Predictors of Bone Mass in Perimenopausal Women A Prospective Study of Clinical Data Using Photon Absorptiometry Charles W. Slemenda, DrPH; Siu L. Hui, PhD; Christopher Longcope, MD; Henry Wellman, MD; and C. Conrad Johnston, Jr, M D
Study Objective: To determine whether clinically available data on risk factors are adequate to identify perimenopausal women with either low or high bone mass. Design: Cross-sectional observational study of a cohort of perimenopausal women (mean age, 50.8 years). Setting: Community volunteers in a university hospital. Subjects: One hundred twenty-four white volunteers established as perimenopausal by history and serum concentrations of estrogens and follicle-stimulating hormone. Measurements and Main Results: Models were constructed to predict bone mass in the radius, lumbar spine, and hip using risk factors (age, height, weight, calcium and caffeine intake, alcohol and tobacco use, and urinary markers of bone turnover). Although highly significant predictive models were developed for all skeletal sites, none of the models correctly identified more than 70% of women with low bone mass at any site. However, for the radius, a model was constructed that never overestimated bone mass by more than 0.10 g/cm. A small subgroup ( 7 % ) with short stature, low body weight, low calcium intake, and who were heavy smokers always had low radial bone mass. Using these models, about 30% of our population could be assessed without bone mass measurements. Predictions for the spine and femur were less efficient, suggesting that direct measurements are required if therapy decisions are to be based on bone mass at these sites. Conclusions: Risk factors for osteoporosis are of limited use in identifying women with low bone mass around the time of menopause. Measurements of bone mass are probably necessary if the risk for osteoporosis is to be the basis for deciding on estrogen replacement therapy.
Annals of Internal Medicine. 1990:112:96-101. From Indiana University School of Medicine, Indianapolis, Indiana; and University of Massachusetts Medical School, Worcester, Massachusetts. For current author addresses, see end of text.
96
© 1990 American College of Physicians
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
Osteoporosis exacts a huge toll in suffering and health care costs; hip fractures are the most serious and costly outcome of this process ( 1 ) . There has been considerable debate about the relation of bone mass to osteoporotic fractures and the value of bone mass measurements in the diagnosis of osteoporosis. Bone mass measurement to screen all perimenopausal women has been generally rejected (2-10), but several recent prospective studies have clearly shown that low bone mass is predictive of fracture (11, 12), and one study has shown that a single bone mass measurement can predict the probability of hip fracture (13). Because of the role of bone mass in osteoporotic fracture, and because there are effective therapies (such as estrogen and calcitonin) for preserving existing bone mass, identifying low bone mass is important in determining who is most likely to benefit from therapy. Measurements of bone mass using the common techniques (single- and dual-photon absorptiometry, dual-energy roentgenographs absorptiometry, and quantitative computed tomography) can provide precise estimates of mineral content at many skeletal sites, including the radius, ulna, lumbar spine, hip, and os calcis, as well as total body mineral content. How effectively clinical values may be used to reduce the proportion of persons requiring bone mass measurements to establish fracture risk is not clear. Riggs and Melton (14) have suggested that women at high risk may be identified through an analysis of risk factors for osteoporosis and that these patients should then have further studies, including measurements of bone mass. However, the risk factors for osteoporosis have not been clearly quantified. Who is thin? How much must a woman smoke to be at increased risk? Are these risk factors independent (for example, smoking and weight)? Can knowledge of a patient's height, weight, tobacco and alcohol consumption, dietary calcium intake, and other potential risk factors be used, without bone mass measurement, to identify patients with greatly increased risk for having low bone mass and who are, therefore, likely to benefit from effective therapy. Can this knowledge be used to identify patients who probably have bone mass so high that there would be no obvious benefit from further tests or treatment? We try to answer these questions for women around the age of menopause, when treatment is known to be effective.
addition to these risk factors, we have added serum and urine markers of bone turnover and serum chemistry values (see laboratory measurements below). Height and weight were measured to the nearest 0.5 cm using a wall-mounted headboard and 0.1 kg using an electronic scale; body mass index was calculated as kilograms per square meter. Cigarette smoking was assessed by questionnaires requesting information on years of smoking and average and maximum packs per day. Alcohol consumption was estimated from a standardized questionnaire that focused on beer, wine, and hard liquors in separate categories. Dietary calcium was estimated from a food-frequency questionnaire that included the 14 foods that account for more than 9 5 % of calcium intake in the American diet ( 1 8 ) . The correlation between repeated calcium questionnaires from the same person at 4month intervals was r = 0.84 (n = 9 6 ) . This questionnaire correlated with 3-day diet diaries (based on 2 weekdays and 1 weekend day) at r > 0.60 ( P < 0.001, n = 57). Associations with bone mass, using the questionnaire or the 3-day diary, did not differ. Caffeine was also estimated from foodfrequency questionnaires that detailed intake of coffee, tea, and soft drinks, and differentiated caffeine-containing from decaffeinated beverages. Questionnaires rather than diet diaries were used to enhance the clinical applicability of our approach. We took personal and family fracture histories, but these did not contribute to the prediction models. Technicians made anthropometric measurements under the supervision of a physical anthropologist. These measurements included estimates of frame size (including bicofemoral width and wrist b r e a d t h ) , circumference measurements (hip, waist, chest), and skinfold thicknesses at several sites. The coefficients of variation for the anthropometric measurements were always less than 8%.
