Has Adult Sleep Duration Declined Over the Last 50+ Years?

Summary

Introduction

It has been widely stated that modern industrialized societies have become sleep-deprived. Some studies have suggested that average sleep duration has declined over the last few decades [1–4]. Such findings, combined with extensive epidemiologic evidence associating short sleep with health risks [5–7] and experimental evidence of adverse effects of sleep deprivation [8–10], have provoked widespread concern that chronic insufficient sleep has become a public health crisis.

However, recent reviews of self-reported data have cast doubt on whether nighttime sleep or 24-h sleep has decreased in recent decades, and whether there has been an increased prevalence of short sleep (<6 h), for which risks have been most clearly established. For example, a review of eight studies by Knutson et al. found no significant 31-year trend (1975–2006) towards a higher prevalence of self-reported nighttime sleep of ≤ 6 h [11]. Bin et al. reviewed 12 studies from 15 countries assessed from the 1960s–2000s, and found that sleep duration had increased in 7 countries, decreased in 6 countries, and had not clearly changed in 2 countries [12]. In a subsequent meta-analysis of 38 studies conducted in 10 countries in the 1970s–2000s, Bin et al. [13] found that average 24-h sleep duration had increased in most countries (including the US), and that the prevalence of sleeping ≤ 6 h had decreased in most countries (including the US). Rowshan Ravan et al. studied 36-year trends (1968–2004) in sleep duration among Swedish women, and found no change in 50-year old women, and a decline of only 15 minutes in 38-year old women [14]. Moreover, Bonke reviewed five representative time-use studies spanning 1964–2009, and concluded that “the same number of hours is slept today as in the mid-1960s, with nearly the same prevalence of short and long sleepers” [15].

Discrepancies between studies of population temporal trends in sleep duration can be attributed to multiple factors, including characteristics and representativeness of the respondents, wording of the questions, and instructions given to respondents [16, 17]. Perhaps the biggest limitation of this literature is that it has been limited to self-reports of sleep duration (some of which were retrospective), which can be inaccurate [18, 19] due in part to response biases. The aim of this review was to examine whether there has been a decline over the past 5 decades in sleep duration, as indexed by objective data.

Methods

The search of the literature was modeled after a previous meta-analysis by Ohayon et al., which assessed objective sleep patterns across age [20]. PubMed, PsychLit, selected journals, and reference lists of located manuscripts were searched for studies published between 1960–2013 which met the following criteria: 1) inclusion of presumably healthy adults (as described by the authors), participant ages ≥18 y without sleep problems; 2) report of all-night average total sleep time (TST) measured by polysomnography (PSG) or actigraphy; 3) assessment of sleep under minimally-disturbed conditions, including baseline or placebo conditions, and not involving particularly invasive procedures (e.g., catheterization). Many of the studies included a control group of presumably normal sleepers who had been compared with participants with sleep disorders. Studies involving individuals with extremely high levels of physical fitness were excluded under the assumption that sleep of such individuals might not be representative of the population. Key search words were sleep with normal, normative, healthy, controls, and adults.

The literature searches were performed by two of the authors: either EEG or NK. Questions regarding whether a study met inclusion criteria were resolved in discussions between EEG and SDY or AMR and SDY. Data from the studies were extracted by EEG and AMR.

The search identified >3,500 studies, of which 168 met the inclusion criteria, generating 257 data points across 6,052 individuals. Studies were separated into PSG (Table 1) and actigraphic studies (Table 2). Citations for all included studies are listed in the reference list (#55–222). Coding for each study included the mean sample age (or mid-point of the age range if the mean age was not available), number of men and women subjects, mean sample total sleep time (min), and estimated year of study. Studies with multiple age groups generated multiple data points for the analyses. When available, separate data points for men and women were used. Since most of the studies recorded sleep in the laboratory, only the laboratory data were used for studies that included both home and laboratory data, except for separate analysis of the actigraphy data.

