obesity 1

nature publishing group articlesepidemiology

IntroductIonIncreases in childhood obesity have been linked to higher prevalence of metabolic health risks during childhood and adolescence (1), and obesity during childhood is an important predictor of adult obesity, metabolic syndrome, type 2 diabetes, and cardiovascular disease (2,3). However, because widely used growth data, such as the Centers for Disease Control and Prevention growth charts, are based on cross-sectional data, we know little about the developmental patterns of weight change within children that increase risk of obesity or about the lifestyle correlates and metabolic consequences of interindividual differences in these trajectories. Previous studies have used accelerated weight gain during infancy and early childhood, defined as a weight, weight-for-age or weight-for-height change score greater than +1 s.d., to predict obesity and metabolic risk during late childhood, adolescence, and adulthood (4–6); this approach assumes children with early accelerated growth are a homogeneous group and that early growth holds the same predictive value for the health status of all children, regardless of growth patterns during later childhood. Approaches that explore the heterogeneity in children’s full patterns of weight change may provide more accurate information about the etiologic pathways of obesity

onset and development, and the relation of these pathways to metabolic health risk later in life.

Recent research has provided evidence for heterogeneity in the development of obesity by illustrating the presence of several distinct patterns of weight status across childhood and adolescence. Li et al. tracked a sample of children from ages 2 to 12 years and identified three patterns of weight status based on children’s categorization as normal or overweight: 84% of children were consistently normal weight, 11% exhib-ited an early onset of overweight, and 5% exhibited a late onset of overweight (7). Using a similar categorical approach, Mustillo et al. tracked a sample of adolescents from ages 9 to 16 years and identified four patterns of weight status change: 73% of adolescents were never obese, 15% were chronically obese, 5% were obese at earlier ages but normal weight at later ages, and 7% were normal weight at earlier ages and obese at later ages (8).

These two studies employed growth mixture modeling techniques to model developmental trajectories of over-weight and obesity in longitudinal data (7,8). Growth mix-ture model analyses fit multiple growth curves based on the distinct (but latent) profiles of growth present within a sample. Identification of the latent trajectory groups can

Developmental Trajectories of Girls’ BMI Across Childhood and AdolescenceAlison K. Ventura1–3, Eric Loken2 and Leann L. Birch2,3

This study describes qualitatively distinct trajectories of BMI change among girls participating in a longitudinal study of non-Hispanic, white girls (n = 182) and their parents, assessed at daughters’ ages 5, 7, 9, 11, 13, and 15 years. Height, weight, body fat, fasting blood glucose and lipids, blood pressure, waist circumference, and pubertal status were measured, and participants self-reported dietary, physical activity, and television (TV) viewing patterns. Growth mixture models were used to model heterogeneity in girls’ BMI trajectories over 10 years. Statistical support was strongest for four distinct BMI trajectories: (i) upward percentile crossing (UPC; n = 25, 14%); (ii) delayed downward percentile crossing (DDPC; n = 37, 20%); (iii) 60th percentile tracking (60PT; n = 52, 29%); and (iv) 50th percentile tracking (50PT; n = 68, 37%). Girls in the UPC group had more metabolic risk factors at age 15 years, even after adjusting for concurrent weight status. Girls in the UPC group had mothers with the highest BMIs at study entry and were breast-fed for a shorter duration. This novel approach for examining differences in growth trajectories revealed four distinct BMI trajectories that predicted adolescent metabolic health outcomes in girls. The present study provides support for BMI monitoring in girls and for the potential utility of combining data on BMI tracking with data on familial characteristics for the early identification of girls at elevated risk for obesity and metabolic syndrome.

Obesity (2009) doi:10.1038/oby.2009.123

1Monell Chemical Senses Center, Philadelphia, Pennsylvania, USA; 2Department of Human Development and Family Studies, The Pennsylvania State University, University Park, Pennsylvania, USA; 3Center for Childhood Obesity Research, The Pennsylvania State University, University Park, Pennsylvania, USA. Correspondence: Leann L. Birch (llb15@psu.edu)

Received 3 July 2008; accepted 24 March 2009; advance online publication 7 May 2009. doi:10.1038/oby.2009.123

2 www.obesityjournal.org

articlesepidemiology

illustrate patterns of weight status change associated with obesity and related health outcomes (9). The present study also uses growth mixture model but differs from previous studies in that it uses a continuous indicator (BMI) rather than a categorical indicator (normal weight vs. overweight). Additionally, previous studies provided information on early life predictors (7) and psychiatric outcomes (8) associated with trajectories of overweight and obesity classification, and the present study examines modifiable behaviors and meta-bolic outcomes associated with differences in trajectories of BMI change during childhood.

In sum, the objective of this study is to explore discrete within-person BMI growth trajectories across childhood in a nonclinical sample of non-Hispanic, white girls. The specific aims of this research are to (i) assess whether distinctly different BMI trajectories across ages 5–15 years exist within our sample; (ii) describe the metabolic health outcomes at age 15 years associated with differing BMI trajectories; and (iii) explore family demographic, dietary, physical activity, and television (TV) viewing patterns associated with differing BMI trajectories.

