Objective Short sleep duration may contribute to childhood obesity. Amenable to intervention, sleep thus provides a potential path for prevention. The authors aimed to determine the impact of a behavioural intervention that successfully reduced parent-reported infant sleep problems on adiposity at age 6.
Design 5-year follow-up of a previously reported population-based cluster randomised trial. Participant allocation was concealed to researchers and data collection blinded.
Setting Recruitment from well-child centres in Melbourne, Australia.
Participants 328 children (174 intervention) with parent-reported sleep problems at age 7–8 months drawn from 49 centres (clusters).
Intervention Behavioural sleep strategies delivered over one to three structured individual nurse consultations from 8 to 10 months, versus usual care.
Main outcomes at age 6 years Body mass index (BMI) z-score, percentage overweight/obese and waist circumference.
Analyses Intention-to-treat regression analyses adjusted for potential confounders.
Results Anthropometric data were available for 193 children (59% retention) at age 6. There was no evidence of a difference between intervention (N=101) and control (N=92) children for BMI z-score (adjusted mean difference 0.2, 95% CI −0.1 to 0.4), overweight/obese status (20% vs 17%; adjusted OR 1.4, 95% CI 0.7 to 2.8) and waist circumference (adjusted mean difference −0.3, 95% CI −1.6 to 1.1). In posthoc analyses, neither infant nor childhood sleep duration were associated with anthropometric outcomes.
Conclusions A brief infant sleep intervention did not reduce overweight/obesity at 6 years. Population-based primary care sleep services seem unlikely to reduce the early childhood obesity epidemic.
Trial registration ISRCTN48752250
Statistics from Altmetric.com
Around a quarter of Australian 5–17 year olds are overweight or obese.1 Interventions targeting childhood diet and physical activity levels have achieved very limited population success.2 Therefore, attention is broadening to factors associated with obesity that may prove more amenable to modification.3
One topical factor is sleep. Secular trends towards rising childhood obesity have been parallelled by a steady decline in child sleep duration over the last 35 years.4 5 Many studies have shown weak-to-moderate associations between habitually short sleep and body mass index (BMI)/waist circumference.6,–,14 A 2008 systematic review of 17 observational studies yielded a pooled odds ratio of 1.6 (95% CI 1.3 to 2.0) for childhood overweight/obesity presenting with short sleep.15
What is already known on this topic
▶ New approaches are urgently needed to halt the childhood obesity epidemic, because efforts to improve activity and nutrition have had very limited success.
▶ Some studies have reported that short sleep duration is associated with higher childhood body mass index. As sleep problems are amenable to intervention, this has fuelled interest in infant sleep interventions to prevent obesity.
What this study adds
▶ A population-based intervention targeting infants with parent-reported sleep problems improved infant sleep problems, but did not prevent obesity at age 6.
▶ It is premature to assume that sleep interventions will prevent or treat obesity in very young children.
Chen's 2008 systematic review15 proposes a number of plausible mechanisms. Poor sleep could lead to higher BMI, for example, by: (1) fatigue, reducing subsequent physical activity; (2) increased evening and overnight intake of calorie-dense foods; (3) lower nocturnal leptin and higher ghrelin, thus increasing daytime appetite; (4) lower nocturnal growth hormone, decreasing lypolysis.15 Under the reverse hypothesis, obesity could contribute to shorter sleep via such mechanisms as obstructive sleep apnoea16 or lower activity levels leading to less fatigue and thence difficulty falling asleep. Currently, however, no single mechanism is endorsed or causal direction established because most studies are cross-sectional.17 Of the few longitudinal studies suggesting that short sleep precedes obesity,8 11 18 none focuses on infancy.
Despite this lack of robust evidence, the idea that poor sleep may play a causal role in childhood obesity is generating intense interest.17 19 Randomised trials could confirm causal associations if they successfully manipulate the predictor then measure its effect on overweight/obesity.20 Therefore, at least two ongoing trials aiming to lessen weight gain are incorporating infant sleep interventions that specifically target infant sleep problems and sleep duration (NCT00892983, New Zealand, and NCT00359242, Pennsylvania, USA). Regardless of their outcomes, anecdotally sleep is already being targeted clinically to reduce obesity.