Since the year of publication of a study often differed from the year in which a study was conducted, the following rules were used to estimate the year that a study had been conducted. 1) Year of study was estimated by subtracting 10 months from the posted date of journal receipt of the manuscript for studies with <50 subjects, 14 months for studies with 50–99 subjects, 18 months for studies with 100–149 subjects, and 22 months for studies with ≥150 subjects. 2) If information was available regarding the date a paper was accepted, but not the date that it was received, the median across-the-literature duration in months between date received and date accepted (4 months) was subtracted from the date of publication, and Rule 1 was followed. 3) If neither date accepted nor date received information was available, the median number of months between date received and date published (11 months) was subtracted from the date of publication, and Rule 1 was followed.

The TST data were first assessed by comparing the intercepts of the regression lines of age vs. TST for studies conducted between 1960–1989 vs. 1990–2013. We chose this split to obtain a more balanced number of data points across the years split. Another reason for the1989/1990 split was that it has been posited that the obesity epidemic, which started shortly after this time, can be partly attributed to declines in sleep. Examining the intercepts allowed an assessment of temporal differences in TST across all data points (without adjustment for age). A temporal decline in TST would be revealed by a smaller intercept for the 1990–2013 studies compared with the 1960–1989 studies. Another rationale for the1989/1990 split was that it has been posited that the obesity epidemic, which started shortly after this time, can be partly attributed to declines in sleep [21].

To further assess a temporal trend of TST across all data points, a linear regression analysis of year of study (weighted for sample size) and participants’ age vs. TST was calculated. To plot these data, age-adjusted TST was determined based on the slope of the linear regression between TST and age across all data points. An a priori decision was made to remove outlying samples, for which mean age-adjusted TST was ≥ 2 standard deviations from the mean value across the literature. Two data points were removed based on this criterion. Weighted linear regression analyses were also conducted for year of study vs. TST across 10-year age categories (e.g, ages 18–27 years, 28–37 years, etc.).

Separate weighted linear regression analyses were conducted for data from studies in which participants followed their usual sleep schedules and for studies involving a fixed sleep period; for polysomnographic and actigraphic data; and for data involving men only and women only. Plots of year of study vs. age-adjusted TST were performed for each of these analyses.

Results

The intercepts and slopes of the regression lines of age vs. TST did not differ for studies conducted between 1960–1989 and 1990–2013 (Figure 1). In the regression analysis across all data points (n=257), there was no significant association of year of study with TST (b=.03, p=0.56) (Figure 2), nor was there a significant association of study year with TST for any of the 10-year age categories (Figure 3) (p=0.40–0.92). Likewise, there was no significant association of year of study in analyses restricted to PSG (n=225) (b=0.03, p=0.63) or to actigraphic data (n=32) (b=−0.17, p=0.38) (Figure 4); or in analyses involving only men (n=71) or only women (n=17) (Figure 5). Finally, there was no significant association in analyses derived from studies in which subjects followed their usual sleep periods (n=154) (b=0.13, p=0.10) or a fixed sleep period (n=68) (b=−0.14, p=0.24) (Figure 6).

Open in a separate window

Figure 1

Association of mean age of participants with total sleep time (min) for studies conducted between 1960–1989 (open circles) and 1990–2013 (closed circles).

Open in a separate window

Figure 2

Association of year of study with age-adjusted total sleep time (min) for all data points. The regression line and 95% confidence intervals are displayed.

Open in a separate window

Figure 3

Association of year of study with total sleep time (with regression line and 95% confidence intervals) for participants ages 18–27 years (a), 28–37 years (b), 38–47 years (c), 48–57 years (d), 58–67 years (e), 68–77 years (f), and ≥78 years (g).

Open in a separate window

Figure 4

Association of year of study with age-adjusted total sleep time (min) for polysomnographic data (a) and actigraphic data (b). The regression line and 95% confidence intervals are displayed.

Open in a separate window

Figure 5

Association of year of study with age-adjusted total sleep time (min) for women subjects only. The regression line and 95% confidence intervals are displayed.

Open in a separate window

Figure 6

Association of year of study with age-adjusted total sleep time (min) for studies in which subjects followed their usual sleep schedule (a), and for studies in which subjects followed a fixed sleep schedule of 470–480 min (b). The regression line and 95% confidence intervals are displayed.