Methods and ProceduresParticipantsAll participants resided in central Pennsylvania. Families were recruited for participation in the study using fliers and newspaper advertisements. Families with age-eligible female children within a 5-county radius also received mailings and follow-up phone calls (Metromail, Liverpool, NY). Eligibility criteria for girls’ participation at the time of recruitment included living with two biological parents, the absence of severe food allergies or chronic medical problems affecting food intake, and the absence of dietary restrictions involving animal products; the sample was not recruited based on parent or child weight status. At study entry, the total sample included 197 5-year-old non-Hispanic, white girls and their parents. Families were reassessed every 2 years (ages 7, 9, 11, 13, and 15 years). The primary analyses presented below used data from girls who had weight status measurements obtained at three or more of the six waves of data collection, yielding a sample of 182 families. Table 1 contains background information for the sample at study entry. The full sample (n = 197) has been described in more detail elsewhere (10–14).

MeasuresBackground information. At study entry, parents reported family demographic characteristics, daughters’ birth weight, and whether their daughters had been exclusively breast-fed, exclusively formula fed, or fed a mix of breast milk and formula. Parents also reported their own childhood weight status.

Weight status and adiposity. For parents and daughters, height and weight measurements were taken in the laboratory and were used to calculate BMI (weight (kg)/height (m2)). Because daughters were all the same age and gender, BMI, rather than BMI percentiles or z-scores, were used following the suggestions of Cole et al. (15). Age- and sex-specific BMI percentiles were also calculated and used to determine the preva-lence of overweight and obesity in this sample. A BMI ≥85th percentile was used to classify girls as overweight and ≥95th percentile was used to classify girls as obese. To estimate body fat mass, triceps and subscapular skinfold measurements were collected when girls were 5, 7, 9, and 11 years of age, and dual-energy X-ray absorptiometry scans were admin-istered at 9, 11, 13, and 15 years of age. Fat mass estimates were used to calculate fat mass index (FMI: fat mass (kg)/height (m2)) (16). Trained nurses measured girls’ waist circumference at age 15.

Blood assays. At age 15 years, girls’ blood samples were collected and processed for glucose, triglycerides, cholesterol, and insulin. An index of insulin resistance was calculated according to the homeostasis model assessment formula (17): homeostasis model assessment of insulin resistance = fasting insulin (µU/ml) × fasting glucose (mg/dl)/405.

Blood pressure. At each assessment, girls’ seated blood pressure was measured after 5 min of quiet resting; the average of two right arm readings was recorded.

Pubertal status and timing. At ages 7, 9, 11, and 13 years, girls’ breast development was assessed by visual inspection by trained nurses using the Tanner rating scale (18).

Dietary intake. At each visit, 24-h recalls were conducted to assess girls’ dietary intake, and the Minnesota Nutrition Data System for Research (NDS-R) version 4.06_34 (2003) (University of Minnesota Nutrition Coordinating Center, Minneapolis, MN) was used to calculate nutrient intakes. Girls provided three 24-h recalls within a 2–3-week period, and the three recalls included two weekdays and one weekend day.

Physical activity tendency. At ages 7, 9, 11, and 13 years, the Children’s Physical Activity scale was used to assess girls’ general tendency or inclination toward physical activity (e.g., “I would rather watch TV or play in the house than play outside”) (19).

TV viewing. Starting at age 7, girls reported number of hours of typical TV viewing on both school and nonschool days.

statistical analysesDescriptive information was generated for all variables of interest and each outcome variable was assessed for normality. Latent growth mixture models were fit to the BMI scores from ages 5, 7, 9, 11, 13, and 15 using Mplus (version 4.1) (Muthén & Muthén, Los Angeles, CA) and AMOS (version 16.0; SPSS, Chicago, IL). Mplus estimates mixture models using maximum likelihood, whereas AMOS imple-ments a Bayesian algorithm. The Mplus and AMOS solutions were in close agreement. In what follows, we report the solution from Mplus.

table 1 Background information for the total sample at study entry (n = 182)

Mean s.d. Percent n

Daughters’ age (years) 5.3 0.3

Mothers’ age (years) 35.4 6.0

Fathers’ age (years) 37.5 4.1

Average parent education (years)

14.7 2.1

Family income level

Less than $20,000 6.6 12

$20–35,000 21.0 38

$35–50,000 35.9 65

Over $50,000 36.5 66

Daughters’ BMI 15.9 1.5

Daughters’ BMI percentile 59.9 26.7

Mothers’ BMI 26.4 6.0

Fathers’ BMI 28.0 4.1

Prevalence of overweight

Daughtersa 19.2 35

Mothersb 52.8 96

Fathersb 74.7 136aDefined as BMI ≥85th percentile; bDefined as BMI >25.

obesity 3

articlesepidemiology

Mplus allows for estimation of models with missing data using maximum likelihood estimation under missing completely at random and missing at random assumptions. Thus, all available data for girls who did not complete the study could still be retained. Girls with fewer than three weight status measurements were excluded, yielding a total sample of 182 girls. We modeled a mixture of latent basis models (20), where basis coefficients for the growth factor were estimated from the data and not fixed. Factor loadings and means were unconstrained across groups, but the variances and covariances of the factors were constrained to be equal across groups. The number of latent groups (BMI trajectory groups) was determined by several model fit indices including Bayesian information criteria, Akaike information criteria, and entropy, along with other relevant evidence such as replicability and latent group sizes. Girls were classified into their most probable groups based on posterior probabilities, given the best-fitting model.

