Delayed bedtime on non‐school days associates with higher weight and waist circumference in children: Cross‐sectional and longitudinal analyses with Mendelian randomisation

Sleep duration has been linked with obesity in population‐based studies. Less is known about bedtimes and, especially, if discrepancy between bedtimes on school and non‐school days associate with adiposity in children. The associations of self‐reported bedtimes with the body mass index z‐score (BMIz) and waist‐to‐height ratio (WtHr) were examined among children with a mean (SD) age of 11.2 (0.85) years in cross‐sectional (n = 10,245) and longitudinal (n = 5085) study settings. The causal relationship of whether BMIz contributes to bedtimes, was further examined in a subset of 1064 participants by exploiting Mendelian randomisation (MR). After adjusting for sleep duration and other confounders, every 0.5 h later bedtime on non‐school nights and a delay in bedtime in non‐school nights compared with school nights associated with 0.048 (95% CI 0.027; 0.069) and 0.08 (95% CI 0.056; 0.105) higher BMIz as well as 0.001 (95% CI 0; 0.002) and 0.004 (95% CI 0.003; 0.005) with higher WtHr, respectively. Moreover, every 0.5‐h delay in bedtime in non‐school nights compared with school nights associated with 0.001 (95% CI 0; 0.002) greater increase in WtHr in the 2.5 years follow‐up. Thus, a 2‐h delay in bedtime at the age of 11 years corresponds with a 0.6 cm increase in waist circumference. The MR analysis did not indicate an opposite causal relationship: higher BMIz was not causing delayed bedtimes. Later bedtime on non‐school days and discrepancy in bedtimes associated with increased BMIz and WtHr, while longitudinally these predicted higher WtHr, independently of sleep duration. Promoting early bedtimes, especially on weekends, should be considered in obesity prevention among school‐aged children.

Short-term sleep deprivation or poor-quality sleep are associated with poor physical and academic performance at all ages (Chaput et al., 2018), while these are also linked with adverse chronic health outcomes, e.g.depression (Zhai et al., 2015), and all-cause mortality (Shen et al., 2016).Furthermore, several unhealthy sleep habits seem to cluster with being overweight (Garfield, 2019;Garfield et al., 2019) without clear causal relationship.
The relationship between short sleep duration and overweight appears bidirectional.Excess weight, especially in the trunk area, can disturb sleep in terms of breathing problems and even cause sleep apnea, which may lead to excessive sleepiness and tiredness during daytime, also in children (Li et al., 2017).To compensate for tiredness, people tend to eat more than required or to desire more unhealthy foods than when not tired (Dashti et al., 2015).In addition, due to tiredness they may skip exercise routines, which possibly contributes to the vicious cycle by enhancing weight gain (Machal et al., 2018).Besides, sleep deprivation itself alters energy metabolism in otherwise healthy young individuals (Diethelm et al., 2011).Naturally, children are more sensitive to cognitive, physical, and emotional effects of sleep deprivation than adults (Bayon et al., 2014;Diethelm et al., 2011;Talbot et al., 2010).
Self-reporting of sleep habits is used widely in population-based studies despite imprecisions compared with objective measures (Nascimento-Ferreira et al., 2016).In children's health surveys, sleep habits are typically addressed with questions concerning bedtime and waking hours during school days and non-school days (Gariepy et al., 2020).In general, self-reporting of health behaviours among school-aged children is considered more reliable than parent-reports of children's health behaviours (Koning et al., 2018;Nascimento-Ferreira et al., 2016).
A meta-analysis of Li and colleagues (Li et al., 2017) summarised 12 longitudinal cohort studies with 44,000 children, and reported that short sleep duration increased the likelihood for childhood obesity by 45%.The strongest association between short sleep duration and obesity has been observed in children with deprived backgrounds (Collings et al., 2017), while sleep interventions have not been shown to be effective in the treatment of obesity (Li et al., 2017).The role of sleep timing, in practice bedtime, in the development of adiposity has been poorly investigated (Dutil et al., 2022).Therefore, this study focusses on bedtimes on school nights and non-school nights, and the discrepancy between bedtimes on school and non-school nights.The objectives are to examine the associations of these bedtimes with body mass index (BMI) and waist-to-height ratio (WtHr) in 9-12-year-old children using cross-sectional (n = 10,245) and longitudinal (n = 5085) study settings.The causal relationship is further examined in a subsample by exploiting Mendelian randomisation, assessing the genetically predicted BMI on bedtime.

