• Open Access

Day type and the relationship between weight status and sleep duration in children and adolescents

Authors


Correspondence to:
Carol A. Maher, School of Health Sciences, Division of Health Sciences, University of South Australia, GPO Box 2471, Adelaide SA 5001. Fax: (08) 8302 6558; e-mail: carol.maher@unisa.edu.au

Abstract

Objective: This study aimed to explore sleep duration in young Australians on different types of days across weight classes.

Methods: Use of time and anthropometric data were collected on 8,866 nights from 3,884 9–18 year old Australians. The association between sleep duration and weight status was examined using factorial ANOVA for four day types: S-S (to bed and waking on school days); S-NS (to bed on school day and waking on non-school day); NS-NS (to bed and waking on non-school days); NS-S (to bed on non-school day and waking on school day).

Results: Sleep duration varied with weight status when all day types were considered together (p=0.0012). Obese adolescents slept less than normal and underweight adolescents. However, the relationship varied for different day types; with the strongest relationship for NS-S days (on which obese children slept 65 min less than very underweight children, p<0.0001).

Conclusions: The association between weight status and sleep duration showed consistent gradients across weight categories, but only for certain day types.

Implications: These patterns cast light on the direction of causation in the obesity-sleep duration relationship. Findings suggest that short sleep duration contributes to obesity, or that a third unidentified factor has an impact on both.

Chronic reduced quality and duration of sleep have been associated with a range of physical and psychosocial disturbances in both adults and children, including: impaired attention, memory, creativity, learning and academic performance;1 motor skill deficits;2 greater emotional lability, increased impulsivity, aggression and hyperactivity;3 and increased potential for alcohol and drug abuse in adulthood.4 Studies in Japan and the UK have found that poorer quality sleep is associated with poorer overall health in adults, perhaps mediated by compromised immune function, and is more likely among those of lower socio-economic position (SEP).5 In particular, there is a growing body of evidence that short sleep duration is associated with overweight in young people and adults. Sleep duration has been found to be inversely associated with the risk of childhood overweight and obesity in 5–6 year olds from France6 and Germany,7 3–6 year old Japanese children,8 10–15 year old Australian adolescents,9 and 9 and 11–14 year old US children.10,11 Additionally, Landhuis and colleagues showed in their longitudinal study that sleep restrictions in childhood increases the longterm risk for obesity in adulthood. This relationship remained after adjustment for adult sleep time and other potential confounders.12

The mechanisms for this relationship are unclear. It is possible that overweight causes poor sleep, through dysfunctions such as obstructive sleep apnea.13 It is also possible that both overweight and poor sleep are caused by some third factor. Agras et al.10 for example, found that adolescents who slept poorly were also reported to be less active. The authors speculated that low activity levels might lead both to a positive energy balance and to reduced tiredness, and hence poor sleep. However, a recent analysis of 10–15 year old Australians revealed that the association of school day sleep duration and weight status was independent of self-reported physical activity and dietary behaviours.14 Most studies have found that shorter sleep duration is due to later bedtimes rather than earlier wake up times, suggesting that late night activities such as television viewing or computer and video game use may be implicated.15

However, evidence is growing that poor sleep itself may be a causal factor in overweight. Chronic sleep debt can affect health through disruption of endocrine and metabolic functions.11 In adults where sleep duration was experimentally manipulated, short sleep duration induced disruption of glucose homeostasis, increased sympatho-vagal drive, higher levels of thyroid-stimulating hormone and cortisol, and reduction in leptin concentrations.16 Increases in ghrelin secretion have also been reported.17 These hormonal disturbances can effect weight status in a number of ways. Lower levels of leptin and higher levels of ghrelin are associated with increased hunger,17 and preference for high energy-density foods.18 Chronic increases in blood glucose concentrations can lead to insulin resistance and diabetes. Increased levels of cortisol, which have been associated with acute sleep loss, are also associated with accumulation of abdominal fat.19

There is evidence that in high-income countries sleep duration has been decreasing. Iglowstein et al.20 found declines of 20–40 minutes in night time sleep in Swiss children aged between six months and 14 years between 1974–78 and 1986–93. Similarly, Dollman and colleagues21 reported declines of about 30 minutes in the time in bed on school days of Australian children and adolescents aged 10–15 between 1985 and 2004. These declines in sleep duration were mainly due to later bedtimes, driven perhaps by academic demands, ‘always on’ electronic media such as television, the internet and mobile phones and less parental regulation. The Sleep in America survey noted that children with a television in their bedroom slept about 40 minutes less each day than those without a television in their room.22

None of the pediatric studies has examined the relationship between sleep duration and weight status across the full range of weight status classes, from underweight to obese. Rather they have compared obese (or obese and overweight) children and “normal” weight (i.e. non-obese or non-overweight) children, without distinguishing between non-overweight and underweight children.

