The Interactive Role of Socioeconomic Status,Race/Ethnicity,and Birth Weight on Trajectories of Body Mass Index Growth in Children and Adolescents

Abstract

This study assessed how socioeconomic status (SES), race/ethnicity, and birth weight interacted to predict differential patterns of body mass index (BMI) growth among U.S. children born in the early 1990s. Three BMI growth trajectories emerged—one above the 50th percentile across the age range of 5 to 14, one in which children rapidly became obese before adolescence, and one where children started out and remained seriously obese. Hispanic and African American children were more likely to show accelerated patterns of weight gain as were those of lower SES and/or higher birth weights. However, SES interacted with both race/ethnicity and birth weight. For girls of all race/ethnicity groups tested, lower SES and higher birth weights predicted membership in the seriously obese BMI growth trajectory. For African American and Asian boys, however, the higher the SES the more likely they were to be on a trajectory for rapidly developing obesity by early adolescence.

Keywords

developmental trajectories ethnic/racial obesity/overweight/weight control socioeconomic status (SES)

The prevalence of childhood obesity in the United States has been increasing at an alarming rate (Danner, 2008; Ogden, Carroll, Curtin, Lamb, & Flegal, 2010), and childhood obesity dramatically increases the probability of obesity in adulthood (Guo, Wu, Chumlea, & Roche, 2002; Singh, Mulder, Twisk, van Mechelen, & Chinapaw, 2008). Furthermore, child and adolescent obesity is not just a predictor of later obesity but is associated with a host of health risks including hyperlipidemia, glucose intolerance, high blood pressure, asthma, sleep apnea, and depression (Baker, Olsen, & Sorensen, 2007; Dietz, 1998; Stunkard, Faith, & Allison, 2003). Therefore, it is critical to get a clear picture of the current scope of the problem and to better characterize the children most at risk for developing obesity so that interventions can be targeted appropriately.

There are a number of approaches one can use to identify these at-risk children and adolescents. One approach is to classify children and adolescents as overweight or obese if their body mass index (BMI) equals or exceeds that of the 85th or 95th percentile for their age and gender and then determine the characteristics of those children who exceed these cutoffs. Ogden, Flegal, Carroll, and Johnson (2002), for example, reported that 22% of Hispanic children and 20% of African American children aged 6 to 11 years were overweight compared with 14% of non-Hispanic White children. Similarly, studies have looked at associations between birth weight (Oken & Gillman, 2003) or measures of family socioeconomic status (SES; Whitaker & Orzol, 2006; Zhang & Wang, 2004) with overweight in children and adolescents. While traditional BMI percentile cutoffs and the BMI measure itself provide only crude approximations of body types and body fat distribution (Flegal, Tabak, & Ogden, 2006), they are generally accepted as indirect measures of adipose tissue in population-based studies (Dietz & Robinson, 1998; Lobstein, Baur, & Uauy, 2004).

A second approach is to use growth curve or multilevel modeling techniques (Raudenbush & Bryk, 2002; Singer & Willett, 2003) to show a single BMI growth trajectory, which is presumed to represent the average growth pattern in a sample. One can then assess how time-invariant variables (e.g., race/ethnicity or birth weight) or time-varying variables (e.g., hours of TV viewing or hours of exercise measured at multiple time points) influence this average trajectory, and estimate how well various combinations of child characteristics predict who is likely to exceed obesity cutoffs (cf. Cecil-Karb & Grogan-Kaylor, 2009; Danner, 2008). However, this approach assumes that there is only one growth trajectory around which individual children vary.

A third approach that does not assume one growth trajectory, but allows one to identify multiple underlying latent trajectories, is known as latent class multiple-groups growth curve modeling (Jones & Nagin, 2007; Muthén, 2004). Each underlying latent trajectory represents a group (or “class”) of individuals whose intercept (i.e., initial status), growth pattern, and estimated population prevalence differs. This latent class multiple-groups growth curves approach to the study of the development of obesity in children and adolescents is very promising because it can help identify groups of children whose patterns of BMI growth substantially differ.

Only a small number of studies have applied latent-class multiple-group growth curve modeling approaches to the study of obesity in children. Mustillo et al. (2003) identified four latent growth patterns of obesity in a sample of 991 White, 9- to 16-year-olds from rural North Carolina during 1993 to 2000. The largest group was consistently normal (i.e., was not obese at any time point from ages 9-16 years) and comprised approximately 73% of the sample. The second largest group (15%) was chronically obese during this age span. The final two groups were smaller and represented those who had normal weight at early ages and late onset obesity (7%), or early onset obesity and normal weight at later ages (5%). Similarly, Ventura, Loken, and Birch (2009) identified four latent patterns of BMI growth from a convenience sample of non Hispanic, White girls (n = 182) ages 5 to 15 years. Two groups were labeled as staying in the healthy range of BMI for the entire 10 years, one that tracked at the 50th percentile (37% of the girls), and one that hovered near the 60th percentile (29%) across this age span. However, a third group reached the 95th percentile and then fell back below it (20%), whereas a fourth group was consistently overweight and continued to accelerate well beyond the 95th percentile (14%). These studies are helpful in understanding the obesity problem, but they did not use nationally representative samples.

