In a population-based, cross-sectional study, we assessed age- and sex-specific changes in bone structure by QCT. Over life, the cross-sectional area of the vertebrae and proximal femur increased by ∼15% in both sexes, whereas vBMD at these sites decreased by 39–55% and 34–46%, respectively, with greater decreases in women than in men.
Introduction: The changes in bone structure and density with aging that lead to fragility fractures are still unclear.
Materials and Methods: In an age- and sex-stratified population sample of 373 women and 323 men (age, 20–97 years), we assessed bone geometry and volumetric BMD (vBMD) by QCT at the lumbar spine, femoral neck, distal radius, and distal tibia.
Results: In young adulthood, men had 35–42% larger bone areas than women (p < 0.001), consistent with their larger body size. Bone area increased equally over life in both sexes by ∼15% (p < 0.001) at central sites and by ∼16% and slightly more in men at peripheral sites. Decreases in trabecular vBMD began before midlife and continued throughout life (p < 0.001), whereas cortical vBMD decreases began in midlife. Average decreases in trabecular vBMD were greater in women (−55%) than in men (−46%, p < 0.001) at central sites, but were similar (−24% and −26%, respectively) at peripheral sites. With aging, cortical area decreased slightly, and the cortex was displaced outwardly by periosteal and endocortical bone remodeling. Cortical vBMD decreased over life more in women (∼25%) than in men (∼18%, p < 0.001), consistent with menopausal-induced increases in bone turnover and bone porosity.
Conclusions: Age-related changes in bone are complex. Some are beneficial to bone strength, such as periosteal apposition with outward cortical displacement. Others are deleterious, such as increased subendocortical resorption, increased cortical porosity, and, especially, large decreases in trabecular vBMD that may be the most important cause of increased skeletal fragility in the elderly. Our findings further suggest that the greater age-related decreases in trabecular and cortical vBMD and perhaps also their smaller bone size may explain, in large part, why fragility fractures are more common in elderly women than in elderly men.
Measurements of BMD have helped define the patterns of age-related bone loss that lead to fragility fractures and have contributed greatly to a better understanding of the pathogenesis of osteoporosis.(1) In most studies, BMD measurements were made at the lumbar spine and hip using DXA.(2) It has become clear, however, that DXA has major limitations in assessing the structural mechanisms that lead to decreased bone strength. In particular, DXA measures only areal BMD and, because wider bones are also thicker, overestimates substantially the BMD of larger bones, significantly confounding the interpretation of age- and sex-related changes.(3) Attempts to correct areal BMD data for differences in bone size have used empirical corrections or theoretical corrections of idealized bone geometries.(4) Both approaches are oversimplifications of the underlying structural relationships. Clearly, the size, shape, and structure of bone have major effects on bone strength, but these components of so-called “bone quality” are not well captured by DXA.(5) Beck et al.(6,7) have developed a method called hip structural analysis (HSA) that uses areal BMD measurements made by DXA at the femoral neck to estimate cross-sectional geometry and indices of bone strength, but this approach involves many unproved assumptions and provides only rough approximation of these structural indices.(8) Indeed, there is evidence(9) that direct measurement of hip BMD is as effective as HSA in predicting fracture risk.
In the present population-based, cross-sectional study, we have used central and peripheral QCT and novel image analysis software to assess age- and sex-specific changes in volumetric BMD (vBMD), bone size, geometry, and structure at the lumbar vertebrae, femoral neck, distal radius, and distal tibia, the sites of four commonly occurring fractures associated with osteoporosis.(10) Using these data, we have attempted to resolve three important, but incompletely understood, questions relative to the pathogenesis of age-related osteoporosis. First, what are the age-related changes in vBMD and bone structure and geometry that are associated with increased bone fragility in the elderly? Second, how do these age-related changes differ between sexes and are these differences consistent with the greater fracture risk in elderly women compared with elderly men? Finally, when do age-related losses of cortical and trabecular bone begin and, in women, what is the relationship of the onset of bone loss to menopause?
