Chapter  28:  Osteoporosis in Postmenopausal Women: Diagnosis and Monitoring: Evidence Report/Technology Assessment Number 28

Risk Factors

After a review of the relevant articles, the findings about the risk factors that predict bone density, bone loss, and fractures are as follows:

• Factors that are consistently associated with increased risks of low bone density and fractures in postmenopausal women include increasing age, white race, low weight or weight loss, nonuse of estrogen replacement, history of previous fracture, family history of fracture, history of falls, and low scores on one or more measures of physical activity or function.

• Other factors are less consistent predictors across studies, but also have statistically significant associations with low bone density and fractures. These include smoking, alcohol use, caffeine use, low calcium and vitamin D intake, and use of certain drugs.

• Predictors of low bone density are similar to those for fracture, except for those factors related to physical function and falls.

• Some clinical risk factors, especially those related to physical function and falls, are as powerful as bone density in the prediction of hip fracture.

• Women with multiple risk factors and low bone density have an especially high risk of hip fracture.

• Most of the strongest risk factors are consistently related to outcomes, regardless of the racial and ethnic population, although there are few studies of nonwhite women.

A second set of findings about risk factors concerned the accuracy of methodologies (also called instruments or tools) for assessing risk factors in identifying women at risk of fracture. These included:

• In contrast to the extensive research about determining clinical risk factors for osteoporosis and fractures, there are fewer studies available that evaluate how to use these risk factors to identify individual women at risk for fracture.

• Methologies designed to assess risk of low bone density or fractures generally have low to moderate sensitivity and specificity. Those that perform well when originally tested either have performed less well in other populations or have not yet been widely tested. Some methodologies -- especially those developed in large community populations and containing variables known to be strong predictors -- may ultimately be applicable to the clinical setting once they are tested there.

A third set of findings, exploring whether risk factors are useful in treatment decisions, included:

• The authors did not identify any studies that examined whether treatment decisions based on clinical risk factors lead to better or worse health outcomes than those based on bone measurement tests or a combination of bone tests and risk factors.

Bone Measurement Tests

Regarding bone measurement tests, the major findings related to the capacity of different bone measurement tests at different sites to predict fractures included:

• Among different bone measurement tests at various sites, bone density measured at the femoral neck by dual energy x-ray absorptiometry (DXA) is the best predictor of hip fracture and is comparable to forearm measurements for predicting fracture at other sites.

• In recent prospective studies, quantitative ultrasound (QUS) measured at the heel predicted hip fracture and all nonspine fractures as well or nearly as well as DXA measured at the femoral neck. For each of these tests, a result in the osteoporotic range is independently associated with an increased short-term probability of hip fracture. Individuals who have low scores by one of these tests, but not the other, have a higher risk of fracture than those who have higher scores by both tests, and a lower risk of fracture than those whose results on both tests are low.

• While other peripheral measures may approach QUS in predicting hip fracture, there are no recent prospective studies that directly compare prediction of hip fracture of these tests with DXA of the hip. Radiographic absorptiometry (RA) or quantitative microdensitometry (QMD) of the hand can predict the risk of nonspine fractures in general, many of which are in the forearm, but there are no recent data about the ability of hand measurement to predict hip fracture.

• Correlations between different bone measurement tests are generally too low to be accepted as evidence that one test will identify patients at similar risk to those identified by another test.

Major findings on identification of the factors related to bone testing that influence diagnosis included:

• The likelihood of being diagnosed with osteoporosis varies greatly, depending on the site and type of the bone measurement test, on the brand of densitometer, and on the relevance of the reference range to the local population.

• The likelihood of being diagnosed as having osteoporosis also depends on the number of sites tested. Testing in the forearm, hip, spine, or heel will generally identify different groups. A physician cannot say, based only on one of these tests, that the patient "does not have osteoporosis."

• The results of bone measurement tests are often inaccurately reported to patients.

Major findings related to how bone measurement test results affect patients' and physicians' decisions and actions included:

• One randomized trial suggests that women who undergo densitometry are more likely to start hormone replacement therapy than women who do not.

• In a randomized trial and a large, uncontrolled case series, women who had densitometry and were told they had osteoporosis were more likely to start or continue hormone replacement therapy than women who were told they had normal bone density.

• In one randomized trial, physicians found densitometry reports confusing and were not confident that their interpretations of T-scores were correct.