Methods Subjects We originally recruited 84 women around the age of menopause to participate in a longitudinal study of sex steroids and perimenopausal bone loss. Sixty-five of these women had very low estrogens (mean estradiol, < 32 p g / m L ) and elevated follicle-stimulating hormone ( F S H ) concentrations ( > 40 m l U / m L ) , whereas the remaining 19 women had higher concentrations of estrogens (mean estradiol, 90 p g / m L ) and only occasionally elevated concentrations of FSH. None were cycling regularly. We recruited another 40 menopausal women later. These women also were cycling irregularly at entry to the study and resembled the 65 perimenopausal subjects in estrogen and F S H ( > 40 m l U / m L ) concentrations (15, 16). T h e F S H concentrations were measured using standard kits from Serono Laboratories Company (Braintree, Massachusetts). All of the women were white, not taking estrogens or other medications thought to affect bone (including corticosteroids, insulin, and anticonvulsants), and were in good health. Bone Measurements Single-photon absorptiometry (Norland Instruments, I N O Tech, Fort Atkinson, Wisconsin) was used to estimate bone mineral content at the midshaft and distal radius as previously described ( 1 7 ) . These measurements were begun on the original 84 women at entry to the study; all data on radial measurements related to these women. Bone mineral content, corrected for area scanned (yielding bone mineral density), of the spine and hip was estimated by dual-photon absorptiometry ( L u n a r Radiation, Madison, Wisconsin). The measurements were made over a period of almost 2 years; the gadolinium-153 source was changed at approximately 6-month intervals and the iodine sources at 4-month intervals. These measurements were first made when the initial group had been in the study approximately 3 years (range, 24 to 44 m o n t h s ) . Based on measurements made on the same day, the correlations between sites were between 0.42 and 0.60 (radius to spine r = 0.49; radius to hip r = 0.42; spine to hip r = 0.60). Repeated measurements at the same site were always strongly correlated ( r > 0.9). The short-term precisions (coefficient of variation of repeat measurements made over a short period, usually 1 week) of these methods were 2 % to 3 % for the midshaft and distal radius, 1.6% for the lumbar spine, and 2.4% for the femoral neck.
Laboratory Measurements T o improve the likelihood of correctly classifying women with low bone mass, we measured variables that might properly be referred to as markers rather than risk factors. Urinary calcium, hydroxyproline, and creatinine were measured at the baseline examination using standard methods (1921). The urine samples used in the prediction of radial bone mass were 24-hour collections. The later collections for the prediction of lumbar spine and hip bone mineral density were from second morning voids, collected after an overnight fast. The correlation between the baseline measurement and the overnight collection taken at the time of the first spine measurement (approximately 3 years later) was 0.48 (P < 0.001).
Risk Factors
Statistical Analyses
Risk factors for our study were based on factors suggested by Riggs and Melton (14) and on various other sources. In
Because the radius and the spine and hip measurements were begun at different times (about 3 years a p a r t ) , the risk
Table 1. Values of Risk Factors in Women with Low Bone
Mass*
Variables
Midshaft Radius High Low
Low
162 62.5 51.6 16.7 14 15.9 676 5.5 0.122 0.418
163 61.7 50.1 11.5 12 16.5 739 5.0 0.105 0.376
Height, cm Weight, kg Age,y Months since last menstrual period Flushing, episodes/mo Smoking, pack-years Dietary calcium, mg/d Alcohol intake, g/d Urinary hydroxyproline/creatinine, mmol/mmol Urinary calcium/creatinine, mmol/mmol
166t 68.4f 50.6 7.8f 11 9.1% 790 7.4 0.109 0.367
Lumbar Spine High 164 68.8| 51.1 10.5 12 9.5 682 7.1 0.097 0.449
Femoral Neck Low High 164 63.8 50.8 10.9 12 15.0 683 6.4 0.105 0.492
164 67.2 51.2 11.3 12 10.3 746 5.6 0.088 0.347
* Persons in the bottom one-third for the specified sites (see Methods) compared with persons with greater bone mass. fP < 0.01. % P < 0.05 for low bone mass compared with higher bone mass by two-tailed Mest.
15 January 1990 • Annals of Internal Medicine • Volume 112 • Number 2
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
97
Table 2. Correlation
Coefficients
Variable Height Weight Age Smoking, pack-years Dietary calcium Urinary hydroxyproline/ creatinine Urinary calcium/creatinine Wrist breadth
Midshaft Lumbar Femoral Radius Spine Neck + 0.53* + 0.49* -0.20t -0.11 -0.03
+ 0.17f + 0.32* -0.09 -0.10 -0.12
+ 0.04 + 0.26J + 0.06 -0.04 + 0.06
-0.06 -0.17f + 0.45*
-0.04 -0.15t + 0.19§
-0.02 -0.23J + 0.15
* P < 0.001. + 0.05 < P < 0.10. XP < 0.01. § P < 0.05.
factors used to predict bone mineral content or bone mineral density were those collected at the time of the bone measurement. For example, the serum calcium concentrations used to predict radius bone mineral content were obtained from blood drawn at the time of the baseline examination; the serum calcium measurements used to predict lumbar spine bone mineral density and hip bone mineral density were obtained at the time of those measurements. We examined risk factors for low bone mass by comparing women with low bone mass with the remainder of the study population. Low bone mass was defined as any value in the lowest tertile of bone mineral content at each site (radius, < 0.778 g/cm midshaft or < 0.845 g/cm distal) or bone mineral density at each site (spine, < 1.056 g/cm 2 ; hip, < 0.80 g/cm 2 ). We did Mests on all potential risk factors using low bone mass as the class variable. Univariately significant variables (P < 0.10) were entered into general linear models to derive prediction equations for bone mass. The best possible model was made (the model that correctly classified the greatest proportion of perimenopausal women with low bone mass at each site). Thus, the predictors used varied from model to model. Finally, to create a simple rule for clinical use, scores were assigned to risk factors. The scores were based on the strength (the coefficient of the predictor and the standard error of the predictor) of each factor in predicting bone mass at a particular site. These scores were rounded to make the scoring system potentially useful for clinical practice (see Appendix Table for scores and equations).
en with higher bone mass. They had also smoked more (P < 0.05) and consumed about 100 m g / d less calcium ( P is not significant). Women with lower bone mass in the spine were 7 kg lighter ( P < 0.01) and had greater cumulative exposure to cigarette smoking, but did not differ significantly in other risk factors. Women with low femoral neck bone mass showed trends similar to women with low bone mass in the other sites studied in body weight and cigarette smoking, and they also consumed about 70 m g / d less calcium; however, none of these trends were significant. We then constructed general linear models to properly weight each of the many independent variables in multiple variable equations to maximize the correct classification of women with low bone mass. The simple correlation coefficients for independent variables that entered the regression equations are shown in Table 2. The regression equations and the results of these models are as follows: For midshaft radius bone, mineral content = -1.1 + (0.0075 X height) + (0.0022 X weight) — (0.00044 X pack-years) + (0.12 X wrist breadth) — (0.00016 X calcium/creatinine) + (0.00004 X dietary calcium). This model correctly identifies 6 8 % of persons with low bone mass and 7 7 % of persons with high bone mass. For lumbar spine bone, mineral density = + 0.69 + (0.00099 X height) + (0.0043 X weight) — (0.00046 X pack-years) — (0.00020 X calcium/ creatinine) + (0.00002 X dietary calcium). This model correctly identifies 6 1 % of persons with low bone mass and 4 5 % of persons with high bone mass. For femoral neck bone, mineral density = + 0.79 + (0.001 X height) + (0.0041 X weight) (0.00018 X pack-years) (0.00023 X calcium/creatinine) + (0.00005 X dietary calcium). This model
Results The study population at baseline had a mean age of 51 years (range, 44 to 55 years). The average height (164 cm) and weight (65 kg) were similar to published means for random population samples of women of similar ages. They consumed 710 m g / d of calcium, about 100 mg less than the recommended daily allowance, and were very light consumers of alcohol (6 g / d ) and tobacco (13 pack-years cumulative mean lifetime exposure). Slightly less than 5 0 % were more than 6 months from their last menstrual period at entry to the study, and none were more than 3 years postmenopausal. Table 1 shows the values for the risk factors studied for women with low bone mass in the radius, spine, or hip compared with women who have greater bone mass (as defined in Methods). Women with lower bone mass at the midshaft radius were 4 cm shorter (P < 0.01), 6 kg lighter (P < 0.01), and were nearly 1 year further past menopause (P < 0.01) than wom-
98
15 January 1990 • Annals
of Internal Medicine
Figure 1: Observed radial bone mass plotted against predicted. The line of identity is shown. BMC = bone mineral content.