In 2003–2005, we conducted a large, community-based, secondary prevention trial of a behavioural sleep intervention designed to improve infant sleep at 8–10 months of age. Drawing from a population-based sample recruited at age 4 months, all infants with parent-reported sleep problems at 7–8 months were eligible (see figure 1). Compared with controls, intervention parents reported fewer sleep problems at infant age 10 months (56% vs 68%, adjusted OR (adj OR) 0.6, 95% CI 0.4 to 0.9) and 12 months (39% vs 55%, adj OR 0.50, 95% CI 0.3 to 0.8).21 By age 2 years, the impact on sleep was less (27% vs 33%, adj OR 0.8, 95% CI 0.5 to 1.4) but a reduction in maternal depression was still apparent (15% vs 26%, adj OR 0.4, 95% CI 0.2 to 0.9).22 Because these benefits occurred across a broad sociodemographic range and at lower healthcare system cost, the government recently extended the programme to the entire state of Victoria.
In this paper, we aimed to determine whether, by age 6 years:
this successful infant sleep intervention led to better anthropometric outcomes, ie:
lower mean BMI z-score
a smaller proportion of children overweight/obese
lower mean waist circumference
any observed intervention effects were mediated by parent-reported child sleep problems (yes/no) and/or sleep duration from baseline to 6 years
Design and setting
The Kids Sleep Study is a 5-year follow-up of the Infant Sleep Study, a randomised controlled trial (ISRCTN48752250) of a behavioural sleep intervention delivered at child age 8–10 months. Methods were previously reported for outcomes at ages 1221 and 2422 months. Briefly, we aimed to recruit all mothers with children born in June to July 2003 who attended the free, universal, 4-month well-child check with their maternal and child health nurse in six sociodemographically diverse local government areas. At 7–8 months, mothers reporting an infant sleep problem (see Predictors, below) were eligible for the trial. We excluded infants born before 32 weeks gestation and mothers with insufficient English to complete questionnaires.
Following baseline recruitment, we cluster randomised (described elsewhere)21 22 the 49 maternal and child health nurse centres which, in turn, determined the participant allocation. Participant allocation was concealed to researchers, and data collection and entry were blinded; however, the nurses and parents could not be blinded to their group allocation.
Intervention maternal and child health nurses were trained to deliver a brief, behavioural sleep intervention at the routine 8-month well-child check to mothers reporting infant sleep problems (figure 1).21 Control families received usual care.
Three hundred and twenty-six of the 328 6-year-old children from the Infant Sleep Study21 22 were eligible; the remaining two had been diagnosed with intellectual disability and autism (prespecified exclusion criteria) since the last follow-up at age 2 years (figure 2).
From April to October 2009, we re-contacted all families. Parents who returned written informed consent were mailed a questionnaire and phoned to arrange a home-based assessment as close as practicable to the child's sixth birthday.
The primary outcomes were BMI (kg/m2) and BMI status, and the secondary outcome was waist circumference (cm). Trained researchers measured weight to the nearest 100 g using digital scales (Model THD-646; Tanita, Tokyo, Japan), and height and waist circumference (both measured twice) to the nearest 0.1 cm using a portable rigid stadiometer (Model IPO955; Invicta, Leicester, UK) and Lufkin Executive Steel Tapes (W606PM), respectively. If height measurements differed by ≥0.5 cm or waist measurements by ≥1.0 cm, a third measurement was taken and the mean of the closest two used. Six parents (three intervention, three control) declined a home visit but agreed to collect anthropometric data, so were mailed a plastic tailor's tape and standardised instructions mirroring the researchers' protocol.
BMI z-scores were calculated using the US Centers for Disease Control (2000) sex- and age-specific charts,23 and BMI status classified as not overweight/overweight/obese according to International Obesity Task Force criteria.24 25
At ages 4, 7, 10, 12 and 24 months, parents answered the question ‘Over the last 2 weeks, has your baby's sleep generally been a problem for you?’ (yes/no), which was also the trial's entry criterion at age 7–8 months.21 22 26 If yes, they proceeded to a 7-point severity rating (‘hardly any problem at all’ to ‘very severe problem’), which we subsequently trichotomised to mild (1–2), moderate (3–4) or severe (5–7). At age 6 years, as used in the Longitudinal Study of Australian Children27 and elsewhere,28 parents completed a single 4-point severity rating (none/mild/moderate/severe), subsequently dichotomised into sleep problem (moderate/severe) versus no sleep problem (none/mild). At 7, 10, 12 months and 6 years, parents also reported the child's usual bedtime, time taken to fall asleep, morning wake time, number of night wakings and length of wakings. We calculated usual night-time sleep duration as the time between sleep and wake times. At 6 years, parents reported these separately for school and non-school days, so we used a weighted average (5/7 of the school day data plus 2/7 of the non-school day data).
