点击显示 收起
1 From the Department of Social Medicine (ISR, ARN, CDS, PME, and GDS), University of Bristol, Bristol, United Kingdom; the MRC Childhood Nutrition Research Centre, Institute of Child Health, London, United Kingdom (JCKW); the Division of Developmental Medicine, University of Glasgow, Yorkhill Hospitals, Glasgow, United Kingdom (JRR); and the Department of Clinical Sciences at South Bristol, University of Bristol, Bristol Royal Infirmary, Bristol, United Kingdom (JT)
2 Supported by a grant from the Wellcome Trust. The UK Medical Research Council, the Wellcome Trust, and the University of Bristol provide core support for the Avon Longitudinal Study of Parents and Children (ALSPAC). 3 Address reprint requests to A Ness, Department of Social Medicine, University of Bristol, 24 Tyndall Avenue, Bristol, BS8 1TQ United Kingdom. E-mail: andy.ness{at}bris.ac.uk.
ABSTRACT
Background: Birth weight has been positively associated with risk of overweight in later life. However, little information exists on how weight and length at birth are associated with subsequent lean and total body fat.
Objective: We investigated the association between weight and length at birth and body composition and fat distribution in childhood.
Design: Body composition was measured by using dual-energy X-ray absorptiometry in 9–10-y-old subjects (n = 3006 boys and 3080 girls). Weight and length at birth were measured or taken from hospital records.
Results: Birth weight was positively associated with both lean body mass (LBM) and total body fat at 9–10 y of age in both sexes. LBM rose by 320 g per 1-SD increase in birth weight (P < 0.001), and total body fat rose by 2.5% (P = 0.001), but birth weight was unassociated with the fat-to-lean mass ratio (FLR). Ponderal index (PI) at birth (ie, weight/length3) was positively associated with LBM, total body fat, and the FLR in both sexes; the FLR increased by 2.7% in boys (P = 0.021) and by 5.0% in girls per 1-SD increase in PI (P < 0.001). Weight and length at birth did not predict central adiposity; although trunk fat had a strong positive association with PI at birth, this association disappeared after adjustment for total body fat.
Conclusions: Higher PI at birth is associated with both higher fat and lean mass in childhood but also with an increase in the FLR. PI at birth is a better predictor of subsequent adiposity than is birth weight.
Key Words: Birth weight • ponderal index • programming • body composition • fat distribution • risk of overweight
INTRODUCTION
In the past 30 y, the worldwide prevalence of obesity has risen rapidly (1). The precise reasons for this epidemic remain unclear, but the prenatal period has been suggested to be a "critical period" for obesity development (2), with the conditions experienced in utero having a lifelong influence on the propensity to develop obesity. The most commonly used indicator of prenatal exposures is birth weight, and several large studies have reported a positive association between birth weight and subsequent body mass index (BMI) and risk of obesity or overweight (3). This association has been found in adults and children in a variety of ethnic groups and is linear in some studies and j-shaped or u-shaped in others (3).
The relation between birth weight and subsequent obesity is also of interest because of the well-documented associations between birth weight and chronic disease risk. Higher birth weight is negatively associated with the incidence of several chronic diseases in adulthood, including coronary heart disease, coronary risk factors (4-6), and type 2 diabetes (7-9). This appears paradoxical, because risk of these diseases increases with obesity (10-12). However, most studies associating birth weight with obesity defined obesity by using BMI, a measure of relative weight that does not necessarily reflect body composition. Birth weight–BMI associations could be mediated by increases in total or central adiposity but also, theoretically, by increases in lean body mass (LBM).
The association between birth weight and the occurrence of some later disorders is dependent on adjustment for adult BMI. If birth weight were positively associated with LBM, and this explained the birth weight–BMI association, then at any given BMI, the LBM would be directly proportional to birth weight, whereas percentage body fat would be inversely proportional. This would be consistent with the high risk of coronary heart disease and diabetes observed among those who were small at birth but who became overweight as adults—ie, the association with birth weight would be mediated by increased adiposity. Several studies have now been conducted on the relation between birth weight and later lean and total body fat. These have consistently shown positive associations between LBM and birth weight (13-17), whereas associations with adiposity were more variable, but generally used relatively inaccurate methods of assessing body composition, such as skinfold-thickness measurements. Only 3 studies, 2 in elderly subjects and 1 in children, directly measured lean and total body fat by using dual-energy X-ray absorptiometry (DXA) (13, 18, 19). These studies were small and had limited ability to look at confounding factors and possible sex differences.
