Systematic review and meta-analyses of the relationship between short sleep and incidence of obesity and effectiveness of sleep interventions on weight gain in preschool children

The aim of this study is to determine (a) whether short sleep is associated with the incidence of obesity and (b) whether interventions beneficial for sleep reduce weight gain in preschool children. We systematically searched PubMed, Embase, Web of Science and Cochrane up to 12/09/2019. (a) Studies that were included were prospective, had follow-up ≥ 1 year, with sleep duration at baseline and required outcome measures. (b) Intervention trials with sleep intervention and measures of overweight or obesity were included. Data were extracted according to PRISMA guidelines. (a) The risk of developing overweight/obesity was greater in short sleeping children (13 studies, 42 878 participants, RR: 1.54; 95% CI, 1.33 to 1.77; p < 0.001). Sleep duration was associated with a significant change in BMI z-score (10 studies, 11 cohorts and 29 553 participants) (mean difference: − 0.02 unit per hour sleep; − 0.03 to − 0.01; p < 0.001) . (b) Four of the five intervention studies reported improved outcomes: for BMI ( − 0.27 kg/m 2 ; − 0.50 to − 0.03; p = 0.03); for BMI z-score ( − 0.07 unit; − 0.12 to − 0.02; p = 0.006). Short sleep duration is a risk factor or marker of the development of obesity in preschool children. Intervention studies suggest that improved sleep may be beneficially associated with a reduced weight gain in these children.

Like most obesity prevention programmes, the UK Government's Childhood Obesity plan highlights food, healthy eating and physical activity as key target areas 3 but, given the limited success of such programmes, they may be insufficient to tackle the observed rise in childhood obesity levels. [1][2][3][4] A child's usual sleep duration is affected by age and a number of different factors including social, psychological, behavioural, pathophysiological and environmental ones. Previous meta-analysis conducted in 2017 and 2018 showed that short sleep is associated with obesity and adverse changes in body mass index (BMI) in infants, children and adolescents. 5,6 The aims of this study were, firstly, to conduct an updated systematic review and meta-analysis of sleep and obesity exclusively in pre-school children, and, secondly, to conduct a novel systematic review and meta-analysis of intervention studies in which a sleep component had been included. The aim of the latter study was to determine whether sleep improvements are associated with beneficial effects on weight gain.

| Search strategy and selection criteria
Systematic searches using the Preferred Reporting Items for Systematic Reviews and Meta-analyses guidelines (PRISMA) 7   with 'Study type terms (prospective OR cohort OR longitudinal) were used. The articles identified by the searches were reviewed along with any relevant references cited within them. There were no language restrictions.

| Included studies
These had to fulfil the following criteria: (a) original published article, (b) study in preschool children (baseline mean age ≤5 years), (c) observational prospective design or intervention trials, (d) assessment of duration of sleep quantity as baseline exposure (aim 1) or sleep modification as intervention (aim 2), (e) follow-up of ≥1 year for incident outcomes and (f) one of the following outcomes: (i) incident cases of overweight and/or obesity (ii) prospective changes in BMI zscore or (iii) changes in BMI.

| Excluded studies
Studies excluded were (a) case-control by design, (b) cross-sectional reports, (c) had only a meeting abstract or unpublished material available or (d) included individuals with sleep disordered breathing.
Where multiple reports from the same study have been published, we used those with the most detail and/or longest follow-up.
Authors were contacted for raw data when not readily available from published reports.  Figure 1). One hundred and thirty studies proceeded to full-text evaluation from which we identified 28 prospective observational studies that had data suitable for meta-analysis (13 for overweight/obesity, 11 for BMI z-score and 10 for BMI). As per previous studies, relevant data were extracted and tabulated. 6 Five intervention studies were identified that were suitable for the second analysis.

| Exposure
Sleep in preschool children is different from that of adults. 8 The definition of 'short sleep' varied between studies and was defined by age as stated in Table 1. Risk ratios for short sleepers were compared to both middle and long sleepers or to a reference category.

| Outcome
Unless stated otherwise in Table 1, preschool children with obesity were defined either by a BMI > 95th and children who were overweight as >85th percentile according to local national growth charts or by international growth charts (see previous study for more detail).
BMI z-scores also called BMI standard deviation (s.d.) scores, and changes in BMI, were also used for outcome data.

| Confounders
Each study adjusted for different confounders as shown in Table 1.
The most adjusted model was used for the meta-analysis after consideration of possible causal pathways. Country, gender and sample size were collected and used in stratified analyses of heterogeneity, publication bias and sensitivity.

| Intervention studies
The five studies included in the systematic review of intervention studies are described in Table 2. Of these, four studies were suitable for meta-analysis. Three for change in BMI and three for change in BMI z-score following intervention.

