The prevalence of Class 3 obesity (BMI ≥40 kg/m2) has more than doubled in the past 25 years. In a 14-year prospective study from age 10 to 24 of a biracial schoolgirl cohort (293 black, 256 white), we assessed childhood correlates of Class 3 BMI at age 24. Of 42 girls with Class 3 BMI at age 24, 36 (86%) were black. By logistic regression, significant explanatory variables of Class 3 BMI at age 24 included top decile waist circumference at age 11 (odds ratio (OR) 5.7, 95% confidence interval (CI) 2.3–13.9, P = 0.0002), age 10 BMI ≥ the Center for Disease Control (CDC) 2000 top 15% (OR 7.0, 95% CI 2.5–19.3, P = 0.0002), and a three-way interaction between race, childhood insulin, and average caloric intake from age 10 to age 19 (for each unit increase, OR 1.7 95% CI 1.3–2.2, P = 0.0003). Age 10 BMI, age 11 waist circumference, and interaction of race, childhood insulin, and childhood caloric intake predict Class 3 obesity in young adulthood, facilitating childhood identification of girls at high risk for developing Class 3 obesity.
The obesity explosion of the past two decades has markedly increased the prevalence of Class 3 obesity (BMI ≥40) (1). A comparison of BMI in 3rd–5th graders in the Cincinnati Princeton Lipid Research Clinics Prevalence Study in the 1970s and in the same schools 15 years later in 1990 showed that the secular shift in BMI was most pronounced at BMI percentiles above the 1970 40th percentile (2). The secular trend of increasing BMI has continued, and the prevalence of obesity has increased 300% since National Health and Nutrition Examination Survey II in 1976 and over 70% since National Health and Nutrition Examination Survey III in 1994 (3). These shifts were not uniform across sex-race groups, but were greater in males than females and in African Americans and Mexican Americans than whites (3,4).
Obesity is a pervasive risk factor for cardiovascular disease (5), hypertension (6), and type 2 diabetes (T2DM) (7). Class 3 obesity causes difficulty in tasks of daily living (8), and is the level at which an adult can qualify for bariatric surgery without substantial comorbidities. Although the majority of overweight adults were not overweight as youth, as many as 80% of overweight youth remain overweight as adults (9). Thus, the recent increase in adolescent obesity portends still greater adult obesity.
Hyperinsulinemia appears to play a central supporting and/or etiologic role in development of obesity and Class 3 obesity in young adult years (10). In our previous 10-year prospective study, when black and white girls were matched by BMI (or fat mass), maturation status, and fasting serum insulin at ages 9–10, insulin interacted positively with caloric intake to produce a greater increase in BMI, and this effect was greater in black than white girls (11). We speculated that a 2-way insulin–race interaction and a 3-way insulin-caloric intake–race interaction was central to an intrinsic black–white metabolic difference (12,13,14,15,16,17,18) underlying the greater increase in fat mass and percent body fat in black than white girls during adolescence (19).
Questions to be addressed in the current 14-year follow-up report include the following:
What are the significant childhood explanatory variables for development of Class 3 obesity in young adulthood, and how do they interact with race?
What group of factors could a clinician use for early identification of female children-adolescents at high risk for developing Class 3 obesity in young adulthood?
Given the findings on the strength of BMI tracking (9,20), to what extent do young women with Class 3 obesity at age 24 years come from the ranks of overweight and obese girls at age 10?
Methods and Procedures
The study population
The NHLBI Growth and Health study (NGHS) has been described previously (21). Briefly, NGHS was a 10-year (1987–1997) multi-center cohort study to explicate origins of black–white disparities in obesity and its effects on cardiovascular disease risk factors in women (21). Race was self-declared and enrollment was restricted to 9- and 10-year-old girls from racially concordant households, i.e., to girls who said they were black or white and whose parents or guardians said that they were black or white, respectively. In an ancillary project, the Cincinnati clinic measured fasting insulin and glucose at mean ages 10, 16, and 19 (11). After completion of NGHS, the Cincinnati clinic conducted investigator-initiated studies that extended follow-up with measurement of insulin, glucose, and BMI to age 24 and insulin only to age 25.
