Predicting Clinical Discordance of Bone Mineral Density

Original Article

Predicting Clinical Discordance of Bone Mineral Density GEORGE LARCOS,

M.B.,B.S.

• Objective: To identify clinical and lifestyle factors that may predict clinical discordance of bone mineraI density (BMD) in otherwise healthy perimenopausal and postmenopausal women referred for bone densitometry. • Material and Methods: Data from 304 white women referred for bone densitometry were retrospectively reviewed in order to determine predictors of BMD status at the lumbar spine, femoral neck, and distal radius. In addition, a cross-validation study in a further independent sample of 50 patients was undertaken. Dualenergy x-ray absorptiometry of all three sites was performed, and T-scores were determined with use of standard criteria established by the World Health Organization. Covariables including age, postmenopausal status, years since menopause, use of alcohol and cigarettes, family history of osteoporosis, exercise, height, weight, and body mass index were analyzed by canonical discriminant functions.

• Results: Seventy-six patients (25%) had normal BMD at all three sites (group A); 55 patients (18%) had osteopenia or osteoporosis at all sites (group B); and 173 patients (57%) showed regional discordance of BMD (group C). Menopausal status, years since menopause, use of alcohol and cigarettes, exercise levels, and weight allowed distinct separation of these three groups by using the plot of one canonical discriminant function against the other. When tested, this method of assignment correctly predicted the regional BMD status in 38 of 50 women (76%) in the independent sample. • Conclusion: Thus, the combination of certain clinical and lifestyle factors may be helpful in predicting variations in the clinical classification of BMD in an ambulatory healthy peri menopausal or postmenopausal woman. Mayo Clin Proc 1998;73:824-828

I BMD = bone mineral density

T

he different rates of occurrence of involutional trabecular and cortical bone IOSSI.2 account, in part, for the modest correlation in bone mineral density (BMD) between various parts of the skeleton.v' This fact has been emphasized recently by Davis and associates," who showed that 15% of elderly Japanese-American women displayed a pronounced regional variation of BMD, with intrapatient values ranging from normal to osteoporotic. Furthermore, in a recent report from the World Health Organization," the prevalence of osteoporosis was shown to be partly dependent on the number and location of sites measured. These data contribute to uncertainty about the number and location of skeletal sites that should be scanned as a screen for osteoporosis.' The hip is undoubtedly important," but various factors":" influence the pattern of bone mineral loss and potentially undermine the value of measurements obtained at a single site. For instance, 14% of women with normal BMD of the hip have either osteopenia or osteoporosis of the lumbar spine. 12 At present, women with regional discordance of BMD are difficult to identify prospectively. Accordingly, the purpose of this

report was to develop a model that could predict variations in the clinical classification of BMD status in healthy perimenopausal and postmenopausal women and thus facilitate decisions about the number of sites to be scanned.

MATERIAL AND METHODS The records of white women were retrieved with use of a computer database if the patients fulfilled the following criteria: (l) had no history of diseases or medications (including hormone replacement) known to affect BMD, (2) had no prior atraumatic fractures, (3) were ambulatory, perimenopausal or postmenopausal subjects (most within 15 years of the menopause), (4) had been referred by a medical practitioner for assessment of osteoporotic fracture risk, and (5) had technically satisfactory views of the spine, hip, and radius available for review. In particular, no confounding factors, such as degenerative changes or scoliosis, were evident in the lumbar spine; the hip was appropriately positioned, with regions of interest through the narrowest portion of the femoral neck; and the wrist showed no rotation of the ulna over the distal radius (forearm flat on the scanning bed). Subjects were considered postmenopausal if they had experienced amenorrhea lasting longer than 6 months after the age of 40 years. Patients were considered perimenopausal if they noted irregular intermenstrual intervals or vasomotor symptoms (or both).

