Lean Mass and Not Fat Mass Is Associated With Male Proximal Femur Strength

Abstract

Obesity is suspected to confer protection against fracture, but evidence is mixed. We examined proximal femur geometry and body composition measures in a diverse group of 1171 men (30–79 yr of age). Analyses showed that nonbone lean mass, but not fat mass, is independently associated with measures of proximal femur density, axial and bending strength, and resistance to buckling.

Introduction

Obesity is often said to confer protection against fracture, but the mechanisms driving such an association remain poorly understood. We hypothesized that the effect of increased body mass on bone structure would be accounted for by total and/or appendicular nonbone lean mass, and that once these trends were removed, fat mass would show no protective influence. To test this hypothesis, we examined body composition and geometric indices of proximal femur strength in an ethnically diverse (black, Hispanic, and white) sample of randomly selected men, 30–79 yr of age.

Materials and Methods

Data were obtained from N = 1171 community-dwelling subjects enrolled in the cross-sectional Boston Area Community Health/Bone study. Body composition was obtained by DXA. Hip geometry parameters at the narrow neck, intertrochanter, and shaft were obtained using Hip Structural Analysis of DXA images. These measures included BMD, bone material in cross-sections (cross-sectional area), bending strength (section modulus), and propensity to buckle under compression (average buckling ratio). Analyses controlled for age, race/ethnicity, height, and physical activity.

Results

In exploratory analyses, lean mass, fat mass, and BMI were each positively associated with hip strength. However, controlling for lean mass was sufficient to remove the positive, and induce a negative, association for fat mass or BMI. Associations between lean mass and hip strength were strongest and resistant to control for other measures. Lean mass alone was sufficient to account for a substantial proportion of racial/ethnic difference in hip strength measures, whereas fat mass exhibited no comparable explanatory power.

Conclusions

The positive association between relative weight and proximal femur strength is accounted for by lean mass, suggesting that, in men, the protective effect of BMI in preventing fracture is mediated not by adipose tissue but by the influence of increased muscle mass accompanying elevated BMI.

INTRODUCTION

In common use, obesity connotes excess adiposity, but in epidemiologic research, it is typically designated by elevated body mass index (BMI) or some other easily measured anthropometric criterion. The association of elevated BMI with BMD, in combination with several lines of evidence suggesting that adipose tissue exercises a metabolic influence over bone growth,(1–4) motivates the hypothesis that adiposity is a prime determinant of bone strength. Indeed, body weight is one of the strongest predictors of BMD, and elevated weight and BMI are linked to decreased risk of fracture in adults.(5–8) These associations, however, have long been understood to be mediated not only by crude skeletal load—provided by fat and lean tissue in combination—but also by the mechanical stresses specifically provided by lean mass action.(9,10)

Existing research studying the relative importance of lean and fat tissue in stimulating osteogenesis has established no consensus. Although several studies suggest an important or primary role for fat in determining bone mass and density,(11–19) associations perhaps most pronounced in postmenopausal women,(3) others have indicated that lean mass is dominant in women,(19–21) men,(11,13,22,23) or both.(10,24–27) Consequently, the precise biomechanical and metabolic relations between bone mass and components of body composition remain the focus of substantial attention.

Our previous analyses of data from the Boston Area Community Health/Bone study showed that positive associations between male BMC or BMD and BMI, waist-to-hip ratio, fat mass, and other generalized weight measures were diminished or reversed once total body lean mass was taken into account.(28) As much as the latter measures can only approximate adiposity, however, so BMD, despite its strong association with fracture,(29) is limited as a summary of bone strength and resistance to mechanical failure. The strength of bone depends both on its material composition and its architecture,(30) of which only the latter may be assessed by noninvasive means. These architectural properties, in turn, are defined both by the amount or density of material present and by the structural arrangement of that material. Attention to both of these factors allows for multidimensional assessments of strength, capturing variation in fracture risk that could be lost through consideration of BMC or BMD alone.(9,31–37)