• Volume 112 • N u m b e r 2
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
We then investigated whether the stereotypic highrisk woman (defined as small [ < 165 c m ] , thin [ < 58 k g ] , a smoker [ > 20 pack-years of cumulative exposure], with inadequate dietary calcium [ < 800 m g / d ] ) had low bone mass. These women represented less than 7% of the population studied, but all of them had radial bone mass of less than 0.76 g/cm and, as a group, they averaged 15% less radial bone mass than the remainder of the population. At the spine, this small group averaged 3% less bone. At the femoral neck, the difference was slightly less than 10%. Discussion
Figure 2. Observed radial bone mass plotted against risk factor score (see Appendix Table). BMC = bone mineral content.
correctly identifies 65% of persons with low bone mass and 53% of persons with high bone mass. All of the models were statistically significant with R 2 values ranging from 17% (femoral neck bone mineral density) to 47% (midshaft radius bone mineral content). For example, at the midshaft radius, the general linear model predicts that a subject 1 SD taller (6 cm) than the mean height would have 0.048 g/cm (5.8%) greater than average bone mass. Similarly, a 1 SD increase in weight (12 kg) would yield a 0.035 g/cm (4.2%) increase in bone mass, and 1 SD increase in calcium to creatinine ratio would yield a 0.028 g/cm (3.4%) decrease in bone mass. Figure 1 shows the predicted bone mass values for the midshaft radius plotted against the actual values; more than 50% of the subjects were within 5% (0.42 SD) of their actual bone mass value. Figure 1 also shows that none of the women with predicted bone mass greater than 0.9 g/cm had an actual bone mass lower than 0.8 g/cm. To simplify the system for classification of subjects, scores based on the regression coefficients (see Appendix Table) were constructed for risk factors. Figures 2, 3, and 4 show the observed bone mass for the radius, spine, and femoral neck plotted against the risk scores. For the radius, 18 women had scores above 17, and none of these women had low bone mass ( < 0.78 g / c m ) . Only 1 of the 7 women with radius scores of 17 had low bone mass. In contrast, 6 of the 7 women with the lowest scores (12 or less) had low bone mass. For the spine and hip, however, the scoring system was less efficient. At the spine, 4 of the 14 women (29%) with the highest scores (12 and 13) had low bone mass that would be missed if this scoring system were used. At the femoral neck, 5 of the 24 women with the highest scores (9 and 10) had low bone mass.
We intended to determine whether clinically available information could be used to identify women at high risk for having low bone mass around the time of menopause. Although low bone mass is a strong risk factor for fractures, including hip fracture (13), this paper addresses only low bone mass, not fracture risk. Although an approach addressing fracture risk has been recommended (14), quantification of risk factors that would allow physicians to ascertain risk, either with risk scores or regression equations, has not previously been done. We have provided both risk scores and regression equations for a group of perimenopausal women. The data from our study clearly show that the analysis of multiple risk factors in perimenopausal women is an inadequate substitute for bone mass measurements. This conclusion is based on two findings. The identification of high-risk women, missing more than 30% of women with low bone mass at any site, was not sensitive enough. In addition, the sites where most important osteoporotic fractures occur (the spine and hip) were associated with greater errors in prediction. Although untested in other data sets, these predic-
Figure 3. Observed lumbar spine bone mineral density plotted against risk factor score. BMD = bone mineral density.
15 January 1990 • Annals of Interna1 Medicine
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
• Volume 112 • N u m b e r 2
99
cannot be obtained through other clinically available data. The only therapeutic approaches to the prevention of osteoporotic fractures are approaches that preserve bone mass. Clinicians can, therefore, only focus on identifying and treating those patients with low bone mass. Correctly identifying 7 0 % of women with low bone mass would be considered good in the absence of better methods, but direct measurements obviously identify nearly all women with low bone mass. Having low bone mass at any site greatly increases the likelihood both of having low bone mass at other sites (correlations range from 0.4 to 0.7) and of fracture; however, bone mass differs enough between sites to necessitate measurements at the site of interest if, for example, the prevention of hip fracture is the goal of therapy. However, whether measurements of the site of interest (for example, the hip) are better predictors of fractures at that site than are measurements at some other site remains to be shown, although this seems to be likely. Estrogen therapy prevents bone loss and reduces fracture risk, but has some risks. Knowledge of a patient's bone mass is currently the only basis for the application of estrogen therapy to prevent osteopenia. We have shown that, for most patients, risk factors fail to provide adequate information upon which to base such a decision. However, risk factors may possibly be Figure 4. Observed femoral neck bone mineral density plotted against risk factor score. BMD = bone mineral density.