A central distribution of body fat is associated with disease risk (20, 21), independently of obesity. If increased birth weight were associated with a less central pattern of obesity, this could also explain the birth weight–disease associations. Limited evidence suggests that after adjustment for BMI, birth weight is associated with a less central pattern of obesity as measured by the subscapular-to-triceps skinfold ratio (22, 23) but not with waist-to-hip ratio (22). However, studies with accurate measures of fat distribution, such as by DXA, are lacking.
The aim of the current study was to investigate the relation between birth weight and lean and total body fat and fat distribution measured by DXA at 9 y in a group of >6000 children. In addition, we investigated how length and ponderal index at birth were associated with later body composition.
SUBJECTS AND METHODS
Subjects
These analyses used data from the Avon Longitudinal Study of Parents and Children (ALSPAC), a geographically based cohort study that is described in detail elsewhere (24). All pregnant women resident within a geographically defined area of southwest England with an expected date of delivery between April 1991 and December 1992 were eligible, of whom 14541 pregnancies were enrolled (25). Ethical approval for the study was obtained from the ALSPAC law and ethics committee and the 3 local research ethics committees covering the area. Data in ALSPAC are collected through self-completion postal questionnaires, by abstraction from and linkage to medical records, and from examination of the children at research clinics.
Measurement of size at birth
Birth length was measured by ALSPAC staff shortly after birth (mean age at measurement: 1.9 d; range: 0–7 d), birth weight was either measured by ALSPAC staff (67.5%) or extracted from hospital records (32.5%), and the sex of the child and gestational age were extracted from hospital records. Duration of gestation was assessed on the basis of the date of the last menstrual period. If there was a discrepancy by ultrasound assessment or other clinical indicators of 2 wk, the clinical records were reviewed and a best estimate was made by an experienced obstetrician. Ponderal index (PI) was calculated from birth weight (kg)/birth length (m)3 and was used as a measure of relative weight-for-length at birth.
Measurement of anthropometric and body-composition variables
Measurements of anthropometric and body-composition variables were taken at the "Focus @ 9" research clinic, to which all ALSPAC children were invited, which was held between January 2001 and January 2003. Of the 7725 children who attended the Focus @ 9 clinic, 7470 agreed to undergo a whole-body DXA scan, which was performed by using a Prodigy scanner (Lunar Radiation Corp, Madison, WI), at which time weight and height were also measured. Height was measured to the last complete mm by using a Harpenden stadiometer (Holtain Ltd, Crosswell, United Kingdom). Weight was measured by using a body fat analyzer (model TBF 305; Tanita UK Limited, Viewsley, United Kingdom). Body mass index was calculated from weight (kg)/height (m)2. Waist and hip circumferences were measured to the nearest mm by using Harpenden anthropometric tape (Holtain Ltd). Waist circumference was measured to the nearest mm at the minimum circumference of the abdomen between the iliac crests and the lowest ribs. Hip circumference was measured at the point of maximum circumference around the child's hips and buttocks.
The DXA scans were evaluated and reanalyzed as necessary to ensure that borders between adjacent subregions were optimally placed. After the exclusion of scans with anomalies such as movement artifacts, complete scans were available for 7336 children. The results of the DXA scan were used to derive total LBM, total body fat, and truncal fat. Truncal fat was defined as fat in the ribs, spine, and pelvis subregions. The CVs for LBM, total body fat, and truncal fat were 1.1%, 2.3%, and 6.2%, respectively. By regressing log LBM on the log of total body fat, the ratio total body fat/(total body lean mass)n was calculated, where n gave the least correlation between the ratio and lean mass. The value of n used was 2.20 for girls and 2.59 for boys, because the relation between fat and lean mass differed between boys and girls (P for interaction < 0.001). This fat-to-lean ratio (FLR) was used as a measure of relative adiposity.
