Association Between Bone Mass and Fractures in Children: A Prospective Cohort Study

Abstract

This is the first prospective cohort study of the association between bone mass and fracture risk in childhood. A total of 6213 children 9.9 years of age were followed for 24 months. Results showed an 89% increased risk of fracture per SD decrease in size-adjusted BMC.

Introduction

Although previous case-control studies have reported that fracture risk in childhood is inversely related to bone mass, this has not been confirmed in prospective studies. Additionally, it remains unclear which constituent(s) of bone mass underlie this association. We carried out a prospective cohort study to examine the relationship between DXA measures in children 9.9 years of age and risk of fracture over the following 2 years.

Materials and Methods

Total body DXA scan results obtained at 9.9 years of age were linked to reported fractures over the following 2 years in children from a large birth cohort in southwest England. DXA measures consisted of total body less head (TBLH) BMD, bone area, and BMC, and results of subregional analysis of the humerus. Analyses were adjusted for age, sex, ethnicity, and social position.

Results

Complete data were available on 6213 children. There was a weak inverse relationship between BMD at 9.9 years and subsequent fracture risk (OR per SD decrease = 1.12; 95% CI, 1.02–1.25). In analyses examining the relationship between fracture risk and volumetric BMD, fracture risk was inversely related to BMC adjusted for bone area, height, and weight (OR = 1.89; 95% CI, 1.18–3.04) and to estimated volumetric BMD of the humerus (OR = 1.29; 95% CI, 1.14–1.45). Fracture risk was unrelated to both TBLH and humeral bone area. However, in analyses of the relationship between fracture risk and bone size relative to body size, an inverse association was observed between fracture risk and TBLH area adjusted for height and weight (OR = 1.51; 95% CI, 1.17–1.95).

Conclusions

Fracture risk in childhood is related to volumetric BMD, reflecting an influence of determinants of volumetric BMD such as cortical thickness on skeletal fragility. Although bone size per se was not related to fracture risk, we found that children who fracture tend to have a smaller skeleton relative to their overall body size.

INTRODUCTION

FRACTURES IN CHILDREN are common and increasing. The reported incidence in the United Kingdom ranges from 1.6%(1) to 3.6%(2) per year, and the incidence seems to be increasing over time.(3) There is a peak in fracture risk around 14 years of age in boys and 11 years of age in girls.(4) Although previous studies into the public health impact of fractures have largely focused on elderly populations, fracture incidence during childhood is similar to that in the elderly. For example, the incidence of fractures of the ulna and radius in boys 5–15 years of age in the United Kingdom in 1991–1992 was 23/10,000 compared with 24/10,000 in men ≥85 years of age.(5) Fractures in children can affect health and development as a consequence of complications such as malalignment of the fractured bone, limb overgrowth,(6) and acute compartment syndrome.(7) Childhood fractures also result in time off school, activity-restricted days (14 and 26 days for arm and leg fractures, respectively),(​8) and long-term consequences arising from complications such as secondary osteoarthritis.(9)

Understanding the etiology of fractures in childhood may provide opportunities for developing population-based interventions aimed at reversing the recent increase in incidence. For example, it has been suggested that psychosocial factors leading to an increase in risk-taking behavior contribute to the higher risk of fractures in puberty.(10) There is also evidence that fractures in childhood are related to underlying skeletal fragility. We recently performed a systematic review and meta-analysis of the literature. The pooled standardized mean difference (SMD) for bone mass in children with and without fractures was −0.32 (95% CI, −0.43 to −0.21), suggesting that fracture risk is linked to low bone mass in children.(11) If fracture risk in childhood is related to underlying skeletal fragility, the rise in fracture incidence during puberty may reflect a transient deterioration in skeletal strength as a consequence of rapid skeletal growth, such as a lag in cortical thickness and/or mineralization relative to linear skeletal growth.(12)

Although childhood fractures have been found to be associated with a reduction in bone mass, previous studies addressing this relationship were case-control studies,(13–22) and to our knowledge, no previous prospective studies have examined this relationship. Case-control studies are more prone to bias and thus the effect size estimate may be unreliable. For example, selection of appropriate controls that are matched for potential confounding factors, but are otherwise representative of the general population, may be problematic. In addition, because case-control designs result in bone mass being measured after the fracture, it is possible that reductions in bone mass that were observed reflect transient reductions caused by inactivity as a result of the fracture.