Existing evidence suggests that the geometric properties of the proximal femur are more closely adapted to lean mass than to body weight(9,38,39) and that lean mass is the most likely mediator of the effects of weight on male proximal femur strength.(40) This makes intuitive sense because the stresses applied by muscle to bone typically far exceed those imposed by weight alone.(41,42)

We have previously(43,44) showed substantial age-related and racial/ethnic variation in cross-sectional measures of BMD and proximal femur geometry among subjects enrolled in BACH/Bone. Motivated by these findings and those mentioned above, we undertook a close examination of body composition in these men to determine the specific contribution of lean tissue to the cross-sectional association between weight and proximal femur strength. We hypothesized that overall or appendicular lean mass would account for the association between BMI and geometric indices of proximal femur strength and that fat mass would exhibit little positive association with those indices once lean mass was taken into account. We further hypothesized that, consequently, lean mass would dominate fat mass in accounting for racial/ethnic heterogeneity in hip geometry. To evaluate these hypotheses, we examined body composition parameters and measurements of hip geometry at the narrow neck (NN), intertrochanter (IT), and shaft (S) regions of the proximal femur.

MATERIALS AND METHODS

Design

The BACH/Bone study(43) is a cross-sectional study of bone health among 1219 community-dwelling black, Hispanic, and white men. BACH/Bone enrolled subjects from the larger BACH survey,(45) a study of 5503 randomly selected residents of greater Boston, MA, with an age range of 30–79 yr. BACH used a multistage stratified cluster random sampling scheme in obtaining approximately equally sized subsamples by sex, race (black, Hispanic, and white), and age group (30–39, 40–49, 50–59, and 60–79 yr). BACH data were collected between April 2002 and June 2005.

At the conclusion of BACH data collection visits, male subjects were asked if they would be willing to participate in BACH/Bone. Enrollment required that subjects weigh ≤300 lb, be able to travel to the Boston University School of Medicine (BUSM), consent to a DXA scan, and be able to lift themselves onto the scan table. Between November 2002 and July 2005, 1219 (65%) of 1877 eligible BACH subjects were enrolled in BACH/Bone; 1209 subjects completed DXA scans.

The BACH Survey protocol was approved by the New England Research Institutes (NERI) Institutional Review Board (IRB), and the BACH/Bone protocol was approved by IRBs at both NERI and BUSM. Written informed consent for each study was obtained independently. Extensive information concerning the BACH and BACH/Bone designs and sampling procedures has been previously published.(43,45)

Data collection

During the BACH interview, subjects provided demographic information, a summary of their normal level of physical activity using the Physical Activity Scale for the Elderly (PASE),(46,47) and other measures. Subjects' race and ethnicity were determined by self-identification according to U.S. Office of Management and Budget guidelines.(48,49)

During the BACH/Bone visit, BUSM General Clinical Research Center staff performed the DXA scan and obtained anthropometric data. Height was measured to the nearest 0.1 cm using a stadiometer, and weight was measured to the nearest 0.1 kg using a digital scale.

DXA and HSA

Hip scans and measures of body composition were obtained using a Hologic QDR 4500W densitometer (Hologic, Waltham, MA, USA). Body composition measures included total body and regional fat mass (FM) and nonfat mass. All mass quantities reported here exclude the head. Lean mass (LM) was calculated by subtracting BMC from nonfat mass. Appendicular lean and FM (ALM and AFM) quantities were generated by adding the relevant quantities for subjects' arms and legs.

Hip geometry parameters were derived using the Hip Structural Analysis (HSA) program(50) based on principles first articulated by Martin and Burr.(42) Generation of HSA data was conducted by trained staff at the Johns Hopkins University School of Medicine under the supervision of one of the authors (TJB).