Appendix Table
tion models may, nonetheless, be useful in reducing the number of women needing bone mass measurements. Although the identification of the high-risk population (women with low bone mass) was not efficient enough for clinical use ( > 3 0 % misidentified), this approach could be used without bone mass measurements in two areas. First, none of the women with predicted radial bone mass greater than 0.9 g / c m or scores greater than 17 had low bone mass when measured. Therefore, women predicted to have high bone mass (about 2 0 % of our sample) might be excluded from further measurements based on their reduced probability of having low bone mass. Because radial bone mass predicts fractures (12, 13), this approach would reduce the population needing a measurement. Second, small, light smokers with low calcium intake might be considered at high risk for having low radial bone mass without resorting to actual measurements of bone mass. The risk factor approach has been unsuccessful in distinguishing between active remodeling and inactive osteoporosis patients (22), although comparisons between patients and controls were not made. The implications of these data for clinicians are straightforward. In evaluating risk for osteoporosis in menopausal women, clinically available data are of limited utility in identifying women with the lowest bone mass. For most candidates for estrogen therapy, a single bone mass measurement at any site provides information on which to base a therapeutic decision, information that
100
Variable Range Height, cm
Weight, kg
Calcium/creatinine ratio Dietary calcium, g/d Cigarette smoking, pack-years Wrist breadth, cm
Hip
< 159 159 to 161 162 to 164 165 to 167 168 to 170 171 + < 59 59 to 63 64 to 68 69 to 73 74 +
1 2 3 4 5 6 1 2 3 4 5
1 1 1 2 2 2 1 2 3 4 5
NAf NA NA NA NA NA 1 2 3 4 5
< 0.367 0.367 to 0.477 0.478 +
3 2 1
3 2 1
3 2 1
1+
1 2
NA NA
1 2
< 2 2 to 14 15 + < 4.9 4.9 to 5.2 5.3 +
3 2 1 1 2 3
3 2 1 NA NA NA
NA NA NA NA NA NA
< 1
* Scores were assigned according to the strength of the predictors in the general linear models in predicting bone mass at each site. Thus, strong predictors (such as height at the midshaft radius) were assigned higher scores than weak predictors. For example, the stereotypical osteoporotic patient described in the text would have a minimum radius score of 6 (1 point in each category) and a maximum of 10 (with the highest scores possible in the unmeasured categories, calcium to creatinine ratio and wrist breadth). + NA = not applicable.
15 January 1990 • Annals of Internal Medicine • Volume 112 • Number 2
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
Score* Radius Spine
useful in reducing the number of women needing bone mass measurements. Grant Support: By NIH grants AG02927, P01AG05793, and RR00750, and a grant from the Ancient and Accepted Scottish Rite, Valley of Indianapolis. Requests for Reprints: Charles W. Slemenda, Dr PH, Regenstrief Health Center, Fifth Floor, 1001 West Tenth Street, Indianapolis, IN 462022859. Current Author Addresses: Drs. Slemenda and Hui: Regenstrief Institute, R G / 5 t h floor, 1001 W. 10th St., Indianapolis, IN 46202-2859. Dr. Longcope: University of Massachusetts Medical School, 55 Lake Avenue North, Worcester, MA 01655. Dr. Wellman: Indiana University School of Medicine, UH P 018, 926 W. Michigan St., Indianapolis, IN 46202-5253. Dr. Johnston: Indiana University School of Medicine, EM 421, 545 Barnhill Dr., Indianapolis, IN 46202-5124.
References 1. Baker SP, Harvey AH. Fall injuries in the elderly. Clin Geriatr Med. 1985;1:501-12. 2. Cummings SR, Kelsey JL, Nevitt MC, O'Dowd KJ. Epidemiology of osteoporosis and osteoporotic fractures. Epidemiol Rev. 1985;7:178-208. 3. Ott S. Should women get screening bone mass measurements? Ann Intern Med. 1986;104:874-6. 4. Hall FM, Davis MA, Baran DT. Bone mineral screening for osteoporosis. N Engl J Med. 1987;316:212-4. 5. Davis M R . Screening for postmenopausal osteoporosis. Am J Obstet Gynecol. 1987;156:1-5. 6. Riggs BL, Melton L. (A reply from authors to Letters to the Editor on involutional osteoporosis.) NEngl J Med. 1987;316:216-7. 7. Sites VR. Involutional osteoporosis. [Letter]. AT Engl J Med. 1987;316:215. 8. Raisz LG, Lorenzo J A, Smith J . Bone mineral screening for osteo-
porosis. [Letter]. N Engl J Med. 1987;317:315-6. 9. Tohme J F , Lindsay R. Bone mineral screening for osteoporosis. [Letter]. N Engl J Med. 1987;317:316. 10. Hall FM. Bone mineral screening for osteoporosis. [Letter]. NEngl J Med. 1987;317:316. 11. Ross P D , Wasnich RD, Heilbrun LK, Vogel J M . Definition of a spine fracture threshold based upon prospective fractive risk. Bone. 1987;8:271-8. 12. Hui SL, Slemenda CW, Johnston CC Jr. Age and bone mass as predictors of fracture in a prospective study. / Clin Invest. 1988;81:1804-9. 13. Hui SL, Slemenda CW, Johnston CC Jr. Baseline measurement of bone mass predicts fracture in white women. Ann Intern Med. 1989;111(5):355-61. 14. Riggs BL, Melton LJ. Involutional osteoporosis. N Engl J Med. 1986;314:1676-86. 15. Longcope C, Watson D, Williams KI. The effects of synthetic estrogens on the metabolic clearance and production rates of estrone and estradiol. Steroids. 1974;24:15-30. 16. Pratt JH, Longcope C. Effect of adrenocorticotropin on production rates and metabolic clearance rates of testosterone and estradiol. J Clin Endocrinol Metab. 1978;47:307-13. 17. Slemenda C, Hui SL, Longcope C, Johnston CC. Sex steroids and bone mass. / Clin Invest. 1987;80:1261-9. 18. Block G, Dresser CM, Hartman AM, Carroll MD. Nutrient sources in the American diet: quantitative data from the N H A N E S II survey. I. Vitamins and minerals. Am I Epidemiol. 1985;122:13-26. 19. Zettner A J, Seligson D. Application of atomic absorption spectrophotometry in the determination of calcium in serum. Clin Chem. 1964;10:869-90. 20. Kivirikko KL, Laitinen O, Prockop DJ. Modifications of a specific assay for hydoxproline in urine. Anal Biochem. 1967;19:249-55. 21. Bartels E, Cikes M . Uber chromogene der kreatininbestimmung nachjaffe. Clin Chim Acta. 1969;26:1-10. 22. Whyte M P , Bergfeld MA, Murphy WA, Avioli LV, Teitelbaum SL. Postmenopausal osteoporosis. A heterogeneous disorder as assessed by histomorphometric analysis of iliac crest bone from untreated patients. Am J Med. 1982;72:193-202.
