Obesity in British children with and without intellectual disability: cohort study

The UK’s Millennium Cohort Study (MCS) is the fourth in the series of British birth
cohort studies. It aims to follow throughout their lives a cohort of over 18,000 children
born in the UK between 2000 and 2002. MCS data are managed by the Centre for Longitudinal
Studies at the University of London (www.cls.ioe.ac.uk/) and are available to researchers registered with the UK Data Service (www.ukdataservice.ac.uk) (www.esds.ac.uk). Full details of the design of MCS are available in a series of reports and technical
papers 36]–42] key aspects of which are summarised below.

Sampling

Participant families were randomly selected from Child Benefit Records, a non means-tested
welfare benefit available to all UK children at the time the cohort was established.
Sampling was geographically clustered to include all four countries of the UK (England,
Wales, Scotland, Northern Ireland), and disproportionately stratified to over-sample
children from ethnic minority groups and disadvantaged communities 43]. Children and families were drawn from 398 randomly selected electoral wards in the
UK. The first survey (MCS1) took place when children were nine months old and included
a total of 18,552 families. Children were followed up at ages three (MCS2; 15,590
families, 78 % response rate), five (MCS3; 15,246 families, 79 % response rate), seven
(MCS4; 13,857 families, response rate) and eleven (MCS5; 13,287 families, 69 % response
rate). For each family, information was collected on the target child falling within
the designated birth date window. For multiple births (e.g., twins, triplets) information
was collected on all children. To avoid the statistical problems associated with the
clustering of multiple births within households, the present analyses are restricted
to the first named target child in multiple birth households. All analyses used sampling
weights provided with MCS data to adjust for the initial sampling design and biases
in recruitment and retention at specific ages.

Identification of children with intellectual disability

Child cognitive ability was assessed at age three using the Bracken School Readiness
Assessment 44] and Naming Subscale of the British Ability Scales (BAS) 45], selected subscales of the BAS at ages five and seven, and the NFER Progress in Maths
test at age seven 46]. At age 11 children were given three cognitive tests; verbal similarities (BAS),
the Spatial Working Memory task and the Cambridge Gambling task, both from the Cambridge
Neuropsychological Test Automated Battery. Of the age eleven tests, only verbal similarities
is closely related to traditional measures of IQ.

For ages five and seven we extracted the first component (‘g’) from a principal component
analysis of all age-standardised subscale/test scores 47]. The first component accounted for 63 % of score variance at age seven and 55 % of
score variance at age five. We identified children as having intellectual disability
if they scored two or more standard deviations below the mean on the first principal
component at age seven (n?=?419 [3.3 %] of 12,820 children for whom test results were available).

If cognitive test scores were missing at age seven, we identified children as having
intellectual disability if they scored two or more standard deviations below the mean
on the first principal component at age five (n?=?146 [6.5 %] of 2,250 children). If cognitive test scores were missing at age five
and at age seven, we identified children as having intellectual disability if they
scored two or more standard deviations below the mean on the Bracken School Readiness
Assessment at age three (n?=?49 [4.4 %] of 1105 children). If Bracken scores were not available, we identified
children as having intellectual disability if they scored two or more standard deviations
below the mean on the BAS Naming Subscale at age three (n?=?54 [7.6 %] of 711 children). This process allowed us to classify intellectual disability
on the basis of cognitive test scores for 99.1 % of children participating at age
seven.

For 125 children no cognitive test results were available at any age. Interviewers
did not administer the assessments if the child ‘has a learning disability/serious
behavioural problem (e.g., severe ADHD, autism) which prevents them from carrying
out the assessments’, ‘is unable to respond in the required manner for each assessment,
e.g., reading, writing, manipulating objects’, ‘is not able to speak or understand
English (or Welsh if applicable)’ or if consent and co-operation were not forthcoming.
For these children we identified intellectual disability on the basis of parental
report at age seven. A child was identified as having learning disabilities if both
of the following two criteria were met: (1) the child was reported to be receiving
special education due to their ‘learning difficulty’ (the term used in educational
services in the UK to refer to intellectual disability); (2) the child was reported
to have ‘great difficulty’ in all three areas of reading, writing and maths. This
led to the identification of another 11 children as having intellectual disability
(8.8 % of children for who no test scores were available).

