3.4 Statistical analysis
4.1.2 Sample characteristics
On initial screening of the data (n=1296) 98 cases were removed due to incomplete data, as per ethics requirements, and 12 cases were removed as it was indicated that the child met the exclusion criteria based on parent’s response to the item ‘Does your child have a disability or medical condition which affects their growth, development or metabolism?’
147 4.1.2.1 Child BMIz
Based on the method of detecting biologically implausible values described in section 3.2.5.1, 18.5% (n=208) of the initial sample were excluded.
Figure 9: Screening for biologically implausible values
Initial data = 1296
Removal of incomplete data (n=110; leaving n= 1186 with complete data) Remove height Z scores less than 3 (n=35)
Remove height Z scores greater than 3 (n=144) Dataset = 1007
Remove weight Z scores less than 3 (n=4) Remove weight Z scores greater than 3 (n=17) Dataset = 986
Regress weight Z score on height Z score remove residuals greater of less than 2.5 (n=8) Screened dataset = 978
This rate of biologically implausible values is similar to that reported in a study of children 2 – 5 years (20.5% - 16.5% implausible), based on parent reported data, although other studies have reported much lower rates (7.9% - 9.7%). [388, 389] Limited data availability makes comparison difficult as does no standard method of reporting biologically implausible values. Similar to what has been reported in other studies, misreporting of anthropometric data was higher in boys, although, contrary to other studies, implausible data were higher in younger children. [390, 391]
From the retained sample, child BMI categories were additionally determined according to Cole, 2000 and 2007. [47, 48] Findings from the previously reported meta-analysis further indicated that the use of self-reported data are more likely to result in under-identification of positive cases (i.e., overweight and obese participants), but less likely to include negative cases (i.e., non-overweight and non-obese participants). [392] This should be kept in mind with regard to the BMI categories derived in this study.
148 Once cases of biologically implausible values were removed, rates of overweight and obese children in this study (11.1% and 6.5%, respectively) were similar to those reported in national data of children age 4 - 5 years of 15.2% overweight and 5.5% obese. [10]
Rates of underweight children in this sample (n=22.4%), however, are likely over-represented compared with national data (n=7.55%), which similarly reduced rates of normal weight children in this sample (n=59.9%) compared with nations data (67.75%;
table 6, section 4.1). [10] Child BMIz were checked for normality and was deemed skewed so transformed accordingly.
4.1.2.2 Parent BMI
Parent’s BMI were initially visually screened for very high or very low values. Based on this, 2 cases were removed from further analysis due to very low values (BMI 9.26kg/m2 and 10.33kg/m2, created due to extremely high height with very low weight). No cases were removed based on very high BMI values. Based on this method, parent BMI data were available for 1184 participants. This sample was used for analysis that was independent of child BMIz. It should be noted that, because of the exclusion of parents with implausible BMI an additional child was lost from analysis that included both parent and child BMI data. This gave a final sample of 977 cases with both parent and child BMI data. The Cronbach α presented in section 3.2.5 are based on the sample including both parent and child (n = 977) as this was the sample predominantly used for further analysis in this thesis.
Despite parental BMI being self-reported, rates of overweight (26.0%) and obese (32%) were similar to those reported in national samples of Australian adults (27.6% overweight, 33.7% obese, women 35 - 44 years, table 8). [10, 19, 20] On checking for normality, parent’s BMI were also deemed shewed so were transformed accordingly.
4.1.2.3 Demographic characteristics
All states in Australia and geographic regions have been represented in this sample. Rates of single parents in this study were 12%. Distribution of participants in the high and middle-income groups were similar (45% and 42%, respectively), however, low middle-income families appear underrepresented (13%). While this study refers to ‘parents,’ mothers were prominent respondents in this study (94%). This under-representation of fathers’ limits ability to conduct analysis of this sub-sample. See table 8 for further details of the sample characteristics.
