1 INTRODUCTION

Cross-sectional studies, and meta-analyses of prospective studies, report that longer sleep duration is associated with lower adiposity and obesity risk in youth.1-3 There is considerable heterogeneity in the results of individual prospective studies, however, and investigations rarely include adiposity markers other than weight-for-height indices, even though centrally stored adiposity is metabolically more harmful than total adiposity.4 In addition, most investigations continue to solely focus on sleep duration. Sleep is a multidimensional construct of partly overlapping dimensions, including sleep duration, timing, quality, and variability. Investigation of diverse sleep parameters and their links with health is needed.5

There is some inconsistency in the literature, but an increasing number of cross-sectional studies,6, 7 and a small but emerging prospective evidence-base,8-11 have recently reported that later bedtimes are associated with higher adiposity and increased overweight and obesity in youth. Some studies even suggest that the associations are distinct from sleep duration.12-14 This information could be pivotal in terms of refining obesity prevention efforts. It may also have important implications for sleep recommendations, which have historically focussed on duration as opposed to any other sleep parameter.15 More studies are required to investigate sleep dimensions simultaneously, but it is challenging to investigate the independent or interactive effects of bedtimes and sleep durations, because they are often highly correlated. This is particularly true in school-aged youth, the majority of whom will wake-up within a narrow set of times on weekday mornings to attend school, thus bedtimes largely dictate variation in sleep duration. Reassuringly, this issue can be overcome in large cohorts of children, if the number of individuals with atypical behaviours is sufficient to allow robust estimation of closely related exposures with outcomes.16 As an example, Olds and colleagues conducted a cross-sectional analysis of 2200 9- to 16-year-olds from Australia who were grouped into four sleep–wake groups. Relative to a late-bed/late-rise sleep pattern, an early-bed/early-rise pattern (which was characterized by an approximately equal sleep duration but 86 minutes earlier bedtime) was associated with lower BMI z-score and lower likelihood of overweight and obesity.12 Similar studies are warranted, as are prospective investigations of sleep durations and bedtimes, to quantify the influence of both dimensions on youth adiposity over time.6

It is also important to shed light on the possible pathways by which sleep parameters may relate to adiposity. The best available evidence remains of low quality, but later bedtimes and short sleep are hypothesised to create more opportunities for sedentary time, including screen-time such as TV viewing, as well as fatigue-induced reductions in daily physical activity.17 Previous studies have either failed to consider the role of awake-time movement behaviours or have typically estimated activity metrics using data from questionnaires,1-3 which are prone to bias and random error.18 Providing additional evidence to support one or more of the hypothesised pathways could help to ameliorate the consequences of poor sleep by identifying modifiable intermediate behaviours that can be targeted for mitigation.

The International Children's Accelerometry Database (ICAD) provides a unique opportunity to analyse sleep–adiposity associations across a large and heterogeneous international dataset of children and adolescents. The dataset also benefits from device-measured estimates of habitual sedentary time and physical activity.19 We examined the cross-sectional and prospective associations of sleep durations and bedtimes with total adiposity, central adiposity, and overweight and obesity in ICAD. We also investigated combinations of behaviours and scrutinized associations when adjusted for device-measured sedentary time and physical activity, and reported TV viewing.

2 METHODS 2.1 Study design and participants

This study is based on the second release of ICAD, a repository of harmonized data from 21 studies that assessed sedentary time and physical activity by waist-worn accelerometers in children aged 3 to 18 years.19, 20 Eleven contributing studies, from 6 European countries (UK, Switzerland, Denmark, Estonia, Norway, Portugal) and Australia, also collected information about sleep duration and/or bedtimes and were included in this study.21-28 With the exception of two cluster randomized controlled trials24, 25 (from which only control participant data, or pre-intervention data from intervention-arm participants, were used) all studies were observational. Extensive descriptions of all source data and the harmonization procedures for study variables can be accessed via the ICAD website (http://www.mrc-epid.cam.ac.uk/research/studies/icad/data-harmonisation/).

