Body Mass Transitions Through Childhood and Early Adolescence: A Multistate Life Table Approach

Abstract

The growing prevalence of overweight and obesity among children is well documented, but prevalence estimates offer little insight into rates of transition to higher or lower body mass index (BMI; weight (kg)/height (m)²) categories. We estimated the expected numbers of years children would live as normal weight, overweight, and obese by race/ethnicity and sex, given rates of transition across BMI status levels. We used multistate life table methods and transition rates estimated from prospective cohort data (2007–2013) for Denver, Colorado, public schoolchildren aged 3–15 years. At age 3 years, normal-weight children could expect to live 11.1 of the following 13 years with normal weight status, and obese children could expect to live 9.8 years with obese status. At age 3 years, overweight children could expect to live 4.5 of the following 13 years with normal weight status, 5.1 years with overweight status, and 3.4 years with obese status. Whites and Asians lived more years at lower BMI status levels than did blacks or Hispanics; sex differences varied by race/ethnicity. Children who were normal weight or obese at age 3 years were relatively unlikely to move into a different BMI category by age 15 years. Overweight children are relatively likely to transition to normal weight or obese status.

The prevalence of overweight and obesity among children and adolescents in the United States has increased steadily over the past 30 years, with recent studies suggesting that the trend may be reaching a plateau (1, 2). As of 2011–2012, 15% of children aged 2–19 years were overweight and 17% were obese (2). The prevalence of obesity is higher among non-Hispanic blacks and Hispanics than among non-Hispanic whites and Asians, and it is slightly higher among males than among females, although sex differences vary across racial/ethnic groups (2–5). Overweight and obesity in childhood have been linked to diabetes (6), cardiovascular disease (7), and premature mortality (8–10) in adulthood.

Most existing research on overweight and obesity among children and adolescents provides prevalence estimates. However, prevalence estimates offer little insight into incidence rates for new cases of overweight or obesity (i.e., transitions from normal weight to overweight status or from overweight status to obesity), or rates of “recovery” (i.e., transitions from obesity to overweight status or from overweight to normal weight status). Children and adolescents, however, experience rapid increases in height and weight due to physical maturation and behavioral factors, including changing exercise and dietary patterns (11). Policies and interventions that seek to reduce excess body mass among children and adolescents could benefit from a better understanding of how long children and adolescents can expect to live with normal weight, overweight, or obese status, given transition rates to higher and lower body mass index (BMI) levels.

The few studies that have examined the incidence of obesity in children have found that incidence is higher in boys than in girls and higher among blacks than among whites or Hispanics (12–14). Gordon-Larsen et al. (12) conducted the only study (to our knowledge) that included Asians and that estimated transitions from obese status to nonobese status, but they focused on adolescents and young adults rather than younger children, who have different growth patterns, and they did not distinguish between normal weight and overweight status. To our knowledge, no existing studies have estimated the number of years children can expect to occupy specific BMI categories.

Multistate life table methods allowed us to use age-, sex-, and race/ethnicity-specific rates of transition across BMI categories to estimate the number of years of living at each status level. By allowing for recovery, multistate life tables provide more valid estimates of the duration of time for which children can expect to remain at a given BMI level than models that only consider transitions to higher BMI categories (12–16). We aimed to estimate the number of years schoolchildren in Denver, Colorado, could expect to spend in each of the normal-weight, overweight, and obese BMI categories between the ages of 3 and 15 years, by race/ethnicity and sex.

METHODS

We used data from a cohort of prekindergarten through ninth grade students attending Denver Public Schools (DPS) who had height and weight measurements taken by school nurses during routine health assessments from 2007 to 2013. DPS is a large, racially/ethnically diverse, urban school district that served over 87,000 students from prekindergarten through 12th grade during the 2013/2014 school year. Approximately 70% of DPS students qualify for the free or reduced-price lunch program for low-income students. Children aged 5 years or younger were representative of those who subsequently enrolled in kindergarten in DPS, due to subsidized prekindergarten programs. DPS is among a handful of school districts nationally that have succeeded in systematically collecting longitudinal measurements of height and weight and thus BMI (17–20).

