Dr Leslie has received speaker fees, research honoraria, and grants from Merck Frosst Canada, in addition to honoraria and grants from Sanofi-Aventis and Procter & Gamble Pharmaceuticals Canada. All other authors state that they have no conflicts of interest.
Site-discordance in BMD assessment is common and significantly affects patient categorization. Greater number of osteoporotic sites correlates with lower T scores at each index site. This largely explains the positive association between number of osteoporotic sites and fracture risk.
Introduction: Site-discordance in BMD is common when used to classify patients based on a cut-off T score of −2.5. It is unclear whether fracture risk assessment is improved by considering BMD information from multiple sites. Our objective was to assess the contribution of number of osteoporotic sites to overall fracture risk.
Materials and Methods: The study population was drawn from the regionally based clinical database of the Manitoba Bone Density Program that includes all clinical DXA test results for the Province of Manitoba, Canada. Analyses were limited to 16,505 women ≥50 years of age at the time of baseline DXA of the spine (L1–L4) and hip (three sites). During follow-up (3.2 ± 1.5 years), longitudinal health service records showed 765 women with at least one osteoporotic fracture code (hip, forearm, spine, or humerus).
Results: Of 5012 women classified as osteoporotic by at least one site (T score −2.5 or lower), almost one half (2370; 47%) were abnormal at only a single site. Among the 1856 women with an osteoporotic total hip measurement, mean total hip T scores decreased as the number of additional osteoporotic sites increased (−2.58, no other osteoporotic sites; −2.69, one other site; −2.87, two other sites; −3.17, three other sites; Spearman r = −0.44, p < 0.0001). Age-adjusted fracture risk from a Cox proportional hazards model increased as the number of osteoporotic sites increased (p < 0.0001), but number of osteoporotic sites was no longer an independent predictor after total hip BMD was included as a covariate (p = 0.19). Covariate adjustment for other sites of BMD measurement attenuated, but did not eliminate, the effect of number of osteoporotic sites.
Conclusions: Site-discordance is common and significantly affects patient categorization when different skeletal sites are used for diagnosis. Greater number of osteoporotic sites correlates with lower T scores at each index site. This largely explains the positive association between number of osteoporotic sites and fracture risk.
Bone densitometry plays a central role in fracture risk assessment in older individuals. BMD measurements from multiple skeletal sites have been shown to predict subsequent fractures.(1,2) Central DXA instruments can assess several anatomic sites but most commonly the lumbar spine and proximal femur are assessed in combination. BMD measurements from different skeletal sites show only moderate correlation, and discordance in site-specific diagnostic categorization is common.(3) There is uncertainty over whether diagnosis should be based on the lowest site or a standardized site.(4) Whether and how to integrate BMD information for multiple sites for fracture risk assessment is equally unclear. On theoretical grounds, it has been suggested that there would be minimal incremental use in using more than one site,(5) and one large meta-analysis found that using the minimum site for fracture risk categorization was no better than using a single site.(6)
One report suggests that the number of osteoporotic sites confers additional fracture risk information beyond whether a single site is abnormal or not,(3) and if confirmed this would provide some justification for assessment of multiple skeletal sites. We hypothesized that there may be a statistical bias in the latter, if patients with multiple abnormal measurements have more severe osteoporosis (and therefore lower bone density at all sites) compared with an individual with a single abnormal site (possibly just within the abnormal range). A cohort follow-up study was performed undertaken to explore the effect of site-discordance and impact of number of osteoporotic sites on fracture outcomes in a large clinical population.
MATERIALS AND METHODS
The study population was drawn from the Manitoba Bone Density Program and its regionally based clinical database, both of which have been previously described in detail.(7) All clinical bone densitometry in the Province of Manitoba, Canada, is performed within a single program structure that maintains uniform testing indications, requisition process, and reporting. The Program's database includes all DXA test results since the first instrument was installed in 1990. This database is >99% complete and accurate as judged by chart audit.(8) Canada has a publicly funded health care system and BMD testing, along with other essential health services, are available to all Manitoba residents without charge. BMD testing in Manitoba is considered a diagnostic test, not a screening test, and requires a physician referral. Criteria for testing are broadly consistent with most published guidelines and emphasize the importance of female sex, ≥65 years of age, premature ovarian failure, prior fragility fractures or X-ray evidence of osteopenia, prolonged corticosteroid use, and other clinical risk factors (see www.gov.mb.ca/health/programs/mbd). Access to testing is not restricted to these indications, however, and most clinical justifications are accepted.