15 January 1990 • Annals
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
of Internal Medicine
• Volume 112 • N u m b e r 2
101
Study Objective: To determine whether clinically available data on risk factors are adequate to identify perimenopausal women with either low or high bone mass. Design: Cross-sectional observational study of a cohort of perimenopausal women (mean age, 50.8 years). Setting: Community volunteers in a university hospital. Subjects: One hundred twenty-four white volunteers established as perimenopausal by history and serum concentrations of estrogens and follicle-stimulating hormone. Measurements and Main Results: Models were constructed to predict bone mass in the radius, lumbar spine, and hip using risk factors (age, height, weight, calcium and caffeine intake, alcohol and tobacco use, and urinary markers of bone turnover). Although highly significant predictive models were developed for all skeletal sites, none of the models correctly identified more than 70% of women with low bone mass at any site. However, for the radius, a model was constructed that never overestimated bone mass by more than 0.10 g/cm. A small subgroup ( 7 % ) with short stature, low body weight, low calcium intake, and who were heavy smokers always had low radial bone mass. Using these models, about 30% of our population could be assessed without bone mass measurements. Predictions for the spine and femur were less efficient, suggesting that direct measurements are required if therapy decisions are to be based on bone mass at these sites. Conclusions: Risk factors for osteoporosis are of limited use in identifying women with low bone mass around the time of menopause. Measurements of bone mass are probably necessary if the risk for osteoporosis is to be the basis for deciding on estrogen replacement therapy.
Annals of Internal Medicine. 1990:112:96-101. From Indiana University School of Medicine, Indianapolis, Indiana; and University of Massachusetts Medical School, Worcester, Massachusetts. For current author addresses, see end of text.
96
© 1990 American College of Physicians
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
Osteoporosis exacts a huge toll in suffering and health care costs; hip fractures are the most serious and costly outcome of this process ( 1 ) . There has been considerable debate about the relation of bone mass to osteoporotic fractures and the value of bone mass measurements in the diagnosis of osteoporosis. Bone mass measurement to screen all perimenopausal women has been generally rejected (2-10), but several recent prospective studies have clearly shown that low bone mass is predictive of fracture (11, 12), and one study has shown that a single bone mass measurement can predict the probability of hip fracture (13). Because of the role of bone mass in osteoporotic fracture, and because there are effective therapies (such as estrogen and calcitonin) for preserving existing bone mass, identifying low bone mass is important in determining who is most likely to benefit from therapy. Measurements of bone mass using the common techniques (single- and dual-photon absorptiometry, dual-energy roentgenographs absorptiometry, and quantitative computed tomography) can provide precise estimates of mineral content at many skeletal sites, including the radius, ulna, lumbar spine, hip, and os calcis, as well as total body mineral content. How effectively clinical values may be used to reduce the proportion of persons requiring bone mass measurements to establish fracture risk is not clear. Riggs and Melton (14) have suggested that women at high risk may be identified through an analysis of risk factors for osteoporosis and that these patients should then have further studies, including measurements of bone mass. However, the risk factors for osteoporosis have not been clearly quantified. Who is thin? How much must a woman smoke to be at increased risk? Are these risk factors independent (for example, smoking and weight)? Can knowledge of a patient's height, weight, tobacco and alcohol consumption, dietary calcium intake, and other potential risk factors be used, without bone mass measurement, to identify patients with greatly increased risk for having low bone mass and who are, therefore, likely to benefit from effective therapy. Can this knowledge be used to identify patients who probably have bone mass so high that there would be no obvious benefit from further tests or treatment? We try to answer these questions for women around the age of menopause, when treatment is known to be effective.
addition to these risk factors, we have added serum and urine markers of bone turnover and serum chemistry values (see laboratory measurements below). Height and weight were measured to the nearest 0.5 cm using a wall-mounted headboard and 0.1 kg using an electronic scale; body mass index was calculated as kilograms per square meter. Cigarette smoking was assessed by questionnaires requesting information on years of smoking and average and maximum packs per day. Alcohol consumption was estimated from a standardized questionnaire that focused on beer, wine, and hard liquors in separate categories. Dietary calcium was estimated from a food-frequency questionnaire that included the 14 foods that account for more than 9 5 % of calcium intake in the American diet ( 1 8 ) . The correlation between repeated calcium questionnaires from the same person at 4month intervals was r = 0.84 (n = 9 6 ) . This questionnaire correlated with 3-day diet diaries (based on 2 weekdays and 1 weekend day) at r > 0.60 ( P < 0.001, n = 57). Associations with bone mass, using the questionnaire or the 3-day diary, did not differ. Caffeine was also estimated from foodfrequency questionnaires that detailed intake of coffee, tea, and soft drinks, and differentiated caffeine-containing from decaffeinated beverages. Questionnaires rather than diet diaries were used to enhance the clinical applicability of our approach. We took personal and family fracture histories, but these did not contribute to the prediction models. Technicians made anthropometric measurements under the supervision of a physical anthropologist. These measurements included estimates of frame size (including bicofemoral width and wrist b r e a d t h ) , circumference measurements (hip, waist, chest), and skinfold thicknesses at several sites. The coefficients of variation for the anthropometric measurements were always less than 8%.
Methods Subjects We originally recruited 84 women around the age of menopause to participate in a longitudinal study of sex steroids and perimenopausal bone loss. Sixty-five of these women had very low estrogens (mean estradiol, < 32 p g / m L ) and elevated follicle-stimulating hormone ( F S H ) concentrations ( > 40 m l U / m L ) , whereas the remaining 19 women had higher concentrations of estrogens (mean estradiol, 90 p g / m L ) and only occasionally elevated concentrations of FSH. None were cycling regularly. We recruited another 40 menopausal women later. These women also were cycling irregularly at entry to the study and resembled the 65 perimenopausal subjects in estrogen and F S H ( > 40 m l U / m L ) concentrations (15, 16). T h e F S H concentrations were measured using standard kits from Serono Laboratories Company (Braintree, Massachusetts). All of the women were white, not taking estrogens or other medications thought to affect bone (including corticosteroids, insulin, and anticonvulsants), and were in good health. Bone Measurements Single-photon absorptiometry (Norland Instruments, I N O Tech, Fort Atkinson, Wisconsin) was used to estimate bone mineral content at the midshaft and distal radius as previously described ( 1 7 ) . These measurements were begun on the original 84 women at entry to the study; all data on radial measurements related to these women. Bone mineral content, corrected for area scanned (yielding bone mineral density), of the spine and hip was estimated by dual-photon absorptiometry ( L u n a r Radiation, Madison, Wisconsin). The measurements were made over a period of almost 2 years; the gadolinium-153 source was changed at approximately 6-month intervals and the iodine sources at 4-month intervals. These measurements were first made when the initial group had been in the study approximately 3 years (range, 24 to 44 m o n t h s ) . Based on measurements made on the same day, the correlations between sites were between 0.42 and 0.60 (radius to spine r = 0.49; radius to hip r = 0.42; spine to hip r = 0.60). Repeated measurements at the same site were always strongly correlated ( r > 0.9). The short-term precisions (coefficient of variation of repeat measurements made over a short period, usually 1 week) of these methods were 2 % to 3 % for the midshaft and distal radius, 1.6% for the lumbar spine, and 2.4% for the femoral neck.