From the Department of Nuclear Medicine and Ultrasound, Westmead Hospital, Sydney, New South Wales, Australia. Address reprint requests to Dr. George Larcos, Director, Department of Nuclear Medicine and Ultrasound, Westmead Hospital, Sydney, NSW 2145, Australia. Mayo Clin Proc 1998;73:824-828

824

© 1998 Mayo Foundation for Medical Education and Research

Mayo Clin Proc, September 1998, Vol 73

Subjects with prior atraumatic fractures and disorders or medications affecting BMD were excluded from this study because these factors influence the pattern of bone mineral loss. Furthermore, the intent of the study was to examine normal women, inasmuch as they constitute the majority of patients screened for osteoporosis in our hospital. Three hundred four patients fulfilled the entry criteria and form the basis of the retrospective phase of this report. BMD was determined with use of a dual x-ray absorptiometer (Norland XR-26, Inderlec, Australia); regions of the lumbar spine (anteroposterior L2-4), femoral neck , and distal nondominant radius were assessed. During the examination, details of patients' age, smoking history (number of cigarettes per day), intake of alcohol (grams per day), current levels of weight -bearing exercise (hours per week), and family history of osteoporosis were recorded. Patients (wearing light clothing) were also weighed (recorded in kilograms) and had their height(in meters) measured. Body mass index (kilograms per square meter) was also calculated. In our institution, the densitometer is assessed daily by a 25-minute calibration process, consisting of scanner selfdiagnostics, a scan of a quality-control phantom, and a printout of results. Specifically, accuracy and precision are examined, and results beyond predetermined levels are highlighted (more than 1.5% for accuracy and more than 2 standard deviations from the mean for precision). Data supplied by Norland (for the hip and spine) or locally generated (wrist) mean values in young adults were used for provision of T-scores. The validity of using mixed cohorts for normal ranges is supported by several reports that have shown the similarity of BMD between Australian and North American women; 13 thus, the development of institution-specific normal cohorts is unnecessary." Subjects were classified as having normal, osteopenic, or osteoporotic BMD , in ·accordance with recently reported World Health Organization criteria." In brief, a T-score of less than - 2.5 indicated osteoporosis, -1 to -2.5 denoted osteopenia, and more than -I signified normal BMD. Patients were allocated to one of three groups on the basis of the following criteria: group A if BMD was normal at all sites (that is, T-scores of more than -1); group B if all sites showed either osteopenic (T-scores of -I to -2.5) or osteoporotic (T-scores of less than - 2.5) BMD; and group C if BMD in an individual patient varied from normal to osteopenic or osteoporotic. The thre shold was selected as a T-score of -1 because this is the level at which treatment is recommended.v" The data were analyzed by using the statistical software package SPIDA (version 6). A plot of the first two canonical discriminant functions' ? for the sample of 304 patients was used to define three regions that were associated with

Discordance in Regional Bone Mineral Density

825

groups A, B, and C, respectively. In brief, canonical discriminant functions are linear combinations of the original variables, chosen in such a way that the first function (x ) reflects as much difference between group s as possible. The second function (y) captures as much as possible of the group differences not already displayed by the first. Each significant variable receives a "loading," which is then multiplied by the patient's actual data (for example, weight in kilograms or years since menopause). The sum of the variables is then used to plot a point with use of the x and y functions. A further independent sample of 50 ambulatory healthy patients (group D) was used to test this method in a crossvalidation exercise. The se subjects fulfilled the same entry criteria as previously and were assigned to group A, B, or C on the basis of their canonical discriminant values . The predicted group was compared with the actual group for each of the 50 patients. Group D subjects were compared with those in groups A, B, and C by using the Wilcoxon rank sum test; P value s less than 0.05 were con sidered significant. RESULTS The clinical and den sitometric characteristics of patients in groups A, B, and C are summarized in Table 1. Eighteen percent of subjects had T-scores of less than -1 (that is, osteopenia or osteoporosis) at all sites; 25% had normal BMD (that is, T-scores of more than -1) at all three sites. The other 57% of subjects had a combination of BMD values above and below -I; as shown in Table 1, patients in group C had a spread of data points, with a mean T-score of -1.8 for the hip, -0.4 for the wrist, and -1 for the lumbar spine. The minimal and maximal within-patient standard deviations for T-scores were 0.058 and 4.98, respectively. For the entire group, the pooled standard deviation for Tscores was 0.943. The three patient groups could be plotted distinctly with use of menopausal status, years since menopause, rate of alcohol and cigarette use, exercise, and weight as canonical variate loadings (Fig. 1). The se loadings are characterized in Table 2. Other variables including age, height, body mass index, and family history of osteoporosis did not contribute further to the model. . The model was cross-validated in an independent cohort of 50 healthy ambulatory women (group ·D; Table 3). These subjects were slightly younger than the women in groups A, B, and C (mean age, 52 years versus 55 years; P = 0.02) but were otherwi se similar, statistically. The pattern of regional BMD was correctly predicted in 38 (76%) of these patients (8 of9 concordantly normal, 6 of II concordantly osteopenic or osteoporotic, and 24 of 30 with regional discordance of BMD) .

