Introduction

Childhood obesity has become a global epidemic. It is seriously affecting the young generation’s health and future, as well as social and economic developments in many countries worldwide [1,2,3]. Over the past two decades, the prevalence of childhood overweight and obesity has been rising in many countries [4, 5]. Worldwide 42% of adults and approximately one-fifth of children are overweight or obese [5]. The US leads other countries in youth overweight/obesity rates, with 34.5% of 12–19-year-olds are overweight or obese in 2011–2012; childhood obesity is more common among certain racial and ethnic groups such as Hispanic boys and Black girls [6].

Ethnic disparities in childhood body mass index (BMI) trajectory may have their origins in the earliest stages of life [7,8,9,10,11]. In previous research, household socio-economic status (SES), family rules for children’s behavior, physiological/ genetic differences and early exposure in utero have been studied as predictors of ethnic disparities in childhood obesity but mainly separately and using small study samples, or/and cross-sectional data. No studies could identify individual’s BMI milestones from infancy to adolescence and use nationally representative data [12,13,14,15,16] due to difficulties in modeling fluctuations in childhood BMI curve and limited data with repeated anthropometric measurements during childhood. No study has examined ethnic disparities in childhood obesity and related causes using recent, nationally representative data in the US.

This study aimed to (1) provide recent estimates of sex- and ethnic-specific childhood BMI trajectories and dynamic of obesity status with age in the US; and (2) study causes of ethnic disparities in childhood BMI trajectories and obesity rates.

Methods and materials

Study design and study sample

Longitudinal data from two US nationally representative cohorts, the Early Childhood Longitudinal Study-Birth Cohort (ECLS-B) and the ECLS Kindergarten Class of 1998–99 (ECLS-K) were used. The ECLS-B followed 10,700 children born in the US in 2001 through kindergarten entry (at 9 months, 2–4–5–6 year; 5 waves); ECLS-K 1998–99 followed 21,410 kindergarteners from kindergarten to middle school (at kindergarten fall and spring, 1–3–5–8th grade; 6 waves) in diverse socio-economic and ethnic backgrounds. The present study included children with available information on sex, ethnicity, and one or more times of height and weight measurements (The final sample size was 29,250; of them 9,350 [88% of total subjects] from ECLS-B and 19,900 [93% of total subjects] from ECLS-K). All reported numbers about sample size were rounded to the nearest 50 for ECLS-B and 10 for ECLS-K according to the confidentiality policy of the US Department of Education.

Assessment and measures

Child demographic characteristics

We classified child ethnicity as non-Hispanic White, non-Hispanic African American (AA), Hispanic and Asian. For purposes of ethnic-specific analysis and to ensure sufficient statistical power, we excluded those children with multiple races or rare race (less than 3% of total population, including Native Hawaiian, other Pacific Islander, American Indian, or Alaska Native).

BMI and milestones of BMI trajectory characteristics

At each wave in ECLS-B or ECLS-K, each child’s weight and height were measured twice by trained test administrators during the direct child assessments to reduce measurement errors. Children’s BMIs were obtained as composite variables of ECLS [17] and we treated BMI < 10 or > 50 as implausible value/outliers for the analysis. Overweight/obesity for children aged ≥ 2 years was defined as the BMI ≥ 85th percentile for the same sex and age in the 2000 US CDC Growth Charts [18].

Childhood BMI trajectory is not easy to describe due to the fluctuation in childhood BMI curve with simple parameters. However, the major milestones–infancy peak (at the end of the first rapid increase of BMI; age at the infancy peak represents tempo of growth [19], and high infancy peak indicates accelerated postnatal growth and an accumulation of fat [20]) and adiposity rebound (AR, the nadir before the second rise in BMI [21]; early AR reflects accelerated growth and is associated with an increased risk of overweight [22]) has helped to describe the critical events of children’s growth, such as the time of high peaks and nadir and the degree of changes between the high and low points in previous research [12, 13, 15]. We used four parameters to characterize the two milestones in childhood BMI trajectory: (1) age and (2) BMI and (3) velocity (defined as difference in BMI divided by difference in age), and (4) area under curve (AUC, defined as the definite integral between the two age points; indicating child’s cumulative exposure to excessive body weight and predicting later risk of obesity and obesity-related diseases [14]) between infancy peak, AR and 16 years of age (the end point of data). According to previous findings, (1) earlier age at AR, (2) higher BMI at AR, (3) steeper velocity, and (4) wider AUC from AR to adolescence were defined as unfavorable BMI trajectory, which are associated with high risk of later obesity [12,13,14,15,16].