Discussion

The results indicate relative stability of objectively-recorded sleep durations in healthy sleepers assessed over the last half-century. Similar results were found across all age groups; in both men and women; for both PSG and actigraphic data; and under conditions of fixed sleep periods and participants’ usual sleep schedules. These data are consistent with recent comprehensive reviews that found no consistent or compelling evidence of significant decrements in self-reported sleep duration and/or prevalence of short sleep over a similar range of years [11–15]. Together, these data cast doubt on the notion of a modern epidemic of insufficient sleep.

There were several limitations of the literature, which might have confounded demonstration of temporal changes in sleep duration. First, although virtually all of the studies failed to describe the racial/ethnic composition of the samples, it is a reasonable assumption that participants in most of these studies were not representative of the population. Recent research has suggested that the prevalence of short sleep is relatively high among Blacks, and that this prevalence might be increasing more among Blacks than among Whites [22]. Furthermore, most of the studies either excluded women or failed to report separate data for women and men. Thus, there was an insufficient number of data points (n=17) to adequately assess whether there was a temporal decline in women’s sleep duration, which might have occurred as more women have entered the workforce over the past 50 years [11, 15]. Study samples have also likely been unrepresentative of the population in other factors which have been associated with sleep duration, including employment status, education, occupation, and socioeconomic status.

A second limitation is that most of the studies assessed sleep with PSG in the laboratory, a process that can result in curtailed sleep duration. The confound was reduced in most of the PSG studies by disregarding data obtained during the first night of laboratory recording (eliminating “first night effects”) [23]. Interestingly, in a post-hoc assessment of studies that measured sleep objectively both at home and in the laboratory, the median difference between home and laboratory TST was only 3.2 min (Table 3). However, the use of PSG recording could have inhibited sleep, and sleep might have been more disrupted in earlier PSG studies due to greater novelty associated with PSG, as well as less technologically advanced methods, such as the use of collodion for securing electrodes.

Constraints of PSG recording might not capture a decline in nighttime sleep that has occurred at home when people are more able to follow their customary habits, which might involve staying up later. Roenneberg et al.’s surveys of thousands of adults assessed from 2002–2010 have found a decline of approximately 30 min in reported sleep duration on weekdays [24]. However, the present review did not find a similar change in home actigraphic sleep duration over the past 10–20 years. Likewise, a recent study by Gubelmann et al. found no decline in reported time in bed from 2005–2011 among a large Swiss sample (n=3,853) [25].

A third limitation is that studies with fixed sleep periods (usually 8 h) could have resulted in sleep restriction for some individuals, particularly if the timing of the sleep periods was not consistent with the participants’ usual sleep schedule. This restriction could have been generally greater in earlier studies if sleep duration truly had declined. However, a similar age-adjusted mean TST was observed for studies involving fixed (443.3±31.7 min) and habitual sleep schedules (435.1±37.4 min), and there was a similar absence of a significant secular trend in TST for fixed and habitual sleep schedules (Figure 6). Figure 6A might reflect a societally-imposed or custom-imposed 8-hr ceiling in how long people usually spend sleeping. It is also possible that PSG technicians have been reluctant to extend the night shift beyond 8 h.

A fourth limitation is that compared with more recent studies, it is possible that earlier studies did not screen as well for absence of sleep apnea and other sleep disorders; this difference in screening methods might have resulted in lower estimates of sleep duration. However among adults above middle age, a small amount of sleep apnea or periodic limb movements is so common that it might be considered normal. Relatively more drug studies in recent years could have contributed to more extensive participant screening of normal sleepers, resulting in samples that sleep longer than population norms. However, a similar absence of a decline in sleep duration was found in the 18–27 year old adults, for whom the prevalence of sleep apnea and other health problems is relatively low. Also contrary to the hypothesis that more recent studies have had more homogenous samples of good sleepers, a post-hoc analysis showed no significant correlation between year of study and sample standard deviation of TST (r=−0.01).