Bone mass as reflected by DXA-derived measures of BMC is affected both by overall bone size and the amount of bone mineral per unit volume (i.e., volumetric BMD [vBMD]). vBMD is in turn influenced by cortical thickness, cortical porosity, trabecular bone volume, and the level of bone tissue mineralization. Biomechanical strength indices such as cross-sectional moment of inertia (CSMI) depend both on components related to bone size such as cortical width and those related to vBMD such as cortical thickness (see Appendix). Various methods have been developed for estimating vBMD from DXA measurements based on more complete adjustment of BMC for bone size,(23) suggesting that it may be possible to distinguish the contributions of bone size and vBMD to fracture risk using DXA-derived data. In addition, as well as the influence of bone size per se, we recently found that fracture risk in childhood is related to bone size relative to body size, as calculated by adjusting bone area for height and weight.(24)

The objective of this study was to prospectively study the association between bone mass as measured by DXA and fractures in children using a large population-based cohort in the United Kingdom. A secondary aim was to examine whether any such association predominantly reflects an influence of vBMD, bone size, or bone size relative to body size on fracture risk.

MATERIALS AND METHODS

Study population

The Avon Longitudinal Study of Parents and Children (ALSPAC) is a geographically based UK cohort that recruited pregnant women residing in Avon (southwest England) with an expected date of delivery between April 1, 1991 and December 31, 1992. A total of 14,541 pregnancies were initially enrolled, with 14,062 children born. This represented 80–90% of the eligible population (see www.alspac.bris.ac.uk for further details).(​25) Of these births, 13,988 were alive at 12 months. This study is based on total body DXA scans performed at a research clinic to which the whole cohort was invited at a mean age of 9.9 years. This clinic was attended by 7725 children. Ethical approval was obtained from the ALSPAC Law & Ethics Committee, local research ethics committees, and Central Office for Research Ethics Committees (COREC). Parental consent and child’s assent were obtained for all measurements made.

Measurement of height, weight, and DXA-derived parameters

At the research clinic, height was measured to the last complete millimeter using a Harpenden stadiometer. Weight was measured to the nearest 50 g using a Tanita Body Fat Analyzer (model TBF 305). Total body less head bone area (TBLH BA), total body less head BMC (TBLH BMC), and total body less head BMD (TBLH BMD) were measured using a Lunar Prodigy DXA in 7444 of the 7725 children. Total body DXA scans were not used, because the head is not responsive to environmental stimuli such as physical activity.(26) A total of 111 DXA scans were not interpretable because of the presence of large movement artifacts or other anomalies, yielding 7333 (98.5%) useable scans. The CV for TBLH BMC was 0.8% based on 120 repeat scans.

Regional DXA measures at the right humerus were derived from total body DXA scans in 549 children reporting fractures over the following 2 years and an additional randomly selected group of 836 children. As well as humeral area, length width and aspect ratio (AR; length divided by width), humeral CSMI, cross-sectional area (CSA), and relative bone strength were calculated from equations obtained from personal communication with TJ Beck (see Appendix).(​27) In addition, humeral vBMD was calculated assuming the humerus is cylindrical by dividing humeral BMC by humeral volume derived from the product of humeral length and total cross-sectional area, of which the latter was calculated from the equation πr².

Fracture incidence

DXA results were linked to results of a questionnaire administered on subsequent attendance at research clinics ~12 and 24 months later, where children were asked if they had broken a bone since they visited for their DXA scan. Children who indicated they had sustained a fracture were sent a further questionnaire to collect information on the nature and circumstances of the injury and for consent to obtain a copy of the radiology report. Because not all children who reported fractures could be contacted to confirm the fracture, the initial report was used to identify those with a fracture. In those where radiology reports were available (~40%), 87% of reported fractures were confirmed.