Finally, we used the normalised verbal similarities standard score at age eleven to
attempt to address potential errors in classification in the W2-4 variables. Specifically,
all children who had been identified as having intellectual disability who scored
at or above the population mean on verbal similarities at age eleven were reclassified
as not having intellectual disability. Similarly, all children identified as not having
intellectual disability but who scored three or more standard deviations below the
population mean on verbal similarities at age eleven were reclassified as having intellectual
disability.

This procedure led to the identification of 647 of the 18,495 (3.5 %) children participating
at Wave 1 where the child’s mother was the primary informant as having intellectual
disability. As expected, boys were significantly more likely than girls to be identified
as having intellectual disability (4.3 % vs 2.6 %; OR?=?1.67, 95 % CI 1.42–1.96).

Overweight obesity

Measurements of child height and weight were taken and BMI calculated at ages three
(MCS2), five (MCS3), seven (MCS4) and eleven (MCS5). MCS data releases include categorical
measures of overweight and obesity based on the International Task Force childhood
obesity reference charts and cut-off points 48]. At ages three and five, 5 % of cohort members were identified as obese 49], 50]. However, studies have suggested that: (1) the international reference lacks specificity
due to the sample size used to define the population; (2) their use may lead to the
under-reporting of rates of obesity; and (3) international reference charts may ignore
population differences in the relationship between BMI and adiposity 51]. As a result, more recent studies of child obesity using MCS data have used UK-specific
growth reference charts to recalculate overweight and obesity from BMI, age and gender
data 52].

In the present study we used UK 1990 gender-specific growth reference charts and the
LMS Growth programme to identify age and gender-specific overweight and obesity BMI
thresholds for each child at ages 5, 7 and 11 years 53], 54]. The use of UK 1990 BMI growth reference charts to calculate obesity is not recommended
under 4 years of age 55]. Children whose BMI fell at or above the 85th percentile of the UK 1990 reference
population were defined as overweight, and those at or above the 95th percentile were
defined as obese 55], 56].

Predictor variables

Previous research undertaken with the MCS has reported that increased risk of overweight
and obesity was associated with a range of variables including ethnicity, higher birth
weight, early introduction of solid foods, missing breakfast, sedentary lifestyle,
maternal smoking (including during pregnancy), parental overweight (including pre-pregnancy
overweight), lone motherhood, maternal employment, low parental educational attainment,
low income, material hardship, living in more socially deprived neighbourhoods 23], 49], 50], 52]. The following potential predictor variables were extracted from MCS data collected
in Waves 1–4.

Child ethnicity

Child ethnicity was based on parental report and coded using the six category UK Census
scheme: White; Mixed Ethnicity; Indian; Pakistani or Bangladeshi; Black or Black British;
Other.

Child birth weight

Parental report of child birth weight was collected at Wave 1. These data were recoded
into a binary measure of higher birth weight based on falling within the upper quartile
of the weighted sample distribution (?3.70 kg).

Child nutrition

At Wave 1 (9 months) information was collected from the main parental informant on
the age at which the child first had solid foods. These data were recoded into a binary
measure of introduction of solid foods in the first three months of life 52].

At Waves 3 and 4 information was collected from the main parental informant on: (1)
the number or portions of fruit eaten daily; (2) the main type of between-meal snacks
the child eats; and (3) the number of days the child has breakfast. Following inspection
of the distribution of these data, binary measures were derived of: (1) low levels
of fruit consumption (less than two portions per day); (2) higher calorie snacks (the
main type of snack being crisps, cakes or biscuits, sweets); (3) child skips breakfast
one or more days per week.

Child participation in sport physical activity

Parental report of child participation in sport was collected at Waves 3 and 4. Following
inspection of the distribution of these data, binary measures of low levels of participation
were derived based on participation on less than one day per week (age 5, Wave 3)
and participation on less than two days per week (age 7, Wave 4). Parental report
of child participation in physical activity sport was collected at Wave 4. Following
inspection of the distribution of these data, a binary measure of low participation
was derived based on participation on less than five days per week.

Child tv watching and playing computer games

Parental report of the time the child spends watching TV was collected at Waves 2–4.
Parental report of the time the child spends playing computer games was collected
at Waves 3 and 4. Binary measures of high levels of TV watching or computer game playing
were derived based on watching/playing for three or more hours daily. At Wave 4 information
was also collected on whether the child had a TV in their bedroom.

Child exposure to bullying

The Strengths and Difficulties Questionnaire 57] was completed by parents at Waves 2–4. One item ‘Picked on or bullied by other children’ was used to identify exposure to bullying at ages three, five and seven.

