Numerous attempts have been made to define the determinants of BMD. However, most of these studies have involved relatively small selected cohorts, and the relative importance of the many proposed factors has been difficult to assess. We previously described the determinants of appendicular bone mass (proximal and distal radius, calcaneus) in older women participating in the Study of Osteoporotic Fractures [8]. We now examine the same study group to identify correlates of axial BMD (in the proximal femur and spine) and to compare the factors associated with BMD at axial and appendicular sites.

Methods

Top Methods Results Discussion Author & Article Info References

Participants

From September 1986 to October 1988, 9704 women who were at least 65 years of age were recruited for participation in the Study of Osteoporotic Fractures. As previously described [9], eligible women were identified in Portland, Oregon; Minneapolis, Minnesota; Baltimore, Maryland; and the Monongahela Valley near Pittsburgh, Pennsylvania, from population-based listings (such as voter registration and vehicle registration lists). We excluded black women because of the low incidence of hip fractures in this group [10], women who could not walk without the assistance of another person, and women who had bilateral hip replacements. The demographic characteristics of the study group have been described [8]. Approximately 2 years after the baseline visit, the entire cohort was invited back to the clinical sites for a second visit. At that time, axial BMD measurements were obtained.

Measurement of Bone Mass

Bone mineral density (g/cm²) of the lumbar spine and total hip and its subregions (the femoral neck, intertrochanteric region, trochanter, and the Ward triangle) were measured with dual-energy x-ray absorptiometry (QDR 1000, Hologic, Inc., Waltham, Massachusetts) using previously described methods [11]. These measurements were obtained at each participant's second visit. The in vivo inter- scanner coefficient of variation was 1.52% for the spine and 1.2% for the femoral neck, and the inter-scanner coefficient of variation was 0.9% for an anthropometric hip phantom [11].

Other Measurements

At the second visit, participants completed a self-administered questionnaire and were interviewed and examined at the clinical centers. Information obtained included elements of the participants' medical history. Some aspects of family history and reproductive history were previously obtained at the baseline visit. Participants were asked to bring all prescription and nonprescription medications for verification and for determination of the doses and duration of use. A detailed history was obtained concerning lifetime smoking history, alcohol use, and caffeine intake. Women who had smoked fewer than 100 cigarettes in their lifetime were considered nonsmokers. We assumed that one cup of coffee contained 95 mg of caffeine and that tea and cola drinks contained 55 and 45 mg, respectively.

We assessed calcium intake using a method developed from the HANES (National Health and Nutrition Examination Survey) II survey [12]. Food models were used to estimate portion sizes, and foods that account for 80% of calcium intake in most adults were included. Milk consumption at several ages was assessed by questionnaire. Physical activity was estimated using a modified Paffenbarger scale that assesses participation in many typical settings [13]. Intensity-weighted activity measures were calculated by designating activities as low intensity (for example, walking or gardening), medium intensity (dancing or tennis), or high intensity (jogging or skiing) and by multiplying the reported frequency of the activity by 2.5, 5.0, and 7.5, respectively.

Weight was measured in indoor clothing with shoes removed using a balance beam scale. Knee height was measured as the distance from the floor to the anterior tibial plateau at the first visit. Height loss since age 25 years was defined in cm as the self-reported height at age 25 years minus the current height. Quadriceps strength was measured using a leg-extension chair (Bodymasters MD 110, Lafayette Instruments Co., Lafayette, Indiana). Maximal triceps extension and hip abduction strength were measured with a hand-held isometric dynamometer (Sparks Instruments and Academics, Coralville, Iowa). Grip strength was assessed as the average of two attempts with a Preston dynamometer (Sammons-Preston Company, Burridge, Illinois), and gait speed was determined as the time in seconds needed to walk 12 meters.