The study population was limited to women ≥50 years of age who had baseline lumbar spine (L1–L4) and proximal femur (total hip, femoral neck, and trochanter) bone densitometry performed before October 31, 2002, with one of the Program's primary instruments (DPX or Prodigy; GE Lunar, Madison, WI, USA) and who had medical coverage with Manitoba Health during the observation period ending March 31, 2004. The complete BMD database contained 33,929 records (one record per patient visit). We excluded nonresidents (N = 204), males (N = 2833), individuals <50 years of age on the date of BMD testing (N = 4027) or where the scanner or scan date were not recorded (N = 72). We also excluded records that did not provide valid results for all four measurement sites (L1–L4, total hip, femoral neck, and trochanter), which included most scans analyzed with software versions before May 1998 because these did not provide total hip measurements (N = 6234). For patients with more than one densitometry record, only the first measurement for that individual was retained, and all follow-up scans were excluded (N = 4230). Some records had multiple reasons for exclusion. The final study population consisted of 16,505 women ≥50 years of age at the time of baseline BMD assessment with complete spine and hip BMD information. The study was approved by the Research Ethics Board for the University of Manitoba and the Health Information Privacy Committee of Manitoba Health.
Before 2000, DXA measurements were performed with a pencil-beam instrument (N = 4428 [27%]; Lunar DPX; GE Lunar, Madison, WI, USA), and after this date, a fan-beam instrument was used (N = 12,077 [73%]; Lunar Prodigy; GE Lunar). Instruments were cross-calibrated in vivo with 59 volunteers and no clinically significant differences were identified (T score differences < 0.2). Therefore, all analyses are based on the unadjusted numerical results provided by the instrument. Lumbar spine T scores (number of SDs above or below young adult mean BMD) and Z scores (number of SDs above or below age-matched mean BMD) used the manufacturer U.S. white female reference values. Hip T scores and Z scores were based on the revised NHANES III reference data (Prodigy version 8.8).(9) The number of sites within the WHO osteoporosis range (T score ≤ −2.5) was determined from the four available skeletal sites (lumbar spine L1–L4, total hip, femoral neck, and trochanter). All equipment and technologist performance is subject to a quality assurance program developed from published models and monitored by a medical physicist.(10–12) Densitometers underwent daily assessment of stability using an anthropomorphic spine phantom and each showed stable long-term performance (CV = SD/mean < 0.5%).
Manitoba Health maintains computerized databases of physician billing claims and hospital separations for all residents of the province eligible to receive health services. Each health system contact includes information on a patient's demographics, date of service, type of service and/or procedure, and diagnoses coded using the International Classification of Disease-9-Clinical Modification (ICD-9-CM). Anonymous linkage of these databases to the BMD database was possible through a unique scrambled health identification number, thereby allowing for the creation of a longitudinal record of health services and outcomes.
Each subject's longitudinal health service record was assessed from the date of bone density measurement to March 31, 2004, for the presence of noncraniofacial ICD-9-CM fracture codes using previously described definitions.(13) Fractures were classified as incident if they occurred after the BMD test. The mean observation period to identify fracture outcomes after BMD assessment was 3.2 ± 1.5 years. Specific fracture sites of interest were the hip (ICD-9-CM 820–821), spine (ICD-9-CM 805), forearm (ICD-9-CM 813), and proximal humerus (ICD-9-CM 812) because they are the basis for the 10-year absolute fracture risk estimates published by Kanis et al.(14) We excluded fractures associated with ICD-9-CM trauma codes (ICD-9-CM E800-E879 and E890-E999). In addition, we required that hip fractures and forearm fractures be accompanied by a site-specific fracture reduction, fixation, or casting code. This would exclude less severe fractures such as isolated trochanteric fractures not requiring surgical fixation and distal radius fractures not requiring immobilization. Hip, spine, forearm, and proximal humerus fractures defined in this way were collectively designated as “osteoporotic” fractures. Our use of administrative health data to define fractures in this way shows that BMD measurements predict fractures in our clinical cohort and has been reported in large meta-analyses.(15) All other fracture codes (i.e., fractures not involving the hip, spine, forearm, or proximal humerus) were studied as a separate subgroup and collectively designated as “miscellaneous” fractures (approximately one half affected the small bones of the hands/feet and one half affected more proximal sites).