826

Mayo Clin Proc, September 1998, Vol 73

Discordance in Regional Bone Mineral Density

Table I.-Clinical Characteristics of Study Groups A, B, and C* Group A

GroupB

GroupC

Overall

Mean (SD) or % of patients

Covariab1e Age (yr)

52 (6.2)

61 (9.0)

55 (8.8)

55 (8.8)

Postmenopausal (%) Years since menopause

60 3.0 (4.8)

96 15.1 (10.2)

81 7.1 (7.7)

78 7.4 (8.6)

70 6.1 (13.5)

89 2.0 (5.0)

78 3.9 (7.9)

78 4.2 (9.4)

90 1.8 (6.4)

91 2.8 (7.8)

92 1.8 (6.1)

91 2.0 (6.5) 30

Nondrinkers (%) Amount of alcohol (g/day) Nonsmokers (%) No. of cigarettes/day Family history of osteoporosis (%)

33

26

31

3.4 (4.3)

2.3 (2.8)

2.4 (3.0)

2.7 (3.4)

Height (m)

1.63 (0.06)

1.60(0.08)

1.61 (0.07)

1.61 (0.07)

Weight (kg)

71.6 (12.6)

62.0 (10.9)

66.1 (12.5)

66.8 (12.6)

27.1 (5.0)

24.4 (4.4)

25.7 (4.5)

25.8 (4.7)

0.40 (0.87) -0.09 (0.91) -0.01 (0.41)

-2.42 (0.83) -2.60 (0.80) -1.60 (0.60)

-1.00 (LlO) -1.80 (0.85) -0.41 (0.52)

-0.89 (1.40) -1.50 (1.20) -0.50 (0.70)

Exercise (h/wk)

Body mass index (kg/m') T-score Spine Hip Wrist

*BMD = bone mineral density; SD = standard deviation; Group A = concordantly normal BMD; Group B = concordantly osteopenic or osteoporotic; Group C = heterogeneity of BMD.

DISCUSSION The major finding in this study is that the variation in BMD between measurement sites in a healthy non-drug-taking perimenopausal or postmenopausal woman may be accurately predicted by certain clinical and lifestyle factors. In this series, postmenopausal status, years since menopause, weight, current levels of exercise, consumption of alcohol, and smoking were valuable in predicting the variability in BMD among the three measured anatomic sites in individual patients. These data may allow the type of bone densitometric examination (single or multiple sites) to be tailored on the basis of a woman's predicted BMD status and could potentially contribute to cost savings. In a previous report, the variation in bone mass between subjects could be predicted in 83% by using a complex statistical technique known as principal components analysis." With this technique, the "average" BMD of seven skeletal sites was related to age, body mass index, and ethnicity. In contrast, the statistical model in the current study is relatively simple to use, incorporates a wider variety of clinical and lifestyle factors, and predicts regional variation of BMD within an individual patient. Although a relationship exists between age and years since menopause, age per se did not contribute to the model tested in this report. As shown in Table 1, substantial overlap was evident in ages of subjects in groups A, B, and C. Moreover, age was not an independently significant

factor in the canonical variate loadings (Table 2). As an illustration, a 49-year-old woman who is 1 year postmenopausal and who neither consumes alcohol nor smokes and exercises 1 hour each week would be expected to have concordantly normal BMD (group A; y =0.023, x =0.064). If the same woman was assumed to be 9 years postmenopausal (with no other change in clinical or lifestyle factors), she would be predicted to show regional discordance of BMD (group C; y = 0.071, x = -0.136). Age may have emerged as an important variable had older subjects been included in the cohort examined. Results of this study indicate that more than 50% of healthy perimenopausal or postmenopausal women display BMD that varies from normal to osteopenic or osteoporotic, depending on the measurement site. This finding supports prior observations.v? Furthermore, results from the current study as well as those from Davis and associates' and Feyerabend and Lear'? emphasize that regional heterogeneity in BMD can occur with any pair of skeletal sites-for example, spine and hip, wrist and hip, or wrist and spine. Our data suggest that certain clinical and lifestyle factors may enable physicians to predict this heterogeneity of BMD and could be useful in determining the number of skeletal sites that should be scanned. Perhaps the accuracy of the model could be improved by incorporation of other variables, including type of physical activity" and parity and age at menarche and menopause.t''" Fat

