Background
Obesity is a serious public health problem which adversely affects children’s health [
1]. Childhood obesity is related to physical diseases, such as cardiovascular disease [
2‐
4], type 2 diabetes [
3,
5], some cancers [
6], stroke [
2], arthritis [
6], sleep apnea [
7], early adult mortality [
8], and poorer mental health [
9,
10], e.g. negative self-image [
11] and peer perceptions [
12].
Although the prevalence of childhood obesity may have leveled off among U.S. children in recent years [
13,
14], it remains high: currently 34 % of school-aged children 6 to 11 years old were overweight or obese [
13]. Children who were overweight or obese were more likely to be overweight or obese in adulthood [
6,
15‐
18]. Little is known, however, about body mass index (BMI) developmental trajectories during childhood. Approximately 15 % were chronically obese from 9 to 16 years old, while 7 % became obese in adolescence [
19]. A semi-parametric clustering procedure has categorized children using annual BMI into trajectory groups [
20,
21]. Among Taiwanese elementary school children, four groups were identified with persistent relative weight status over time. Lin et al. [
22], we applied this relatively new statistical procedure to semiannual elementary school children data in a community in the southwestern U.S.
Seasonal variation has been demonstrated in children’s BMI or standardized BMI z-score (BMIz): children’s overweight or obesity status increased during the summer months of the elementary school years [
23]. Overweight and ethnic minority children gained more weight in the summer [
24]. Weight status transition probabilities also differed by demographics, i.e., boys had higher probability of transitioning into the overweight/obese category than girls [
25].
Gender differences have been reported in trends in children’s weight status [
13,
26‐
28]. Ethnic disparities emerged at very young ages [
29‐
31] and racial differences were found in longitudinal tracking patterns [
24,
25,
32,
33]. Low SES groups disproportionately became obese [
29,
34‐
36]. To our knowledge, no research has examined childhood BMI developmental trajectories among US elementary school children using semiannually collected data. This secondary analysis aimed to: 1) replicate in a US sample of elementary school children from 5 to 12 years old previous trajectories in BMIz scores over a five-year period; 2) identify the number and type of distinct categorical BMI percentile (overweight/obese versus others; obese versus others) trajectories; and 3) examine whether these trajectories were associated with gender, ethnicity, or socioeconomic status (SES). The identification of childhood weight status trajectory groups and their demographic correlates could inform obesity prevention interventions.
Discussion
This secondary analysis identified distinctive weight status trajectory groups for children by using GBTM to identify groups of children with different probabilities BMIz score and of overweight or obesity across the 5 years of assessments during elementary school.
Use of BMIz has been preferred over BMI percentile in longitudinal population-based analyses [
43,
47]. Six stable trajectories were found when assessing BMIz over 11 time points. In contrast to our findings, among similar aged children from Canada [
48] a four trajectory solution was determined with increasing BMIz for the lower three trajectories and stability only for the highest trajectory. Seasonal differences were not possible to detect in their biannual and annual data. Their high stable trajectory reflected higher child overeating, more mothers smoking during pregnancy, and more rapid weight gain in infancy. These variables were not available in our data set, but need to be assessed in future research.
In our GBTM model assessing the probability trajectories of overweight and obesity (BMI ≥ 85%ile), five overweight/obesity status trajectories were identified: 1) 51 % of the children who had a near zero probability of being overweight/obese at any time point, 2) 22 % who had a very high probability of being overweight/obese across all five years, 3) 10 % who increased their probability of overweight/obese starting at the summer after second grade, 4) 9 % transitioned their probability of overweight/obese in the early elementary school, and 5) 8 % who decreased their probability of overweight/obese (e.g. transitioned into healthy weight status). In the GBTM model assessing the probability trajectories of obesity (BMI ≥ 95%ile), three-quarters of children were classified as having a near zero probability of being obese at any time point (e.g. persistently non-obese weight) and 13 % made up each of the remaining two trajectory groups.
With the GBTM model for overweight or obesity (BMI
≥ 85%ile), the finding of five trajectory groups was different from the previously identified four obesity trajectories (boys: normal or slightly underweight, persistently normal weight, overweight becoming obese, and persistently obese; girls: persistently obese, persistently overweight, persistently normal weight, and persistently slightly underweight [
22]) or three obesity trajectories for both boys and girls (gradual onset of overweight/at risk of overweight, always overweight/at risk of overweight, and normal weight [
49]). Furthermore, our results identified two weight status trajectory groups which differed on when the children’s probability of overweight/obesity transitioned (starting in the summer right after Kindergarten or another right after 2
nd Grade). Some events (e.g., becoming healthy) were sufficiently rare that were not captured by the semiparametric mixture models in previous studies. Since the other study used data collected annually or biennially, it is possible that our larger sample or the more frequent semiannual longitudinal data enabled us to detect more nuanced trajectories. With the GBTM model for obesity (BMI
≥ 95%ile), one previous study also identified a three-trajectory group model, but different trajectory patterns (early onset overweight, late onset overweight, and never overweight [
20]) using a different modeling approach (i.e., latent growth mixture modeling). The three trajectory groups (persistently healthy weight, becoming obese, and chronically obese groups) in this study were also inconsistent with two other prior findings identifying four-trajectory group models (no obesity, chronic obesity, childhood obesity, and adolescent obesity [
19]; chronically obese, decreasing, increasing, and non-obese [
21]). It is not clear if the number of data points, method of analysis, characteristics of sample population, or some other factor, accounted for these differences.