A fifth limitation is that mean nighttime sleep duration data for a sample might not reflect temporal changes in the prevalence of short or long sleep, nor changes in 24-hr sleep duration which might have occurred. Interestingly, Figures 1–3 suggest a higher prevalence of sleep of ≤6 h over the last 20 years, particularly among 18–27 year old participants.

In recent decades, the siesta tradition has waned considerably in some countries [26]. Without corresponding increases in nighttime sleep, this could have resulted in a temporal decline in 24-h sleep in these countries. Partial support for this hypothesis was provided by Bin et al., who found in a meta-analysis that 24-h sleep duration decreased by 22 min from 1989–2002 in Italy [13], whereas there was not a decline in 24-h sleep in 8 of the other 9 countries assessed, none of which has had a notable siesta tradition (Australia, Canada, Finland, Germany, Netherlands, Norway, Sweden, United Kingdom, United States).

However, there has been limited empirical investigation of temporal trends in napping. Wolf-Meyer traces a historical decline in napping to the industrial revolution, increased structure of the work day, and the origins of sleep medicine which has promoted a theoretical need for 8 hours of sleep at night [27, 28]. Thus, through much of the 20^th century, napping in many industrialized countries was regarded as a sign of laziness [28]. However, attitudes and practices of napping have apparently changed over the past 10–20 years, as evidenced by formal sanctioning of work-day napping and commercial napping services in some cities.

Napping is relatively more common among older adults who have less nighttime sleep and less consolidation of the sleep-wake cycle than young adults. Compared with previous older cohorts, some factors could have resulted in less napping in contemporary seniors, such as later retirement age, more physically and socially active lifestyles, and greater rates of residence in senior living facilities.

Nonetheless, the present review is the first to explore historical patterns of objective sleep duration, which has long been regarded as the gold standard for defining sleep duration [18]. Further, the findings have several implications. Although historically 8 h of sleep was thought to be optimal for health and well-being, an extensive epidemiologic literature has indicated that 7 h of self-reported sleep is associated with the lowest health risks [29], with progressively higher risks associated with shorter as well as longer reported sleep. However, since objectively-recorded sleep duration is generally 30–60 min less than self-reported sleep, optimal objective sleep duration for longevity and health might be only 6–6.5 h. For example, Kripke et al. recently found 5–6.5 h of actigraphic sleep was associated with lower mortality than <5 h and > 6.5 h [30]. The present review adds to recent reviews of self-reported data, which have also indicated no decline in sleep duration over the last 50 years. If the optimal duration of objective sleep is indeed between 6–6.5 hours, the review also suggests that more participants in these studies might be at risk due to long sleep than to short sleep.

Had sleep duration truly declined by 1–2 hours over the last 50 years, as many sleep researchers have claimed, the signal to detect this would be at least as great as that associated with age, which shows only a decline of about 1 h from young adulthood to the elderly (Figure 1). The results also contradict the hypothesis that such a decline in sleep is a probable culprit in modern epidemics of obesity and diabetes [21].

Notwithstanding these findings, assumptions about a steady decline in sleep duration over the past few decades persist, and could be explained by many factors. First, increased public awareness about sleep and the dangers of inadequate sleep, coinciding with an exponential increase in sleep disorders diagnoses with the emergence of sleep medicine [31], could have partly shaped these perceptions. Greater knowledge about sleep, perhaps especially a greater ability to distinguish between sleep and time spent in bed, could lead to perceptions of less sleep.

Second, sleep is commonly considered in the context of leisure time and being a respite from daily stressors [32]. In what seems to many to be an increasingly fast-paced and stressful world, there is a perception of having less free time for “rest.” Third, evidence indicates that the prevalence of depression has increased over time [33], and depression is associated with reports of poor or inadequate sleep [34].

Fourth, self-reported behavior is influenced by perceived social norms [26, 35], and the perception that we have become a sleep-deprived society has likely been shaped partly by promotion of this message in the popular media and by sleep scientists. However, much of the narrative regarding an epidemic of declining sleep has been based on arguments which have not been well-supported by empirical data. We address some of these arguments in the following section, although much of this discussion is also not well-supported by empirical observations.

Acknowledgments

List of Abbreviations

Footnotes

References