MATERIALS AND METHODS
We recruited subjects from an age-stratified, random sample of Rochester, MN, residents who were selected using the medical records linkage of the Rochester Epidemiology Project.(11) This community is highly characteristic of the U.S. white population but underrepresented with respect to blacks and Asians. The sample spanned ages from 20 to 97 years and included 373 women and 323 men. Reflecting the ethnic composition of the community, 96% of the men and 99% of the women were white. Thirty-two percent of the men and 29% of the women were obese as defined by a weight of >30% of the ideal weight for their height. Ninety-four postmenopausal women were receiving estrogen therapy and six postmenopausal women and three men were receiving bisphosphonate therapy for osteopenia. Because of the large number of postmenopausal women receiving estrogen at the time of recruitment, we oversampled in the 50- to 69-year age range to have adequate numbers of untreated women for analysis. There was an offsetting undersampling of young adult women and men.
QCT of the central skeleton
Single-energy CT scans were made at the lumbar spine and proximal femur with a multi-detector CT scanner (Light Speed QX-I; GE Medical Systems, Wakesha, WI, USA), using a tube potential of 120 kVp, tube current of 80 mA, rotation time of 0.8 s, table speed of 7.5 mm/rotation, detector collimation of 4 × 2.5 mm, and pitch of 0.75. Data sets were obtained in 20 s for the spine and in 20 s for the hip, eliminating significant motion artifacts. For the lumbar vertebrae, we analyzed a single slice obtained at the midportion of the L1, L2, and L3 vertebrae, and a mean of these values is given in Table 1. For the femoral neck, we assessed a single reformatted oblique section contiguous at the mid-portion of and orthogonal to the femoral neck, between the superior aspect of the head of the femur and the inferior aspect of the inferior trochanter. For all scanning sites, slice width was 2.5 mm and the in-plane voxel size was 0.74 mm. Calibration standards scanned with the patient were used to convert CT numbers directly to equivalent vBMD (mg/ml3).(12) To validate our image processing algorithm, we made 10 scans of the European spine phantom (ESP), which is composed of hydroxyapatite.(13) The correlation between BMD results determined by our algorithm and that of the ESP was r = 0.998.
Table Table 1. Central QCT Measurements at the Lumbar Spine and Proximal Femur
To study age- and sex-specific structural changes in bone mineral distribution and structure, we developed software for the analysis of bone structure, geometry, and volumetric density from the CT images.(14) The CT data are treated as a 3D volume and are reformatted in an orientation that is perpendicular to the primary loading forces on the bone in the region of interest. On this plane, the software program automatically places a single image line that extends from the centroid of the bone to outside the periosteal surface, and this line is rotated about its centroid end in 3° increments. From the range of gray levels in the image, the cortex is identified as the maximal brightness found on this line. The full-width half-maximal points on either side of this maximum, which are defined above, are interpreted as the periosteal and endocortical boundaries of the cortex. The resulting points are joined to create 2D cortical and subcortical regions. The BMD and area of the two regions are measured independently, and a circular central trabecular region is also measured.
Single-energy CT scans were made at two scanning sites in the distal radius and at two scanning sites in the distal tibia using the Densiscan 1000 (Scanco Medical AG, Bassersdorf, Switzerland), as previously described.(15) All scans had a slice thickness of 1.5 mm and an in-plane voxel size of 0.35 mm for the radius and 0.45 mm for the tibia. From a digital image (scout view) of the lower forearm and lower leg, the joint space is visualized and a reference point is set electronically at the intersection of the joint space with radius-ulnar junction for the forearm and the tibial-fibular junction for the distal leg. From this line, an automated program then selects a more distal and a more proximal scanning site at both the distal radius and distal tibia. For the distal radius, the more distal of the two scanning sites (termed Rad-D) was located 7–20 mm, and the more proximal scanning site (termed Rad-P) was located 48–55.5 mm from the reference line. For the distal tibia, the more distal scanning site (termed Tib-D) was located 20–33.5 mm, and the more proximal scanning site (termed Tib-P) was located 63–70.5 mm from the reference line. Ten consecutive slices were made at the Rad-D and Tib-D sites, and six consecutive slices were made at the Rad-P and Tib-P sites. A surface detection algorithm delineates the bone that then is peeled pixel-by-pixel until core areas of 90%, 70%, and 50% remain. The outer 10% of bone is excluded to avoid partial volume effects from the bone edge. The 70–90% cross-sectional area is completely inside cortical bone within all normal subjects below the age of 75 years, and the inner 50% contains only trabecular bone in all subjects.