Confounding variables
The following confounding variables were considered: maternal smoking in pregnancy, maternal education, paternal social class, housing tenure, maternal age, parity, and stage of puberty. These were selected on the basis of evidence that they are associated both with birth weight and with subsequent body composition or obesity (26-29). Information on maternal smoking in pregnancy (at 32 wk of gestation), highest maternal educational qualification, paternal occupational social class, housing tenure, and parity was obtained by questionnaire. Parity was measured as the mother's number of previous pregnancies that resulted in either a live birth or a stillbirth. Maternal age at delivery was derived by subtracting the mother's date of birth from the child's date of birth. These variables were classified as follows: smoking (0, 1–9, 10–19, or 20 cigarettes/d), education [certificate of secondary education (CSE) or no qualifications, vocational qualifications, O level or equivalent, A level or equivalent, or degree], paternal social class (I, II, III nonmanual, III manual, IV, V, or Armed Forces), parity (0, 1, 2, or more), maternal age (19, 20–29, 30–39, or 40), housing tenure [owned or mortgaged, council-rented (ie, government housing), or other rented]. (CSE and O level were, respectively, lower and higher levels of qualifications taken at 16 y of age. A levels were the standard academic qualifications taken at 18 y of age.) Stage of puberty was assessed by a questionnaire sent out when the children were 9 y old.
Statistical analyses
These analyses were restricted to white children from singleton births. There is evidence that both birth weight (30, 31) and body composition (32, 33) differ according to ethnic group, and there were too few children from nonwhite ethnic groups (4% of the cohort) to analyze separately.
All birth size variables (eg, birth weight, ponderal index, and birth length) were converted to sex-specific z scores (internally referenced). The birth size variables were examined in relation to the following outcomes: LBM, total body fat, FLR, and trunk fat mass. Total body fat, FLR, and trunk fat mass were transformed to the natural logarithm to reduce skewness in the distribution. The regression models used were adjusted for the following confounders: 1) Model 1 was adjusted for sex, duration of gestation, current age, and measures of the child's current body size, ie, height and height2 (height2 was used in addition to height because there was evidence of nonlinearity in the relation between height and all outcomes investigated). 2) Model 2 was adjusted as was model 1 plus measures of social position, ie, maternal education, paternal social class, and housing tenure. 3) Model 3 was adjusted as was model 2 plus maternal factors, ie, maternal age at delivery, maternal parity, and smoking in late pregnancy. 4) Model 4 was adjusted as was model 3 plus stage of puberty. (This model was used in girls only, because nearly all the boys were prepubertal. Information on stage of puberty was available for 1862 girls with DXA measurements and information on birth weight; 36.5% of girls were in Tanner stage 2 or above). 5) In analyses with trunk fat as the outcome, we used an extra model, ie, model 5 = model 3 + total body fat, to investigate whether any variation in total fat associated with birth dimensions was independent of variation in total body fat. To allow comparison with the results of other studies, we also examined the associations between size at birth and BMI and WHR after adjustment for gestation and age, and between size at birth and WHR after further adjustment for BMI. BMI was transformed to the natural logarithm to reduce skewness in the distribution. The interaction between sex and each birth measure in model 1 was also tested, and if evidence existed that the strength of the association between birth size and DXA outcome differed between the sexes, then the results for boys and girls are presented separately.
Regression analyses including both the linear and the quadratic term for each birth size measure (adjusted for the factors in model 1) were also performed to look for evidence of nonlinearity in any of the associations. In the figures, we fitted a line across the deciles using the "2 way qfit" command in STATA. This calculates the prediction of a y variable on the basis of a linear regression of the y variable on the x variable and the square of the x variable and plots the resulting curve. All statistical analyses were performed by using SPSS version 12.0 (SPSS Inc, Chicago, IL) and STATA version 8.0 (Statacorp LP, College Station, TX).
RESULTS
Characteristics of the study children
Of the 7336 children with DXA measurements of body composition, 3006 boys and 3080 girls were white singleton births for whom data on birth weight were available (94% of the 6470 white singleton subjects attending the Focus @ 9 clinic). Descriptive statistics on the characteristics of these children are summarized in Table 1. Compared with the whole ALSPAC cohort, the girls included in the analyses had slightly higher birth weights and birth lengths: mean (±SD) birth weights were 3.40 ± 0.48 and 3.37 ± 0.51 kg for those included and not included, respectively (P = 0.009), and the figures for birth length were 50.3 ± 1.9 and 50.1 ± 2.1 cm (P = 0.005). There was no significant difference in PI at birth in girls or in any measure of size at birth in boys. Compared with the whole cohort, the children included in the analyses were more likely to live in owned or mortgaged homes (86.7% compared with 71.3%; P < 0.001), were less likely to have a mother whose highest qualification was CSE or less (12.6% compared with 26.3%; P < 0.001), were more likely to have a father in social class I or II (49.2% compared with 39.6%; P < 0.001), were less likely to have a mother who smoked during pregnancy (13.4% compared with 26.1%; P < 0.001), and were more likely to be a first child (46.0% compared with 42.7%; P < 0.001). However, among those children attending the Focus @ 9 clinic, there was no significant difference in height, weight, or BMI between those children who were and those were not included in the analyses.