Household SES

We used the derived variable of ECLS SES scale Z-score, which incorporates information on parents’/primary caregivers’ education, occupation, and household income at the household level with mean of 0 and a standard deviation of 1. The SES scale Z-score reflects the household SES at the time of data collection.

Parenting practice

ECLS-B 2-year collection and the ECLS-K kindergarten spring collection shared three survey questions about parenting practice that may be related to childhood obesity. Parents/primary caregivers reported in a typical week: (1) whether they have routines about what time the child goes to bed (yes/ no); (2) which television programs the child can watch (yes/ no); and (3) the number of days the child has evening meals at a regular time. Accordingly, we defined three parenting practice based on whether the child had a regular bedtime, rules about TV program selections, and dinner at a regular time ≥ 5 days/week (according to the assumption that parents might keep regular dinner on weekdays and the distribution of children’s response).

Statistical analysis

First, we described socio-demographic characteristics of children and their families and presented mean BMI and prevalence of overweight and obesity by cohort, wave, sex, and ethnicity using SAS surveymeans and surveyfreq procedures to accounting the multistage stratified clustered sampling scheme in two ECLS cohorts with longitudinally consistent weights (Tables 1 and 2; Appendix 2).

Table 1 The US children’s mean BMI and overweight/obesity prevalence (%) by cohort and wave during 1998–2008: the Early Childhood Longitudinal Study-Birth Cohort (ECLS-B) and the ECLS Kindergaten class of 1998–99 (ECLS-K) combined
Full size table
Table 2 Prevalence of overweight and obesity (%) of the US children during 1998–2008 by wave, sex, and ethnicity: the ECLS-B and ECLS-K
Full size table

Second, to define sex- and ethnic-specific children’s BMI growth curve, we pooled the ECLS-B and K data sets. We reasoned that combining the ECLS-B and ECLS-K data sets could yield a full perspective on childhood BMI trajectory from infancy to adolescence, given their similarities in nationally representative sampling as well as weight and height measures. And we used the best combination of fractional polynomial functions (using diverse combinations of age terms—Agepj; powers pj were selected from a fixed set of eight candidate values, including –2, –1, –0.5, or log, 0.5, 1, 2, and 3) from our previous work using a separate data set with more frequent BMI measures from ages at 1 week to 18 years old: [14] age(-2), age(-1), age(-0.5), log(age), age0.5, and age [14], mainly because it used frequent data points (at least a total of 18 BMI measures) for each child through the whole childhood from 1 week to 18 years. The best combination of fractional polynomial functions from our previous work is valid to reflect the whole childhood BMI trajectory among general US children. Appendix 1 presents the model fit with current data as predicted (curves) vs. observed (circles) childhood BMI trajectories.

Then a specific fixed effect of each fractional polynomial term was estimated, which reflected the population-average trend for each of the eight subgroups by child sex and ethnicity (i.e., White boys/girls, AA boys/girls, Hispanic boys/girls, and Asian boys/girls). With the subgroup-specific fixed beta coefficients for fractional polynomial terms, we fit and plotted eight BMI trajectory curves to present sex- and ethnic-specific predicted mean BMI by age (in months, Fig. 1). To estimate each child’s BMI trajectory characteristics from his or her fitted mixed model we included the random effects of fractional polynomial terms with autoregressive correlation structure (the lowest BIC value among several candidate structures) to reflect his or her deviation from his or her sex and ethnic group average. From the model, we derived the parameters (age, BMI, velocity, and area under the curve) of two milestones for each individual and compared their mean levels by sex-ethnic groups (Table 3). To facilitate estimation of the two BMI milestones related parameters, we chose fractional polynomial model over other alternatives such as piecewise linear model and cubic spline model. We recognized its limitation of unstraightforward interpretation of the polynomial terms, which was not the focus of this study.