2.2 Sleep duration and timing

Five studies21, 24, 25, 27, 28 used proxy-reports by parents to capture sleep parameters (including all studies that involved young children—who would have struggled to provide accurate self-reported information), otherwise sleep data were self-reported.22, 23, 26 It has previously been shown that self- and parent-reported sleep durations are equally valid in 9–17 year olds, and that data from both sources agree substantially with polysomnography which is the gold-standard for sleep assessment.29 Seven studies collected information about the time that children and adolescents usually went to bed and got out of bed22, 24-26 and two studies asked when participants usually went to sleep and woke up.21, 23 Four studies collected free-text (hour and minute) responses21, 23-25 and five studies offered a choice of categories.22, 26 Two studies collected free-text responses to the question ‘How many hours per night does your child usually sleep?’.27, 28 The current analyses are based on weekday data only, which was collected by all studies (only five collected information about weekends21-25), and is considered to be more indicative of habitual sleep behaviours in school-aged youth because of set school day routines.30 Due to the broad age range in ICAD, and because there appear to be age-related differences in the associations of sleep with adiposity,1-3 study participants were categorized as children (<12 years) or adolescents (≥12 years). Data from all studies were harmonized to create categorical variables for bedtimes (after 22:00/21:00 to 22:00/before 21:00) and age-group specific sleep durations (children: <10 h/≥10 h and <11 h/≥11 h; adolescents: <9 h/≥9 and <10 h/≥10 h). Bedtime and sleep duration categories were further collapsed to age-specific categories signifying earlier (children: before 21:00; adolescents: before 22:00) and later bedtimes (children: 21:00 or later; adolescents: 22:00 or later) and shorter (children: <10 h; adolescents: <9 h) and longer sleep durations (children: ≥10 h; adolescents: ≥9 h). These data were combined to create four age-group specific combinations of bedtimes and sleep durations (later-shorter/earlier-shorter/later-longer/earlier-longer).

2.3 Adiposity markers

In each study, trained personnel measured height and weight using standardized techniques and calibrated equipment.21-28 The resulting data were used to calculate body mass index (BMI, kg/m2) which was converted to z-scores and weight status categories.31 In all European studies (n = 9), waist circumferences were measured at the end of gentle expiration, with an anthropometric tape placed at the midpoint between the lowest rib margin and the iliac crest.21-26 Waist circumferences were also converted to z-scores.32 Eight studies collected follow-up data, which were used to calculate changes in adiposity (ΔBMI and Δwaist z-scores), by subtracting follow-up from baseline values.21-27

2.4 Covariables

All studies collected information about putative confounders or mediators, including participant age, sex, parental education, sedentary time and physical activity. Parental education was based on maternal education (or paternal qualifications given missing data) and was harmonized to three categories (School/College/University). Sedentary time and physical activity were assessed with uniaxial, waist-mounted Actigraph (Pensacola) accelerometers that were deployed for 4–7 consecutive days. Raw accelerometer files were collated and reprocessed centrally using commercially available software (KineSoft v3.3.80, Loughborough, UK).19 Files were initially reintegrated to 60s epochs, and non-wear periods were identified and excluded by scanning the data array for periods of consecutive zeros ≥60 min, allowing for 2 min of non-zero interruptions. To account for children and adolescents who wore monitors overnight whilst asleep, all registered data between midnight and participant-specific getting out of bed times were discarded. If participant-specific getting out of bed times were missing, then group-level times which represented the average time of getting out of bed for a given study, age and sex, were used. Validated cut-points were incorporated to estimate the average daily time that children spent sedentary, in light intensity physical activity, and moderate-to-vigorous physical activity (MVPA) on weekdays.33 All days with ≥500 min of monitor wear time were considered valid, and participants were required to provide ≥1 valid weekday of data. Seven studies gathered information about child ethnicity. Parents either reported their own ethnicity which was used a proxy21 or their child's ethnicity,23, 26 and in one study, ethnicity was determined by visual inspection by the research team.22 The data were harmonized to two categories (white/other). Maternal height and weight were self-reported in all but one study22 and the data were used to calculate maternal BMI (kg/m2). Sexual maturity was collected in eight studies, via Tanner staging by researchers,26 or by parent- and/or self-assessment.21, 22, 25, 27 For this investigation, pubertal status was based on breast development for girls and pubic hair for boys, and children aged ≤6 years with missing data were assigned a Tanner score of 1. Tanner stages were collapsed to three groups (pre-puberty (Tanner score 1)/in puberty (scores 2 and 3)/completing puberty (scores 4 and 5).34 Eight studies collected parent- or self-reported information about whether participants had a TV set in their bedroom and the amount of daily time spent watching TV (including DVDs and videos).22, 23, 26-28