In addition to secular declines, there is accumulating evidence of chronic sleep deprivation in some children and adolescents on school days, coupled with “catch-up” sleeping on weekends and holidays.23 This indicates that they are recovering their sleep debt on weekends as they are not receiving adequate sleep for their age.24 Differences in sleep duration on weekdays and weekend days of 1–3 hours have been reported for Korean,25 American26 and Belgian27 adolescents. In spite of radically different sleeping patterns on school and non-school days, no studies have examined separately the relationship between weight status and sleep on different types of day. Such information would assist in the development of higher resolution sleep hygiene guidelines for use by health professionals and parents.

The aim of this study was to explore the relationship between weight status and sleep duration in 9 to 18 year old Australian children and adolescents, analysing different types of days separately, and considering underweight as well as overweight and obese participants.

Methods

Participants

The participants for this study were 4,053 Australians aged between 9 and 18 years, who were interviewed as part of a series of surveys conducted between 2001 and 2007. Three main clusters of surveys were involved:

  • 1) The Australian National Children's Nutrition and Physical Activity Survey (n=2,350), a household-based, random national survey of 9–16 year old children and adolescents conducted between March and August 2007. The response rate was 41%.
  • 2) The Health of Young Victorians Survey (HOYVS; n=878), the third wave of a school-based cohort study involving young people aged 13 to 19. Details of this survey have been published elsewhere.28
  • 3) A series of school-based South Australian studies of children and adolescents aged 10 to 15 using a variety of sampling frames (n=656). In most of these studies, schools were randomly selected from a list of all schools in the state and all children from a particular age group were invited to participate. In these surveys, the average response rate was 69% for schools, and 92% for children within the chosen age group within each school. In a small number of cases, data were collected from students at individual schools at the invitation of the schools themselves, or as part of pilot projects. Ethical approvals were obtained for each of the studies from which data were drawn, and all participants' parents provided written informed consent.

Participant characteristics across all the surveys are shown in Table 1. The proportion of participants in each weight class closely matched 2001–2008 Australian estimates (underweight 5.0%, normal weight 74.5%, overweight 15.1% and obese 5.3%, calculated on the basis of 17,266 children in 14 studies29).

Table 1.  Participant characteristics.
  VUWUWNWOWOBAll
  1. VUW = very underweight   UW = underweight   NW = normal weight   OW = overweight   OB = obese

  2. NB. Values are shown as percentages, or means (standard deviations). SEIFA (Socio-Economic Indicators For Areas) is a method used by the Australian Bureau of Statistics to characterise socio-economic status. It uses a basket of indicators such as education and income, and is calculated at the postcode level. The national mean is 1000, and the standard deviation is 100.

  3. S-S = days when adolescents went to bed on a school day and woke on a school day

  4. S-NS = days when adolescents went to bed on a school day and woke on a non-school day

  5. NS-NS = days when adolescents went to bed on a non-school day and woke on a non-school day

  6. NS-S = days when adolescents went to bed on a non-school day and woke on a school day

Boys n (%) 27 (1.4%)119 (6.2%)1,376 (72.2%)287 (15.1%)96 (5.0%)1905
Girls n (%) 25 (1.3%)99 (5.0%)1,333 (67.4%)392 (19.8%)130 (6.6%)1979
Mean age in years (SD) 13.8 (1.7)14.0 (2.0)14.1 (2.2)14.0 (2.2)13.9 (2.2)14.0 (2.2)
Total number of data pointsAll days125536619115184968866
 S-S3918219514841522808
 S-NS1769844203721205
 NS-NS4719025055971903529
 NS-S2295891234821324
Mean SEIFA (SD) 1,025 (42)1,021 (69)1,023 (65)1,013 (71)990 (77)1,016 (62)

Data collection

Height and body mass were measured by trained interviewers according to the protocols of the International Society for the Advancement of Kinanthropometry.30 Participants aged 9–18 years completed at least one 24 hour time use recall, using either face-to-face interviews, telephone interviews, self-completion under guidance in school computer laboratories or in a small number of cases self-completion at home using a CD.