To date, only two large multiple-group latent class growth modeling studies of obesity have been published. Li, Goran, Kaur, Nollen, and Ahluwalia (2007) identified three distinctive patterns of obesity among a more representative sample of 1,739 children ages 2 to 12 years during 1986 to 2000. The largest group was consistently not obese (84%), whereas two smaller groups evidenced a pattern of either early (11%) or late (5%) onset of obesity. Similarly, Balistreri and Van Hook (2010) estimated latent trajectories of overweight and obesity among children who participated in the Early Childhood Longitudinal Study–Kindergarten Cohort (ECLS-K) by first categorizing the children as not overweight, overweight, or obese. They found evidence for a model that split children into three groups that were never overweight (59% boys, 55% girls), became overweight gradually (15% boys, 20% girls), or remained overweight throughout the duration of the study (27% boys, 25% girls). However, neither of these studies used the raw BMI to measure growth but instead used the probability of overweight/at risk for overweight (BMI > 85th percentile) across age.

These studies indicate that there are multiple patterns of weight gain in children and that those from some racial/ethnic groups are more likely to be obese by the time they reach adolescence. A better picture of the social-demographic characteristics of children with differential BMI growth patterns would help target intervention efforts. Furthermore, determining the timing of patterns of excessive BMI growth would indicate when interventions should begin.

The purpose of our study was to apply growth mixture modeling (GMM; Muthén & Muthén, 1998-2010) techniques to examine differential BMI growth patterns among children and young adolescents (age 5 to 14 years) from a nationally representative sample of U.S. children and to describe the social-demographic characteristics that are likely to put them at risk for developing obesity. Similar to previous studies (Balistreri & Van Hook, 2010; Li et al., 2007), we used GMM to model heterogeneity in BMI growth. The GMM approach allows for the estimation of distinct subgroup trajectories and their population prevalence. We chose to use GMM versus latent class growth analysis (LCGA) given its ability to estimate random variance terms around intercepts and slopes (i.e., not constrain them to zero), along with the estimation of residuals at each data-collection point (for a detailed discussion of differences between LCGA and GMM, see Peugh & Fan, 2010).

Although we used the same ECLS-K data set as Balistreri and Van Hook (2010), our study used the raw BMI scores versus a binary outcome for overweight status and the actual ages of the children (rather than grade levels) to identify latent trajectory groups which could then be overlaid on a graph of the 50th, 85th, and 95th percentile BMI score trajectories from Centers for Disease Control and Prevention (CDC) norms. The advantage of this approach is that it allowed us to more fully describe the paths of physical growth without relying too heavily on standard, but somewhat arbitrary BMI cutoff scores, which have changed in years past. This also means different standards or norms can be applied to our results. In addition, our study extends previous research by examining the extent to which SES, race/ethnicity, and birth weight interact to place children at increased risk for developing obesity by the time they reach adolescence. It is particularly important to consider interactions between these predictors as they have been shown to interact in complex ways in cross-sectional studies (Cramer, 1995; Kallan, 1993; Nepomnyaschy, 2009).

Method

Sample and Procedure

Data were derived from the ECLS-K that began in 1998 with a nationally representative sample of U.S. children followed Kindergarten through eighth grade (spring of 2007) by the National Center for Education Statistics. The cohort sample was selected from about 1,000 schools (i.e., public and private schools and all-day and part-day programs; ECLS-K, 2001). The sampling procedure involved a multistage probability sampling design whereby geographical regions such as counties or groups of counties were selected, then schools within counties, and ultimately children within schools. Specific details regarding the sampling procedure, study design, and measures are available through the National Center for Educational Statistics (NCES, 2010; see also ECLS-K, 2001).

Children were selected for inclusion if they were first-time kindergartners at the beginning of the study. Careful height and weight measurements were made on this cohort sample on six occasions (fall and spring of kindergarten, and spring of Grades 1, 3, 5, and 8). BMI values for five children (in Grade 8) were set to missing because the values recorded for their heights appear to be in error as they were well-below 3 ft which resulted in excessively high BMI values exceeding 100. Although a few BMI values across all six time points were rather low or high (full range was 5-53), the ECLS-K sample was initially selected to represent the entire population of U.S. kindergarten children and a few extreme BMIs are to be expected in such a large sample over a 10-year age range.

All analyses reported here used the appropriate longitudinal weights that compensate for attrition across waves of data collection, maintaining the national representation of this cohort of children who were first-time kindergartners in 1998 (ECLS-K, 2009). We conducted analyses comparing mean BMI for children used in the final data analysis to mean BMI for children not used in the final data analysis and did not find any difference in initial BMI for either boys or girls. Also, the amount of missing data across waves varied from 1.5% (Wave 1) to 8.5% (Wave 6) for both boys and girls. All missing data in the trajectory models were treated by Mplus (Muthén & Muthén, 1998-2010) as missing at random. We also descriptively examined the various missing data patterns and found no systematic pattern related to BMI scores (i.e., there was no evidence that missing data were conditional on overweight status).