Fig. 1
figure 1

Sex- and ethnic-specific predicted mean BMI from 7 months to 16 years among the US children: the ECLS-B and ECLS-K cohorts (n = 29,250). Data collected during 1998–2008 from ECLS-B and ECLS-K were pooled and analyzed. The mixed effect model with fractional polynomial functions was used for the sex- and ethnic-specific BMI growth curves with considering the complex sample design of ECLS. The US CDC 2000 Growth Charts, the WHO child Growth Standards 2006 (birth to 5 years old) and the WHO Reference 2007 (5 to 19 years old) were used as references: 1) In White boys, expected value of BMI = 4465.1 – 9232.0*age(-2) + 11,336.0*age(-1) – 11,197.0*age(-0.5) – 1085.7*log(age) + 183.2*age0.5 – 2.7*age; 2) In AA boys, expected value of BMI = 1735.6 – 2614.1*age(-2) + 3822.1*age(-1) – 4132.2*age(-0.5) – 433.0*log(age) + 76.2*age0.5 – 1.1*age; 3) In Hispanic boys, expected value of BMI = 3733.7 – 7070.9*age(-2) + 9025.0*age(-1) – 9191.7*age(-0.5) – 924.0*log(age) + 161.7*age0.5 – 2.5*age; 4) In Asian boys, expected value of BMI = 4435.5 – 7541.3*age(-2) + 10,357.0*age(-1) – 10,815.0*age(-0.5) – 1109.1*log(age) + 198.3*age0.5 – 3.1*age; 5) In White girls, expected value of BMI = 2685.6 – 5699.2*age(-2) + 6795.1*age(-1) – 6702.3*age(-0.5) – 653.0*log(age) + 110.3*age0.5 – 1.6*age; 6) In AA girls, expected value of BMI = 4913.2 – 10,888.0*age(-2) + 12,809.0*age(-1) – 12,426.0*age(-0.5) – 1188.8*log(age) + 198.6*age0.5 – 2.9*age; 7) In Hispanic girls, expected value of BMI = 5647.5 – 12,254.0*age(-2) + 14,575.0*age(-1) – 14,236.0*age(-0.5) – 1374.2*log(age) + 233.0*age0.5 – 3.5*age; 8) In Asian girls, expected value of BMI = 1901.8 – 3475.6*age(-2) + 4615.0*age(-1) – 4689.2*age(-0.5) – 459.0*log(age) + 75.4*age0.5 – 1.0*age)

Full size image
Table 3 Mean (SD) of individual BMI trajectory characteristics of the US children by sex and ethnicity: based on pooled data from the ECLS-B and ECLS-K cohorts
Full size table

Based on the causal step method for the mediation criteria [23], a substantial mediation effect was defined as a 10% or greater reduction in the estimated coefficient for ethnic disparities in a BMI trajectory characteristic after adjusting for a potential mediator. Since parenting practice for the child’s regular meal times and TV program selections were not significantly associated with milestones of childhood BMI trajectory in bivariate regression analysis (p > 0.1), they were not included in the final model. Therefore, we assessed the mediating roles of household SES and parenting practice for child’s bedtime on the ethnic disparities in major milestones of BMI trajectory characteristics with a series of multivariable linear regression models with different sets of covariates: model 1 adjusted for cohort effect, child sex, number of siblings, and family structure; model 2 additionally adjusted for household SES level after model 1; model 3 additionally adjusted for parenting practice for child’s regular bedtime after model 1; model 4 simultaneously adjusted for all covariates (Table 4 and Appendix 3). Each parameter of BMI trajectory characteristics had a separate model to test the mediation effect and ethnic dummy variables were used to compare ethnic difference in BMI trajectory characteristics. Analyses were conducted using SAS 9.3 (SAS Institute, Cary, NC).

Table 4 The roles of household SES and parenting practice in the ethnic disparities in selected childhood BMI trajectory characteristics at and after adiposity rebound: based on pooled data from the US ECLS-B and ECLS-K cohorts
Full size table

Results

Sample characteristics at baseline

Sex and ethnicity distributions were similar between the two cohorts, but ECLS-K had a higher proportion of full-time working mothers (46.9% vs. 32.6%) and a lower proportion of families under the federal poverty threshold (19.5% vs. 23.4%) than ECLS-B (Appendix 2).

Mean BMI and overweight and obesity by cohort and survey wave

Table 1 shows overall mean BMI and overweight and obesity prevalence by cohort and wave. Mean BMI of ECLS-K increased from 16.3 (2.2) kg/m2 at Kindergarten fall to 23.0 (5.2) kg/m2 at 8th grade. ECLS-B did not show a clinically considerable change in mean BMI (between 16.7 kg/m2 at preschool and 17.9 kg/m2 at 9 months).

Overweight and obesity rates ranged from 25.6% to 38.9% by age and cohort; the 5th graders had the highest prevalence (38.9%, Table 1). Hispanic boys and AA girls had the highest prevalence of overweight and obesity in most of the waves (e.g., Hispanic boy: 52.5%, AA girl: 49.1% at 5th grade [around the age of 11]) (Table 2).