| Participants
This study utilised data from the Finnish Health in Teens study (Fin-HIT), which is a school-based cohort study of initially 9-12 year old Finnish children.In 2011In -2014, 11,407 , 11,407 children participated in the baseline data collection at schools, and about 2.5 years later 54% of them took part in the first follow-up survey online.The details of the Fin-HIT cohort are described elsewhere (Figueiredo et al., 2019).The Coordinating Ethics Committee of the Hospital District of Helsinki and Uusimaa has approved the study protocol (169/13/03/00/10), and a written informed consent was obtained from all participants and their parents.

| Anthropometry
Children's anthropometry, including height, waist (centimetres, cm) and weight (kilograms, kg) were measured at baseline in a standardised way by trained field workers.Children's body mass index (BMI) (kg/m 2 ) was calculated and age-and sex-specific z-scores (BMIz) derived based on the International Obesity Task Force (IOTF) guidelines (Cole & Lobstein, 2012).The waist-to-height ratio (WtHr) was calculated to reflect central adiposity (Savva et al., 2000).At the follow-up, the anthropometry was home-measured by one of the parents.We have previously reported the validity of home-measured anthropometry among 113 children (Sarkkola et al., 2016).BMIz and WtHr were used as continuous variables in our analysis.Collectively these are referred to as anthropometry in this text.

| Sleep habits
At baseline participants filled in a comprehensive questionnaire on lifestyle factors including sleep habits, adapted from the WHO's Health Behaviour in School-aged Children study questionnaire for 13and 15-year-olds with minor modifications (Gariepy et al., 2020).
Bedtimes on school nights were asked with a question: When do you usually fall asleep in the evenings on a school night?with 12 response options varying from 19:30 or earlier [1] to 01:00 or later [12], each differing by a half hour.Similarly, bedtime during non-school days was asked with the following question: When do you usually fall asleep in the evenings before a day off?with 14 response options varying by 0.5 h from 20:30 or earlier [1] to 03:00 or later [14].A discrepancy in bedtimes was calculated by subtracting the school night bedtime from the non-school night bedtime for each child, which is called a delay in bedtime in this text.
To calculate sleep duration, typical waking hours on school days were asked from the child with the following question: When do you usually wake up on school days?The seven response options varied between 5:30 or earlier [1] and 8:30 am or later [7].For non-school days the question was: When do you usually wake up on a day off?The 14 response options varied from 6:30 or earlier [1] to 13:00 or later [14], each differing by 0.5 h.Thus, typical sleep duration (with 0.5 h accuracy) was calculated separately for school days and nonschool days.Weighted mean for sleep duration was calculated as following: (5 Â sleep duration on school days +2 Â sleep duration on non-school days)/7 and was used when comparing sleep duration with recommendations (Hirshkowitz et al., 2015) and the literature.

| Other background information
Screen time and leisure-time physical activity (LTPA) were covered by the baseline questionnaire and thus, self-reported by the child as described previously (Engberg et al., 2020).Screen time was calculated as a sum of TV viewing and computer use reported as h/day for school days and non-school days separately, whereas LTPA duration was reported for the whole week (h/week).
The questionnaire included evaluation of pubertal development based on Tanner stage with pictorial assessment of breast development and pubic hair for girls and only pubic hair for boys with a scale of 1-5 (Tanner, 1962).Due to several incomplete responses, the categorisation was recoded into prepuberty (T1-2), puberty (T3-4) and postpuberty (T5) to describe the puberty phase.Additionally, the child's age and sex were included in the analysis.