Once BMI trajectory group membership was determined for each girl, the remainder of analyses were conducted using SAS version 9.0 (SAS Institute, Cary, NC) (21). Mixed model analyses examined differ-ences among BMI trajectory groups on patterns of change for BMI and FMI across ages 5–15 years. ANOVA with a general linear model was used to assess BMI trajectory group differences for metabolic health outcomes at age 15 years and family demographics and parent weight status at study entry. Additionally, follow-up analyses of covariance were used to adjust models examining the association between BMI trajectory groups and metabolic health outcomes for weight status at age 15 years. To compare the BMI trajectory group solution to tradi-tional measures of early accelerated weight gain (see ref. 22), girls were divided into two groups based on their BMI change scores for early (prepubertal) growth: (i) Accelerated weight gain: Girls whose BMI change across ages 5–9 years was >1 s.d. of the mean BMI change score for that period; or (ii) Normal weight gain: Girls whose BMI change across ages 5–9 years was ≤1 s.d. of the mean BMI change score for that period. ANOVA with a general linear model was used to assess differ-ences between these two groups on metabolic health outcomes at age 15 years. Repeated measures ANOVA was used to examine BMI trajec-tory group differences on patterns of puberty, dietary intake, physical activity tendencies, and TV viewing across ages 5–15 years.

resultsGrowth mixture model solutionA four-group solution was the best-fitting model based on the Bayesian information criteria and on considerations regarding the replicability of the five-group model (Table 2). For illustrative purposes, BMI trajectories for the four groups are plotted on the Centers for Disease Control and Prevention growth chart for girls ages 2–20 years (23) (Figure 1). Groups

were labeled based on how their BMI trajectory related to the Centers for Disease Control and Prevention growth chart percentile curves: (i) upward percentile crossing (UPC; n = 25, 14% of the sample); (ii) delayed downward percentile crossing (DDPC; n = 37, 20% of the sample); (iii) 60th percentile tracking (60PT; n = 52, 29% of the sample); and (iv) 50th percentile tracking (50PT; n = 68, 37%). Figure 1 illustrates the average trajectory of the girls classified within each group; as evidenced by the error bars around each group trajectory, there are individuals within each group who display a similar patterning of BMI change, but track along different percentile lines.

Girls in the UPC group exhibited a consistent pattern of upward BMI percentile crossing across childhood (Figure 1). Girls in the DDPC group also exhibited upward BMI percentile crossing between ages 5 and 9 years, but diverged from this pattern at age 11 years, exhibiting a downward BMI percentile crossing trajectory across ages 11–15 years. From ages 5 to 15 years, the 60PT group tracked, on average, along the 60th percentile for BMI and the 50PT group tracked, on average, along the 50th percentile for BMI.

table 2 Indices of goodness-of-fit for GMM solution

Number of latent groups AIC BIC SA-BIC Entropy

1 4,209.15 4,254.00 4,209.66 —

2 3,848.62 3,906.30 3,849.29 0.87

3 3,779.48 3,862.78 3,780.44 0.78

4 3,721.83 3,830.77 3,723.09 0.81

5a 3,705.45 3,840.02 3,707.00 0.86

Four-group solution (boldface) was determined the best fit based on BIC and replicatability.AIC, Akaike information criteria; BIC, Bayesian information criteria; GMM, growth mixture model; SA-BIC, sample-size adjusted BIC.aAlthough the 5-group model had lower AIC and SA-BIC indices, it was only found once after generating 1,000 random start values. Furthermore, this model did not terminate normally and contained an exceptionally small class (5% of the sample).

UPC (14%; n = 25)

DDPC (20%; n = 37)

60PT (29%; n = 52)

50PT (37%; n = 68)

BMI

27

26

25

24

23

22

21

20

19

18

17

16

15

14

13

12

2 3 4 5 6 7 8 9 10 11 12 13

Age (years)

14 15 16

kg/m2

BMI

27

28

29

30

26

25

24

23

22

21

20

19

18

17

16510

25

50

75

85

90

95

15

14

13

12

kg/m2

Figure 1 Growth mixture model solution of BMI trajectory groups plotted on the Centers for Disease Control and Prevention growth reference chart for girls ages 2–20 years. Intercepts for the UPC and DDPC group trajectories were significantly higher than the 50PT group trajectory (P < 0.001); the intercept for the 60PT group trajectory was not significantly different than the 50PT group trajectory. Slopes for the UPC and DDPC group trajectories were significantly greater than for the 50PT group trajectory (P < 0.01), but the slope for the 60PT group trajectory was no different than that for the 50PT group trajectory. 50PT, 50th percentile tracking; 60PT, 60th percentile tracking; DDPC, delayed downward percentile crossing; UPC, upward percentile crossing.

4 www.obesityjournal.org

articlesepidemiology

Patterns of FMI change across ages 5–15 yearsPatterns of change in FMI across ages 5–15 years for BMI trajectory groups reflected patterns of BMI change over this same period (Figure 2). Specifically, girls in the UPC and DDPC groups had similar FMI trajectories across ages 5–9 years but diverged after age 9 years. Girls in the 60PT group showed a slight but steady increase in FMI, whereas girls in the 50PT group maintained a consistent percent body fat across ages 5–15 years.