In the NGHS (11) and Cincinnati extended follow-up, procedures followed were in accordance with the ethical standards of the institutional review boards of the Centers, which approved the study. Signed informed consent was obtained from the girls' parents or guardians and assent from the girls as minors as well as their signed consent as adults.
Clinical variables and elective behavior methodologies
The NGHS used BMI to assess overweight and obesity annually according to a standard protocol (21), and waist circumference beginning at mean age 11 as an indicator of fat patterning.
Fasting glucose was measured at mean age 10 using a hexokinase reagent (Boehringer–Mannheim, Mannheim, Germany) and at mean age 19 through mean age 24 using the glucose oxidase method with the Hitachi 704 Chemistry Analyzer (Roche Diagnostics, Indianapolis, IN). Coefficients of variation ranged from 2 to 7% for glucose.
Fasting insulin was measured at mean age 10 by the Michigan Diabetes Research and Training Center (Ann Arbor, MI) and in at age 16 by the Endocrine Lab (Children's Hospital Medical Center, Cincinnati, OH) by competitive protein-binding radioimmunoassays. Since outcomes were virtually the same whether insulin levels were transformed into Z scores (data not shown), only insulin data is displayed. We used fasting insulin and homeostatic model assessment of insulin resistance as measures of insulin resistance. Since outcomes were virtually the same whether insulin or homeostatic model assessment of insulin resistance was used (data not shown), only data for insulin is displayed.
NGHS protocols for the collection of data on elective behaviors, including physical activity, and TV viewing time have been described previously (2). An accelerometer was used over a 24-h period, including a school day, to record activity (Table 1). A validated habitual physical activity questionnaire was used to measure leisure-time physical activity at age 10 including a school day, and separately for a summer day, once per year from ages 10 to 17 (Table 1). A general physical activity questionnaire was filled out at age 19 (Table 1) (22). Participants checked off the TV programs watched during the previous week, plus time spent watching videos and music TV shows. The total time spent watching TV and videos was then calculated for the week, both at age 10 and for ages 10–19 (11,22). Household income IN 1987-8 was categorized as <$10,000/year, ≥$10,000 but <$30,000, ≥ $30,000 but <$50,000, and ≥ $50,000. Parents' education level was categorized as 1 (≤high school), 2 (some college), and 3 (college and beyond).
Table 1. Class 3 obesity at age 24, and characteristics of 549 NHLBI Growth and Health study girls at study entry (ages 10, 11) and during follow-up, ages 10–19
Dietary data was collected annually at mean ages 10, 11, 12, 13, 14, 16, 17, and 19 (Table 1). Dietary data was obtained using 3-day dietary records, retrieved by registered dietitians and coded using the most recent version of the Nutrition Data System for Research Software (2,21) developed for the Nutrition Coordinating Center, U Minnesota, Minneapolis, MN, for calculation of total calories and calories from protein, fat, and carbohydrate.
Diagnosis and exclusion of type 1 diabetes
As previously described (23), NGHS subjects having fasting blood glucose ≥126 mg/dl (24) at mean age 10 or type 1 diabetes mellitus at any time from mean age 10 through mean age 25 were excluded from this report. Diagnosis of type 1 diabetes was based on World Health Organization criteria, fasting glucose ≥126 mg/dl at entry, and self-reported diabetes with treatment by a physician (24).
Based on previous reports by our group (2,11,19) and by others (1,3,4,9,21,22,25,26,27,28,29), we evaluated multiple factors at study entry (age 10) and from ages 10 to 19 thought to be associated with development of Class 3 obesity (Table 1). Wilcoxon tests were used to compare each of these variables between two groups (girls who attained Class 3 BMI at age 24 and girls who did not).
The cohort was categorized by BMI at age 10 using Center for Disease Control (CDC) 2000 age-specific 85th percentiles, and was further characterized by BMI at age 24 (Table 2). Mantel–Haenszel χ2 analysis was used to measure the association of categories of BMI at age 10 with BMI at age 24 (Table 2). Also, Spearman correlations were calculated between age 10 BMI with age 24 BMI and with childhood insulin in girls categorized by BMI at age 10 using the CDC 2000 cutpoints (Table 2).