Measures

Demographics

Gender, race/ethnicity, SES, birth weight, and birth dates were available in the database. The primary source of this information was the parent who, in approximately 90% of the cases, was the mother. Race/ethnicity categories were White, non-Hispanic; African American, non-Hispanic; Hispanic; Asian; Native Hawaiian or other Pacific Islander; American Indian or Alaskan Native; and more than one race specified, non-Hispanic. SES, computed at the household level, was based on the parent’s education, occupation, and household income and was recorded as a standardized score (ECLS-K, 2001). Age at each of the six assessment points was calculated (to an accuracy of 1 month) by subtracting the participant’s birth date from the assessment date.

Weight status

Birth weight of the child was reported in pounds and ounces by the child’s parent. Height and weight were measured twice at each of the six waves of data collection by trained assessors using the Shorr Board (accuracy = .01 cm) and a Seca digital bathroom scale (accuracy = .1 kg). A BMI for each assessment point was calculated for each child from the average of these two height and weight measurements as follows:

BMI = (average weight in kilograms) / {(average height in meters)}^{2}

A SAS program for the CDC growth charts (CDC, 2010) was used to calculate BMIs for each child.

Data Analysis Strategy

Data analyses proceeded in three steps. First, simple descriptive analyses were done of children’s BMI and obesity status in kindergarten and in Grade 8 as a function of gender, race/ethnicity, SES, and birth weight to obtain an overview of the possible influence of the obesity risk variables under consideration in the present study. Second, growth mixture modeling (GMM; Muthén & Muthén, 1998-2010) was used to model heterogeneity in BMI growth. There are multiple methods for determining the optimal number of latent groups. We relied on Bayesian Information Criterion (BIC), adjusted BIC, Akaike’s Information Criterion (AIC), the accuracies of Bayesian posterior probability classification, entropy, previous findings in the obesity-research literature, interpretability, and the relative size of each latent class. Other likelihood statistics, such as the bootstrap likelihood ratio test (BLRT), are sometimes used to determine the appropriate number of groups to select, but Mplus software does not currently allow this test when time (i.e., age in our case) is treated as random. This limitation may not be a problem, however, as a recent simulation study by Nylund, Asparouhov, and Muthén (2007) showed the BIC and adjusted BIC performed similar to the BLRT for identifying the correct number of classes for a GMM with continuous outcomes when N was large (i.e., N = 1,000).

The use of multiple pieces of evidence for identifying the number of latent trajectory groups is consistent with current recommendations (Nagin, 2005; Peugh & Fan, 2012) because no single piece of evidence is the definitive marker of a model with acceptable fit. AIC measures the goodness of fit of a model relative to other models, but includes a penalty for the number of model parameters. Similarly, BIC and adjusted BIC add in a penalty for sample size. For all three indices, values closer to zero indicate a better model fit. The posterior probability classification estimates each person’s probability for membership into each of the latent trajectory groups and the ideal is to correctly place all individuals with 100% accuracy. Nagin (2005) suggests that a reasonably good model is one in which the average of these Bayesian posterior probabilities across all groups exceeds .70. To assess the adequacy of the GMM solution we also estimated entropy, which is a standardized index of model-based accuracy, with values close to 1 indicating improved accuracy. Entropy is similar to average model Bayesian posterior probability. Based on the limited number of obesity studies using this analysis technique, we expected that a good model to data fit might include three to four latent-class growth trajectories, the largest of which would track in a range considerably below the 85th percentile. Our goal was to identify separate models for boys and girls with the fewest groups necessary to reasonably describe distinct and meaningful patterns in the data.

GMM analyses were conducted using the Mplus 6.1 software (Muthén & Muthén, 1998-2010). Mplus allowed us to use the actual ages at each assessment rather than assuming that all children were the same age at each wave of BMI data collection. This is important because there was considerable BMI growth across the age range of this study, ages varied at each data collection, and Mplus could account for variations in initial ages and ages at subsequent measurement occasions. Boys and girls were analyzed separately because normal BMI growth patterns differ by gender (NCES, 2009). Based on these trajectory patterns, we assigned participants to groups. We chose to use absolute assignment of participants to group based on a simulation study by Clark and Muthén (2009) that showed when entropy is .80 or greater the findings between most likely class membership versus using posterior probabilities were similar. In our study, entropy was .93 for both girls and boys.

In the final data analysis step, race/ethnicity, SES, birth weight, and interactions between these three variables were tested as predictors of placement in the different BMI growth trajectory groups, separately for girls and boys. A series of multinomial logistic regression models were tested, beginning with one which contained all two-way interactions. These analyses were done using STATA 10.1 software to take full advantage of the complex survey design and to more accurately estimate standard errors. To determine which sets of interactions improved trajectory group predictions, a series of adjusted Wald tests were conducted (rather than likelihood ratio tests), as recommended on the STATA website (www.stata.com/support/faqs/stat/lrtest.html).

Results

Descriptive Statistics

Table 1 presents kindergarten and Grade 8 BMI and obesity status as a function of gender, race/ethnicity, SES, and birth-weight category. Children were considered obese if their BMI exceeded the 95th percentile for their age and gender (NCES, 2009). More than 10% of the children were already obese in kindergarten; almost 20% by Grade 8. The actual numbers of children who were categorized as Hawaiian, Native American, or more than one race/ethnicity were quite small (70-150 in each category). As all subsequent analyses were done separately for boys and for girls and split the sample even further into separate growth trajectories, these three race/ethnicity categories were dropped due to small absolute cell sizes.