Previous studies reported seasonal differences in child’s weight or BMI [
23‐
25,
50‐
55]. Significant differences in BMIz score in our study were found between the school year and summer months, but the pattern varied by trajectory group, with decreasing probability of increasing BMIz during summer months than in the school year in the lowest two trajectory groups for the first summer between Kindergarten and first grade. Future research will need to replicate these trajectories and assess seasonal differences in obesogenic behaviors and family and other influences on those obesogenic behaviors.
Inconsistent findings have been reported on gender differences in weight status trajectory groups. When assessing the probability of being overweight or obese (BMI
≥ 85%ile), one study showed that gender differed by obesity trajectory [
21,
22]; however, another study indicated no gender difference [
56]. The present study revealed that girls were less likely to be in the early onset overweight/obese, late onset overweight/obese, or becoming healthy trajectory groups, similar to a prior finding which utilized a slightly different developmental range (ages 7–12) [
19,
21,
22]. In the models assessing the probability of obesity (BMI
≥ 95%ile) the finding in this study was in line with past research for adolescents (ages 6–18) [
21,
22], indicating that boys were more likely to be chronically obese or becoming obese during both elementary schools and adolescence. Ethnic and school SES differences by overweight/obese or obese trajectory groups were consistent with past research. When BMI 95%ile was the cut-point, non-Hispanic White children were less likely to be chronically or becoming obese than non-Hispanic Black and Hispanic [
21]. Ethnic differences among overweight/obese trajectory groups (BMI 85%ile cut-off) were not examined in previous studies.
Bivariate chi-square analysis revealed that overweight/obese or obese trajectory significantly varied among SES groups, which was consistent with previous findings, indicating an inverse relationship between the probability of becoming or remaining overweight/obese and SES [
19,
57]. However, the effect was no longer significant after adjusting for other demographics. This was in contrast to that of previous studies [
19,
22]. For BMI 85%ile cut-point, high SES (higher income or higher education level) was less likely to follow a becoming obese trajectory group than a persistently healthy weight trajectory group [
21,
22]. For BMI 95%ile cut-point, children from low SES families were more likely to be in the chronic obesity or childhood obesity trajectory groups [
19].
We previously applied different statistical procedures to analyze these data: mean change in BMIz over time [
24], and Markov category transition probabilities over time [
25]. Each procedure provided different insights into the dynamics of BMI change. The current procedure may provide the most interesting results for targeting interventions. Children who were chronically overweight and/or obese likely require an obesity treatment intervention while children who have a high probability of becoming overweight/obese at earlier (9.2 %) or later (9.7 %) points in elementary school may benefit from an obesity prevention intervention if they can be identified before or early in their transition. More research is needed to identify what factors (e.g., diet, physical activity, sedentariness, sleep) may account for who belongs in these groups, and how soon the differences emerge, so that interventions can be targeted to what may be expected to have the most preventive effect.
A strength of the current study was the use of objective measures with a large sample over five years, and the use of GBTM analysis. GBTM is conceptually similar to
K-means cluster analysis, but has several advantages including measurement invariance [
58], no prior specification of the number of groups to be extracted, and rigid model selection criteria. Alternatively, GBTM assumes zero within-class variances implying the trajectories of the individual-level group members are not allowed to vary about the group’s mean trajectory and the individual level heterogeneity can be expressed in terms of group differences [
40,
59], and does not take into account cluster sampling [
60]. Compared to several national surveys, the current sample size is relatively small; however, sample sizes of 500 can yield good estimates in the use of GBTM [
61]. Limitations of the present study include an a priori assumption of the existence of distinct trajectories. Over- and under- fitting models were identified, implying the identified trajectories may merely reflect random variation, but unusual patterns were not identified. Trajectory assignments were based on probability of BMI %ile status category and are not absolutes [
40,
59]. The findings are limited by the attrition and missing data which are relatively common in longitudinal studies. In addition, the sample was restricted to one school district in southeast Texas in the United States. For example, the summer changes may have been due to unusual heat or severe weather. Therefore, our findings may not be generalizable to the population of U.S. elementary school aged children, but the prevalence of obesity was similar.
Acknowledgement
The authors thank Sandy Bristow, Sonya Kaster, RD, LDN, SNS, and Thomas R. Woehler, MD, from the Oliver Foundation for their dedication to improving the health of children. The Oliver Foundation assisted with study concept and design and assisted in supervision and implementation of the study procedures. They did not participate in analysis or interpretation of the data. DW from the Oliver Foundation provided a critical review of the manuscript.