JOURNAL OF BONE A N D MINERAL RESEARCH Volume 7, N u m k r 6,1992 Mary Ann Llekrt. he., Publishers
Axial and Appendicular Bone Density Predict Fractures in Older Women DENNIS M. BLACK,' STEVEN R. CUMMINGS,'.' HARRY K. GENANT,' MICHAEL C. NEVITT,' LISA PALERMO,' and WARREN BROWNER'
ABSTRACT
To determine whether measurement of hip and spine bone mass by dual-energy x-ray absorptiometry (DEXA) predicts fractures in women and to compare the predictive value of DEXA with that of single-photon absorptiometry (SPA) of appendicular sites, we prospectively studied 8134 nonblack women age 65 years and older who had both DEXA and SPA measurements of bone mass. A total of 208 nonspine fractures, including 37 wrist fractures, occurred during the follow-up period, which averaged 0.7 years. The risk of fracture was inversely related to bone density at all measurement sites. After adjusting for age, the relative risks per decrease of 1 standard deviation in bone density for the occurrence of any fracture was 1.40 for measurement at the proximal femur (95% confidence interval 1.20-1.63) and 1.35 (1.15-1.58) for measurement at the spine. Results were similar for all regions of the proximal femur as well as SPA measurements at the calcaneus, distal radius, and proximal radius. None of these measurements was a significantly better predictor of fractures than the others. Furthermore, measurement of the distal radius was not a better predictor of wrist fracture (relative risk 1.64: 95% CI 1.13-2.37) than other sites, such as the lumbar spine (RR 1.56; CI 1.07-2.26). the femoral neck (RR 1.65; CI 1.12-2.41), or the calcaneus (RR 1.83; C1 1.26-2.64). We conclude that the inverse relationship between bone mass and risk of fracture in older women is similar for absorptiometric measurements made at the hip, spine, and appendicular sites.
INTRODUCTION EVERAL PROSPECTIVE STUDIES have established that measurements of bone mass in the appendicular skeleton are inversely related to the risk of subsequent fractures of the hip and most other types of fractures in women."-5) One of these studies"' suggested that measurements of the spine using dud-photon absorptiometry might have a somewhat weaker relationship to risk of fracture than appendicular measurements. There have been no prospective studies, however, of the predictive value of densitometry of the proximal femur in comparison to measurements at other sites or of the predictive value of bone mass measured by dual-energy x-ray absorptiometry (DEXA). To determine whether older women with lower bone mineral density (BMD) in the proximal femur and spine
S
would have a higher incidence of fractures and to compare the predictive value of measurements of BMD in the hip and spine with those of appendicular measurements, we conducted a prospective multicenter study of a cohort of women aged 65 years or older, the Study of Osteoporotic Fractures. We measured BMD in a cohort of women and followed them every 4 months for the occurrence of fractures.
METHODS Subjects From September 1986 to October 1988, 9704 women at least 65 years of age were recruited for the Study of Osteoporotic Fractures in four areas of the United States: Portland, Oregon; Minneapolis, Minnesota; Baltimore County,
'Department of Epidemiology and Biostatistics, University of California, San Francisco. 'Division of General Internal Medicine, University of California, San Francisco. 'Department of Radiology, University of California, San Francisco.
633
634 Maryland, and the Monongahela Valley near Pittsburgh, Pennsylvania. Details of recruitment have been published We excluded black women because of the low incidence of hip fractures in this group, women who were unable to walk without the assistance of another person, and those who had bilateral hip replacements. All women in the cohort were invited to return for a second examination approximately 2 years after the baseline examination, from November 1988 through December, 1990; 8241 (87%) of 9483 surviving members of the cohort attended that examination. At the second visit, bone mineral density in the hip and spine was measured. In this analysis we include the 8134 participants who had adequate measurements of the hip or of the spine at the second examination.
Measurement of bone mass
BLACK ET AL. were unable to provide an interview, we obtained this information from the closest available relative or other surrogate. We excluded reports of spine fractures from this study because self-report may not be reliable and we did not have baseline and comparison spine x-rays to confirm incident spine fractures. We also excluded any fractures that resulted from major trauma, such as a motor vehicle accident. A total of 18 fractures in 12 women were excluded because of major trauma. Only fractures that occurred after the second examination are included in these analyses. Copies of x-ray reports were obtained for all reported fractures, and copies of preoperative radiographs were obtained for all reported hip fractures. X-ray reports were reviewed by physicians at the clinical center and again at the coordinating center to confirm and classify fractures.