Table 1.

Children’s Kindergarten and Grade 8 Mean BMI and Percent Obese by Demographic Category.

		K		Grade 8
Demographic variable	N	Mean BMI	Percent Obese	Mean BMI	Percent Obese
Gender
Boys	3,320	16.31	11.4	22.44	20.2
Girls	3,280	16.24	11.6	22.96	16.9
Race/ethnicity
White	4,410	16.12	9.8	22.16	16.1
African American	530	16.48	12.5	23.67	21.6
Hispanic	1,020	16.60	15.3	23.49	22.5
Asian	290	16.01	12.5	21.89	17.2
Hawaiian	70	16.38	16.6	22.56	17.1
Native American	120	16.98	19.4	24.55	30.1
More than one	160	16.35	9.4	23.31	24.1
SES
Low (Quintile 1)	800	16.40	13.3	23.80	25.8
Middle (Quintile 3)	1,250	16.40	13.1	23.08	21.0
High (Quintile 5)	1,890	15.91	7.2	21.41	12.0
Birth weight
Low (<2,500 g)	430	15.62	6.8	21.56	14.1
Middle (>2,500 & <4,000 g)	5,160	16.20	10.5	22.58	17.8
High (>4,000 g)	810	17.24	21.6	24.56	29.1
Total for all children	6,600	16.28	11.5	22.70	18.6

Note: N = actual number of children. There were some missing data regarding race/ethnicity and birth weights. Means and percentages were calculated using ECLS-K longitudinal weights. Mean and standard deviations for birth weight were 7.38 lbs and 1.32 lbs, respectively. Standard classifications for low and high birth weight were used in this table.

Trajectory groups for girls

Given that prior researchers have explored and found three or four latent trajectories, we expected to have as many. We explored a two-group model as a baseline model to compare with the three-group model. Cubic polynomials were selected for each model because the general shape of BMI growth across childhood is cubic (NCES, 2009). Also, based on previous studies, we should roughly expect the majority of trajectories falling into an average trajectory and smaller percentages for other latent trajectory groups. Based on BIC, adjusted BIC, and AIC scores, the four-group model had the best fit. However, the four-group model resulted in one group with less than 1% of the sample. The three-group model was selected because it resulted in a distinct set of interpretable trajectories, had posterior probabilities approaching one (.87-.98), entropy of .93, had groups with adequate numbers (i.e., >1% of the sample), and was consistent with previous studies. Table 2 contains the parameter estimates for each group of girls and Figure 1 shows the trajectories for each group and reference lines for average (50th percentile), overweight (85th percentile), and obese (95th percentile) BMI levels (CDC, 2010). The three groups were characterized as follows: (a) 88.0% comprised an above average group, whose BMI growth tracked between the 50th and 85th percentiles; (b) 7.6% were identified as an obese group, who began at the 85th percentile, quickly exceeded the 95th percentile by age 10, and continued to stay obese through their early teens; and (c) 4.5% fell into a seriously obese group, who were already over the cutoff for obesity at age 5 and who accelerated toward very high levels of BMI by age 11 (i.e., BMI > 30).

Table 2.

Final Parameter Estimates for Group Trajectories for Girls.

Group	Parameter	Estimate	SE
1. Above average	Intercept	15.735	0.066***
	Linear	−0.075	0.054
	Quadratic	0.165	0.014***
	Cubic	−0.010	0.001***
2. Obese	Intercept	17.116	0.306***
	Linear	0.181	0.405***
	Quadratic	0.387	0.113
	Cubic	−0.022	0.007*
3. Seriously obese	Intercept	22.422	0.941***
	Linear	1.011	0.529
	Quadratic	0.155	0.141
	Cubic	−0.012	0.010

p < .05. ***p < .001.

Figure 1.

BMI growth trajectories for girls.

Trajectory groups for boys

The same analysis steps were followed as those used for girls. Based on adjusted BIC and AIC scores, the four-group model had the best fit, while BIC indicated the three-group model had the best fit. However, as was the case for girls, the four-group model resulted in one group with less than 1% of the sample, whereas the three-group model resulted in an interpretable set of trajectories, posterior probabilities approaching one (.83-.98), entropy of .93, and groups with adequate numbers (i.e., >1% of the sample). Therefore, based on interpretability, posterior probabilities, entropy, group sizes, and consistency with previous studies, the three-group model was retained for boys. Table 3 contains the parameter estimates identified for each latent group for boys whereas Figure 2 displays the trajectories for each group and reference lines for average, overweight, and obese BMI levels (CDC, 2010). We characterized the three groups as follows: (a) 89.4% comprised an above average group, whose BMI growth tracked just above the 50th percentile and began to approach the 85th percentile during their early teens; (b) 8.0% were in an obese group, who began just below the 95th percentile, quickly exceeded it by age 7, and continued to stay obese through their early teens; and (c) 2.6% fell into a seriously obese group, who were already well over the cutoff for obesity at age 5 and continued to accelerate toward very high levels of BMI by age 11 (i.e., BMI > 30).

Table 3.

Final Parameter Estimates for Group Trajectories for Boys.

