点击显示 收起
1 From the Department of Social Medicine, University of Bristol, Bristol, United Kingdom (AM, RH, KT, GDS, and YB-S), and the Department of Child Health, Wales College of Medicine, Cardiff University, Cardiff, United Kingdom (DD)
2 Supported by grants from Diabetes UK (ref 1192) and the British Heart Foundation (ref 97020). 3 Reprints not available. Address correspondence to A McCarthy, Academic Unit of Primary Care, The Grange, 1 Woodland Road, Bristol, BS8 1AU, United Kingdom. E-mail: anne.mccarthy{at}bristol.ac.uk.
ABSTRACT
Background: Birth weight has been shown to be positively associated with adult obesity, but relatively few studies have examined the associations with growth in specific periods of early childhood.
Objective: The objective was to assess the association of measures of growth between birth and 5 y of age with adult measures of adiposity.
Design: We conducted a longitudinal study of young adults from Barry and Caerphilly, United Kingdom, who had previously taken part between 1972 and 1974 in a randomized controlled trial of milk supplementation. We reexamined 679 men and women (72% of the target population) to measure body mass index (BMI; in kg/m2), waist-to-hip ratio, sagittal abdominal diameter, and waist circumference.
Results: An increase in weight velocity from 1 y and 9 mo to 5 y of age was the most important predictor of BMI, waist circumference, and sagittal abdominal diameter. A z-score increase in weight gain in this period was associated with an increase in BMI of 1.13 (95% CI: 0.69, 1.57; P < 0.001). Infant weight gain from 5 mo to 1 y and 9 mo was the strongest predictor of waist-to-hip ratio (0.51; 95% CI: 0.00, 1.02; P = 0.05).
Conclusions: Birth weight does not predict adiposity on the basis of weight gain in childhood. The association between adult adiposity and weight gain in different periods is variable and depends on the measure of adiposity that is used. It remains unclear whether early childhood is the optimum period in the life course for the primary prevention of adult adiposity.
Key Words: Early life • adiposity • life course
INTRODUCTION
Obesity is a principal risk factor for many adult chronic diseases (1). The prevalence of childhood obesity has increased dramatically in many developed countries over the past 20 y (2). This is of particular concern because longitudinal studies indicate that overweight and obese children are likely to grow into overweight and obese adults (3). Epidemiologic models in the post–World War II era focused primarily on adult risk factors in studying the causes of chronic diseases (4). However, over the past decade, life course models have highlighted the potential importance of risk factors in early life either in a critical or sensitive period or through risk accumulation (5). Two sets of findings are relevant to the present study: the first, a longitudinal follow-up of children that showed how cardiovascular risk factors track from childhood into adulthood (6), and the second, a study of the associations between small size at birth and a variety of adult chronic diseases, such as ischemic heart disease, type 2 diabetes, insulin resistance syndrome, and obesity (7). The latter research led to the hypothesis that influences on fetal development could program adult disease (8).
The fetal or developmental origins hypothesis is based on the concept that biological programming arises when the trajectory of cellular or organ growth is altered to compensate for an explicit insult occurring at vulnerable stages of embryonic development (9). The consequential adjustment, which may be adaptive in the short term, results in adverse metabolic or structural alterations that lead to later disease (10). Several studies have determined that individuals who go on to develop ischemic heart disease, diabetes, or insulin resistance show different patterns of weight gain from those of their peers (11). It has also been proposed that accelerated growth after birth, especially in the first 2 wk of life, may increase insulin resistance–related risk processes (12, 13). Alternatively, genetic factors may determine both the rate of fetal growth and the risk of chronic disease in later life (14). We examined whether different patterns of weight gain in the first 5 y of life predict a range of adult measures of adiposity and to what degree, if any, these can be explained by parental characteristics.
SUBJECTS AND METHODS
The Barry Caerphilly Growth Study is a follow-up of a dietary intervention randomized controlled trial. The original trial participants were pregnant women and their offspring, who were followed until the offspring were 5 y of age. In the late 1990s, when the offspring were 25 y of age, they were contacted and invited to participate in a follow-up study.
Original study
The original study was undertaken from 1972 to 1974 in 2 small towns in southern Wales: Barry, a seaside town, and Caerphilly, a largely industrial town. Pregnant women and their offspring were recruited through primary care practices and were randomly assigned to the "supplemented" or control group. Milk tokens were provided to women in the supplemented group throughout pregnancy and to each child (index case) until the age of 5 y (15).
