Distinct patterns of socio‐economic disparities in child‐to‐adolescent BMI trajectories across UK ethnic groups: A prospective longitudinal study

Summary Background In many high‐income countries, body mass index (BMI)/obesity levels are inversely associated with socio‐economic position (SEP). Little is known whether socio‐economic patterns in BMI trajectories throughout childhood differ by ethnicity, especially in the United Kingdom. Objectives To investigate socio‐economic disparities in child‐to‐adolescent BMI trajectories and risks of overweight and obesity during adolescence across ethnic groups. Methods Mixed‐effects fractional polynomial and multinomial regression models were applied to estimate socio‐economic differences in BMI trajectories (3‐14 years) and risk of overweight/obesity at 14 years, respectively, in the UK Millennium Cohort Study (n = 15 996). Analysis was stratified by ethnicity. Result Poverty was associated with higher BMI in children of White and South Asian origins, with a small difference at 3 years, which widened with age to 0.75 kg/m2 (95% CI, 0.59‐0.91) and 0.77 kg/m2 (0.26‐1.27) at 14 years for the White and South Asian groups, respectively. There was a reverse income‐BMI association in children of Black (African‐Caribbean) origin with the poverty group having a lower BMI (−0.37 kg/m2 [−0.71 to ‐0.04] at 5 years; −0.95 kg/m2 [−1.79 to −0.11] at 14 years). These patterns also presented with maternal education as a SEP indicator and for obesity at 14 years. Conclusions Socio‐economic advantage may not be universally associated with lower BMI, which should be considered when planning obesity interventions. The positive SEP‐BMI association in children of Black origin requires replication and merits further investigation into underpinning mechanisms.


| INTRODUCTION
Obesity is a global public health challenge. 1 In many higher-income countries, levels of body mass index (BMI) 2,3 and obesity 4 are socially patterned, with disadvantaged population groups having higher mean BMI and more likely to be affected by obesity, possibly because of their disproportionally greater exposure to risk factors such as consumption of energy-dense foods. 5 Recent evidence also suggests that socio-economic differences in BMI have widened across generations and are emerging at younger ages. 6,7 High BMI in childhood tends to track into adulthood, and adult obesity is associated with a number of health outcomes, 8

especially cardiometabolic diseases.
Recent research suggests that the association between socioeconomic position (SEP) and BMI in children may differ by ethnic group. Studies, mainly from the United States, showed that the inverse SEP-BMI association in children and adolescents is less strong for Asian American populations and inconsistent for Hispanic and non-Hispanic Black populations, compared with non-Hispanic White populations. [9][10][11] The differential associations may be attributable to that cultural, environmental and biological factors related to obesity development have different socio-economic patterns across ethnic groups. 12 Little evidence is available in in children in the United Kingdom (UK). One study used data from the National Child Measurement Programme and found that the variation in BMI by area deprivation group is smaller in the South Asian and Black groups than in the White group in London. 13 A recent cross-sectional analysis of the UK Millennium Cohort Study (MCS) 14 showed the relationship between low SEP and increased risk of overweight/obesity at 7 years in White children reversed in Black African/Caribbean children. Examination of these patterns throughout childhood will not only lend greater support to these findings (if replicated) but also point towards when (and thus, potentially, why) these differences occur and whether they persist into adolescence.
Minority ethnic populations represent a growing group in the United Kingdom and are projected to make up a fifth of the population by 2051. 15 It is therefore necessary to gain a better understanding of how socio-economic disadvantage impacts on adiposity at different stages in childhood, rather than at one age, across different ethnic groups to provide information for public health policies and targeted interventions. Using longitudinal data from a UK national prospective cohort study, we aimed to investigate (a) socio-economic disparities in child-to-adolescent BMI trajectories across ethnic groups and (b) whether similar patterns of socio-economic disparities were also found for risks of overweight and obesity during adolescence.