Bone mineral content (BMC, g/cm) and density (BMD, Statistical analysis g/cm') of the proximal and distal forearm and calcaneus We estimated the independent contributions of bone were measured using OsteoAnalyzers (Siemens-Osteon, Wahiawa, HI). Details of the appendicular measurements density and age in two ways. First, we stratified women by have been published elsewhere."I Measurements of the 5 year age groups and then calculated the age-specific distal radius site begins at the point at which the ulna and quartiles of bone mineral density. Each woman was asradius are separated by 3.2 mm and moves proximally signed to one of those quartiles for each measurement. We scanning four times. 3 mm apart. All measurements were then summed the number of fractures and the womanconducted on the right side except in women who had suf- years of exposure in each age-specific quartile of bone denfered a fracture, stroke, or severe injury involving that sity and used these figures to calculate age-adjusted rates limb; in these instances, bone mineral density was mea- of fracture for each quartile of bone density. Second, we sured on the left side. Bone mineral content (BMC, g/cm) included bone mineral density and age as continuous variaand density (BMD, g/cm') of the proximal femur and bles in Cox proportional hazards models. [ W Separate modspine were measured in the anteroposterior (AP) projec- els were analyzed for each measurement site. We contion using Hologic QDR lo00 scanners (Hologic, Inc., ducted similar analyses for bone mineral content. Relative hazards from the proportional hazards model are reported Waltham, MA). We also assessed the association between measurements as relative risk. To compare the strength of association beof appendicular bone mass made approximately 2 years tween bone mass at each of the measurement sites and apart by remeasuring appendicular bone mass in 470 fractures, we analyzed the areas under receiver-operating women at the second examination using the same instru- characteristic (ROC) curves for each site and tested the statistical significance of the difference between these ments and protocols as at the baseline examination. Based on measurements of research staff who visited all areas."." In comparisons of axial and appendicular meafour clinical centers, the mean coefficients of variation (be- surements of bone density, we included only those women tween centers) were 0.4% for the distal radius, 0.5% for who had both types of measurements. We used the method the proximal radius, 1.2% for the calcaneus, 1.2% for the of Rosner et al.[gl to estimate the attenuation in the relaproximal femur, and 1.5% for the lumbar spine. We tive risk for the appendicular measurements resulting from the 2 year interval between the first and second examinafound that one clinic (Portland) had slightly higher (34%) mean appendicular BMD measures than the other three tions. clinics. Thus, we also included clinical center in our multivariate analyses. This had no effect on our results, so for RESULTS simplicity, we have omitted this variable from our presenCharacteristics of the 8134 women who completed all tation of the results. measurements of bone density at the baseline and second examinations are described in Table 1. In the 470 women Assessment of fractures for whom appendicular values were repeated at the second To ascertain fractures that occurred after the baseline examination, the correlation between appendicular meaexamination, we asked participants to notify the local clin- surements made at the first and second examinations ical center as soon as possible after any fracture. In addi- ranged from r = 0.89 for the distal radius to r = 0.97 for tion, we contacted participants every 4 months by letter or the calcaneus. In the 8134 women with hip or spine BMD, the correlatelephone to ask if they had sustained a fracture; these contacts were over 99% complete. As soon as possible tion between mean BMD of the lumbar spine (L1-4)and after a report of a fracture, we interviewed participants total proximal femur was 0.66, and correlations between about the type of fracture and how it occurred. If they BMD of the spine and regions of the proximal femur
AXIAL BONE DENSITY AND NONSPINE FRACTURE
ranged from 0.58 for BMD of Ward‘s triangle to 0.65 for trochanteric BMD. Correlations between BMD of the regions of the proximal femur ranged from 0.76 to 0.87. Participants were followed (after the second examination) for an average of 0.72 years (range 0.1-1.9 years). Over 99% of follow-up contacts about vital status and fractures were completed. During this follow-up period, 191 participants suffered 208 nonspine fractures (confirmed by x-ray reports) that were not a result of major trauma. Types of fractures are detailed in Table 2. We report results for all fractures combined and for wrist fractures.
TABLE1. BASELINE CHARACTERISTICS OF SUBJECTS
N subjects (Yo)
Characteristic
8,134 71.4 3,601 (44) 2.566 (32) 1,265 (16) 544 (7) 158 (2)
Number Age, years (mean) 65-69 (VO) 70-74 (To) 75-79 (Yo) 80-84 ( 7 0 ) 2 85
Race White Bone mineral density, g/cm’a Lumbar spine Proximal femur Femoral neck Ward’s triangle Intertrochanteric Trochanter Calcaneus Proximal radius Distal radius *Values are mean
8,111 (99.7) 0.857 0.757 0.648 0.427 0.883 0.557
f 0.169 f 0.131
f 0.111
f f f 0.406 f 0.636 f
0.110 0.160 0.102
0.094
0.103 0.363 f 0.084
standard deviation.
635
Bone mass of proximal femur, spine, and risk of all nonspine fractures Decreased bone mineral density at both the hip and the spine was associated with an increased risk of any subsequent fractures (Table 3). For example, the relative risk for BMD at the femoral neck was 1.41 after adjusting for age. Thus, a woman whose BMD was 1 standard deviation (SD) below the mean BMD at that site had a risk 1.41 higher than a woman whose BMD was at the mean. Similar associations were observed for all other regions of the proximal femur (Table 3). The age-adjusted relative risk of fracture increased by 1.35 for each standard deviation decrease in BMD of the lumbar spine. Furthermore, the relationship between age-adjusted measurements of calcaneal and forearm bone density were very similar to those observed for the proximal femur and spine: the relative risks of fracture for a decrease of 1 SD ranged from 1.32 for the proximal radius to 1.51 for the calcaneus. In general, there appeared to be a uniform gradient of increasing risk from the highest to lowest quartile of BMD for all sites (Fig. 1). The results for wrist fracture are similar to those for all fractures. After age adjustment, the standardized relative risks are similar for BMD at all sites (Table 4). Specifically, the measurement of the distal radius was not a better predictor of wrist fracture (relative risk 1.64; 95% CI 1.132.37) than other sites, such as the lumbar spine (RR 1.56; CI 1.07-2.26), the femoral neck (RR 1.65; CI 1.12-2.41), or the calcaneus (RR 1.83; CI 1.26-2.64). Since the estimates are based on only 37 wrist fractures, however, the confidence intervals for all these measurements are wide. There were no significant differences in the areas under the ROC curves between any of the measurement sites for all fractures (proximal femur = 0.60, spine = 0.59, calcaneus = 0.62, distal radius = 0.62, and proximal radius = 0.59; P 2 0.15 for all comparisons) or for wrist fractures. Therefore, none of the measurements of bone mineral density in the proximal femur, spine, calcaneus, or radius were substantially or significantly better than any others for predicting all fractures or wrist fractures.
