Contribution of In Vivo Structural Measurements and Load/Strength Ratios to the Determination of Forearm Fracture Risk in Postmenopausal Women
Bone structure, strength, and load-strength ratios contribute to forearm fracture risk independently of areal BMD.
Introduction: Technological and conceptual advances provide new opportunities for evaluating the contributions of bone density, structure, and strength to the pathogenesis of distal forearm fractures.
Materials and Methods: From an age-stratified random sample of Rochester, MN, women, we compared 18 with a distal forearm fracture (cases) to 18 age-matched women with no osteoporotic fracture (controls). High-resolution pQCT was used to assess volumetric BMD (vBMD), geometry, and microstructure at the ultradistal radius, the site of Colles' fractures. Failure loads in the radius were estimated from microfinite element (μFE) models derived from pQCT. Differences between case and control women were assessed, and the risk of fracture associated with each variable was estimated by logistic regression analysis.
Results: Given similar heights, estimated loading in a fall on the outstretched arm was the same in cases and controls. However, women with forearm fractures had inferior vBMD, geometry, microstructure, and estimated bone strength. Relative risks for the strongest determinant of fracture in each of the five main variable categories were as follows: BMD (total vBMD: OR per SD change, 4.2; 95% CI, 1.4–12), geometry (cortical thickness: OR, 4.0; 95% CI, 1.4–11), microstructure (trabecular number: OR, 2.3; 95% CI, 1.02–5.1), and strength (axial rigidity: OR, 3.8; 95% CI, 1.4–10); the factor-of-risk (fall load/μFE failure load) was 24% greater (worse) in cases (OR, 3.0; 95% CI, 1.2–7.5). Areas under ROC curves ranged from 0.72 to 0.82 for these parameters.
Conclusions: Bone geometry, microstructure, and strength contribute to forearm fractures, as does BMD, and these additional determinants of risk promise greater insights into fracture pathogenesis.
Areal BMD (aBMD) at the distal radius can predict subsequent wrist fractures, with a relative risk of ∼1.7 per SD change,(1) but the actual basis for this association is somewhat uncertain. In particular, it is not clear whether fracture risk is determined by BMD per se or by some related parameter. Previously available technologies allowed some assessment of bone size (a confounder of aBMD) and geometry. However, it is now possible to explore this issue in more detail using high-resolution pQCT (HRpQCT), which can measure numerous micro- and macrostructural variables in the distal radius, along with volumetric BMD (vBMD) of cortical and trabecular bone separately. Indeed, Boutroy et al.(2) showed that distal radius vBMD and microstructural parameters better discriminated 35 postmenopausal women with mixed fractures from 78 postmenopausal women without fracture than did aBMD of the hip or spine. Moreover, we recently found that relating bone strength indices at the ultradistal radius (UDR) to estimated impact loads in a fall simulated the age- and sex-specific pattern of forearm fracture incidence rates over life better than did changes in radius vBMD.(3) In addition, it is now possible to estimate bone strength directly with microfinite element (μFE) analysis at the UDR(4) and therefore to calculate the factor-of-risk (ϕ, the ratio of applied loads to bone failure loads) at the distal forearm in vivo.(5) Thus, as a guide to better understanding the structural determinants of fracture risk, the purpose of this preliminary report is to evaluate these diverse measures (BMD, bone geometry, bone microstructure, bone strength, and fall load to bone strength ratios) in a population sample of postmenopausal women with and without a prior distal forearm (Colles') fracture.
MATERIALS AND METHODS
After approval by Mayo Clinic's Institutional Review Board, subjects were recruited from an age-stratified random sample of Rochester, MN, women, as described elsewhere.(6) This analysis focused on the 248 postmenopausal women recruited in 2000 ± 1 yr, 98% of whom were white, reflecting the ethnic composition of the population. In conjunction with a follow-up visit, 214 of these women had an HRpQCT scan of the wrist (mean age, 68.1 ± 11.4 [SD] yr; range, 41–99 yr). Twenty-two of them had experienced a prior distal forearm fracture after age 50 yr that was caused by moderate trauma (i.e., a fall from standing height or less) from 1.9 to 35.8 yr previously (median age at fracture, 66 yr). However, four women were excluded because of prior fracture in both wrists (n = 2), inadvertent scanning of a fractured wrist (n = 1), or poor scan quality (n = 1). The remaining 18 women were age-matched to an equal number of study women with no history of a prior hip, spine, or wrist fracture after age 35 yr, although one had a torus (“buckle”) fracture of the contralateral forearm at 10 yr of age. Written informed consent was obtained from all subjects before participation in the study.