2.5 Statistics

Multilevel linear and logistic regression models, with random intercepts to accommodate study-level clustering of data, were used to investigate associations of sleep durations (reference: children: <10 h; adolescents: <9 h), bedtimes (reference: after 22:00) and combinations of behaviours (reference: later-shorter) with adiposity markers and weight status. Models were initially adjusted for age and sex (model A) and subsequently also for sedentary time and MVPA (model B). Likelihood ratio tests (LRT) were used to identify improved fit (p < 0.05) if models were specified with random study-level slopes or sex-by-exposure interactions. If model fit was improved by including a sex-by-exposure interaction term, the model was stratified by sex. Models were also stratified by sex if, for any level of an exposure category, there was some evidence for a sex-by-exposure interaction (Wald test p < 0.1). Additional adjustments were made in turn for parental education, ethnicity, pubertal status, maternal BMI, and TV viewing duration and having a bedroom TV. These extra models were framed as sensitivity analyses as they involved smaller analytical samples due to missing data. The analysis was also replicated to include adjustment for total active minutes (the sum of light and MVPA minutes) rather than solely MVPA. All analyses were a priori stratified by children and adolescents. Prospective associations between baseline sleep duration and bedtimes with subsequent changes in adiposity (ΔBMI and Δwaist z-scores from baseline to follow-up) were quantifiable only in children due to limited longitudinal data for adolescents. Longitudinal models were specified as described above but were consistently adjusted for baseline age, follow-up time, and the baseline value of the outcome variable. All analyses were performed with Stata/SE 17.1 software (StataCorp, College Station, TX). No corrections were made for multiple comparisons, but exact p-values are reported. Readers are encouraged to focus on patterns of results, and on the range of plausible values of associations, as indicated by confidence intervals.35

3 RESULTS 3.1 Descriptive characteristics

The characteristics of included participants are shown in Table 1. The sample was predominantly white, and more than one-quarter of participants were overweight or obese. The prevalent bedtimes were 21:00 to 22:00 and after 22:00 in children (age range: 3.8 to 11.9 years) and adolescents (12.0 to 18.4 years), respectively. The average sleep durations across combined bedtime and sleep length categories were: later-shorter (children: 9.2 h; adolescents: 8.3 h), earlier-shorter (children: 9.7 h; adolescents: 8.7 h), later-longer (children: 10.2 h; adolescents: 9.3 h) and earlier-longer (children: 10.8 h; adolescents: 9.9 h). Tables S1 and S2 provide individual cohort characteristics. In total, 5819 children who are described in Table 2, contributed to prospective analyses. Children were followed-up over a mean of 2.3 (range: 0.5 to 8.0) years.