TABLE2. TYPESOF FRACTURES OBSERVED DURINGFOLLOW-UP
Age category Type of fracture
N (% of fractures)
65-69
70-74
75-79
80+
~
Distal radiudwrist Foot Ribs Humerus Ankle Toes Hip Intertrochanteric Femoral neck Other Other types (s10 each)
7 12 3 3 3 3 3 1 2
15 9 7 7 6 5 4 2 2
11 4 8 3 8 5 5 3 2
13
27
5
4 3 5 6
1 2 6 4 1
1 5
BLACK ET AL.
636
TABLE3. ASSOCIATION BETWEEN BONEMINERALDENS^
AND
ALL~RACTURESIN 8134 WOMEN Relative riska (95 % confidence interval)
Standard deviation Measurement (g/cm ’)
W)
Unadjusted
Adjusted for age
Proximal femur (total) Femoral neck Ward’s triangle Intertrochanteric Trochanteric Lumbar spine Calcaneus Distal radius Proximal radius
0.131 0.111 0.110 0.160 0.102 0.169 0.094 0.084 0.103
1.43 (1.23,1.66) 1.44 (1.23,1.69) 1.45 (1.25,1.69) 1.42 (1.22,1.64) 1.41 (1.21,1.64) 1.36 (1.17,1.60) 1.53 (1.31, 1.79) 1.45 (1.24,1.69) 1.35 (1.17,1.56)
1.40 (1.20,1.63) 1.41 (1.20,1.66) 1.43 (1.22,1.67) 1.39 (1.19,1.62) 1.38 (1.18,1.61) 1.35 (1.15,1.58) 1.51 (1.29,1.77) 1.42 (1.21,1.67) 1.32 (1.14,1.53)
.Relative hazards for bone mineral density (BMD) pcr standard deviation decrease in BMD.
We considered the possibility that the longer interval between SPA measurements and fractures may have weakened that association; however, the correlation between measurements of appendicular sites 2 years apart was r L 0.89,and thus the effect would be very small. For example, based on the observed relative risk of 1.42 for each standard deviation decrease in BMD of the distal radius, we estimate that the relative risk for measurements made 2 years later would not exceed 1.48.
60
40
!
rn
F
20
C 0
!
a,
a
DISCUSSION 0
0
Proximal Femur: Quartiles of BMD
0 0 7
L
60
a,
a a, 0
C
a,
2 u C -
40
20
0
1
2
3
4
Distal Radius: Quartiles of BMD
Incidence of all fractures by age-adjusted quarFIG. tile of bone mineral density (BMD).
We have demonstrated that reduced bone mineral density of the proximal femur and spine is strongly associated with an increased risk of all nonspine and of wrist fractures in older women. The strength of the relationship between BMD and all nonspine fractures in our cohort (range in relative risk from 1.3 to 1.5) appears to be similar to the relative risk of 1.5 reported by Hui et al.,”’ although somewhat lower than the range of values (1 3-1.9) reported by Wasnich et a l . “ O ’ However, the Wasnich study was performed on a younger cohort of Asian-American women. The association between radial BMD and wrist fractures in our cohort is somewhat higher than that found by Hui et al.[’’ We found that the magnitude of the association between BMD and all fractures or wrist fractures was similar, regardless of the site or method of measurement. This suggests that measurements of bone mass at the proximal femur, spine, and appendicular sites have similar predic.. tive value for the overall risk of nonspine fracture in postmenopausal women. Our results differ somewhat from those of Eastell and colleagues,[11’who found that wrist fractures were somewhat more strongly associated with BMD of the distal radius than BMD of the spine by dualphoton absorptiometry. This difference might be due, in part, to differences in the methods of the two studies. Eastell and colleague’s study was a retrospective case-con-
637
AXIAL BONE DENSITY AND NONSPINE FRACTURE TABLE4. ASSOCIATION BETWEENBONEMINERAL DENSITY AND WRIST FRACTURES IN 8134 WOMEN ~
~~
Relative rkka (95% confidence interval) Standard deviation Measurement (g/cm’) ~~~~
~
(SO)
Unadjusted
Adjusted for age
0.131 0.111
1.49 (1.06, 2.09) 1.62 (1.12, 2.35) 1.32 (0.94,1.86) 1.49 (1.06, 2.10) 1.39 (0.99.1.95) 1.57 (1 .W, 2.28) 1.80 (1.26, 2.57) 1.64 (1.14, 2.35) 1.79 (1.29, 2.47)
1.49 (1.05. 2.12) 1.65 (1.12, 2.41) 1.32 (0.93, 1.88) 1.51 (1.06, 2.14) 1.39 (0.98, 1.98) 1.56 (1.07, 2.26) 1.83 (1.26, 2.64) 1.64 (1.13, 2.37) 1.82 (1.30, 2.55)
~
Proximal femur (total) Femoral neck Ward’s triangle Intertrochanteric Trochanteric Lumbar spine Calcaneus Distal radius Proximal radius
0.110
0.160 0.102 0.169 0.094
0.084 0.103
PRelative hazards for bone mineral density (BMD) per standard deviation decrease in BMD.