All participants provided use of time data using the Multimedia Activity Recall for Children and Adolescents (MARCA).31 The software allows young people to recall everything they did on the previous day from wake-up to bedtime. It uses a segmented day format with self-determined anchor points, including wake up time and bedtime. Sleep duration was calculated from the bed time and wake-up times reported when participants provided use-of-time data for sequential days. Bed time was defined as the time when adolescent turned out the light and went to sleep and wake-up time as the time they woke up (not necessarily the time they got out of bed). “Lying awake in bed” was an activity option and was not included in sleep time. Intra-day naps were included in sleep time. Young people could report activities in time-slices as fine as five minutes. The MARCA has a same-day test-retest reliability of ICC=0.84–0.92 for major outcome variables (ICC=0.87 for sleep time), and criterion validity with reference to accelerometry of rho=0.45 for physical activity level.31

Sleep data treatment

Sleep data were adjusted for age and sex by regressing sleep time against age for boys and girls separately, fitting a fourth-order polynomial and retaining the residuals in the analysis. In the presentation of results, the residuals have been retransformed to actual sleep times for ease of understanding. Children were classified as either meeting or not meeting the guidelines of the United States' National Sleep Foundation32 for the recommended amount of sleep, viz. 10–11 hours per night for 5–12 year olds, and ≥ 9 hours per night for >12 year olds.

Anthropometric data treatment

Body mass index (BMI) scores were categorised into five classes according to the guidelines of Cole et al 2000,33 and Cole et al. 2007:34 1) obese (equivalent to an adult BMI of ≥ 30), 2) overweight (equivalent to an adult BMI of 25.0–30.0), 3) normal weight (equivalent to an adult BMI of 18.5–25.0), 4) Grade 3 thinness (equivalent to an adult BMI of 17.0–18.5), and 5) Grade 2 thinness (equivalent to an adult BMI of 16.0–17.0). No participants fell into the Grade 1 thinness category (adult BMI < 16.0). BMI values were converted to z-scores using the UK 1990 reference standards.35

Statistical analysis

One-factor factorial ANOVA was used to assess univariate relationships between weight status category and both overall sleep duration and bedtime on all days considered together, and on the four separate day types. Fisher's PLSD t-test was used in post-hoc analysis. Linear regression was used to describe the relationship between sleep duration and BMI z-scores for all days combined, and for each of the day types separately. Chi-square was used to compare the frequencies with which children from different weight classes met the National Sleep Foundation guidelines on all days combined, and on each day type separately. Alpha was set at 0.05 for all analyses.

Results

Sleep patterns

Sleep duration varied significantly (p<0.0001) between all day types. The shortest sleep (mean = 555 minutes) occurred on NS-S days (such as Sundays), and the longest (601 minutes) on NS-NS days, such as Saturdays and holidays. Girls slept on average seven minutes per night more than boys, the differences being greatest (10–11 minutes) on S-NS and NS-NS nights, mainly because girls went to bed earlier. Overall, participants failed to achieve the minimum recommended amount of sleep on 32% of nights. This figure was significantly higher (43%) on NS-S days.

Sleep duration and weight status categories

Sleep duration was significantly related to weight status category across all days taken together (p=0.0012), with sleep duration decreasing linearly across weight status categories from 591 (very underweight) to 571 minutes per night (obese) (Table 2). Results from post-hoc analyses are indicated in Table 2 with superscript; within each row, values with the same superscript were significantly different. For example, when all day types were considered together, obese participants slept significantly less than normal weight, underweight and very underweight participants. In addition, overweight participants slept significantly less than normal weight and underweight participants. Sleep duration was also significantly related to weight status on S-S days (p=0.05), NS-NS days (p=0.03) and NS-S days (p<0.0001), but not on S-NS days (p=0.29). On NS-S days (e.g. Sundays), obese participants slept 65 minutes less than very underweight participants. Figure 1 shows the relationship between weight status and sleep duration on all days, and on S-S, S-NS, NS-NS and NS-S days.