All results were analyzed with and without the inclusion of 94 postmenopausal women who were receiving estrogen therapy and 9 subjects who were receiving bisphosphonate therapy. Because there was no significant difference in any of the results with their exclusion, we retained them in the final analysis. The relationships between bone mass/geometry/density and age were studied using Pearson correlation and linear regression, where age was modeled using natural splines. Each model was compared with a linear relationship, and the simplest model was used for analysis. Changes in variables between ages 20 and 90 years were based on predicted values from these models. Differences in changes over age between men and women were tested using an age-sex interaction term in a regression model. An adjustment for the effect of differences in bone size on bone mass and area was made by dividing the original values by height; the need for this adjustment was determined by analyzing the correlation of different variables with height among subjects 20–35 years of age. The Student's t-test was used to compare the mean values for young males versus young females. S-plus function lowess,(16) essentially a type of moving average, was used to explore the data in Figs. 1, 2, 3, 4, and 5.
Based on our measurements in young adults 20–29 years of age, we estimated the composition of bones at central scanning sites before age-related bone loss begins to be >80% trabecular at the vertebrae and ∼65% trabecular/∼35% cortical at the femoral neck. For the peripheral scanning sites, we estimate the composition at the distal radius scanning sites to be 67% trabecular and 33% cortical at Rad-D and >95% cortical at Rad-P. For the distal tibia scanning sites, these values were 78% for trabecular and 22% for cortical at Tib-D and >95% for cortical at Tib-P.
The age- and sex-specific changes for the QCT measurements at central (lumbar spine and proximal femur) sites are described in Table 1, and those for the peripheral (distal radius and distal tibia) sites are described in Table 2. Based on regression analysis between height and each dependent variable in women or men between 20 and 29 years of age, several variables relating to bone area required adjustment and, for these, both unadjusted and adjusted values are given in Tables 1 and 2. These provide means and SD for the absolute values for variables in women and men who are 20–35 years of age. They also provide the absolute (in units of the variable) and relative (percentage) changes with age between 20 and 90 years as well as the significance of the age regression and differences between women and men. Individual values for some variables are shown in Figs. 1, 2, 3, 4, 5, and 6 and are color-coded to distinguish among men, premenopausal women, and postmenopausal women.
Table Table 2. Peripheral QCT Measurements at the Distal (D) and Proximal (P) Scanning Sites of the Distal Radius and at the Distal (D) and Proximal (P) Scanning Sites of the Distal Tibia
At both central and peripheral sites, young adult women had a bone cross-sectional area that was 25–33% less (p < 0.001) and a bone mass that was 18–21% (p < 0.001) less than young adult men. These relative differences were largely maintained over life (Fig. 1). At central sites, total cross-sectional bone area increased over life by 7–15% (p < 0.001), and this increase served as an index of periosteal apposition. The increases were not statistically different between sexes (Fig. 1). At the peripheral sites, total bone area increased over life by 2–16%. The increase was more in men than in women at the Rad-D scanning site but was similar at the other three peripheral scanning sites.
There were large decreases in trabecular vBMD over life (p < 0.001) which seemed to begin before middle life, especially at central sites. These decreases were greater in women (∼55%) than in men (∼45%, p < 0.001) at central sites but were similar (24% and 26%, respectively) at peripheral sites. At the vertebrae (Fig. 2), there was an apparent small midlife acceleration in the slope of the decrease in women that accounts for much of their significantly greater decrease in vertebral vBMD over life compared with men. Endocortical resorption, as estimated by the expansion of the total medullary area, also began to increase before middle age and continued throughout life (Fig. 3). The increases in marrow expansion with aging in women was 38%, and in men was 25%, but these were not significantly different from each other. With aging, the apparent rate of periosteal apposition (estimated by the increase in bone diameter) was less than that for endocortical resorption. Consequently, cross-sectional cortical area decreased slightly with aging at the femoral neck, and the decrease was similar between sexes (Fig. 4). These changes displaced the cortex outward from the central axis.