View this table:
TABLE 1. Characteristics of the children included in the analyses
Size at birth and lean body mass, total body fat, and the fat-to-lean ratio
Birth weight was positively associated with LBM among both boys and girls (Table 2). This association was equivalent to a 320–390-g increase in LBM per 1-SD increase in birth weight. The association between birth weight and LBM did not differ between the sexes. PI at birth was also strongly positively associated with LBM in both sexes, and again this association persisted after adjustment for potential confounders. The strength of the association between PI and LBM was stronger in girls than in boys, being equivalent to a 345–377-g increase in LBM per 1-SD increase in PI in boys and a 274–288-g increase in girls.
View this table:
TABLE 2. Associations of size at birth with lean body mass, total body fat, and the fat-to-lean mass ratio at 9–10 y of age1
Birth length was positively associated with subsequent LBM in boys. This association was diminished after adjustment for potential confounders. Among girls, on the other hand, there was a weak negative association between birth length and LBM, although this association became positive after adjustment for stage of puberty. There was strong evidence that the birth length–LBM association differed between the sexes (Table 2).
Both birth weight and PI were positively associated with total body fat. The association with birth weight was equivalent to a 2–3% increase in total body fat per 1-SD increase in birth weight, and that with PI was an 7% increase per 1-SD increase in PI. There was no evidence that the associations between birth weight or ponderal index and total body fat differed between the sexes. Birth length was negatively associated with total body fat. The association was stronger in girls than boys (Table 2), with total body fat decreasing by 3% for every 1-SD increase in birth length in boys and by 6% in girls.
There was no association between the FLR and birth weight. PI was positively associated with the FLR in both sexes (with increases of 2–3% and 5–6% per 1-SD increase in PI in boys and girls, respectively), and was largely unchanged after adjustment for potential confounders. Birth length was negatively associated with the FLR in both sexes. These associations were largely unaffected by adjustment for potential confounders (except stage of puberty in girls, which diminished the association), and were stronger in girls, being equivalent to an 6% decrease in total body fat or FLR per 1-SD increase in birth length.
Size at birth and trunk fat mass
There was a weak positive association between birth weight and trunk fat mass, which was strengthened after adjustment for confounders (models 3 and 4) but abolished after control for total body fat (Table 3). PI at birth was strongly positively associated with trunk fat mass in both sexes, equivalent to an 7–8% increase per 1-SD increase in PI. This association was also abolished after control for total body fat. There was no evidence that the strength of the associations between birth weight or PI and trunk fat differed between the sexes. Birth length was negatively associated with trunk fat mass in both sexes, with these associations being stronger in girls than in boys (Table 3), equivalent to an 4% fall in trunk fat per 1-SD increase in birth length in boys, and an 8% fall in girls. However, in both sexes, the negative association with birth length disappeared after control for total fat.
View this table:
TABLE 3. Associations of size at birth with trunk fat at 9–10 y of age1
A previous study in adolescent girls reported that the negative association between birth weight and central adiposity measured by the subscapular-to-triceps skinfold ratio was stronger among fatter children (ie, those with a BMI > 25) (34). To look for a similar effect, we repeated the trunk fat analyses investigating interactions with birth size in those children in the top 20% for total body fat. There was evidence of differences in the relation with birth size between fatter and less-fat children (birth weight, P < 0.001; PI, P = 0.007; birth length, P = 0.007; model 5). Among this group of fatter children, birth size was inversely associated with trunk fat. The association between birth weight and trunk fat was present in girls only (P for interaction 0.016) and persisted after adjustment for total body fat and stage of puberty. Among girls, the association was equivalent to an 2.2% (95% CI: –3.0%, –1.3%) fall in trunk fat per 1-SD increase in birth weight (P < 0.001; model 5). There was no evidence that the association with PI or birth length varied between boys and girls. There was a 1.8% (95% CI: –2.5%, –1.1%) and 0.9% (95% CI: –1.83%, –0.04%) decrease in trunk fat per 1-SD increase in PI at birth and in birth length, respectively, in both sexes combined.