Our study included students who were enrolled in DPS during at least 2 school years between fall 2007 and spring 2013 and had at least 2 BMI observations. We included children aged 3–15 years who were non-Hispanic white, non-Hispanic black, non-Hispanic Asian, or Hispanic. We excluded 783 children who were in the 9th grade and were aged 16 years or older—children who failed to advance and were atypical of 9th grade students in DPS. Because of sample sizes that were too small to provide stable estimates, we excluded 448 American Indians/Alaska Natives, 128 Native Hawaiians/Pacific Islanders, and 2,035 multiracial children. Our final sample included 65,672 students.

Given the longitudinal structure of our data, individual students were observed (i.e., had their heights and weights measured) multiple times throughout the study period. For each observation, we flagged BMI values that were implausible based on the 2000 Centers for Disease Control and Prevention (CDC) Growth Charts (21). A macro in the CDC Growth Charts flags values that are “too low” or “too high” to be considered biologically plausible. Of those flagged as implausible, chart reviews indicated that some of the especially high BMI values were consistent with a child's prior or subsequent height and weight measurements; we retained those records in the analyses. We excluded the remaining 1,313 (0.7%) observations with implausible BMIs, resulting in an average of 2.7 observations for each student during the study period. The primary determinant of the number of observations for each child was the study design—children who were older in earlier years of the study aged out of the sample, and children who entered prekindergarten in later years were eligible for fewer observations before the study ended. Notably, students in our analytical sample had similar distributions of race/ethnicity, sex, and age as children in the same grades in DPS.

Variables

Our outcome variable was BMI status. Nurses measured height and weight during routine hearing and vision screenings for prekindergarten, kindergarten, first, second, third, fifth, seventh, and ninth grade students. Nurses used a standardized screening protocol to weigh and measure children and to record results in a school records database. We calculated BMI as weight (kg) divided by the square of height (m²) and used a standard SAS program (SAS Institute, Inc., Cary, North Carolina) to calculate BMI percentiles by age and sex (21). For each observation, we classified participants as normal weight if their BMI was below the 85th percentile, overweight if their BMI was between the 85th and 95th percentiles, and obese if their BMI was at or above the 95th percentile.

Our covariate data came from school records, as reported by parents. Sex was a dichotomous variable. Race/ethnicity was categorized as non-Hispanic white, non-Hispanic black, non-Hispanic Asian, or Hispanic. Age was coded as years (i.e., age in months divided by 12).

Event-history analysis

We used discrete-time event-history models to estimate annual incidence rates for 4 possible transitions: 1) from normal weight to overweight, 2) from overweight to normal weight, 3) from overweight to obese, and 4) from obese to overweight. A few students (0.60%) appeared to transition directly from normal weight to obesity or from obesity to normal weight between observations. We assumed that those children had transitioned to overweight status halfway through the interval before moving on to normal weight or obesity, because transitions between normal weight and obesity require a child be overweight sometime in the interim.

We created a person-period data set to estimate our discrete-time event-history models. Person-period data allowed us to flexibly account for unequal observation windows across respondents as they entered or exited the risk set at different times (22). After children entered the study, they contributed observations to the risk set until July 2013 (the end of the follow-up period), until they left DPS, or upon completion of the ninth grade, after which DPS no longer collected data on student height and weight. We used logistic regression to predict the risk of becoming overweight among children who were normal weight and the risk of becoming overweight among children who were obese. Because children who were overweight were at risk of 2 outcomes, we used multinomial logistic regression to predict their risk of becoming either normal weight or obese (22, 23).

In our models, age was a time-varying factor. Because the risk of a BMI transition may vary nonlinearly with age, we compared models (not shown) that raised age to each of and then each pair of the following exponents: 0.5, 1, 1.5, 2, 2.5, and 3 (24). Raising age to the exponent of 0.5 best fitted the data based on the Bayesian Information Criterion (22). We also fitted models that tested for all 2-way and 3-way interactions between age, race/ethnicity, and sex. We found support for models that included all 2-way interactions (i.e., age ×sex, age × race/ethnicity, and sex × race/ethnicity) but not 3-way interactions. Interactions for age × race/ethnicity were significant in all models, although interactions for sex × race/ethnicity and sex × age were only significant in the logistic regression models for transitions to overweight (from either normal weight or obese status). To preserve assumptions about the associations between the covariates and the outcomes across the models, we included interactions for sex ×race/ethnicity and sex × age in the multinomial logistic regression model for transitions from overweight to either normal weight or obese status.