There were substantial decreases in cortical vBMD with aging at both the femoral neck and at the peripheral scanning sites. These decreases show changes in vBMD at the Rad-P scanning site, which are composed almost entirely of cortical bone (Fig. 5). There was little decrease in vBMD at these sites until midlife in either women or men. Thereafter, there were linear decreases in both sexes, but the decreases were greater in women (28%) than in men (18%, p < 0.001). The role of menopause in inducing this loss and the lack of interaction with weight-bearing are shown in Fig. 6, in which decreases of vBMD in women at the Rad-P scanning site of the distal radius and at the Tib-P scanning site of the distal tibia, both sites containing largely cortical bone, are plotted as years since menopause. At both sites, vBMD decreased linearly (p < 0.001) and to a similar degree, although the Rad-P site is non-weight-bearing and the Tib-P site is weight-bearing.
Although skeletal QCT measurements have been available for almost 30 years, recent innovations in instrumentation and software have improved resolution and precision.(12,17) Nonetheless, only a few large studies on skeletal changes with aging have been previously reported using QCT,(18–24) and only two of them assessed changes in men.(20,22) These earlier studies were largely restricted to measurements of total and central trabecular vBMD. There is very little information on population changes in cortical vBMD. This lack likely relates to the difficulty in defining the endocortical border accurately on cortical structures ≤2.5 mm wide (e.g., the femoral neck), especially in older subjects with low BMD values.
In this study using spiral QCT, we used our newly developed image analysis program(14) to quantify changes in the width and vBMD of the proximal femoral cortex with an accuracy superior to commercially available software (data not shown). There was an excellent correlation between the values that we obtained from scans of the European Spine Phantom (composed of hydroxyapatite) and the actual hydroxyapatite values. We did not evaluate the vertebral cortical rim, which was too thin to measure accurately.(25) We also concurrently assessed changes in vBMD, structure, and geometry at the distal radius and tibia by peripheral QCT. The instrument used had a precision of <0.5%, had a pixel size small enough to measure cortical structures accurately, and was capable of detecting individual rates of loss in postmenopausal women within 1 year.(19) Although age-related changes of bone at the distal radius have been studied using peripheral QCT,(19,26) this is the first population-based study to concurrently apply central and peripheral QCT measurements to assess the mechanisms of the development of osteoporosis.
Although the importance of bone loss to the pathogenesis of fragility fractures in the elderly is broadly accepted,(3) a full understanding of the pathophysiology of osteoporosis also requires assessment of pubertal changes that establish the characteristics and sex-specific differences of the adult skeleton. We find that young adult women have a substantially smaller cross-sectional bone area and bone mass at the lumbar spine and proximal femur, consistent with previous reports.(27) Smaller bones are less strong than larger bones, even when there is an equivalent vBMD. However, we also found that, in young adulthood, women have average values for vBMD of total, trabecular, and cortical bone at the lumbar spine and proximal femur that were ∼10% higher than in the young adult men, but this was more variable at the peripheral scanning sites. Because young women are shorter and lighter than young men, their smaller bones may be appropriate for their lesser skeletal loading. Nonetheless, the decreased bone size persists into old age and may predispose elderly women to a greater fracture risk than elderly men after substantial age-related bone loss has occurred. A detailed analysis relating mechanical loading to estimates of bone strength will be required to evaluate this possibility quantitatively.
There are three major age-related processes that lead to bone loss, all of which were evaluated in this study. The first and most important is trabecular bone loss: vBMD of trabecular bone decreased by about one-half with aging at central sites but by only about one-quarter at peripheral sites. The decrease in trabecular vBMD is caused both by thinning of trabeculae and, especially in early postmenopausal women, by disruption of trabecular microstructure and loss of trabecular elements.(28,29) Unfortunately, our measurements did not have the resolution to assess these mechanisms separately. The second process is continued net resorption at the endocortical surface. Bone loss over life from this process, as assessed by expansion of the marrow space, was about 25–40% at the femoral neck and distal radius but was less at the distal tibia, perhaps because of the effect of weight-bearing. Over the 70-year age range of our subjects, this process would result in an estimated cortical bone resorption rate of about 0.5%/year, which is similar to that reported from longitudinal density measurements at the metacarpal cortex in placebo-treated postmenopausal women.(30) The third process contributing to bone loss is a decrease in cortical vBMD. Because the density measurements were volumetric, they were theoretically independent of changes in cortical area. Because the density of interstitial bone does not decrease appreciably with aging,(31) these decreases are almost certainly caused by increased porosity from both an increase in resorption cavities(32,33) and an accumulation of incompletely closed osteons with aging.(31) It is also of interest that the slopes for decreases in cortical vBMD after menopause were identical at the distal radius, a non-weight-bearing bone, and at the distal tibia, a weight-bearing bone, indicating that the effect of estrogen deficiency is dominant over the effect of mechanical loading, at least in this circumstance.