| Subjects
The MCS is a nationally representative cohort study, which included over 19 000 children who were born between September 2000 and January 2002 and were living in the United Kingdom at the age of 9 months. The first contact was carried out when participants were 9 months old (baseline). They were followed up at ages of 3, 5, 7, 11, and 14 years. Ethics approval as well as informed consent in writing from parents (and from participants themselves as they were getting older) were obtained. 16 The MCS oversampled children who lived in the less advantaged socioeconomic circumstances and in England those from minority ethnic backgrounds. Details of the study design are described elsewhere. 17 We included singletons from White, South Asian, and Black African-Caribbean backgrounds with at least one BMI measurement at follow-up visits. Participants whose ethnicity was "mixed" (n = 510, 3.0%), "others" (eg, Chinese) (n = 297, 1.8%), or missing (n = 11, 0.07%) were excluded, resulting 16 082 children available for this analysis (eligible sample). After removing participants who had missing data on exposure variables (n = 86), a total of 15 996 participants (99% of eligible sample) with 62 051 BMI measurements were included in the study sample.

| Exposure: family SEP indicators
Information on family income was collected at baseline during parental interview and weighted using Organisation for Economic Co-operation and Development (OECD) scales to take into account family size. 16 Poverty was defined as OECD equivalized family income below 60% of national median household income-a commonly used measure of relative poverty. 18 At baseline interviewers, mothers were asked to self-identify their highest academic qualification from the list-"higher degree," "first degree," "diploma in higher education," "A/AS/S levels," "O level/GCSE grades A-C," "GCSE grades D-G," "other academic qualifications (including overseas)," and "none of these qualifications." Maternal education level was categorized as "higher (GCSE grades A*-C & above)," "lower (GCSE grades D-G & below)," and "others." General Certificate of Secondary Education (GCSE) is a subject-specific qualification in the United Kingdom typically taken by students at ages 14 to 16 years.

| Outcomes: BMI between 3 and 14 years
Height and weight were measured with light clothing and without shoes by trained interviewers at follow-up visits when children were 3, 5, 7, 11, and 14 years old. 19 Height was measured to the nearest 0.1 cm using Leicester Stadiometers with heels to the back of the base plate and head in the Frankfurt Plane position. Weight measurements were measured to the nearest 0.1 kg by Tanita BF-522 W scales. BMI (kg/m 2 ) at each age was derived. Weight status at 14 years was categorized as "normal" (including thinness), "overweight" (not including those with obesity), and "obesity" using International Obesity Task Force (IOTF) BMI-for-age cut-offs, 20 and alternatively World Health Organisation (WHO) BMI-for-age cut-offs. 21

| Ethnicity
We considered ethnicity to be a potential effect modifier of the relationship between poverty and childhood BMI. Participants' ethnicity was defined by parents at baseline using the 2001 UK Census ethnicity classes and subsequently grouped as "White," "South Asian (Indian, Bangladeshi and Pakistani)," and "Black African-Caribbean (Black African and Black Caribbean)".

| Statistical analyses
Mixed effects fractional polynomial models were applied to capture the non-linear trends of BMI changes with age between 3 and 14 years. We considered both second-order and third-order fractional polynomials. The third-order fractional polynomials fitted the data better (Table S1), and the best-fitting powers were age 2 , age 2 * log(age), and age 3 . Mixed effects models were used to take into account correlations of BMI measurements within individuals and permit the inclusion of cases with missing BMI measurement at some ages under a missing at random assumption. 22 Random effects were used for age 2 and age 3 . Unstructured covariance matrix for the random coefficients was used. Inclusion of an additional random effect for age term (ie, age 2 * log(age)) led to nonconvergence. The assessment of model residuals is provided in Table S2. Model selection was guided by deviance, Akaike and Bayesian Information Criterion statistics.
The first model included sex, fractional polynomial age terms, poverty, and the interaction between poverty and each age term. To assess whether the relationship between income poverty and BMI differed by ethnic group, we tested the interaction between ethnicity and poverty. As the P value for ethnicity-poverty interaction was less than.01, the analysis was stratified by ethnicity (ie, the model was repeated for each ethnic group separately).
We also applied multinomial logistic regression models to examine the association between income poverty and overweight/obesity at 14 years. We estimated relative risk ratios (RRRs) and 95% confidence intervals (CIs) for overweight or obesity (vs normal weight) in the poverty group, compared with the nonpoverty group. The models included age at measurement and sex. In addition, to assess whether BMI disparities differed across the BMI distribution, we applied quantile regressions to BMI at 14 years to estimate differences in BMI centiles (ie, 10th, 25th, 50th, 75th, and 90th) between poverty and nonpoverty group by ethnicity, adjusting for sex and age at measurement. We examined whether the relationship between income and BMI was non-linear by repeating quantile regressions using the continuous income variable and its quadratic term (instead of the binary poverty variable). There was little evidence of nonlinear associations in all ethnic groups (P value for quadratic term = 0.01 in White, 0.73 in South Asian, and 0.14 in Black African-Caribbean). Both multinomial logistic regression and quantile regression models were weighted to take into account clustered sampling design and attrition at 14-year visits.