Monitoring

The major findings regarding whether bone measurement tests are effective for monitoring response to therapy and for guiding decisions about changes in management included:

• The weight of the evidence is currently against repeating bone density tests within the first year of treatment. There is insufficient evidence to determine whether repeating bone density tests 2 years after starting therapy is useful.

• There are also no studies about the effect of either monitoring responses to therapy using densitometry or the choice of test on the outcome of therapy.

Biochemical Markers

Findings regarding whether biochemical markers can be used instead of bone measurement tests in identifying women at risk for osteoporosis included:

• No single marker or cluster of markers accurately predicted the results of densitometry in individuals. Densitometry measures current bone status, whereas markers measure the process of bone turnover.

Major findings as to how well biochemical markers predict fracture included:

• No marker was associated with increased fracture risk consistently across all studies. One study provides evidence that using markers in conjunction with densitometry may increase predictability, but this result has not been otherwise confirmed.

In addition, major findings as to whether markers can help select patients for treatment included:

• Studies correlating marker results and bone loss indicated no clear trend. Furthermore, sensitivity and specificity of markers were too low to be useful for the purpose of selecting patients for treatment.

• Some studies found better test accuracy when a combination of two or more markers and/or other risk factors was used to predict bone loss.

The report also investigated what is known about the adverse effects of using markers to identify women at risk for osteoporosis. Major findings for this issue included:

• The primary adverse effect of biochemical markers is the potential for false-positive and false-negative results. Rates of false-positive and false-negative test results vary widely among the biochemical markers. (If markers are used to select women for treatment, a false-positive test could lead to the initiation of unnecessary treatment, while a false-negative result could lead to lack of needed treatment.)

Major findings regarding whether markers can predict a patient's response to therapy included:

• There is a small correlation between response to therapy as measured by densitometry and marker results, but no marker is accurate enough to reliably identify those individuals who will fail to respond to treatment.

Evaluation for Secondary Causes of Osteoporosis

Major findings from a literature review, assessment of published guidelines, and a secondary analysis regarding which diagnostic or laboratory tests are appropriate for evaluating patients with osteoporosis for secondary causes included:

• There is no evidence from controlled trials on which to base recommendations for a strategy of testing to determine secondary causes of osteoporosis.

• Some guidelines and experts support extensive testing to rule out major concomitant disease, while others suggest a limited testing strategy based on findings in the history and physical examination. Because the diagnosis of primary osteoporosis is often seen as a diagnosis of exclusion, the pattern of diagnostic testing may continue to be costly until the diagnostic yield is fully demonstrated.

• Assumptions about the probability of a secondary disease or disorder to explain the occurrence of osteoporosis vary by practice type and specialty.

• Thyroid stimulating hormone measurement, chemistry battery, and complete blood count were the most frequently ordered tests to rule out secondary causes of osteoporosis cited by respondents in our supplemental analysis. These were also the most frequently recommended tests in our review of expert guidelines.

Costs

The report's findings regarding the costs and cost-effectiveness of diagnostic strategies for identifying women with osteoporosis included:

• Published economic assessments suggested that diagnosis and treatment of women at risk for osteoporosis would be more cost-effective by targeting treatment to those with the lowest bone measurement results. Inclusion of another assessment, such as a risk profile or additional bone measurement test, prior to DXA may improve the cost-effectiveness of diagnosis.

• Using data from two large studies, the authors conducted cost-effectiveness analyses to estimate cost per hip fracture prevented. These analyses suggested that a sequential diagnostic approach may be more cost-effective than DXA alone. The sequential approach that the authors considered was QUS of the heel followed by DXA of the femoral neck only for those with low values of QUS/BUA. A range of QUS/BUA measures was used because there are no established cut points that separate high risk from normal risk. The authors used QUS as an example of a less expensive and more widely available diagnostic approach than DXA.

• Diagnosis with DXA of the femoral neck alone prevented more fractures, in most cases, but at a higher cost per hip fracture prevented compared with the sequential approach.

• All current treatment trials have inclusion criteria based on diagnosis with DXA. In sensitivity analyses, if treatment efficacy following diagnosis with QUS (using QUS/BUA cut points between 65 and 75 dB/mHz) were 5%-15% less than that following diagnosis using DXA, diagnosis with QUS alone would have higher costs to prevent fewer hip fractures than other diagnostic options.

The Evidence-based Approach

An evidence report focuses attention on the strength and limits of evidence from published studies about the effectiveness of a clinical intervention. The development of an evidence report begins with a careful formulation of the problem. In this phase, a preliminary review of the literature and input from experts, stakeholders, and patients may be used to identify the patient populations, interventions, health outcomes, and harms. These parameters are summarized in an analytic framework, which is used in turn to generate a list of key questions to examine in a systematic review of the published literature.