A total of 1288 women were eligible for the study; 37 (2.9%) declined to participate, which left 1251 subjects. Of these, 88 women were subsequently excluded from the study if they had a miscarriage, if they were not pregnant, or if they had moved away from the research area. A total of 1163 pregnant women were enrolled in the trial, and a total of 951 (88.8%) singleton children completed the trial at 5 y (18.2% loss). An analysis of participants who dropped out did not reveal evidence of any systematic bias in terms of birth weight or maternal characteristics (15).
The women were visited at the time of first reporting the pregnancy. At approximately the 20th and 36th weeks of pregnancy, data on anthropometric measures, adult health behaviors, and socioeconomic characteristics were gathered by using a health visitor–administered questionnaire. After birth, the infants were visited at 10 d, at 6 wk, and at 3, 6, 9, and 12 mo, and thereafter at 6-mo intervals, for a total of 14 home visits by their fifth birthday. Birth weights were obtained from the hospital records, and thereafter weight was measured with a portable beam balance. Trained study nurses, who were monitored throughout the study, made all measurements, other than birth weight.
Follow-up study
From 1997 to 1999, we attempted to contact all of the participants who had completed the original study. The participants were traced through their parents, by obtaining information from the National Health Service Central Register, or by checking the local voter registration lists. The participants were invited to take part in the follow-up study; if they agreed, they were sent a self-completion questionnaire, which contained questions on lifestyle variables. Female participants were asked about oral contraceptive use and the date of the last menstrual period.
Each participant gave written informed consent. Ethical approval for the study was given by the Bro Taf Health Authority Local Research Ethics Committee.
The participants were invited to a screening clinic in which standard anthropometric measures were recorded. Standing height was measured by using a stadiometer (Harpenden, Crymych, United Kingdom). The participants were weighed while wearing a light gown, and weights were recorded to the nearest kilogram on a scale that was calibrated monthly. Waist circumference was measured at the narrowest point between the costal line and the iliac crest. Hip circumference was measured around the bony markings of the greater trochanters. A plastic tape measure attached to a spring balance maintained at a constant tension of 600 g was used for both measures. These measures were used to derive the waist-to-hip ratio (WHR). Sagittal abdominal diameter (SAD) was measured by using an abdominal caliper (Holtain-Kahn, Crymych, United Kingdom) while the participant was resting in the supine position and during expiration. All measures were repeated twice. If there was difference of >0.5 unit, then up to 2 additional measures were recorded to reduce measurement error.
Statistical analysis
We used 2 different approaches to model the effects of prenatal and postnatal growth on adult adiposity. The first approach, which is the most commonly used, was simple linear regression in which the effects of early growth on later adiposity were examined by using z scores for the early growth variables as exposures. We initially examined the z score from each time point alone. Because of the multitude of measurements, we reduced the data by including only the measures at birth, 3 mo, 1 y, and 5 y. These time points were chosen because we were specifically interested in testing the accelerated early growth hypothesis (13) and in comparing our findings with the original Hertfordshire cohort, who had data on birth weight and weight at 1 y (16). We later altered the third time point to 1.5 y on the basis of the spline regression models (see below). For birth weight, the z scores were derived for all births that took place between 36 and 44 wk of gestation, and they were standardized by sex and the gestational age (in weeks). For the other growth measures, the z scores were standardized by sex and the age at the time of the measurement.
The second approach was more complex, but it used all 14 childhood measurements in developing a linear spline random-effects model with 2 knots (thus dividing the follow-up into 3 time periods, each with its own gradient). Spline models with knots positioned at different time points were compared with a third-order fractional polynomial model. The positioning of the 2 knots was chosen by selecting the spline model with the highest percentage of predicted values within 5% of those of the fractional polynomial model. With the use of the same procedure, 3-knot spline models were investigated. However, because the fit of the 3-knot model was comparable with that of the 2-knot model, the simpler 2-knot model was chosen. The spline model was adjusted for sex, and the interaction terms between sex and each of the 3 time periods were considered. The random-effects model allowed the 4 coefficients, namely, the birth weight and the slopes for each of the 3 time periods, to vary between subjects. In addition, the model allowed variation in measurement between occasions and within subjects, thereby capturing the change in the variance of measurements with age. The model was estimated by using Markov Chain Monte Carlo methods with diffuse priors, which can be used to approximate maximum likelihood estimation.