149
Table 8: Demographic data
Full Sample % (n = 1184) Sample excluding biologically implausible values % (n = 977) Gender
Single parents (binary coded) 11% (133) 12% (114)
Parent BMI categoryc
150
Capital city
Metro (population over 100,000) Large rural (population 25,000 – 99,999) Small rural (population 10,000 – 24,999) Large remote (population 5,000 – 9,999) Small remote (population less than 5,000)
25% (298) 31% (369) 21% (244) 13% (154) 4% (45) 6% (74)
26% (255) 31% (301) 19% (188) 13% (128) 4% (41) 7% (64) Sleep
Minimum Maximum Mean
4 hours 14 hours 10.67 (SD 1.11)
4 hours 14 hours 10.67 hours (SD 1.11)
Mean number of children in the home 2.06 (SD 0.88) 2.07 (SD 0.88)
N (%) reported for dichotomous variables Mean (SD) reported for continuous
aCut offs per Cole, TJ. (2000 and 2007)
b2000 CDC growth charts
cCut offs per WHO classifications for adults (2000)
151 Text and tables within section 4.1.3 are a reproduction of the manuscript published in
the journal Appetite except for edits which have been made to the following sections to improve the statistical rigour of the paper.
Edits made to section Statistical Method, Relation [Correlation] between psycho-social variables, child BMI z-score and income and Table 11 – 12.
The citation is as follows:
Boswell, N., Byrne, R., & Davies, P. S. W. (2018). Eating behaviour traits associated with demographic variables and implications for obesity outcomes in early childhood.
Appetite, 120, 482-490. Doi: https://doi.org/10.1016/j.appet.2017.10.012
Contributor Statement of contribution
Nikki Boswell Conception and design of the study (60%) Analysis and interpretation of the data (60%) Drafting and critical review of manuscript (60%) Rebecca Byrne Conception and design of the project (20%)
Analysis and interpretation of the data (20%) Drafting and critical review of manuscript (20%) Peter Davies Conception and design of the project (20%)
Analysis and interpretation of the data (20%) Drafting and critical review of manuscript (20%)
152 4.1.3 Paper 2: Eating behaviour traits associated with demographic variables and
implications for obesity outcomes in early childhood
4.1.3.1 Background
Despite the attention that childhood overweight and obesity has been given over the past few decades, it remains an issue of major public health concern, with rates reaching around 20% among Australian children aged 2 – 3 years (2007) and around 27% among children aged 5 -17 years (2014 – 2015). [10, 11, 393] The emergence of overweight and obesity at such early stages of life is particularly concerning given that being overweight and obese during childhood significantly increase the risk of being overweight and obese as an adult, as is associated with an increased risk non-communicable disease in both the long and short term. [394]
According to the behavioural susceptibility theory, obesity emerges when genetic susceptibility interacts with environmental circumstances and ‘obesogenic’ behaviours ensue. [2] In accordance with this, eating behaviours provide a potential intermediary pathway from which obesity development can be better understood and prevention initiatives targeted. In particular, the early childhood period offers a unique and critical window for such intervention, as it is during this period that eating behaviours emerge and are reinforced to provide a foundation for obesogenic behaviour throughout the lifespan.
[65] This intermediary role of eating behaviours in childhood obesity can be seen across the literature which shows food approach eating behaviours such as food responsiveness and enjoyment of food to be positively associated with overweight and obesity, while food avoidance eating behaviours such as satiety responsiveness, slowness in eating and food fussiness, are associated with reduced overweight and obesity outcomes. [5, 7, 57]
In attempting to understand this role of eating behaviours in early childhood obesity, attention has largely been given to proximal, micro-environmental factors, such as parent’s feeding strategies, which are considered to have a pivotal influence on child behaviour.
[294, 395] Far less attention, however, has been given to exploring psycho-social variables which may underpin eating behaviours, such as parent’s stress and depression, low income status, single-parent status, and/or short sleep duration, which have been seen to internally alter appetite regulatory systems. [13, 91, 100-102, 104-107, 396-398] For the purposes of this paper, a distinction will be made between micro-environment
153 determinants of eating behaviours and psycho-social determinants of eating behaviours, by using the term eating behaviour traits to refer to the latter.
