Sleep is essential to a variety of domains of health, including weight management, mood regulation, and longevity [1-3]. National survey research purports that 25% of adults in the United States do not meet the recommended duration of sleep, which is 7 to 9 hours; however, these findings rely on self-reported data [4], which has been shown to vary widely from objectively monitored sleep [5]. The increasingly low-cost technologies embedded in mobile phone and wearable devices provide users with the ability to track their behaviors, such as sleep [6-9]. Sleep tracking with commercially available technology has become popular among the general population in recent years, which presents researchers with an opportunity to analyze the big data captured by these trackers and examine trends in population sleep health.

Behavioral monitoring or the ability to document and reflect on one’s own behavior is an essential part of health behavior according to prominent theories. Specifically, cognitive behavioral therapy [10] and self-regulation theory [11] articulate the importance of behavioral modeling for achieving a desired health outcome. Sleep tracking on a mobile device could be viewed as a form of behavioral monitoring. Although sleep-tracking devices were initially very crude approximations of sleep duration, several of the consumer-facing tracking devices now offer robust visualizations, often including sleep staging and overall sleep performance in the form of a score summarizing such components as sleep quality [12,13]. Indeed, an emerging literature shows the results of studies comparing sleep output from consumer-facing sleep trackers with polysomnography, the gold standard for sleep measurement [14,15].

Sleep tracking via mobile technology is increasingly available to our global population. Research has shown that ownership of smartphones, which feature components that afford the ability to sleep track, such as accelerometers, is reported by approximately half of those living in developing areas of the world and by as many as 90% of those living in developed areas [16]. National data in the United States show approximately one-third of the population report regularly using a smartphone or other devices to track their sleep [17], suggesting sleep tracking is a common practice among individuals in the United States.

Another area of research has examined the design of apps for aiding in sleep disorder symptom identification, such as snoring as a risk factor for sleep apnea [18,19]. Moreover, 1 study analyzed a large volume of sleep tracker data and compared sleep before and after major political events [20]; yet otherwise, no research has explored trends over time in sleep data captured by a sleep tracker app or device.

We undertook an analysis of 4 years of data from a popular wearable sleep tracker to examine patterns of use among users in New York City.

We explored potential seasonal differences in sleep as captured by the sleep tracker. We also examined differences in sleep during the weeknight and weekend nights and contrasted the potential effects of age, sex, and life stage.

We performed historical research using data obtained from Sleep Cycle app. This app is a sleep tracker that uses accelerometer and auditory input from a smartphone device to detect sleep duration and stages. Sleep Cycle is a low-cost app (US $1) available for iPhone and Android operating platforms. This analysis was conducted at the New York University School of Medicine and Kean University. The data used in this analysis were obtained directly from the Sleep Cycle app.

Eligibility criteria for this study included living in a major urban center (New York City) and age 13 years or older. Location was detected using built-in GPS, which allows the phone to detect the users and their location in the form of latitude and longitude coordinates. We requested data for users living within the latitude and longitude that signify New York City and its 5 boroughs for this analysis.

We chose New York City as a location for several reasons. First, we chose 1 geographic location, as this would avoid having to control for varying weather, light, noise, or other environmental factors if we were to compare multiple urban centers. Second, inhabitants of the central region of New York City (Manhattan) include those with a higher average income, whereas the outer regions have inhabitants with a different socioeconomic profile. Therefore, New York City provides a backdrop for examining sleep across different socioeconomic regions in a major urban center.

Data were obtained in an anonymous and aggregated dataset from the developers of the tracker without personal identifiers. Users provided their email address and name to create an account. These sensitive data were removed from the dataset we received and replaced with a random number. According to the Sleep Cycle app’s privacy policy, users consented to provide access to their location while using the tracker. We requested all data on users based in New York City over the past 4 years for this analysis. We analyzed 2,161,067 nights of data, which were provided by 160,963 participants during a 4-year period. The app developers cautioned that a drop in users in 2016 is observed because of a change in the company’s privacy policy.

Users of the sleep tracker place their smartphone device either at their bedside or on their mattress during their sleep, where the app is able to detect motion and sound as inputs. These inputs were scored using proprietary algorithms. Users received detailed statistics and visuals on their sleep from the night before. The app developers maintain that the algorithm to score sleep duration and quality is identical when users track their sleep from either their mattress or bedside.