Relation of size at birth with BMI and WHR at 9–10 y of age
BMI was strongly positively associated with birth weight and PI at birth, with BMI increasing by 2.7% per 1-SD increase in birth weight (Table 4). The association between birth length and BMI differed between the sexes: there was a positive association with BMI at 9-10 y in boys but not in girls.
View this table:
TABLE 4. Association of size at birth with BMI and waist-to-hip ratio (WHR) at 9–10 of age1
Both birth weight and birth length were negatively associated with WHR after adjustment for current BMI. Ponderal index at birth was positively associated with WHR in analyses adjusted for gestation and age, but this association was lost after adjustment for current BMI.
Evidence of nonlinearity
There was no evidence of nonlinearity for any of the associations between PI or birth length and any outcome. There was also no evidence that any of the birth weight–body composition associations in boys were nonlinear. However, the birth weight2 term was significantly associated with each of the outcomes investigated in girls (P = 0.022, P = 0.037, P = 0.038 for LBM, total body fat, and trunk fat, respectively). Figure 1 shows adjusted LBM, total body fat, and trunk fat in girls according to decile of birth weight for gestational age. These appear to show a j-shaped relation, with an excess of total and truncal fat in the girls with the lowest birth weights.
FIGURE 1.. Lean body mass, total body fat, and trunk fat in girls by decile of birth weight for gestational age. Lean body mass, total body fat, and trunk fat were adjusted for gestational age, age, height, height2, maternal education, paternal social class, housing tenure, parity, maternal age, and maternal smoking during pregnancy. The mean value of birth weight in each decile is shown on the x axis. Error bars are 95% CIs. The line was fitted by using mean birth weight and mean birth weight squared for each decile. n = 2717.
DISCUSSION
In the present study, we showed that after adjustment for current height, both LBM and total body fat at 9–10 y increased with increasing weight or PI at birth. A greater PI at birth was also associated with becoming a relatively fatter child among both sexes. Associations with birth length were less consistent and tended to operate in a direction opposite of those with birth weight and PI, ie, longer babies tended to become children with reduced total body fat and a lower FLR. Birth weight was negatively associated with WHR after adjustment for current BMI. Using DXA measures of central obesity, however, we found a positive association between size at birth and truncal fat mass, although we found little evidence that this association was independent of changes in total body fat.
This is one of very few studies, and 1 of only 2 epidemiologic studies in children, to have DXA measurements of body composition and body fat distribution. Although DXA has high precision, accuracy is not always ideal (35). However, most epidemiologic studies of relations between birth weight and adiposity have depended on more crude proxies for these variables such as BMI. Our study is by far the largest DXA study and included detailed background information on social and maternal factors. It is unusual in having measures of length as well as weight at birth. It has been suggested that anthropometric disproportionality at birth (eg, being long and thin) increases the risk of later cardiovascular disease, whereas proportionate smallness does not (36, 37). So far, however, few data have been available to test this hypothesis. If the risk of cardiovascular disease is mediated by unfavorable body composition, then our results would suggest the opposite. Birth length was inversely and PI was directly associated with the FLR; thus, long, thin babies would be expected to have a reduced FLR compared with proportional babies.
Birth weight per se was only weakly associated with the FLR in this contemporary population, and so our results do not suggest that birth weight is likely to be strongly associated with the subsequent risk of chronic disease if this association is mediated via changes in body composition. Our results imply that the associations between birth weight and disease will vary according to how much of the variation in birth weight results from variation in birth length and how much from relative weight-for-length at birth. When most of the variation results from changes in birth length, heavier (and longer) babies would be expected to have a reduced FLR and a lower risk of chronic disease in later life. When most of the variation results from variation in PI, the heavier babies would be expected to have a higher FLR and a higher subsequent risk of chronic disease. It seems likely that the relative proportion of variation in birth weight accounted for by length and PI would vary over time and between populations according to nutritional and other environmental factors. This may mean that the associations between birth weight and subsequent body composition will differ between historical and contemporary cohorts.
