The Skeleton, Race, and Ethnicity

Differences in fracture risk between different racial and ethnic groups have been noted and investigated for many years. Despite this, our understanding of the basis for these differences is still incomplete. A full explanation will undoubtedly provide important insights about the etiology of osteoporosis, so additional research is clearly relevant.

In this issue of JCEM, two interesting papers from the Study of Women’s Health Across the Nation (SWAN) add to our knowledge base on these topics. Finkelstein et al. (1, 2) examined bone mineral density (BMD) and bone turnover markers in a racially and ethnically mixed sample of more than 2000 pre- and perimenopausal women. Data are provided separately for Chinese and Japanese women living in the United States in addition to African American and Caucasian women, which allow a direct comparison between these four groups. More notably, these studies avoid problems associated with treating Asians as a single group rather than distinct groups with different cultural backgrounds, lifestyles, and fracture rates (3, 4). Various interesting findings regarding the rank order of the different race/ethnic groups and the effect (or lack thereof) of the different explanatory factors emerged. Among the most striking findings was that ethnicity per se was not associated with differences in lumbar spine BMD in the three nonwhite groups examined. Furthermore, all three nonwhite groups had higher lumbar spine BMD than Caucasians of comparable body weight, which is noted as being consistent with observed racial and ethnic differences in vertebral fracture rates. Interestingly, even after accounting for size differences, the patterns of femur neck BMD did not mirror fracture patterns at this skeletal site quite as well, although the discrepancy was smaller than when unadjusted BMD values were compared between groups. In addition, ethnic patterns of bone marker levels did not parallel ethnic patterns of BMD, leading the authors to conclude that other factors are likely responsible for much of the ethnic variation in adult BMD.

A few topics were curiously omitted from the two SWAN studies. First, for reasons not stated, the authors did not include data from their Hispanic sample. This is disappointing because it would have extended the comparison to include this important ethnic group, which has now drawn equal with blacks in population size in the United States (5). Second, as noted above, the authors speculated that differences in bone turnover may play a role in observed BMD differences between the groups, but they did not address this question directly by including marker levels in their regression analyses of BMD. Hopefully these omissions will be addressed in future publications from this research group.

With these minor quibbles aside, the two studies from SWAN are well done and provide some interesting insights regarding possible mechanisms for the differences in fracture rates between race/ ethnic groups. They also provide the opportunity to consider some of the general issues that can arise when performing research on racial and ethnic differences in skeletal status. One of the thornier problems affecting research in this area is confounding caused by differences in bone size between the groups. In specific, areal BMD measurements are based on only two of the three dimensions of bone, namely length and width. Because they do not include bone depth, they are only partially adjusted for bone size. The SWAN study on BMD is commendable because the authors used several strategies to try to address this issue. A review of these strategies is useful in illustrating the complexity of the problem as well as suggesting possible targets for improvement in future research.

First, bone mineral apparent density (BMAD) was calculated. BMAD is an estimate of volumetric BMD made from the two-dimensional BMD measurements that are made by dual-energy x-ray absorptiometry. Specifically, it is the ratio of bone mineral content (BMC) divided by an estimate of the bone volume, which is calculated from the two-dimensional measurement of bone area after making certain assumptions about the geometric proportionality for different bones (6, 7). BMAD is intended to address the confounding effect of bone size on areal BMD measurements, but its value remains unclear. For example, Nevill et al. (8) recently suggested that BMAD may not address bone size differences appropriately in circumstances where the groups being compared differ in other factors that are known to affect BMC, such as age or body mass. More importantly, estimates of BMAD have not been found to improve fracture prediction over that obtained when BMD is used (9, 10). Thus, it does not seem likely that volume density constructs provide a means for normalizing important mechanical differences between bones of different size.

The problem of size differences between the race ethnic groups was also addressed in the SWAN study on BMD by repeating the comparisons in women who weighed less than 70 kg. The sample was restricted to this weight range because the majority of the Asian women in their sample had body weights below this level. However, slightly less than half of the white women in the SWAN sample had body weights below this value, and less than a quarter of the black women fell into this range. Thus, Asian women representing the entire distribution of weight were being compared with black and white women who represented only the lower portion of their respective weight distributions, and who may have differed from the majority of the white or black women who fell above this weight cutpoint in important ways. The authors did make exclusions for conditions that could affect BMD when enrolling their sample and also adjusted for several possible confounders in their multivariate models. These strategies likely helped to reduce possible biases in this comparison, but probably cannot completely rule them out.