trol study, and their site of measurement was somewhat more distal and closer to the usual site of Colles’ fracture than ours. The prospective design of this study avoids many of the potential biases inherent in case-control studies of the association between measurements of bone mass and fracture.tL0’Our study has several limitations, however. We recruited subjects from population-based listings in four cities across the United States, but participants in the Study of Osteoporotic Fractures are not a representative sample of older women; they are volunteers who are likely to be healthier than those who did not participate. Participants did not include women who were homebound or living in institutions, and our results might not apply to women living in long-term care institutions. Nevertheless, our findings are generalizable to the vast majority of white women. Finally, we were unable to determine whether DEXA of the hip has a better predictive value for hip fractures than measurements at other sites. Further follow-up of this cohort may allow us to address this issue. Measurements of the appendicular sites were made 2 years before DEXA of the axial sites. Thus, although it is theoretically possible that this may have caused us to underestimate the association between appendicular BMD and fracture in comparison with DEXA, we found that measurements of BMD made 2 years apart are so highly correlated that the 2 year difference between time of measurement could have, at most, a trivial effect on this comparison. The interval between measurement of bone mass and fracture in our study is very short. This provides a good assessment of the role of decreased bone density in the pathogenesis of fracture, but it will overestimate the predictive value of bone density for fractures over much longer periods of time. Finally, our observational study cannot exclude the possibility that low bone density is a marker for other causes of fracture, such as reduced quality of bone and accumulation of microscopic damage.(*a’
We conclude that there is an inverse relationship between both axial and appendicular bone mass and the risk of nonspine fractures in older women. BMD at all these sites appears to have similar predictive accuracy for all nonspine fractures and wrist fractures.
ACKNOWLEDGMENTS Supported by Public Health Service Grants I-ROI-AG05407, 1-R01-AR35582, 5-RO1-AG05394, 1-ROl-AM35584, and 1-R01-AR35583. Investigators in the Study of Osteoporotic Fractures research group: University of California, San Francisco (coordinating center): S.R. Cummings (principal investigator). M.C. Nevitt (project director), D. Black (study statistician), H.K. Genant (director, central radiology laboratory), C. Arnaud, W. Browner, K. Faulkner. C. Fox, C. Glueer, S. Harvey, S.B. Hulley, L. Palermo, D. Seeley, and P. Steiger. University of Maryland: R. Sherwin (principal investigator), J. Scott (project director), K. Fox (study coordinator), J. Lewis (clinic supervisor), M. Bahr, S. Trusty, B. Hohman, L. Emerson, D. Rebar, and E. Oliner. University of Minnesota: R. Grimm, Jr. (principal investigator), K. Ensrud (coinvestigator), C. Bell (projector director), D. Thomas, K. Jacobson, S. Jackson, E. Mitson, L. Stocke, and P. Frank. University of Pittsburgh: J.A. Cauley (principal investigator). L.H. Kuller (coprincipal investigator). L. Harper (project director), M. Nasim (clinic coordinator), C. Bashada, L. Buck, A. Githens. D. Medve, and S. Rudovsky. The Kaiser Permanente Center for Health Research, Portland, Oregon: T.M. Vogt (principal investigator), W.M. Vollmer, H. Glauber, E. Orwoll (coinvestigators), J. Blank (project director), B. Mastel-Smith (clinic coordinator), R. Bright, and J. Downing.
BLACK ET AL.
638
REFERENCES 1. Cummings SR. Black DM, Nevitt MC, et al. 1990 Appendicular bone density and age predict hip fracture in women. JAMA M3:665-668. 2. Gardsell P , Johnell 0.Nilsson BE 1989 Predicting fractures in women by using forearm bone densitometry. Calcif Tissue Int 44:235-242. 3. Hui SL, Slemenda CW, Johnston CC 1988 Age and bone mass as predictors of fracture in a prospective study. J Clin lnvest 81:1804-1809. 4. Ross PD. Wasnich RD, Heilbrun LK, Vogel JM 1987 Definition of a spine fracture threshold based upon prospective fracture risk. Bone 8:271-278. 5. Wasnich RD, Ross PD. Heilbrun LK, Vogel JM 1985 Prediction of postmenopausal fracture risk with use of bone mineral measurements. Am J Obstet Gynecol 153:745-751. 6. Cox DR 1972 Regression models and life tables (with discussion). J R Stat Soc B34:187-220. 7. Hanky JA, McNeil BJ 1983 A method of comparing the areas under receiver operation characteristic curves derived from the same cases. Radiology 148:839-843. 8. McNeil BJ. Hanky JA 1984 Statistical approaches to the analysis of receiver operating characteristic curves. Med Decis Making 4:137-150.
9. Rosner B, Willett WC, Spiegelman D 1989 Correction of logistic regression relative risk estimates and confidence intervals for systematic within-pcrson measurement error. Stat Med 8:1051-1069. 10. Wasnich RD. Ross PD, Heilbrun LK, Vogel JM 1985 Prediction of postmenopausal fracture risk with use of bone mineral measurements. Am J Obstet Gynecol 153(7):745-751. 11. Eastell R, Wahner HW, O’Fallon WM, Amadio PC, Melton LJ, Riggs BL 1989 Unequal decrease in bone density of lumbar spine and ultradistal radius in Colles’ and vertebral fracture syndromes. J Clin lnvest 83:168-174. 12. Parfitt AM 1987 Trabecular bone architecture in the pathogenesis and prevention of fracture. Am J Med 82(suppl 18): 68-72.