Current understanding of the eating behaviour traits that underpin childhood overweight and obesity postulates that alteration in the homeostatic regulation of food intake, as coordinated by neuroendocrine feedback loops, involving nutrient and hormonal signals, results in a down regulation of food avoidance eating behaviour traits and/or an up regulation of food approach eating behaviour traits. [5, 73, 74] These alterations may be a consequence of internal susceptibility to inappropriate responses to homeostatic regulatory systems or a vulnerability of these systems to be overridden by external influences. Additionally, vulnerability to hedonic eating (eating for pleasure in the absence of energy deficits), neurologically, can contributes to overweight and obesity through excess energy intake. [95]
These systems, particularly during early childhood, are vulnerable to epigenetic changes and/or changes in neurological structure in response to certain environmental circumstances. [3, 182, 396, 399, 400] That is, while eating behaviour traits are estimated to have approximately 50% heritability, it is possible that they are manipulated and shaped through shared environment which is similarly estimated to account for approximately 45%
of variance in eating behaviour traits. [3] For instance, alterations in the appetite regulating hormones leptin, ghrelin and cortisol, under control of the hypothalamic-pituitary-adrenal [HPA] axis, have been noted as a result of chronic stress, reduced breastfeeding, reduced sleep duration, and general ‘disadvantaged’ life circumstances. [13, 91, 100-102, 104-107, 396-398] Furthermore, it is understood that children from low socio-economic backgrounds often experience greater neurological impulsivity and reward seeking behaviour, as has been associated with increased food approach behaviours and obesity development. [74, 401, 402] This underpinning of chronic stress and adversity on appetite and eating behaviours traits highlights a potential pathway from which higher rates of obesity in disadvantaged sub-population groups could be explained, however are yet to be extensively explored. [10, 19, 146]
Given this, this study aims to determine psycho-social demographic variables associated with eating behaviours traits and the relationship these traits have with obesity development in Australian children during early childhood. It is hypothesized that low income status, single-parent status, short sleep duration, parent’s depression, stress and
154 anxiety, and breastfeeding duration, will be associated with obesogenic eating behaviour traits in children. Gaining understanding of such interaction of these psycho-social variables provides a new perspective in approaching childhood obesity and support alternative/novel preventative focus.
4.1.3.2 Materials and Method Recruitment
Between July and November, 2016, Australian parents of children aged 2.0 – 5.0 years self-enrolled to complete an online, cross sectional survey. Participants were invited to enrol in the survey through advertising on the social media website Facebook®. The advertisement provided brief information about the survey and provided a link to a website which contained further details about the research project as well as the plain language statement, participant consent form, and access to the online survey, hosted by Checkbox®. Children were excluded from this study if parents reported the child had a medical condition likely to affect the child’s growth, development or metabolism. In the instance that a parent had more than one child within the target age, parents were asked to refer to the child whose birthday occurred next.
Measures
Self-selected parents responded to questions regarding their child’s age, gender, parent/respondent gender, single parent status, income, state and region of residency, sleep duration, breastfeeding history, parent depression, anxiety and stress, and children’s eating behaviours. These variables were selected for inclusion in analysis as identified to be psycho-social variables associated with eating behaviours and/or obesity development across the literature.
Participants were prompted to use household measures (e.g. bathroom scales/ household tape measure) to report child weight and height which were subsequently used to calculate weight, height and BMI z-scores according to the 2000 CDC growth charts. [381] Use of CDC growth charts for children from 2 years of age is consistent with the recommendations of the National Health and Medical Research Council (NHMRC). [17]
Child BMI categories were additionally determined according to Cole, 2000 and 2007. [47, 48]
155 Parents were also prompted to use household measures to report their weight and height which were used to calculated BMI scores and BMI categories in accordance with the World Health Organizations classifications (Underweight <18.50kg/m2; Normal weight 18.50 - 24.99kg/m2; Overweight ≥25.00kg/m2; Obese ≥30.00kg/m2). [49]
Data Screening and sample size
As child height and weight were by parental report it was deemed necessary to screen the data for biologically implausible values (BIVs). Although there is no standard approach to assessing BIVs, Flegal and Cole (2013) reported that using the CDC BMI reference data to calculate BMI z-scores that are then screened for BIVs will lead to errors if a modification of the BMI z-score is not undertaken. [382, 383] The process of calculating what is referred to as a “modified Z score” is complicated and not universally supported, thus the process used to screen for BIV’s in this paper involved a twostep approach.