Laboratory Measurements T o improve the likelihood of correctly classifying women with low bone mass, we measured variables that might properly be referred to as markers rather than risk factors. Urinary calcium, hydroxyproline, and creatinine were measured at the baseline examination using standard methods (1921). The urine samples used in the prediction of radial bone mass were 24-hour collections. The later collections for the prediction of lumbar spine and hip bone mineral density were from second morning voids, collected after an overnight fast. The correlation between the baseline measurement and the overnight collection taken at the time of the first spine measurement (approximately 3 years later) was 0.48 (P < 0.001).
Risk Factors
Statistical Analyses
Risk factors for our study were based on factors suggested by Riggs and Melton (14) and on various other sources. In
Because the radius and the spine and hip measurements were begun at different times (about 3 years a p a r t ) , the risk
Table 1. Values of Risk Factors in Women with Low Bone
Mass*
Variables
Midshaft Radius High Low
Low
162 62.5 51.6 16.7 14 15.9 676 5.5 0.122 0.418
163 61.7 50.1 11.5 12 16.5 739 5.0 0.105 0.376
Height, cm Weight, kg Age,y Months since last menstrual period Flushing, episodes/mo Smoking, pack-years Dietary calcium, mg/d Alcohol intake, g/d Urinary hydroxyproline/creatinine, mmol/mmol Urinary calcium/creatinine, mmol/mmol
166t 68.4f 50.6 7.8f 11 9.1% 790 7.4 0.109 0.367
Lumbar Spine High 164 68.8| 51.1 10.5 12 9.5 682 7.1 0.097 0.449
Femoral Neck Low High 164 63.8 50.8 10.9 12 15.0 683 6.4 0.105 0.492
164 67.2 51.2 11.3 12 10.3 746 5.6 0.088 0.347
* Persons in the bottom one-third for the specified sites (see Methods) compared with persons with greater bone mass. fP < 0.01. % P < 0.05 for low bone mass compared with higher bone mass by two-tailed Mest.
15 January 1990 • Annals of Internal Medicine • Volume 112 • Number 2
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
97
Table 2. Correlation
Coefficients
Variable Height Weight Age Smoking, pack-years Dietary calcium Urinary hydroxyproline/ creatinine Urinary calcium/creatinine Wrist breadth
Midshaft Lumbar Femoral Radius Spine Neck + 0.53* + 0.49* -0.20t -0.11 -0.03
+ 0.17f + 0.32* -0.09 -0.10 -0.12
+ 0.04 + 0.26J + 0.06 -0.04 + 0.06
-0.06 -0.17f + 0.45*
-0.04 -0.15t + 0.19§
-0.02 -0.23J + 0.15
* P < 0.001. + 0.05 < P < 0.10. XP < 0.01. § P < 0.05.
factors used to predict bone mineral content or bone mineral density were those collected at the time of the bone measurement. For example, the serum calcium concentrations used to predict radius bone mineral content were obtained from blood drawn at the time of the baseline examination; the serum calcium measurements used to predict lumbar spine bone mineral density and hip bone mineral density were obtained at the time of those measurements. We examined risk factors for low bone mass by comparing women with low bone mass with the remainder of the study population. Low bone mass was defined as any value in the lowest tertile of bone mineral content at each site (radius, < 0.778 g/cm midshaft or < 0.845 g/cm distal) or bone mineral density at each site (spine, < 1.056 g/cm 2 ; hip, < 0.80 g/cm 2 ). We did Mests on all potential risk factors using low bone mass as the class variable. Univariately significant variables (P < 0.10) were entered into general linear models to derive prediction equations for bone mass. The best possible model was made (the model that correctly classified the greatest proportion of perimenopausal women with low bone mass at each site). Thus, the predictors used varied from model to model. Finally, to create a simple rule for clinical use, scores were assigned to risk factors. The scores were based on the strength (the coefficient of the predictor and the standard error of the predictor) of each factor in predicting bone mass at a particular site. These scores were rounded to make the scoring system potentially useful for clinical practice (see Appendix Table for scores and equations).
en with higher bone mass. They had also smoked more (P < 0.05) and consumed about 100 m g / d less calcium ( P is not significant). Women with lower bone mass in the spine were 7 kg lighter ( P < 0.01) and had greater cumulative exposure to cigarette smoking, but did not differ significantly in other risk factors. Women with low femoral neck bone mass showed trends similar to women with low bone mass in the other sites studied in body weight and cigarette smoking, and they also consumed about 70 m g / d less calcium; however, none of these trends were significant. We then constructed general linear models to properly weight each of the many independent variables in multiple variable equations to maximize the correct classification of women with low bone mass. The simple correlation coefficients for independent variables that entered the regression equations are shown in Table 2. The regression equations and the results of these models are as follows: For midshaft radius bone, mineral content = -1.1 + (0.0075 X height) + (0.0022 X weight) — (0.00044 X pack-years) + (0.12 X wrist breadth) — (0.00016 X calcium/creatinine) + (0.00004 X dietary calcium). This model correctly identifies 6 8 % of persons with low bone mass and 7 7 % of persons with high bone mass. For lumbar spine bone, mineral density = + 0.69 + (0.00099 X height) + (0.0043 X weight) — (0.00046 X pack-years) — (0.00020 X calcium/ creatinine) + (0.00002 X dietary calcium). This model correctly identifies 6 1 % of persons with low bone mass and 4 5 % of persons with high bone mass. For femoral neck bone, mineral density = + 0.79 + (0.001 X height) + (0.0041 X weight) (0.00018 X pack-years) (0.00023 X calcium/creatinine) + (0.00005 X dietary calcium). This model
Results The study population at baseline had a mean age of 51 years (range, 44 to 55 years). The average height (164 cm) and weight (65 kg) were similar to published means for random population samples of women of similar ages. They consumed 710 m g / d of calcium, about 100 mg less than the recommended daily allowance, and were very light consumers of alcohol (6 g / d ) and tobacco (13 pack-years cumulative mean lifetime exposure). Slightly less than 5 0 % were more than 6 months from their last menstrual period at entry to the study, and none were more than 3 years postmenopausal. Table 1 shows the values for the risk factors studied for women with low bone mass in the radius, spine, or hip compared with women who have greater bone mass (as defined in Methods). Women with lower bone mass at the midshaft radius were 4 cm shorter (P < 0.01), 6 kg lighter (P < 0.01), and were nearly 1 year further past menopause (P < 0.01) than wom-
98
15 January 1990 • Annals
of Internal Medicine
Figure 1: Observed radial bone mass plotted against predicted. The line of identity is shown. BMC = bone mineral content.