This evidence report emphasizes the patient's perspective in the choice of outcome measures. Studies that measure health outcomes (events or conditions that the patient can feel, such as quality of life, functional status, and fractures) are emphasized over studies of intermediate outcomes (such as changes in bone density). Such a report also emphasizes measures that are easily interpreted in a clinical context. Specifically, measures of absolute risk or the probability of disease are preferred to epidemiologic measures such as relative risk or events per 1,000 women-years.

An evidence report also emphasizes the quality of the evidence, giving more weight to studies that meet high methodological standards that reduce the likelihood of biased results. In general, the results of well-done, randomized controlled trials are regarded as better evidence than results of cohort, case-control, or cross-sectional studies. These studies, in turn, are considered better evidence than uncontrolled trials or case series. An evidence report pays particular attention to the generalizability of efficacy studies performed in controlled or academic settings. Studies that reflect actual clinical effectiveness in unselected patients and community practice settings are highlighted. Finally, an evidence report considers the net benefit, after a thorough effort to assess both the benefits and the harms of a service or technology.

In the context of developing clinical guidelines, evidence reports are useful because they define the limits of the evidence, clarifying when the assertions about the value of the intervention are based on strong evidence from clinical studies. The quality of the evidence on effectiveness is a key component, but not the only component, in making decisions about clinical policies. Additional criteria include acceptability to physicians or patients, the potential for unrecognized harms, and cost-effectiveness.

Osteoporosis Analytic Framework and Key Questions

Figure 1. Osteoporosis diagnosis and monitoring (more...)

   Figure 1. Osteoporosis diagnosis and monitoring analytic framework

Figure 1

Rationale

Practicing clinicians often initiate an evaluation for osteoporosis with an assessment of risk factors for individual patients. Arrow 1 in the analytic framework addresses the role of risk factors in identifying women at high risk for osteoporosis. This review includes a survey of the medical literature identifying risk factors predictive of bone density, bone loss, and fractures. It also evaluates evidence of the usefulness of risk factor assessment tools in identifying women at risk for fractures.

We also examined how well the various techniques for measuring bone at different anatomic sites predict fractures, and what effect the choice of test has on the diagnosis of osteoporosis (Arrow 2). We reviewed the ability of markers of bone turnover to identify women at high risk of fractures.

Women eligible for treatment based on risk factors, bone measurement tests, or marker studies would then undergo further evaluation for secondary causes, selection and initiation of medical therapies, and subsequent monitoring if indicated. Arrow 3 in the analytic framework represents several questions relating to treatment and monitoring, including: (1) the role of risk factors and bone measurement tests in guiding initial treatment decisions, (2) the appropriateness of diagnostic or laboratory tests for evaluating patients with osteoporosis for secondary causes, (3) the role of markers of bone turnover in selecting patients for treatment, (4) the role of markers of bone turnover for monitoring response to therapy, and (5) the effectiveness of bone measurement tests for monitoring response to treatment and for guiding decisions about changes in management. Arrow 4 represents how the choice of diagnostic strategy affects the costs of care.

Cohort studies (prospective studies) or randomized controlled trials

A "good" rating was achieved if the study met all the following criteria: comparable groups were assembled initially and maintained throughout the study (followup at least 80 percent); reliable and valid measurement instruments were used and applied equally to the groups; interventions were spelled out clearly; important outcomes were considered; appropriate attention was given to confounders in the analysis; and intention-to-treat analysis was used in randomized controlled trials.

A study received a "fair" rating if any of the following problems were seen: generally comparable groups were assembled initially, but some question remains whether some (although not major) differences occurred in followup; measurement instruments were acceptable (although not the best) and generally applied equally; some, but not all, important outcomes were considered; some, but not all, potential confounders were accounted for; and intention-to-treat analysis was used in randomized controlled trials.

Studies were given a "poor" rating if any of the following flaws existed: groups assembled initially were not close to being comparable or were not maintained throughout the study; unreliable or invalid measurement instruments were used or instruments were not applied equally among groups (including not masking outcome assessment); key confounders were given little or no attention; and intention-to-treat analysis was lacking in randomized controlled trials.