Child's sex, mother's education (did not complete high school, completed high school, tertiary degree), family language other than English (yes/no) and Socio-Economic Indexes for Areas (SEIFA)29 Index of Relative Disadvantage were all recorded at baseline. SEIFA is a national index derived from census data for all individuals living in a postcode, with a median (IQR) for Australia of 1000 (962–1044) and for Victoria of 1025 (989–1058).30
The original Infant Sleep Study was powered (estimating that 132 children and 12 maternal and child health clusters are required in each trial arm) to detect a difference of 20% (70% and 50% for control and intervention groups, respectively) between the proportions of mothers reporting infant sleep problems at each of the 10- and 12-month follow-ups with 80% power at the 5% level of significance, taking account of the clustered design and assuming an expected cluster size of 11 and an intracluster correlation coefficient of 0.02.
We compared trial arms by fitting random effects linear regression models estimated using maximum likelihood31 for the quantitative outcomes, BMI z-score and waist circumference. For the binary overweight/obese status outcome, marginal logistic regression models were fitted using generalised estimating equations (GEEs), assuming an exchangeable correlation with information sandwich (‘robust’) estimates of standard error.32 Both the random effects model and the GEE method allow for correlation between outcomes of participants from the same cluster. Unadjusted analyses and analyses adjusted for potential prognostic factors were conducted. We applied the intention-to-treat principle as far as possible, given missing data, and analysed participants according to their randomisation status. Intracluster correlation coefficients are reported for the outcomes in keeping with the recommendations of the CONSORT statement for cluster randomised trials.
Aim 2 did not proceed due to the negative results reported below for Aim 1. Instead, in posthoc analyses we explored the relationship between sleep duration at 7, 10, 12 months and 6 years and the adiposity outcomes at 6 years. Data were reshaped (up to four records – one for each sleep measurement – per participant) so that the association between sleep duration across 7 months to 6 years (a single predictor) and adiposity (outcome) was estimated in a longitudinal analysis. We used random effects linear regression (continuous outcomes) and random effects logistic regression (binary outcome) to allow for correlation between records from the same participant across time. The wave at which sleep was measured was adjusted for in this analysis. Tests of interaction provided little evidence that the timing of sleep measurement modified the relationship between sleep and adiposity, so there was no need to stratify this analysis by study wave. We used Stata 11.1 software for all analyses.
The original trial (23067B) and 6-year-old follow-up (28137F) were approved by the Human Research Ethics Committee of The Royal Children's Hospital, Melbourne.
Figure 2 shows the participant flow: 225 families (69% of the eligible 326) participated at age 6 years, and 193 (59%) provided both child physical measurements and sleep data. We were unable to contact 49 families (15%), and 52 (16%) families declined for reasons including ‘too busy’ (n=26), ‘not interested’ (n=6), ‘personal reasons’ (n=6), child illness (n=1) or no reason (n=13).
Table 1 shows the sample characteristics. In the control arm, children of mothers with a tertiary degree were over-represented and children from very disadvantaged backgrounds were under-represented among those who were retained compared to those lost to follow-up, and children of families who spoke a language other than English at home were under-represented in both arms among those who were retained. Follow-up occurred at a mean age of 6.01 years (SD 2.04 months).
Child sleep variables
During infancy, intervention parents reported fewer child sleep problems, less severe sleep problems, fewer night-time wakings and shorter waking periods than control parents (table 2). Sleep duration remained similar in the two groups throughout. At the 6-year-old follow-up, intervention and control groups were similar across all sleep variables.
The intervention and control groups were similar at age 6 years with respect to BMI z-score and waist circumference (table 3). Regarding BMI z-score, the lower 95% CI bound of −0.1 indicates sufficient precision to exclude any important benefit of the intervention (table 3). Intervention children were slightly more likely to be overweight/obese at outcome (OR 1.4, 95% CI 0.7 to 2.8), although here the lower bound of the 95% CI suggests we cannot rule out a small but important benefit.
As no intervention effects were detected, we were unable to address our stated second aim, and proceeded to posthoc analyses to investigate whether lack of an intervention effect was due to lack of association between sleep and adiposity. These did not support an association between childhood sleep duration and any of BMI z-score (mean increase per 1 h increase in sleep time = 0.001 (95% CI −0.01 to 0.01), p=1.0), overweight/obese status (OR for 1 h increase in sleep time = 0.9 (95% CI 0.3 to 2.4), p=0.8) or waist circumference (mean increase per 1 h increase in sleep time = −0.001 (95% CI −0.07 to 0.07), p=1.0) at age 6.
A population-based intervention that reduced parent-reported sleep problems during infancy neither prevented overweight/obesity, nor reduced BMI or waist circumference, at age 6 years. Thus, while there are good mental health reasons to support community-based efforts to tackle infant sleep problems, our findings should temper hopes that such efforts will also address the childhood obesity epidemic.