Discordance in Regional Bone Mineral Density

Mayo Clin Proc, September 1998, Vol 73

0.4

827

c---------------------------, Group A

Group B

o

0.3

<>

o

<>

o

<> <> <>

0.2 <>

0.1

o Group C -0.1

-0.3

-0.2

-0.1

-0

0.2

0.1

First canonical discriminant function Fig. 1. Scattergram showing distribution of patient data points (N = 304) into three distinct groups-Group A (concordantly normal bonemineraldensitymeasurements), Group B (concordantly osteopenic or osteoporotic), and Group C (heterogeneity of bone mineral density measurements)-by using menopausal status, years since menopause,rate of cigarette use and alcoholconsumption, exercise,and weightas canonical variate"loadings." mass has also been shown to be of importance, but this variable necessitates use of a whole-body scan'? rather than the history or clinical examination. Although the model used in the current study predicted regional bone mineral status reasonably accurately, bone densitometry remains an important investigation because it defines the magnitude and site of bone loss and allows the efficacy of treatment to be evaluated objectively. In addition, the suitability of the model in other groups of women (premenopausal subjects, other racial groups, those with disorders or medications known to affect BMD, or elderly subjects) has not been evaluated. In general, the preferred site for bone mineral measurement should include the femoral neck because of the morbidity and mortality associated with fractures at this site." Although current densitometers can scan multiple sites in less than 20 minutes, the development of compact, mobile dual-energy x-ray absorptiometers for assessment of peripheral sites such as the calcaneus or wrist, as well as the

continued use of computed tomographic densitometry for the lumbar spine, indicates that many patients will have BMD evaluated on the basis of a single skeletal site. The model presented in the current study may assist in the management of patients who are referred to a physician with a scan of one skeletal site having previously been obtained. For example, a subject with a normal wrist BMD Table 2.-Canonical Variate "Loadings"* Variable

x-axis

y-axis

Menopausal status Years since menopause Alcohol (g/day) Smoking(cigarettes/day) Exercise (h/wk) Weight (kg) P value

0.056 0.005 -0.001 0.001 -0.005 -0.002

-0.112 0.006 0.000 0.002 0.009 0.002

<0.001

0.005

*Seetext for further information.

828

Discordance in Regional Bone Mineral Density

Mayo Clin Proc, September 1998, Vol 73

Table 3.-Clinical Characteristics of 50 Healthy Ambulatory Women (Group D) and Comparison With Study Groups A, B, and C*

Covariable

Mean (SD) or % of patients

P value versus

groups A, B, andC

ACKNOWLEDGMENT Dr. Karen Byth provided statistical advice, Louise G. Baillon managed the collection of data, and Lyn Crotty assisted with manuscript preparation.

REFERENCES

Age (yr)

52 (8.8)

0.02t

Postmenopausal (%) Years since menopause

76 5.1 (7.8)

0.9 0.06

Nondrinkers (%) Amount of alcohol (g/day)

73 4.1 (7.2)

0.9 0.9

3.

Nonsmokers (%) No. of cigarettes/day

96 1.3 (4.7)

0.9 0.9

4. 5.

Family history of osteoporosis (%)

22

0.3

2.2 (3.7)

0.2

Height (m)

1.62 (0.06)

0.3

Weight (kg)

65.3 (l0.4)

0.9

25.0 (3.5)

0.5

-0.67 (1.30) -1.51 (1.20) -0.44 (UO)

0.5 0.5 0.6

Exercise (h/wk)

Body mass index (kg/m') T-score Spine Hip Wrist

1.

2.

6. 7. 8.

*SD = standard deviation. For definitions of groups A, B, and C, see Table 1. tSignificantly different in comparison with groups A, B, and C.

9. 10.