Address reprint requests to: Dennis M . Black, Ph.D. Box 0886 University of California Medical Center Son Francisco, C A 94143-0886 Received for publication March 25, 1991; in revised form November 26. 1991; accepted November 26. 1991
Axial and Appendicular Bone Density Predict Fractures in Older Women DENNIS M. BLACK,' STEVEN R. CUMMINGS,'.' HARRY K. GENANT,' MICHAEL C. NEVITT,' LISA PALERMO,' and WARREN BROWNER'
ABSTRACT
To determine whether measurement of hip and spine bone mass by dual-energy x-ray absorptiometry (DEXA) predicts fractures in women and to compare the predictive value of DEXA with that of single-photon absorptiometry (SPA) of appendicular sites, we prospectively studied 8134 nonblack women age 65 years and older who had both DEXA and SPA measurements of bone mass. A total of 208 nonspine fractures, including 37 wrist fractures, occurred during the follow-up period, which averaged 0.7 years. The risk of fracture was inversely related to bone density at all measurement sites. After adjusting for age, the relative risks per decrease of 1 standard deviation in bone density for the occurrence of any fracture was 1.40 for measurement at the proximal femur (95% confidence interval 1.20-1.63) and 1.35 (1.15-1.58) for measurement at the spine. Results were similar for all regions of the proximal femur as well as SPA measurements at the calcaneus, distal radius, and proximal radius. None of these measurements was a significantly better predictor of fractures than the others. Furthermore, measurement of the distal radius was not a better predictor of wrist fracture (relative risk 1.64: 95% CI 1.13-2.37) than other sites, such as the lumbar spine (RR 1.56; CI 1.07-2.26). the femoral neck (RR 1.65; CI 1.12-2.41), or the calcaneus (RR 1.83; C1 1.26-2.64). We conclude that the inverse relationship between bone mass and risk of fracture in older women is similar for absorptiometric measurements made at the hip, spine, and appendicular sites.
INTRODUCTION EVERAL PROSPECTIVE STUDIES have established that measurements of bone mass in the appendicular skeleton are inversely related to the risk of subsequent fractures of the hip and most other types of fractures in women."-5) One of these studies"' suggested that measurements of the spine using dud-photon absorptiometry might have a somewhat weaker relationship to risk of fracture than appendicular measurements. There have been no prospective studies, however, of the predictive value of densitometry of the proximal femur in comparison to measurements at other sites or of the predictive value of bone mass measured by dual-energy x-ray absorptiometry (DEXA). To determine whether older women with lower bone mineral density (BMD) in the proximal femur and spine
S
would have a higher incidence of fractures and to compare the predictive value of measurements of BMD in the hip and spine with those of appendicular measurements, we conducted a prospective multicenter study of a cohort of women aged 65 years or older, the Study of Osteoporotic Fractures. We measured BMD in a cohort of women and followed them every 4 months for the occurrence of fractures.
METHODS Subjects From September 1986 to October 1988, 9704 women at least 65 years of age were recruited for the Study of Osteoporotic Fractures in four areas of the United States: Portland, Oregon; Minneapolis, Minnesota; Baltimore County,
'Department of Epidemiology and Biostatistics, University of California, San Francisco. 'Division of General Internal Medicine, University of California, San Francisco. 'Department of Radiology, University of California, San Francisco.
633
634 Maryland, and the Monongahela Valley near Pittsburgh, Pennsylvania. Details of recruitment have been published We excluded black women because of the low incidence of hip fractures in this group, women who were unable to walk without the assistance of another person, and those who had bilateral hip replacements. All women in the cohort were invited to return for a second examination approximately 2 years after the baseline examination, from November 1988 through December, 1990; 8241 (87%) of 9483 surviving members of the cohort attended that examination. At the second visit, bone mineral density in the hip and spine was measured. In this analysis we include the 8134 participants who had adequate measurements of the hip or of the spine at the second examination.
Measurement of bone mass
BLACK ET AL. were unable to provide an interview, we obtained this information from the closest available relative or other surrogate. We excluded reports of spine fractures from this study because self-report may not be reliable and we did not have baseline and comparison spine x-rays to confirm incident spine fractures. We also excluded any fractures that resulted from major trauma, such as a motor vehicle accident. A total of 18 fractures in 12 women were excluded because of major trauma. Only fractures that occurred after the second examination are included in these analyses. Copies of x-ray reports were obtained for all reported fractures, and copies of preoperative radiographs were obtained for all reported hip fractures. X-ray reports were reviewed by physicians at the clinical center and again at the coordinating center to confirm and classify fractures.