Group	Parameter	Estimate	SE
1. Above average	Intercept	16.043	0.104***
	Linear	−0.426	0.205*
	Quadratic	0.241	0.017***
	Cubic	−0.014	0.001***
2. Obese	Intercept	16.914	2.879***
	Linear	2.044	0.867*
	Quadratic	−0.028	0.242
	Cubic	−0.006	0.014
3. Seriously obese	Intercept	23.377	1.941***
	Linear	0.486	0.524
	Quadratic	0.252	0.101*
	Cubic	−0.018	0.007*

p < .05. ***p < .001.

Figure 2.

BMI growth trajectories for girls.

Assessing Risk of Belonging to Overweight Trajectory Groups

For both girls and boys, Group 1 (the above average group whose trajectory was above the 50th percentile but never exceeded the 85th) served as the reference category for the multinomial logistic regression analyses.

Risks for girls

A multinomial predictive model containing all two-way interactions fit the data better than did one with only main effects, F(14, 360) = 2.10, p = .01. This indicates that at least some two-way interactions contributed to model fit. Neither race/ethnicity by SES nor race/ethnicity by birth weight interactions were statistically significant, but SES by birth weight was, F(2, 370) = 3.60, p = .03. Parameter estimates and statistical tests for girls’ risks of belonging to group trajectories that fell into the obese or seriously obese BMI range are presented in Table 4. For the obese group trajectory (Group 2), there were statistically significant main effects for race/ethnicity and SES; African American girls and those from lower SES families were more likely to be in this obese group trajectory than in Group 1. African American and Asian girls were more likely to be in the seriously obese trajectory (Group 3). In addition, the combination of relatively low SES and relatively high birth weights increased the odds of girls being in the seriously obese BMI trajectory group. The magnitude of this interaction between SES and birth weight was similar for all race/ethnicity groups. An illustration of this interaction (using Hispanic girls as an example) is presented in Figure 3.

Table 4.

Final Parameter Estimates and Tests for Risk of Belonging to Obese Trajectory Groups for Girls.

Group	Parameter	Estimate	SE	t	p
2. Obese	African American	1.220	0.281	4.35	<.000
	Hispanic	−0.182	0.287	−0.63	.527
	Asian	0.107	0.491	0.22	.826
	SES	−0.513	0.210	−2.44	.015
	Birth weight	0.127	0.130	0.97	.331
	SES × Birth weight	−0.205	0.156	−1.31	.191
	Constant	−2.913	0.186	−15.66	<.001
3. Seriously obese	African American	1.404	0.402	3.49	.001
	Hispanic	0.602	0.319	1.89	.060
	Asian	1.075	0.506	2.12	.034
	SES	−0.509	0.167	−3.04	.003
	Birth weight	0.430	0.117	3.68	<.001
	SES × Birth weight	−0.309	0.123	−2.51	.012
	Constant	−3.815	0.179	−21.36	<.001

Note: SES = socioeconomic status. Overall reference group is Group 1 (above average). Race/ethnicity reference group is White. SES and Birth weight are mean centered.

Figure 3.

Proportion of Hispanic girls predicted to be in seriously obese trajectory (Group 3) as a function of SES and birth weight.

Risks for boys

As with girls, a multinomial predictive model containing all two-way interactions fit the data better than did one with only main effects, F(14, 360) = 4.02, p < .001, and the SES by birth weight interaction also contributed to model fit, F(2, 380) = 3.60, p = .03. Contrary to the girls findings, the race/ethnicity by SES interaction did improve model fit, F(6, 380) = 3.56, p = .002. Results for the final model are presented in Table 5.

Table 5.

Final Parameter Estimates and Tests for Risk of Belonging to Obese Trajectory Groups for Boys.

Group	Parameter	Estimate	SE	t	p
2. Obese	African American	0.523	0.281	1.86	.064
	Hispanic	0.117	0.225	0.52	.603
	Asian	0.695	0.569	1.22	.223
	SES	−0.378	0.187	−2.03	.043
	Birth weight	0.308	0.064	4.79	<.001
	Afam × SES	1.233	0.372	3.31	.001
	Hisp × SES	−0.219	0.259	−0.85	.398
	Asian × SES	0.804	0.615	1.31	.192
	SES × Birth weight	0.295	0.085	3.47	.001
	Constant	−2.614	0.121	−21.54	<.001
3. Seriously obese	African American	0.245	0.572	0.43	.668
	Hispanic	0.716	0.393	1.82	.069
	Asian	0.517	0.719	0.72	.473
	SES	−0.713	0.238	−3.00	.003
	Birth weight	0.458	0.155	2.96	.003
	Afam × SES	1.520	0.387	3.93	<.001
	Hisp × SES	0.525	0.417	1.26	.209
	Asian × SES	0.330	0.376	0.88	.379
	SES × Birth weight	0.118	0.158	0.75	.453
	Constant	−3.932	0.240	−16.36	<.001

Note: SES = socioeconomic status. Overall reference group is Group 1 (above average). Race/ethnicity reference group is White. SES and Birth weight are mean centered. Afam = African American; Hisp = Hispanic.

