This secondary analysis uses data from a randomised controlled trial of early-life obesity prevention (Prevention of Overweight in Infancy [POI]) in Dunedin, New Zealand (2-year intervention [21] with follow-up at 3.5 and 5 years of age) [22]. The current data have been analysed using the entire cohort with appropriate adjustment for randomisation group, as no significant differences were observed in physical activity, sedentary behaviour, or sleep following the intervention [23–25].
Detailed information regarding POI is available in study protocols [21, 22] and published findings [24, 25]. The intervention was approved by the Lower South Ethics Committee (LRS/12/08/063) and the follow-up study by the University of Otago Human Ethics Committee (12/274). Written informed consent was obtained from the parent/guardian of all child participants. We invited all mothers who had booked into the single maternity hospital (> 97% of all births) in Dunedin to participate when in the latter stages of their pregnancy. The final sample included 802 women (58% response rate) randomised to Usual care; Sleep; Food, Activity and Breastfeeding; or Combination groups. Anthropometric assessments (primary outcome) were performed by researchers blinded to group allocation. Demographic information obtained at baseline included maternal age, education, ethnicity, self-reported pre-pregnancy height and weight, and level of household deprivation. Information on infant gestational age, sex and birth weight was obtained from hospital records.
Measures obtained at 2, 3.5 and 5 years of age
Anthropometric measurements were obtained by trained measurers following standard protocols [26]. Duplicate measures of weight (Tanita WB-100 MA/WB-110 MA) and height (Harpenden stadiometer, Holtain Ltd, UK) were obtained with children wearing light clothing. Body mass index (BMI) z-scores were calculated using the WHO growth standards [27], with overweight defined as a BMI z-score ≥ 85th but < 95th percentile, and obesity as BMI ≥ 95th percentile.
Sleep, physical activity, and sedentary behaviour were assessed using Actical (Mini-Mitter, Bend, OR) accelerometers (initialized using 15 second epochs), worn around the waist 24-hours a day for one week. Data were scored using an automated count-scaled algorithm that estimates sleep onset (start of first 15 continuous minutes of sleep preceded by 5 min of awake) and offset (last of 15 continuous minutes of sleep followed by 5 min of awake) specific to each individual each day. Total sleep time is the difference between sleep onset and offset, excluding waking after sleep onset (WASO). Naps were determined in children at 1 and 2 years of age only defined as at least 30 min of continuous sleep, preceded by 5 min of awake between 9am and 5pm [28]. Awake time was divided into non-wear time (at least 20 minutes of consecutive zeros [29]), sedentary time (0–6 counts/15 seconds), light physical activity (LPA, 7-286 counts/15 seconds), and moderate-to-vigorous physical activity (MVPA, ≥ 287 counts/15 seconds) [30, 31]. As each 24-hour ‘day’ was determined from the time the child woke up on day 1 to the time they woke up on day 2 (and so on), a day was considered valid if the participant had 20–28 hours of data to allow for changing wake times. Participants had to have at least three valid days to be included in analyses. Snoring at 3.5 and 5 years of age was controlled for in analyses as it is one of the most common forms of sleep disturbance at this age [32], and can lead to problems with behavioural and emotional regulation [33]. Parents were asked ‘how often does your child snore loudly’ with 7 answer options ranging from ‘never’ to ‘every night’.
Measures obtained at 5 years of age only
Psychosocial factors were measured using laboratory assessment and parental report. We determined levels of inhibitory control using the ‘Statue’ component of the NeuroPSYchological Assessment (NEPSY-2) [34] and the Head-Toes-Knees-Shoulders task [35]. NEPSY-2 is a test battery that is well-normed, reliable, and appropriate for use with 5-year-old children. The Head-Toes-Knees-Shoulders task [35] is a measure of behavioural regulation and inhibition, which determines the ability of a child to follow opposing instructions (e.g. touch the head when directed to touch the toes).
Parental ratings were obtained using the Parent Rating Scale of the Behavioral Assessment System for children (BASC-2), a well-validated and normed scale [36]. We used the ‘Hyperactivity’ (11 items, α = 0.79), ‘Emotional Self-Control’ (6–8 items, α = 0.79), ‘Executive Functioning’ (13 items, α = 0.79) and ‘Attentional Problems’ (6 items, α = 0.80) subscales as measures of self-regulation, and the ‘Anxiety’ (13 items, α = 0.82), Depression (11 items, α = 0.77), and ‘Resilience’ (12 items, α = 0.82) subscales as indicators of the child’s mental health.
Statistical analyses
Demographic characteristics were described for those with measures at 2, 3.5, or 5 years of age, and those who provided at least one measure (full analysis sample). Differences between the full analysis sample and the remaining POI participants were assessed using a t-test for continuous variables and a chi-squared test for categorical variables.
Time use components were normalised to sum to 24-hours, with non-wear time first reallocated proportionally to wake-time components only [37]. Compositional analyses were undertaken using a 3-component composition (sleep, sedentary, physical activity) because international guidelines for preschoolers focus on light-to-vigorous physical activity (LMVPA), and because most (80%) of our measurements at 5 years were obtain just prior to the child’s birthday. However, a 4-component composition (sleep, sedentary, light physical activity [LPA], and moderate-to-vigorous physical activity [MVPA]) was also undertaken for the cross-sectional analyses at 5 years, and longitudinal analyses using the 3.5 year data, based on the Level 2 Canadian guidelines for 3–4 year olds which recommend that at least 60 of the 180 minutes a day in LMVPA is spent in energetic play [8]. Compositional means for each component were calculated as geometric means normalised to 24 hours [38]. To be able to include all co-dependent compositional variables in a regression model together, compositional data analysis (CoDA) methods were used. This involves using isometric log-ratios of the components [38] and including these coordinates in a linear regression model as the independent variables with the relevant outcome as the dependent variable. Models were adjusted for sex, household deprivation, randomisation group, and BMI z-score at 5 years of age, and snoring at 3.5 and 5 years of age as previously mentioned.
To report meaningful estimates of association between time-use components and mental or psychosocial health, the regression coefficient of the first isometric log-ratio coordinate (which contains the ratio of one component to all others) was back-transformed to represent the mean difference in the dependent variable for a 10% greater time spent in the component of interest relative to all others [39]. Separate regression models were generated to report estimates for each time-use component, with the isometric log-ratios calculated by different permutations of components.
Longitudinal associations between time-use at 2 and 3.5 years and mental or psychosocial health at 5 years were assessed as for the cross-sectional associations with no adjustment for 5-year time-use. All mental and psychosocial health variables were standardised so that estimates are presented in units of standard deviations. Standardised mean differences and 95% confidence intervals were calculated, estimating the mean difference for a 10% greater time in the component relative to all other components. P-values are not reported and hence adjustment for multiplicity was not undertaken. Effect estimates and confidence intervals are considered in the interpretation of results. Residuals of all regression models were plotted and visually assessed for homogeneity of variance and normality. All statistical analysis was carried out in Stata 17.0 (StataCorp, Texas).