Table 2. Relationships of BMI at age 10 ≥ and < the Center for Disease Control 2000 age-specific 85th percentile to Class 1, 2, and 3 obesity at age 24 in 549 NHLBI Growth and Health study girls
The variables in Table 1 that differed between those with Class 3 BMI at age 24 and not Class 3 (by Wilcoxon tests), were then assessed for sensitivity and specificity in predicting Class 3 BMI at age 24 (Table 3). The variables were categorized by the CDC age-specific top 15th percentile for BMI, top 10th percentile for waist circumference at age 11 and childhood insulin, bottom two categories for household income, then sensitivity, specificity, positive predictive value, and negative predictive value in prediction of Class 3 obesity at age 24 year were calculated (Table 3). Only those variables which significant (P < 0.05, by χ2 test) associated with Class 3 obesity at age 24 were displayed in Table 3.
Table 3. Sensitivity, specificity, positive and negative predictive values for Class 3 BMI (≥40 kg/m2) at age 24 in 549 NHLBI Growth and Health study girls
To assess significant explanatory variables for Class 3 obesity at age 24, we used those categorical variables that were significantly associated with age 24 Class 3 obesity in Table 3. We added average dietary calories and percent calories from fat from age 10 to 19 (all as categorical variables, top decile = 2 vs. other 9 deciles = 1). We also added two- and three-way interaction terms. The two-way interaction term included interaction of race (white = 1, black = 2) with childhood insulin (top decile =2, other 1), and the 3-way interaction terms included race, childhood insulin, and average calorie or percent calorie from fat, all as categorical variables. Significant explanatory variables from the stepwise logistic regression for the development of Class 3 BMI were displayed in Table 4. In the resultant model (Table 4), there were three significant explanatory variables, including a three-way interaction term race×childhood insulin×calorie intake. The odds ratio (OR) (95% confidence interval (CI)) for age 24 Class 3 obesity between groups defined by this three-way interaction term were calculated. The OR (95% CI) for all groups according to the three-factor levels vs. the lowest risk group (white, both insulin and calories were not in top decile—0 factors in the upper level) were exhibited in Figure 1.
Table 4. Stepwise logistic regression: significant explanatory variables for BMI (≥40 kg/m2, ≥40) at age 24 in 549 NHLBI Growth and Health study girls
For each of the three significant explanatory variables from stepwise logistic regression, the relative risk (RR) for developing Class 3 obesity by age 24 was individually assessed (Table 5). For age 11 waist circumference and age 10 BMI, there were only two levels for each variable, and RR was calculated as risk in high level vs risk in the low level (Table 5). For the three-factor interaction term, race×childhood insulin×calorie intake, there were four levels, since the highest level had only five girls, we combined the top two levels as level C (82 girls), and displayed the RRs of levels B and C vs. the lowest level A (Table 5).
Table 5. Relative risk for Class 3 BMI (≥40 kg/m2) by age 24 in 549 NHLBI Growth and Health study girls, using explanatory variables significant from stepwise selection
To examine the possible effect of missing data in the NGHS girls not included in the analysis sample due to missing childhood insulin or BMI at age 24, we compared age 10 entry data on the girls included (n = 549) and not included (n = 316) in analyses.
After exclusion of girls with type 1 diabetes (n = 7), there were 865 girls in the NGHS Cincinnati cohort (436 black, 429 white), of whom 549 (64%, 293 black, 256 white) were included in analyses and 316 not included due to missing childhood insulin or BMI by age 24. There were no differences (P > 0.05) in any age 10 entry measurements between the girls included and not included in the analyses except for race, a greater percentage of black (67%) than white (60%) participants were included in study (P = 0.019).
Characteristics of the study population
Descriptive characteristics of the study population at baseline (ages 10, 11) and, for some variables, from ages 10 to 19, as displayed in Table 1, categorized by Class 3 obesity present or not present at age 24. Of 256 white girls in the cohort, by age 24, 6 (2.3%) had Class 3 obesity at age 24, in contrast to 36 of 293 black girls (12.3%), χ2 = 19.1, P < 0.0001 (Table 1). Of the 42 girls with Class 3 obesity at age 24, 36 (86%) were black (Table 1).
Girls who had Class 3 obesity at age 24 were taller and heavier at age 10, and had higher age 10 BMI and waist circumference, all P ≤ 0.0002 (Table 1).