Table 2.  Mean sleep durations (minutes per day, adjusted for age and sex) across weight status categories on all days, and on each of the four day types.
Day typeVUWUWNWOWOBP
  1. VUW = very underweight   UW = underweight   NW = normal weight   OW = overweight   OB = obese

  2. S-S = days when adolescents went to bed on a school day and woke on a school day

  3. S-NS = days when adolescents went to bed on a school day and woke on a non-school day

  4. NS-NS = days when adolescents went to bed on a non-school day and woke on a non-school day

  5. NS-S = days when adolescents went to bed on a non-school day and woke on a school day

All days591c589be583ad577de571abc0.0012
S-S574571abc561c558b556a0.05
S-NS6126155845956100.29
NS-NS594607b606a597585ab0.03
NS-S596cd567a555bd549ac531abcd<0.0001
Figure 1.

The relationship between weight status category and sleep duration on various types of days.

Sleep duration and BMI z-scores

When weight status was construed as a continuous variable (expressed as a z-score relative to the UK 1990 standards), the relationship between weight status and sleep duration adjusted for age and sex was also significant (p<0.0001). For every unit of BMI z-score, sleep declined by 3.2 minutes. The relationship was much stronger on NS-S days (p<0.0001, 7.5 minutes per unit of BMI-score) than on S-S (p=0.06, 2.8 minutes), S-NS (p=0.92, 0.2 minutes), or NS-NS (p=0.04, 2.9 minutes) days.

Sleep guidelines and weight status categories

Overall, participants failed to meet the National Sleep Foundation guidelines on 32% of days. The proportion of nights failing to meet guidelines increased as weight status increased from very underweight to obese on all day types taken together (p<0.0001), NS-NS days (p=0.04) and especially NS-S days (p=0.005) (Table 3). Overall, obese participants were 50% less likely to reach the minimum recommended sleep durations than very underweight participants, and more than four times more likely to fail to reach recommended sleep durations on NS-S days.

Table 3.  Percentage of adolescents in each weight class failing to meet the National Sleep Foundation guidelines on all days considered together, and on each of the four day types, and associated probabilities from chi-square analysis.
Day typeVUWUWNWOWOBP
  1. Notes:

  2. VUW = very underweight   UW = underweight   NW = normal weight   OW = overweight   OB = obese

  3. S-S = days when adolescents went to bed on a school day and woke on a school day

  4. S-NS = days when adolescents went to bed on a school day and woke on a non-school day

  5. NS-NS = days when adolescents went to bed on a non-school day and woke on a non-school day

  6. NS-S = days when adolescents went to bed on a non-school day and woke on a school day

All days2325313536<0.0001
S-S31293740380.11
S-NS24192828310.52
NS-NS19202428290.04
NS-S18334145550.005

Relationship between weight status and wake up and bed times

Most of the differences in sleep duration among school days and weight status categories were due to variation in bedtimes. Mean bedtime ranged from 9.36 pm on S-S days to 10.37 pm on NS-NS days. There were significant differences in bedtimes among weight status categories on all days considered together (p<0.0001), S-NS days (p=0.03), NS-NS days (p=0.0004) and NS-S days (p=0.003), but not on S-S days (p=0.39) (Table 4). Considering all day types together, obese participants went to bed significantly later than underweight (by 23 minutes), normal weight (12 minutes), and overweight participants (nine minutes). On NS-S days, obese participants went to bed 46 minutes later than very underweight participants. There were no significant differences across weight status categories in wake up times.

Table 4.  Bedtimes of adolescents on all days considered together, and on each of the four day types, and associated probabilities.
Day typeVUWUWNWOWOBP
  1. VUW = very underweight   UW = underweight   NW = normal weight   OW = overweight   OB = obese