| Sensitivity analyses
We conducted several additional analyses. First, we repeated mixed effects models using the mother's highest educational level as an alternative SEP indicator. Since the size of the "others" maternal education group was small, for plotting purpose, we estimated BMI differences between "higher" and "lower" maternal education groups.
Second, we repeated mixed effects models for boys and girls separately to compare socio-economic disparities between sexes. BMI is positively correlated with height in children and adolescents. 23 In our study, children from the nonpoverty group were found to be taller than those from the poverty group in all ethnic groups. Therefore, we additionally adjusted for height in the mixed effects model to examine whether some of the socio-economic differences in BMI was explained by socio-economic differences in height.
Analyses were conducted in Stata V.15.1 (Stata Corp., College Station, Texas) and the R software environment for statistical computing V3.6.1. 24 The following Stata commands were used in this analysis: fp for fitting optimal fractional polynomial functions, mixed for fitting mixed effects models, and svy: mlogit for weighted multinomial logistic regression. The R packages survey 25 and quantreg 26 were used to fit weighted quantile regression and estimate bootstrap variances.

| Participant characteristics by socio-economic groups
Of the included 15 996 children, 86.5% were from White, 10.0% from South Asian, and 3.5% from Black African-Caribbean ethnic backgrounds ( Table 1). The percentage of children living in families with relative poverty was higher in the South Asian (64%) and Black African-Caribbean (59%) groups compared with the White group (32%). Mothers of ethnic minorities were more likely to be in the "lower (GCSE grades D-G & below)" educational group than those of White ethnic group (South Asian 47.6%, Black African-Caribbean 39% vs White 26.8%).

| Socio-economic inequalities in BMI trajectories according to ethnicity
Overall, children of South Asian ethnic origin had the lowest mean BMI compared with those of White and Black African-Caribbean origins. Trajectories of mean BMI differed between poverty and nonpoverty groups, and the patterns of these socio-economic disparities varied by ethnicity (Figure 1). In the White group, a BMI difference between poverty and nonpoverty groups was  Table S3.

| Overweight and obesity at 14 years
Overall, the prevalence of overweight and obesity was 31% in the poverty group and 24% in the nonpoverty group (35% and 29%, respectively, when using WHO references). Table 2

| Sensitivity analyses
Socio-economic patterns in BMI trajectories across ethnic groups largely remained when using maternal education level as an alternative family socio-economic indicator. However, the increment in estimated BMI differences with age was smaller, and the standard errors of the estimates were greater for minority ethnic groups ( Figure S1). Analysis further stratified by sex revealed the same socio-economic patterns for boys and girls. Estimated BMI differences between poverty groups were slightly greater in girls than in boys for White and South Asian groups but were similar between boys and girls in the Black African-Caribbean group ( Figure S2).
However, the 95% CIs of these estimates for boys and girls overlapped across all ages. The 95% CIs were markedly wider F I G U R E 2 Estimated mean body mass index (BMI) difference and 95% confidence intervals (95% CI) between income poverty and nonpoverty (reference) groups. Models were stratified by ethnic group and included sex, poverty, age terms, and poverty-age interactions