Metabolic outcomes at age 15 yearsWe hypothesized the metabolic health risk profile for girls in the UPC group would differ from other girls in the sample because girls in the UPC group had a pattern of continued UPC. Thus, ANOVAs were conducted with a priori contrasts specified between the UPC group and all other groups. These analyses revealed that girls in the UPC group had significantly higher values for waist circumference, total cholesterol, low-density lipoprotein cholesterol, fasting insulin, homeostasis model assessment of insulin resistance, triglycerides, and blood pressure at age 15 years compared to girls in the DDPC, 60PT, and 50PT groups (Table 3).

BMI at 15 years of age was tested as a covariate in all models to determine whether the effect of BMI trajectory group mem-bership on metabolic health outcomes remained after adjusting

for concurrent weight status. Concurrent weight status was not a significant predictor of high-density lipoprotein, glucose, and triglyceride levels. Even after accounting for concurrent weight status, girls in the UPC group had significantly higher waist

table 3 BMI trajectory groups vs. prepubertal accelerated weight gain as predictors of metabolic outcomes at age 15 years

Total sample (n = 182)

BMI trajectory groups

F valuea

Prepubertal weight gain

F valueUPC

(n = 24)DDPC (n = 37)

60PT (n = 52)

50PT (n = 68)

Accelerated ∆BMI >1 s.d.

(n = 27)

Normal ∆BMI <1 s.d.

(n = 155)

Waist circumference (cm)

78.9 (10.1) 97.6 (8.6) 77.3 (8.5) 76.7 (6.5) 74.8 (5.5) 179.05***,b 90.0 (13.3) 77.0 (8.1) 42.24***

Total cholesterol (mg/dl)

158.4 (25.3) 175.1 (24.7) 155.3 (27.8) 157.0 (21.8) 154.7 (25.1) 11.62***,b 164.1 (28.1) 157.4 (24.8) 1.35

HDL (mg/dl) 50.6 (10.4) 48.7 (11.1) 50.6 (11.9) 49.7 (7.9) 52.0 (11.1) 0.76 48.9 (10.7) 50.9 (10.3) 0.73

LDL (mg/dl) 90.6 (23.3) 105.8 (22.0) 87.6 (26.8) 91.6 (17.9) 85.3 (23.9) 11.33***,b 96.1 (24.3) 89.6 (23.1) 1.50

Fasting glucose (mg/dl)

84.4 (11.0) 83.9 (5.7) 83.1 (8.1) 85.4 (16.6) 84.4 (7.4) 0.02 83.5 (5.8) 84.6 (11.7) 0.17

Fasting insulin (µU/ml)

9.1 (4.6) 14.3 (7.1) 8.0 (4.1) 8.0 (2.5) 8.4 (3.1) 40.60***,b 11.6 (7.1) 8.7 (3.8) 7.86**

Insulin resistance (HOMAIR)

1.9 (1.1) 3.0 (1.5) 1.7 (1.0) 1.8 (0.9) 1.8 (0.7) 26.91*** 2.4 (1.5) 1.9 (1.0) 4.70*

Triglycerides (mg/dl)

86.4 (42.9) 103.7 (39.8) 85.1 (32.9) 78.5 (27.1) 86.9 (56.2) 4.17* 95.4 (38.3) 84.8 (43.6) 1.19

Systolic blood pressure (mm Hg)

101.3 (10.0) 105.5 (12.6) 101.3 (10.6) 100.5 (9.2) 100.4 (9.0) 4.28* 102.8 (12.6) 101.0 (9.5) 0.62

Diastolic blood pressure (mm Hg)

64.9 (6.7) 68.9 (7.8) 64.8 (6.1) 64.2 (6.2) 64.0 (6.6) 8.92** 66.0 (7.5) 64.7 (6.6) 0.70

Prepubertal defined as 5–9 years of age.50PT, 50th percentile tracking; 60PT, 60th percentile tracking; DDPC, delayed downward percentile crossing; HDL, high-density lipoprotein; HOMA, homeostasis model assessment; LDL, low-density lipoprotein; UPC, upward percentile crossing.aF-test value for the unadjusted model; in both the unadjusted and adjusted models, a priori contrasts were specified for the UPC group vs. the DDPC, 60PT, and 50PT groups. bModel remained significant, or trended toward significance, after adjusting for weight status at age 15 years.*P < 0.05; **P < 0.01; ***P < 0.001.

12

10

8

6

Fat m

ass

inde

x sc

ore

(fat

mas

s (k

g)/h

eigh

t (m

2 ))

4

2

05 7 9 11

Age13 15

UPCDDPC60PT50PT

Figure 2 Fat mass index across ages 5–15 years reflects patterns of BMI change for BMI trajectory groups. Intercepts for the UPC and DDPC group fat mass index (FMI) trajectories were significantly different from the trajectory for the 50PT group (P < 0.001). The 60PT group FMI trajectory intercept was not significantly different than that for the 50PT group. Slopes for the UPC, DDPC, and 60PT group FMI trajectories were significantly higher than that for the 50PT group (P < 0.001). 50PT, 50th percentile tracking; 60PT, 60th percentile tracking; DDPC, delayed downward percentile crossing; UPC, upward percentile crossing.

