 58%, 53%, and 47% of groups 1, 2, 3, 4, 5, and 6 and 58.4% of participants overall had an irregular sleep-wake schedule (Supplementary Fig. 1 in the online-only Data Supplement). We also compared physiological sleep parameters (HR and RR) as well as sleep architecture parameters (TST, SE, and WASO) in participants with typically regular (mean TST SD <60 minutes) vs. irregular (≥60 minutes) sleep duration. We found a significant difference in all parameters (Fig. 2), with unpaired t-statistics for regular vs. irregular sleep duration groups of -7.9 for HR, -5.6 for RR, +11.9 for TST, +33.2 for SE%, and -32.7 for WASO, with the corresponding p-values all below 2e-8. These results remained significant (p<6e-4) when the participants were divided into subgroups of based on sex (Supplementary Fig. 2 in the online-only Data Supplement).
We then compared these same sleep parameters in regular scheduled weeks vs. irregular scheduled weeks, averaging the per-participant means for each of these across all participants, and found that all parameters again had a significant difference, this time by paired t-test, paired by participant (Fig. 3). The paired t-statistics for regular vs. irregular scheduled weeks were -6.9 for HR, -5.1 for RR, +24.3 for TST, +54.6 for SE%, and -40.0 for WASO, with the corresponding p-values all below 4e-7. These results for the sleep architecture parameters remained significant when participants were divided into subgroups based on sex (p-values all infinitesimal, with t-statistics >20 in absolute value); however, the results for the physiological sleep parameters only remained significant (by paired t-test) for male participants (p<1-e9) but not for female participants, for whom neither HR (t-statistic -1.9, p>0.05) nor RR (t-value -1.0, p>0.3) showed significant differences (Supplementary Fig. 3 in the online-only Data Supplement). Note that because this analysis split nights into regular and irregular nights by participant, seasonal confounders may have been present, unlike the previous analysis (comparing across participants instead of comparing across weeks), where seasonal confounders were less likely to be present because we included only participants with recordings for a significant majority (300) of nights of the year. Another caveat in this analysis comparing across weeks is that each participant may have had different numbers of regular and irregular weeks, leading to varying levels of certainty in each per-participant paired comparison. Nevertheless, we included it because the paired analysis by participant provides additional detail beyond the conclusions of the across-participant comparison (between participants with typically regular vs. typically irregular sleep durations) without the demographic confounders that may inherently arise from comparing across participants.
Regular vs. irregular sleep timing
In addition to quantifying regular versus irregular sleep in terms of duration, we considered regular versus irregular SO. When again categorically divided into two groups with regular or irregular SO (with mean SO SD <60 minutes and ≥60 minutes, respectively), we found that 47%, 39%, 37%, 34%, 28%, 22% of groups 1, 2, 3, 4, 5, and 6 and 34.9% of participants overall had an irregular SO schedule (Fig. 4). The unpaired t-statistics for regular vs. irregular sleep timing comparison were -9.0 for HR, -8.3 for RR, +18.6 for TST, +26.8 for SE%, and -25.4 for WASO, with the corresponding p-values all below 1e-16.
We again compared physiological sleep parameters and sleep architecture parameters as above in typically regular participants with a regular (mean SO SD <60 minutes) vs. irregular (≥60 minutes) SO. We found a significant difference for all parameters (Fig. 5) and when divided into subgroups based on sex (Supplementary Fig. 4 in the online-only Data Supplement), with the p-values all below 2e-9.
Finally, we compared these parameters in weeks in which the SO timing was regular or irregular by the same standards, averaging the per-participant means for each of these across all participants. We found that all parameters again had a significant difference, this time by paired t-test, paired by participant (Fig. 6), with paired t-statistics for regular vs. irregular SO timing of -11.6 for HR, -2.8 for RR, +45.4 for TST, +37.1 for SE%, and -30.2 for WASO, with the corresponding p-values all below 1e-30 except for the one for RR, which was below 0.005. These results remained significant for HR and the sleep architecture parameters when divided into subgroups based on sex, with the p-values all infinitesimal for sleep architecture parameters and p-values for HR below 1e-33 (male participants) or below 4e-6 (female participants). However, the results for RR was only significant (by paired t-test) for male participants (p<1e-33) but not for female participants (p>0.8) (Supplementary Fig. 5 in the online-only Data Supplement). Similar caveats applied to the cross-week comparison, as noted at the regular vs. irregular sleep duration section.