Multistate life tables

We used multistate life table methods to calculate the number of years between ages 3 and 15 years for which a typical child would occupy each BMI status, conditional on BMI status at age 3 years. The number of years for which a typical child occupied a given BMI status depended on the age-, sex-, and race/ethnicity-specific rates of transition to higher or lower BMIs estimated from our event-history analyses (16). Multistate life tables also typically incorporate the risk of transitioning to death. Mortality rates vary across the age ranges examined and by sex and race/ethnicity (25). Because DPS does not collect vital status information on children, we used age-, sex-, and race/ethnicity-specific mortality rates from the CDC's Wide-Ranging Online Data for Epidemiologic Research (WONDER) system (26). We assumed that mortality rates were the same regardless of BMI because we are not aware of any data on BMI-specific death rates among children. We used mortality rates from the 2006–2010 US population because 5 years’ worth of data provided stable incidence rates by single year of age and by race/ethnicity and sex, because 2010 was the most recent year for which data were available, and because the 2006–2010 data were centered over the years for which we had BMI observations.

Formulas for the calculation of multistate life tables are presented in the Web Appendix, available at http://aje.oxfordjournals.org/. Although students must have had at least 2 height and weight observations to be included in our analyses, none were observed continually from age 3 years to age 15 years. Thus, our estimated transition rates incorporated both within-individual (over time) and between-individual variability (27). We estimated 95% confidence intervals around our estimated life expectancies by simulating 5,000 sets of rates from our event-history models, calculating our multistate life tables for each set of values, and using the 2.5th and 97.5th centiles of our estimated life expectancies as the lower and upper bounds of our confidence intervals, respectively (28, 29).

Ethical approval

The institutional review board at the University of Colorado, Denver, Anschutz Medical Campus approved this research. Parents gave consent for DPS students to participate in vision and hearing screening and to have weight and height measured.

RESULTS

Table 1 presents descriptive statistics by the child's BMI status at the beginning of each person-period record. Normal-weight children were younger and more likely to be female, white, or Asian than overweight or obese children.

Table 1.

Characteristics of Children Aged 3–15 Years Enrolled in Denver Public Schools, Denver, Colorado, 2007–2013

Variable	BMIa Statusb
	Normal Weight	Overweight	Obese
Age, yearsc	8.7 (2.7)	9.3 (2.8)	9.6 (2.8)
Sex, %
Female	51.2	49.6	42.8
Male	48.8	50.4	57.2
Race/ethnicity, %
White	23.4	12.7	6.6
Asian	3.6	2.5	1.8
Black	12.5	12.1	10.5
Hispanic	60.5	72.7	81.1
No. of person-years	100,200	22,013	22,777

Abbreviation: BMI, body mass index.

^a Weight (kg)/height (m)².

^b Normal weight was defined as BMI below the 85th percentile; overweight was defined as BMI between the 85th and 95th percentiles; and obesity was defined as BMI at or above the 95th percentile.

^c Values are presented as mean (standard deviation).

Table 2 presents odds ratios from the discrete-time event-history models. Model 1 examined transitions to overweight status among children who were normal weight. For example, the odds ratio for age raised to the exponent of 0.5 shows that white females (the reference group for race/ethnicity and sex) at age 4 years had 1.15 ([1.67 × 4^0.5]/[1.67 × 3^0.5] = 1.15) times the odds of becoming overweight as white females at age 3 years, although white females at age 15 years had just 1.04 ([1.67 × 15^0.5]/[1.67 × 14^0.5] = 1.04) times the odds of becoming obese as white females at age 14 years. Model 2 examined transitions from overweight status to either normal weight or obesity, and model 3 examined transitions from obesity to overweight.