U.

and predicted group A status (all T-scores of more than -1) would probably not benefit from measurements of other sites; conversely, a patient with predicted group C status (some T-scores above and below -1) should undergo measurement of BMD of the lumbar spine and femoral neck. Finally, although many current densitometers can acquire scans of several sites in a timely fashion and with low radiation exposure, reliance on a single skeletal site still has considerable appeal because of its inherently lower cost. CONCLUSION The data from this study demonstrate that the combination of certain clinical and lifestyle factors may predict the discordance in BMD in healthy perimenopausal or postmenopausal women. The model may be used to tailor the number of sites assessed by bone densitometry. Additional clinical factors may allow refinement of the technique.

12. 13. 14. 15. 16. 17. 18. 19. 20.

Riggs BL, Wahner HW, Melton U III, Richelson LS, Judd HL, Offord KP. Rates of bone loss in the appendicular and axial skeletons of women: evidence of substantial vertebral bone loss before rnenopause. ) Clin Invest 1986;77:1487-1491 Riggs BL, Melton U III. Involutional osteoporosis. N Engl ) Med 1986;314:1676-1686 Pocock NA, Eisman JA, Yeates MG, Sambrook PN, Eberl S, Wren BG. Limitations of forearm bone densitometry as an index of vertebral or femoral neck osteopenia. ) Bone Miner Res 1986;1:369-375 Seldin OW, Esser PO, Alderson PO. Comparison of bone density measurements from different skeletal sites. ) Nucl Med 1988;29:168-173 Davis JW, Ross PO, Wasnich RD. Evidence for both generalized and regional low bone mass among elderly women. ) Bone Miner Res 1994;9:305-309 Kanis JA (WHO Study Group). Assessment of fracture risk and its application to screening for postmenopausal osteoporosis: synopsis of a WHO report. Osteoporos Int 1994;4:368-381 Need AG, Nordin BEC. Which bone to measure? [editorial). Osteoporos Int 1990;1:3-6 Cummings SR, Black OM, Nevitt MC, Browner W, Cauley J, Ensrud K, et al (Study of Osteoporotic Fractures Research Group). Bone density at various sites for prediction of hip fractures. Lancet 1993;341:72-75 Harris S, Dallal GE, Dawson-Hughes B. Influence of body weight on rates of change in bone density of the spine, hip, and radius in postmenopausal women. Calcif Tissue Int 1992;50:19·23 Reid JR, Ames R, Evans MC, Sharpe S, Gamble G, France JT, et al. Determinants of total body and regional bone mineral density in normal postmenopausal women-a key role for fat mass. ) Clin Endocrinol Metab 1992;75:45-51 Bauer DC, Browner WS, Cauley JA, Orwoll ES, Scott JC, Black OM, et al (Study of Osteoporotic Fractures Research Group). Factors associated with appendicular bone mass in older women. Ann Intern Med 1993;118:657-665 Lai K, Rencken M, Drinkwater BL, Chesnut CH III. Site of bone density measurement may affect therapy decision. Ca/cif Tissue Int 1993;53:225-228 Pocock NA, Eisman JA, Mazess RB, Sambrook PN, Yeates MG, Freund J. Bone mineral density in Australia compared with the United States. ) Bone Miner Res 1988;3:601-604 Sheth RD, Hobbs GR, Riggs JE, Penney S. Bone mineral density in geographically diverse adolescent populations. Pediatrics 1996;98:948-951 Kanis JA, Melton U III, Christiansen C, Johnston CC, Khaltaev N. The diagnosis of osteoporosis. ) Bone Miner Res 1994;9:1137-1141 Sambrook PN. Osteoporosis. Med) Aust 1996;165:332-336 Manly BFJ. Multivariate Statistical Methods: A Primer. London: Chapman and Hall; 1986 Feingold M, Nelson DA, Parfitt AM. Composite index of skeletal mass: principal components analysis of regional bone mineral densities. ) Bone Miner Res 1992;7:89-96 . Feyerabend AJ, Lear JL. Regional variations in bone mineral density as assessed with dual-energy photon absorptiometry and dual x-ray absorptiometry. Radiology 1993;186:467-469 Davis JW, Ross PO, Wasnich RD. Relation of height and weight to the regional variations in bone mass among Japanese-Americanmen and women. Osteoporos Int 1995;5:234-238