As previously described,(7) the nondominant wrist (for cases, the nonfractured wrist) was scanned using a high-resolution pQCT device (a prototype of the Xtreme CT; Scanco Medical AG, Bassersdorf, Switzerland) to assess vBMD (mg/cm3) at the ultradistal radius (UDR), the site of Colles' fractures of the wrist.(8) A 3D stack of 116 QCT slices was acquired at the distal end of the radius, using an effective energy of 40 keV, slice thickness of 89 μm, image matrix of 1024 × 1024 pixels, and pixel size of 89 μm per side. Trabecular vBMD was assessed for the central 50% of bone at the more distal (7–20 mm from the reference line in the joint space) of two scanning sites; cortical vBMD was determined at the more proximal site (48–55.5 mm from the reference line) after the cortex was identified by a surface detection program and the outer 10% excluded to avoid volume averaging artifacts.
Areal BMD (aBMD) of the “arms” region was measured on a whole body scan using DXA with the Lunar Prodigy instrument (Madison, WI, USA) and software version 6.10.029. In a subset of subjects ∼2.4 yr later, we also obtained a dedicated DXA wrist scan from this instrument.
From the HRpQCT data, we determined total cross-sectional area (CSA), endocortical area, cortical area, and cortical thickness. We also evaluated the polar moment-of-inertia (I), which assesses the distribution of bone material relative to the center of the bone cross-section, and the related section modulus, the moment-of-inertia divided by the maximum distance from the centroid to the bone edge. These variables reflect the ability of a tubular structure like the radius to resist bending and torsional loading.(9)
Analysis of the HRpQCT images for trabecular microstructure has been described and validated.(10–13) Briefly, bone tissue volume to total volume (BV/TV) is derived from trabecular vBMD. Because individual trabecular thickness cannot be correctly resolved because of partial volume effects, a thickness-independent structure extraction is used.(11) Trabecular number (Tb.N) is taken as the inverse of the mean spacing of the ridges.(12) Trabecular thickness is derived as BV/TV ÷ Tb.N, and trabecular separation (Tb.Sp) is derived as (1 − BV/TV) ÷ Tb.N, as in standard histomorphometry.(14) Tb.Sp.SD is the standard deviation of Tb.Sp. There is very high correlation (r2, 0.96–0.99) between the HRpQCT methodology and 28-μm resolution μCT.(13)
Indices of bone strength (i.e., structural rigidity) were also derived from the HRpQCT data as described in detail elsewhere.(3) Structural rigidity measurements combine the intrinsic mechanical behavior of bone material (i.e., elastic or Young's modulus [E]) with relevant cross-sectional geometric properties (i.e., CSA for compression and tension and moment of inertia for bending and torsion). Each bone voxel is assigned a unique value for E based on its vBMD, using a previously published regression between QCT-derived vBMD and the Young's modulus for human trabecular bone.(15) Axial rigidity (EA) reflects the resistance of bone to tensile or compressive loading and was computed as the product of the E of each bone voxel times its CSA, summing over all bone voxels in the cross-section. We used a polar I for the cortical and trabecular regions separately (with mean E values for trabecular and cortical bone) in a “two-compartment” model computation of torsional rigidity (EI), a measure of resistance to bending and twisting forces. Axial and torsional rigidities predict structural failure(16,17) but have not been defined relative to whole bone strength at the UDR.
Bone failure load at the UDR was calculated directly from μFE models, the most widely used computational method in engineering for structural analysis.(18,19) Conceptually, a complicated object is divided into a finite number of small and manageable pieces (i.e., elements), which are assigned appropriate material properties; the resulting set of mathematical equations is solved. We created μFE models from each HRpQCT dataset by converting each voxel into a cubic element. Two material properties are needed to describe the relationship between element stress and strain; these are the Young's modulus and Poisson's ratio (v), the ratio of transverse to longitudinal strains. Their values were set to E = 8.7 GPa and v = 0.3, respectively, such that the mean factor-of-risk (see below) for the 36 bones was 1.0. The models were subjected to simulated compression, with nodes at the proximal surface fully fixed and nodes at the distal surface constrained to move in an axial direction only. The models were solved using the element-by-element method(20) as implemented by in-house software,(21) from which whole bone strength was estimated.(22) Failure loads calculated from such μFE models correlated highly (r2 = 0.66) with compressive loads producing Colles' fractures in 54 cadaveric forearms.(22) The relative importance of the cortical and trabecular bone compartments was assessed by calculating the strain energy in the cortex as a fraction of total strain energy to estimate the load carried by the cortex.