Pearson correlation coefficients were computed for BMD from different measurements sites and for BMD with age. Mean T scores according to the number of osteoporotic sites were compared using ANOVA. The correlation between T score and the number of osteoporotic sites was assessed using Spearman's rank order correlation statistic. Tests for the independence of categorical variables were conducted using the χ2 test. Concordance between sites in the designation of osteoporosis (i.e., T score = −2.5 or lower) was assessed with Cohen's κ statistic, which corrects for chance agreement.(16) Because age was hypothesized to affect site discordance between the spine and hip sites, we examined the relationship between a woman's age and indices of discordance as reflected by the range in her T scores (from among the four measured sites) and the numerical differences between the T scores for the spine and individual hip sites.
Crude (unadjusted) fracture rates per 1000 person-years of observation were calculated as a function of the number of the osteoporotic sites, and the proportion of women with fractures was compared with the Cochran-Armitage test for trend. Cox proportional hazards models were constructed to assess the importance of the number of the osteoporotic sites on time to fracture for each site (i.e., spine, hip, forearm, and humerus). Age (in years) and number of osteoporotic sites (values ranged from 0 to 4) were explanatory variables in all initial models. Additional models were constructed that included BMD as a covariate. The total hip site was designated a priori as the primary covariate for these models, but secondary analyses were performed using the other skeletal sites as covariates. Subgroup analyses were conducted separately for hip and spine fractures because these have a greater impact on health-related quality of life than fractures of the forearm of forearm.(17) The hazard ratio (HR) per SD in the population was obtained by scaling the β coefficient. All analysis was performed with SPSS for Windows (version 12.0; SPSS, Chicago, IL, USA).
Subject characteristics are summarized in Table 1. Although this was a clinically selected population, mean Z scores for the spine and hip were very close to the manufacturer age-matched mean reference values (i.e., Z score close to zero).
Table Table 1.. Study Population Baseline Characteristics (N = 16,505)
There were statistically significant correlations among the BMD measurements for different sites (p < 0.0001 for all correlations). Correlations were higher between different hip sites (r = 0.82–0.95) than between lumbar spine and hip sites (r = 0.66–0.71; Table 2). Osteoporosis diagnostic concordance also showed better percent agreement between two hip sites (89.3–93.9%, Cohen's κ = 0.59–0.74) than between lumbar spine and hip sites (82.0–83.4%, Cohen's κ = 0.40–0.42).
Table Table 2.. Between-Site Correlation (Pearson's r) in BMD and Agreement (Cohen's κ) in Osteoporosis Categorization (T Score −2.5 or Lower vs. T score > −2.5)*
Among the study population, 5012 (30.4%) of the women met a densitometric diagnosis of osteoporosis for at lease one site. Site-discordance was common, and of those women with at least one osteoporotic measurement, 2370 (14.4%) were osteoporotic at a single site, 856 (5.2%) at two sites, 727 (4.4%) at three sites, and 1059 (6.4%) at all four sites. Almost one half of the women with a densitometric diagnosis for osteoporosis (2370 [47.3%]) were osteoporotic at only a single site, and this was most commonly the lumbar spine (1518 [64.1%]). Older age correlated with a greater number of osteoporotic sites (mean age: 67 ± 9 years for one, 70 ± 9 years for two, 71 ± 9 years for three, 73 ± 9 years for four osteoporotic sites; Spearman r = 0.34, p < 0.0001; Table 3)
Table Table 3.. Crude Fracture Rates per 1000 Person-Years According to Number of Osteoporotic Sites
Among women osteoporotic at only a single site, there was minimal difference in age between those osteoporotic at the spine only versus those osteoporotic at a hip site only (67 ± 8 versus 68 ± 10 years, p < 0.0001). Overall and site-specific differences between T scores were only weakly correlated with age. Older age was associated with a slight increase in the range of T scores across all four sites (mean, 1.27 ± 0.70; r = 0.05, p < 0.0001). The lumbar spine showed slightly lower T scores than the total hip (mean difference, −0.17 ± 1.09, p < 0.0001), but greater than the femoral neck (mean, 0.29 ± 1.13; p < 0.0001) and trochanter (mean, 0.05 ± 1.17, p < 0.0001). The difference between lumbar spine and hip T scores showed a variable relationship with older age: larger differences for the total hip (r = 0.05, p < 0.0001), no systematic difference effect for the femoral neck (r = −0.01, p = 0.11), and greater differences for the trochanter (r = −0.04, p < 0.0001). The net effect was less discordance between lumbar spine and total hip T scores in older women than in younger women (least squares mean difference, −0.25 at age 50 versus −0.03 at age 90). Lumbar spine and trochanter T scores also showed smaller differences with increasing age, and a cross-over point was reached at age 75 (least squares mean difference, 0.13 at age 50 versus −0.09 at age 90).