Statistical Analysis

To detect potential associations among predictor variables and bone density, we first analyzed the data using bivariate age-adjusted models. These associations were further adjusted for weight. We used linear regression, analysis of variance, and the Student t-test to test significance. Because some participants were not available for measurements at the second clinic visit, complete data were available for bivariate analyses in 7664 participants for spinal BMD and in 7963 for femoral BMD. To express the strength of the bivariate associations listed in Table 1 and Table 2, units of change were chosen to be approximately one standard deviation in the distribution of a given variable. We used multiple regression techniques to determine which variables were independently associated with BMD. Variables with bivariate associations in which the P value was less than 0.05 were included in the multivariable models. Variables with bivariate associations in which the P value was greater than 0.05 were excluded to limit the number of independent variables and the likelihood of nonspecificity. Because of the cumulative effect of missing values (participants were not included if variables were missing in any field), the sample size for the multivariable regression was 5405. All variables considered in the multivariable analysis are listed in the Appendix (Table 1)

The data collected included sets of variables related to broad areas of interest, such as measures of calcium intake, estimates of physical activity, and measures of estrogen exposure. Each of these sets contained several factors that were associated with BMD in bivariate analyses. To minimize the complexity of the overall statistical model, we first examined variables in these sets for multicollinearity and then did multiple regression analyses within each set to determine which factor or factors best explained the variability in BMD. We then included these factors in the overall multivariable analyses. For example, in the calcium intake category, two factors (calcium from milk between the ages of 18 and 50 years and the three variables for level of use indicating no, past, or current use of calcium supplements) both independently and significantly explained variance in femoral neck BMD. We therefore included both these variables in the overall multivariable analysis. To avoid bias from specific therapies for osteoporosis, we did not include participants with osteoporosis (osteoporotic fractures) in the multivariable analyses. Subanalysis showed no major differences between participants with and those without osteoporosis. We tested possible interactions between variables, suggested by the literature or initial review of the data, after selecting the initial model. These included weight and alcohol use, weight and smoking, weight and estrogen use, estrogen use and smoking, caffeine intake and calcium obtained from supplements, and caffeine intake and calcium obtained from food. If an interaction to be tested included effects that were not in the final model, we also included these main effects in the model. We used graphic displays and discontinuous regression models to examine potential threshold effects.

Results

Top Methods Results Discussion Author & Article Info References

Results of the analyses were similar for all four subregions of the proximal femur. For the sake of clarity, we only report results for the femoral neck region. Factors associated with spine and femoral neck BMD are summarized in Table 1 and Table 2, and factors not associated with BMD are listed in (Table 3).

Age, Height, and Weight

Increasing age was associated with a reduced BMD at all measurement sites. The relation was considerably weaker in the multivariable model than in the bivariate analysis, presumably because some of the age effect is the result of more proximate variables. Knee height was positively associated with BMD at all sites but the femoral neck after adjustment for weight, but the association was weaker than that with overall height.

Weight was the strongest predictor of BMD at all sites, accounting for 9% to 21% of the difference in BMD between participants. For example, a 10-kg weight increase was associated with an increase in BMD of about 6% at spinal and femoral neck sites.

Reproductive History and Estrogen Use

Earlier menarche and later menopause were associated with an increase in both spinal and femoral BMD. Whether a participant had ever given birth was not associated with BMD at either measurement site after adjustment for weight.

Current use of oral estrogen was strongly positively associated with BMD, as was the number of years of estrogen use or a history of birth control pill use. In women who had stopped using estrogen, BMD was inversely associated with the number of years since discontinuation of therapy, suggesting that the beneficial effects of estrogen are not permanent. There was no interaction between calcium intake (from food or supplements) and caffeine intake on BMD. When the concurrent use of estrogens was considered, no association was seen between progestin use and BMD.

Family History

Bone mineral density at the spine and femoral neck was lower among women whose mother, sister, or brother had a history of hip fracture. Having a sister with a history of wrist fracture was associated with reduced BMD at both spinal and femoral sites. Hair color and Northern European ancestry were not related to bone density.