Diagnostic Accuracy Studies

A study received a "good" rating if the following criteria were met: one or more relevant available diagnostic tests were evaluated; a credible reference standard was used; the reference standard was interpreted independently of the diagnostic test; the reliability of the test was assessed; indeterminate results were few or handled in a reasonable manner; and a large number (more than 100) of broad-spectrum patients with and without disease was included.

A study received a "fair" rating if it evaluated one or more relevant available diagnostic tests; used a reasonable, although not the best, reference standard; interpreted the reference standard independently of the diagnostic test; and consisted of a moderate sample size (50 to 100 subjects) and a "medium" spectrum of patients.

A diagnostic accuracy study with a "poor" rating was characterized by having one or more of the following flaws: an inappropriate reference standard was used; the diagnostic test was administered improperly; a biased ascertainment of the reference standard was shown; or a very small sample size of a very narrowly selected spectrum of patients was included.

Case-Control Studies

A "good" rating included the following: an appropriate ascertainment of cases and a nonbiased selection of case and control participants were obtained; exclusion criteria were applied equally to cases and controls; the response rate was equal to or greater than 80 percent; diagnostic procedures and measurements were accurate and applied equally to cases and controls; and appropriate attention was given to confounding variables.

A "fair" rating included those studies that were recent, relevant, and without major apparent selection or diagnostic workup bias; had response rates of less than 80 percent; or provided attention to some but not all important confounding variables.

A "poor" rating was given to a study in this category if it had major selection or diagnostic workup biases, response rates were less than 50 percent, or little or no attention was provided to confounding variables.

Other considerations in study quality

For studies of the relation between risk factors and fracture or other outcomes, we also assessed completeness and length of followup; the use of co-interventions in compared groups; the training level of the personnel conducting the intervention; the inclusion and exclusion criteria used in individual studies; and the methods used to establish the outcome measure.⁵¹

Table 8. Summary of six principles for economic evaluations (more...)

Table 8. Summary of six principles for economic evaluations^a

Principle	Description
Perspective stated	Whose costs and consequences are considered?
Benefit described	What are noneconomic consequences of program?
Costs included	What are the intervention costs, morbidity or side-effect costs, averted costs, and induced costs?
Discounting included	Are future costs and consequences adjusted for timing?
Sensitivity analyzed	For values that are uncertain (e.g., assumed), are analyses performed using alternative values?
Incremental cost-effectiveness comparisons presented	Are alternatives compared in a way that allows decisions on prioritization to be made?

a

Adapted from Udvarhelyi⁵²

Quantitative Information

For each study of risk factors, we recorded the relative risk (RR) or odds ratio (OR) for fracture and, when available, the correlation coefficient for fracture or bone density. For prospective studies of fracture prediction using bone density and QUS test results, we calculated the prevalence of fractures in the sample and the probability of a fracture, given various bone density and QUS test results. We then calculated the "risk difference," which we defined as the difference between the probability of a fracture if a test revealed osteoporosis and the probability of a fracture otherwise. If a study reported results for different decision limits (cutoffs), we calculated a risk difference for each threshold.

For studies of the agreement between tests, we recorded or calculated the correlation coefficient, sensitivity, specificity, likelihood ratio (LR), and slope of the regression line for bone density and QUS tests to predict other tests, or of markers to predict the results of bone density or QUS testing. The LR for a positive test was calculated using the formula LR={[PV/(1-PV)]/[p*(d)/1-p*(d)]}, where PV is the probability of the disease, given a positive test result, and p*(d) is the apparent prevalence of disease, estimated as p(d)=(number of true positives+number of false negatives)/(number of patients screened).⁵¹

What Risk Factors Predict Bone Density, Bone Loss, and Fractures?

Many studies have been published about risk factors for low bone density, bone loss, and fractures in postmenopausal women. These studies vary by design, population, measurement of risk factors and outcomes, and methods of analysis. No single study includes all risk factors and outcomes of interest, although longitudinal studies and well-done case-control studies that use multiple regression methods provide some of the best evidence about the independent contribution of risk factors.

Figure 4. The combined influence of bone density (more...)