Bone mineral content (BMC, g/cm) and density (BMD, Statistical analysis g/cm') of the proximal and distal forearm and calcaneus We estimated the independent contributions of bone were measured using OsteoAnalyzers (Siemens-Osteon, Wahiawa, HI). Details of the appendicular measurements density and age in two ways. First, we stratified women by have been published elsewhere."I Measurements of the 5 year age groups and then calculated the age-specific distal radius site begins at the point at which the ulna and quartiles of bone mineral density. Each woman was asradius are separated by 3.2 mm and moves proximally signed to one of those quartiles for each measurement. We scanning four times. 3 mm apart. All measurements were then summed the number of fractures and the womanconducted on the right side except in women who had suf- years of exposure in each age-specific quartile of bone denfered a fracture, stroke, or severe injury involving that sity and used these figures to calculate age-adjusted rates limb; in these instances, bone mineral density was mea- of fracture for each quartile of bone density. Second, we sured on the left side. Bone mineral content (BMC, g/cm) included bone mineral density and age as continuous variaand density (BMD, g/cm') of the proximal femur and bles in Cox proportional hazards models. [ W Separate modspine were measured in the anteroposterior (AP) projec- els were analyzed for each measurement site. We contion using Hologic QDR lo00 scanners (Hologic, Inc., ducted similar analyses for bone mineral content. Relative hazards from the proportional hazards model are reported Waltham, MA). We also assessed the association between measurements as relative risk. To compare the strength of association beof appendicular bone mass made approximately 2 years tween bone mass at each of the measurement sites and apart by remeasuring appendicular bone mass in 470 fractures, we analyzed the areas under receiver-operating women at the second examination using the same instru- characteristic (ROC) curves for each site and tested the statistical significance of the difference between these ments and protocols as at the baseline examination. Based on measurements of research staff who visited all areas."." In comparisons of axial and appendicular meafour clinical centers, the mean coefficients of variation (be- surements of bone density, we included only those women tween centers) were 0.4% for the distal radius, 0.5% for who had both types of measurements. We used the method the proximal radius, 1.2% for the calcaneus, 1.2% for the of Rosner et al.[gl to estimate the attenuation in the relaproximal femur, and 1.5% for the lumbar spine. We tive risk for the appendicular measurements resulting from the 2 year interval between the first and second examinafound that one clinic (Portland) had slightly higher (34%) mean appendicular BMD measures than the other three tions. clinics. Thus, we also included clinical center in our multivariate analyses. This had no effect on our results, so for RESULTS simplicity, we have omitted this variable from our presenCharacteristics of the 8134 women who completed all tation of the results. measurements of bone density at the baseline and second examinations are described in Table 1. In the 470 women Assessment of fractures for whom appendicular values were repeated at the second To ascertain fractures that occurred after the baseline examination, the correlation between appendicular meaexamination, we asked participants to notify the local clin- surements made at the first and second examinations ical center as soon as possible after any fracture. In addi- ranged from r = 0.89 for the distal radius to r = 0.97 for tion, we contacted participants every 4 months by letter or the calcaneus. In the 8134 women with hip or spine BMD, the correlatelephone to ask if they had sustained a fracture; these contacts were over 99% complete. As soon as possible tion between mean BMD of the lumbar spine (L1-4)and after a report of a fracture, we interviewed participants total proximal femur was 0.66, and correlations between about the type of fracture and how it occurred. If they BMD of the spine and regions of the proximal femur
AXIAL BONE DENSITY AND NONSPINE FRACTURE
ranged from 0.58 for BMD of Ward‘s triangle to 0.65 for trochanteric BMD. Correlations between BMD of the regions of the proximal femur ranged from 0.76 to 0.87. Participants were followed (after the second examination) for an average of 0.72 years (range 0.1-1.9 years). Over 99% of follow-up contacts about vital status and fractures were completed. During this follow-up period, 191 participants suffered 208 nonspine fractures (confirmed by x-ray reports) that were not a result of major trauma. Types of fractures are detailed in Table 2. We report results for all fractures combined and for wrist fractures.
TABLE1. BASELINE CHARACTERISTICS OF SUBJECTS
N subjects (Yo)
Characteristic
8,134 71.4 3,601 (44) 2.566 (32) 1,265 (16) 544 (7) 158 (2)
Number Age, years (mean) 65-69 (VO) 70-74 (To) 75-79 (Yo) 80-84 ( 7 0 ) 2 85
Race White Bone mineral density, g/cm’a Lumbar spine Proximal femur Femoral neck Ward’s triangle Intertrochanteric Trochanter Calcaneus Proximal radius Distal radius *Values are mean
8,111 (99.7) 0.857 0.757 0.648 0.427 0.883 0.557
f 0.169 f 0.131
f 0.111
f f f 0.406 f 0.636 f
0.110 0.160 0.102
0.094
0.103 0.363 f 0.084
standard deviation.
635
Bone mass of proximal femur, spine, and risk of all nonspine fractures Decreased bone mineral density at both the hip and the spine was associated with an increased risk of any subsequent fractures (Table 3). For example, the relative risk for BMD at the femoral neck was 1.41 after adjusting for age. Thus, a woman whose BMD was 1 standard deviation (SD) below the mean BMD at that site had a risk 1.41 higher than a woman whose BMD was at the mean. Similar associations were observed for all other regions of the proximal femur (Table 3). The age-adjusted relative risk of fracture increased by 1.35 for each standard deviation decrease in BMD of the lumbar spine. Furthermore, the relationship between age-adjusted measurements of calcaneal and forearm bone density were very similar to those observed for the proximal femur and spine: the relative risks of fracture for a decrease of 1 SD ranged from 1.32 for the proximal radius to 1.51 for the calcaneus. In general, there appeared to be a uniform gradient of increasing risk from the highest to lowest quartile of BMD for all sites (Fig. 1). The results for wrist fracture are similar to those for all fractures. After age adjustment, the standardized relative risks are similar for BMD at all sites (Table 4). Specifically, the measurement of the distal radius was not a better predictor of wrist fracture (relative risk 1.64; 95% CI 1.132.37) than other sites, such as the lumbar spine (RR 1.56; CI 1.07-2.26), the femoral neck (RR 1.65; CI 1.12-2.41), or the calcaneus (RR 1.83; CI 1.26-2.64). Since the estimates are based on only 37 wrist fractures, however, the confidence intervals for all these measurements are wide. There were no significant differences in the areas under the ROC curves between any of the measurement sites for all fractures (proximal femur = 0.60, spine = 0.59, calcaneus = 0.62, distal radius = 0.62, and proximal radius = 0.59; P 2 0.15 for all comparisons) or for wrist fractures. Therefore, none of the measurements of bone mineral density in the proximal femur, spine, calcaneus, or radius were substantially or significantly better than any others for predicting all fractures or wrist fractures.