Table 2.

Odds Ratios From Discrete-Time Event-History Models of Transitions Between Body Mass Index Categories Among Children Aged 3–15 Years, Denver Public Schools, Denver, Colorado, 2007–2013

Variable	Model and BMIa Status Transitionb
	Model 1c— Normal Weight to Overweight		Model 2d				Model 3c— Obese to Overweight
			Overweight to Normal Weight		Overweight to Obese
	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
Age,e years	1.67f	1.42, 1.97	0.59f	0.48, 0.73	1.23	0.99, 1.54	0.57f	0.42, 0.79
Sex
Female	1.00		1.00		1.00		1.00
Male	2.28f	1.62, 3.22	0.90	0.56, 1.47	1.43	0.91, 2.23	1.74	0.87, 3.46
Race/ethnicity
Non-Hispanic white	1.00		1.00		1.00		1.00
Asian	1.69	0.56, 5.13	0.40	0.09, 1.73	9.06g	2.22, 36.96	0.12h	0.02, 0.77
Non-Hispanic black	4.19f	2.20, 7.99	0.41h	0.17, 1.00	3.31g	1.37, 7.98	0.30	0.08, 1.16
Hispanic	5.83f	3.53, 9.63	0.39g	0.21, 0.72	3.92f	1.91, 8.01	0.07f	0.03, 0.20
Interactions
Agee × race/ethnicity
Asian	0.87	0.61, 1.23	1.41	0.88, 2.26	0.50g	0.32, 0.78	2.18h	1.19, 3.98
Black	0.83	0.68, 1.03	1.22	0.90, 1.64	0.82	0.62, 1.09	1.16	0.75, 1.79
Hispanic	0.74f	0.63, 0.87	1.22	0.99, 1.50	0.74g	0.59, 0.92	1.90f	1.37, 2.63
Male sex × race/ethnicity
Asian	1.26	0.89, 1.79	0.67	0.44, 1.01	1.33	0.80, 2.20	0.79	0.46, 1.36
Black	0.64f	0.53, 0.78	0.97	0.75, 1.24	0.72h	0.54, 0.96	1.49h	1.03, 2.16
Hispanic	0.91	0.78, 1.06	0.97	0.82, 1.16	1.03	0.82, 1.30	1.01	0.77, 1.34
Male sex × agee	0.82f	0.74, 0.91	1.09	0.93, 1.28	0.94	0.83, 1.07	0.80h	0.64, 1.00
Constant	0.00f	0.00, 0.01	0.79	0.43, 1.45	0.04f	0.02, 0.09	0.65	0.25, 1.74

Abbreviations: BMI, body mass index; CI, confidence interval; OR, odds ratio.

^a Weight (kg)/height (m)².

^b Normal weight was defined as BMI below the 85th percentile; overweight was defined as BMI between the 85th and 95th percentiles; and obesity was defined as BMI at or above the 95th percentile.

^c Discrete-time hazard model via logistic regression.

^d Discrete-time hazard model via multinomial logistic regression.

^e Age raised to the exponent of 0.5.

^f P < 0.001.

^g P < 0.01.

^h P < 0.05.

The interaction terms in Table 2, models 1–3, complicate interpretation but flexibly allowed us to estimate the age-, sex-, and race/ethnicity-specific transition rates that were necessary for calculating our multistate life tables. To illustrate, Figure 1 shows estimated transition rates per 100 person-years, for ages 3–15 years, when sex and race/ethnicity were set to the proportions observed in our data. The solid lines show rates of transition to higher BMI categories. Transitions from normal weight status to overweight status (solid line with boxes) were relatively rare, but they were the only rates that increased with age. In contrast, transitions from overweight to obesity (solid line with stars) were relatively common and declined only slightly with age. The dashed lines show rates of transition to lower BMI categories. Transitions from overweight to normal weight (dashed line with boxes) were the most common transition at age 3 years, but they declined markedly with age. Finally, transitions from obese status to overweight (dashed line with stars) declined slightly with age.

Figure 1.