| Statistical methods
Normality of the main variables (= BMIz, WtHr, bedtimes and sleep duration on school and non-school nights at baseline) were visually inspected.
Extreme outliers (> ±4 SD) were omitted from these to truncate their distributions.Missing values were not allowed in the main variables.The number of missing values are shown for other variables in Table 1.In the analysis, bedtimes were considered as continuous variables.
The associations between bedtimes/delay in bedtime and anthropometry were studied with linear regression using three models: 1, crude association; 2, adjusted for sleep duration (bedtimes) or difference in sleep duration (delay in bedtime); and 3, fully adjusted analysis including additional adjustment for age, sex, puberty phase, LTPA, and the corresponding screen time (bedtime)/difference in screen time (delay in bedtime) to consider only significant confounding factors in cross-sectional analyses.In longitudinal analyses, similar modelling was applied, except that model 3 included also baseline anthropometry, which allowed interpreting the change in anthropometrics.In model 3 there were 20.5%, 0.2%, and 0.2% of missing values in the following covariates: puberty phase, LTPA, and screen time, respectively.We imputed these by using pooled values across five logistic regression models.These models included age, sex, height, waist, BMIz, and bedtimes during school nights and non-school nights.
Correlations between anthropometrics, bedtimes, and sleep durations were tested with Pearson correlation.Multicollinearity was measured with the variable inflation factors (VIF) using VIF >5 as an indication of problematic multicollinearity in linear regression models.There were no concerns of multicollinearity in our analysis.The highest VIF of 3.3 was observed in models 2 and 3 including both bedtime and sleep duration during school nights.
Mendelian randomisation (MR) was used to assess potential causality between BMI and sleep in a subset of the cohort with Metabochip data available (Viljakainen et al., 2019).For this, single nucleotide polymorphisms (SNPs) with a known effect on BMI were selected (Locke et al., 2015).For one-sample MR analyses, a genetic risk score (GRS) was calculated for each participant by weighting the risk allele count by the reported per-allele effect in Plink v1.9.One-sample MR analyses were performed using individual-level data in 2-stage least squares regression (2SLS) analysis.The F statistic was used to validate the association between GRS and BMIz, and a value ≥10 was used as an indication of the validity of the MR analysis (Staiger & Stock, 1997).For two-sample MR analyses, we used summary data (β coefficients and standard errors) from Locke et al. (Locke et al., 2015) for the SNP-exposure (BMI) associations and Fin-HIT data for the SNP-outcome associations (Viljakainen et al., 2019).Two-sample MR analyses were conducted using the "TwoSampleMR" R package.
Other statistical analyses were performed with IBM SPSS Statistics Version 27.Significance level with 5% uncertainty was adopted.

| Characteristics of study population
The study population consists of 10,245 Fin-HIT participants who had all main variables available at the mean (SD) age of 11.2 (0.85) years (Table 1).Supplementary Tables 1 and 2 in Appendix S1 describe the sub-samples with genotype data (n = 1064) and longitudinal data on anthropometry (n = 5085), respectively.Baseline characteristics, including screen time, LTPA, and puberty phase were similar across the entire study sample and the sub-samples.Weighted mean (SD) for sleep duration was 9.7 (0.8) hours per night.Of the participants, 16.4% (n = 1681) did not meet the sleep recommendation for age, which is 9-11 h/night (Hirshkowitz et al., 2015).
Median bedtime was at 21:30 and at 22:30 for school nights and Determinants of bedtimes and the delay in bedtime are shown in Supplementary Figure 2 in Appendix S1.The delay in bedtime was highest in those engaged in higher amounts of screen time, followed by higher age, and advanced puberty.

| Cross-sectional associations of bedtime with anthropometry
Table 2 shows cross-sectional associations of bedtimes and delay in bedtime with BMIz and WtHr.On school nights, the associations between bedtimes and anthropometry varied between the models.After adjusting for sleep duration and other confounders, every 0.5 h later bedtime during non-school days, associated with 0.048 (95% CI 0.027; 0.069) higher BMIz and 0.001 (95% CI 0; 0.002) higher WtHr.Correspondingly, a delay in bedtime between non-school and school nights was consistently and positively associated with BMIz and WtHr.Our finding concerning the delay in bedtime in Model 3 may be interpreted that a 2 h delay in bedtime results in 0.32 higher BMIz (= 4 Â 0.08) and a 2.4 cm higher waist circumference in a 148 cm tall child (= 4 Â 0.004 Â 148 cm) independently of other lifestyle factors (Table 2).