Other measures

The mother’s, partner’s, and grandparent’s race and ethnic group was recorded by the mother on self-reported questionnaires sent out at ~32 weeks of gestation. Sex was obtained from birth notifications. Mothers’ highest educational qualifications was assessed at 32 weeks of gestation and were coded on an ascending five point scale as follows: 1, none/CSE; 2, vocational; 3, O-level; 4, A-level; 5, university degree level (levels 1, 2, and 3 refer to educational qualifications generally gained at school at age 16 years and level 4 to qualifications gained at school at age 18). Paternal social class was derived using the 1991 Office of Population Censuses and Surveys (OPCS) occupation-based classification, based on the fathers/partners current or last job at 32 weeks of gestation.

Puberty was assessed by self-completion questionnaires using diagrams based on Tanner staging of pubic hair distribution for boys and breast development for girls. Only pubertal measurements completed within 3 months of the DXA scan were used. At the time of the DXA scan and measurement of the anthropometric variables, the child’s baseline age was calculated from the date of birth and date of attendance at the research clinic.

Statistical analyses

Statistical analyses were performed using STATA 8.0. Logistic regression was used to calculate ORs and 95% CIs to describe the associations between variables and the risk of fracture over the following 2 years. Weight of the children was skewed toward the right and therefore was log-transformed. Interactions between variables were assessed by including a multiplicative interaction term in the regression models and calculating the likelihood ratio test (LRT). Z scores were used to provide standardization of an individual’s bone mass and to allow calculation of ORs for fracture risk per SD unit change. The Z scores were produced for the continuous bone mass measures (TBLH BA, TBLH BMC, TBLH BMD) and height and weight by subtracting from the mean and dividing by the SD. Logistic regression analyses were performed unadjusted, adjusted for age, sex, ethnicity, and socio-economic status as assessed by maternal education and paternal social class (minimally adjusted model), and for size variables as described for specific analyses.

RESULTS

Of the 7333 children with useable DXA scans at 9.9 years of age , results were available for reported fracture over the following 2 years in 6213 (84.7%), of whom 3042 (49.0%) were male and 203 (3.6%) were from nonwhite ethnic backgrounds. Five hundred fifty (8.9%) children reported at least one fracture over the 2-year period, of whom 85 (1.4%) reported more than one fracture. Based on analysis of questionnaire returns, 44.6% of fractures were in the forearm, 12.8% were in fingers or thumb, 12.6% were in toes, 5.9% were in the elbow, 5.2% were in the clavicle, 4.0% were in the tibia or fibula, and 2.9% were in the humerus.

Girls and children of nonwhite ethnic backgrounds had a lower proportion of reported fractures, both in crude un-adjusted analyses and after adjustment for age, socioeconomic status, and sex/ethnicity. The adjusted OR for fracture risk over the 2-year follow-up period for girls compared with boys was 0.79 (95% CI, 0.65–0.95; p = 0.014). For nonwhite children compared with whites, the OR was 0.52 (95% CI, 0.25–1.06; p = 0.070), but this was based on only nine fractures in nonwhite children. No difference was found in maternal education or paternal social class between children who did and did not report fractures (results not shown). TBLH BMD was associated with a modest increase in fracture risk after adjusting for age, sex, ethnicity, and socio-economic status, whereas no association was seen for TBLH BMC or TBLH BA, suggesting that childhood fracture risk is related to areal BMD (aBMD) but not skeletal size (Table 1).

Table 1.