Age-specific rates of transition across body mass index (BMI; weight (kg)/height (m)²) categories among children aged 3–15 years (both sexes and all racial/ethnic groups combined), Denver Public Schools, Denver, Colorado, 2007–2013. Normal weight was defined as BMI below the 85th percentile; overweight was defined as BMI between the 85th and 95th percentiles; and obesity was defined as BMI at or above the 95th percentile.

Web Table 1 shows multistate life expectancies in the normal-weight, overweight, and obese BMI categories between the ages of 3 and 15 years, given BMI status at age 3 years. The first row shows life expectancies for children of both sexes and all racial/ethnic groups, combined. Children who were normal weight or obese at age 3 years could expect to live most of the following 13 years with those respective statuses. Children who were normal weight at age 3 years could expect to live 11.1 of the following 13 years with normal weight status, only 1.3 years with overweight status, and just 0.6 years with obese status. Further, children who were obese at age 3 years could expect to live 9.8 of the following 13 years with obese status, only 2.1 years with overweight status, and just 1.1 years with normal weight status. In contrast, children who were overweight at age 3 years routinely transitioned to other BMI categories. Children who were overweight at age 3 years could expect to spend 4.4 of the following 13 years with normal weight status and 3.4 years with obese status.

Web Table 1 also shows clear disparities by race/ethnicity, with white children being most likely to transition into (or remain in) lower BMI categories, followed by Asian, black, and Hispanic children. For example, among children who were obese at age 3 years, whites could expect to live 3.1 of the following 13 years with normal weight status, compared with 1.9 years among Asians, 1.4 years among blacks, and 0.9 years among Hispanics. In contrast, Hispanic children were most likely to transition into (or remain in) higher BMI categories, followed by black, Asian, and white children. Indeed, among children who were overweight at age 3 years, Hispanics could expect to live 3.9 of the following 13 years with obese status, as compared with 3.4 years among blacks, 3.1 years among Asians, and 1.4 years among whites.

Sex differences varied across racial/ethnic groups. Among white children, females and males could expect to live similar numbers of years in each BMI status group. Among both Asians and Hispanics, males could expect to live more years with higher BMI statuses than females, although sex differences were more pronounced among Asians. For example, among Asians who were overweight at age 3 years, females could expect to live 1.8 more of the following 13 years with normal weight status than males and 1.8 fewer years with obese status than males. Among blacks, females could expect to live more years with higher BMI statuses than males. Black females who were obese at age 3 years could expect to live 9.8 years with obese status, as compared with 8.5 years among black males.

DISCUSSION

Findings

To our knowledge, our study is the first to have examined the expected number of years that children aged 3–15 years would live in the normal-weight, overweight, and obese BMI status groups. Three major findings emerged from our analyses. First, children who were normal weight or obese at age 3 years could expect to live most of the next 13 years with those respective statuses. The relative persistence of normal weight and obesity between ages 3 and 15 years suggests the importance of promoting normal weight and preventing obesity at very early ages (14, 30). Relatively few weight control interventions have been tested among children younger than 3 years (31), although some evidence suggests that interventions may be particularly effective among younger children (32, 33).

Second, children who were overweight at age 3 years could expect to live most of the following 13 years with different BMI statuses. Only limited research has examined transitions to higher BMI categories (12–14), and just 1 other study has specifically considered recovery from obesity (12). Our analyses demonstrated that children more frequently moved out of the overweight status group (to either normal weight or obese status) than out of either normal weight status or obese status. Thus, interventions that target children who are overweight may be particularly effective at facilitating transitions into healthier BMI groups throughout adolescence. Future research might examine whether the factors that predict transitions from overweight to normal weight status differ from the factors that predict transitions from obesity to overweight status.

Third, our results identified important racial/ethnic and sex differences in the numbers of years children could expect to live with each BMI status. Among the racial/ethnic groups examined, white children had the highest expectation of remaining in or transitioning to lower BMI categories, followed by Asians. In contrast, Hispanic children had the highest expectation of remaining in or transitioning to higher BMI categories, followed by blacks. Racial/ethnic disparities in the expected number of years of living in each BMI category are consistent with racial/ethnic differences in the prevalence of overweight and obesity (2–5). However, our findings demonstrate that children in all racial/ethnic groups regularly transition to higher or lower BMI categories—a pattern that prevalence estimates cannot reveal.

