To examine independent and combined cross-sectional associations between movement behaviors (physical activity (PA), sedentary time, sleep duration, screen time and sleep disturbance) and fat mass index (FMI), as well as to examine longitudinal associations between movement behaviors and FMI.
Cross-sectional and longitudinal analyses were done using data from the OPUS school meal study on 785 children (52% boys, 13.4% overweight, ages 8–11 years). Total PA, moderate-to-vigorous PA (MVPA), sedentary time and sleep duration (7 days and 8 nights) were assessed by an accelerometer and FMI was determined by dual-energy X-ray absorptiometry (DXA) on three occasions over 200 days. Demographic characteristics, screen time and sleep disturbance (Children’s Sleep Habits Questionnaire) were also obtained.
Total PA, MVPA and sleep duration were negatively associated with FMI, while sedentary time and sleep disturbances were positively associated with FMI (P0.01). However, only total PA, MVPA and sleep duration were independently associated with FMI after adjustment for multiple covariates (P<0.001). Nevertheless, combined associations revealed synergistic effects among the different movement behaviors. Changes over time in MVPA were negatively associated with changes in FMI (P<0.001). However, none of the movement behaviors at baseline predicted changes in FMI (P>0.05), but higher FMI at baseline predicted a decrease in total PA and MVPA, and an increase in sedentary time (P0.001), even in normal-weight children (P0.03).
Total PA, MVPA and sleep duration were independently associated with FMI, and combined associations of movement behaviors showed a synergistic effect with FMI. In the longitudinal study design, a high FMI at baseline was associated with lower PA and higher sedentary time after 200 days but not vice versa, even in normal-weight children. Our results suggest that adiposity is a better predictor of PA and sedentary behavior changes than the other way around.
Several movement behaviors involving various aspects of physical activity (PA), sedentary behavior and sleep have been linked to the recent development of overweight and obesity among children in many parts of the world.1, 2, 3, 4, 5 These potential risk factors for increased adiposity have been investigated in several studies; however, existing evidence from accelerometers are primarily obtained from cross-sectional studies. Most of these studies investigated PA, sedentary behavior or sleep in isolation with suboptimal adiposity indicators (for example, body mass index (BMI) or skinfold thickness) and failed to adjust for important covariates (for example, diet). Furthermore, the independent as well as combined contribution of these different risk factors for adiposity is largely unknown.
As BMI in 10-year-old children consists of approximately three-quarters fat-free mass and one-quarter fat mass, changes in fat mass could be partly overlooked when only BMI is examined. Among longitudinal studies of children and adolescents using accelerometer-assessed movement behavior and adiposity determined by dual-energy X-ray absorptiometry (DXA), we identified only three studies with PA,1, 6, 7 one with sedentary time6 and one with sleep duration;4 however, no study to date has investigated all three movement behaviors at the same time. In addition, early puberty is a period characterized by rapid changes in body composition and movement behaviors, which stress the need to explore the association between these, as they may be important for successful prevention of early fat accumulation in children.
The aim of this study was to examine independent and combined cross-sectional associations between movement behaviors (PA, sedentary time, sleep duration, screen time and sleep disturbance) and fat mass index (FMI), and to further identify longitudinal associations between changes in movement behaviors and changes in FMI, as well as movement behaviors from baseline as predictors of change in FMI and vice versa.
Materials and methods
The initial sample comprised 834 of the 1021 invited third- and fourth-grade students (8–11 years old) from nine Danish municipal schools enrolled in the OPUS (Optimal well-being, development and health for Danish children through a healthy New Nordic Diet) school meal study. The main aim of the OPUS project was to investigate the health effects of a New Nordic Diet served at school.8 It was a cluster-randomized cross-over study with a number of measurements performed at baseline (August to November 2011), before the end of the first dietary period approximately 100 days later and before the end of the second dietary period after another 100 days. Owing to missing data (no body composition data at baseline (n=22), no body composition data during the last visit (n=77) and no measure of pubertal status (n=27)), our cross-sectional and longitudinal sample comprised data from 785 and 708 children, respectively.