Applied loads and factor-of-risk
For these estimates, we used the loading condition for the type of trauma most commonly associated with wrist fractures—a forward fall.(23,24) The load applied to the wrist was estimated from predicted impact forces on the upper extremity during a fall on the outstretched hand.(25) Individual height data were used to estimate subject-specific applied loads using the formula,
We assessed the ratio of fall load to overall bone strength, as determined by μFE, as an estimate of fracture risk (ϕ). Because the fall load is in the numerator and bone strength is in the denominator, higher values indicate increasing fracture risk.(26)
Bone variables were summarized using means and SDs. The Student's t-test was used to compare differences in means between fracture cases and controls. Z-scores for the cases were calculated using the means and SDs of the controls. Pearson correlation coefficients were used to evaluate relationships between key BMD, structure, and strength variables.
The relative risk of fracture associated with different bone parameters was estimated by ORs obtained from logistic regression models where case status was the dependent variable and BMD, geometry, microstructure, and strength (all per SD decrease) and the load to strength ratio (per SD increase) were the potential predictors. Univariate relationships were assessed first, and stepwise methods with forward selection and backward elimination were used to choose independent variables for the final models.
As an additional expression of fracture discrimination, the area under a receiver operating characteristic (ROC) curve was also obtained from the logistic regression models. All analyses were performed using SAS (SAS Institute, Cary, NC, USA) and Splus (Insightful, Seattle, WA, USA).
The 18 postmenopausal women with a history of Colles' fracture (median age, 82.1 yr) were matched to comparably aged control women without a prior Colles' fracture, proximal femur fracture, or thoracolumbar vertebral body fracture (median, 78.6 yr). Because the fracture case and control women had similar heights (mean, 159 ± 8 versus 159 ± 6 cm; p = 0.960), the estimated mean traumatic load on their distal radius, given a fall forward onto the outstretched arm, was the same (2645 ± 64 versus 2645 ± 52 N; p = 0.970). Therefore, only skeletal variables discriminated case from control women.
Thus, the forearm fracture cases had significantly lower total vBMD values at the UDR than did controls, and there was a greater percentage reduction in trabecular (−22%) than cortical (−10%) vBMD (Table 1). Z-scores were correspondingly reduced in cases compared with controls. There was no significant difference in aBMD of the “arms” region of the DXA total body scan which is also a cortical assessment. Similar results were seen for total radius aBMD (0.49 versus 0.55 g/cm2; p = 0.233) from a dedicated DXA wrist scan (not the total body scan) made on a subset of the subjects (10 cases and 11 controls) at a later date; a greater reduction in aBMD at the UDR site by DXA in the cases compared with controls (0.32 versus 0.38 g/cm2) did not achieve statistical significance (p = 0.095).
Table Table 1.. Comparison of 18 Rochester, MN, Women With a Distal Forearm Fracture (Cases) to 18 Age-Matched Community Women (Controls) With Respect to the Five Main Variable Categories
With respect to bone geometry, the fracture cases had slightly larger cross-sectional and endocortical areas but significantly smaller cortical area, cortical thickness, and polar moment of inertia (Table 1). The women with forearm fractures also had inferior values for most microstructural variables, which differed from controls by one half an SD or more (Table 1). In particular, their trabecular number was smaller and trabecular separation was greater. Variability in trabecular separation (Tb.Sp.SD) was also 36% greater among the cases, but this difference did not achieve statistical significance (p = 0.168) with the limited sample size.
Bone strength estimates in the wrist were about an SD lower in the women with a fracture (Table 1). For example, there was a 20% reduction in the overall radius failure load as estimated by μFE. However, the percentage of the load borne by the cortex was the same for cases and controls (63% in each instance), and the μFE-based bone failure load was best predicted by a linear combination of trabecular plus cortical bone volume (R = 0.90). When the μFE failure load was related to the estimated traumatic load in a fall for each subject, the applied load to failure load ratio (factor-of-risk, ϕ) was 24% higher (worse) among the fracture cases (Table 1).