The HSA methodology uses engineering principles to extract data on the geometry of the proximal femur from DXA images, as has been described in detail elsewhere.(50,51) This involves analysis of cross-sectional slabs of bone within each of the three regions of interest (NN, IT, S). Measurements are averaged over five parallel profiles each of 1 pixel (∼1 mm) thickness, thus corresponding approximately to 5-mm-thick cross-sections. For each of these, estimated BMD (g/cm²) is measured as the average pixel intensity in the profile (not to be confused with conventional femoral neck or total hip BMD by DXA alone), and the outer diameter (OD) as the distance between the profile outer margins after correcting for image blur. The mineral profile is converted from mineral thickness to linear thickness by dividing by the average mineral density of adult cortical bone. The resulting integral measures the cortical tissue equivalent surface area of the bone in the cross-section excluding soft tissue spaces, which do not provide mechanical support. This is expressed as bone cross-sectional area (CSA, cm²).

After the centroid has been located, the cross-sectional moment of inertia (CSMI, cm⁴) is computed as the integral weighted by the square of distance from the centroid. The section modulus (Z, cm³), an index of bending strength, is derived for each cross-section as the CSMI/d_max, where d_max is the maximum distance from the centroid to the medial or lateral cortical margin. Finally, average buckling ratio (ABR), an index of cortical stability under compressive loads, is calculated by dividing d_max by the estimated mean cortical thickness derived from a circular or elliptical annulus model of the cortex with a fixed fraction of measured mass in the cortex.

Using these methods, we obtained measures of BMD, cross-sectional width (OD), material (CSA), bending strength (Z), and propensity to buckle under compression (ABR) for use in this report. Because of the form of the computations by which they are derived, higher values of BMD, Z, CSA, and OD correspond to increases in strength, broadly speaking, whereas increases in ABR, conversely, denote weaker bone.

Analytic sample and statistical methods

Of 1209 potential DXA images, 17 were unsuited to processing by HSA because of the presence of artifacts or lack of image clarity. An additional 19 subjects lacked lean and FM values, and 2 subjects were removed from analyses because of the fact that they had several outlying hip geometry measurements (>4 SD from the corresponding sample means). These exclusions left N = 1171 subjects for consideration in the analyses described here.

Before formal modeling, we used exploratory data analysis to assess the strength and form of statistical associations. Graphical summaries were obtained by smoothing the data using generalized additive models (GAMs),(52) which assume a continuous relationship between dependent and independent variables that is allowed to be curvilinear (e.g., U-shaped) if the data so imply. The form of such relationships is known to be heavily influenced by the choice of the smoothing parameter, indexing the degree of “wiggle” in the fitted line or plane. To reduce any influence of such subjectivity, we used the version of GAM given by Wood,(53) which automates the choice of a reasonable degree of smoothness as part of the algorithm by which the fit is generated.

Once the functional form of relationships between components of body composition and outcomes was established, estimation and inference were conducted using multiple regression analyses. To enhance interpretability, regression estimates were scaled to the overall sample means of the hip geometry parameters (i.e., regression parameters were divided by the relevant sample means). The resulting point and interval estimates may therefore be interpreted on a proportionate scale, where the denominator of that proportion is the sample average. These analyses were conducted using SUDAAN 9.0.1 (Research Triangle Institute, Research Triangle Park, NC, USA), which accommodates weighting according to the complex sampling design, so that results may be interpreted as representative of the greater Boston population. Statistical significance was determined using Wald-type tests.

RESULTS

Sample characteristics are presented in Table 1. The analytic sample was made up of 348 black men, 391 Hispanic men, and 432 white men. Weighted mean age was 47.5 ± 12.8 (SD) yr. Although white subjects had slightly lower mean BMI than other subjects, they exhibited the highest overall total and appendicular FM. Black subjects, on the other hand, exhibited the greatest total and appendicular LM, despite being 2.7 cm shorter than white subjects on average.

Table 1.