Height, Weight, and Bone Mass in 6213 Children With and Without Fractures and OR for Fracture per SD Decrease in Each Variable

Parameter	Children not reporting a fracture (N = 5663) [mean (SD)]	Children reporting a fracture (N = 550) [mean (SD)]	OR for risk of fracture [OR (95% CI) p value]
Height (cm)	139.6 (6.4)	139.7 (6.4)	0.98 (0.88, 1.09) 0.696
Weight (kg)	34.6 (7.4)	35.0 (7.4)	0.96 (0.86, 1.07) 0.435
TBLH BMC (g)	891 (185)	884 (185)	1.07 (0.97, 1.19) 0.186
TBLH BMD (g/cm2)	0.778 (0.055)	0.773 (0.055)	1.12 (1.02, 1.25) 0.023
TBLH BA (cm2)	1137 (165)	1136 (166)	0.94 (0.84, 1.05) 0.250

Mean (SD) height, weight, TBLH BMC, BMD, and BA measured at 9.9 years of age in children who did and did not report fractures over the following 24 months and associations between these variables and risk of fracture. The ORs for fracture, which are per SD decrease in each variable, were calculated using logistic regression and minimally adjusted for age, sex, ethnicity, maternal education, and paternal social class.

Subsequently, we explored the association between TBLH BMC or TBLH BA and fracture risk using different models of size adjustment. Progressive adjustment for body size led to a greater inverse association between TBLH BMC and fracture risk (Table 2). For example, after adjusting for height, weight, and TBLH BA, there was an 89% increased risk of fracture per SD decrease in TBLH BMC. Similar results were seen in this model after excluding those children with one or more fractures before 9.9 years of age (irrespective of whether they reported a fracture during the 2-year follow-up period; OR = 1.84; 95% CI, 0.92–3.70; p = 0.086; n = 4379). Similar results were also seen in this model after excluding those with two or more fractures in the 2-year follow-up period (OR = 2.02; 95% CI, 1.21–2.38; p = 0.007; n = 5145). Adjusting for pubertal status also did not significantly affect the results (OR, 2.08; 95% CI, 1.19–3.64; p = 0.010; n = 4016).

Table 2.

Association Between Estimated Volumetric BMD or Bone Size at 9.9 Years of Age and Fracture Risk Over the Following 2 Years

Model	OR for risk of fracture [OR (95% CI) p value]
Model 1: TBLH BMC minimally adjusted	1.07 (0.97, 1.19) 0.186
Model 1 additionally adjusted for height	1.25 (1.06, 1.48) 0.009
Model 1 additionally adjusted for weight	1.40 (1.16, 1.69) 0.001
Model 1 additionally adjusted for height and weight	1.57 (1.26, 1.96) <0.001
Model 1 additionally adjusted for height, weight and TBLH BA	1.89 (1.18, 3.04) 0.009
Model 2: TBLH BA minimally adjusted	1.04 (0.94, 1.15) 0.495
Model 2 additionally adjusted for height	1.18 (0.99, 1.42) 0.071
Model 2 additionally adjusted for weight	1.28 (1.05, 1.56) 0.013
Model 2 additionally adjusted for height and weight	1.51 (1.17, 1.95) 0.002

Association between TBLH BMC and TBLH BA measured at 9.9 years of age and risk of fracture over the following 2 years. Table shows ORs for fracture per SD decrease in TBLH BMC or TBLH BA calculated by logistic regression. Minimally adjusted models adjusted for age, sex, ethnicity, and socio-economic status. Additional models are shown in which analyses are adjusted for height and/or log-transformed weight and for TBLH BMC additionally adjusted for TBLH BA.

Likewise, adjustment for body size led to an inverse relationship between TBLH BA and fracture risk, as exemplified by a 51% increased risk of fracture per SD decrease in TBLH BA after adjusting for height and weight. Adjusting for pubertal status also did not affect these results (OR = 1.52; 95% CI, 1.12–2.06; p = 0.007; n = 4016). To further explore the relationship between BMC, bone size, and fracture risk, we analyzed OR across tertiles of TBLH BMC and BA. There was an inverse linear relationship with fracture risk after adjusting for skeletal area, height, and weight in the case of TBLH BMC (Fig. 1A) and height and weight in the case of TBLH BA (Fig. 1B).