In general, SES and birth weight were statistically significant negative and positive predictors, respectively, for being in the obese and seriously obese trajectories relative to the above average BMI trajectory, holding all other predictors constant in the model. However, the statistically significant association between SES and placement in Groups 2 and 3 must be interpreted with caution given the statistically significant two-way interactions with SES. Figure 4 illustrates the interaction between SES and race/ethnicity in which African American and Asian boys show a different pattern than do Hispanics and Whites. Figure 5 shows how the combination of SES and birth weight relates to placement in Group 2 for African American boys.

Figure 4.

Proportion of boys predicted to be in obese trajectory Group 2 as a function of the interaction of race/ethinicity and SES.

Figure 5.

Proportion of African American boys predicted to be in obese trajectory (Group 2) as a function of SES and birth weight.

Discussion

This study assessed the interactive associations between SES, race/ethnicity, and birth weight and differential patterns of BMI growth among a recent (1998-2007) nationally representative sample of U.S. children and adolescents born in the early 1990s. More than 10% of children were already obese in kindergarten; almost 20% by age 14. For both boys and girls, three distinct BMI growth trajectories emerged—one in which children were, on average, not obese across the age range of 5 to 14, one in which children rapidly became obese prior to early adolescence, and one where children started out and remained seriously obese. Although a large majority of children fell into the growth trajectory group that did not place them above cutoffs for overweight or obesity, even this group was, on average, well above the 50th percentile of CDC norms for BMI growth. These results suggest that the entire population distribution for weight gain among children and especially early adolescents has shifted upward since the last time BMI growth norms, based on National Health and Nutrition Examination Study (NHANES III) data collected in the 1990s, were published (CDC, 2010).

The study confirms and extends with a nationally representative sample what numerous smaller scale studies have suggested, namely, that there are a number of substantially different patterns of weight gain across childhood and a rapid growth in the percentage of children who are obese by early adolescence. In general, Hispanic and African American children were more likely than White and Asian children to show accelerated patterns of weight gain as were those of lower SES and/or higher birth weights. However, SES interacted with both race/ethnicity and with birth weight to predict BMI growth trajectories. For girls, the combination of low SES and high birth weight predicted higher BMI growth trajectories among all race/ethnicities tested. Surprisingly, this was not the case for boys. Higher SES African American and Asian boys, especially those with higher birth weights, substantially exceeded population base rates for placement in a trajectory that rapidly led to obesity prior to early adolescence.

The study has a number of strengths. First, it includes data from the most recent wave of data collection from a large, nationally representative cohort of children. The fact that the demographic risk factors it highlights are consistent with previous smaller scale studies is not surprising, but the levels of childhood overweight and obesity found and the extent to which these factors interact to predict it are. Second, it revealed some surprising anomalies regarding the prediction of SES onto BMI growth trajectories which appear to place high SES African American and Asian boys at relatively high risk for developing a pattern of rapidly accelerating BMI growth prior to adolescence. We did not expect to find such an association and cannot offer any convincing explanation for it. Further research on possible differential cultural and contextual norms concerning nutritional and behavioral practices associated with higher SES seems warranted.

There are also a number of limitations to this study. First, it relied on BMI as the major dependent variable and particular BMI percentile cutoff scores to characterize children and adolescents as overweight and/or obese. As noted earlier, BMI provides only a crude proxy for measurement of body fat and its distribution. One cannot assume that all children who have high BMIs have too much body fat or are otherwise unhealthy (Eissa et al., 2009; Freedman & Sherry, 2009). Furthermore, the standards used as a reference for normal and excessive BMI growth levels were based on U.S. norms which were published in 2000 and based on population samples measured several years before that.

A review by Wang and Lobstein (2006) indicates that there has been a worldwide acceleration of weight gain among children and adolescents that is particularly evident in developed countries and urbanized populations. Therefore, it is highly likely that when new U.S. norms are developed, children’s BMI growth curves will shift upward and the corresponding percentile-based BMI cutoffs for overweight and obesity will move up as well. The results of our study are consistent with this trend. As the study was based on a nationally representative sample of children and adolescents and used the 95th percentile as a cutoff for obesity, only 5% of the children should have been obese. Yet more than 10% were already obese in kindergarten and approximately 20% were by Grade 8. Unless overweight and obesity are defined away by continuing to raise BMI cutoffs, it would appear that they will soon become the new norm for children.

Another limitation of our study is that it relied on a simple model employing only three social-demographic variables. It was our intention to focus on these variables to determine, on a population basis, which children are most at risk for obesity by the time they reach adolescence. Other known predictors of children’s obesity such as maternal weight, daily nutrition, and activity levels were either not consistently available in the database or were collected in a manner that was insufficiently precise to develop credible measures. Therefore, our results show only which particular groups of children have trajectories of accelerated weight gain rather than why.

Finally, when conducting GMM, it is not uncommon to find two or three latent trajectories. Thus, we cannot be confident that exactly three groups in the sample clearly reflect the population or if these groups possibly reflect a more complex BMI distribution (Bauer & Curran, 2003). However, based on previous studies’ findings of multiple groups, we believe such a structure can exist. In addition, the predictions from the model we used are probabilistic (i.e., one cannot make confident predictions about any individual child’s specific developmental trajectory based on the demographic risk factors listed in the model nor does every child within any trajectory follow it exactly). The trajectories are, by definition, latent which means that they reflect statistical attempts to capture distinct patterns of growth within the population rather than individual variations within these patterns, and only some of those who fall into the highest risk categories are predicted to be overweight or obese. It would be particularly interesting to study high-risk children who do not follow the predicted path toward obesity to determine what they, their parents, their communities, or their schools are doing that promotes more healthy outcomes.