  2. S-S = days when adolescents went to bed on a school day and woke on a school day

  3. S-NS = days when adolescents went to bed on a school day and woke on a non-school day

  4. NS-NS = days when adolescents went to bed on a non-school day and woke on a non-school day

  5. NS-S = days when adolescents went to bed on a non-school day and woke on a school day

All10:05 PM9:55 PMade10:06 PMbe10:09 PMcd10:18 PMabc<0.0001
S-S9:38 PM9:35 PM9:41 PM9:42 PM9:44 PM0.39
S-NS10:02 PM9:47 PMabc10:10 PMc10:18 PMb10:13 PMa0.03
NS-NS10:49 PMe10:21 PMcde10:29 PMb10:33 PMad10:50 PMabc0.0004
NS-S9:22 PMde9:45 PMa9:50 PMbe9:56 PMcd10:08 PMabc0.003

Discussion

The key findings of this study were: 1) sleep duration was significantly associated with weight status; 2) not only did obese participants sleep less than normal weight participants, but normal participants slept less than underweight participants; 3) day type was an important moderator variable in the relationship between weight status and sleep duration; and 4) the differences in sleep duration were primarily due to differences in bedtime.

Sleep across the day types

There were sleep gradients across weight status categories when all day types were considered together, and especially on NS-S days, but not on S-NS days. This pattern is hard to reconcile with the theory that obesity may cause shorter sleep duration. If this were the case, we would expect to see shorter sleep durations in obese children only (as opposed to linear gradients across weight status categories) and on all day types. One possible explanation for the observation that obese adolescents fell asleep later on NS-S days may be anxiety associated with returning to school (given the self-esteem issues widely recognised to be related to obesity). However, this explanation does not explain why the normal weight adolescents' bedtimes were significantly later than underweight adolescents, and it also does not hold up on S-S days, when obese adolescents also have to return to school the next day. The sleep patterns described here sit much more comfortably either with the theory that short sleep duration predisposes towards overweight and obesity, or that some third factor (perhaps screen time or low physical activity) contributes both to overweight and short sleep duration, in a way which interacts with day type (for example, screen times are much higher on non-school days).

If this is so, accumulated sleep debt, as a result of very short duration sleep on some days of the week, may be a more significant predictor of obesity than sleep duration. Given this pattern, we should be concerned with the age-related increase in the gap between school and non-school day sleep. The difference between NS-NS day sleep (e.g. Saturdays or holidays) and NS-S day sleep (e.g. Sundays) increased from seven minutes for 10 year olds to 106 minutes for 18 year olds. In other words, as adolescents get older, their school day sleep declines rapidly (at the rate of about 16 minutes per night per year of age), whereas their non-school day sleep remains fairly constant (declining at only four minutes per night per year of age). This increasing gap may be associated with the demands of study, the attractions of socialising, or the perceived need for some ‘chill-out’ time among older adolescents. It has also been proposed that a delay of circadian phase during adolescent development also contributes to the delay of sleeping times in adolescents.24 Solutions might include more flexible starting times in secondary schools, bedtime curfews at home, technology-free bedrooms and “sleep hygiene” education, including the advantages of strategic napping over extended recovery sleep on the weekend.36

It may be argued that the differences in sleep duration identified in this study (which averaged 20 minutes when all day types were considered together, though was up to 65 minutes on NS-S days) are statistical differences as a function of the large sample size, rather than being of clinical significance. However, several other authors have reported clinically significant relationships between a number of health and performance issues and sleep reductions in the vicinity of those reported here. For example, Lumeng et al.37 found that every extra 30 minutes sleep per night at nine years of age reduced the risk of overweight at 12 years by 20%. Agras et al.10 found the mean difference in sleep time between infants who did and did not become overweight by nine years was 30 minutes. In addition, a 25 minutes sleep reduction in adolescents has been associated with poor academic performance23 and experimental reduction of sleep duration by 60 minutes has been linked to memory and reaction time deficits.38

Despite the relationship between weight status and sleep duration found in this, and other studies, it is interesting to note that two thirds of obese participants met the National Sleep Foundation guidelines. Sub-group analyses comparing obese participants who did and did not meet the sleep guidelines showed there were significant differences in socioeconomic status (on the basis of SEIFA codes; 1008 versus 960, p=0.007), total screen time (238 versus 303 minutes, p=0.001) and television time (174 vs 210 minutes, p=0.02), confirming that other factors are interacting with the relationship between weight status and sleep. There were no differences in moderate to vigorous physical activity, total energy expenditure, computer time or video game time.