Periosteal apposition increased bone size from 20 to 90 years of age by ∼15% at the various measurement sites. However, the finding that cortical area and width decreased slightly over life indicates that the gain in cortical bone caused by periosteal apposition failed to completely offset the adverse effect of cortical bone loss caused by endosteal resorption. Because the periosteal perimeter is more extensive than the endocortical perimeter, this implies that the rate of endocortical resorption over periosteal apposition may be even more than we observed. As a result of opposing effects of net bone formation on the periosteal and net bone resorption on the endocortical surfaces, the cortex was displaced outwardly with increasing age. Both the increase in bone cross-sectional area and outward displacement of the cortex enhance bone strength, because the bending strength of bone is proportional to its area moment of inertia, a geometric property that is influenced much more profoundly by periosteal than by endocortical apposition.(5) Determination of the relative quantitative effects of periosteal apposition and bone loss with aging on fracture susceptibility must await further studies in which biomechanical indices of bone strength are measured directly.
We next evaluated whether the various age-related changes in vBMD and bone structure differed between sexes. We found that the substantial increases in cross-sectional area did not differ between sexes at central sites but were more in men than in women at three of the four peripheral scanning sites. Women lost more trabecular bone than men did at the vertebrae and more trabecular and cortical bone at the femoral neck and at most, but not all, peripheral sites. Contrary to conventional wisdom, the excess loss over life in women as expressed in the equation (total loss in women-total loss in men/total loss in women) was 2- to 3-fold greater for cortical bone than for trabecular bone. This suggests that the high differential for trabecular versus cortical bone loss observed during the early postmenopausal interval(19,34) may not be maintained long term in postmenopausal women.
Duan et al.(8,35) have reported that age-related increases in cross-sectional area of the vertebrae and femoral neck were 3-fold more in men than in women, but that the rates of bone loss with aging were similar in both sexes, just the opposite of our findings. Previous studies on sex-specific differences in periosteal apposition with aging have been made only at sites in the peripheral skeleton and have given conflicting results, showing greater increases in men,(36,37) no difference between sexes,(38) or greater increases in women.(39) The reason for the disparity between our results and those of Duan et al. at central sites is unclear but probably is methodological. They used DXA measurements and determined bone geometry indirectly using algorithms, whereas we used QCT to make direct 3D measurements.
There continues to be uncertainty about the time of onset of age-related skeletal changes. In contrast to prevailing belief, our data strongly suggest that trabecular vBMD begins to decrease before middle life at all scanning sites, but most prominently at central sites. However, this finding is consistent with earlier longitudinal DXA measurements(40) and with the cross-sectional single- and dual-energy QCT measurements at the lumbar spine in men.(20,24) Cortical bone is lost both by increases in net endocortical resorption and by increases in porosity. We find that that the total marrow space, a surrogate for the rate of endocortical bone resorption, also begins to expand before middle age at femoral sites in both sexes, although this is less definite than the onset of trabecular bone loss. In contrast, cortical porosity, as assessed by decreases in cortical vBMD, either commences or accelerates in middle life. Thus, this process has a later onset than the other two and, in women, is closely associated with menopause. This pattern was evident both in the central and peripheral skeleton but was more prominent at the peripheral skeleton, owing to the greater precision of the measurements.
The mechanism for the onset of decreases in cortical vBMD in the perimenopausal interval in women almost certainly is the increase in bone turnover induced by acute estrogen deficiency, which leads to increased cortical porosity.(41) In the male cohort, there was also a decrease in cortical vBMD in men beginning in mid-life, although the slope of the decrease was less than in women. Although men do not experience the equivalent of menopause, their circulating levels of biologically available sex steroids fall progressively with aging and lead to increases in bone resorption and bone loss in those with lower circulating levels.(42,43) There was a suggestion in some plots that the progressive age-related decrease in trabecular vBMD accelerates during the perimenopausal interval, but this is much less apparent than in cortical bone. However, menopausal effects on bone loss are notoriously difficult to show in cross-sectional observational studies because decreases in ovarian secretion begin well before onset of amenorrhea and may continue for several years thereafter. Thus, further studies are needed.