Of the 549 measures of childhood insulin, 407 (74%) were first measured at mean age 10, 142 first measured at age 16. Of the 277 girls who had insulin measured at both age 10 and age 16, mean ± s.d. (median) insulin was 13.5 ± 12.7 (10) µU/ml at age 10 and was 1 µU/ml higher (P = 0.03, paired Wilcoxon) at age 16 (14.5 ± 15.7 (11.2) µU/ml. In these 277 girls, insulin maintained its quintile rank from age 10 to age 16, Mantel–Haenszel χ2 = 6.4, P = 0.011.
Childhood insulin was higher in girls who had Class 3 obesity at age 24 (Table 1).
The percent of total calories as fat was marginally higher at age 10 (P = 0.053), and higher from ages 10 to 19 (P = 0.0019) in girls who had Class 3 obesity at age 24, P = 0.053 (Table 1).
At age 10, girls who had Class 3 obesity at age 24 did not differ from girls without Class 3 obesity at age 24 for pedometer score, or reported summer physical activity, and did not differ by average reported physical activity pattern score from age 10 to 17, average summer physical activity pattern score from age 10 to 17, or general activity level at age 19 (Table 1). In contrast to these findings, girls who had Class 3 obesity at age 24 had higher reported physical activity pattern score at age 10 (Table 1).
Girls who later had Class 3 obesity had higher TV hours weekly, both at age 10 and from ages 10 to 19 years (Table 1).
Household income category was lower in girls who had Class 3 obesity at age 24 (Table 1).
Distributions of BMI at ages 10 and 24
At mean age 10, the distribution of BMI in black girls (mean ± s.d. =18.9 ± 4.2, median 17.9 kg/m2) was shifted towards higher values than in white girls (17.5 ± 3.0, median 16.9 kg/m2, P = 0.0004). By age 24, the shift in the BMI distribution toward higher BMI in black girls was more exaggerated than at age 10, (29.8 ± 8.5, median 28.0 kg/m2 in black, 24.7 ± 5.8, median 23.1 kg/m2 in white, P < 0.0001).
BMI at age 10 was closely correlated with BMI at age 24, r = 0.69, P < 0.0001 for blacks, r = 0.66, P < 0.0001 for whites (Table 2). BMI at age 10 was correlated with childhood insulin, r = 0.44, P < 0.0001 in blacks, r = 0.35, P < 0.0001 in whites (Table 2).
Consistent with their higher BMI at age 10, the prevalence of Class 3 obesity was greater in black girls than in white girls at age 24 (Table 2). Of 42 girls with Class 3 obesity at age 24, 36 (86%) were black (Table 2). There was a strong association between BMI status at ages 10 and 24 in both black and white girls (P < 0.0001 by Mantel–Haenszel χ2 test); girls ≥ the CDC 2000 85th percentile for BMI at age 10 were more likely than girls with BMI < the CDC 85th percentile to have Class 1, 2, or 3 obesity at age 24 (Table 2).
Age 10 high BMI tracked more closely with age 24 high BMI for black girls than it did for white girls (Table 2). Of white girls with age 10 BMI ≥ the CDC 2000 85th percentile, 23 of 45 (51%) had BMI ≥30 at age 24, vs. 71 of 95 (75%) of black girls, (Table 1), χ2 = 7.7, P = 0.005. Conversely, BMI below the CDC 85th percentile at age 10 was strongly associated with adult BMI <30, but more so in whites than blacks (Table 2). Of 211 white girls with BMI < the CDC 85th percentile at age 10, at age 24, 157 (74%) had BMI <25 kg/m2 (normal weight) and 196 (93%) were <30 kg/m2 (obese) at age 24 (Table 2). Of 198 black girls with BMI < the CDC 85th percentile at age 10, at age 24, 91 (46%) had BMI <25 and 157 (79%) had BMI <30, (Table 2), χ2 = 16.0, P < 0.0001.
Sensitivity, specificity, and positive and negative predictive values for categorical variables to predict Class 3 obesity at age 24 by bivariate analysis are summarized in Table 3. High negative predictive values which indicated low risk for developing Class 3 obesity were seen for white race, bottom 85th percentile BMI at age 10, bottom nine deciles waist circumference at age 11, bottom nine deciles childhood insulin, and the two highest household income categories for 1987–8 (≥$30,000, ≥$50,000/year) (Table 3).