Sleep patterns in underweight participants

Aside from the possibly protective effect of sleep against the risk of overweight, it is not easy to understand why underweight participants should experience more sleep than other participants. Contrary to the findings of Agras et al.10 there was no relationship between physical activity (minutes of moderate to vigorous physical activity, daily pedometer steps or estimated daily energy expenditure) and sleep duration in the sample used in the current study, therefore tiredness due to physical activity is unlikely to be the cause. Human growth hormone is mainly secreted during sleep, so the extra sleep in underweight participants may be a response to a perceived need for growth.39 It is also possible that undiagnosed or unreported pathologies in the underweight may lead to greater tiredness. However, underweight participants reported fewer health problems on almost all of 17 specific health probes in the National Children's Nutrition and Physical Activity Survey, which constituted a large part of this sample.

Sleep duration, bedtime and wake-up time

Reduced sleep duration was closely associated with later bedtimes. It has been suggested that this association may be mediated by increased evening screen time.15 There is emerging evidence of a mechanistic link between screen exposure and sleep disturbance. The sleep-promoting hormone melatonin is suppressed by exposure to bright light, and to blue light in particular.40–42 Blue light exposure increases alertness and wakefulness.43,44 Television and computer screens may emit greater proportions of blue light wavelengths45 and therefore exposure to screens in the evening may have the potential to inappropriately reset the body's internal clock, increase alertness, decrease the body's ability to initiate sleep and deregulate sleep/wake rhythms. In addition, the stimulating nature of screen-based games (often combative and characterised by violent imagery) would appear likely to impair sleep initiation.

Strengths and limitations

This study is the first to distinguish between the relationships of weight status with sleep patterns on different day types, and the first to compare sleep patterns across the full gamut of weight status categories from underweight to obese. Other strengths of the study are the large and reasonably representative sample and the wide age range.

It is important to acknowledge that the sampling methods varied among the included datasets. The largest dataset from which approximately 60% of participants were drawn involved randomly sampled children from throughout Australia (Australian National Children's Nutrition and Physical Activity Survey). The remaining participants were part of a cohort from Victoria (Health of Young Victorians Survey) and mainly randomly sampled schools throughout South Australia. Data collected in different surveys may have been susceptible to confounding due to factors such as seasonal variation and secular trends. When adjusted for age, sex and day type, sleep time varied by less than five minutes across seasons. Secular reductions in sleep time in the order of 1 to 1.5 minutes per year have been found for children,21 however in our combined data set, the average year of measurement varied by less than three months across weight classes and the less than seven months across day types (equivalent to secular trends of less than 30 seconds sleep per night and less than one minute sleep per night respectively). Therefore, these factors appear unlikely to have affected our results.

Self-report measures are always potentially subject to recall bias. However, recall of bedtimes and wake-up times have been shown to be reliable and accurate,46 and diaries have been shown to be as accurate as actigraphy for time in bed,47 but not for nocturnal wake times. Recall of sleep times is highly reliable for the MARCA (ICC = 0.87).

Each participant provided up to four days' use of time data (representing a maximum of three nights of sleep data), thus data from each of the four day types was not available for any individual participants. In the future, a repeated measures methodology where each child provides sleep data for each of the day types would afford a more powerful design with less risk of residual confounders.

This study focused on sleep duration, and did not attempt to assess sleep quality. It is possible that some participants who report short duration sleep may enjoy high quality, undisturbed sleep, while others may spend a long time in bed but with broken sleep. Furthermore, it remains possible that the critical variable may not be actual sleep duration, but some other characteristic of sleep, such as bedtime, for example.

Conclusion

The association between weight status and sleep duration showed consistent gradients across all weight categories, from obese to underweight, but only for certain types of days. These patterns do not support the notion that overweight and obesity cause short sleep duration through mechanisms such as sleep apnea. Rather, it appears that short sleep may predispose individuals to overweight and obesity, or alternatively, that a third factor (perhaps screen time) contributes to both overweight and short sleep duration. Sleep intervention studies, and cross-sectional studies examining the relationship between screen time, weight status and sleep would assist to clarify these issues.

Acknowledgements

Melissa Wake, for making available the HOYVS dataset.

This study was supported in part by the Australian Commonwealth Department of Health and Ageing, the Department of Agriculture, Fisheries and Forestry; and by the Australian Food and Grocery Council.

Ancillary