Questionnaire data were first collected, followed by simultaneous registrations of movement behaviors and dietary intake within the following 2 weeks, after which anthropometric measurements were collected the following week. The study was approved by the Committees on Biomedical Research Ethics for the Capital Region of Denmark (J.nr. H-1-2010-123). Child assent and written informed parental consent of both custody holders were obtained for all participants. The study was registered in the database www.clinicaltrials.gov (no. NCT01457794).
The children were asked to wear an ActiGraph tri-axis accelerometer monitor (GT3X+ or GT3X, Pensacola, FL, USA) tightly on the right hip using an elastic belt for 7 consecutive days and 8 nights (entire 24-h period), and to remove it only during water activities (that is, showering or swimming). At the end of the observation period, data were reintegrated to 60-s epochs and analyzed using ActiLife6 (ActiGraph, Pensacola, FL, USA). Before analysis of PA and sedentary time, we removed (1) data between midnight and 0600 hours as this was expected to be non-awake time, (2) periods of at least 15 min of consecutive zero counts using tri-axial vector magnitudes to remove non-wear time and non-awake time and (3) consecutive wear time periods of <60 min to remove non-awake time as sleep for most children is characterized by minor periods of movement that we did not want to include in our analysis of PA and sedentary time. Total PA (counts per min (cpm)) was expressed as total vertical counts from monitor wear time, divided by monitor wear time. As a secondary variable, total tri-axial PA (cpm) was expressed as a vector magnitude of the total tri-axial counts from monitor wear time, divided by monitor wear time. Time spent in a sedentary state was defined as all minutes showing 100 vertical cpm or less, which is a widely used cutoff point.9 Moderate-to-vigorous PA (MVPA) was defined as 2296 vertical cpm, which is a recently suggested pediatric cutoff point.9 The percentage of time spent in a sedentary state was calculated by dividing the sedentary time by monitor wear time and multiplying by 100. The weekly averages of total PA, MVPA and sedentary time were calculated in the proportion 5 to 2 between weekdays (Monday to Friday) and weekend days (Saturday and Sunday). Total PA, MVPA and sedentary time were only considered valid if monitor wear time was at least 10 h day–1 for a minimum of 3 weekdays and 1 weekend day. We reanalyzed a random subsample of 105 children where self-reported sleep was removed and non-wear time was defined as 60 min of consecutive zeros, allowing for 2 min of non-zero interruptions, and found total PA to be 28 cpm higher (P<0.001), MVPA to be 0.3 min lower (P=0.03) and sedentary time to be 1.6% lower (P<0.001). However, the two different approaches were very closely correlated (total PA (r2=0.98), MVPA (r2=1.00) and sedentary time (r2=0.92)), and did not change the associations with FMI.
The parents and children were instructed to keep logs for bedtime (‘lights off’ and trying to sleep) and waking time (‘lights on’) during the week in which the monitor was worn. To estimate accelerometer-determined sleep duration, self-reported bedtimes and waking times were used as the possible window of sleep and accelerometer data within this window were scored in ActiLife6 using the algorithm by Sadeh et al.10 Self-reported sleep logs were missing for 6% (113 out of 1795) of the sleep measurements used in this study; in these cases, sleep was scored visually from the individual actograms as the difference between time when activity stopped and time when activity resumed. In a random subsample of 105 individuals, we found that the mean difference in sleep duration between these two approaches was small (3.8 min, P<0.001). The weekly average of sleep duration was calculated in the proportion 5 to 2 between weekdays (Sunday to Thursday) and weekend days (Friday and Saturday). Sleep duration was only considered valid if it was measured for a minimum of 3 weekdays and 1 weekend day. On any of the three measurement occasions, 85% of individuals had valid activity and sleep registrations for a minimum of 7 days.