| Statistical analysis
The quality of the studies included was evaluated by the Downs and Black Quality Index score system, with a maximum score of 20 for nonrandomized studies. 42 We calculated an estimate of relative risk (risk ratio RR) from either odds ratios (OR) or hazard ratios (HR) with 95% confidence intervals or regression coefficient β (95% CIs) for changes in BMI z-score or BMI as a continuous outcome so as to estimate the quantitative relation between short sleep duration and overweight or obesity.
Data were requested from authors if necessary. If required, the SE of either the RR or β was calculated algebraically from the 95% CIs. A random effect model weighted by the inverse of the variance was used. 43 By comparison with the reference category to estimate, the pooled risk and 95% CI of risk of development of overweight/obesity or the mean difference in BMI z-score or BMI with each additional hour of sleep.
Heterogeneity between studies was tested by Q-statistic and quantified by H-statistic and I 2 -statistic. 44 Publication bias was detected using funnel plot asymmetry and quantified using the Egger's regression test 45,46 where appropriate. Publication bias was corrected by the 'Trim and fill' method. 47

| Characteristics of study cohorts
After exclusion of ineligible studies (Table S1 for details of the studies and the reason for exclusion), we identified a total of 1719 studies from the searches (Figure 1.

| Incidence of overweight and/or obesity in short sleepers
The relationship between sleep and preschool children who were overweight and/or obesity is shown in Figure 2A.

| Short sleep and BMI Z-score
We

| Short sleep and BMI
Similarly, we observed a significant negative relationship between Body mass index, BMI z-scores. Obesity as defined as BMI ≥ 95th percentile.
Sleep duration measured by 24-h accelerometry data Significant intervention effect was observed for BMI at 5 y Significant intervention effect for BMI z-score at 5 y Protective effect for obesity among those receiving the 'sleep intervention' v control: Relative risk, 0.82 (95% CI, 0.36-1.84) at 5 y.
Note. Description of interventions studies that included a sleep component and obesity-related outcomes.
F I G U R E 2 Forest plots of meta-analysis of results from prospective studies looking at sleep and measures of overweight or obesity. Forest plot of prospective observational studies on the effect of short sleep duration on the incidence of overweight and/or obesity in preschool children. Results are relative risks (and 95% CI). Forest plot of prospective observational studies on the effect of sleep duration on the change in BMI z-score over time in preschool children. Results are mean BMI z-score difference per an additional hour of sleep (and 95% CI). Forest plot of prospective observational studies on the effect of sleep duration on the change in BMI (kg/m 2 ) over time in pre-school children. Results are mean BMI difference per an additional hour of sleep (and 95% CI) and evidence of publication bias (Eggers test p < 0.001) with six possible missing studies identified by 'trim and fill'.

| Short sleepers with overweight and/or obesity
Subgroup analysis by continent yielded risk ratios ranging from 1.50 to 1.77 across the four subgroups, each with statistically significant results. Heterogeneity ranged from 30% to 73% (Table S2A).

| Short sleepers and BMI Z-score
Subgroup analysis by continent yielded risk ratios ranging from −0.01 to −0.087 across three subgroups with the exception of the Australia and Oceanic group, each with statistically significant results. Heterogeneity ranged from 0% to 67% (Table S2B).

| Short sleepers and BMI
Subgroup analysis by continent yields risk ratios ranging from −0.03 to −0.39 across three subgroups, which apart from both the Australia and Oceania and East Asian groups were statistically significant. Heterogeneity estimates ranged from 38% to 83% (Table S2C).
3.6 | Sensitivity analysis 3.6.1 | Short sleepers with overweight and/or obesity For the sensitivity analysis, we deleted one study at a time. The heterogeneity was reduced to 33% (p = 0.12) by the removal of the study of Bolijn et al. 24 The removal of this study yielded risk ratios that were still statistically significant (1.59 [1.42 to 1.79], n = 41,220; p < 0.001).

| Short sleep and BMI Z-score
The removal of Snell (younger) 11 reduced the heterogeneity to 32% (p = 0.16) with the effect being highly significant (p < 0.001). The removal of Bolijn, Snell (younger) and Taveras 11,12,24 reduced the heterogeneity to zero (p = 0.68) with a highly significant overall effect (p < 0.001).

| Intervention study analysis
Five intervention studies with different strategies to improve sleep duration and quality and the effect on weight-related prospective outcomes were published between 2011 and 2018. All contained boys and girls. Three were from the United States and one each from Canada and New Zealand with sample sizes ranging from 48 to 1461 (see Table 2). The individual results and meta-analyses are shown in  37 However, these interventions were not effective on their own ( Figure 3A). There was no evidence of publication bias.
The second panel shows the results from the three studies 38,39,41 which looked at change in BMI following interventions that aimed to improve sleep. There was a significant overall effect of −0.27 (−0.50 to −0.03), p = 0.03, with no evidence of heterogeneity between studies or publication bias ( Figure 3B).
In the two intervention studies (three cohorts) 40,41 reporting changes in BMI z-scores, there was a favourable effect of the interventions of −0.14 (−0.36 to 0.08), p = 0.006, with no evidence of heterogeneity ( Figure 3C) or publication bias.
In the study by Taylor et al, 37 when looking at the effect of sleep compared to control on overweight or obesity outcomes, a favourable outcome was observed ( Figure 3D).