Firstly, all weight or height z-scores greater or less than 3 were discarded. Secondly, after regressing weight z-score on height z-score any residuals greater or less than 2.5 were also discarded.
Parents BMI were also visually screened for very high or very low values, however no cases were excluded based on this visualization.
Children’s eating behaviour traits
Child eating behaviour traits were measured using five of the eight sub-scales of the children’s eating behaviour questionnaire (CEBQ) and showed acceptable internal reliability in the present study; enjoyment of food (4 items; Cronbach α 0.868); food responsiveness (5 items; Cronbach α 0.921); satiety responsiveness (5 items; Cronbach α 0.800); slowness in eating (4 items; Cronbach α 0.709); and food fussiness (6 items;
Cronbach α 0.677). This shortened version of the CEBQ was chosen to minimize participant burden and as each sub-scale has previously shown good psychometric properties and good internal reliability with Cronbach’s alpha values between 0.73 and 0.91 when validated in an early childhood population (1 – 5 years) in Australia. [179] The sub-scales selected were nominated so as to allow comparison across the literature. [5, 67] Items were scored on a 5-point scale, with higher scores indicating higher values of each trait. Mean scores for each sub-scale were calculated. [67]
Parental depression, anxiety and stress
156 Parental depression, anxiety and stress was measured using the DASS-21, depression, anxiety and stress scale. [373] The DASS-21 is a 21 item self-report questionnaire designed to measure the severity of a range of symptoms common to depression, anxiety and stress over the previous week. [403] This study similarly showed high internal reliability across all 3 scales (stress [7 items; Cronbach α 0.837]), anxiety [7 item;
Cronbach α 0.742], depression [7 items; Cronbach α 0.886]).
Statistical method
All analyses were carried out using SPSS v24 (SPSS Inc., Chicago, IL, USA). The distribution of predictive variables were examined for multicollinearity and normality (skewness and kurtosis between 1 and -1). Parent mean depression scores, parent mean anxiety scores, and parent BMI were deemed skewed so transformation was performed on these variables accordingly. Upon transforming parents BMI, one outlier was identified and thus excluded from further analysis.
CEBQ sub-scales were normally distributed (skewness and kurtosis between 1 and -1) and had good internal consistency, with Cronbach’s α over 0.677.
To examine whether BMI categorization showed linear associations with eating behaviour traits a one-way between-groups multivariate analysis of variance was performed with BMI category as the independent variable and the three CBEQ sub-scales significantly correlated with child BMI category (food responsiveness, enjoyment of food, satiety responsiveness, and food fussiness) as dependent variables. Pillai’s Trace was examined for significance, homogeneity of variance assumption examined with Levene’s F tests, a series of one-way ANOVA’s on each of the CEBQ sub-scales was conducted as a follow-up tests to the MANOVA, and post-hoc contrasts (LSD) performed. [7]
Pearson’s correlations were used to assess relationships between CEBQ scores and child BMI z-score, and between continuous psycho-social variables and child BMI z-score, as appropriate [5, 55]. Where significant, linear regression was conducted and coefficients, confidence intervals and mean scores inspected to check the direction and pattern of the association. [5] To examine the relationship between child BMI z-score and categorical psycho-social variables ANOVA and Chi-square tests were conducted as appropriate.
Multiple regression analysis was applied to examine associations between select variables (child age, child gender, duration of sleep, parent BMI, breastfeeding less than 6 months
157 (binary coded with breastfeeding longer than 6 months), low income status, single parent status, and parental DASS scores) and CEBQ scores. Stepwise regression was used to determine eating behaviour traits and demographic variables associated with child BMI z-score. All hypotheses will assume a 0.05 significance level and a two-sided alternative hypothesis.