It is important to note that the longitudinal weights we used to compensate for missing data (i.e., those that dropped out during the course of the study) and to maintain national representation mean that our analyses can only be generalized to a cohort of U.S. children who entered kindergarten in 1998, children born nearly 20 years ago. Therefore, even though the ECKS-K database is one of the most up-to-date longitudinal studies available, the sample does not represent the current population of children. We expect that longitudinal studies which began more recently are likely to find even more children who are overweight or obese. If so, then the use of the word “epidemic” to describe the problem of childhood obesity does not appear to be an exaggeration. The fact that so many of the children in the present nationally representative study could already be categorized as obese when they were in kindergarten clearly suggests that interventions to reverse this unhealthy trend will need to begin well before the age of five.

Based on these findings and the analytic technique used, we would make the tentative inference that the latent trajectories for the variable obesity among children and adolescents aged 5 to 14 years, as measured by BMI, has a three-group structure. We were able to use social-demographic variables to predict membership in these three groups and to therefore indicate which children are most at risk of becoming obese by adolescence. Unfortunately, such variables are not subject to intervention. Future studies should apply this technique to study trajectories of growth based on measures that provide more valid indices of body fatness and health risks than does BMI and link differential growth patterns to behavioral variables, such as nutrition and activity levels, that are amenable to change.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Author Biographies

Fred W. Danner is professor and chair of the Department of Educational, School, and Counseling Psychology at the University of Kentucky. His areas of research interest include cognitive development, adolescence, health psychology, the relationship between fitness and intellectual performance, and the effects of sleep deprivation on learning and behavior.

Michael D. Toland earned a PhD in educational psychology (quantitative, qualitative, and psychometric methods) from the University of Nebraska–Lincoln in 2008. He is an assistant professor in educational psychology at the University of Kentucky. His research efforts focus on scale development and evaluation, item response theory, and multilevel modeling.

References

Baker

Olsen

Sorensen

(2007). Childhood body-mass index and the risk of coronary heart disease in adulthood. New England Journal of Medicine, 357, 2329-2337. doi:10.1056/NEJMoa072515

Balistreri

Van Hook

(2010). Trajectories of overweight among U.S. school children: A focus on social and economic characteristics. Maternal and Child Health Journal. doi:10.1007/s10995-010-0622-7

Bauer

D. J.

Curran

P. J.

(2003). Distributional assumptions of growth mixture models: Implications for overextraction of latent trajectory classes. Psychological Methods, 8, 338-363. doi:10.1037/1082-989X.8.3.338

Cecil-Karb

Grogan-Kaylor

(2009). Childhood body mass index in community context: Neighborhood safety, television viewing, and growth trajectories of BMI. Health and Social Work, 14, 169-177. Retrieved from http://www.naswpress.org/publications/journals/hsw.html

Centers for Disease Control and Prevention. (2010). CDC growth charts computer programs. Retrieved from http://www.cdc.gov/growthcharts/computer_programs.htm

Clark

Muthén

(2009). Relating latent class analysis results to variables not included in the analysis. Manuscript submitted for publication.

Cramer

(1995). Racial and ethnic differences in birth weight: The role of income and financial assistance. Demography, 32, 231-247. doi:10.2307/2061742

Danner

(2008). A national longitudinal study of the association between hours of TV viewing and the trajectory of BMI growth among U.S. children. Journal of Pediatric Psychology, 34, 1-8. doi:10.1093/jpepsy/jsn034

Dietz

W. H.

(1998). Health consequences of obesity in youth: Childhood predictors of adult disease. Pediatrics, 101, 518-525. Retrieved from http://pediatrics.aappublications.org/

10.

Dietz

Robinson

(1998). Use of the body mass index (BMI) as a measure of overweight in children and adolescents. Journal of Pediatrics, 132, 191-193. doi:10.1016/S0022-3476(98)70426-3

11.

Early Childhood Longitudinal Study–Kindergarten Cohort. (2001). User’s manual for the ECLS-K base year public-use data files and electronic codebook (Document No. 2001-029, revised). Retrieved from http://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2001029rev

12.

Early Childhood Longitudinal Study–Kindergarten Cohort. (2009). Combined user’s manual for the ECLS-K eighth-grade and K-8 full sample data files and electronic codebooks. Retrieved from http://nces.ed.gov/ecls/data/ECLSK_K8_Manual_part1.pdf

13.

Eissa

Dai

Mihalopoulos

Day

Harrist

Labarthe

(2009). Trajectories of fat mass index, fat free-mass index, and waist circumference in children: Project HeartBeat! American Journal of Preventive Medicine, 37(Suppl. 1), S34-S39. doi:10.1016/j.amepre.2009.04.005

14.

Flegal

K. M.

Tabak

C. J.

Ogden

C. L.

(2006). Overweight in children: Definitions and interpretation. Health Education Research, 21, 755-760. doi:10.1093/her/cyl128

15.