obesity 5

articlesepidemiology

circumference (F (1, 158) = 8.65, P < 0.01) and total cholesterol (F (1, 147) = 3.74, P = 0.05) values compared to girls in the DDPC, 60PT, and 50PT groups. Trends were seen for an effect of BMI trajectory group membership, over and above current weight status, on low-density lipoprotein cholesterol (F (1, 148) = 3.66, P = 0.06) and insulin (F (1, 133) = 3.30, P = 0.07).

comparison of BMI trajectory grouping to traditional measures of accelerated weight gainFifteen percent of the sample (n = 27) exhibited accelerated weight gain during early childhood (prepubertal growth, across ages 5–9 years) based on traditional definitions of accel-erated weight gain (BMI change >1 s.d.; see (22)). Fifteen of these girls were classified in the UPC group and the other 12 of these girls were classified in the DDPC group; no girls with accelerated weight gain across ages 5–9 years were classified in the 60PT or 50PT groups.

Accelerated weight gain during early childhood was not a strong predictor of obesity at age 15. Of the 27 girls exhibit-ing accelerated weight gain during early childhood, 42% (n = 10) were classified as obese at age 15 whereas 58% (n = 14) were classified as normal weight. As shown in the Table 3, girls with accelerated weight gain during early childhood had significantly higher waist circumference, fasting insulin, and homeostasis model assessment values at age 15. However, also

shown in Table 3, mean values for the accelerated weight gain group were a compromise between the mean values for the UPC and DDPC group.

Family demographic and parent weight status differences at study entryAt study entry, BMI for mothers of girls in the UPC group was significantly higher than for mothers of all other groups and, on average, mothers of girls in the UPC group were obese (Table 4). Fathers of girls in the UPC group were heav-ier than fathers of girls in the 50PT group. Parents of girls in the UPC and DDPC group reported lower education levels compared to parents of girls in the 60PT and 50PT groups, and girls in the UPC group were breast-fed for a significantly shorter duration.

Patterns of pubertal status and timing across ages 7–13 yearsA significant main effect of group membership (F (3, 139) = 15.51, P < 0.001) and interaction between group member-ship and time (F (9, 417) = 6.63, P < 0.001) were noted for pubertal status (Table 5). Post hoc analyses revealed that, overall, girls in the UPC group matured earliest, followed closely by girls in the DDPC group, and the 50PT group matured the latest.

table 4 Means (s.d.) for BMI trajectory group differences on family demographics at study entry

BMI trajectory group

UPC (14%, n = 24) DDPC (20%, n = 37) 60PT (29%, n = 52) 50PT (37%, n = 68)

Maternal BMI 31.4a (8.9) 26.5b (5.9) 26.4b (5.3) 24.6b (4.2)

Paternal BMI 29.4a (3.6) 29.0a (4.6) 28.4a (4.1) 26.8b (3.8)

Maternal retrospective reported weight as child1

3.4a (0.9) 2.9a,b (0.9) 2.7b (0.8) 2.9b (0.9)

Paternal retrospective reported weight as child1

2.9a,c (0.9) 3.0b (1.0) 2.9b (1.0) 2.5c (0.8)

Family income2 1.9a (0.8) 2.1a (0.9) 2.0a (0.9) 2.1a (1.0)

Average parent education (years) 13.6a (1.3) 14.3a (2.1) 15.0b (2.3) 15.1b (2.2)

Breast-feeding duration (months) 4.4a (2.9) 5.8b (1.7) 5.7b (2.1) 5.6b (1.8)

Birth weight (pounds) 7.5a (1.2) 7.2a (1.2) 7.7a (1.2) 7.7a (1.2)

Means in a row followed by the same superscript (a, b, c) were not significantly different at P < 0.05 in the Bonferroni post hoc test.50PT, 50th percentile tracking; 60PT, 60th percentile tracking; DDPC, delayed downward percentile crossing; UPC, upward percentile crossing.1Variable coding: 1 = markedly underweight to 5 = markedly overweight. 2Variable coding: 0 = less than $20,000; 1 = $20–35,000; 2 = $35–50,000; 3 = over $50,000.

table 5 Mean (s.d.) pubertal status (tanner staging score) across ages 7–13 years by BMI trajectory group

Total sample

BMI trajectory group

UPC (14%, n = 24) DDPC (20%, n = 37) 60PT (29%, n = 52) 50PT (37%, n = 68)

Age 7 1.1 (0.4) 1.4a (0.5) 1.2a,b (0.4) 1.1b (0.2) 1.0b (0.2)

Age 9 1.8 (0.8) 2.2a (0.7) 2.5a (0.6) 1.8b (0.7) 1.2c (0.5)

Age 11 2.2 (0.8) 2.5a (0.8) 2.5a (0.7) 2.4a (0.7) 1.9b (0.7)

Age 13 3.8 (0.9) 3.8a (0.9) 3.7a (0.8) 3.9a (0.9) 3.7a (0.9)

Means in a row followed by the same superscript (a, b, c) are not significantly different at P < 0.05 in the repeated measures ANOVA Bonferroni corrected post hoc test.50PT, 50th percentile tracking; 60PT, 60th percentile tracking; DDPC, delayed downward percentile crossing; UPC, upward percentile crossing.