The 4 between-subject random effects from the spline model were thus a summary of an individual's growth curve from birth to 5 y, denoting the deviance from the average predicted birth weight and that from the average predicted growth rate (kg/y) in each of the 3 time periods. These were defined as "immediate weight velocity" (from birth to 5 mo), "infant weight velocity" (from 5 mo to 1 y and 9 mo), and "childhood weight velocity" (from 1 y and 9 mo to 5 y). Again, we converted these variables into z scores so that the sizes of the coefficients were directly comparable. The analyses of the adult outcome measures were undertaken by using linear regression with the 4 random effects (the deviance from the average predicted birth weight and that from the average predicted growth rate in each of the 3 time periods) as exposures.
The 4 exposures [the deviance from the average predicted birth weight and the deviance from the average predicted growth rate (kg/y) in each of the 3 time periods] may have been estimated more accurately in the subjects with a greater number of childhood weight measurements than in those with fewer childhood weight measures. The accuracy of the estimates was examined by carrying out weighted linear regression models of the adult outcomes on all exposures (as described above), weighting each subject by a measure of the precision with which the random effects were estimated. The average of the 4 SDs of these random effects was used as a measure of precision.
We examined the effects of early growth by using 5 different models (Figure 1). Model 1 was adjusted for adult age, sex, and gestational age. Model 2 was adjusted as was model 1, but with the addition of maternal and paternal weight and height, because these may determine prenatal and postnatal growth and later adiposity. Model 3 was adjusted as was model 2, but with the addition of parental socioeconomic status in childhood, because this is a marker of the parental socioeconomic status and may determine parental obesity, childhood growth, and later adiposity. Model 4 was adjusted as was model 3, but with the addition of maternal smoking in pregnancy, because this may be determined by parental socioeconomic status and will affect birth weight and hence subsequent growth. Model 5 was adjusted as was model 4, but with the addition of current adult smoking status of the participant, because this may be related to both maternal smoking and parental socioeconomic status.
FIGURE 1.. Potential pathways that may explain the associations between weight gain in early life and adult obesity.
RESULTS
Of the 951 subjects who completed the original study, 23 were untraceable or had died or emigrated; of the remaining 928 subjects, 679 (73%) agreed to attend a follow-up clinic. Male infants were heavier (3.44 compared with 3.30 kg; difference in means = 0.16; 95% CI: –0.08, –0.23; P value < 0.0001) than were female infants (Table 1). In adulthood, the men had greater central adiposity (as measured by larger waist circumferences, WHR, and SAD; P < 0.0001 for all) than did the women.
View this table:
TABLE 1. Basic descriptive data on anthropometric measures from the participants of the Barry Caerphilly Growth Study by sex1
Every weight measure except birth weight showed a statistically significant positive association with adult body mass index (BMI; in kg/m2), which increased in magnitude with time (Table 2). Similar patterns of association with BMI were seen for SAD and waist circumference (Table 3) but not WHR (Table 2), for which meaningful associations emerged only with childhood measures from 1.5 y onward, and the strength of the association was relatively consistent for measures from 2.5 y onward.
View this table:
TABLE 2. Associations of z-scored anthropometric measures between birth and age 5 y with adult BMI, waist circumference, sagittal abdominal diameter (SAD), and waist-to-hip ratio (WHR)1
View this table:
TABLE 3. . Associations of birth weight and weight in infancy and childhood with adult BMI and waist-to-hip ratio (WHR)1
The results of the simple linear regression models with BMI as the adult measure of adiposity are shown in Table 3. Weight at 5 y was the only growth measure that was a strong and consistent predictor of adult BMI in all the models. The same pattern of results (data not shown) was seen with SAD and waist circumference, although the initial inverse birth weight association was stronger for SAD (–3.1 mm; 95% CI: –5.8, –0.4; P = 0.02). Different patterns were seen for WHR. The simple model shows an inverse association with birth weight and a positive association with weight at 5 y. With the addition of parental height and weight, the effect of weight at 5 y was greatly attenuated. In the final model, weight at 5 y was no longer a predictor. In contrast, weight at 1.5 y was not predictive in model 1, but after further adjustment, its effects became stronger with some weak evidence against the null hypothesis (P between 0.04 and 0.07).