Freedman

D. S.

Sherry

(2009). The validity of BMI as an indicator of body fatness and risk among children. Pediatrics, 124(Suppl. 1), S23-S34. doi:10.1542/peds.2008-3586E

16.

Guo

S. S.

Chumlea

W. C.

Roche

A. F.

(2002). Predicting overweight and obesity in adulthood from body mass index values in childhood and adolescence. American Journal of Clinical Nutrition, 76, 653-658. Retrieved from http://www.ajcn.org/

17.

Jones

Nagin

(2007). Advances in group-based trajectory modeling and an SAS procedure for estimating them. Sociological Methods Research, 35, 542-571. doi:10.1177/0049124106292364

18.

Kallan

(1993). Race, intervening variables, and two components of low birth weight. Demography, 30, 489-506. doi:10.2307/2061653

19.

Goran

Kaur

Nollen

Ahluwalia

(2007). Developmental trajectories of overweight during childhood: Role of early life factors. Obesity, 15, 760-771. doi:10.1038/oby.2007.585

20.

Lobstein

Baur

Uauy

(2004). IASO International Obesity Task Force. Obesity in children and young people: A crisis in public health. Obesity Reviews, 5(Suppl. 1), 4-104. doi:10.1111/j.1467-789X.2004.00133.x

21.

Mustillo

Worthman

Erkanli

Keeler

Angold

Costello

(2003). Obesity and psychiatric disorder: Developmental trajectories. Pediatrics, 111, 851-859. doi:10.1542/peds.111.4.851

22.

Muthén

(2004). Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data. In Kaplan

(Ed.), Handbook of quantitative methodology for the social sciences (pp. 345-368). Newbury Park, CA: Sage.

23.

Muthén

L. K.

Muthén

B. O.

(1998-2010). Mplus user’s guide (6th ed.). Los Angeles, CA: Authors.

24.

Nagin

(2005). Group-based modeling of development. Cambridge, MA: Harvard University Press.

25.

National Center for Educational Statistics. (2009). National Center for Health Statistics, Centers for disease control growth charts: United States. Retrieved from http://www.cdc.gov/growthcharts/

26.

National Center for Educational Statistics. (2010). Early Childhood Longitudinal Study. Retrieved from http://nces.ed.gov/ecls/kindergarten.asp

27.

Nepomnyaschy

(2009). Socioeconomic gradients in infant health across race and ethnicity. Maternal and Child Health Journal, 13, 720-731. doi:10.1007/s10995-009-0490-1

28.

Nylund

Asparouhov

Muthén

(2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling, 14, 535-569. doi:10.1080/10705510701575396

29.

Ogden

Carroll

Curtin

Lamb

Flegal

(2010). Prevalence of high body mass index in U.S. children and adolescents, 2007-2008. Journal of the American Medical Association, 303, 242-249. doi:10.1001/jama.2009.2012

30.

Ogden

Flegal

Carroll

Johnson

(2002). Prevalence and trends in overweight among U.S. children and adolescents, 1999-2000. Journal of the American Medical Association, 288, 1728-1732. doi:10.1001/jama.288.14.1728

31.

Oken

Gillman

(2003). Fetal origins of obesity. Obesity Research, 11, 496-506. doi:10.1038/oby.2003.69

32.

Peugh

Fan

(2010). Growth mixture modeling: Concepts and implementation. Advances and Applications in Statistical Sciences, 3, 255-284. Retrieved from http://www.mililink.com/journals_desc.php?id=41

33.

Peugh

Fan

(2012). How well does growth mixture modeling identify heterogeneous growth trajectories? A simulation study examining GMM’s performance characteristics. Structural Equation Modeling: A Multidisciplinary Journal, 19, 204-226.

34.

Raudenbush

Bryk

(2002). Hierarchical linear models: Applications and data analysis methods (2nd ed.). Thousand Oaks, CA: Sage.

35.

Singer

Willett

(2003). Applied longitudinal data analysis: Modeling change and event occurrence. New York, NY: Oxford University Press.

36.

Singh

Mulder

Twisk

van Mechelen

Chinapaw

(2008). Tracking of childhood overweight into adulthood: A systematic review of the literature. Obesity Review, 9, 474-488. doi:10.1111/j.1467-789X.2008.00475.x

37.

Stunkard

A. J.

Faith

M. S.

Allison

K. C.

(2003). Depression and obesity. Biological Psychiatry, 54, 330-337. doi:10.1016/S0006-3223(03)00608-5

38.

Ventura

Loken

Birch

(2009). Developmental trajectories of girls’ BMI across childhood and adolescence. Obesity, 17, 2067-2074. doi:10.1038/oby.2009.123

39.

Wang

Lobstein

(2006). Worldwide trends in childhood overweight and obesity. International Journal of Pediatric Obesity, 1(1), 11-25. doi:10.1080/ 17477160600586747

40.

Whitaker

Orzol

(2006). Obesity among U.S. urban preschool children: Relationships to race, ethnicity, and socioeconomic status. Archives of Pediatric and Adolescent Medicine, 160, 578-584. doi:10.1001/archpedi.160.6.578

41.

Zhang

Wang

(2004). Socioeconomic inequality of obesity in the United States: Do gender, age, and ethnicity matter? Social Science and Medicine, 58, 1171-1180. doi:10.1016/S0277-9536(03)00288-0