Calcium and Vitamin D Intake

Persons with a history of fractures (who may have been prescribed calcium supplements as therapy for osteoporosis) were excluded. Current calcium intake from milk, all dietary sources, or supplements was associated with increased femoral neck and spine BMD, but no evidence suggested a threshold effect of the relation between calcium consumption and BMD. The effect of calcium intake on BMD was similar in those receiving and not receiving estrogen replacement therapy. The interaction between calcium intake (from food or supplements) and caffeine intake did not change the effect of calcium intake on BMD. Participants who were currently taking a vitamin D supplement had higher BMD at the lumbar spine.

Strength and Physical Activity

Results of tests of strength were associated with all sites of BMD measurement. Walking speed was positively correlated with BMD at the femoral neck. Level of activity, expressed as the number of times the participant was active in the last year or the number of times per year at age 50 the participant engaged in high-intensity activity, was also associated with BMD.

Smoking, Caffeine, and Alcohol

Tobacco use was inversely associated with femoral neck BMD in bivariate analyses, an association that remained apparent at the femoral neck after adjustment for weight. Previous smokers did not differ from participants who had never smoked, but the number of pack-years of cigarette use was negatively related to bone mass at the femoral neck. Caffeine intake was negatively correlated with BMD at both proximal femoral and lumbar spine measurement sites. Positive associations between BMD and alcohol consumption (expressed as either current alcohol use or the total number of drinks in the participant's lifetime) became apparent after adjustment for weight.

Medications

After adjustment for weight, women who were currently taking glucocorticoids had lower spinal BMD but not lower femoral neck BMD. Use of anticonvulsant agents was associated with lower spinal (but not femoral) BMD. Thyroid hormone use was not associated with BMD. No association was seen between antacid use and spinal or femoral BMD.

Bone mineral density was greater in women who currently use or had previously used thiazide diuretics. In addition, use of nonthiazide diuretics was associated with higher BMD, even in participants with no history of thiazide use.

Multivariable Model

Multivariable correlates of BMD in the spine and femoral neck are listed in Table 4 and Table 5. Multivariable regression models explained 21% and 25% of the overall differences between participants in BMD at the femoral neck and spinal measurement sites, respectively. At both sites, weight was most highly associated with bone mass. In addition, increased age and a history of osteoporotic fracture in a participant's mother or sibling were associated with lower BMD, whereas estrogen therapy, later menopause, height, physical activity, and use of thiazide diuretics were all positively related to BMD. Other variables associated with higher BMD at one of the two measurement sites include calcium intake from milk since age 18 years (hip), younger age at menarche (spine), quadriceps strength (hip), alcohol intake (spine), non-insulin-dependent diabetes mellitus (hip), and use of nonthiazide diuretics (spine). Spinal density was higher in participants who reported a history of any type of arthritis.

Discussion

Top Methods Results Discussion Author & Article Info References

Hip and spine fractures are the most common sequelae of osteoporosis in elderly women and cause considerable morbidity and mortality [1]. Bone mass at these sites is an important predictor of risk for fracture [5]. To better understand the factors that influence bone mass, we characterized environmental and anthropometric variables in elderly women and examined the relation of those variables to BMD. Several features of the study are unique, including its large size, the careful delineation of the variables studied, and the wealth of information available about the study cohort. The results validate associations described earlier in smaller series, such as the major importance of weight and estrogen exposure, but also show previously unrecognized relations.

Our study also had several limitations. We examined elderly, independently living volunteers whose characteristics may differ from those of other groups. The study was cross-sectional in design and partially depended on information obtained by recall. Confirmation of our results with longitudinal evaluations would be useful.

The correlates of spinal and proximal femur BMD in the multivariable regression analyses were similar to one another and to correlates previously reported for appendicular sites in this study group [8] (Table 6). Several variables, including weight, height, estrogen exposure, non-insulin-dependent diabetes mellitus, and thiazide use were consistently associated with greater bone mass. A positive effect of calcium intake was clearly present at femoral and radial sites and, in the bivariate models, at the spine. The association of these factors with BMD at several skeletal sites emphasizes their strength and the widespread nature of their effects; both cortical and trabecular bone are positively influenced. On the other hand, a family history of fracture had a negative effect at all skeletal sites, and aging was associated with lower BMD in all areas except the spine, where artifact can obscure that trend [15].