Study Sample Characteristics* (N = 1171)

Subject characteristics	Overall	Black men (N = 348)	Hispanic men (N = 391)	White men (N = 432)
Age (yr)	47.5 ± 12.8	48.0 ± 12.6	44.5 ± 11.0	48.0 ± 13.2
BMI (kg/m2)	28.4 ± 4.6	28.7 ± 5.1	28.3 ± 4.7	28.2 ± 4.4
Total LM (kg)	55.1 ± 7.7	56.3 ± 8.8	51.8 ± 7.2	55.4 ± 7.1
Total FM (kg)	22.0 ± 8.6	20.6 ± 9.0	19.7 ± 7.4	23.1 ± 8.5
ALM (kg)	26.6 ± 4.1	28.0 ± 4.9	25.1 ± 3.9	26.4 ± 3.7
AFM (kg)	9.3 ± 3.5	9.2 ± 4.0	8.1 ± 3.0	9.6 ± 3.4
Height (cm)	175.7 ± 7.4	174.7 ± 7.3	169.6 ± 6.1	177.4 ± 6.9
PASE	187 ± 109	190 ± 112	190 ± 100	185 ± 108

Values are mean ± SD.

* Estimates weighted according to sampling design.

Exploratory analysis: LM versus FM, age, and height

Simple graphical displays initially indicated enhanced bone strength among younger men and those with greater overall fat or LM. The left panels of Fig. 1 provide an illustration, with specific focus directed to NN CSA. Unadjusted CSA decreases with age, increases quite sharply with LM, and exhibits a moderate, curvilinear increase with FM. A curvilinear relationship with BMI (data not shown) was also observed, similar to the associations between BMI and BMC that were documented previously.(28)

FIG. 1.

Narrow neck. Crude association between bone material (CSA) and age, FM, and LM, singly and in multivariate models. Smooth terms and residual associations are obtained using GAM.(53) Shaded regions depict adjusted CIs for mean CSA. Left panels give unadjusted associations, whereas right panels detail residual associations; for instance, the middle right panel shows the association between LM and CSA after the influences of age, height, and FM have been removed. There is substantial association between CSA and each of the other variables when values are unadjusted, but the association between FM and CSA reverses direction when the other variables are held constant.

When these parameters were examined simultaneously, LM seemed to be most strongly associated with CSA, as was consistent with our hypothesis. The right panels of Fig. 1 show residual associations between age, FM, and LM once the other factors have been entered into the model. The association between LM and CSA is enhanced somewhat when age and FM are added as a control factors (middle right panel), but that between FM and CSA becomes negative once LM is taken into account (bottom right panel), although this negative association is mild. Controlling for LM induced a stronger negative association between BMI and CSA (data not shown).

Examining FM or LM simultaneously produced similar exploratory results across hip regions and the different indices of proximal femur strength. Figure 2 depicts age- and height-adjusted proportionate increases (as defined in the Materials and Methods section) in BMD, CSA, OD, Z, and ABR associated with a 10-kg increases in total FM and LM from models including each individually (estimates displayed in gray) and simultaneously (displayed in black). That the marginal, unadjusted associations between FM and proximal femur strength are positive is indicated by the fact that the gray CIs lie to the right of the zero line. These trends are eliminated or reversed, however, when models also account for LM, so that the black point estimates lie on or to the left of that line. (Results for ABR are similar, although the signs of effects are inverted, as noted in the Materials and Methods section.) In contrast, for all regions, the positive association between total LM and hip strength is essentially unchanged when models control for the effects of FM. As is displayed in Fig. 2 for NN specifically (insets), results are also similar when AFM and ALM are substituted for FM and LM, respectively. Taken as a whole, these results showed that LM effects persist when the effect of mechanical loading by FM is taken into account, but FM effects are eliminated, or even rendered negative, by control for LM effects.

FIG. 2.

Geometric strength indices: age- and race/ethnicity-adjusted associations with fat and LM, singly and in concert. Symbols indicate estimated proportionate differences in geometric strength indices between men whose total fat or LM differs by 10 kg. Models controlling only for age and race/ethnicity (gray symbols) show a strong positive association between strength indices (BMD, CSA, Z, ABR) and both lean and FM. However, models including both lean and FM and height (black symbols) reverse the association between FM and strength, whereas the association between LM and strength is essentially unchanged. Results are nearly identical when AFM and ALM are substituted for FM and LM, respectively (NN shown, insets).

