We address an important public health problem using a uniform modeling approach and a common analytic strategy in five well-characterized cohorts representing different political-economic, cultural, and nutritional circumstances. Despite differences in economic and other conditions across populations and in methods used to measure body composition, we found certain consistencies across the five sites. We also found that results were contingent upon the body composition measure employed. When adult body composition was modeled as separate compartments of fat-free and fat mass, birth weight tended to be a stronger predictor of adult fat-free mass than of fat mass irrespective of gender and site, consistent with prior evidence of beneficial effects of fetal growth on adult body composition (Singhal et al., 2003
; Wells et al., 2007
). In contrast, CW measured in mid-childhood was not differentially related to fat-free or fat mass in adulthood in any site or gender, pointing to fewer potential benefits to lean tissue in relation to weight gain at older ages. Modeling % body fat as an outcome revealed that the correlations of most BW and CW scores with fat mass, although weaker than correlations with those with fat-free mass when both are expressed in standardized units, were sufficient to modestly increase % body fat. In males, BW was generally the weakest predictor of % body fat. Although there was heterogeneity across models, CW at 12 months, reflecting weight gain in the first year of life, and mid-childhood CW, reflecting early childhood weight gain, tended to be the strongest predictors of adult % body fat in both genders. These findings show that patterns of early life weight gain predict adult body composition, while also documenting heterogeneity in these relationships by gender and across populations.
Modeling relationships between early weight gain and adult fatness requires grappling with several issues, including the choice of outcomes, the age ranges of early life weight gain to be investigated as predictors, and the related statistical problem that weight gain in a given age interval tends to be highly correlated with gain in adjacent intervals. With respect to the first issue, adult body mass index (BMI) is a commonly reported outcome in studies of early life determinants of adult body composition, with consistent positive associations being reported with birth weight and faster weight gains in infancy and childhood (Baird et al., 2005
; Monteiro and Victora 2005
). Because BMI does not separate fat from fat-free mass, such results must be interpreted with caution. As an alternative approach, we investigated fat mass as an outcome, and evaluated whether early weight trajectories are associated more strongly with fat than with fat-free mass in adulthood. In our view, this is a useful approach, because one would like children to build up fat-free mass without gaining excess body fat as adults. Finally, we also considered the % of body weight that is fat as an outcome, which provides insight into how differential effects on lean and fat compartments contribute to changes in body composition.
The second set of issues involve how best to model early weight trajectories as predictors of adult body composition. Past work has shown that it is important to separate the long-term effects of early and later childhood weight gain (Victora et al., 2008
). Some of the first studies on the long-term consequences of growth patterns in childhood reported on weight gains from birth to school age (Eriksson et al., 2001
; Parsons et al., 2001
), and were therefore unable to identify age ranges associated with different risks. We opted to model the relationship between early weight during different intervals from birth to mid-childhood, which allowed us to tease apart the relative importance of change in weight at different ages to adult fat and fat-free mass. However, because weight at the end of an age range is the same as the weight at the beginning of the next range, collinearity would be unavoidable if simple weight gains or weight velocities were modeled as independent predictors. To avoid this problem, we used conditional weight scores, which estimate the effect of that component of weight gain in each interval that is independent of gain in prior intervals.
We found that associations relating early CW scores to adult body composition were not consistent for males and females, and in addition varied substantially across sites. We anticipated heterogeneity across sites in light of the many factors that differentiated the sites and that we were not able to fully control in our models. Notably, the age at which subjects had their adult body composition measured varied by site, as did the method used to estimate fat and fat-free mass. In addition, the populations varied in prevalence of low birth weight, early life undernutrition and adult overweight, reflecting diverse settings in which the nutrition and epidemiologic transitions have had differential influences on adult health and nutritional status. The significant heterogeneity in the pattern and magnitude of coefficients across site and by gender required that we pursue stratified rather than pooled models. Nevertheless, underlying consistencies were apparent across the sites that point to the differential importance of weight gain in different early life age intervals as predictors of distinct components of adult body composition.
Conditional weight scores are positive when growth in the preceding interval was faster than expected based upon that individual’s weight trajectory entering that interval and the mean population growth during the interval. Thus, we interpret our conditional weight models as indicating that faster than expected weight gains from birth to 12 months and from 24 months to mid childhood tend to be more strongly associated with adult fat mass than are birth weight or relatively rapid weight gain from 12-24 months. Adjusting for height attenuated coefficients predicting fat-free mass, especially in males, but left coefficients predicting fat mass largely unchanged. This suggests that early weight gain influences fat-free mass in part via effects on adult stature, consistent with prior analysis in these cohorts (Stein et al., 2009
Certain consistencies emerged across sites when we modeled fat and fat-free mass simultaneously as outcomes. Here fat and fat-free mass were modeled as site- and sex-specific SD scores, thus allowing for a direct comparison of effect sizes of each CW score predicting fat and fat-free mass while also taking into account the average differences in adult fat-free mass between males and females. In these models, birth weight and to a lesser extent CW in the first two years of life tended to be more strongly associated with fat-free than with fat mass, but this differential effect on fat-free mass was absent in 8 of 9 models by mid-childhood.