Children have had a limited time compared with adults for lifestyle factors to affect body composition and thus may be particularly suitable for the study of programming effects. On the other hand, children have also had relatively little time for programming effects to be amplified, which may make such effects more difficult to detect [although, if anything, it appears that birth weight–BMI associations diminish with age (3)]. One possible problem with this age group is that some of the children will be entering puberty, which has a marked effect on body composition and fat distribution. Indeed, adjustment for stage of puberty had a major effect on many of the associations observed in girls. Only self-reported puberty data were available, which affected our ability to control for this effectively [although self-reported Tanner stage in girls was strongly associated with growth and bone density, which suggests that it was reasonably valid (38)]. Our study cannot cast any light on the reasons for the observed associations, ie, do these reflect programming effects on body composition and fat distribution or shared genetic influences on size at birth and body constitution in later life? Measurements of parental body composition would be helpful in this respect, but unfortunately were unavailable. This study is also unable to investigate possible ethnic differences in the associations between birth size and later body composition because of the very small number of nonwhite children in the cohort.
Our results are consistent with a growing number of studies showing a positive association between size at birth and LBM in later life. This has now been shown in 3 studies with DXA measures of body composition, 1 in children aged 7 y (19) and 2 in elderly adults (13, 18), and in 1 study of adult Pima Indians in whom LBM was estimated by using a combination of DXA and underwater weighing (39). Several studies in which LBM was estimated by anthropometric methods or bioimpedance have also shown positive associations with birth size in young and elderly subjects from both the developed (14, 15, 40) and the developing (41, 42) world.
The literature relating birth size to subsequent adiposity is less consistent. Of the other 3 studies with DXA measurements of body composition, 2 found a negative association between birth size and adiposity after adjustment for current weight or height and weight (13), and 1 found no association (19). In a large study of 5–11-y-old children in the United States, birth weight was negatively associated with the sum of skinfold thicknesses after control for current BMI (43). Studies in Brazilian (41) and Italian (44) children, however, found no association between birth size and adiposity. Similarly, in a recent study of English men in their sixties, body weight was positively correlated with total body fat (r = 0.10) and fat-free mass (r = 0.27), but was not correlated with percentage body fat (r = 0.02) (40) We are aware of 3 studies other than our own that found positive associations between birth weight and subsequent adiposity. Among young Guatemalan women (45) after control for a range of confounders, a 1-kg increase in birth weight was associated with a 1.27% increase in body fat (P = 0.03), and a 1-g/cm3 increase in PI at birth was associated with a 2.28% increase in body fat. Stettler et al (29) found that increased birth weight was associated with increased sum of skinfold thicknesses among young African Americans, and Stern et al (46) found a positive correlation (r = 0.16, P < 0.0002) between birth weight and percentage body fat among 16–44-y-old Mexican American subjects. It is possible that failure to adjust for parity accounts for some of the inconsistency in the reported relations between size at birth and adiposity; firstborn children are at increased risk of obesity (29) and are also smaller at birth, which may obscure a positive association between birth size and adiposity.
Two studies have related DXA measures of body composition in neonates to birth size and provide indirect support for our findings. One compared 47 appropriate-for-gestational-age (AGA) and large-for-gestational-age (LGA) neonates and found that LGA infants have both more fat and more lean mass but also a higher percentage of body weight as fat (47). The other compared small-for-gestational-age (SGA), AGA, and LGA neonates and found percentage body fat to be roughly twice as high in LGA as in AGA and SGA infants (22.5%, 11.9%, and 10.0%, respectively) (48).
In conclusion, we found that increased weight or PI at birth is associated with both increased lean mass and increased adiposity at 9 y of age, and that PI at birth is a better predictor of subsequent body composition than is birth weight. We conclude that the well-documented positive associations between birth weight and subsequent BMI result from increases in both lean mass and total body fat.
ACKNOWLEDGMENTS
We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting the families, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses.
This publication is the work of the authors, and ARN and ISR serve as guarantors for the contents of this article. ARN and ISR conceptualized the study. ISR conducted the analyses and wrote the initial draft of the paper. CDS provided statistical expertise, and JCKW advised on the analyses. JT raised the funds for collection of the DXA data and made a substantial contribution to its interpretation and analysis. All authors contributed to the writing of the manuscript. None of the authors had any conflict of interest.
REFERENCES