6 www.obesityjournal.org

articlesepidemiology

Patterns of dietary intake across ages 5–15 yearsNo significant differences among groups were noted for self-reported dietary energy intake, grains, vegetables, fruits, dairy, meats, fats, sweets, added sugars, sweetened beverages, and eating in front of the TV.

Patterns of physical activity and tV viewingBMI trajectory groups did not significantly differ on tenden-cies toward physical activity or TV viewing patterns.

dIscussIonWe used a data-driven growth mixture modeling approach to explore distinct patterns of BMI change across childhood and early adolescence in a nonclinical sample of non-Hispanic, white girls. The four growth trajectories identified by this approach were described in reference to the Centers for Disease Control and Prevention BMI-for-age percentile curves (23): (i) UPC, (ii) DDPC, (iii) 60PT, and (iv) 50PT. The validity of these trajectories is supported by their similarity to trajectories reported in previous studies using growth mixture models with different approaches and samples (7,8); for example, Mustillo et al. noted the presence of a group of children similar to the DDPC group found in the present study in that they were obese during childhood, but normal weight during adolescence (8). Although the patterns presented in the present study represent groups, not individuals, the UPC growth pattern for the UPC and DDPC groups across ages 5–9 years would likely have been highlighted in a clinical setting as reflecting accelerated early weight gain (5,6,24–26). The problematic nature of the UPC group’s growth trajectory was confirmed by the finding that girls in this group, on average, accrued higher levels of body fat and had the highest metabolic health risk compared to all other girls in this sample. Girls in the UPC group were further distinguished from all other groups in that their mothers had the highest weight status. Girls in the DDPC group had metabolic outcomes at age 15 years and familial characteristics at study entry similar to girls who tracked along percentile lines, despite having a change in BMI across ages 5–9 years similar to that of the UPC group.

Studies examining the predictive ability of growth patterns have typically focused on associations between accelerated growth during critical periods (e.g., infancy, adiposity rebound) and later obesity and metabolic risk factors (4–6,27). We took a similar stance by examining the implications of growth patterns, but used an exploratory, data-driven approach to identify latent trajectories across ages 5–15 years rather than quantify change scores based on a priori assumptions about sensitive periods and problematic growth rates. This approach illustrated that, even in the absence of these a priori assumptions, a subset of girls with accelerated weight gain during early childhood could be identified. However, by examining girls’ growth trajectories across both early and later childhood, we illustrated that girls with accelerated weight gain during early childhood were not a homogeneous group: early accelerated weight gain was only associated with metabolic risk factors when that pattern persisted into adolescence. Thus, through exploration of the

latent heterogeneity in growth trajectories, we were able to incorporate information about both early and later growth to gain a clearer picture of the nature and implications of patterns of growth.

The two groups, UPC and DDPC, differed on maternal weight status at study entry, despite their similarly higher weight status at study entry and accelerated growth from study entry to age 9 years. It is well established that children of obese parents have a higher risk of becoming obese than children of normal weight parents (14,28). Although familial similarity in weight status is partially attributable to genetics (29), characteristics of parents and family environments also have a substantial influence on childhood weight status (30). Additionally, maternal weight status may be more strongly associated than paternal weight status to child weight status because mothers have more opportunities to influence children’s growth and development through their direct influence on the prenatal environment (31), or through their typically greater involvement in children’s postnatal feeding and caregiving environment (32). Thus, the findings from the present study, in conjunction with previous research, suggest researchers should combine BMI monitoring with examination of family factors, specifically maternal weight status, to identify girls at higher risk for sustained upward BMI percentile crossing across childhood and higher metabolic syndrome risk in adolescence.

In national data, pubertal onset for non-Hispanic, white girls has been found to occur at an average of 9–10 years of age (33); our sample was consistent with this national data. However, in our sample, membership in trajectory groups with higher initial weight and fat gain (the UPC and DDPC groups) predicted earlier pubertal development. In this same sample of girls, Davison et al. showed that higher fat mass at age 5 predicted earlier pubertal onset using a combination of pubertal measures at ages 7 and 9 years (34). The fact that we found similar trends using different indicators of weight status and pubertal timing is an additional validation for the BMI trajectory groups yielded by our data-driven analysis. Early menarche is associated with higher metabolic risk later in life independent of weight status, and the combination of higher weight status and earlier puberty may create a cumulatively higher risk for metabolic health problems in children (35). Although this hypothesis was not explicitly tested by the current study, our findings suggest that the culmination of higher weight status and earlier pubertal onset exhibited by girls in the UPC group did put these girls at higher metabolic risk at age 15 years.

Despite significant differences in growth among BMI trajectory groups, we could not identify any consistent group differences in dietary intake, physical activity, or TV viewing that corresponded with BMI trajectories or explained the downward deflecting trajectory for girls in the DDPC group. We did note, however, that based on our estimation of reporting bias in our dietary recall data (36), the tendency for all girls to underreport dietary intake increased with time and that, at later ages, underreporting bias was higher for girls in the UPC group (data not shown). It is well established that the presence

obesity 7

articlesepidemiology

of bias in self-reported dietary data can mask relations between diet and health status (37,38), and a systematic bias is created when underreporting differs by weight status, as in the present study. Thus, dietary differences that could explain the divergent trajectory of girls in the DDPC group and the UPC trajectory of the UPC group may not have been detected due to heavier girls’ tendency to underreport to a greater extent.