   Figure 4. The combined influence of bone density and clinical risk factors on the annual risk of hip fracture^a

Figure 4

These data demonstrate that factors other than bone density significantly affect the risk of fracture. Other investigations support this observation. A good-quality, large prospective study of fall-related factors and hip fracture from the Epidemiologie de l'Osteoporose (EPIDOS) cohort of European women age 75 years and older found that independent fall-related predictors of hip fracture included slower gait speed, difficulty in performing a tandem walk, reduced visual acuity, and small calf circumference.⁵⁵ After adjustment for femoral bone density, all functional measures remained significant. Another study from the same cohort reported that femoral bone density (adjusted RR [ARR] for hip fracture 1.6; CI 1.38-1.95), calcaneal BUA (ARR 1.4; CI 1.2-1.7), gait speed (ARR 1.5; CI 1.3-1.7), and age (ARR 1.5; CI 1.2-1.8) have approximately the same discriminant value to identify women at high risk of hip fractures.⁵⁶

Figure 5. 2-year incidence of hip fractures (more...)

   Figure 5. 2-year incidence of hip fractures based on calcaneal ultrasound and cognition scores^a

Figure 5

Although most studies about risk factors and fractures have focused on hip fracture outcomes, risk factors for other types of fractures also have been reported.^58-61 Other fractures, such as wrist and vertebral fractures, are more common in postmenopausal women than hip fractures and may be especially dominant in younger peri- and postmenopausal women. In a population-based prospective cohort of 3,140 Finnish women age 47-56 years (mean age 53), DXA of the lumbar spine and femoral neck and other clinical risk factors were assessed at baseline, and incident fractures were recorded over 2.7 years of followup.⁶² Fractures of the wrist were the most common type of fracture in this cohort (n=46), followed by fractures of the ankle (n=30) and rib (n=26); vertebral (n=7) and hip (n=1) fractures were less common. In this study, women with fractures had significantly lower bone density at lumbar and femoral sites compared with those without fractures. History of a previous fracture was a strong predictor of further fractures (OR 2.83 [CI 1.95-4.10]). Women with fractures also had higher weekly alcohol intake (100 g/week or more; OR 1.70 [CI 1.08-2.67]), and were more likely to be younger than 45 years when they had undergone bilateral oophorectomy (OR 3.64; CI 1.01-13.04). Women on hormone replacement therapy (HRT) were less likely to have a fracture (OR 0.94;0. CI 88-0.99). Smoking, type of menopause, age at menarche, parity, and lactation history did not differ between the groups.

A prospective study of incident vertebral fractures in 5,807 postmenopausal women 65 years or older from the SOF found several differences between women with fractures and those without.²⁵ Women experiencing their first vertebral fracture were significantly older and thinner, had a decline in grip strength compared with previous values, were less likely to take estrogen, and reported more mean height loss since age 25 years than women without fractures. A case-control study of vertebral fracture risk factors in 1,012 women in the United Kingdom found that those with fractures were significantly older, shorter, had an earlier menopause, lower parity, and a greater frequency of hyperthyroidism than those without vertebral fractures.⁶³

Figure 6. Correlates of lumbar, femoral neck, (more...)

   Figure 6. Correlates of lumbar, femoral neck, and distal radius bone density^a

Figure 6

Numerous studies about single or related risk factors for osteoporosis and fractures also have been published. These studies were conducted in a variety of populations and present additional insights about individual factors that were not considered in studies investigating multiple predictors.

Are Tools for Assessing Risk Factors Accurate in Identifying Women at Risk for Fractures?

In contrast to the extensive body of literature about determining clinical risk factors for osteoporosis and fractures, few studies evaluate how to use these risk factors to identify individual women at risk for fracture.²⁵⁹ Ideally, tools for assessing risk factors would aid clinicians in determining who should and should not undergo bone density testing, reduce modifiable risk factors, and consider treatment.

Variables in the models differed among studies. Most studies determined which variables to include by finding the most significant predictors within their study population and combining them to create a score. These scores were developed by assigning a certain number of points to risk factors based on the magnitude of their odds ratios, standardizing the beta coefficients of the regression results, or stratifying the results and designating a threshold for a positive score. Some models focused on certain types of variables, such as a model that included weight over 70 kilograms as the only variable,⁸³ and another that included four cognitive and functional variables.²⁶¹ Others included a broader range of variables such as age, weight, activity level, family history, and smoking status.²⁶⁷

Four studies evaluated hip fracture outcomes, two studies evaluated vertebral fractures, and two studies evaluated all types of fractures. These studies determined how well risk factors were associated with fractures known to have occurred already (four case-control studies), or how well they would predict fractures in the future (four prospective studies). The other nine studies evaluated bone density outcomes. These studies determined how well risk factors identified women with low bone density obtained by densitometry at the time of the study (all cross-sectional studies). Only 3 of the 17 studies validated the model in a population different from that in which it was developed.