The spline model in the present study found that, on average, boys gained 1.0 kg/mo and girls gained 0.9 kg/mo in the immediate postnatal period. Infant weight gain was similar in boys and girls at 0.28 kg/mo, whereas childhood weight gain was more modest at 0.17 kg/mo. Standardized (z score) immediate weight gain (0.69; 95% CI: 0.31, 1.07), infant weight gain (0.79; 95% CI 0.41 to 1.16), and childhood weight gain (1.37; 95% CI 1.00 to 1.74) were univariably associated (P < 0.0001 for all) with adult BMI. When we mutually adjusted for birth weight and immediate, infant, and childhood weight velocity, we found no evidence of an association between adult BMI and birth weight or infant weight velocity (Table 4). Immediate and childhood weight velocity showed positive associations with adult BMI, although the latter effect was stronger. Further adjustment did not change the basic pattern, although the influence of infant growth was strengthened. Increased childhood weight velocity remained the strongest predictor. These results were similar for waist circumference, although the effects of infant velocity were slightly but significantly stronger (1.08; 95% CI: 0.19, 1.97; P = 0.02 in model 5). For abdominal circumference, only childhood weight velocity was a meaningful predictor (4.22; 95% CI; 1.43, 7.01; P = 0.003 in model 5), whereas WHR again showed a different pattern, so that only infant weight velocity was a predictor (see Table 4). The results were essentially unchanged after we undertook a weighted regression analysis whereby we used the inverse of the SD for the 4 random effects as the weighting factors.
View this table:
TABLE 4. Associations of birth weight and immediate, infant, and childhood weight velocities with adult BMI and waist-to-hip ratio (WHR)1
DISCUSSION
We showed in this cohort that the most important predictor of BMI, waist circumference, and SAD was either weight at 5 y or increased weight velocity in childhood from 1 y and 9 mo to 5 y; both were conditional and unconditional of earlier growth and were not attenuated after adjustment for other confounders. These 2 measures were not predictive of WHR, for which, if anything, weight at 1.5 y and infant weight gain from 5 mo to 1 y and 9 mo were the strongest predictors. We did not find any association with immediate accelerated growth in the first few months of life.
This discordant pattern between different patterns of early growth and adult measures of adiposity was previously reported in another study. The Delhi birth cohort was a high-quality cohort study with detailed anthropometric data on men and women from birth until 21 y of age and a reexamination of the subjects between the ages of 26 and 32 y (17). Despite the ethnic differences between this cohort and the Barry Caerphilly Growth Study, both cohorts found that, in the first 5 y of life, the strongest associations with adult BMI were between 0 and 6 mo and between 2 and 5 y. The Delhi birth cohort reported the adjusted coefficients for BMI rather than for weight. Similarly, the strongest effect on adult WHR in the Delhi cohort was the 1-2-y growth period (see Figure 2 in Reference 17), which is analogous to the infant growth period in the present study. WHR is a composite measure that captures both central adiposity and pelvic skeletal development. If the association of early weight gain with skeletal growth differs from that with adiposity, then one would expect that the association for this measure would differ from those associations using a purer measure of central adiposity, such as waist circumference or sagittal abdominal diameter. This is indeed what was seen in the Delhi birth cohort, in which weight gain between 6 mo and 5 y was inversely associated with adult height.
The present study has several important strengths. First, we have 14 measures of growth between 0 and 5 y. This allowed us to differentiate patterns of growth by using more complex spline regression methods rather than using growth measures at arbitrary time points simply because those measures were all that were available. This was a more sensible approach because it may have also identified potential critical or sensitive periods for intervention, which were based on biological growth patterns defined before the examination of the association with adult obesity. This approach was also better able to manage when 1 or 2 data points were missing and thereby maximized statistical power.
We presented our results by using a more conventional analytic strategy to aid comparability with the existing literature. The 2 methods in some cases produced similar answers, such the importance of weight at 5 y and of childhood weight velocity for BMI. In other cases, the effects were seen only with the more complex method, such as the positive association between infant weight velocity and adult BMI.
We also had a wide range of covariates, including both maternal and paternal measures of adiposity, which may have reflected shared genetic determinants, and perinatal factors such as maternal smoking and socioeconomic conditions. In some cases, the adjustment for covariates attenuated the initial associations, eg, weight at 5 y and WHR, whereas we noted that the effects of infant weight gain on BMI and WHR were strengthened after adjustment.