Daily food and beverage intake was recorded over 7 consecutive days using a Web-based Dietary Assessment Software for Children (WebDASC) tool that has been validated for fruits and vegetables.11 Dietary intake was recorded at the end of each day (no later than midnight the following day) using pictures of different portion sizes. Further description of the WebDASC is available elsewhere.12 The energy density of the diet, at baseline and changes during the 200 days, was a covariate in this study. This was calculated as energy in kilojoules (kJ) divided by the weight in grams (g) of solid food and liquids consumed as food (for example, soups and yoghurts). No individuals were excluded from the analytical sample because of a low (<1.05) or high (>2.29) reported energy intake, defined as energy intake divided by basal metabolic rate.13
The children were weighed to the nearest 0.1 kg (Tanita BWB-800S, Tokyo, Japan) while fasting, barefoot and wearing light clothes, and their heights were measured three times to the nearest 0.1 cm (average used) (CMS Weighing Equipment LTD, London, UK). The BMI Z-score was calculated based on the World Health Organization Growth Reference data from 2007.14 The prevalence of underweight, normal-weight, overweight and obese children was calculated based on age- and sex-specific cutoffs defined to pass through a BMI of 18.5, 25 and 30 kg m–2 at 18 years of age.15, 16 Finally, body fat was determined by DXA-scanning (Lunar Prodigy Pro, GE Medical Systems, Madison, WI, USA) using EnCore software version 13.5 (Encore, Madison, WI, USA), and FMI was calculated as fat mass divided by height squared. As FMI is relatively independent of fat-free mass17 it was chosen over, for example, percentage body fat.
A baseline questionnaire ascertained age, sex, school grade, highest education of the household (divided into four groups according to years of education: 10 years, 11–12 years, 13–16 years, 17 years), number of parents born in Denmark (a proxy for ethnicity) and screen time. Screen time was computed based on the parent-reported time that the child spent watching television, playing passive video games or using the computer for leisure activities on weekdays and weekend days. Time spent on playing the Nintendo Wii or similar active video game devices was not included in screen time. The weekly average of screen time was calculated in the proportion 5 to 2 between weekdays and weekend days. Pubertal status was self-reported (parent and child) based on breast development among girls and pubic hair growth among boys on a scale from 1 to 5.18 A dichotomous variable indicating whether or not the child had entered puberty was used in the statistical analyses (1 or 2). Finally, the parents were asked to fill out the 33-item Children’s Sleep Habits Questionnaire (CSHQ) that screens for common sleep disturbances. On a three-point scale, parents reported the frequency of their child’s habits. Items were summed, with higher scores suggesting the presence of sleep disturbances. The scale has previously demonstrated test–retest reliability and validity in school-aged children.19
Descriptive characteristics of the study sample were presented as means and s.d., median (interquartile range) or as proportions. Sex differences were assessed using two-sample t-tests (variables were logarithmically transformed if they were not normally distributed) or Pearson’s χ2 tests. As no intervention effect was found in movement behaviors, a linear mixed model with subject as a random factor (cross-sectional analysis) was used to test the association between movement behaviors (average of baseline, day 100 and day 200) and FMI on all three occasions (except for screen time and CSHQ, which were only obtained at baseline). These analyses were adjusted for a number of covariates (model 1: baseline age, sex, pubertal status, sex–pubertal status interaction and month of baseline measurement; model 2: as in model 1+number of parents born in Denmark and highest education of the household; model 3: as in model 2+energy density of the diet, MVPA, sedentary time, screen time, sleep duration and CSHQ (total PA was not adjusted for MVPA and sedentary time)). To assess whether the cross-sectional associations were synergistic or not, movement behaviors were divided into quartiles using the first and fourth quartiles as the reference group and risk group, and up to three variables were combined and shown as differences in absolute values of FMI using the same adjustments as in model 1. In the cross-sectional analyses, FMI was logarithmically transformed and, because of the rather small effect that was observed, the unstandardized regression coefficients (β) were back-transformed using the exponential function. Partial correlation coefficients (r) were calculated as average of r from baseline, day 100 and day 200, except for screen time and CSHQ, which were only measured at baseline.