| Longitudinal associations of bedtimes with anthropometry
Table 3 shows longitudinal associations of baseline bedtimes and delay in bedtime with BMIz and WtHr after a 2.5 year follow-up.
In the fully adjusted model, each 0.5 h later bedtime on school nights associated with 0.042 (95% CI 0.007; 0.077) higher BMIz, but not with WtHr.On non-school nights, later bedtime associated consistently with 0.001 (95% CI 0; 0.002) higher WtHr.Correspondingly, the delay in bedtime between non-school and school nights was associated with a greater increase in WtHr.Based on our findings, we calculated that a 2 h delay in bedtime at 11 years of age (Model 3) predicted a 0.6 cm increase in waist circumference (Table 3).   3 in Appendix S1 (Table 4).

| DISCUSSION
The associations of self-reported bedtimes with anthropometry were investigated among 9-12-year-old children using both cross-sectional and longitudinal study designs as well as Mendelian randomisation to identify a potential causal effect of BMIz on bedtimes.The crosssectional analysis revealed consistent, moderate associations between later bedtime and higher BMI/WtHr on non-school days.Findings concerning the delay in bedtime on non-school nights compared with school nights confirmed these.Moreover, the longitudinal analyses demonstrated that a later bedtime on non-school nights and a delay in bedtime at baseline associated with higher increases in WtHr 2.5 years later.These associations were independent of sleep duration and other cofounders.Furthermore, MR analysis did not support causality, i.e.BMI impacting bedtimes.
In the present study, we focussed on bedtime, which is one aspect of sleep habits (Gariepy et al., 2020;Grummon et al., 2021).Of the 9-11-year-old children investigated here, 84% met the sleep recommendation for their age (Hirshkowitz et al., 2015), which is somewhat higher than shown in a pan-European report on sleep habits among 13.5-year-old children (Gariepy et al., 2020).A chance for over reporting exists due to methodological issues of self-reports (Nascimento-Ferreira et al., 2016).Children were likely to meet the minimum sleep recommendation if they went to bed before 21:30 on school nights or before 23:00 on non-school nights.Bedtime correlated strongly with sleep duration, but still was independently related to anthropometry.Weaker correlation on non-school days implies that some children compensated the delayed bedtime with later wake up time.Still, even a short-term sleep deprivation may induce stress, emotional distress, and behavioural problems in children (Medic et al., 2017).Biologically, a late bedtime is associated with fewer sleep cycles and less stage 3 sleep, also known as slow wave sleep, which is essential for the body to recover and grow (Colten & Altevogt, 2006).There is evidence that slow wave sleep promotes health in multiple ways e.g., by inducing immunity, insightful thinking, creativity, and memory (Yordanova et al., 2010).
The associations of late bedtimes with high anthropometry were seen consistently on non-school days.We calculated a delay in bedtime on non-school days compared with school days for each child, and showed that irregular sleep timing associated with adiposity, as reported previously (Quach et al., 2016).On average, a moderate delay of about an hour in bedtime when moving from the school week to days-off was observed, but every eight child's bedtime was delayed by at least 2 h.The delay in bedtime and late bedtimes in general were more common in those engaging in higher amounts of screen time, which is in line with previous results as reviewed recently (Dutil et al., 2022).Our findings on LTPA were somewhat contradictory, as LTPA associated with later bedtime during school nights.Accordingly, higher activity in sport is shown to postpone bedtime in Czech adolescents (Machal et al., 2018).However, the overall effect of physical activity on various aspects of sleep seems positive as discussed elsewhere (Machal et al., 2018).Our findings on higher age and advanced puberty in this context are in line with others reporting later bedtime in older teenagers, which may be due to peers, referred to as social jetlag (Gariepy et al., 2020), but also changes in sleep architecture have been established during puberty (Colrain & Baker, 2011).
In this study, the longitudinal analyses with 5085 children confirmed the cross-sectional findings, especially for WtHr.The later bedtime on non-school days and the discrepancy in bedtimes experienced at the age of 11 years alone predicted a 0.6 cm higher waist circumference 2.5 years later.This calculation assumes a stable delay in bedtime during adolescence, but as the general tendency of social jetlag is to increase with age (Gariepy et al., 2020), the actual increase might be even higher.A systematic review by Dutil et al. pointed to three previous longitudinal studies, which investigated the association of bedtime with adiposity, including altogether 7766 children aged between 4 and 18 years (Dutil et al., 2022).Positive associations were seen in two studies using self-reported bedtime (Asarnow et al., 2015;Quach et al., 2016), while no association was shown in a study using an objective measure of sleep-onset among 823 8-year-old children in New Zealand (Taylor et al., 2020).Besides these, we found two other longitudinal studies with 16,000 and 5819 participants (Collings et al., 2022;Swindell et al., 2021).Late self-reported bedtime during week contributed to higher body fat percentage in a 10 year follow-T A B L E 2 Cross-sectional associations of bedtime and delay in bedtime on school and non-school nights with BMIz and WtHr with three models (n = 10,245) using linear regression   up of initially 9-11-year-old children from United Kingdom, and the effect was independent of known covariates (Swindell et al., 2021).