We have no data on growth after 5 y of age or on the timing of the pubertal growth spurt (18), both of which may determine later adiposity. Growth during these periods may itself be determined by earlier growth (19) or may have an independent effect on later adiposity. Although we had some specific measures of central adiposity, we did not undertake body scans of the subjects, so we were limited in our ability to differentiate tissue composition (ie, lean and fat tissue).
Two recent systematic reviews examined the association between anthropometric measures in early life and later obesity (3, 20). Those studies concluded that large infants or infants who grew rapidly in infancy were more likely to be obese in later life. However, in most of the articles that were reviewed, obesity was measured in childhood or adolescence. In addition, few studies had detailed repeat measures on growth at different time periods. For example, Baird et al (3) identified only 10 such studies, of which only 2 studies (21, 22) had measures for subjects >18 y of age. The review by Baird et al did not include the Delhi birth cohort (17). We did not find a strong association between birth weight and any of the 4 measures of adiposity. In most studies, the association was weakly positive. For example, in a 1958 birth cohort, a 0.5-kg increase in birth weight, 1 SD, was positively associated with a BMI increase of 0.4 (23), which is consistent with the results of the present study.
The simple univariable analysis in the present study showed that standardized weight at later time points had stronger associations with all measures of later obesity except WHR. This finding is consistent with the data from the Helsinki cohort, which showed that children who became obese in adulthood had greater weight z scores at all ages that became progressively larger and more divergent with age (24). This association was also seen in the Delhi birth cohort (17).
The most detailed study of infant growth comes from a cohort of children with 7 measures of infant growth and adult obesity measured by BMI at around age 25 y (22). That study noted that weight gain in the first 8 d predicted later adiposity independent of later weight gain in the first 4 mo. Although we could not specifically examine the first week of life, the spline regression did not suggest that this was a critical period of growth, but rather that growth up to 5 mo showed a different pattern than did subsequent growth. We also found that increased weight gain in the first 5 mo predicts later BMI, which is consistent with a study of 300 African Americans (21). However, unlike both of these studies, we were able to adjust for growth in childhood and other potential confounders, such as parental adiposity. These more complex models attenuated the effects of increased weight gain in the first 5 mo. The best determinant of adult obesity was weight at 5 y or increased weight gain from 1 y and 9 mo to 5 y. This finding is similar to a study from Brazil, which noted that rapid weight gain between 0 and 20 mo and between 20 and 43 mo was associated with obesity, although the mutually adjusted odds ratios were not presented (25). In addition, a study from Guatemala found similar results, with the strongest associations being seen in later childhood (3–7 y), followed by infancy (0–1 y), and then birth weight (26). Similarly, the Delhi birth cohort (17) noted that the largest coefficients for adult BMI were for the periods between 0 and 6 mo and between 2 and 5 y. It is possible that changes in weight in the initial postnatal period may have greater importance for later obesity in populations in the developing world than in those in the developed world because of the former's greater adversity during the intrauterine period.
In conclusion, the data in the present study indicated that adult obesity is more strongly related to weight gain from the age of 1.5 to 5 y. We did not find that birth weight or rapid weight gain in the immediate postnatal period was as important a predictor once weight gain in childhood was taken into account. These conclusions may be sensitive to different population settings. Too few studies have detailed growth measures across the whole life course to allow more complex patterns to be examined. The results of the present study highlight the importance of preventing childhood obesity, although it remains unclear whether early intervention, eg, in the preschool period, is more, less, or equally cost-effective than is waiting and intervening at a later period.
ACKNOWLEDGMENTS
We are grateful to the subjects who participated in the original survey and who were willing to continue to be followed in early adulthood. We thank the Bro Taf Health Authority for help with contacting the subjects. P Elwood undertook the original study within the MRC Epidemiology Unit (South Wales). The original study was funded by the UK Department of Health.
The authors' responsibilities were as follows—DD: undertook the original randomized controlled trial; AM, DD, GDS, and YB-S: designed the follow-up study; AM: conducted the fieldwork under the supervision of YB-S and GDS; AM, YB-S, RH, and KT: undertook the statistical analyses; AM, RH, and YB-S: drafted the manuscript; and all authors: commented on and finalized the manuscript. None of the authors had any conflict of interest.
REFERENCES