The Relation of Childhood BMI to Adult Adiposity: The Bogalusa Heart Study

METHODS

RESULTS

Thirty-seven percent of all eligible children were reexamined in adulthood (Table 1), and compared with children lost to follow-up, the study cohort was slightly older (10.0 vs 8.9 years, boys) and participated in earlier (1975 vs 1978, boys) examinations. However, mean levels of weight-for-age, height-for-age, and BMI-for-age were close to 0 in both groups. After adjustment for age and examination year, the only significant difference between the 2 groups was a slightly higher prevalence of overweight (BMI ≥ 95th centile) among boys in the study cohort (7% vs 6%; P = .02).

Of the 1115 boys and 1495 girls in the study cohort, 76% were initially examined in 1973–1974, and 94% participated in either the 1988–1991 or 1995–1996 screening of adults. Ages at baseline ranged from 2 to 17 years, and the mean age at follow-up was 27.5 years (range: 18–37 years). At follow-up, 23% were obese (BMI ≥ 30 kg/m²), and 25% were overfat (mean SF in the upper quartile).

Childhood levels of both BMI-for-age and triceps SF were significantly associated with adult levels of BMI and mean SF, with the overall correlations ranging from r = 0.44 to 0.64 (Table 2). Although the strongest associations were seen between BMI levels in childhood and adulthood, childhood BMI was also associated with adult levels of mean SF in all age groups including 2- to 5-year-olds (boys: r = 0.41; girls: r = 0.33). Furthermore, childhood BMI was almost as strongly associated with adult mean SF, as was childhood triceps SF; correlations among 6- to 8-year-old boys, for example, were r = 0.48 (BMI-for-age) and r = 0.50 (triceps SF). In general, the magnitudes of the correlations were stronger among boys than among girls and stronger among older (9–17 years) children than younger children. Between the ages of 2 and 5 years and 9 and 11 years, the association between BMI-for-age and adult mean SF increased from r = 0.41 to 0.56 among boys and from r = 0.33 to 0.51 among girls.

We then examined the ability of BMI to predict which children would be obese (BMI ≥ 30 kg/m²) or overfat (upper quartile of mean SF) in adulthood (Table 3). The positive predictive value for adult obesity and overfat consistently increased as childhood BMI-for-age increased, and this was evident among even the youngest children. Among 2- to 5-year-olds girls, for example, 15% (18 of 121) of those with a childhood BMI ≤ 50th centile became overfat adults, whereas 53% of overweight (BMI ≥ 95th centile) girls became overfat adults. Among overweight children, positive predictive values were generally higher for adult obesity than for adult overfat; of the 51 overweight 9- to 11-year-old boys, for example, 76% were obese in adulthood, and 67% were overfat. With the exception of estimates among 2- to 5-year-old boys, which were based on small numbers, the positive predictive values of childhood overweight for adult obesity generally increased with age. In contrast, the positive predictive values for adult overfat showed little variation with age, and similar estimates were seen among 6- to 8-year-olds (70–84%) and 15- to 17-year-olds (65–81%). Additional analyses (data not shown) indicated that changes in (age-adjusted) levels of BMI and triceps SF during childhood showed similar associations with adult overfat.

To examine further the importance of BMI levels among the 2- to 9-year-olds (n = 1362), various regression models were constructed to determine if levels of BMI-for-age and triceps SF were independently associated with adult mean SF. (Models also included race, gender, childhood age, adult age, and various interaction terms.) Although the inclusion of both BMI-for-age and triceps SF in a model only slightly improved the prediction of adult mean SF as compared with models containing only 1 of these variables (multiple R² = 0.44 vs 0.39 and 0.42), both childhood characteristics were independently associated (P < .001) with adult adiposity.

These independent effects are shown in Fig 1 for boys (left) and girls (right). Predicted levels of adult mean SF (y-axis) are shown among white adults according to childhood levels of BMI-for-age (x-axis) and the 10th, 50th, and 90th centiles of triceps SF. For example, among boys with a triceps SF of 10 mm (median value), estimated adult levels of mean SF increased from 14 to 17 mm as the BMI z score increased from 0 to 1.645 (95th centile of BMI). Similarly, among boys with a BMI z score of 1.0, the predicted adult mean SF increased by ∼5 mm as triceps SF increased from 10 to 17 mm. The observed associations with adult mean SF did not significantly differ by race or gender, but adult levels were higher among women (Fig 1 right) than men and among black women than white women (data not shown).

• Download figure
• Open in new tab
• Download powerpoint

Fig 1.

Predicted adult levels of mean SF according to childhood levels of BMI-for-age (x-axis) and triceps SF; these analyses are restricted to 2- to 9-year-old boys (left) and girls (right). Predicted levels of mean SF are shown for children at the 10th, 50th, and 90th centiles of triceps SF; the regression models also controlled for race, childhood age, adult age, and various interactions. Childhood levels of both BMI-for-age and triceps SF were independently associated with adult mean SF (P < .001 for each association) in these regression models.

DISCUSSION

Our results show that childhood levels of both BMI and triceps SF are associated with adult adiposity, as assessed ∼18 years later by the mean thickness of the triceps and subscapular SFs. Although the magnitude of the correlations between childhood and adult levels increased with the age at which the childhood measurement was obtained, the BMIs of even 2- to 5-year-olds were moderately associated (r = 0.33–0.41) with adult adiposity. Furthermore, overweight 2- to 5-year-old boys and girls were >4 times as likely to become overfat adults (15 of 23 [65%]) as were those who had a BMI < 50th centile (30 of 201 [15%]). Childhood triceps SF showed a slightly stronger association with adult adiposity than did childhood BMI, but both characteristics provided independent information on adult adiposity.