If confirmed by longitudinal studies, the demonstration that decreases in trabecular and cortical vBMD begin at different ages and have different patterns of change with age would call for a re-examination of the relationship between sex steroid levels and bone loss. The onset of trabecular, and possibly also endocortical, bone loss in young adulthood of both sexes when serum sex steroid levels are, by definition, “normal” cannot be caused only by menopausal-induced estrogen deficiency. One possibility is that bone surfaces in contact with marrow are genetically programmed to favor net bone resorption. If so, trabecular bone loss would inevitably commence after completion of the pubertal skeletal growth. Another possibility is that sex steroid production in women may in fact decrease well before the menopause(44,45) and lead to bone loss. Clearly, trabecular bone loss is at least partially caused by estrogen deficiency in both sexes. This is most clearly shown by the large losses of trabecular bone occurring in premenopausal women after ovariectomy,(34) but also is shown by the direct correlation of serum biologically active estradiol levels with bone loss at the lumbar spine in aging men.(46) In contrast, we found that, in women, substantial decreases in cortical vBMD did not occur until the perimenopausal period and were linear when plotted as a function of years since menopause, suggesting that the predominant mechanism was estrogen deficiency.
Our study has several limitations. The most obvious one is that we are projecting rates of bone loss from cross-sectional data. One problem with doing this is the secular increase in stature that has occurred during the last century. We have attempted to adjust by using height as a covariate for those variables where a relationship could be shown in young adulthood. Despite these concerns, Melton et al.(47) showed in a population sample of women that there was good correspondence between the rate of bone loss at the femoral neck estimated from cross-sectional data and that determined directly from longitudinal data over a 16-year observation period. Also, because our population sample was almost completely white, extrapolations of our data to changes in other racial and ethnic groups cannot be made. Another limitation is the effect of volume averaging on cortical vBMD at central sites. Because of the larger pixel size of the spiral CT relative to the thickness of the femoral cortex, the apparent vBMD is reduced. Thus, it is reassuring that the patterns of change in cortical vBMD at central sites were very similar to those observed at peripheral sites where volume averaging is less of a problem. Finally, there is the theoretical possibility that single-energy CT would be affected by increases in the fat content of the marrow with aging.(48) However, reference values for changes in vertebral vBMD over life have shown good agreement between single- and dual-energy CT.(24) Also, our data for decreases in vertebral vBMD with age in women are similar to those reported for measurements with lateral DXA(49) and with dual-energy QCT,(24) but are somewhat more than histological values for trabecular bone volume reported from an autopsy study.(50)
In conclusion, age-related changes in bone are complex. Some have beneficial effects on bone strength, such as periosteal apposition with outward cortical displacement, whereas others have deleterious effects, such as increased endocortical resorption, increased cortical porosity, and especially, large decreases in trabecular vBMD, which may be the most important cause of increased skeletal fragility in the elderly. Our findings further suggest that the greater age-related decreases in trabecular and cortical vBMD and perhaps also a smaller bone size at the end of puberty in women compared with men explain, in large part, why fragility fractures are more common in elderly women than in elderly men. Last, current paradigms attribute bone loss with aging mainly to the menopause in women and to the slow decrease in levels of serum biologically active sex steroids after mid-life in men.(41,42) Our finding that trabecular vBMD begins to decrease before middle age and proceeds linearly in both sexes, whereas cortical vBMD begins to decrease at menopause, suggests that these paradigms may be incomplete and require further study.
We thank Margaret Holets for making the peripheral QCT measurements and Lisa McDaniel, RN, and Louise McCready, RN, for assistance in recruitment and management of the study subjects. Bruno Koller (Scanco, Inc.) provided substantial advice regarding the peripheral QCT measurements. Ronald A Karwoski and Mahlon C Stacy (Biomedical Imaging Center) provided valuable assistance in analysis of the spiral CT scans. This work was supported in part by National Institutes of Health Grants AR-27065 and M01 RR00585.