• Volume 112 • N u m b e r 2
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
We then investigated whether the stereotypic highrisk woman (defined as small [ < 165 c m ] , thin [ < 58 k g ] , a smoker [ > 20 pack-years of cumulative exposure], with inadequate dietary calcium [ < 800 m g / d ] ) had low bone mass. These women represented less than 7% of the population studied, but all of them had radial bone mass of less than 0.76 g/cm and, as a group, they averaged 15% less radial bone mass than the remainder of the population. At the spine, this small group averaged 3% less bone. At the femoral neck, the difference was slightly less than 10%. Discussion
Figure 2. Observed radial bone mass plotted against risk factor score (see Appendix Table). BMC = bone mineral content.
correctly identifies 65% of persons with low bone mass and 53% of persons with high bone mass. All of the models were statistically significant with R 2 values ranging from 17% (femoral neck bone mineral density) to 47% (midshaft radius bone mineral content). For example, at the midshaft radius, the general linear model predicts that a subject 1 SD taller (6 cm) than the mean height would have 0.048 g/cm (5.8%) greater than average bone mass. Similarly, a 1 SD increase in weight (12 kg) would yield a 0.035 g/cm (4.2%) increase in bone mass, and 1 SD increase in calcium to creatinine ratio would yield a 0.028 g/cm (3.4%) decrease in bone mass. Figure 1 shows the predicted bone mass values for the midshaft radius plotted against the actual values; more than 50% of the subjects were within 5% (0.42 SD) of their actual bone mass value. Figure 1 also shows that none of the women with predicted bone mass greater than 0.9 g/cm had an actual bone mass lower than 0.8 g/cm. To simplify the system for classification of subjects, scores based on the regression coefficients (see Appendix Table) were constructed for risk factors. Figures 2, 3, and 4 show the observed bone mass for the radius, spine, and femoral neck plotted against the risk scores. For the radius, 18 women had scores above 17, and none of these women had low bone mass ( < 0.78 g / c m ) . Only 1 of the 7 women with radius scores of 17 had low bone mass. In contrast, 6 of the 7 women with the lowest scores (12 or less) had low bone mass. For the spine and hip, however, the scoring system was less efficient. At the spine, 4 of the 14 women (29%) with the highest scores (12 and 13) had low bone mass that would be missed if this scoring system were used. At the femoral neck, 5 of the 24 women with the highest scores (9 and 10) had low bone mass.
We intended to determine whether clinically available information could be used to identify women at high risk for having low bone mass around the time of menopause. Although low bone mass is a strong risk factor for fractures, including hip fracture (13), this paper addresses only low bone mass, not fracture risk. Although an approach addressing fracture risk has been recommended (14), quantification of risk factors that would allow physicians to ascertain risk, either with risk scores or regression equations, has not previously been done. We have provided both risk scores and regression equations for a group of perimenopausal women. The data from our study clearly show that the analysis of multiple risk factors in perimenopausal women is an inadequate substitute for bone mass measurements. This conclusion is based on two findings. The identification of high-risk women, missing more than 30% of women with low bone mass at any site, was not sensitive enough. In addition, the sites where most important osteoporotic fractures occur (the spine and hip) were associated with greater errors in prediction. Although untested in other data sets, these predic-
Figure 3. Observed lumbar spine bone mineral density plotted against risk factor score. BMD = bone mineral density.
15 January 1990 • Annals of Interna1 Medicine
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
• Volume 112 • N u m b e r 2
99
cannot be obtained through other clinically available data. The only therapeutic approaches to the prevention of osteoporotic fractures are approaches that preserve bone mass. Clinicians can, therefore, only focus on identifying and treating those patients with low bone mass. Correctly identifying 7 0 % of women with low bone mass would be considered good in the absence of better methods, but direct measurements obviously identify nearly all women with low bone mass. Having low bone mass at any site greatly increases the likelihood both of having low bone mass at other sites (correlations range from 0.4 to 0.7) and of fracture; however, bone mass differs enough between sites to necessitate measurements at the site of interest if, for example, the prevention of hip fracture is the goal of therapy. However, whether measurements of the site of interest (for example, the hip) are better predictors of fractures at that site than are measurements at some other site remains to be shown, although this seems to be likely. Estrogen therapy prevents bone loss and reduces fracture risk, but has some risks. Knowledge of a patient's bone mass is currently the only basis for the application of estrogen therapy to prevent osteopenia. We have shown that, for most patients, risk factors fail to provide adequate information upon which to base such a decision. However, risk factors may possibly be Figure 4. Observed femoral neck bone mineral density plotted against risk factor score. BMD = bone mineral density.