Body weight was most highly related to BMD, a finding that was present regardless of age, estrogen status, or any other variable examined. This finding corroborates other reports of the importance of weight in predicting bone mass [8, 16, 17] and reinforces the need to further understand the mechanism of this effect. Weight itself, not body mass index, is most highly predictive of BMD in this study group [14]. The mechanism underlying the strong association of weight with BMD is undoubtedly complex. It is largely the result of mechanical loading [14] but may also be caused by metabolic concomitants of body composition and diet [18, 19]. Lower body weight is associated with an increased risk for hip fracture that is independent of bone mass [20, 21], and the strong association of lower weight with low bone mass reinforces the importance of this variable. Patients should avoid excessive thinness, and clinicians should be alert to the increased risk for fracture in slender women.

Estrogen status was strongly associated with BMD, as has been repeatedly shown in former evaluations [22]. In addition to the beneficial effects of estrogen replacement therapy, other variables that probably reflect the total duration of estrogen exposure (such as early age of menarche and later age of menopause) were also positively related to BMD, a finding that emphasizes the strength and complexity of the association. The inverse relation between years since discontinuation of estrogen therapy and BMD corroborates other evidence suggesting the transience of the protective effect of estrogen on BMD [23].

Other variables were related to BMD. Greater strength, particularly quadriceps strength, was strongly associated with higher BMD. This finding is in accord with other reports of the positive relation between bone mass and muscle strength [8, 24-26]. In the multivariable model, indices of activity were independently associated with higher BMD, supporting the contention that the relation between exercise and bone mass may be mediated by several pathways. For example, mechanical effects on bone may result from direct effects of loading (weight bearing, impact) or more indirectly through muscle tension.

A personal history of previous fracture was associated with lower BMD, a finding consistent with those of previous reports [8, 27] and the operational definition of osteoporosis (low bone mass and fracture). In addition, several groups have documented the value of a previous fracture in predicting subsequent risk, independent of the presence of low bone mass [4, 28]. A family history of osteoporotic fractures was also strongly associated with lower bone mass. Although a maternal history of fracture was most clearly related to reduced BMD, a history of fractures in siblings (male or female) was also associated [29]. Studies of twins and families have described the heritability of bone mass [30], although the mechanism of this effect is uncertain.

We found that a higher calcium intake through diet or supplement use was related to increased bone mass in the proximal femur. A 400-mg increment in calcium consumption was associated with a 1.3% increase in femoral neck bone mass. No threshold intake level was apparent. Despite other reports [31], we did not detect an effect of caffeine intake on the relation between dietary calcium and BMD. Heaney [32] has also suggested the presence of an association among calcium, estrogen use, and BMD. Such a relation was not apparent in our study group, and the beneficial effects of calcium intake with BMD was similar in participants who received estrogen supplements and those who did not. In our study, and in longitudinal comparisons of the effects of estrogen and dietary calcium in postmenopausal women [33], the effect of estrogen on BMD is considerably greater than the effect of increased dietary calcium. A positive effect of increased calcium intake may therefore be obscured or replaced by estrogen supplementation.

Cigarette smoking was a risk factor for lower BMD in the bivariate model, a finding in accord with those of various other studies [24, 34, 35]. The number of pack-years of smoking was negatively correlated with BMD in current smokers, but BMD did not appear to be permanently decreased in previous smokers. However, as in previous studies [25, 26], the negative effect of smoking was not seen in the multivariable model, perhaps indicating the influence of intervening variables.

Alcohol intake was positively associated with bone mass. Many studies have documented the adverse skeletal effects of alcohol abuse [36, 37], but others suggest that moderate alcohol intake is related to higher bone mass [38]. Whether this is a direct effect of ethanol on bone remodeling or is related to other inapparent but related variables is unclear. The relation of alcohol to BMD, however, only became apparent after adjustment for weight.