Exploratory analysis: FM, LM, and racial/ethnic heterogeneity in geometry

To assess the likelihood of statistical interaction between LM and FM in models of proximal femur geometry, we examined multidimensional depictions of the joint height-adjusted association of FM and LM with strength indices. Results are presented in Fig. 3, with specific focus on NN bending strength (section modulus, Z) and are quite similar to those related to CSA in Fig. 1. Here we observe (Fig. 3A) that, with age and height held constant, Z exhibits a sharp linear increase with increasing LM and a milder decrease with increasing FM. Results indicate similar associations between Z and LM at various levels of FM, and likewise, that the mild negative relation between Z and FM is consistent across the possible levels of LM, so that there is little evidence of statistical interaction. (Results for other regions and indices, not shown, are similar.)

FIG. 3.

Narrow neck: joint association of lean and FM with bone bending strength. Age- and height-adjusted estimates obtained by GAM show that trends of bending strength, as measured by section modulus (Z), increases rapidly with increasing total LM, but exhibit mild decreases with increasing FM once LM is taken into account (A). Variation in total LM can account for a substantial variation in bending strength by age and race/ethnicity; height-adjusted Z is greater in younger black men than in others (B), a difference is sharply reduced when LM is controlled (C), but exhibits only mild further reductions when FM is taken into account (D). SE bars accompany mean estimates in B–D.

Exploratory analyses also indicated that previously reported racial/ethnic variation in age-specific Z(44) (Fig. 3B) is sharply reduced in models controlling for total LM (Fig. 3C), but that models adding FM in addition to LM produced little further reduction in racial/ethnic differences (Fig. 3D). A depiction (data not shown) of results controlling for FM but not LM was nearly identical to Fig. 3B, indicating the minimal impact of introducing FM alone into exploratory models. Once again, we observed similar results (data not shown) when AFM and ALM were substituted for FM and LM. In addition, results were similar when BMI was substituted for FM (and height removed from the model) so that we observed that controlling for LM was sufficient to reverse the positive association between BMI and NN Z.

Formal results: multivariate regression analyses

Exploratory analyses such as those presented in the figures indicated that multivariate linear models would reasonably approximate trends in strength indices as a function of LM, FM, and covariates.

Based on previous findings,(43,44) when constructing regression models, we allowed both mean bone strength parameters and their age trends to vary by race/ethnicity. The resulting estimation for NN BMD, CSA, Z, and ABR is described in Table 2. (Associations between body composition and OD, while consistent with those for the other parameters, were of more limited magnitude [see Fig. 2] and were not further pursued.)

Table 2.