TABLE2. TYPESOF FRACTURES OBSERVED DURINGFOLLOW-UP
Age category Type of fracture
N (% of fractures)
65-69
70-74
75-79
80+
~
Distal radiudwrist Foot Ribs Humerus Ankle Toes Hip Intertrochanteric Femoral neck Other Other types (s10 each)
7 12 3 3 3 3 3 1 2
15 9 7 7 6 5 4 2 2
11 4 8 3 8 5 5 3 2
13
27
5
4 3 5 6
1 2 6 4 1
1 5
BLACK ET AL.
636
TABLE3. ASSOCIATION BETWEEN BONEMINERALDENS^
AND
ALL~RACTURESIN 8134 WOMEN Relative riska (95 % confidence interval)
Standard deviation Measurement (g/cm ’)
W)
Unadjusted
Adjusted for age
Proximal femur (total) Femoral neck Ward’s triangle Intertrochanteric Trochanteric Lumbar spine Calcaneus Distal radius Proximal radius
0.131 0.111 0.110 0.160 0.102 0.169 0.094 0.084 0.103
1.43 (1.23,1.66) 1.44 (1.23,1.69) 1.45 (1.25,1.69) 1.42 (1.22,1.64) 1.41 (1.21,1.64) 1.36 (1.17,1.60) 1.53 (1.31, 1.79) 1.45 (1.24,1.69) 1.35 (1.17,1.56)
1.40 (1.20,1.63) 1.41 (1.20,1.66) 1.43 (1.22,1.67) 1.39 (1.19,1.62) 1.38 (1.18,1.61) 1.35 (1.15,1.58) 1.51 (1.29,1.77) 1.42 (1.21,1.67) 1.32 (1.14,1.53)
.Relative hazards for bone mineral density (BMD) pcr standard deviation decrease in BMD.
We considered the possibility that the longer interval between SPA measurements and fractures may have weakened that association; however, the correlation between measurements of appendicular sites 2 years apart was r L 0.89,and thus the effect would be very small. For example, based on the observed relative risk of 1.42 for each standard deviation decrease in BMD of the distal radius, we estimate that the relative risk for measurements made 2 years later would not exceed 1.48.
60
40
!
rn
F
20
C 0
!
a,
a
DISCUSSION 0
0
Proximal Femur: Quartiles of BMD
0 0 7
L
60
a,
a a, 0
C
a,
2 u C -
40
20
0
1
2
3
4
Distal Radius: Quartiles of BMD
Incidence of all fractures by age-adjusted quarFIG. tile of bone mineral density (BMD).
We have demonstrated that reduced bone mineral density of the proximal femur and spine is strongly associated with an increased risk of all nonspine and of wrist fractures in older women. The strength of the relationship between BMD and all nonspine fractures in our cohort (range in relative risk from 1.3 to 1.5) appears to be similar to the relative risk of 1.5 reported by Hui et al.,”’ although somewhat lower than the range of values (1 3-1.9) reported by Wasnich et a l . “ O ’ However, the Wasnich study was performed on a younger cohort of Asian-American women. The association between radial BMD and wrist fractures in our cohort is somewhat higher than that found by Hui et al.[’’ We found that the magnitude of the association between BMD and all fractures or wrist fractures was similar, regardless of the site or method of measurement. This suggests that measurements of bone mass at the proximal femur, spine, and appendicular sites have similar predic.. tive value for the overall risk of nonspine fracture in postmenopausal women. Our results differ somewhat from those of Eastell and colleagues,[11’who found that wrist fractures were somewhat more strongly associated with BMD of the distal radius than BMD of the spine by dualphoton absorptiometry. This difference might be due, in part, to differences in the methods of the two studies. Eastell and colleague’s study was a retrospective case-con-
637
AXIAL BONE DENSITY AND NONSPINE FRACTURE TABLE4. ASSOCIATION BETWEENBONEMINERAL DENSITY AND WRIST FRACTURES IN 8134 WOMEN ~
~~
Relative rkka (95% confidence interval) Standard deviation Measurement (g/cm’) ~~~~
~
(SO)
Unadjusted
Adjusted for age
0.131 0.111
1.49 (1.06, 2.09) 1.62 (1.12, 2.35) 1.32 (0.94,1.86) 1.49 (1.06, 2.10) 1.39 (0.99.1.95) 1.57 (1 .W, 2.28) 1.80 (1.26, 2.57) 1.64 (1.14, 2.35) 1.79 (1.29, 2.47)
1.49 (1.05. 2.12) 1.65 (1.12, 2.41) 1.32 (0.93, 1.88) 1.51 (1.06, 2.14) 1.39 (0.98, 1.98) 1.56 (1.07, 2.26) 1.83 (1.26, 2.64) 1.64 (1.13, 2.37) 1.82 (1.30, 2.55)
~
Proximal femur (total) Femoral neck Ward’s triangle Intertrochanteric Trochanteric Lumbar spine Calcaneus Distal radius Proximal radius
0.110
0.160 0.102 0.169 0.094
0.084 0.103
PRelative hazards for bone mineral density (BMD) per standard deviation decrease in BMD.