Sex- and ethnic-specific BMI trajectory

Figure 1 shows sex- and ethnic-specific BMI trajectory curves from ages of 7 months to 16 years by the best fit mixed effect models using pooled data from the two cohorts (N = 29,250).

The BMI trajectories of different ethnic groups crossed each other several times between 7 months of age and 16 years old due to different ages of infancy peak and AR and BMI growth velocity between milestones. However, after AR, ethnic disparities on BMI trajectory became clear.

Overall, Hispanic boys and AA girls maintained the highest mean BMI after AR than did others of the same sex. The BMI curve of Asian boys showed a level-off tendency at 13.6 years of age.

Compared to the US CDC 2000 Growth Charts (2 to 20 years of age), the WHO 2006 Child Growth Standards (birth to 5 years of age), and WHO Reference 2007 (5–19 years of age), our fitted childhood BMI trajectories of US children were higher after infant peak.

Table 3 shows the diverse characteristics of BMI trajectory by sex and ethnicity. Hispanic boys and AA girls had a similarity of early AR (mean (SD) = 57.9 (7.8) and 59.0 (7.2) month) and high AUC (2724.5 (489.8) and 2681.2 (426.7) kg/m2*month) from AR to 16 years of age. AAs had the steepest increase in the BMI growth after AR (boy: 6.2 (1.5); girl: 7.0 (1.5) 10-2 kg/m2/month). Asian boys and girls had the earliest infancy peak compared to other ethnic groups. Their velocity and AUC from AR to 16 years of age were lower than those of Hispanic and AA children.

Factors associated with ethnic disparities in childhood BMI trajectory characteristics

As indicated in Table 4, overall, household SES level and child’s regular bedtime significantly influenced child BMI trajectory characteristics. Children with higher household SES and regular bedtimes were more likely to have later AR (e.g., 1.3 (0.3) month in the 3rd tertile vs. the 1st tertile in household SES; 1.4 (0.5) month when having bedtime rule vs. no rule, both p < 0.05; Model 4), lower BMI at AR, lower velocity and AUC between AR and 16 years of age.

After adjusting for covariates, AA children had the earliest age of AR and steepest BMI growth, and Hispanic children had the largest AUC between adiposity and 16 years of age (Model 1). Adjustment of household SES and to a lesser extent, a family routine for child’s bedtime, led to substantial reductions in ethnic disparities in childhood BMI trajectory characteristics (Models 2–4 in Table 4). These reductions were consistently the largest in Hispanic children (Appendix 3). For example, their 16.9% of age at AR difference, 38.5% of velocity difference and 18.5% of AUC difference between AR and 16 years of age compared to those in White children were attenuated by household SES and child’s regular bedtime (AA: 8.0, 5.5, 11.0%, respectively, Appendix 3).

Asian children had the latest age at AR and lowest BMI growth velocity from adiposity to 16 years of age. The differences in these characteristics between Asians and whites were not mediated by household SES or child’s regular bedtime (Table 4 and Appendix 3).

Discussion

For the first time, using unique data from two large nationally representative cohorts, we constructed the recent US children’s BMI trajectory curves from ages of 7 months to 16 years. We reported considerable ethnic disparities in BMI trajectories and obesity rates. Hispanic boys and AA girls maintained the highest overweight and obesity rate and mean BMI after AR vs. others of the same sex. The ethnic disparities in BMI trajectory characteristics became attenuated significantly after adjusting for individual household SES level and child’s regular bedtime. In particular, the difference between White children and Hispanic children was largely mediated by household SES. Childhood BMI levels including obesity status track into adulthood and childhood BMI affect later chronic disease risks [24,25,26].

However, due to the nature of fluctuation in childhood BMI curve and the lack of data sets with repeated anthropometric measurements from infancy to adolescence, only few studies have been able to identify individual’s BMI milestones in relation to later risk of obesity, including infancy peak, AR, and some features of BMI change [12,13,14,15,16]. Evidence from these studies has indicated that children with early AR, fast increase in BMI and longer exposure to high BMI are at increased risk for subsequent obesity. In particular, AAs and Hispanics showed this adverse characteristics in BMI trajectory from early ages [7, 14, 27], and have been identified as the ethnic groups with the highest risk of childhood obesity in the US [28,29,30].

Our study also found that AA children had all of the adverse BMI trajectory characteristics after adjusting for the effects of child sex, household SES level, children’s regular bedtime, and other covariates. In contrast, as the second most adverse group for childhood BMI trajectory characteristics, Hispanics had the largest attenuation of the differences between their BMI trajectory characteristics and those of White children after adjusting for the effects of household SES and the child’s regular bedtime (e.g., 38.5% of Hispanic-White difference in BMI growth velocity from AR to 16 years old).