Multiple Linear Regression Results*: Narrow Neck

Outcomes and covariates	Base model		Model II		Model III
	Percent difference (95% CI)	p	Percent difference (95% CI)	p	Percent difference (95% CI)	p
BMD (g/cm2)
Fat mass, 10 kg					−1.0 (−2.6, 0.6)	0.21
Lean mass, 10 kg			7.9 (6.0, 9.7)	<0.001	8.7 (6.3, 11.0)	<0.001
Race/ethnicity, age = 50 yr		<0.001		<0.001		<0.001
Black	11.2 (8.5, 13.9)		9.0 (6.3, 11.7)		8.6 (5.8, 11.4)
Hispanic	5.1 (2.2, 8.1)		4.8 (2.2, 7.4)		4.6 (2.0, 7.2)
White	Ref		Ref		Ref
Age, 10 yr		<0.001		<0.001		<0.001
Black	−3.1 (−4.9, −1.2)		−2.7 (−4.3, −1.0)		−2.5 (−4.2, −0.9)
Hispanic	−5.1 (−6.5, −3.7)		−4.5 (−5.9, −3.1)		−4.4 (−5.8, −3.0)
White	−2.1 (−3.5, −0.8)		−1.7 (−2.9, −0.4)		−1.6 (−2.8, −0.3)
Height (cm)	0.4 (0.2, 0.6)	<0.001	−0.04 (−0.2, 0.1)	0.63	−0.1 (−0.2, 0.1)	0.47
PASE, 100 units	1.7 (0.7, 2.8)	0.002	1.6 (0.5, 2.6)	0.003	1.4 (0.4, 2.5)	0.01
CSA (cm2)
Fat mass, 10 kg					−2.7 (−4.2, −1.1)	0.001
Lean mass, 10 kg			10.2 (8.2, 12.3)	<0.001	12.2 (9.8, 14.7)	<0.001
Race/ethnicity, age = 50 yr		<0.001		<0.001		0.001
Black	8.5 (5.8, 11.2)		6.0 (3.4, 8.5)		5.0 (2.3, 7.6)
Hispanic	2.5 (0.5, 5.5)		2.1 (−0.4, 4.5)		1.5 (−0.9, 4.0)
White	Ref		ref		Ref
Age, 10 yr		<0.001		<0.001		<0.001
Black	−2.4 (−4.2, −0.6)		−1.8 (−3.2, −0.3)		−1.4 (−2.8, 0.1)
Hispanic	−4.3 (−5.7, −2.8)		−3.5 (−4.8, −2.3)		−3.3 (−4.6, −2.0)
White	−1.9 (−3.4, −0.5)		−1.3 (−2.6, −0.03)		−1.0 (−2.3, 0.2)
Height (cm)	0.8 (0.6, 1.0)	<0.001	0.2 (−0.003, 0.4)	0.05	0.1 (−0.1, 0.3)	0.18
PASE, 100 units	1.4 (0.2, 2.6)	0.02	1.2 (0.1, 2.3)	0.04	0.9 (−0.2, 2.0)	0.13
Z (cm3)
Fat mass, 10 kg					−4.6 (−6.6, −2.6)	<0.001
Lean mass, 10 kg			11.4 (8.8, 14.0)	<0.001	14.9 (11.7, 18.1)	<0.001
Race/ethnicity, age = 50 yr		0.001		0.07		0.29
Black	6.3 (2.8, 9.8)		3.6 (0.2, 6.9)		1.8 (−1.6, 5.3)
Hispanic	0.6 (−3.0, 4.2)		0.1 (−3.0, 3.3)		−0.8 (−3.9, 2.3)
White	Ref		ref		ref
Age, 10 yr		0.02		0.06		0.19
Black	−1.6 (−3.7, 0.5)		−0.8 (−2.6, 0.9)		−0.2 (−1.9, 1.6)
Hispanic	−3.2 (−5.5, −1.0)		−2.4 (−4.2, −0.7)		−2.0 (−3.8, −0.2)
White	−0.7 (−2.5, 1.2)		0.02 (−1.7, 1.7)		0.5 (−1.2, 2.2)
Height (cm)	1.1 (0.9, 1.4)	<0.001	0.5 (0.2, 0.8)	<0.001	0.4 (0.1, 0.7)	0.01
PASE, 100 units	1.6 (−0.1, 3.4)	0.07	1.4 (−0.3, 3.1)	0.10	0.9 (−0.8, 2.6)	0.31
ABR
Fat mass, 10 kg					−1.3 (−3.6, 1.0)	0.26
Lean mass, 10 kg			−6.1 (−8.7, −3.5)	<0.001	−5.1 (−8.6, −1.6)	0.005
Race/ethnicity, age = 50 yr		<0.001		<0.001		<0.001
Black	−15.0 (−19.0, −11.1)		−13.1 (−17.2, −9.1)		−13.6 (−17.8, −9.4)
Hispanic	−10.3 (−14.6, −6.1)		−10.2 (−14.3, −6.1)		−10.4 (−14.6, −6.3)
White	Ref		ref		ref
Age, 10 yr		<0.001		<0.001		<0.001
Black	4.3 (2.4, 6.1)		4.0 (2.1, 5.9)		4.2 (2.3, 6.1)
Hispanic	6.0 (3.9, 8.0)		5.5 (3.4, 7.5)		5.6 (3.5, 7.7)
White	3.7 (1.6, 5.7)		3.3 (1.2, 5.4)		3.5 (1.4, 5.6)
Height (cm)	0.004 (−0.2, 0.2)	0.97	0.3 (0.1, 0.6)	0.02	0.3 (0.02, 0.6)	0.04
PASE, 100 units	−2.5 (−3.9, −1.0)	0.001	−2.3 (−3.8, −0.9)	0.002	−2.5 (−4.0, −1.0)	0.001