A limitation of the present study is that we did not collect data on metabolic health at all assessments, which would have allowed a more detailed examination of the concurrent development of BMI trajectories and metabolic profiles across ages 5–15 years. Although we found growth trajectories were associated with metabolic outcomes at age 15 years even after controlling for weight status at age 15, a caveat to our findings remains in that concurrent BMI was highly correlated with membership in our UPC group, and was a strong predictor of the age 15 metabolic outcomes (data not shown). Additionally, the generalizability of our findings may be compromised due to the homogeneity of our sample, which was created by the fact that we recruited non-Hispanic, white girls who resided in the same geographical location, lived with both biological parents and had no severe food allergies, chronic medical problems affecting food intake, or dietary restrictions involving animal products. Future research is needed to examine these issues further in more diverse populations.

Despite these limitations, two methodological aspects of this study are key in illustrating the contribution of the growth mixture model solution presented here: (i) this method summarized individual differences in the developmental progression of weight status in the form of four distinct trajectories; and (ii) the form of the trajectories was not imposed ex ante but rather emerged from the data. Categorical grouping of individuals is a way to provide a heuristic summary of a complex developmental process (39,40). Through use of a person-centered technique (growth mixture modeling), we were able to illustrate the natural heterogeneity in the development of weight status across childhood and adolescence. Additionally, by combining this person-centered approach with variable-centered methods, we were also able to understand this natural heterogeneity within the context of early predictors and later metabolic outcomes associated with individual differences in growth trajectories (41).

In summary, our findings support the importance of long-term tracking of children’s BMI; by considering girls’ age 5–15-year growth trajectories, we illustrated that girls with accelerated growth (i.e., upward BMI percentile crossing) across all of childhood had undesirable metabolic outcomes at age 15 years, whereas girls showing a downward deflection in their BMI trajectory at age 9 avoided these same undesirable outcomes. Additionally, our findings support the need to consider girls’ individual patterns of growth in combination with evidence available regarding other risk factors, particularly maternal weight status, that influence risk and may distinguish girls in need of early intervention for high-risk patterns of growth.

acknowledGMentsThis research was supported by National Institutes of Health grants M01 RR10732 and HD32973. We also like to thank the families who participated in this research study and the General Clinical Research Center of The Pennsylvania State University, which provided aid and services.

dIsclosureThe authors declared no conflict of interest.

© 2009 The Obesity Society

reFerences1. Gidding SS, Bao W, Srinivasan SR, Berenson GS. Effects of secular trends

in obesity on coronary risk factors in children: the Bogalusa Heart Study. J Pediatr 1995;127:868–874.

2. Berenson GS, Srinivasan SR, Bao W et al. Association between multiple cardiovascular risk factors and atherosclerosis in children and young adults. The Bogalusa Heart Study. N Engl J Med 1998;338:1650–1656.

3. Srinivasan SR, Myers L, Berenson GS. Predictability of childhood adiposity and insulin for developing insulin resistance syndrome (syndrome X) in young adulthood: the Bogalusa Heart Study. Diabetes 2002;51:204–209.

4. Demerath EW, Schubert CM, Maynard LM et al. Do changes in body mass index percentile reflect changes in body composition in children? Data from the Fels Longitudinal Study. Pediatrics 2006;117:e487–e495.

5. Ekelund U, Ong KK, Linné Y et al. Association of weight gain in infancy and early childhood with metabolic risk in young adults. J Clin Endocrinol Metab 2007;92:98–103.

6. Stettler N, Kumanyika SK, Katz SH, Zemel BS, Stallings VA. Rapid weight gain during infancy and obesity in young adulthood in a cohort of African Americans. Am J Clin Nutr 2003;77:1374–1378.

7. Li C, Goran MI, Kaur H, Nollen N, Ahluwalia JS. Developmental trajectories of overweight during childhood: role of early life factors. Obesity (Silver Spring) 2007;15:760–771.

8. Mustillo S, Worthman C, Erkanli A et al. Obesity and psychiatric disorder: developmental trajectories. Pediatrics 2003;111:851–859.

9. Greenbaum PE, Del Boca FK, Darkes J, Wang CP, Goldman MS. Variation in the drinking trajectories of freshmen college students. J Consult Clin Psychol 2005;73:229–238.

10. Abramovitz BA, Birch LL. Five-year-old girls’ ideas about dieting are predicted by their mothers’ dieting. J Am Diet Assoc 2000;100:1157–1163.

11. Fisher JO, Birch LL. Eating in the absence of hunger and overweight in girls from 5 to 7 y of age. Am J Clin Nutr 2002;76:226–231.

12. Mannino ML, Lee Y, Mitchell DC, Smiciklas-Wright H, Birch LL. The quality of girls’ diets declines and tracks across middle childhood. Int J Behav Nutr Phys Act 2004;1:5.

13. Davison KK, Francis LA, Birch LL. Links between parents’ and girls’ television viewing behaviors: a longitudinal examination. J Pediatr 2005;147:436–442.

14. Francis LA, Ventura AK, Marini M, Birch LL. Parent overweight predicts daughters’ increase in BMI and disinhibited overeating from 5 to 13 years. Obesity (Silver Spring) 2007;15:1544–1553.