Finally, Finkelstein et al. (1, 2) adjusted for body weight in their regression analyses of BMD and BMAD. Adjusting for differences in overall body size, such as standing height, weight, or body mass index, is frequently done in studies of racial and ethnic differences, but this strategy may not completely control for all relevant aspects of size differences between groups. For example, body proportions differ between the different race/ethnic groups (e.g. when compared with whites, blacks have longer legs and narrower pelvises, whereas Asians have shorter legs than whites) (11). Some of these differences in body proportions may have important biomechanical consequences in terms of the loads placed on the skeleton and on the adaptation to those loads (12). Furthermore, controlling for overall body size may not completely control for all relevant differences in bone size. We previously found that adjusting for body size removed or reduced areal BMD differences at the femur between men and women, but it did not eliminate gender differences in bone widths or cortical thickness (13). Thus, estimates of overall body size may not necessarily function well as surrogates for bone size (11).

Thus, the authors used all strategies to address size differences that were available to them, but it is not clear whether these strategies completely removed confounding due to size in the SWAN BMD study. More importantly, the focus needs to be broadened beyond bone density to identify differences in bone structure and geometry that may affect bone strength. Areal BMD provides information about the average amount of bone in a particular skeletal region but does not indicate its structural distribution, and so by itself may not always provide an accurate or complete assessment of bone strength. For example, we found that femur neck BMD declined with age, but bone strength, as reflected by the section modulus, was maintained via periosteal expansion despite BMD loss (14). Furthermore, bone strength may vary due to differences in geometry (e.g. in cortical thickness or bone width), even when areal bone density is the same. Beck et al. (15) recently found that size-adjusted BMD did not differ between stress fracture cases and noncases in males, but did in females. The reasons for these contradictory findings became clearer when bone structure was examined. Fracture cases of both sexes had lower section moduli than noncases. But size-adjusted BMD values in males did not differ by fracture status because cases and noncases had similar cortical thickness values. Nonetheless, male cases had lower section moduli because their bones were narrower. In contrast, female cases had bones of similar diameter but with thinner cortices, which resulted in lower values for both BMD and section modulus.

Finally, the structural basis for BMD differences between racial and ethnic groups may vary depending on the skeletal site, which in turn might partially explain why race/ethnic differences in fracture rates appear to vary in magnitude at different skeletal sites (16). For example, Nelson et al. (17) found higher bone densities in postmenopausal African American women at the femur neck, not only because there was more bone within the subperiosteal envelope (i.e. higher BMC), but also because the subperiosteal envelope itself was smaller due to a narrower bone. In contrast, the density advantage of these black women at the femur shaft was due solely to a greater amount of bone tissue within a similarly sized diameter. This variability in the structural differences between skeletal sites is consistent with the variation in fracture risk for blacks vs. whites noted for the upper and lower femur by Baron et al. (16).

Widening the focus beyond BMD to consider structure will likely advance the understanding of racial and ethnic differences in fracture risk. But doing so may not be easy at the moment because most bone densitometry methods used in clinical settings or epidemiological studies do not routinely provide estimates of bone geometry (18). More work is needed to develop automated, simple methods to assess bone structure.

Examining differences in bone turnover between racial and ethnic groups also presents certain complexities. First, a choice of markers must be made. There are a large number of markers available, but within the general categories of formation or resorption, it is not completely clear yet which markers are best suited for which purpose. Finkelstein et al. (1, 2) made reasonable choices in their selection of markers, but it is possible that their findings would have been different if different markers had been used. For example, two of the more widely used bone formation markers, osteocalcin and bone-specific alkaline phosphatase, are believed to reflect different aspects of osteoblastic activity (19), which may possibly explain why many studies have found lower osteocalcin levels in whites than in blacks (20–24) but no differences in bone-specific alkaline phosphatase between these races (23, 25–27). Another complexity results from the unclear nature of the relationship between areal BMD and bone turnover markers measured at a single point in time. For example, Finkelstein et al. (1, 2) hypothesized that bone turnover markers might explain racial and ethnic variation in BMD at the spine and hip. This expectation assumes a direct relationship between markers and BMD, but results from studies that have directly tested this have been mixed (28). One possible explanation offered for these inconsistent results is that turnover markers reflect global skeletal activity, whereas most BMD measurements only assess a very small portion of the skeleton (29). Furthermore, the inability of BMD measurements to completely capture bone size and structure may confound the ability to relate bone turnover to bone mass. For example, differing rates of bone turnover on the endosteal and periosteal surfaces of bone could increase the size of the bone via periosteal apposition. An increase in bone size would reduce BMD or BMAD even if the amount of BMC remained constant, because bone size is the divisor in the calculation of BMD or BMAD. More importantly, the biomechanical effect of altering the distribution of bone mass within a bone of fixed size differs from that in a bone that is expanding, but the difference cannot be readily extracted from BMD or BMAD.