Several investigators have examined either the tracking of BMI from childhood to adulthood or the accuracy of childhood BMI in predicting adult obesity. Although there are exceptions,²¹ the larger studies consistently found that as compared with relatively thin children, fatter children are more likely to become obese adults.^22–27 Although most investigators have examined the tracking of weight-height indices, similar (but slightly weaker) longitudinal associations have been observed for various SFs.^22,24,28,29 For example, among 9- to 10-year-olds who were followed for ∼18 years, the tracking of BMI ranged from r = 0.60 (females) to r = 0.73 (males), whereas longitudinal correlations for triceps SF ranged from r = 0.50 to 0.58.²⁴ Other investigators have also reported correlations that range from r = ∼0.4 to 0.5 for the triceps SF²⁹ and biceps SF²² measured at ages 6 to 9 years and 18 to 25 years. These estimates agree well with the associations we observed between childhood triceps SF and adult mean SF (r = 0.38–0.59) over 17.6 (mean) years. We also found that the tracking of triceps SF was weakest among 2- to 5-year- olds, and SFs measured among 1- to 5-year-olds have shown²⁸ little association (r = 0.14) with levels measured 20 years later.

A report from the Newcastle Thousand Families Study¹¹ found that although BMI levels at ages 9 years and 50 years were correlated (r = 0.24), childhood BMI was not associated with adult adiposity (r = 0.10; P > .05) as assessed by bioelectrical impedance analysis. These contrasting associations have led several authors to suggest that childhood BMI is a poor indicator of adiposity, and that the tracking of BMI reflects the persistence of “body build” rather than adiposity.^11–13 Cross-sectional studies, however, have found that childhood BMI is associated with body fatness as assessed by dual-energy x-ray absorptiometry.^3–6 In addition, a previous study³⁰ reported that increases in BMI percentile rankings between the ages of 3 and 9 years among girls was associated with adult adiposity as determined by hydrostatic weighing. Therefore, it is unlikely that childhood BMI is merely an indicator of body build.

It is possible that the negative findings of the Newcastle Thousand Families Study are due to the relative thinness of children born in the United Kingdom in 1947. As compared with 9-year-olds in the current study, who were born between 1963 and 1976, similarly-aged children in Newcastle, on average, weighed 9 lb less and were 2 inches shorter,³¹ resulting in lower BMI levels. Furthermore, 16% of 9-year-olds in the current study had a BMI > 95th centile in Newcastle (boys: 20.5 kg/m²; girls: 18.5 kg/m²) (C. M. Wright, MD, written communication, 2003). As the accuracy of BMI as an index of body fatness increases with the degree of adiposity,^32–34 it is possible that differences in childhood BMI levels between the 2 studies could account for the contrasting associations with adult adiposity. The variability in BMI levels among thinner children largely reflects differences in fat-free mass,³⁴ and if this were the case in Newcastle, childhood BMI levels would not be expected to be associated with adult adiposity. In contrast, the BMI levels of Bogalusa children may reflect differences in fat mass more accurately.

The negative findings of the Newcastle study may also reflect the limitations of bioelectrical impedance analysis. Although levels of various cardiovascular disease risk factors showed fairly similar cross-sectional associations with BMI and body fatness among adults in the Newcastle study, some investigators^35–37 have found bioelectrical impedance estimates of body fat to be less accurate than those based on SFs. In addition, bioelectrical impedance estimates can be influenced by the possible association between electrical conductivity and the amount of fat mass,³⁵ the choice of the calibration equation,³⁸ and various clinical characteristics.³⁹

Although we found that most overweight (BMI ≥ 95th centile) children were obese or overfat in adulthood, with positive predictive values ranging from 53% to 90%, only ∼20% of obese or overfat adults had been overweight children. This estimate is comparable to those reported in other studies,^11,23,28,40 but it should be realized that estimates of sensitivity are greatly influenced by the prevalences and classifications of childhood overweight and adult obesity. For example, use of the 85th centile rather than the 95th centile as the childhood BMI cut point resulted in sensitivities of 42% (overfat) and 51% (obese). Furthermore, of the 114 adults with a BMI ≥ 40 kg/m², 46% had a childhood BMI ≥ 95th centile and 70% had a childhood BMI ≥ 85th centile. Although there are certainly additional factors that operate in early adulthood to promote obesity, a child with a high BMI-for-age is much more likely to become an obese adult than is a relatively thin child.

Several limitations of the current study should be considered when interpreting our results. Although only 37% of age-eligible children were reexamined as adults, their baseline levels of weight, height, and BMI were fairly similar to those among children who were not reexamined. In addition, we were able to follow children only through early adulthood (maximum age: 37 years), and it is likely that a longer follow-up would have reduced the magnitudes of the various associations. Although there are few studies on the tracking of childhood BMI beyond the age of 40 years, BMI levels at ages 30 and 50 years are highly correlated (r = 0.7-0.8).²¹ In addition, the current study did not investigate the associations with adult risk factors for coronary heart disease, but previous studies^11,41 have shown that the most important correlate is the adult BMI level. Childhood BMI, however, may be important in the development of atherosclerosis. We have found, for example, relatively high levels of the carotid artery intima-media thickness among obese adults who had been overweight children but not among obese adults who had been normal-weight children.⁴²

Our findings indicate that childhood BMI is associated with adiposity in adulthood and that overweight children have a greatly increased risk for becoming overfat adults. Our findings also provide a partial explanation for the variability observed in previous studies of BMI tracking from childhood to adulthood. Because the accuracy of childhood BMI as an indicator of body fatness increases with the degree of adiposity, the degree of tracking would be expected to be highest in populations with a high prevalence of overweight.