The association between an increased number of osteoporotic sites and mean T score is shown in Table 4. To assess the effect of increasing concordance in the number of osteoporotic sites, we identified all individuals who had a total hip T score of −2.5 or lower. We subcategorized these individuals as to the number of additional osteoporotic sites (range, 0–3). Mean total hip T score decreased in proportion to the number of additional osteoporotic sites: −2.58 (no other sites), −2.69 (one other site), −2.87 (two other sites), −3.17 (three other sites; Spearman r = -0.44, p < 0.0001). Similar findings were observed when this procedure was repeated for the femoral neck, trochanter, and lumbar spine using subgroups with T scores −2.5 or lower at each of these sites (Spearman r = −0.37 to −0.50, all p < 0.0001).
Table Table 4.. Mean T Scores for an Osteoporotic Index Site According to the Number of Additional Osteoporotic Sites
Crude fracture rates
During 52,616 person-years of follow-up, 765 patients experienced an incident osteoporotic fracture. Table 3 summarizes the numbers of fractures and unadjusted (crude) fracture rates per 1000 person-years of follow-up. Overall and site-specific fracture rates showed a significant positive linear trend in relation to greater number of osteoporotic fracture sites (all p < 0.0001). As noted earlier, this relationship was potentially confounded by the association between greater number of osteoporotic fracture sites and both older age and more severe reductions in BMD.
Cox proportional hazards regression analyses
Age and BMD measurement were significant univariate predictors of osteoporotic fractures. For each decade of age, the osteoporotic fracture rate increased by 1.81 (95% CI, 1.68–1.95). For each T score unit decrease in BMD, fractures increased by 1.90 (1.78–2.02) for the total hip, 2.12 (1.95–2.31) for the femoral neck, 1.82 (1.71–1.94) for the trochanter, and 1.41 (1.34–1.48) for the lumbar spine.
The age-adjusted fracture risk from Cox proportional hazards model increased in proportion to the number of osteoporotic sites (p < 0.0001). Number of osteoporotic sites was no longer significant (p = 0.19) after covariate adjustment for total hip BMD (hazard 1.84 [95% CI, 1.55–2.18] per SD decrease). Adjustment for other sites of BMD measurement attenuated but did not eliminate the effect of number of osteoporotic sites (Fig. 1). Number of osteoporotic sites was still significant after adjusting for the femoral neck BMD (p < 0.0001), trochanter BMD (p = 0.03), and lumbar spine BMD (p < 0.0001).
Subgroup analyses for hip and spine fractures are summarized in Table 5. Number of osteoporotic sites without BMD covariate adjustment showed a strong risk gradient when hip fractures and spine fracture were considered separately or together. Inclusion of L1–L4 BMD as a covariate attenuated but did not eliminate the effect of number of osteoporotic sites as a predictor of spine fractures (p = 0.012) and did not explain the effect of number of osteoporotic on hip fractures alone or the combination of hip and spine fractures (p < 0.0001). In contrast, when total hip BMD was used as the covariate, it essentially eliminated the effect of number of osteoporotic sites as a predictor of hip fractures (p > 0.2) and the combination of hip and spine fractures (p = 0.134), while attenuating (but not eliminating) its effect on spine fractures alone (p = 0.001).
Table Table 5.. Subgroup Analyses for Hip, Spine, and Miscellaneous Fractures
In a separate analysis of all other incident fractures (i.e., fractures other than hip, forearm, spine, and humerus; N = 1273), number of osteoporotic sites showed a weaker gradient of fracture risk than for the osteoporotic fracture definition (hip, forearm, spine, or humerus). However, we observed the same effect of BMD covariate adjustment on these HRs, namely that the number of osteoporotic fracture sites was statistically significant before BMD covariate adjustment (p < 0.0001) but nonsignificant after total hip BMD adjustment (p > 0.2).