Other general barriers exist as well. For example, cross-sectional studies such as these two studies from SWAN cannot definitively answer why fracture rates differ between racial and ethnic groups because the intermediate outcomes they measure cannot be linked directly to fracture. However, longitudinal data with fracture outcomes for nonwhites have been difficult to obtain given the lower fracture rates and younger average age of many nonwhite groups in the United States (4, 30–32). In the absence of longitudinal data, less direct methods must be used when trying to evaluate whether intermediate outcomes such as BMD or bone markers can explain racial and ethnic differences in fracture. Thus, Finkelstein et al. (1, 2) examined the extent of congruence between their results for the rank order of BMD or bone turnover markers in the different race/ethnic groups with the rank order of fracture rates in these groups obtained in other studies. This type of comparison presumes that the rank order of fracture rates by race or ethnicity is well established, but this is not necessarily true for all skeletal sites or nonwhite groups in the United States. For example, data on spine fracture among nonwhites in the United States are scanty. Jacobsen et al. (33) published estimates for blacks and whites using hospitalization data from Medicare, but because many vertebral fractures are asymptomatic or do not require hospitalization, these data are potentially confounded by differences in health care seeking and access (34). Estimates of vertebral fracture for Chinese Americans do not seem to be available. Data on vertebral fracture are available for Japanese Americans from community-based studies, primarily among those living in Hawaii. Interestingly, Ross and Huang (35) recently speculated that the prevalence of certain risk factors for osteoporosis may be lower in Hawaii because hip fracture rates for Caucasians in Hawaii were much lower than those reported for Caucasians on the mainland. It is unclear whether a similar situation exists for vertebral fracture, but in light of this possibility, it would be useful to obtain vertebral fracture data for Japanese Americans living on the mainland. Data are available for Chinese and Japanese living in their home countries, but these may not accurately reflect rates in those who have immigrated to the United States (36).

Data on hip fracture among nonwhites are somewhat more available, but also suffer from limitations. Questions exist about the accuracy of racial and ethnic group coding in national databases such as Medicare or the National Hospital Discharge Survey (37–39). Community-based studies may avoid coding problems, but their results may not represent other samples. Finally, many of the studies on hip fracture rates by race or ethnicity use broad groupings when reporting data for nonwhite groups (e.g. Hispanic or Asian), which may obscure important differences in hip fracture rates between different Hispanic and Asian subgroups (4, 31).

On the other hand, it may be difficult to decide which of the many ethnic subgroups in the United States should be studied. To date, the United States Census Bureau has reported separate population estimates for 24–25 different Hispanic or Asian subgroups and 44 different Native American tribes from the 2000 Census, and has indicated that data on additional Native American tribes will be forthcoming (40–42). The advent of a multi-racial category on the United States Census (5) will likely further increase the number of potential subgroups as well as the complexity of defining ethnic and racial groups in general.

In the end, then, while these two studies add to our progress in understanding racial and ethnic differences in fracture, many questions still remain. Answers to many of these will likely require longitudinal data and further advances in our ability to measure bone size, structure, and turnover. Better data on fracture rates in nonwhites are also needed. In the meantime, well-designed cross-sectional studies such as these from SWAN can continue to provide useful clues.

Acknowledgements

I gratefully acknowledge Dr. Thomas J. Beck (Johns Hopkins University, Baltimore, MD) for insightful review and constructive comments on this manuscript.

Abbreviations:

• BMAD,

  Bone mineral apparent density;

• BMC,

  bone mineral content;

• BMD,

  bone mineral density;

• SWAN,

  Study of Women’s Health Across the Nation.