| Sleep and measures of obesity in preschool children: Strengths and limitations
The findings from this study are consistent with our previous study in infants, children and adolescents 6 and demonstrate that the effect of sleep duration is large, even when restricted to pre-school children.
They confirm that sleep duration is an important predictor of measures of obesity and that the findings are consistent in different populations. The observed significant heterogeneity between studies appeared to be due to only one study. 24 There was no evidence of publication bias. The results from the categorical studies were corroborated by the meta-analysis of regression coefficients looking at both changes in BMI z-score and BMI. In these, longer sleep duration in children was associated with a reduced age-related weight gain. For short sleep duration and BMI z-scores following removal of three studies, 11,12,24 the overall effect was still significant (p < 0.001), and the heterogeneity was reduced to zero. There was no appreciable effect on the results when studies with quality scores <15 out of a maximum of 20 were removed (Table S3).
The study has some limitations. The results are only representative of included studies. The quality of the data within the study is limited by the quality of the individual studies.
There are inherent limitations associated with the measurement of the variables of interest. Twenty six of the 28 studies used parental report to record the duration of sleep of the children, which does require that the parent has a good understanding of their child's sleep schedule. One study used both parental report and an accelerometer, 15 the other two studies used accelerometers. 27,33 This review is also limited to sleep duration as very few studies have examined other dimensions of sleep, which include sleep quality, sleep efficiency and bed/wake times in relation to obesity. It is F I G U R E 3 Forest plot of the results of five intervention trials of different strategies to improve sleep duration and quality and the effect on weight-related prospective outcomes. This is a 2 × 2 factorial RCT of intervention with both solids and soothe and sleep group (n = 22), or Solids (n = 29) or both approaches together (n = 29) compared to a control group (n = 30). The follow-up was to 1 year with outcome of weight for length percentile. (A) These three intervention trials share the same outcome (change in BMI) at the end of follow-up. Haines used a promotion to promote four family routines (family meals, adequate sleep, limiting TV, and removing TV from the child's bedroom) with 6-month follow-up. Walton used an intervention, which focused on lifestyle behaviours with a 9-month follow-up. Taylor used food, activity and breastfeeding (FAB) and sleep interventions. This analysis compares the sleep intervention vs control group with a 5-year follow-up. These two trials have both looked at change in BMI z-score. Woo Baidal is a pretreatment and posttreatment trial using woman, infants, and children (WIC) nutrition and sleep routines intervention to improve BMI z-score over 2 years. Taylor has looked at sleep intervention vs Control with a 5-year follow-up. The graph below shows the protective effect of sleep compared to control on overweight or obesity outcomes therefore important that future studies include these additional dimensions. 48 There  38 The advice given to families as part of the interventions also varied.
One study had a short 6-month follow-up and high attrition rate especially in the combined treatment group and was limited to firsttime mothers. 37 Haines et al 38  One of the studies showing significant results found success in using mobile technology to send weekly texts on the desired household routine (e.g., sleep routine). 38 The study also investigated the to 10-13 h in preschool children (3-5 years). 53 Although it is possible that early-life growth patterns may be associated with childhood and adult obesity, it is not clear whether 'an hour' of sleep reduction or extension effects children from birth to pre-school age in the same way.
There are several plausible mechanisms by which insufficient or poor sleep may lead to obesity in children. 6 These include changes in appetite control and, in particular, the hormones leptin and ghrelin, 54,55 changes in factors which would affect inflammation and metabolism, including the metabolism of insulin, glucose, cortisol, growth hormone and thyroid stimulating hormone. [56][57][58][59][60][61] Sleeping less would give more time to eat energy dense foods and to engage in other sedentary activities such as increasing their screen time as opposed to undertaking physical activity. 56 In a recent study, in parents and preschool children aged 3-5 years, parental screen time was positively associated with child screen time. Furthermore, after controlling for household income, parental occupation, and parental BMI, greater child screen time on weekends, was associated with higher child BMI percentile. 62 Given the observed interaction with ethnicity, 40 the effect of ethnicity also needs to be explored in future studies. Insufficient sleep is associated with alterations in mood, attention, impulse control, motivation and judgment; although all of these factors could potentially influence eating behaviours, energy intake and ultimately BMI in children, it is not clear if they would be as important in pre-school children, where a large part of their food intake is being governed by parents and carers. 63,64

| CONCLUSIONS
These two systematic reviews and meta-analyses show that short sleep duration is prospectively associated with measures of obesity and that interventions to improve sleep may lead to a reduced weight gain in preschool children. Further studies are required to determine whether specifically targeting obesity prevention programmes to include sleep interventions may be effective in reducing the incidence of overweight and obesity in preschool children.