This study found a high rate of discordance in the number of osteoporotic sites for women ≥50 years of age. For patients meeting the WHO criterion for osteoporosis (i.e., T score −2.5 or lower for the lumbar spine or proximal femur measurement), almost one half were classified as osteoporotic based on a single site. For those women designated as osteoporotic, greater concordance (i.e., greater number) of osteoporotic sites was associated with increased osteoporotic fracture risk. This concordance was confounded by an association with older age and with lower average T score at any measurement site. After adjustment for these covariates, the number of osteoporotic sites was attenuated as a risk factor for fracture and was no longer statistically significant after adjustment for total hip BMD.
We believe that the finding of progressively lower mean T score in association with a greater number of osteoporotic sites can be explained from basic clinical and statistical principles. A woman who is osteoporotic at all measurement sites should have (on average) lower BMD than a woman identified as having osteoporosis at a single site. This is because, in the latter case, the single discordant measurement is more likely to lie close to the arbitrary categorical threshold, because of the known correlation in BMD between different anatomic sites. Therefore, even when analyses are restricted to individuals who are osteoporotic at a specified site, an increasing number of osteoporotic sites will be associated with a lower average T score at the index site.
One third to one half of the variance in spine and hip BMD can be attributed to the correlation between these anatomic sites.(5,18) The correlation between two hip regions is even higher.(3) These observations are confirmed in our cohort. Categorical discordance (i.e., osteoporotic versus nonosteoporotic) between the lumbar spine and hip sites was observed in 16.6–18.0% of our cases, with κ values indicating moderate agreement. This is similar to the findings of Woodson(19) in 5051 female patients from a community-based osteoporosis testing center of whom 10.0% were osteoporotic at the spine but not at the total hip, and 8.9% were osteoporotic at the total hip but not at the spine, with a combined discordance rate of 18.9%. Although less frequent for two hip sites, categorical discordance was still observed in 6.1–10.7% of our cases. Wong et al.(20) reported relatively high correlations between the right and left proximal femoral BMD measurements (r = 0.90–0.95) in 78 women with low-impact fractures, with T score discordance 0.5 SD or greater in 7.5% of subjects and discordance 1 SD or greater in only 0.5% of subjects (9% and 2.5% for femur neck, respectively). When categorization was based on the minimum site, use of dual femur measurements and multiple regions increased the number of subjects categorized as being osteoporotic by ∼10%.
O'Gradaigh et al.(21) compared T score classifications for 109 men and 504 women undergoing baseline DXA scanning of the spine and hip. The lumbar spine was noted to categorize a larger proportion of the population as osteoporotic than either the total hip or femoral neck sites. The authors concluded that both sites needed to be assessed to determine an individual's fracture risk, but the study did not specifically assess fracture outcomes and was therefore limited in its ability to support this view. Lu et al.(3) examined site-discordance in the classification of osteoporosis using different reference populations applied to 7671 women from the Study of Osteoporotic Fractures (SOF) using eight BMD variables. Only 25% of subjects were consistently categorized using WHO criteria as implemented with the manufacturer's normative data. Inconsistency was greatly reduced by standardizing on a T score of −1 relative to women 65 years of age at baseline (50% consistent diagnosis) and was further improved using a risk-based approach for the spine and/or hip combined with age (consistency 68%). This study prospectively assessed fracture risk in relation to the number of osteoporotic sites for the hip (range, zero to four) and forearm (range, zero to two). The risk of osteoporotic fracture increased as a function of the number of osteoporotic regions assessed for either hip or forearm scans in all models (p < 0.001). The authors suggested that assessing multiple regions from the same scan site might help physicians better determine the prognosis of a patient.
Site-discordance has also been documented in longitudinal measures of BMD from the Danish Osteoporosis Prevention Study.(22) For 1422 women followed for 5 years without changes in their treatment regimen, correlations in the rate of change for the lumbar spine and total hip were 0.40 for untreated and 0.48 for treated women. When categorized in terms of significant change, agreement was fair (κ = 0.37, p < 0.01). Forearm measurements showed even higher rates of discordance, consistent with the view that it is a relatively nonresponsive site for assessing skeletal response to therapy or assessing bone loss.