TABLE 1. Characteristics of participants contributing to cross-sectional analyses Children Adolescents Characteristic Girls (n = 6501) Boys (n = 5746) Girls (n = 1927) Boys (n = 1636) Age (y) 10.4 ± 1.8 10.4 ± 1.9 14.1 ± 1.6 14.1 ± 1.6 Parental education School 1995 (35.7) 1733 (35.1) 533 (35.3) 465 (37.1) College 2044 (36.5) 1808 (36.7) 486 (32.1) 405 (32.4) University 1556 (27.8) 1392 (28.2) 493 (32.6) 382 (30.5) Ethnicity White 4383 (94.4) 3870 (94.0) 738 (95.8) 641 (96.4) Other 260 (5.6) 246 (6.0) 32 (4.2) 24 (3.6) Pubertal status Pre-pubertal 1689 (45.0) 2046 (65.4) 157 (12.2) 91 (9.0) In puberty 1742 (46.4) 1013 (32.4) 394 (30.8) 265 (26.2) Completing puberty 325 (8.6) 68 (2.2) 730 (57.0) 656 (64.8) Sleep duration (h/night) Children: <10; adolescents: <9 2338 (37.0) 2319 (40.3) 776 (40.3) 642 (39.2) Children: ≥10 and <11; adolescents: ≥9 and <10 3041 (46.8) 2520 (43.9) 647 (33.6) 544 (33.3) Children: ≥11; adolescents: ≥10 1122 (17.2) 907 (15.8) 504 (26.1) 450 (27.5) Bedtimes After 22:00 760 (14.8) 803 (17.7) 792 (49.0) 715 (52.4) 21:00 to 22:00 2677 (52.2) 2437 (53.8) 594 (36.8) 506 (37.1) Before 21:00 1693 (33.0) 1294 (28.5) 229 (14.2) 144 (10.5) BMI z-score 0.3 ± 1.2 0.4 ± 1.1 0.3 ± 1.2 0.4 ± 1.1 Waist circumference z-score 0.9 ± 1.2 0.7 ± 1.0 0.9 ± 1.2 0.5 ± 1.0 Weight status Underweight 108 (1.7) 61 (1.1) 30 (1.6) 22 (1.4) Healthy weight 4696 (72.2) 4031 (70.2) 1428 (74.1) 1160 (70.9) Overweight 826 (12.7) 761 (13.2) 251 (13.0) 231 (14.1) Obese 871 (13.4) 893 (15.5) 218 (11.3) 223 (13.6) Maternal BMI (kg/m2) 23.6 (5.1) 23.8 (5.3) 23.7 (5.4) 23.8 (5.4) TV in bedroom (n [%] yes) 1705 (45.2) 1653 (50.6) 835 (55.0) 801 (64.4) TV viewing duration (h/day) 1.5 (2.0) 1.5 (1.9) 1.5 (1.5) 2.0 (2.0) Valid weekdays of accelerometer data 5 (2) 5 (2) 4 (3) 4 (3) Monitor wear time (min/day) 775 ± 77 776 ± 80 814 ± 93 804 ± 100 Sedentary time (min/day) 353 ± 83 331 ± 84 451 ± 103 412 ± 108 Light physical activity (min/day) 373 ± 70 375 ± 70 319 ± 81 330 ± 85 Moderate-to-vigorous physical activity (min/day) 46 (31) 68 (42) 41 (31) 59 (41) Total physical activity (min/day) 422 ± 81 446 ± 84 363 ± 91 392 ± 99 Note: Data are n (%) for categorical variables, mean ± SD for normally distributed variables, and median (IQR) for non-normally distributed variables. Data were missing and therefore sample sizes were smaller for parental education (children: n = 10 528; adolescents: n = 2764), ethnicity (children: n = 8759; adolescents: n = 1435), pubertal status (children: n = 6883; adolescents: n = 2293), bedtimes (children: n = 9664; adolescents: n = 2980), waist circumference z-score (children: n = 10 098; adolescents: n = 2990), maternal BMI (children: n = 8986; adolescents: n = 2413), TV viewing duration and having a bedroom TV (children: n = 7044; adolescents: n = 2762). Some (n = 1187) participants spanned age groups. Abbreviation: BMI, body mass index. TABLE 2. Characteristics of children contributing to prospective analyses Characteristic Girls (n = 3098) Boys (n = 2721) Baseline age (y) 10.8 ± 1.8 10.7 ± 1.9 Follow-up duration (y) 2.3 ± 1.4 2.3 ± 1.3 Parental education School 720 (28.4) 637 (28.9) College 1088 (43.0) 963 (43.6) University 724 (28.6) 607 (27.5) Ethnicity White 2259 (94.6) 1975 (94.4) Other 129 (5.4) 117 (5.6) Pubertal status Pre-pubertal 736 (33.0) 968 (55.1) In puberty 1241 (55.6) 734 (41.7) Completing puberty 255 (11.4) 56 (3.2) Sleep duration <10 h 1266 (40.9) 1179 (43.3) ≥10 and <11 h 1443 (46.6) 1197 (44.0) ≥11 h 389 (12.5) 345 (12.7) Bedtimes After 22:00 524 (19.3) 494 (20.9) 21:00 to 22:00 1615 (59.6) 1418 (59.9) Before 21:00 570 (21.1) 454 (19.2) ΔBMI z-score 0.1 ± 0.5 0.0 ± 0.5 ΔWaist z-score 0.2 ± 0.8 −0.1 ± 0.6 Maternal BMI (kg/m2) 23.7 (5.3) 23.9 (5.0) TV in bedroom (n (%) yes) 435 (44.6) 438 (51.8) TV viewing duration (h/day) 1.5 (0.9) 1.5 (1.0) Valid weekdays of accelerometer data 5 (1) 5 (1) Monitor wear time (min/day) 788 ± 75 787 ± 77 Sedentary time (min/day) 365 ± 81 342 ± 82 Light physical activity (min/day) 374 ± 69 376 ± 69 Moderate-to-vigorous physical activity (min/day) 48 ± 23 69 ± 31 Total physical activity (min/day) 423 ± 79 446 ± 81 Note: Data are n (%) for categorical variables, mean ± SD for normally distributed variables, and median (IQR) for non-normally distributed variables. Data were missing and therefore sample sizes were smaller for parental education (girls: n = 2532; boys: n = 2207), ethnicity (girls: n = 2388; boys: n = 2092), pubertal status (girls: n = 2232; boys: n = 1758), bedtimes (girls: n = 2709; boys: n = 2366), Δwaist z-score (girls: n = 2797; boys: n = 2455), maternal BMI (girls: n = 2022; boys: n = 1764), TV viewing duration and having a bedroom TV (girls: n = 976; boys: n = 845). Abbreviation: BMI, body mass index. 3.2 Cross-sectional associations with adiposity and weight status in children and adolescents