Given that the order of intervention did not affect changes in movement behaviors during the 200-day period, a linear mixed model with school as a random factor was used to evaluate associations between changes in movement behavior and FMI during the 200-day period (longitudinal analysis). Movement behaviors at baseline were similarly used to predict changes in FMI during the 200 days and vice versa. These longitudinal analyses were adjusted for baseline age, sex, pubertal status, sex–pubertal status interaction, month of first measurement and days of follow-up. Furthermore, baseline FMI and baseline movement behaviors were included in all longitudinal analyses. The assessment of baseline FMI as a predictor of movement behavior was also done in normal-weight children only. Data are reported as r, β and 95% confidence intervals (CI). Linearity between residuals and the dependent variables in the model was visually checked using scatter plots, along with normal distribution and homogeneity of variance of the residuals. As no interaction with sex was observed in the longitudinal analyses (P0.17), data were not stratified according to sex. The level of significance was set at P<0.05 and statistical analyses were done using STATA/IC 11.2 (Houston, TX, USA).
The characteristics of children in the study are presented in Table 1. The majority (65.6%) had not entered puberty and 13.4% were categorized as overweight. Boys were more physically active than girls but engaged in more screen time. When adjusted for baseline age, sex, pubertal status and month of baseline measurement in the cross-sectional analyses, total PA, MVPA and sleep duration were negatively associated with FMI while sleep disturbances (CSHQ) were positively associated with FMI (P0.01) (Table 2). In an unadjusted model or after additionally adjusting for the number of parents born in Denmark and the highest education of the household, sedentary time became positively associated with FMI (P0.01); however, only total PA, MVPA and sleep duration were still associated with FMI after adjustment for all covariates (P<0.001). When combined cross-sectional associations of MVPA, sedentary time, screen time, sleep duration and CSHQ with FMI were assessed (Figure 1), FMI was found to be 3.44 units (CI: 2.39–4.49; P<0.001) higher among children in the lower quartile for MVPA and higher quartile for CSHQ and sedentary time, compared with children in the opposite quartiles.
In the longitudinal analyses, changes in MVPA were negatively associated with changes in FMI (P0.001; Table 3). However, the different movement behaviors at baseline did not predict changes in FMI (P0.11) (only a trend toward significance was observed for screen time; P=0.057). Instead, a high FMI at baseline was associated with lower total PA, MVPA and higher sedentary time after 200 days (P0.001). This was also the case when analyzed in the normal-weight children only (P0.03). The association of FMI with sedentary time was, however, not independent of MVPA (P=0.60). All significant findings remained significant in an unadjusted model and after further adjusting for the number of parents born in Denmark, highest education of the household, changes in height, changes in energy density of the diet and changes in fat-free mass index.
Total PA, MVPA and sleep duration were negatively associated with FMI, while sedentary time and sleep disturbances were positively associated with FMI in the cross-sectional analyses; however, only total PA, MVPA and sleep duration were independently associated with FMI after adjustment for covariates. Nevertheless, combined associations revealed substantial synergistic effects among MVPA, sedentary time, screen time, sleep duration and sleep disturbances, suggesting that these movement behaviors are inter-connected. In the longitudinal analysis, changes in MVPA were negatively associated with changes in FMI; however, as movement behaviors at baseline did not predict changes in FMI while FMI at baseline predicted changes in total PA, MVPA and sedentary time, adiposity may be a better predictor of PA and sedentary behavior changes than the other way around.