Strengths of the trial
To our knowledge, this is the first randomised trial (the gold standard for assessing causality20) to investigate whether an early sleep intervention has lasting effects on children's BMI and waist circumference. Its population-based sampling should mean that our findings are generalisable to English-speaking families across a wide demographic range. In posthoc longitudinal analyses, we examined the impact of sleep duration over multiple time points on overweight/obesity. It has been proposed that effects of short sleep on obesity accumulate over a long time frame.33 Our data suggest that, if so, they commence after the infancy and preschool years.
Parent report somewhat overestimates sleep duration compared with sleep diaries,34 35 actigraphy35 and polysomnography34 in preschool and primary school age children. Although this might weaken true associations between measured sleep and obesity, it should not affect our main findings about relationships between the intervention and obesity. Furthermore, systematic overestimation should not alter the strength of the reported associations.
More important is the 41% loss to follow-up of the original sample, potentially diminishing power to demonstrate true effects. However, our conclusion that the intervention did not benefit BMI z-score was statistically robust and was supported by the tendency to a higher proportion of overweight/obesity in the intervention group (although here the lower bound of the 95% CI could not rule out a small benefit). Loss to follow-up can also introduce internal bias and reduce generalisability. Regarding bias, the retained intervention and control participants seemed fairly balanced (table 1) but, regarding generalisability, non-English-speaking and disadvantaged families were over-represented in those lost to follow-up. For the intervention to differentially benefit the BMI of these subgroups and alter our conclusions, however, the effect would have to be very large.
Interpretation in light of other studies
The infant sleep programme did not reduce sleep problems at age 6 years, perhaps because virtually all sleep problems had already resolved in both groups by this age. Our data suggest that its early benefits reflected improvements in sleep organisation rather than duration.
It is more noteworthy that sleep duration at 7, 10, 12 months and 6 years did not predict BMI at age 6 years. This was unexpected, given the literature cited in the Introduction, and reaffirms the importance of confirming observational associations before changing clinical practice or public policy. These findings are consistent with our subsequent analysis of a much larger national study of Australian children examining longitudinal associations of short sleep with overweight/obesity between ages 0 and 7 years (Hiscock et al, submitted). They differ, however, from most published studies, possibly reflecting their cross-sectional nature and/or positive publication bias, as well as the population nature of our sample and young participant age. Most studies linking short sleep to obesity did not assess infant samples,6 7 9 10 13 14 despite the very high prevalence of sleep problems at this age.36 37 The lack of consistency in ‘short sleep’ definitions and rationale further hampers comparisons. To avoid this problem, we treated sleep duration as a continuous variable in analyses.
Unanswered questions and future research
As current longitudinal studies follow ‘short sleepers’ throughout childhood and adolescence, it will become possible to investigate the impacts of early sleep on later overweight/obesity. Ideally, these would include more objective measures of early sleep duration than was feasible for this community-based infant sleep study.
This intervention achieved all of its original aims (better infant sleep and lower maternal depression and societal costs) by improving the organisation, rather than overall duration, of infant sleep. Although it remains possible that a more intensive or longer-lasting intervention might increase infant sleep duration and reduce subsequent BMI gain, it seems unlikely given that infant sleep duration did not predict any subsequent anthropometric outcome. Furthermore, implementing even the limited sleep intervention described here to an entire state was a major logistic exercise. Delivering a more substantial intervention would require very careful consideration of population costs and feasibility.
Thus, until a relationship between sleep and obesity is clearly demonstrated, quantified and shown to be modifiable, we conclude that infant sleep intervention should not be considered to be an effective means of preventing obesity in young children.
The authors thank all the parents and children who took part in the Infant and Kids Sleep Studies, and the Maternal and Child Health nurses from the cities of Bayside, Darebin, Hobson's Bay, Manningham, Monash and the Shire of Yarra Ranges who helped recruit and deliver the intervention in the Infant Sleep Study.
Funding The Infant Sleep Study was funded by the Australian National Health and Medical Research Council (NHMRC) Project Grant 237120 and the Pratt Foundation, and the follow-up Kids Sleep Study by the Foundation for Children (Project Grant 180 2009). The authors' work was independent of the funders (the funding source had no involvement). MW was supported by NHMRC Population Health Career Development Awards #284556 and #546405, AP by a Melbourne Research Scholarship (The University of Melbourne), SC by the Foundation for Children Grant, and OCU's and HH's postdoctoral positions by NHMRC Population Health Capacity Building Grant 436914.
Competing interests None.
Ethical approval Project approval was obtained from Human Research Ethics Committee of The Royal Children's Hospital, Melbourne, for the Infant Sleep Study (23067B) and 6 year Kids Sleep Study follow-up (28137F).
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.