This analysis was based on sleep duration and sleep quality data recorded by the Sleep Cycle app. Sleep duration was captured objectively and displayed in hours and minutes. Sleep duration was collected for weeknight and weekend night. We scored the sleep duration data using the National Sleep Foundation recommendations, which include recommended, may be appropriate, and not appropriate ranges for different age groups. For instance, recommended duration for teenagers (aged 13-17 years) was 8 to 10 hours, maybe appropriate duration for teenagers was either 7 to 8 hours or 10 to 11 hours, and not appropriate duration for teenagers was either less than 7 hours or more than 11 hours.

This analysis used sleep quality as derived by the Sleep Cycle app. The app reports sleep quality as a score of the overall efficiency of the sleep using an algorithm based on time in bed, deep sleep time, sleep consistency, and amount of times fully awake. Sleep quality scores range from low (0) to high (100). Drawing on the sleep efficiency index, we conceptualized sleep quality scores of 85 and above as good quality and those below 85 as poor sleep quality [21].

Demographic factors are obtained for each user and self-reported in their profile. These data were obtained along with sleep duration and quality for each user. Demographic variables included age and sex. We eliminated individuals aged younger than 13 years and older than 85 years. In addition, each night of sleep included a date and time stamp. Consequently, we created variables to distinguish between sleep during summer, winter, spring, and autumn.

We computed descriptive statistics for sleep duration by independent variables (weeknight vs weekend night, age, sex, and seasonality). Similarly, we computed descriptive statistics for sleep quality by weeknight versus weekend night, age, sex, and seasonality. We used logistic regression to examine differences in sleep duration and quality by independent variables. We created graphical matrix representations of data to graphically display the differences in sleep duration and quality based on weeknight versus weekend night, age, sex, and seasonality. We compared sleep duration and quality by weeknight (vs weekend night), age, sex, and seasonality with analysis of variance or Chi-square and reported the P value to indicate significant differences in sleep parameters.

Finally, we created a heat map, consistent with previous research [22], whereby red indicated greater concentration of users who demonstrated not recommended sleep duration on a day chosen at random from the 4 years of data. Per the National Sleep Foundation recommendations, this cutoff varied by age group. Specifically, users were marked in red if their sleep for the night chosen at random did not meet the criteria for recommended in their respective age group. We matched these individuals, using the latitude and longitude data obtained from the app, to a geographical map of New York City to visually examine the concentration of not recommended sleep duration in New York City. All analyses were performed in R software.

Sleep tracking may confer benefits such as increasing motivation for healthy changes to sleep routines among users of sleep trackers [6-9]. Furthermore, sleep tracking is reported using a smartphone or other devices by approximately one-third of the population in the United States [17]. Although researchers articulate the need for validation of commercially available apps and sleep trackers [15,23], we emphasize the need to understand the nature of the data that users of popular sleep trackers receive but also the opportunity to identify a signal in population-level sleep health vis-à-vis the data provided from commercially available sleep trackers. In this study, we analyze big data from users of a popular sleep tracker to illuminate trends in sleep on this platform over 4 years and over 2 million nights.

First, our results showed that women, according to the sleep-tracking app, experience longer sleep duration and higher sleep quality in nearly every age group than men. This finding is somewhat perplexing, as previous research has shown women are sleepier than men (which might indicate poorer sleep quality) and take longer to fall asleep, and older women sleep less (approximately 20 min less) than men [24]. Some previous studies point to the increased household responsibilities many women shoulder disproportionately compared with men, which would systematically detract from sleep time, as a reason for shorter and lower quality sleep among women compared with men [25]. However, our results showed the contrary that women experienced longer sleep duration and higher quality sleep as collected by the app than their male counterparts. We postulate the perplexing finding that women obtain longer sleep and higher sleep quality because of several factors, but notably, there is a chance that the sleep tracker may be poorly suited for detecting sleep among women. For instance, women may have a lighter frame, and therefore, the tracker is less able to detect their movements. It is also possible that cosleeping couples may confound the tracker and its ability to efficiently capture sleep among both individuals.

Second, our study showed a low proportion of teenagers (25,491/97,156; 26.23% of individuals aged 13-17 years) obtained recommended sleep duration. In addition to documenting insufficient sleep among teenagers, our research also shows the prevalent inconsistent sleep schedules maintained by many teenagers. These findings are consistent with previous research that has demonstrated clear reductions in sleep time during the teenage years [26-28]. Research shows the significant barriers teenagers face to sleep health, including social pressures and academic responsibilities, which place teenagers at a significant disadvantage when it comes to their ability to obtain sufficient sleep. Furthermore, this research also highlights structural barriers, such as early school start times, that compete with teenager physiological preference for later bedtimes and result in teenagers extending sleep on the weekend nights, which can introduce social jetlag and circadian desynchrony hindering sleep and overall health [29]. Our research further emboldens the need for public health and policy efforts to address to the issue of poor sleep health among teenagers.