Correspondingly, a prospective analysis of accelometer-measured data demonstrated that an optimal sleep on-set for adiposity outcomes for about 11-year-old children would be between 21:00 and 22:00, but the effect remained dependent on sleep duration (Collings et al., 2022).Based on our and others' findings, we speculate that inconsistency in bedtimes and sleeping occurs most likely due to screen time and socialising with peers.It may cause "circadian misalignment" that disturbs energy metabolism (Gonnissen et al., 2012;Spaeth et al., 2013), which is noted as increased waist circumference here, although other explanations may apply.
Additionally, we investigated the causality between BMIz and bedtime with MR, in order to explore whether the association of delayed bedtimes on BMI could be reversed.We have previously constructed polygenic risk scores for obesity using various published datasets (Viljakainen et al., 2019), which were further tested here.We found no support for causality between genetically predicted BMIz and bedtimes, suggesting that higher BMIz is not causing delayed bedtimes, but perhaps instead is a consequence.A few studies have studied the causal relationship between obesity and sleep using Mendelian randomisation (Dashti et al., 2021;Dashti & Ordovas, 2021;Wang et al., 2019).Our study was the first to examine a causal effect of BMI on sleep habits in children.Although we were unable to explore the potential causal effect of the inversed relationship (delayed bedtimes contributing to BMIz), one previous MR study in the Hong Kong's "Children of 1997" birth cohort, including 36,000 children, has studied the causal effect of sleep duration on BMI in children (Wang et al., 2019).
Genetically determined longer sleep duration was associated with lower BMI.The observed association of longer sleep with lower BMI in that study was confirmed both cross-sectionally at the age of 11 years and longitudinally at the age of 16 years, which speaks for downstream effects of sleep habits on BMI.
Limitations of our study include self-reported bedtimes, which tend to be earlier than the actual sleep-onset time obtained with objective methods (Gariepy et al., 2020;Grummon et al., 2021).Moreover, a sleep habit questionnaire may provide more reliable information on sleep during weekdays than days-off (Nascimento-Ferreira et al., 2016).
We observed higher variation in bedtimes and sleep duration during non-school days than on school days, which may partly explain our finding on a stronger association for non-school days.As stated, only selfreported measures are feasible in large-scale studies such as ours (Nascimento-Ferreira et al., 2016).Unfortunately, we had no longitudinal data on sleep behaviours, since sleep habits may evolve in adolescence (Colten & Altevogt, 2006;Machal et al., 2018).In addition, genotype data were available only from a sub-sample of 1064 children.
The Metabochip is a custom array designed for the replication of top associations from cardiometabolic GWAS studies (Voight et al., 2012) performed until 2012.Therefore, the Metabochip has poor coverage of SNPs identified in later GWAS studies, such as those on sleep duration or chronotypes.Consequently, we were not able the explore the inverse relationship, which would have required SNP data for these phenotypes.Although we have considered several mediating factors in the modelling, additional ones may exist, such as late bedtime possibly co-occurring with unhealthy eating or temporarily decreased physical activity (Grummon et al., 2021;Machal et al., 2018).Thus, some residual effects likely remained unsolved in our study.Nevertheless, with a large sample size, we detected an independent, but small long-term association between self-reported bedtime and WtHr.
The strengths of our study include the information on various aspects of sleep habits and the longitudinal data, which allowed us to address independent risk factors for adiposity and predict changes in anthropometry in over 5000 adolescents.Anthropometry was measured by field workers at baseline, while measured at home in the follow-up.In the previous validation study, at-home measurements resulted in a lower mean BMI but led to a misclassified BMI category only in 2 (1.8%) participants (Sarkkola et al., 2016).We showed that the association of bedtime with later anthropometry was independent of sleep duration among children, highlighting the need to emphasise and address not only sleep duration but also consistent bedtimes in obesity prevention as well as in future studies.Consistent bedtimes between non-school and school days may be achieved in two ways: either by allowing later school start times (Bowers & Moyer, 2017) or pursuing earlier bedtimes on non-school days.Of these the former seems more feasible for adolescents and their families.In addition, we had incorporated various lifestyle factors that partly explain the relationship between bedtime and anthropometry, and thus we provide a realistic picture of the connections and their magnitudes.