In other studies, thiazide use has been related to a positive effect on BMD at appendicular measurement sites [39-41] and at the lumbar spine [41, 42]. We noted associations in the study group at both the spine and proximal femur. The skeletal effects of thiazide diuretics have largely been attributed to their hypocalciuric action, although investigators have suggested that members of this class of diuretics may have other actions that directly affect the skeleton [43]. Whether the beneficial effects on bone mass are translated into fracture prevention has been controversial, but most reports suggest that the risk for hip fracture is reduced in persons receiving thiazides [39, 44-46]. Surprisingly, nonthiazide diuretics were also related to higher bone mass in our study group, an association that was apparent even when we omitted participants who had ever taken thiazides. Previous smaller studies had found no association between the use of other antihypertensive drugs and bone mass or risk for fracture [44, 45, 47]. In the same group we studied, Cauley and colleagues [39] found no association between the use of nonthiazide diuretics and appendicular bone mass. These findings should prompt a reevaluation of the association of other diuretics on bone metabolism and the mechanism of the diuretic effect. For example, the common association of BMD to diuretics with differing mechanisms of action could suggest the presence of a confounding variable not previously considered.

Other factors appeared to have a less generalized effect. In our study, measures of muscle strength were most closely associated with BMD at anatomically related sites (such as grip strength with radial BMD and quadriceps strength with femoral BMD). This observation is in accord with other observations that muscle action has a local effect [48]. Measures of recent activity were positively associated with both spinal and femoral BMD but not with the distal radius, possibly because these variables primarily reflect weight-bearing action expected to load the hip and spine more than the distal radius. Although tobacco use and gastrectomy (both associated with lower radial BMD) were related to spinal and femoral neck BMD in the bivariate models, multivariable analyses did not support these relations. Other reports have substantiated the adverse effects of these conditions on axial bone mass [24, 49]. Intervening nutritional or metabolic variables considered in these more detailed analyses may be responsible for the mediation of the effects of smoking and gastrectomy on spinal and femoral BMD.

In summary, our study of a large group of older women supports the importance of several factors associated with bone mass. From a clinical perspective, these findings reinforce the skeletal benefits of premenopausal eugonadism, estrogen use after menopause, and adequate lifelong calcium nutrition. Moreover, they suggest that excessive thinness should be discouraged and emphasize the usefulness of physical activity. A family history of fracture was clearly associated with reduced BMD, underscoring the need to identify a genetic mechanism for the heritability of osteoporosis.

Appendix. Members of the Study of Osteoporotic Fractures Research Group

University of California, San Francisco (Coordinating Center): S.R. Cummings, M.C. Nevitt, D. Black, H.K. Genant, C. Arnaud, W. Browner, K. Faulkner, C. Fox, C. Gluer, S. Harvey, S.B. Hulley, L. Palermo, D. Seeley, and P. Steiger.

University of Maryland, Baltimore: R. Sherin, J. Scott, K. Fox, J. Lewis, G. Greenberg, L. Finazzo, T. Page, S. Trusty, B. Hohman, and E. Oliner.

University of Minnesota, Minneapolis: K. Ensrud, R. Grimm Jr., C. Bell, D. Thomas, K. Jacobson, S. Jackson, E. Mitson, L. Stocke, and P. Frank.

University of Pittsburgh, Pennsylvania: J.A. Cauley, L.H. Kuller, L. Harper, M. Nasim, C. Bashada, L. Buck, A. Githens, D. Medve, and S. Rudovsky.

The Kaiser Permanente Center for Health Research, Portland, Oregon: T.M. Vogt, W.M. Vollmer, H. Glauber, E. Orwoll, J. Blank, P. Smith, R. Bright, and J. Downing.

Dr. Bauer: University of California, San Francisco, Prevention Sciences Group, 74 New Montgomery, Suite 600, San Francisco, CA 94105.

Dr. Vogt: Kaiser Permanente Center for Health Research, 3800 North Kaiser Center Drive, Portland, OR 97227-1042

Dr. Fox: University of Maryland, Department of Epidemiology and Preventive Medicine, 660 West Redwood Street, Baltimore, MD 21201.