In general, cross-sectional studies in children using accelerometers have found inverse associations between total PA or MVPA with a broad range of adiposity measures (waist circumference, BMI, FMI and body fat).1, 20, 21, 22, 23, 24, 25, 26 We identified 10 prospective studies in children and adolescents (3–19 years of age, 1–8 years of follow-up time, n=94–2882) examining accelerometer-assessed total PA or MVPA as predictors of weight, BMI, waist circumference, skinfold thickness, percentage body fat, fat mass or FMI, two of which used DXA.1, 7 Three studies found that PA predicted adiposity in the whole study population,1, 27, 28 one only in normal-weight and not overweight individuals,29 one in white and not black girls30 and in the remaining five studies PA could not predict adiposity.7, 31, 32, 33, 34
Only two of these studies looked at the association between changes in PA and changes in adiposity. The ALSPAC study was the largest study and involved almost 3000 children. It found a negative association between changes in PA and changes in body fat for children between 12 and 14 years of age.1 Also, a recent study in overweight individuals that did not find PA to predict BMI Z-score found a negative association between changes in PA and changes in BMI Z-score over 3 years in 5- to 10-year-old children.34 This prospective negative association has also been observed between changes in vigorous PA and changes in fat mass from the age of 5 to 8 years.6 However, none of these studies tested the reverse association (BMI or fat mass as a predictor of change in PA) and could therefore not determine the dominant direction of the relationship. A recent meta-analysis of seven studies tested accelerometer-assessed MVPA as a predictor of waist circumference at follow-up and vice versa but found that neither one was significant.35 Another two studies in children have investigated this and found that percentage body fat at age 7 predicted changes in total PA and MVPA until age 10,7 and percentage body fat at age 8 predicted changes in the daily sum of accelerometer movement counts derived during MVPA at age 11.36 It should be noted that ∼25% of these children were overweight or identified as having high percentage body fat, respectively, and it could be speculated that the large amount of fat already present in these children may impact the results observed. Our study supports this controversial finding, but within a group of children that had a much lower prevalence of overweight and obesity. Of note, we also observed the same findings among children within a BMI range for normal weight when analyzed separately. To our knowledge, this has never been reported before.
MVPA is frequently reported to be more strongly associated with adiposity than total PA in both cross-sectional and longitudinal studies.1 In our study, we also found a higher partial correlation coefficient for MVPA than for total PA in both the cross-sectional and longitudinal analyses. Physiological explanations for this (for example, VPA stimulates post-exercise oxygen consumption and increases muscle mass) are dominant in the literature;1 however, as greater adiposity was found to predict lower MVPA and not the other way around, psychological explanations also appear to be plausible, as increasing adiposity could simply refrain children from engaging in MVPA relative to total PA. Finally, as children with higher FMI are expected to produce greater energy expenditure for a given amount of movement, a possible physiological explanation for the reverse causality could be fatigue and thereby lowering of especially MVPA.
Cross-sectional associations between accelerometer-assessed sedentary time and adiposity in children are often not present26, 35 or disappear after adjusting for MVPA,25 although a recent study reported a positive association between sedentary time and BMI that was independent of MVPA in 2506 Portuguese youth (10–18 years old).37 We identified four longitudinal studies that showed associations between accelerometer-assessed sedentary time and weight,32 BMI34, 38 or percentage body fat.6 One study found sedentary time to predict 1-year weight gain in 4- to 19-year-old children and adolescents, although this was not independent of baseline BMI status.32 Another study found sedentary time to be associated with greater increases in BMI at the 90th, 75th and 50th BMI percentiles between ages 9 and 15 years independent of MVPA.38 The remaining two studies did not find accelerometer-assessed sedentary time for 5- to 10-year-olds to predict percentage body fat or BMI Z-score (in overweight individuals) 3 years later. Despite this, television viewing was found to predict changes in percentage body fat.6 Relative to accelerometer-assessed sedentary time, screen time has been reported more frequently to be associated with adiposity39 and has been found to predict obesity incidence 5 years later,40 with a steeper BMI trajectory during early adolescence in white but not black girls.41 In this study, neither sedentary time nor screen time was independent of MVPA in the cross-sectional analysis, although the influences of sedentary time, screen time and MVPA were synergistic with FMI. In the longitudinal analysis, higher screen time at baseline tended to predict a positive change in FMI (P=0.057), and higher FMI at baseline predicted a positive change in sedentary time (although this was not independent of MVPA). FMI has been reported as a predictor of accelerometer-assessed sedentary time in adults with a mean BMI above 25 kg m–2 (ref. 42) and in a recent meta-analysis of seven studies, waist circumference predicted increased sedentary time at follow-up but not vice versa.35 Overall, our data indicate that sedentary time and screen time capture different parts of sedentary behavior, as they respond differently when adjusting for ethnicity and education in the cross-sectional analyses and also based on the different longitudinal findings.