| CONCLUSION
We showed that bedtime is associated with weight development in children independently of sleep duration and other confounding factors.Late bedtime, especially on non-school days, and inconsistent bedtimes were related to higher adiposity, which were partly retained in longitudinal analysis.Our MR analysis suggested that the association of delayed bedtimes on higher adiposity was not reversed; genetically predicted higher adiposity did not seem to cause delayed bedtimes.Further studies are needed to address clinical relevance of our findings and unveil the mechanistic aspects.

AUTHOR CONTRIBUTION
HV and ED were responsible for data analysis and generation of fig- T A B L E 1 Baseline characteristics of 10,245 participants with mean (SD) and 25th and 75th percentile values, if not indicated otherwise non-school nights, respectively.The distributions of bedtimes during school and non-school nights, and mean sleep duration by the bedtimes are shown in Figure 1.The recommendation for sleep duration (9-11 h/night) was not met with 95% likelihood, if bedtime occurred Distribution of bedtime during (a) school days and (c) non-school days in 10,245 children.Corresponding mean sleep duration by bedtime (b) and (d).Error bars represent SDs.Dashed green lines indicate the sleep recommendation, 9-11 h/night, while dashed red line indicates inadequate sleep duration (<7 hour/night), for this age group.laterthan 21:30 on a school night, or later than 23:00 on a non-school night (Figure1).There was r = À0.83 ( p < 0.001) correlation between bedtime and sleep duration on school nights, and r = À0.48 ( p < 0.001) correlation on non-school nights.Supplementary Figure1in Appendix S1 summarises the associations between bedtime, sleep duration, and BMIz.The mean delay in bedtime was 0.91 (0.68) h, while 12.7% (n = 1301) of children went to bed at least 2 h later on non-school nights than on school nights.
MR analysis was performed to further investigate the causal inference of BMIz with bedtimes.This included one sample MR with Fin-HIT data only and two sample MR with summary data from Locke et al. (Locke et al., 2015) (e.g.associations of obesity-related SNPs with BMIz), which provides tools for performing additional sensitivity analyses.The one sample MR included checking instrumental variables with predefined key assumptions (Supplementary

T A B L E 3
Longitudinal associations of bedtime on school and non-school nights and delay in bedtime with BMIz and WtHr at follow-up using linear regression with three different models (n = 5085) ures and tables.HV and EE did the literature search.HV provided the first draft of the manuscript.All authors were involved in study design, interpretation of the data, writing and editing the paper and had final approval of the submitted and published versions.

Table 3 in
Appendix S1) and causal association of genetically predicted BMIz with bedtimes (Table4).We found no evidence that BMIz would