trol study, and their site of measurement was somewhat more distal and closer to the usual site of Colles’ fracture than ours. The prospective design of this study avoids many of the potential biases inherent in case-control studies of the association between measurements of bone mass and fracture.tL0’Our study has several limitations, however. We recruited subjects from population-based listings in four cities across the United States, but participants in the Study of Osteoporotic Fractures are not a representative sample of older women; they are volunteers who are likely to be healthier than those who did not participate. Participants did not include women who were homebound or living in institutions, and our results might not apply to women living in long-term care institutions. Nevertheless, our findings are generalizable to the vast majority of white women. Finally, we were unable to determine whether DEXA of the hip has a better predictive value for hip fractures than measurements at other sites. Further follow-up of this cohort may allow us to address this issue. Measurements of the appendicular sites were made 2 years before DEXA of the axial sites. Thus, although it is theoretically possible that this may have caused us to underestimate the association between appendicular BMD and fracture in comparison with DEXA, we found that measurements of BMD made 2 years apart are so highly correlated that the 2 year difference between time of measurement could have, at most, a trivial effect on this comparison. The interval between measurement of bone mass and fracture in our study is very short. This provides a good assessment of the role of decreased bone density in the pathogenesis of fracture, but it will overestimate the predictive value of bone density for fractures over much longer periods of time. Finally, our observational study cannot exclude the possibility that low bone density is a marker for other causes of fracture, such as reduced quality of bone and accumulation of microscopic damage.(*a’
We conclude that there is an inverse relationship between both axial and appendicular bone mass and the risk of nonspine fractures in older women. BMD at all these sites appears to have similar predictive accuracy for all nonspine fractures and wrist fractures.
ACKNOWLEDGMENTS Supported by Public Health Service Grants I-ROI-AG05407, 1-R01-AR35582, 5-RO1-AG05394, 1-ROl-AM35584, and 1-R01-AR35583. Investigators in the Study of Osteoporotic Fractures research group: University of California, San Francisco (coordinating center): S.R. Cummings (principal investigator). M.C. Nevitt (project director), D. Black (study statistician), H.K. Genant (director, central radiology laboratory), C. Arnaud, W. Browner, K. Faulkner. C. Fox, C. Glueer, S. Harvey, S.B. Hulley, L. Palermo, D. Seeley, and P. Steiger. University of Maryland: R. Sherwin (principal investigator), J. Scott (project director), K. Fox (study coordinator), J. Lewis (clinic supervisor), M. Bahr, S. Trusty, B. Hohman, L. Emerson, D. Rebar, and E. Oliner. University of Minnesota: R. Grimm, Jr. (principal investigator), K. Ensrud (coinvestigator), C. Bell (projector director), D. Thomas, K. Jacobson, S. Jackson, E. Mitson, L. Stocke, and P. Frank. University of Pittsburgh: J.A. Cauley (principal investigator). L.H. Kuller (coprincipal investigator). L. Harper (project director), M. Nasim (clinic coordinator), C. Bashada, L. Buck, A. Githens. D. Medve, and S. Rudovsky. The Kaiser Permanente Center for Health Research, Portland, Oregon: T.M. Vogt (principal investigator), W.M. Vollmer, H. Glauber, E. Orwoll (coinvestigators), J. Blank (project director), B. Mastel-Smith (clinic coordinator), R. Bright, and J. Downing.
BLACK ET AL.
638
REFERENCES 1. Cummings SR. Black DM, Nevitt MC, et al. 1990 Appendicular bone density and age predict hip fracture in women. JAMA M3:665-668. 2. Gardsell P , Johnell 0.Nilsson BE 1989 Predicting fractures in women by using forearm bone densitometry. Calcif Tissue Int 44:235-242. 3. Hui SL, Slemenda CW, Johnston CC 1988 Age and bone mass as predictors of fracture in a prospective study. J Clin lnvest 81:1804-1809. 4. Ross PD. Wasnich RD, Heilbrun LK, Vogel JM 1987 Definition of a spine fracture threshold based upon prospective fracture risk. Bone 8:271-278. 5. Wasnich RD, Ross PD. Heilbrun LK, Vogel JM 1985 Prediction of postmenopausal fracture risk with use of bone mineral measurements. Am J Obstet Gynecol 153:745-751. 6. Cox DR 1972 Regression models and life tables (with discussion). J R Stat Soc B34:187-220. 7. Hanky JA, McNeil BJ 1983 A method of comparing the areas under receiver operation characteristic curves derived from the same cases. Radiology 148:839-843. 8. McNeil BJ. Hanky JA 1984 Statistical approaches to the analysis of receiver operating characteristic curves. Med Decis Making 4:137-150.
9. Rosner B, Willett WC, Spiegelman D 1989 Correction of logistic regression relative risk estimates and confidence intervals for systematic within-pcrson measurement error. Stat Med 8:1051-1069. 10. Wasnich RD. Ross PD, Heilbrun LK, Vogel JM 1985 Prediction of postmenopausal fracture risk with use of bone mineral measurements. Am J Obstet Gynecol 153(7):745-751. 11. Eastell R, Wahner HW, O’Fallon WM, Amadio PC, Melton LJ, Riggs BL 1989 Unequal decrease in bone density of lumbar spine and ultradistal radius in Colles’ and vertebral fracture syndromes. J Clin lnvest 83:168-174. 12. Parfitt AM 1987 Trabecular bone architecture in the pathogenesis and prevention of fracture. Am J Med 82(suppl 18): 68-72.
Address reprint requests to: Dennis M . Black, Ph.D. Box 0886 University of California Medical Center Son Francisco, C A 94143-0886 Received for publication March 25, 1991; in revised form November 26. 1991; accepted November 26. 1991