Virtually, all cross-sectional studies in children and adolescents have consistently reported an inverse association between accelerometer-assessed sleep duration and adiposity.43, 44, 45, 46 However, we only identified two studies that used a longitudinal design. The FLAME study found short sleep duration between 3 and 5 years of age to predict higher FMI at age 7,4 while the other study did not find short sleep duration to predict weight gain after 1 year in 4- to 19-year-old children and adolescents.32 We detected a cross-sectional association between sleep duration and FMI that persisted after multiple adjustments, and found sleep duration to be synergistic with sleep disturbances and other movement behaviors. Although our cross-sectional association was stronger than the one found in the FLAME study, we did not find a longitudinal association between changes in sleep duration and changes in FMI. In addition, we did not find changes in sleep duration to predict changes in FMI as in the FLAME study4 or FMI at baseline to predict changes in sleep duration. The most obvious differences between our study and the FLAME study that could explain the different results obtained are our shorter follow-up time, our lower BMI Z-scores of the children (by approximately 0.75), the older age of our children and a more sensitive measure of sleep duration used by the FLAME study (average of sleep duration at 3, 4 and 5 years of age). Finally, in our study, there were no parents or children who reported <8 h and 15 min of average sleep; such a small inter-individual variability in sleep duration could have made prospective associations between sleep duration and adiposity difficult to detect.
In a recent study, youth with greater adiposity reported poorer sleep quality, more sleep disturbances and a delayed sleep phase pattern, independent of sleep duration.5 We found sleep disturbances to be cross-sectionally associated with FMI, which supports their finding. These data suggest that sleep measures beyond sleep duration contribute to the negative association between sleep and obesity. However, we did not find any prospective associations between sleep disturbances and FMI in the longitudinal analysis.
Strengths of our study include the three repeated measurements of objectively assessed movement behaviors as well as FMI determined by DXA in a large and well-characterized sample of Danish children. The longitudinal analyses were, however, performed within a nutrition intervention study, which may have influenced the changes in energy density of the diet that was used as a covariate, although it did not influence changes in movement behaviors. As self-reported time spent cycling during weekdays was the same during autumn (first measurement) and spring (third measurement) (14 min day–1; data not shown), the inability of accelerometers to capture cycling is not expected to have influenced our conclusions. Owing to the relatively short follow-up time of 200 days from fall to spring, we cannot rule out that our results could be caused by season-specific differences between children with various FMI rather than an effect of having higher FMI at baseline independent of season.47 However, given that prospective associations were detected between FMI and MVPA this seems to be a minor point. Finally, as behavior is more imprecisely measured than FMI, there is a risk that the true association between baseline movement behavior and change in FMI is slightly underestimated.
In conclusion, low PA and short sleep duration were independently associated with a higher FMI, and combined associations of movement behaviors showed a synergistic effect with FMI. In the longitudinal analysis, low PA and longer sedentary time appeared to be the result of fatness rather than its cause even within normal-weight children. We believe that our findings could be relevant to the successful prevention of early adiposity by changing the focus, so that we not only think of weight status as a result of lifestyle behaviors but also lifestyle behaviors as a result of weight status, even in healthy normal-weight children.
The study is part of the OPUS project 'Optimal well-being, development and health for Danish children through a healthy New Nordic Diet' supported by a grant from the Nordea Foundation. We are very grateful to the participants and would also like to acknowledge the school staff as well as other researchers and staff in the OPUS project.
About this article
Designed research: AA, KFM, IT, S-MD and AS; coordinated data collection: MFH, S-MD and RA; analyzed and interpreted data: MFH; discussed the analysis and interpretation of the data: AS, CR and J-PC; wrote paper: MFH; had primary responsibility of the final content: AS. All authors reviewed the manuscript critically and approved the final manuscript.