DISCUSSION
Short and long sleep durations are associated with unfavorable health outcomes. However, irregular sleep-wake schedules, characterized by high day-to-day variability in sleep timing and duration, have been understudied; when reported, it is mostly only in children and adolescent populations with regard to school times and academic performance [13]. Reports in the adult population are typically limited to sleep patterns on shift workers and the influence of such patterns on their health and well-being; however, knowledge about the general adult population is lacking. Furthermore, most previous studies were based on subjective and self-reported tools and included only data from a limited number of nights. Therefore, this study aimed to characterize sleep-wake schedules in a large U.S. sample using a validated home-based sleep-monitoring device. We hypothesized that most of the U.S. population would have irregular sleep timing. Our results showed that this is indeed the case, as irregular sleep timing was expected, with 58.4% and 34.9% of the population having an irregular sleep-wake schedule based on sleep duration and sleep timing, respectively. This high prevalence was identified using a threshold of over 1-h SD, which was selected based on evidence from Huang et al. [6], who demonstrated that this level of variability significantly increased the risk of CVD. By applying this criterion, we aimed to capture the levels of irregularity associated with meaningful health risks. This association was consistently observed across various strata of the study sample, including the subgroups categorized by age and sex.
With regard to age, we divided our population into six different age groups, and interestingly, our findings followed a clear age-dependent trend, with older age corresponding to more regular sleep-wake schedules. This trend was surprising, as we expected that schedules might become more variable with age, particularly after retirement, potentially increasing the risk of irregular sleep patterns. Instead, the data suggest that older adults tend toward greater regularity, perhaps because of the increased prevalence of age-related sleep disorders, which often require structured routines to manage symptoms. As individuals grow older, sleep-related issues, such as sleep-disordered breathing, insomnia, and REM sleep behavior disorder, become increasingly common, adding complexity to the relationship between sleep irregularity and health risks. Changes in sleep architecture, particularly a reduction in slow-wave and REM sleep stages, further contribute to cardiovascular and metabolic risks in aging populations. The elevated prevalence of these disorders may partly explain the reduced variability in sleep schedules observed among older adults in our study, as sleep challenges often prompt more regimented routines to counter sleep fragmentation and improve sleep quality. However, despite potentially greater schedule regularity, underlying health risks remain, emphasizing the importance of addressing the quality and timing of sleep in older populations.
Furthermore, older adults may have fewer obligations and experience lower energy levels throughout the day, making prioritizing a consistent sleep window easier. Additionally, older populations tend to have more advanced circadian rhythms, which can lead to more early morning awakenings and difficulty falling asleep, thereby reinforcing a more regular sleep schedule. This age-dependent trend highlights the importance of sleep consistency across all age groups. For instance, previous research has shown that in college students—primarily in their 20s (our youngest age group, which exhibited the highest variability)—irregular sleep durations were linked to measures of peripheral vascular function [12]. Our findings align with those of the same study, where over 75% of the population studied demonstrated sleep duration SD values more significant than 1 hour, indicating that most participants experienced night-to-night sleep durations that varied by more than 1 hour from their mean value. Another study identified irregular sleep as a risk factor for poor academic performance among medical students [15].
Regarding the older population, according to our data, they had less variability than the younger cohorts did; however, more than two-thirds of individuals aged over 60 years still had an irregular sleep schedule. Two studies have reported an association between variability in sleep duration and timing with obesity in older adults [11,16]. In another study, greater sleep irregularity was also correlated with a 10-year risk of CVD and greater obesity, hypertension, fasting glucose, hemoglobin A1C, and diabetes status [17]. Recently, Huang et al. [6] reported that among a group of older adults without known CVDs at baseline, those with the most irregular sleep durations had more than double the odds of developing CVD over approximately 5 years of follow-up, compared with those with the most consistent sleep durations. In other words, irregular sleep schedules are common and significant modifiable behaviors across all age groups. Interestingly, longevity in humans is also associated with regular sleep patterns [18].
Various lines of evidence suggest that circadian disruption in specific contexts may increase the risk of CVD and that this risk is not limited to age-related factors. For instance, studies from the U.S. and several European countries reported a significant increase in heart attacks and strokes during the first few days following the spring transition to daylight saving time [19].