Appendix Table
tion models may, nonetheless, be useful in reducing the number of women needing bone mass measurements. Although the identification of the high-risk population (women with low bone mass) was not efficient enough for clinical use ( > 3 0 % misidentified), this approach could be used without bone mass measurements in two areas. First, none of the women with predicted radial bone mass greater than 0.9 g / c m or scores greater than 17 had low bone mass when measured. Therefore, women predicted to have high bone mass (about 2 0 % of our sample) might be excluded from further measurements based on their reduced probability of having low bone mass. Because radial bone mass predicts fractures (12, 13), this approach would reduce the population needing a measurement. Second, small, light smokers with low calcium intake might be considered at high risk for having low radial bone mass without resorting to actual measurements of bone mass. The risk factor approach has been unsuccessful in distinguishing between active remodeling and inactive osteoporosis patients (22), although comparisons between patients and controls were not made. The implications of these data for clinicians are straightforward. In evaluating risk for osteoporosis in menopausal women, clinically available data are of limited utility in identifying women with the lowest bone mass. For most candidates for estrogen therapy, a single bone mass measurement at any site provides information on which to base a therapeutic decision, information that
100
Variable Range Height, cm
Weight, kg
Calcium/creatinine ratio Dietary calcium, g/d Cigarette smoking, pack-years Wrist breadth, cm
Hip
< 159 159 to 161 162 to 164 165 to 167 168 to 170 171 + < 59 59 to 63 64 to 68 69 to 73 74 +
1 2 3 4 5 6 1 2 3 4 5
1 1 1 2 2 2 1 2 3 4 5
NAf NA NA NA NA NA 1 2 3 4 5
< 0.367 0.367 to 0.477 0.478 +
3 2 1
3 2 1
3 2 1
1+
1 2
NA NA
1 2
< 2 2 to 14 15 + < 4.9 4.9 to 5.2 5.3 +
3 2 1 1 2 3
3 2 1 NA NA NA
NA NA NA NA NA NA
< 1
* Scores were assigned according to the strength of the predictors in the general linear models in predicting bone mass at each site. Thus, strong predictors (such as height at the midshaft radius) were assigned higher scores than weak predictors. For example, the stereotypical osteoporotic patient described in the text would have a minimum radius score of 6 (1 point in each category) and a maximum of 10 (with the highest scores possible in the unmeasured categories, calcium to creatinine ratio and wrist breadth). + NA = not applicable.
15 January 1990 • Annals of Internal Medicine • Volume 112 • Number 2
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
Score* Radius Spine
useful in reducing the number of women needing bone mass measurements. Grant Support: By NIH grants AG02927, P01AG05793, and RR00750, and a grant from the Ancient and Accepted Scottish Rite, Valley of Indianapolis. Requests for Reprints: Charles W. Slemenda, Dr PH, Regenstrief Health Center, Fifth Floor, 1001 West Tenth Street, Indianapolis, IN 462022859. Current Author Addresses: Drs. Slemenda and Hui: Regenstrief Institute, R G / 5 t h floor, 1001 W. 10th St., Indianapolis, IN 46202-2859. Dr. Longcope: University of Massachusetts Medical School, 55 Lake Avenue North, Worcester, MA 01655. Dr. Wellman: Indiana University School of Medicine, UH P 018, 926 W. Michigan St., Indianapolis, IN 46202-5253. Dr. Johnston: Indiana University School of Medicine, EM 421, 545 Barnhill Dr., Indianapolis, IN 46202-5124.
References 1. Baker SP, Harvey AH. Fall injuries in the elderly. Clin Geriatr Med. 1985;1:501-12. 2. Cummings SR, Kelsey JL, Nevitt MC, O'Dowd KJ. Epidemiology of osteoporosis and osteoporotic fractures. Epidemiol Rev. 1985;7:178-208. 3. Ott S. Should women get screening bone mass measurements? Ann Intern Med. 1986;104:874-6. 4. Hall FM, Davis MA, Baran DT. Bone mineral screening for osteoporosis. N Engl J Med. 1987;316:212-4. 5. Davis M R . Screening for postmenopausal osteoporosis. Am J Obstet Gynecol. 1987;156:1-5. 6. Riggs BL, Melton L. (A reply from authors to Letters to the Editor on involutional osteoporosis.) NEngl J Med. 1987;316:216-7. 7. Sites VR. Involutional osteoporosis. [Letter]. AT Engl J Med. 1987;316:215. 8. Raisz LG, Lorenzo J A, Smith J . Bone mineral screening for osteo-
porosis. [Letter]. N Engl J Med. 1987;317:315-6. 9. Tohme J F , Lindsay R. Bone mineral screening for osteoporosis. [Letter]. N Engl J Med. 1987;317:316. 10. Hall FM. Bone mineral screening for osteoporosis. [Letter]. NEngl J Med. 1987;317:316. 11. Ross P D , Wasnich RD, Heilbrun LK, Vogel J M . Definition of a spine fracture threshold based upon prospective fractive risk. Bone. 1987;8:271-8. 12. Hui SL, Slemenda CW, Johnston CC Jr. Age and bone mass as predictors of fracture in a prospective study. / Clin Invest. 1988;81:1804-9. 13. Hui SL, Slemenda CW, Johnston CC Jr. Baseline measurement of bone mass predicts fracture in white women. Ann Intern Med. 1989;111(5):355-61. 14. Riggs BL, Melton LJ. Involutional osteoporosis. N Engl J Med. 1986;314:1676-86. 15. Longcope C, Watson D, Williams KI. The effects of synthetic estrogens on the metabolic clearance and production rates of estrone and estradiol. Steroids. 1974;24:15-30. 16. Pratt JH, Longcope C. Effect of adrenocorticotropin on production rates and metabolic clearance rates of testosterone and estradiol. J Clin Endocrinol Metab. 1978;47:307-13. 17. Slemenda C, Hui SL, Longcope C, Johnston CC. Sex steroids and bone mass. / Clin Invest. 1987;80:1261-9. 18. Block G, Dresser CM, Hartman AM, Carroll MD. Nutrient sources in the American diet: quantitative data from the N H A N E S II survey. I. Vitamins and minerals. Am I Epidemiol. 1985;122:13-26. 19. Zettner A J, Seligson D. Application of atomic absorption spectrophotometry in the determination of calcium in serum. Clin Chem. 1964;10:869-90. 20. Kivirikko KL, Laitinen O, Prockop DJ. Modifications of a specific assay for hydoxproline in urine. Anal Biochem. 1967;19:249-55. 21. Bartels E, Cikes M . Uber chromogene der kreatininbestimmung nachjaffe. Clin Chim Acta. 1969;26:1-10. 22. Whyte M P , Bergfeld MA, Murphy WA, Avioli LV, Teitelbaum SL. Postmenopausal osteoporosis. A heterogeneous disorder as assessed by histomorphometric analysis of iliac crest bone from untreated patients. Am J Med. 1982;72:193-202.
15 January 1990 • Annals
Downloaded From: https://annals.org/ by a Tulane University User on 08/25/2017
of Internal Medicine
• Volume 112 • N u m b e r 2
101