Using multivariable logistic regression, we identified the strongest predictor of forearm fracture within each of the five main variable categories (Table 2). The relative fracture risks associated with these choices were as follows: BMD (total UDR vBMD: OR, 4.2; 95% CI, 1.4–12); bone geometry (cortical thickness: OR, 4.0; 95% CI, 1.4–11); bone microstructure (trabecular number: OR, 2.3; 95% CI, 1.02–5.1), and bone strength (axial rigidity: OR, 3.8; 95% CI, 1.4–10). The OR for each SD increase in ϕ, computed as the fall load divided by the μFE failure load, was 3.0 (95% CI, 1.2–7.5).
Table Table 2.. Relative Risk (Age-Adjusted ORs per SD Change) of a Distal Forearm Fracture Among Rochester, MN, Women
ORs relate to the area under an ROC curve and, because the ORs were of similar magnitude, areas under the ROC curves did not differ significantly for these parameters (range, 0.72–0.82) as delineated in Table 2. This is partly because of correlations among these variables (Table 3). For example, total UDR vBMD was highly correlated with cortical thickness (r = 0.82) and with trabecular number (r = 0.71; both p < 0.01). However, with the exception of trabecular number, each of these variables remained an independent predictor of forearm fracture risk after adjusting for “arms” aBMD in the logistic models.
Table Table 3.. Correlations Between Key BMD, Structure, and Strength Parameters at the Ultradistal Radius and “Arms” aBMD, Among 18 Rochester, MN, Women With a Distal Forearm Fracture
Other than gender and falling, it has proven difficult to identify strong risk factors for distal forearm fractures, most of which occur in relatively healthy individuals.(27) The most common type, Colles' fracture of the UDR,(28) almost always results from a fall forward onto the outstretched arm.(24) In an earlier study, we found that fall-related risk factors were associated with a 2.1-fold increased risk of Colles' fracture among women but, in aggregate, accounted for (attributable risk) <20% of all distal forearm fractures observed in the community.(27) Likewise, the forearm fractures observed here had all resulted from a fall, but only four of these women reported a fall in the year before assessment compared with seven control women. Moreover, there was no difference between cases and controls in the estimated load applied to the radius given a forward fall.
Consequently, explanations for fracture pathogenesis have turned to measures of bone strength. UDR aBMD is correlated with bone breaking strength ex vivo(29) and is predictive of subsequent wrist fracture risk.(1) However, aBMD overestimates the true vBMD of larger bones, cannot distinguish trabecular from cortical bone, and is unable to directly assess bone structure. It is also clear that a larger bone diameter is more resistant to bending forces(26) and therefore protective against forearm fractures,(30,31) whereas the rise in forearm fracture risk at menopause has been attributed to an increase in cortical porosity.(32) Despite some recent progress,(33,34) it is not entirely clear how these various factors interrelate. Thus, we sought insights from several new approaches to assessment and found a substantial number of BMD, geometry, microstructure, and strength variables in the forearm that are significantly associated with fracture risk.
Whereas “arms” aBMD was reduced by only 8% among the fracture cases, the reduction in cortical vBMD in the UDR was a similar 10%, the same as the reduction in radius aBMD from a dedicated DXA scan of the wrist in a subset of subjects. In contrast, a 22% reduction was seen in trabecular vBMD, and a 15% reduction in aBMD was observed at the UDR site in the DXA wrist scan. This is consistent with previous reports of a stronger association of forearm fractures with BMD assessed at predominantly trabecular sites.(35–39) Although the latter observation has led to the notion that distal forearm fractures stem mostly from trabecular bone loss, cortical bone strength makes an important contribution to forearm fracture risk as indicated by ex vivo studies.(40) In this analysis, cortical bone carried >60% of the load on the radius, and cortical thickness and area were 33% less and 29% less, respectively, in fracture cases compared with controls. The cortical thickness values were lower than may be expected from earlier pQCT devices but are consistent with other recent reports using HRpQCT.(2,41)
There was also evidence that Colles' fractures are associated with disruption of trabecular microstructure, as a 17% reduction in trabecular number and a 23% increase in trabecular separation were observed among the women with a forearm fracture compared with only a 6% decrease in derived trabecular thickness. The same observation was made for osteoporotic fractures generally by Boutroy et al.,(2) but they found that the greatest percentage change (26%) was in distal radius Tb.Sp.SD, thus documenting a more inhomogeneous bone structure in fracture cases. We saw a 36% increase in Tb.Sp.SD in the women with forearm fractures, albeit not statistically significant given the number of subjects; an even greater increase in Tb.Sp.SD was found by Sornay-Rendu et al.(42) in 34 women with forearm fractures. Disruption of trabecular architecture could explain an increase in fracture risk disproportionate to the loss of BMD.(26) This view is reinforced by the recent demonstration that reduced UDR vBMD in men (whose forearm fracture risk is lower than that in women) is accompanied by trabecular thinning rather than loss of trabeculae,(7) although men have other biomechanical advantages that also reduce their risk.(3)