Sex differences in the expected number of years of living with each BMI status varied across racial/ethnic groups. Among whites, sex differences were trivial in magnitude. Among Asians and Hispanics, males tended to transition to or remain in higher BMI categories than females; those sex differences were much larger among Asians than among Hispanics. In contrast, black females tended to transition to or remain in higher BMI categories than males. Our findings diverge somewhat from the most recently published prevalence estimates of sex differences in overweight and obesity within racial/ethnic groups between ages 2 and 19 years (2). Future research could examine the social and behavioral factors that predict racial/ethnic and sex disparities in BMI transitions throughout childhood.

Strengths and limitations

The strengths of our study included 6 years of cohort data that allowed us to estimate age-, sex-, and race/ethnicity-specific transitions across BMI status categories for children aged 3–15 years. Existing research on the incidence of obesity has focused on older age groups or a smaller subset of age ranges (12–14). Our study also demonstrates the value of working with school systems to acquire surveillance data that are representative and accurate. Finally, we used multistate life tables—a method that has been used to examine interstate migration patterns and active life expectancy among the elderly, but not to examine BMI transitions among children (15, 16).

We note 3 limitations of our study. First, children in Denver public schools may not be representative of children living elsewhere. Indeed, children in the DPS system are more likely to be Hispanic and low-income than children living in the city of Denver (34). However, the racial/ethnic patterns in BMI status in our data are consistent with prevalence data in nationally representative samples (2–5). Further, children in Denver public schools are similar to children in other urban school districts throughout the country. The National Center for Education Statistics (35) identifies school districts that are similar to DPS in terms of the percentage of students living in poverty (e.g., Baltimore, Maryland; Sacramento, California), the percentage of students that are English language learners (e.g., Oklahoma City, Oklahoma; Austin, Texas), student:teacher ratio (e.g., Detroit, Michigan; Phoenix, Arizona), and expenditures per student (e.g., Seattle, Washington; Omaha, Nebraska). The magnitude of the transitions across BMI categories and the racial/ethnic and sex disparities observed in our data are probably not unique to DPS, but future studies should replicate our work in other populations.

Second, our results may be conservative if we underestimated transitions across BMI categories. Our data may have missed short-term fluctuations in BMI status that do not persist until the subsequent sets of measurements are taken. This may be of particular concern with regard to the older children in our sample, because height and weight were measured every other year after the third grade. However, research in other dynamic populations (e.g., transitions in disability among older adults) suggests that transition rates are accurate when observations are made up to 2 years apart (36). Our analyses also ignored changes in body mass that do not result in movement across the thresholds between normal weight, overweight, and obese. Thus, our analyses most likely captured transitions in BMI that persisted over time and that resulted in changes in major BMI status levels that have been linked to adverse health outcomes (6–10).

Finally, our study data did not include information on behavioral, familial, neighborhood, or school-related variables that might explain variability in transitions across BMI categories by age, race/ethnicity, or sex (37, 38). Future studies might systematically explore the diverse sets of factors that may be associated with transitions to higher or lower BMI status.

Conclusion

Our findings demonstrate substantial mobility across BMI status categories during childhood and emphasize the importance of developing prevention and intervention strategies that target children very early in life. Children who were normal weight or obese at age 3 years could expect to live most of the following 13 years with those respective statuses. Given that stability, interventions that prevent obesity and promote normal weight before age 3 years could pay dividends in the form of lower body mass well into adolescence. In contrast, children who were overweight at age 3 years could expect to live numerous years outside of that BMI status, suggesting that interventions could nudge overweight children into lower BMI categories throughout the elementary-school and middle-school ages. Finally, the tendency to remain obese was greatest among Hispanic, black, and (to a lesser extent) Asian children. Thus, targeted interventions that prevent obesity in very early life may play an important role in closing racial/ethnic disparities in body mass, while improving health outcomes throughout childhood and potentially into adulthood.