Using stepwise logistic regression with categorical explanatory variables, three explanatory variables were independently significant for attaining BMI Class 3 at age 24: top decile waist circumference at age 11, CDC top 15th percentile BMI at age 10, and a three-way interaction race×childhood insulin×average caloric intake from age 10 to 19 (Table 4). The logistic regression model for Class 3 obesity by age 24 had high discriminant ability, with area under the receiver-operator curve = 0.90 (Table 4). Based on this logistic regression model, in black girls with top decile caloric intake and top decile insulin, then their odds of developing Class 3 obesity were 7.7 (2.5–23.5) fold greater than if their insulin was not in the top decile. In contrast, in white girls with top decile caloric intake and top decile insulin, then their odds of developing Class 3 obesity were 2.8 (1.6–4.8) fold greater than if their insulin was not in the top decile. Similarly, the OR (95% CI) for each group defined by the three-way interaction term vs. the lowest level (white girls with both childhood insulin and total caloric intake in the bottom nine deciles—zero risk factors in the upper level) was calculated, and displayed in Figure 1. If there were three upper level factors (black race, top decile childhood insulin, top decile total caloric intake for ages 10–19) girls were 35 times more likely to have Class 3 obesity by age 24 (Figure 1). Girls with two of the three factors in upper levels were 4.63 times more likely to have Class 3 obesity by age 24, and those with one upper level risk factor, 1.67 times (Figure 1).
The RR for developing Class 3 obesity by age 24 for each predictor in the logistic model from Table 4 was investigated (Table 5). Girls with top decile waist circumference at age 11 (vs. bottom nine deciles) had RR for age 24 Class 3 obesity = 12.7, 95% CI 7.3–22.1 (Table 5). Girls with age 10 BMI in the CDC top 15th percentile had RR = 14.6, 95% CI 6.64–32.1 (Table 5). For the three-factor interaction term race×childhood insulin×average caloric intake from age 10 to 19, girls were classified as three levels (Table 5). For girls with at least two high factors (n = 82) vs. girls with no high factors (white, insulin not in top decile, calories not in top decile), RR was 11.4 (95% CI 3.9–33.2) (Table 5). When one factor was high vs. all three factors not high, RR was 5.55 (95% CI 1.9–15.8) (Table 5).
Black–white differences in the prevalence of obesity in women are thought to contribute significantly to their differences in cardiovascular disease morbidity and mortality (30), T2DM, and certain cancers. Black–white differences in obesity begin during the peripubertal period and increase during adolescence (31). Evaluating 15-year secular trends in BMI and T2DM during the past 25 years, the prevalence of obesity and T2DM have both increased and the age of onset for T2DM has decreased dramatically, especially in black females (32). T2DM now presents in the teenage years with increasing frequency, perhaps because the obese insulin-resistant state appears earlier in childhood, thus reducing β-cell reserves (33).
Our current study identifies a group of factors in childhood that a clinician could measure, with particular importance in black girls (age 10 BMI, age 11 waist circumference, age 10–16 fasting serum insulin) to identify girls with high or low likelihood of reaching Class 3 obesity status by age 24, facilitating supervised dietary and exercise programs, and, possibly, insulin sensitizing drugs (34), directed to primary prevention of Class 3 obesity at age 24.
How do race and insulin affect obesity risk, and what are the physiologic differences between whites and blacks that create the obesogenic metabolic condition? Higher insulin resistance, higher insulin throughout the BMI distribution, and differences in insulin secretion dynamics appear to be major physiologic difference between blacks and whites (19,35,36,37,38), increasing risk of obesity and T2DM in blacks. Moreover, black girls have higher caloric intake than white girls during adolescence (11,19), despite under-reporting of energy intake (39). Within this frame of reference, as shown in the current report, the three-way interaction between race, caloric intake, and insulin and the two-way interaction between race and insulin results in disproportionate percentages of black girls by age 24 having Class 3 obesity. The findings in the current report, in our previous studies (11,19), and from other reports (40,41,42) provide additional support for the growing idea that primary hyperinsulinemia is a contributing cause of obesity, particularly in blacks. Hence, beyond lifestyle intervention, obese black girls with hyperinsulinemia should benefit optimally from the insulin sensitizer metformin, which would lower their insulin levels (34). Mosca et al. (40) reported that insulin resistance in adults appears to interact with high-fat diets to increase weight gain. Kahn and Flier (41) have proposed that insulin resistance and hyperinsulinemia, in addition to being caused by obesity, can contribute to the development of obesity via an interaction with total and fat calories, as observed in our current and previous studies (11,19). Lustig (43) has theorized that the genesis of obesity may lie in hyperinsulinemia.