The average adult fat-free mass tended to be 3-5 times the mass of fat in the cohorts considered here, while variability for fat mass was also larger. Thus, even relatively weaker relationships between early growth and adult fat mass than for fat-free mass could increase the proportion of adult weight accounted for by fat. Our models predicting % body fat confirmed this expectation. With few exceptions, faster earlier growth at any interval predicted an increase in adult % body fat in both males and females. Although the strength of these effects did vary by age, gender and site, there were certain consistencies across many of the cohorts, and in general findings mirrored trends seen in models predicting fat mass and fat-free mass. Just as CW at 12 months and mid-childhood tended to be strongest predictors of fat mass (), the weight gain represented in these intervals generally predicted the largest % increase in adult body fat. In the males, mid-childhood CW was the strongest predictor of % body fat in all the cohorts, while BW was the weakest predictor of % body fat among all male cohorts but South Africa. Among females, CW at 12 months and in mid-childhood tended to be strongest predictors of % body fat.
Although many early growth measures were significant predictors of adult % body fat, these effects were generally small. A full standard deviation change in most early growth measures, which represents a large change in early weight trajectory, predicted on average a 0.85 % change in adult body fat. The health impacts of a change of this magnitude are likely to be negligible in most of our cohorts, many of whom are relatively lean. Based upon these modest effect sizes, it seems likely that the many benefits to health and function known to result from improvements in early nutrition and growth (e.g. Victora et al 2008
) will outweigh any negative effects on health operating through changes in body composition.
Previously published single-site analyses of our cohorts are generally in line with the findings reported here, despite the fact that these analyses did not rely on conditional weights and none modeled fat and fat-free mass simultaneously. In Brazil, adult height was primarily determined by fetal and infant growth, whereas weight-related indices were more strongly influenced by later growth (Victora et al., 2007
). In India, BMI changes in infancy and early childhood were correlated more strongly with adult lean mass than with adiposity or central adiposity, whereas BMI changes in late childhood or adolescence were associated with increased adult adiposity and central adiposity (Sachdev et al., 2005
). In Guatemala, BMI in infancy and later childhood was positively associated with adult BMI, % fat, abdominal circumference, and fat-free mass; after the age of three years, associations were stronger with BMI, % fat and abdominal circumference than with fat-free mass (Corvalan et al., 2007
). In a recent analysis of the Philippine cohort, weight gain from birth to 6 months of age was found to be an especially strong predictor of adult stature, fat-free mass, and strength in males but these relationships were weaker in females (Kuzawa et al 2010
Other studies in high income populations have addressed these issues. Positive associations between birth weight and lean mass – to a greater extent than with fat mass – have been reported in the United Kingdom (Singhal et al., 2003
) and from a younger cohort in Brazil (Wells et al., 2005
). Also, the finding that weight gain from 24 months to mid-childhood is a stronger predictor of adult adiposity than is earlier weight gains has been reported in cohorts from the United Kingdom (Ong et al., 2000
), Brazil (Wells et al., 2005
), and Sweden (Ekelund et al., 2007
). At the same time, the relatively low average birth weight of most of our study populations underscores the distinct starting point of growth trajectories in these settings compared to most higher income populations. Future studies in high income nations employing the analysis approach used here would help clarify how the predictors of adult body composition might vary across these contexts.
The different strengths of the coefficients linking CW during the first and second years of life with adult body composition may partly reflect absolute differences in weight gain during each growth interval or the increasing stability of growth trajectories as children mature. In our cohorts, absolute gains in body weight during the first year of life were roughly three times the gains in the second year of life (), suggesting a greater potential for variation in early postnatal growth to influence adult body composition. In addition, there may be important biological critical periods during specific developmental windows, such as the early postnatal surge in sex steroid production (Winter et al., 1976
), which could influence body composition development and the pattern of gender differences in adult body composition (Kuzawa et al., 2010
Our observational data can reveal patterns of association but cannot speak to causes. Indeed, it is possible that part of the association between early conditional weight and adult body composition reflects the pleiotropic effects of genes. Polymorphic loci with effects on early weight growth patterns have been identified (Sovio et al., 2009
), and if such genes also influence adult body composition, this could lead to correlations between early conditional weight scores and adult body composition (Suarez et al., 1997
). Although we cannot rule out such effects, prior research conducted in the cohorts analyzed here (Stein et al., 2009
; Victora et al., 1987
), and in populations living under similar socioeconomic and political-economic settings (Billewicz and McGregor 1982
), has underscored the overwhelming importance of environmental factors like nutrition, infant feeding practices, and early life infectious morbidity as determinants of skeletal growth and weight gain during the first 2-3 years of postnatal life, highlighting the potential for environmental contributions to the relationships that we document.
Additional limitations of the analyses reported here warrant mention. As alluded to above, methods for measuring adult body composition varied by site and some of the between-site heterogeneity undoubtedly reflects biases inherent in these methods. Although this limits the conclusions that we may draw regarding population differences, these concerns apply less to our findings of sex differences, which were found within site using common methods. The instruments used to measure diet, lifestyle, or environmental conditions across our five sites were also not directly comparable, which constrained our ability to adjust for potentially important covariates.
Although our analyses were limited by the ages at which weight measurements were available across sites, these measures allowed us to probe impacts at age intervals previously shown to be important as predictors of adult body composition in other populations. The ages of adult body composition measurement were more variable across sites, which likely contributed to the heterogeneity in relationships that we document. In this respect, the generally weaker relationships between early growth and fat-free mass among the South African males might partially reflect their relatively young age (~15.5 years), and the fact that males would have been at various stages of puberty and adult height attainment when body composition was measured. Future analysis of this cohort after they have achieved final adult stature would help clarify the importance of this source of biological variation as an influence on the relationships documented here. While our focus here has been on body composition per se, future work should also evaluate associations with regional fat patterning, as this may be an important link between early growth processes and the metabolic syndrome and cardiovascular disease risk (Despres and Lemieux 2006