FIG. 1.

Relationship between tertile of (A) TBLH BMC or (B) TBLH BA as measured at age 9.9 years and the risk of fracture over the following 2 years. TBLH BMC is adjusted for (i) age, sex, ethnicity, maternal education, and paternal social class, and (ii) additionally adjusted for height, log-transformed weight and TBLH BA. TBLH BA is adjusted for (i) age, gender, ethnicity, maternal education and paternal social class and (ii) additionally adjusted for height and log-transformed weight. Graphs show mean ORs ± 95% CIs for the association between each tertile and fracture risk (1 = lowest tertile, 3 = highest), calculated by logistic regression. The highest tertile was taken as the baseline category, and the p values are for the test for trend. TBLH BMC and TBLH BA were reduced to categorical variables (tertiles) after adjustment. The dotted line represents the null value of 1.

We examined the association between fracture risk and skeletal geometry of the upper limb using parameters obtained at the humerus from total body DXA scans at 9.9 years of age in a subset of 1317 children. Measures of humeral geometry, estimated CSMI, and section modulus relative to lean mass were unrelated to fracture risk, both with and without adjustment for height and weight (Table 3). In contrast, an inverse association was observed between humeral aBMD and fracture risk. In further analyses where we estimated humeral vBMD based on a model assuming that the humerus is a cylinder, a similar association with fracture risk was observed to that seen for aBMD. An inverse association was also present between fracture risk and humeral vBMD tertile, such that fracture risk was ~60% greater in children with a humeral vBMD in the lower compared with upper tertile (Fig. 2).

Table 3.

Association Between Humeral Geometry, Biomechanical Properties, and Estimated Volumetric Density at 9.9 Years of Age and Fracture Risk Over the Following 2 Years

Variable	OR for risk of fracture (minimally adjusted) [OR (95% CI) p value]	OR for risk of fracture (size-adjusted) [OR (95% CI) p value]
Measures of humeral geometry
Humeral area	0.95 (0.85, 1.07) 0.427	1.04 (0.86, 1.25) 0.701
Humeral length	0.99 (0.89, 1.12) 0.935	1.19 (0.98, 1.44) 0.083
Humeral width	0.93 (0.83, 1.05) 0.237	0.96 (0.83, 1.11) 0.568
Humeral CSA	0.94 (0.83, 1.05) 0.265	0.96 (0.84, 1.11) 0.625
Humeral AR	1.08 (0.96, 1.21) 0.212	1.08 (0.96, 1.22) 0.190
Measures of humeral biomechanical properties
Humeral CSMI	0.98 (0.88, 1.11) 0.787	1.05 (0.90, 1.22) 0.541
Relative bone strength	0.93 (0.83, 1.04) 0.215	0.91 (0.81, 1.03) 0.133
Measures of humeral vBMD
aBMD	1.22 (1.08, 1.37) 0.001	1.40 (1.21, 1.62) <0.001
vBMD	1.29 (1.14, 1.45) <0.001	1.28 (1.14, 1.45) <0.001

Associations between indices of humeral geometry, humeral biomechanical properties, and humeral vBMD, as measured in a subset of 1317 children at 9.9 years of age, and the risk of fracture in children over the following 2 years. Table shows ORs for fracture per SD decrease, as calculated by logistic regression. In minimally adjusted analyses, results are adjusted for age, sex, ethnicity, maternal education, and paternal social class. In size-adjusted analyses, results are further adjusted for height and log-transformed weight.

FIG. 2.

Relationship between tertile of humeral vBMD as measured at 9.9 years of age and the risk of fracture risk over the following 2 years, adjusted for age, sex, ethnicity, maternal education, paternal social class, height, and log-transformed weight. Graphs show mean ORs ± 95% CIs for the association between each tertile and fracture risk (1 = lowest tertile, 3 = highest), calculated by logistic regression. The highest tertile was taken as the baseline category, and the p values are for the test for trend. Humeral vBMD was reduced to a categorical variable (tertiles) after adjustment. The dotted line represents the null value of 1.