Moreover, for all ethnic groups, household SES had a larger mediating contribution on BMI disparities than did child’s regular bedtime. Previous evidence also supports that ethnic disparities in obesity could be substantially due to household SES status (e.g., 38% of Black-White differences in risk of overweight/obesity was reduced after adjustment for educational level and income) [31], especially in Hispanic children (e.g., attenuated AAs-Whites and Hispanics-Whites difference in BMI Z-score after adjusting for parental and household socio-economic factors: Hispanics: 37.7%, AAs: 27.1%) [7]. A US study reported that among the independent contribution of socio-demographic, cultural and family routine factors to observed disparities in childhood BMI and obesity, disparities for Hispanic children were explained by socio-economic disadvantage [32]. Further studies are needed to examine the underlying mechanisms of adverse BMI trajectory characteristics in AA children.

We found that Asian American children had the healthiest BMI trajectory among all ethnic groups in the US. They had the latest AR and slowest BMI growth from AR to 16 years of age. Other studies also reported that Asian children had the healthiest BMI in the US [27, 33]. In particular, Asian boys show decreased BMI at the end of the BMI curve (0.03 lower after 190 months compared to 189 months in our study). Similarly, Chinese boys’ BMI curve has a lower and level-off tendency after age 15 [34]. However, the relatively small sample size of Asian boys of older ages in our sample may affect the robustness of our results.

In general, our findings suggested that low-SES US children had a higher risk for developing obesity [10, 11]. We also found parenting practices on children’s bedtime seemed to affect childhood BMI in all ethnic groups. Another longitudinal study of US children found that low parental income and education were risk factors for gradual onset of overweight after beginning kindergarten (OR (95% CI) of overweight: 0.91 (0.82–1.0) by logged base year income; 0.59 (0.47–0.74) in college vs. high school degree) [10]. Parental education can be related to health knowledge and lead to better daily routines for children (e.g., appropriate bedtime, balanced meals) [35]. On the other hand, the potential influences of household income could be bidirectional: families with greater household incomes may spend more money on children’s access to healthier foods and recreational activities, or they may buy a greater variety of foods and smart screen devices for their children, with the latter causing excessive calorie intake and more sedentary time [36]. Nevertheless, results from previous studies suggest that childhood BMI is affected by household SES [7, 11]. Efforts to reduce ethnic disparities in obesity should focus on preventing early life risk factors related to socio-demographic characteristics and parenting practices.

All of our sex- and ethnic-specific mean BMIs for these ECLS US children are higher than those of the 2000 US CDC Growth Charts (describe the growth of children in the US based on data collected during 1963–1994 [37]), the WHO 2006 Child Growth Standards (based on data collected from six countries during 1997–2003 with optimal growth environment [34]) and WHO Reference 2007 (developed based on data collected in the US and the WHO 2006 Child Growth Standards [37]). US children have relatively higher mean BMIs than those of the old reference/standards, which were comprised of earlier birth cohorts and breastfed children in other countries. In addition, different modeling methods could cause differences between results for the current study and the previous reference/standards (which used a variant of the lambda-mu-sigma (LMS) statistical method with data specific statistical smoothing techniques [37]).

This study had some limitations, including no consideration of biological, cultural or environmental determinants of obesity, considerable attrition with child age in both cohorts, and few parenting practices as risk factors for obesity disparities. Another limitation was the use of combined household SES Z-scores within the study population, which did not allow us to estimate the individual effects of parental education, occupation, and household income independently.

Nevertheless, this study is the first to define ethnic disparities in childhood BMI trajectory with visualization as well as parameterized BMI trajectory characteristics using the most recent nationally representative cohort data for the US, which yielded a large analytic sample. Also, we found the mediating effects of parenting practice (i.e., regular bedtime) and household SES level on ethnic disparities in child BMI trajectory, which could potentially contribute to prevention and intervention efforts to reduce ethnic disparities in childhood obesity.

In conclusion, among children in the US, ethnic disparities in obesity rates and BMI trajectory are large and become apparent after adiposity rebound. Hispanic boys and African-American girls have unfavorably higher BMI and obesity rates when compared to other groups. Household SES level and parenting practice seem to contribute to the disparities. Efforts to reduce disparities in childhood obesity should focus on modifiable risk factors including parenting.

Disclaimer

The content of the paper is solely the responsibility of the authors and does not necessarily represent the official views of the funder.