We also considered whether these findings could be attributed to differences between weekends and weekdays, as social jetlag (the variation in sleep timing between workdays and free days) has been linked to adverse cardiometabolic risk factors [20]. Specifically, individuals tend to sleep and wake up later on weekends and might also have higher TST on weekends. With this in mind, we analyzed the data by exclusively examining weekdays, when the schedule tends to be more regular due to social norms (work and school hours are the same every day). Our results were still significant, with only 25.0% (10.9%) of the variance in TST SD explained by the difference between weekends and weekdays and 26.7% (11.3%) for SO SD. In other words, substantial variation remains even when considering only weekdays. Not surprisingly, this variation is primarily driven by the variation in the SO. This aligns with previous findings by Lenneis et al. [21], who reported similar mid-sleep scores (the midpoint between SO and waking up) on both free and work days. Even though mid-sleep and sleep schedule variability are not identical concepts, they both indicate the difference (or lack thereof) between sleep on weekdays and weekends. However, they only studied individuals at the end of their adolescence or early adulthood; therefore, those results were not generalizable. Moreover, they found that, although people have different mid-sleep points on work days and free days, they tend to follow a routine of going to bed and waking up on work days and a different routine on free days [21]. In contrast, Roenneberg et al. [22] reported the opposite, noting that the difference between weekend and weekday sleep timing increases during the second decade of life.
In our study, irregular sleep schedules were analyzed in two different ways: dividing the population into those with mostly irregular vs. mostly regular sleep schedules and dividing the nights for the same individuals into regular vs. irregular nights. Interestingly, both methods arrived at the same conclusion: irregular schedules negatively influence sleep architecture and physiology. Irregular sleep can further disrupt behavioral rhythms, such as the timing and amount of eating or exercise, which may increase the CVD risk in individuals with inconsistent sleep patterns by leading to, for example, nocturnal food intake and breakfast skipping.
Finally, we also examined these differences by dividing our population by sex (when reported). Previous studies have found differences in sleep schedules between sexes, with women having more regular schedules [23]. We found that the frequency of irregular sleep schedules was essentially the same in both sexes.
Our findings suggest that irregular sleep is associated with physiological changes, including HR and RR, decreased TST, lower SE, and increased WASO. These effects may result from disrupted circadian rhythms and the body’s response to sleep variability. Elevated HR and RR could reflect heightened autonomic arousal because irregular sleep schedules are thought to activate the sympathetic nervous system, which can affect cardiovascular function. Additionally, variability in sleep timing can lead to a shortened TST and reduced SE, potentially due to difficulties in achieving restorative sleep. This irregularity may disturb the homeostatic sleep drive, making it more challenging to maintain consolidated sleep periods and increasing WASO. These disruptions highlight how irregular sleep patterns may strain physiological systems, potentially increasing vulnerability to cardiovascular and metabolic conditions due to chronic circadian misalignment.
The strengths of this study include the large sample size and number of nights recorded. Although PSG is widely regarded as the gold standard for assessing sleep, using it to study large groups of participants is challenging. To the best of our knowledge, this is the first study to examine sleep-wake schedules using a home-based sleep-monitoring device that allows individuals to remain in their natural environment. This enhances the validity and generalizability of the findings as participants follow their usual routines. The use of these and other portable devices provides opportunities to incorporate sleep consistency goals into future interventions. Moreover, most previous studies have relied on questionnaires or retrospective recall methods, which often fail to capture daily intra-individual variability and are prone to recall bias. Our use of an objective measure for a large number of nights makes our study highly representative of the real-world scenario. However, there is an obvious socioeconomic bias, as we only studied those who could afford the device. This study also has some other limitations. For example, the lack of detailed demographic and lifestyle information and medical history limited our capability to consider confounders. Another limitation is the possibility of selection bias. As the participants were users of a commercially available sleep tracker, this population was likely to have a greater interest in sleep health or experience sleep disturbances, compared with the general population. Although this may result in an overrepresentation of individuals with irregular sleep patterns or other sleep-related issues, it also provides a unique opportunity to study a population that is actively engaged in sleep monitoring. We also had no way of ensuring that the person sleeping in the bed is the participant, as it might be a child, pet, or another individual on any given night; therefore, we might be reporting the sleep schedule of a different individual. However, we believe that the analysis of data from a large number of nights potentially reduced this error.
In summary, these data support the notion that irregular sleep duration and timing are common across all age categories, indicating that sleep irregularity, as demonstrated in this large sample, represents a potentially modifiable behavior with health implications. Although this study did not directly examine cardiovascular outcomes, the high prevalence of irregular sleep patterns underscores the importance of future research to elucidate their potential health impacts, including treatable risk factors for CVD. This analysis is primarily descriptive; however, if sleep irregularity is truly a risk factor, it offers an essential and actionable target for health policy, as it involves modifiable behavior. Therefore, promoting consistent and sufficient sleep may be a novel strategy for maintaining good health. In addition, the ability to measure sleep parameters in a home setting is a valuable tool for public health initiatives.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation 
Supplement1
Print 