*Point estimates and CIs reflect percentage differences in indices of bone strength. Percentages are derived by dividing regression effects by the global mean of the relevant parameters, as in Fig. 2. Models allow both geometry parameters and their cross-sectional age trends to differ by race/ethnicity. Moving from left to right, changes in the magnitude and significance of effects reflect the impact of controlling for lean (model II), and for fat and lean (model III), mass in successive models.

For each geometry parameter, three models are presented. The first (base) model shows effects for age, race/ethnicity, height, and physical activity score (PASE). The second (model II) presents results controlling for these as well as LM, whereas the third (model III) controls for total FM as well. Comparisons of these models confirm our main hypothesis and quantify the exploratory analyses described above; whereas total FM is positively and consistently associated with NN BMD, CSA, and Z when considered alone in the exploratory models detailed above, it exhibits negative association with these parameters once LM is taken into account (model III). Taking CSA as an example, we observed that it increased by an estimated 2.9% (95% CI: 1.5%, 4.4%) per 10-kg cross-sectional increase in FM with LM unaccounted for (data not shown in the table), but decreased by 2.7% (95% CI: 1.1%, 4.2%) per 10-kg increase in FM in models accounting for total LM (as shown in the rightmost column of Table 2). Negative adjusted associations between FM and geometry indices were statistically significant for CSA and Z but statistically indistinguishable from zero for BMD and ABR.

In contrast, LM exhibited a powerful association with hip geometry parameters, with associations between LM and BMD, CSA, and Z actually increasing in magnitude once FM was entered into models. Table 2 also shows that controlling for LM seems to reduce overall differences by race/ethnicity (as was observed for Z in Fig. 3) and to attenuate age trends.

DISCUSSION

These cross-sectional results support the contention that LM, but not FM, is associated with proximal femur density and strength when age, height, and body composition are accounted for. Exploratory analyses and multivariate regression modeling indicate a nonpositive, and potentially negative, association between FM and hip geometry when LM is held constant. In addition, LM differences across age and race/ethnicity were able to account for a substantial proportion of variation in proximal femur strength between these groups, whereas FM had no comparable explanatory power. Results were similar when FM was replaced by BMI. To our knowledge, this is the first study showing these trends in a large population of white, Hispanic, and black American men.

The results presented here are broadly consistent with our previous, more conventional analyses of BMC and BMD(44) showing that controlling for LM reduces or reverses associations between FM and BMC or BMD (measured by DXA) in the femoral neck and lumbar spine. They are also consistent with recent work by Hsu et al.(10) and Zhao et al,(27) who gave similar findings with regard to BMC and BMD. Our results likewise echo those of Semanick et al.,(40) who showed that LM had a stronger effect on NN and shaft bending and axial strength than did FM among men enrolled in the Tobago Bone Health Study.