Maternal obesity

Maternal BMI was calculated from self-reported height and weight at each of the four
MCS waves and retrospectively at Wave 1 for pre-pregnancy weight. Obesity was defined
as a BMI of 30 or more. Maternal obesity status over time was coded as never, once
or twice, three or more times.

Maternal smoking

Information was collected at each Wave on whether the child’s mother was a current
smoker. Following inspection of the distribution of these data a binary measure of
maternal smoking was derived based on being a current smoker at two or more of the
four waves of data collection.

Maternal educational attainment

Highest level of educational attainment across waves was coded according to National
Vocational Qualifications (NVQ) categories.

No qualifications or NVQ Level 1: Competence that involves the application of knowledge
in the performance of a range of varied work activities, most of which are routine
and predictable (equivalent to one General Certificate of Secondary Education (GCSE)
at grade D-G).

NVQ Level 2: Competence that involves the application of knowledge in a significant
range of varied work activities, performed in a variety of contexts. Collaboration
with others, perhaps through membership of a work group or team, is often a requirement
(equivalent to one GCSE at grade A*-C).

NVQ Levels 3–5: NVQ Level 3 requires competence that involves the application of
knowledge in a broad range of varied work activities performed in a wide variety of
contexts, most of which are complex and non-routine. There is considerable responsibility
and autonomy and control or guidance of others is often required (equivalent to 1–5
Advance Level certificates at grades A*-C).

Overseas qualifications only.

Maternal employment

Information on maternal employment and hours worked was collected at each Wave. High
levels of hours worked were defined at each wave as working for more than 20 h per
week 52].

Single parent family

Information on household composition was collected at each Wave. Following inspection
of the distribution of these data a single binary measure of single parent family
was derived based on being a single parent family at any of the four waves of data
collection.

Household income poverty

Information on household income was collected at each Wave. These were adjusted for
household composition using the modified OECD equivalisation scale 58] and used to define household income poverty (equivalised household income falling
60 % below the population median) 59]. Following inspection of the distribution of these data a single binary measure of
household income poverty was derived based on a household being in income poverty
at any of the four waves of data collection.

Area deprivation

At each wave postcode data were linked to country-specific area-based measures of
multiple deprivation 60]. These were recoded into a binary measure of living in an area characterised by high
levels of local area deprivation based on living in an area in the lowest quintile
of neighbourhoods in a given country at all four waves of data collection (vs. not).

Approach to analysis

In the first stage of analysis we used simple bivariate descriptive statistics to
estimate the prevalence of overweight and obesity at ages five, seven and eleven for
children with and without intellectual disability. In the second stage of analysis
we used simple bivariate descriptive statistics to estimate the persistence of obesity
between successive waves of data collection for children with and without intellectual
disability.

In the third stage of analysis we examined bivariate associations between the predictor
variables listed above and obesity at age 11 in the full sample. Missing data on predictor
variables was imputed using multiple imputation routines in SPSS 20 to create five
parallel data sets. The results of pooled analyses were used to exclude predictor
variables from subsequent analyses if they had no statistically significant association
with child obesity. This led to the exclusion of all measures of maternal employment
and at Wave 4 indicators of having a TV in the child’s bedroom, the number of days
of physical activity and the main type of between-meal snack. In the fourth stage
of analysis we determined the strength and statistical significance of the bivariate
association between remaining predictor variables and child obesity at age 11 for
children with and without intellectual disability.

In the fifth stage of analysis we used multivariate logistic regression to determine
the unique strength and statistical significance of the association between predictor
variables that were either significantly associated or showed moderate or greater
effect sizes in relation to obesity at age 11 among children with intellectual disability.
In the final stage of the analysis we applied the model from the fifth stage to the
sub-sample of children without intellectual disability. The aim of this stage of the
analysis was to investigate whether there were any marked difference in the strength
of association between predictor variables and obesity that were significant for children
with intellectual disability when applied to children without intellectual disability.
We did not undertake any further analyses of the non-intellectual disability subsample
as data has been reported elsewhere on the full sample, the results of which are driven
by the non-intellectual disability subsample 49], 50], 52]. In the final two stages of the analyses missing data on predictor variables was
imputed using multiple imputation routines in SPSS 20 to create five parallel data
sets. The results of pooled analyses were reported.

All analyses used appropriate wave-specific weights to take account of biases in the
sampling frame (e.g., oversampling of households in Scotland, Wales and Northern Ireland),
initial recruitment and attrition over time.