Since BMI at age 10 and waist circumference at age 11 are the preponderant significant explanatory variables for Class 3 obesity at age 24, to what extent does measuring childhood insulin improve the ability to predict Class 3 obesity in young adulthood? In the current report, congruent with results from previous tracking studies (4,44) age 10 BMI and age 11 waist circumference were major significant explanatory variables for Class 3 BMI at age 24. Odds for Class 3 obesity by age 24 were expanded sevenfold by CDC top 15th percentile BMI at age 10, 5.7-fold by top decile waist circumference at age 11, and 1.7-fold by 1 unit increment in the three-way interaction of race×childhood insulin×average caloric intake from age 10 to age 19. Measuring childhood insulin is especially useful in black girls with top decile caloric intake. In the current study, if their insulin was in the top decile then their odds of developing Class 3 obesity were 7.7-fold higher than those in black girls with top decile caloric intake but without top decile insulin. Our data show that childhood insulin improves the ability to predict Class 3 obesity in young adulthood, and that insulin is a useful clinical measure beyond BMI, particularly in black girls. Beyond Class 3 obesity, all classes of obesity pose serious health risks, particularly over a lifetime beginning from childhood (12), and should be identified as early as possible.
In this NGHS cohort of schoolgirls, we had previously reported that childhood homeostatic model assessment of insulin resistance interacted with percentage of calories from fat to predict 10-year change in BMI and in waist circumference in both white and black girls from age 9–10 to age 18–19 (11). Separately, in 215 pairs of black and white schoolgirls matched at age 9–10 by BMI, insulin, and pubertal stage, 10-year increases in BMI were predicted by age 9–10 BMI, 10-year change in insulin, and a three-way interaction of age 9–10 insulin×adolescent caloric intake×race (19). We concluded that there appeared to be intrinsic black–white metabolic differences that lead to greater gains in fat and weight during adolescence in black girls (19). Based on the results from our current study, and from previous studies in the NGHS cohort (11,19), if overweight and obesity are present and if elevated insulin is documented at ages 10–16, and if hyperinsulinemia increases during early adolescence, then steps to restrict diet, increase physical activity, and decrease insulin and insulin resistance (34,45,46,47,48) may be warranted.
Our study has the following limitations. The study design, which is epidemiological, does not enable direct causal inference. Participants were not a random selection of the United States, as in National Health and Nutrition Examination Survey (3), but came from a biracial school population. Thus, the data, while suggestive, need to be confirmed and cannot be extrapolated to all adolescent girls. Three-day dietary diary data may less optimally reflect actual dietary intake than a 7-day record (49).
In our current study, the significant association between Class 3 obesity by age 24 with waist at age 11, age 10 BMI, and a race×caloric intake×childhood insulin interaction highlighted childhood obesity, central obesity, insulin, and caloric intake, as major modifiable childhood targets in the prevention of or reduction of Class 3 obesity in young adulthood, particularly in black girls. We speculate that diet, exercise, and, if needed, metformin (34), have promise in primary prevention of Class 3 obesity if initiated in hyperinsulinemic obese adolescents. We speculate that initiating these interventions in a conservative, stepped fashion might reduce the development of obesity and development of impaired fasting glucose and T2DM, leading to an ultimate goal of primary prevention of both progressive obesity and T2DM (42,50,51,52).
This research was supported in part by NIH-HL55025, 48941, HL52911 and HL66430 (J.A.M. and S.R.D.), and by the Lipoprotein Research Fund of the Jewish Hospital of Cincinnati (C.J.G.).