Models A (data available on request) and B consistently yielded comparable results, highlighting that adjusting for sedentary time and MVPA did not influence any associations. With regard to total adiposity, Table 3 shows that longer sleep durations and earlier bedtimes were dose-dependently associated with lower BMI z-score in both age groups. Figure 1 illustrates that longer sleep durations were associated with lower odds of overweight and obesity in children (≥10 and <11 h: 0.87 (0.80 to 0.95); p = 0.003; ≥11 h: 0.72 (0.63 to 0.83); p < 0.001) and adolescents (≥10 h: 0.76 [0.61 to 0.95]; p = 0.018). Earlier bedtimes were dose-dependently associated with lower odds of overweight and obesity in children (21:00 to 22:00: 0.88 (0.77 to 0.99); p = 0.047; before 21:00: 0.68 (0.59 to 0.80); p < 0.001). With regard to waist z-score, Table 3 further shows that longer sleep durations were dose-dependently associated with lower values in children, and there was some evidence that bedtimes before 21:00 were associated with lower waist z-score in boys. An association of earlier bedtimes with lower waist z-score in girls was entirely attenuated when adjusted for pubertal status (see Table S3). In adolescents, longer sleep durations and earlier bedtimes were associated with lower waist z-score in boys.

TABLE 3. Cross-sectional associations of sleep durations and bedtimes with adiposity in children and adolescents Child sleep durations Participants (n per category) <10 h ≥10 and <11 h ≥11 h BMI z-score 12 247 (4657/5561/2029) Ref −0.10 (−0.14 to −0.05); <0.001 −0.21 (−0.28 to −0.15); <0.001 Waist z-score 10 098 (4029/4550/1519) Ref