| DISCUSSION
In this contemporary UK national cohort, our main findings include (a) child-to-adolescent BMI trajectories were socio-economically patterned, and the pattern differed between ethnic groups. Income deprivation was associated with higher BMI in the White and South Asian groups but with lower BMI in the Black African-Caribbean group.
(b) A difference in BMI between poverty and nonpoverty groups was established as early as 3 years in the White and South Asian groups, and overall increased with age across all ethnic groups. (c) Similar socio-economic patterns presented when using maternal education as the alternative SEP indicator in sensitivity analyses and were found for the risk of obesity at 14 years.
Our findings on socio-economic disparities in BMI in the White and South Asian ethnic groups are consistent with existing evidence from the general population in the United Kingdom. We showed that a modest difference in mean BMI between poverty and nonpoverty groups was established at as early as 3 years and increased with age.
This is in line with previous studies, which suggested that social differences in BMI emerged at younger ages in the general population 6,7 and widened with age. 27 One UK study (approximately 96% participants from White ethnic background) found that a socio-economic difference in BMI presented at about 4 years and became greater with age (0.4 kg/m 2 for boys and 0.9 kg/m 2 for girls at 10 years). 27 The differences in factors such as dietary between social groups from a young age may have are likely to largely explain these BMI differences. However, as detailed dietary measures were not available in the MCS, we were not able to fully investigate this. In addition, we found, as suggested by previous studies, 6,28 that BMI disparities at 14 years were greater at higher end of the BMI distribution in the White and South Asian groups.
Limited studies have investigated socio-economic disparities in BMI trajectories in UK children across ethnic groups. Findings from US studies are mixed-some studies showed a negative 29,30 or no association 9,31,32 between SEP and BMI/obesity for non-Hispanic Black children, other studies reported a positive association as shown in our study. 10,33 A recent analysis of the MCS at age 7 showed that White children in lower income families (bottom 40%) were at increased risk of overweight/obesity at 7 years, compared with their counterparts in higher-income (top 60%) families; the relationship was reversed for Black African/Caribbean children. 14 Our study supports these findings and, in addition, demonstrates that these differences emerged at around age 6 years and persisted to 14 years. Similar patterns were also seen for obesity at 14 years. BMI in children and adolescents is positively correlated with height. 23 Although mean height of Black African-Caribbean children in the nonpoverty group was greater than that of their counterparts in the poverty group, the socio-economic pattern in BMI was not explained by socio-economic difference in height (data not shown addressing socio-economic disadvantage will benefit the health and well-being of these families in many ways, but other approaches may be needed to reduce the higher rates of BMI/obesity observed among children of Black African-Caribbean origin.
Our analysis benefited from using family-level socio-economic indicators collected during early childhood, which have been suggested to be more accurate than those retrospectively collected. 40 We used repeated measures of BMI, which allowed us to explore the age at which socio-economic differences in BMI emerge and how they change with age. However, a few limitations need to be noted.
Despite using a large national study, which oversampled minority ethnic groups, the sample sizes of South Asian and Black African-Caribbean groups are relatively small, which contributed to the wider confidence intervals of estimates in these groups, especially in quantile regression analyses and when analyses were further stratified by sex. Previous research has shown that BMI differences among relatively heavy children are largely driven by differences in fat mass. 41 A social gradient was observed in children's fat mass but not in lean mass or trunk fat mass. 42 BMI does not distinguish between fat mass and lean mass, 23 and there are known ethnic differences in body composition in UK children with those of Black African-Caribbean origin having a lower level of body fat at a given BMI compared with those of White and South Asian origins. 43 The present study was not able to ascertain whether the socio-economic pattern in BMI in the Black African-Caribbean group was primarily attributed to differences in lean mass or fat mass. Mothers of minority ethnic groups were more likely to self-report their education level in the "others" group, which potentially captured a range of different qualifications obtained overseas. Therefore, the estimated BMI differences between lower and higher maternal education groups among children of ethnic minorities in the sensitivity analysis may be underestimated. Detailed information on dietary intakes from a young age was not collected in the MCS, which prevents further investigation on whether social patterns in diet also differed across ethnic groups.

| CONCLUSIONS
Socio-economic disadvantage may not be universally associated with higher child and adolescent BMI. Among children of Black African-Caribbean origin, we found that poverty was associated with lower