DISCUSSION

In this prospective cohort study, we found an inverse relationship between TBLH aBMD, as measured at 9.9 years of age, and risk of fracture over the following 2 years. aBMD measurements in children can be difficult to interpret because they are influenced by both skeletal size and vBMD. Therefore, we attempted to separate possible relationships between fracture risk and bone size from those with vBMD. Progressive adjustment of TBLH BMC for body and bone size led to a greater inverse association with fracture risk, suggestive of an inverse relationship between vBMD and fracture risk. For example, an 89% increase in risk of fracture over the following 2 years was observed for each SD decrease in TBLH BMC adjusted for height, weight, and TBLH bone area. An equivalent relationship was observed between fracture risk and estimated humeral vBMD derived from subregional analysis.

In the absence of measures such as pQCT, it is not possible to distinguish the relative contribution of different constituents of vBMD to fracture risk, such as cortical thickness and trabecular bone volume, on the basis of these results. Theoretically, reduced cortical thickness might be expected to be largely responsible for this association, and in support of this suggestion, in a previous case-control study, cortical width as assessed by metacarpal morphometry was found to inversely relate to fracture risk in boys and girls.(20)

To our knowledge, this is the first prospective study to examine the relationship between DXA-measured parameters of bone mass and fracture risk in children. Our findings are consistent with the results of our recent meta-analysis based on previous case-control studies(11); when this was repeated to include these results for size-adjusted TBLH BMC (mean of 883 ± 38 and 887 ± 38 g, respectively, in the children with and without fractures), the SMD for the association between bone mass and fracture risk in children was marginally reduced from the original value of −0.32 to −0.26 (95% CI, −0.40 to −0.12). It is unclear why the effect of bone mass on fracture risk that we found was less strong than that observed in the majority of previous studies. The association reported here is likely to be more accurate because of the relatively large sample size, and our use of a prospective rather than case-control design. For example, because bone mass was measured before the fracture, there is no possibility that DXA results could be influenced by the fracture. In addition, a cohort study design reduces the possibility of introducing bias through control selection.

Our observations, which suggest that children who fracture have a lower vBMD compared with those who do not fracture, presumably underlies previous observations from case-control studies that fracture risk in children is inversely related to aBMD, in view of the strong relationship between aBMD and vBMD. On the other hand, no association was observed between fracture risk and TBLH BA (minimally adjusted) and humeral bone area (minimally and size adjusted). Similarly, fracture risk was unrelated to measures of humeral geometry such as length and width and derived biomechanical parameters such as CSMI. Thus, our results would seem to be at odds with the report by Skaggs et al.,(17) who found that cross-sectional area of the radius as measured by pQCT in 50 4- to 15-year-old girls who had recently sustained a forearm fracture was lower than in 50 girls with no history of fractures. A possible explanation for this apparent discrepancy is that, in this study, only 2.9% of reported fractures were of the humerus. Therefore, whereas measures of skeletal geometry may relate to the risk of fracture at the site in question, these may provide limited information about fracture risk at other sites.

An alternative explanation for these discrepant results is that, in the study by Skaggs et al., both groups were matched for height and weight, and therefore bone size relative to body size may have been the principle determinant of fracture risk rather than bone size per se. Consistent with this interpretation, in this study, an inverse association was observed between fracture risk and TBLH BA when the latter was expressed relative to body size by adjusting for height and weight. These findings are also consistent with results of our recent case-control study based on 100 children 4–16 years of age with fractures and 50 nonfracture controls, in which bone size was found to be a determinant of fracture risk after adjusting for height and weight.(24)