Faulkner et al.(23) has previously shown that a T score threshold of −2.5 could not be universally applied to different measurement sites and techniques. This analysis estimated that the prevalence of osteoporosis based on a T score of −2.5 or lower could vary from 3% for heel ultrasound up to 50% for spinal QCT. Differences in age-related bone loss at different skeletal sites and different reference population measurements both contributed to this discordance. It was simultaneously noted that sites with the strongest relationship to hip fracture risk (hip and heel) actually showed the smallest age-related change. McMahon et al.(24) further noted that differences in manufacture reference data contribute to discordance for the lumbar spine. In 59 women assessed with two different DXA instruments that showed high correlation in BMD measurement (r = 0.98), there were significant differences in T scores and Z scores. The effect on diagnostic categorization was not stated. Although it is often assumed that manufacture differences in proximal femur T scores have been neutralized by standardizing reference data on the National Health and Nutrition examination Survey III (NHANES), the situation is more complicated. The differences in conversion of BMD values between DXA manufacturer platforms and incomplete adoption of NHANES reference data have been noted to contribute to significant discordance.(9,25) Leslie et al.(26) noted that two versions of software using NHANES reference data from the same equipment manufacturer generated a 2-fold difference in the apparent prevalence of osteoporosis of the femoral neck and a 3-fold difference in the apparent prevalence of osteoporosis for the trochanter.
One of the practical implications of this study relates to the use of multiple skeletal sites for fracture risk assessments. Clinical practice guidelines have often promoted the use of the minimum T score,(27,28) although other groups have favored standardization on the hip.(29) Our findings are consistent with the simulation studies of Blake et al.,(5) which noted that correlations between skeletal sites should attenuate additional fracture risk stratification from using multiple sites versus a single site. Furthermore, a recent meta-analysis showed that minimum T score did not provide improved overall osteoporotic fracture risk stratification compared with standardization on a single skeletal site.(6) Site-specific fracture analyses would be expected to show a stronger association with site-specific BMD measurements. This was clearly evident in the subgroup analysis limited to hip fractures alone, for which total hip BMD eliminated the effect of other sites, whereas lumbar spine BMD did not attenuate this relationship. In contrast, total hip BMD partially accounted for the contribution of multiple measurement sites in the prediction of vertebral fractures alone, with no significant value of number of osteoporotic sites when vertebral and hip fractures were considered together.
There are several limitations to our study. These analyses were limited to a single equipment manufacturer and software implementation; thus, the results may not be directly applicable to other combinations of equipment and reference data. The study population was limited to women ≥50 years of age and included very few nonwhite women; therefore, findings may not be applicable to men, younger women, or nonwhite ethnic populations. Fracture ascertainment from administrative health data has known limitations. Specifically, minor fractures or fractures that produce few symptoms may not lead to a physician interaction and hence are not recorded in the administrative health databases. This is particularly likely to occur with vertebral compression fractures, because the majority are not clinically diagnosed.(30) Our risk estimates are age-adjusted but are not adjusted for treatment history, prior fractures, corticosteroid use, or other clinical risk factors that have been linked to increased fracture risk.(31–34) Recent data suggest there is a larger gradient of hip fracture risk per SD in younger women and a larger gradient of risk for all fractures in older women.(31) We did not observe an age interaction (data not shown) and have assumed that gradient of risk is independent of age.
Our use of clinic-based patients rather than a regionally based cohort has advantages and disadvantages. It accurately reflects the population in which BMD testing is typically used in clinical practice. This population would be expected to have a higher fracture risk than the average population, although we could not confirm this in the mean Z scores, which were very close to the expected age-matched reference values. There may, however, be a higher prevalence of degenerative artifact in the lumbar spine in a clinical population who may be referred for testing because of back pain or abnormal X-rays, and this could degrade performance of lumbar spine bone density testing in this population.
In conclusion, site-discordance is common and markedly affects patient categorization when different diagnostic criteria are used. Greater number of osteoporotic sites correlates with lower T scores at each index site, and this largely explains the correlation between number of osteoporotic sites and fracture risk. The number of osteoporotic sites does not seem to enhance fracture risk prediction when BMD is considered as a continuous risk factor.
We thank Manitoba Health for providing data and Dr Marilyn Cree and Charles Burchill for assistance with data extraction. The results and conclusions are those of the authors, and no official endorsement by Manitoba Health is intended or should be inferred. This article has been reviewed and approved by the members of the Manitoba Bone Density Program Committee. This study was funded in part by an unrestricted educational grant from the CHAR/GE Healthcare Development Awards Programme.