15. Cole TJ, Faith MS, Pietrobelli A, Heo M. What is the best measure of adiposity change in growing children: BMI, BMI %, BMI z-score or BMI centile? Eur J Clin Nutr 2005;59:419–425.

16. VanItallie TB, Yang MU, Heymsfield SB, Funk RC, Boileau RA. Height-normalized indices of the body’s fat-free mass and fat mass: potentially useful indicators of nutritional status. Am J Clin Nutr 1990;52:953–959.

17. Matthews DR, Hosker JP, Rudenski AS et al. Homeostasis model assessment: insulin resistance and βcell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 1985;28:412–419.

18. Marshall WA, Tanner JM. Variations in pattern of pubertal changes in girls. Arch Dis Child 1969;44:291–303.

19. Tucker LA, Seljaas GT, Hager RL. Body fat percentage of children varies according to their diet composition. J Am Diet Assoc 1997;97:981–986.

20. McArdle JJ, Epstein D. Latent growth curves within developmental structural equation models. Child Dev 1987;58:110–133.

21. SAS Institute I. Statistical Analysis Software. SAS Institute: Cary, NC, 2001.22. Monteiro PO, Victora CG. Rapid growth in infancy and childhood and

obesity in later life—a systematic review. Obes Rev 2005;6:143–154.23. Kuczmarski R, Ogden C, Grummer-Strawn L et al. CDC growth charts:

United States. In: Advance Data From Vital and Health Statistics. National Center for Health Statistics: Hyattsville, Maryland, 2000.

8 www.obesityjournal.org

articlesepidemiology

24. Cole TJ. Measuring normal growth and growth variance in infancy and childhood. In: Birch LL, Dietz W (eds). Eating Behaviors of the Young Child: Prenatal and Postnatal Influences on Healthy Eating. American Academy of Pediatrics: Elk Grove Village, IL, 2008, pp 17–31.

25. Berkowitz RI, Stallings VA, Maislin G, Stunkard AJ. Growth of children at high risk of obesity during the first 6 y of life: implications for prevention. Am J Clin Nutr 2005;81:140–146.

26. Dennison BA, Edmunds LS, Stratton HH, Pruzek RM. Rapid infant weight gain predicts childhood overweight. Obesity (Silver Spring) 2006;14: 491–499.

27. Ekelund U, Ong K, Linné Y et al. Upward weight percentile crossing in infancy and early childhood independently predicts fat mass in young adults: the Stockholm Weight Development Study (SWEDES). Am J Clin Nutr 2006;83:324–330.

28. Whitaker RC, Wright JA, Pepe MS, Seidel KD, Dietz WH. Predicting obesity in young adulthood from childhood and parental obesity. N Engl J Med 1997;337:869–873.

29. Maes HH, Neale MC, Eaves LJ. Genetic and environmental factors in relative body weight and human adiposity. Behav Genet 1997;27:325–351.

30. Davison KK, Birch LL. Obesigenic families: parents’ physical activity and dietary intake patterns predict girls’ risk of overweight. Int J Obes Relat Metab Disord 2002;26:1186–1193.

31. Whitaker RC, Dietz WH. Role of the prenatal environment in the development of obesity. J Pediatr 1998;132:768–776.

32. McHale SM, Crouter AC, McGuire SA, Updegraff KA. Congruence between mothers’ and fathers’ differential treatment of siblings: links with family relations and children’s well-being. Child Dev 1995;66: 116–128.

33. Sun SS, Schubert CM, Chumlea WC, et al. National estimates of the timing of sexual maturation and racial differences among US children. Pediatrics 2002;110: 911–919.

34. Davison KK, Susman EJ, Birch LL. Percent body fat at age 5 predicts earlier pubertal development among girls at age 9. Pediatrics 2003;111: 815–821.

35. Frontini MG, Srinivasan SR, Berenson GS. Longitudinal changes in risk variables underlying metabolic syndrome X from childhood to young adulthood in female subjects with a history of early menarche: the Bogalusa Heart Study. Int J Obes Relat Metab Disord 2003;27: 1398–1404.

36. Ventura AK, Loken E, Mitchell DC, Smiciklas-Wright H, Birch LL. Understanding reporting bias in the dietary recall data of 11-year-old girls. Obesity (Silver Spring) 2006;14: 1073–1084.

37. Heitmann BL, Lissner L. Can adverse effects of dietary fat intake be overestimated as a consequence of dietary fat underreporting? Public Health Nutr 2005;8: 1322–1327.

38. Rosell MS, Hellénius ML, de Faire UH, Johansson GK. Associations between diet and the metabolic syndrome vary with the validity of dietary intake data. Am J Clin Nutr 2003;78: 84–90.

39. Li F, Duncan TE, Duncan SC, Acock A. Latent growth modeling of longitudinal data: a finite growth mixture modeling approach. Structural Equation Modeling 2001;8: 493–530.

40. Muthén B. Latent variable analysis: growth mixture modeling and related techniques for longitudinal data. In: Kaplan D (ed). Handbook of Quantitative Methodology for the Social Sciences. Sage Publications: Newbury Park, CA, 2004, pp 345–368.

41. Muthén B, Muthén LK. Integrating person-centered and variable-centered analyses: growth mixture modeling with latent trajectory classes. Alcohol Clin Exp Res 2000;24: 882–891.