Third, our research examined seasonal patterns in sleep duration between summer, winter, spring, and autumn. Interestingly, there were unique patterns of sleep by season for each age group and sex, with the exception. On the one hand, among young adults and adults, both sexes slept longest during winter. Teenagers, on the other hand, perhaps because of summer vacation and being free of academic and school responsibilities, slept most during the summer months. Interestingly, older adult males had almost entirely different patterns of sleep duration compared with female older adults.

Regarding sleep duration and quality over the 4 years of analysis in New York City, we found sleep duration and quality were lowest around 2015 and early 2016, rising steadily to 2018, at which time there appears to be a plateau. One could likely look to national and international events at these times to explain the slight decline in 2016. For instance, the advent of the gig economy and trend toward project-based work may have led some individuals to shorten sleep or sleep less because of increased occupational burden and less stability that is common among these occupational categories [30]. Overall, sleep duration and quality are higher across the age groups on weekend nights than on weeknights. This is also consistent with previous literature, showing that individuals tend to sleep longer on the weekend nights [31]. Our research similarly found overall longer sleep on weekend nights (compared with weeknights).

Our results provide a comprehensive assessment of 4 years of data and over 2 million nights of sleep. We identified trends in seasonality, age, and sex. In so doing, we provide a compelling case for the issues regarding insufficient sleep among teenage populations. Our geospatial analyses revealed a higher concentration of users receiving not recommended sleep duration in outer boroughs of New York City, suggesting sleep that is recommended duration may be more common among higher income inhabitants of central New York City.

Although the strengths of our analysis included big data used to perform the analyses of sleep data from a large population of users over several years, limitations must be noted. First, we analyzed 1 major metropolitan area. Results will differ in different geographical regions. It must also be noted that although ownership of smartphones that allow this type of tracking is high among developed countries, fewer than half of certain populations report access to smartphones. Consequently, our results regarding sleep duration and quality may be disproportionately represented by high-income populations with access to these technologies. We note that the dataset provided included limited details on the study sample. We received only age, sex, sleep duration, and quality. Future research may aim to include additional variables, such as race and ethnicity or health conditions, for a greater understanding of sleep between a richer array of demographic and health variables.

Our results were performed on the summary scores for sleep duration and quality as reported by the developers of the Sleep Cycle app. The authors of this study did not have access to the algorithm used by the developers to detect, measure, or analyze sleep. Furthermore, there have not been any published studies validating the methodology used by the developers of the Sleep Cycle app with established measures of sleep assessment, such as either wrist-based actigraphy or polysomnography. Previous researchers have emphasized the need for validation of sleep tracking devices such as Sleep Cycle [14,15]. Therefore, it is possible that sleep duration is scored using similar metrics to those used for measuring sleep quality. Therefore, there is a possibility that sleep duration and quality data are correlated with one another. Finally, there is a possibility that different types of cell phone hardware and accelerometer technology may produce differences in sleep as detected by the app.

Our study identified several compelling avenues for future research. First, interventions could be designed to target the specific barriers (eg, insufficient sleep or poor-quality sleep) as reported by users of apps such as that studied in this paper. In addition, researchers and developers of sleep-tracking technology could collaborate on the sleep duration and quality algorithms to ensure concordance with the gold standard in sleep research (ie, sleep diary, actigraphy, or polysomnography). Finally, future research may also examine the effect of exposure to sleep-tracking apps. For instance, although we know sleep tracking is increasingly common, we know less about the effects of its exposure, the duration of adherence to the devices, and how helpful (or perhaps harmful) output regarding sleep may be for individuals, particularly those suffering from disorders, such as insomnia or sleep apnea. Among these patient populations, sleep tracking may be ill advised as a source of worry about sleep.

We examined sleep data from more than 2 million nights of data captured during a 4-year period among users of a popular sleep tracker in New York City. Our findings show women slept longer and demonstrated higher sleep quality than men, and that teenagers demonstrated abysmal rates of adherence to sleep duration recommendations. We also showed that sleep quality and duration vary seasonally. We also demonstrated visually that insufficient sleep is observed in greater prevalence among users located in outer boroughs of New York City. Future research may consider the role of sleep tracking for improving motivation to adhere to recommended sleep routines, such as consistent bedtime schedules. Future research may also examine the role of sleep tracking and health profiles of users over time.