The biological significance of this apparent relationship between bone size relative to body size and fracture risk is currently unclear. One possible explanation is that bone size relative to body size represents the level of adaptation to loading requirements. However, measures such as humeral area relative to body size and humeral relative bone strength were unrelated to fracture risk. Because childhood fractures largely affect the upper limbs, if anything, humeral measures are likely to reflect biomechanical determinants of fracture risk in children more accurately than those derived from TBLH BA. Alternatively, TBLH adjusted for body size may represent an interrelationship between skeletal size, fracture risk, and body composition, as suggested by recent evidence that body weight and adiposity are negative determinants of fracture risk in childhood.(28,29)

Our finding that fracture risk in children is higher in boys compared with girls is consistent with several previous studies.(3,4,30) Although this difference has previously been attributed to greater risk taking behavior in boys,(10) sporting injuries have recently been reported to be more common in girls between 3 and 12 years of age.(31) Taken together, these observations raise the possibility that greater skeletal fragility contributes to the higher fracture risk seen in boys. Consistent with this possibility, we found that humeral vBMD tended to be lower in boys compared with girls (0.489; 95% CI, 0.485–0.492 and 0.494; 95% CI, 0.490–0.497, respectively, p = 0.09), which may have contributed to the higher fracture risk in boys.

In terms of possible weaknesses of this study, 15.3% of the study population was lost over the 2-year follow-up period, which may have introduced bias because there was a preferential dropout of children from families of lower socio-economic status (results not shown). In addition, although 87% of subjects in whom we were able to obtain X-ray reports were confirmed as having a fracture, we were not able to verify reported fractures in all cases, and it is inevitable that a small number of children were erroneously classified as having had a fracture. However, any such mis-classification of our outcome is likely to have underestimated the association between DXA measures and fracture risk, rather than to have produced a spurious association. We were not able to exclude children who fractured because of high levels of trauma such as road traffic injury, where fracture is likely to be inevitable and not dependent on bone mass, and this may also attenuate the size of the association. Finally, although we interpreted our findings as indicating a difference between vBMD between children with and without fractures, in the absence of other measures such as pQCT, it is not possible to determine whether these differences are as a consequence of altered cortical thickness, trabecular bone content, or level of bone tissue mineralization.

In conclusion, this is the first report of a prospective cohort study of bone mass and fracture risk in childhood. We found that TBLH aBMD as measured by DXA at 9.9 years of age is inversely related to the risk of fracture over the following 2 years. Further analyses suggested that this relationship predominantly reflects an inverse relationship between vBMD and fracture risk, because a stronger inverse association was observed between fracture risk and size-adjusted TBLH BMC and estimated humeral vBMD. In contrast, skeletal size per se was not related to fracture risk. However, an inverse relationship was found between fracture risk and TBLH bone area relative to body size. Further studies are justified to extend these findings, for example by examining whether the relationship between fracture risk and bone size relative to body size that we observed reflects an interaction between fracture risk, bone size, and body composition in childhood.

APPENDIX

Cross-sectional moment of inertia (CSMI) was calculated by

$\mathrm{CSMI}=\frac{\prod (r_{\mathrm{o}}^4-r_{\mathrm{i}}^4)}{4}$

Where r_o (radius of outer cortex) was one half the width, and r_i (radius of inner cortex) was calculated as:

$r_{\mathrm{i}}=\sqrt{r_{\mathrm{o}}^2-f_{\mathrm{c}}\frac{\mathrm{CSA}}{\pi }}$

where f_c is the fraction of the bone mass in the cortex, taken as 1 for diaphyseal regions such as the shaft of the humerus.

Humeral cross-sectional area (CSA) (excluding soft tissue spaces) was calculated as average thickness (BMD divided by the average mineral density of 1.052 g/cm³) multiplied by the humeral width. It is recognized that this arbitrary constant of 1.052 g/cm³ is derived from adult values and may not be valid in children. Section modulus Z was calculated as CSMI divided by r_o. Z was divided by one half the length of the humerus as a proxy for moment arm length to obtain a measure of a geometric strength index. This was normalized to an estimate of muscle index, taken as upper limb lean mass measured by the whole body DXA scan.(32)