Implications for obesity and fracture risk

Although low BMD is among the strongest predictors of hip fracture, it is not a measure of bone strength per se. The geometric properties of bone such as those described here capture variation in risk that may not be embedded in the more traditional measures of BMC or BMD. These results suggest that LM is the mediator of associations between body weight and bone strength in men. Consideration of body composition parameters in light of these associations may help to explain the nonlinear trend in fracture risk accompanying change in BMI.(8) By virtue of the fact that men with lower BMI tend to have higher proportionate LM, a 1-unit difference in BMI would have greater implications for muscle forces applied to bone among leaner men. Among the subjects considered in this report, for instance, the proportion of nonbone mass that is lean tissue has a strong negative association with BMI (Pearson correlation, −0.74). The practical implication of this is that increases in BMI are not reflected in commensurate increases in LM among the subjects whose BMI is largest; in the BACH/Bone sample, for instance, there is essentially no association between BMI and LM among subjects whose BMI is at least 35 kg/m². Not coincidentally, it is within this range that unadjusted associations between BMI and BMC and BMD are sharply attenuated (as shown in prior results(28)), and that existing research indicates that additional increases in BMI confer no protection against hip fracture. Despite comparable racial/ethnic differences in FM than LM (Table 1), analyses showed that LM was much more successful in explaining ethnic variation in geometric strength indices. This adds further strength to our result and implies that LM differences may play an important role in explaining racial and ethnic variation in hip fracture rates.

Interpretation of negative association between FM and strength indices

LM, FM, and BMI are nearly collinear, and linear models including all three would be nonsensical. We have endeavored here first to establish the superiority of LM vis-à-vis either FM or BMI as a correlate of proximal femur strength, and second to estimate the independent contributions of lean and FM to BMD and hip geometry. Models controlling for both LM and BMI, described briefly above, must be interpreted with care, because holding weight constant (in the form of BMI) dictates that decreased LM implies increases in adiposity (cf. Wang et al.,(19) Appendix A). These models are properly interpreted as indicating that, for fixed BMI, proximal femur strength increases in relation to that portion of nonbone mass that is donated to the total by lean tissue.

To estimate the relative importance of fat and LM, we compared LM and FM effects directly to obtain their controlled associations with hip geometry (Figs. 1–3; Table 2). These results indicate no increase, or mild cross-sectional decreases, in proximal femur strength indices with increasing FM, conditional on a given total body LM. This suggests that two subjects with equal total LM but differing FM may exhibit modest differences in hip geometry, with the heavier subject exhibiting weaker bone structure. The mechanisms by which this occurs are not immediately clear, although the results given in Table 2 indicate that they are not caused by the influence of age, height, or physical activity. This result is in contrast to that found by Wang et al.,(19) who showed mild positive effects of FM in similar models on young women, but is consistent with our previous analyses of BMC in men.(28)

Strengths and limitations

The analyses presented here have several strengths. First and foremost, the data describe a true population-based sample. The BACH/Bone cohort is a large, representative, and randomly selected population of men, diverse in age and race/ethnicity, and inferences should therefore be applicable to the broader community of aging men. Second, we carefully diagnosed the functional form of relationships between parameters, showing that the multiple regression models we apply are reasonable summaries of the associations between hip geometry and body composition parameters. In addition, we showed through careful multivariate analyses that the effects shown here are unlikely to be artifacts of age, height, or physical activity, which did not affect their strength or consistency across measures and regions.

Some limitations must also be acknowledged. Our hip geometric data are obtained through HSA and measurements of geometry using DXA methods, which are subject to certain technical limitations.(50) For instance, bone geometry assessments using relatively low-resolution CT or DXA evaluate macroscopic geometry of cross-sections, but thinned osteoporotic bones may fail at the microstructural level in ways that may not be readily apparent from these measures. Average buckling ratio is an attempt detect propensity for local cortical failure under compressive loads, but this estimate is necessarily crude. Finally, the cross-sectional nature of the study design is a limitation. Confirmation of our results in the longitudinal setting will be an important future endeavor.

Conclusions

We conclude that, in men, LM is more strongly associated with proximal femur strength than is FM. In addition, LM dominates “obesity” (i.e., enhanced weight relative to height) in association with measures of proximal femur strength, which suggests a stronger role for LM in decreasing the propensity for hip fracture. These results imply that the maintenance of LM is critical to maintaining bone strength for aging men, whereas simply maintaining weight, without regard to whether it is composed of loads donated by lean as opposed to fat tissue, is less likely to be successful.