Because bone strength by μFE is in the same units (N) as the estimated fall load, we were able to derive an absolute, as opposed to relative,(3) factor-of-risk estimate for forearm fracture. In engineering terms, ϕ > 1.0 indicates a progressive risk of structural failure.(26) Although no specific ϕ threshold has been established for Colles' fracture, the mean failure load estimated in vivo by μFE in the cases was 2488 N. This is consistent with failure loads of 2642–2852 N determined experimentally in cadaver forearms.(5,40,41) Given an impact load in a fall averaging 2645 N, the resulting value for ϕ was 1.11 among the fracture patients in this study. However, the mean value was 0.89 even among the controls, suggesting a substantial risk for fracture among many elderly women given the right circumstances.
It must be acknowledged that the μFE results, and subsequently the factor-of-risk determination, depend on the Young's modulus for bone tissue. Only limited data are available on Young's moduli in the distal radius. However, this is not of major concern in this analysis because the specific value chosen for the Young's modulus affects the absolute values for calculated bone strength but not the relative difference between cases and controls. We estimated Young's modulus by relating the μFE results to the estimated fall load and derived a value of 8.7 GPa, which is close to values used in μFE models for other sites in the human body. Using this value, we found that 83% of controls (15 of 18) had a value for ϕ < 1, whereas 67% of the cases (12 of 18) had a value for ϕ > 1, further indicating the potential for using ϕ to assess fracture risk.(26) In addition, the μFE analyses simulated an axial compression of the distal radius, which is thought to be an important loading mechanism leading to Colles' fractures.(29) Although bending could play a role, its effect was not simulated in the μFE modeling because of the large cross-section of the radius compared with the relatively small stack of bone measured with HRpQCT, which would not lead to meaningful results for bending.
This analysis was limited mainly by the relatively small number of postmenopausal women in our study population who had a prior distal forearm fracture, some of which had occurred years earlier. This restricted our ability to determine the “best” predictor, or set of predictors, of forearm fracture risk. A much larger prospective study now underway will address this question. For this analysis, we had available measures of BMD, geometry, microstructure, and strength. However, in an in vivo study such as this, we could not assess bone material properties.(43) These might play a role in Colles' fracture pathogenesis,(44) but their contribution will have to be defined ex vivo. In addition, the comparison with DXA is limited by the availability of dedicated DXA wrist scans only for a subset of the subjects. Finally, we estimated the applied load on the forearm by assuming a fall on the outstretched hand,(25) and there is good evidence that this is a characteristic orientation.(23,24) In the new forearm fracture study now underway, we will assess the exact circumstances of each fall so that traumatic loads can be estimated with greater precision.
Despite the limited number of forearm fractures available for study, we found significant associations with fracture risk for a diverse array of BMD, structure, and strength variables. Some of these may prove to be useful clinically if their predictive ability is confirmed prospectively. More important, however, is to better define how the different structure and density parameters interact to determine bone strength and resistance to fracture at the UDR under different stresses. In particular, the standard measurements assessed here may indirectly reflect a smaller number of key underlying parameters that actually distinguish fracture patients from controls. It would be important in future studies to identify these more fundamental determinants of bone strength and fracture risk, which might then be integrated to generate a reliable surrogate index for forearm fracture outcomes.
This work was supported by Research Grants R01-AR27065 and M01-RR00585 from the National Institutes of Health, U.S. Public Health Service. The authors thank Margaret Holets for the peripheral QCT measurements, Lisa McDaniel, RN, and Louise McCready, RN, for assistance in recruitment and management of the study subjects, James M Peterson for assistance with data management and file storage, and Mary Roberts for assistance in preparing the manuscript. The authors also thank Dr SK Boyd, Department of Mechanical and Manufacturing Engineering, University of Calgary, for providing the FAIM finite element solver.