4.1.3.3 Results
From the initial sample of 1186 participants that had completed the survey, once cases of BIV and outliers were removed, a final sample of 977 Australian children, aged between 2.0 and 5.0 years, were retained with plausible BMI data. Excluded cases did not differ significantly based on parent BMI category, parent gender, single parent status, income group, or state or region of residency in one-way ANOVA analysis, however were significantly younger (mean age 3.1 years, compared with 3.4 years, p=0.000) and were significantly more likely to be boys (58.0% in excluded case compared with 49.4% in retained sample, p=0.026).
Participants represented all states and territories of Australia, from a variety of geographic regions. The majority of parent responders were mothers (94.7%) and 11.7% were single parents. Further demographic variables of participants can be seen in table 9.
Table 9: Demographic data (n = 977)
Gender - Boy 483 (49.4)
158
Metro (population over 100,000)
Large rural (population 25,000 – 99,999) Small rural (population 10,000 – 24,999) Large remote (population 5,000 – 9,999) Small remote (population less than 5,000)
255 (26.1) N (%) reported for dichotomous variables
Mean (SD) reported for continuous
aCut offs per Cole, TJ. (2000 and 2007)
b2000 CDC growth charts
cCut offs per WHO classifications for adults (2000)
Correlation between eating behaviour traits and child BMI z-score
Significant correlations were detected between all CEBQ sub-scales and child BMI z-score, except slowness in eating. Significant correlations were also detected between all CEBQ sub-scales, such that, food approach traits (enjoyment of food and food
159 responsiveness) were negatively correlated with food avoidance traits (satiety responsiveness, food fussiness and slowness in eating), while food approach traits were positively correlated with each other (enjoyment of food and food responsiveness), as were food avoidance traits (satiety responsiveness, food fussiness and slowness in eating;
table 10).
Table 10: Correlations between CEBQ Sub-scales and child BMI z-score Food
*. Correlation is significant at the 0.05 level (2-tailed).
**. Correlation is significant at the 0.01 level (2-tailed).
Relation between psycho-social variables, child BMI z-score and income
Significant correlations were seen between child BMI z-score and parent BMI (p=0.029), but no other variables (table 11).
Table 11: Correlations between psycho-social variables and child BMI z-score Sleep
Parent depression .580** .177**
.000 .000
Parent anxiety .141**
.000
160
Parent BMI 1
*. Correlation is significant at the 0.05 level (2-tailed).
**. Correlation is significant at the 0.01 level (2-tailed).
In ANOVA model’s no significant differences were seen in child BMI z-score between categories of single parent status (F=0.318, p=0.573), breastfeeding duration (F=1.22, p=0.269) or income category (F=0.630, p=0.533; table 12). Significant differences were detected, however, between income category and child sleep duration, parent’s BMI, parent’s stress, depression and anxiety, breastfeeding duration, and single parent status (table 12).
Table 12: Psycho-social variable differences by income group (ANOVA and Chi-square)
Low Income
Eating behaviours traits across BMI categories
A statistically significant difference existed between BMI category for the dependent variables (food responsiveness, enjoyment of food, satiety responsiveness, and food fussiness) in the MANOVA, with Pillais’ Trace= 0.022, F(12, 2907) = 1.79, p= 0.044. The multivariate effect size was estimated at 0.007, which implies that 0.7% of the variance in the dependent variable was accounted for by BMI category. Based on a series of Levene’s F tests, the homogeneity of variance assumption was considered satisfied, with all sub-scales failing to reach significance. A series of one-way ANOVA’s on each of the CEBQ
161 sub-scales was finally conducted as a follow-up test to the MANOVA. Only ANOVA’s for enjoyment of food, food responsiveness and satiety responsiveness were statistically significant (p=0.012, p=0.005, p=0.008, respectively), with effect sizes (partial n2) ranging from 0.011 to 0.013.
Finally, a series of post-hoc analyses (Fisher’s LSD) were performed to examine individual mean difference comparisons across BMI categories and CEBQ subscales. The results revealed statistically significant differences in enjoyment of food between underweight and
Finally, a series of post-hoc analyses (Fisher’s LSD) were performed to examine individual mean difference comparisons across BMI categories and CEBQ subscales. The results revealed statistically significant differences in enjoyment of food between underweight and