One risk assessment tool from a good-quality study assigned points to selected risk factors for low femoral neck bone density (age, weight, race, estrogen use, presence of rheumatoid arthritis, history of fractures) to create a summary measure referred to as the Simple Calculated Osteoporosis Risk Estimation (SCORE).²⁷¹ These risk factors were obtained from over 1,200 women from the community and were subsequently tested in a validation group (n=259). SCORE had an area under the ROC curve of 0.81 in the development group, and a sensitivity of 91 percent and a specificity of 40 percent in the validation group.

The performance of SCORE also was evaluated in a separate cohort from the Rancho Bernardo Study.²⁷⁴ These women had a mean age approximately 10 years older than women in the SCORE cohorts. A total of 1,013 postmenopausal white women age 44 to 98 years underwent assessment with SCORE protocol and DXA of the femoral neck. Using the recommended cut point of 6, the sensitivity of SCORE was 98 percent, specificity 12.5 percent, positive predictive value 69 percent, and negative predictive value 75 percent. SCORE also has been used in two studies of cost-effectiveness.^12,275 These applications of SCORE are discussed in the Cost section of this report.

Figure 7. Predictive value of risk scores for (more...)

   Figure 7. Predictive value of risk scores for hip fracture^a

Figure 7

Are Risk Factors Useful in Treatment Decisions?

The use of risk factors other than low bone density in treatment decisions is advocated by experts^7,9 and included in guidelines by the National Osteoporosis Foundation.⁸ In one approach, treatment decisions are based on the overall risk of fracture, rather than on the bone measurement test alone. The assumption underlying this approach is that the number needed to treat is lower, and the cost-effectiveness of treatment is higher, for higher risk patients.

Figure 8. Testing nomogram for 5 years of alendronate (more...)

   Figure 8. Testing nomogram for 5 years of alendronate therapy^a

Figure 9. Treatment nomogram for 5 years of (more...)

   Figure 9. Treatment nomogram for 5 years of alendronate therapy^a

8

9

We found no studies in our literature search testing a risk-based treatment approach. Such studies would directly compare the benefits of treatment for women who have clinical risk factors but normal or mildly low bone density with those for women with few risk factors but lower bone density. Alternatively, we examined the entry criteria of eight trials of alendronate therapy to assess how applicable recent treatment trials are to a risk-based approach. In these trials, treatment was intended to halt or reverse the loss of bone and prevent fractures in women who had osteoporosis (secondary prevention).^276-283 We focused on alendronate trials because many clinicians consider calcium and vitamin D supplementation a standard of care for older women, and because the decision to begin HRT is a complex process involving consideration of risks and benefits unrelated to osteoporosis treatment. Subjects in these trials were predominantly white postmenopausal women, generally in their mid-60s to 70s, and were often selected because of their low bone density and, in some trials, previous vertebral fractures.

One study from the Fracture Intervention Trial (FIT) included 2,027 women who had T-scores of -1.6 or less and preexisting vertebral fractures.²⁷⁶ The 3-year risk of hip fracture in the placebo group was 2.2 percent (1.1 percent in the alendronate group, relative hazard [RH] 0.49 [CI 0.23-0.99]), and the 3-year risk of any clinical fracture was 18.2 percent (13.6 percent in the alendronate group, RH 0.72 [CI 0.58-0.90]).

A second study from FIT included 4,432 women who also had T-scores of −1.6 or less, but did not have preexisting vertebral fractures.²⁸³ The 4-year incidences of hip fracture (1.1 percent) and any clinical fractures (14.1 percent) in the placebo group were less than those of the women in the placebo group of the FIT study that included women with preexisting vertebral fractures. In this second study, only the subgroup of treated patients (n=1,627) who had a T-score under −2.5 had a significant reduction in all clinical fractures, from 19.6 to 13.1 percent (RR 0.64; CI 0.50-0.82). There was no reduction in fractures for patients who had a T-score between −2.0 and −2.5 or for those between −1.6 and −2.5.

These results suggest that women with preexisting vertebral fractures and low bone density have a higher incidence of fractures than women with low bone density alone. Women with more risk factors for fracture, such as those with preexisting vertebral fractures or the lowest bone density, had the greatest benefit with treatment. These results do not indicate that patients with a high overall fracture risk who have a T-score above −2.5 will not benefit from therapy. In the highest risk group of the second study (those who had a T-score under −2.5), the incidence of fractures was still much lower than the estimated risk for an 80-year-old woman with a T-score of −1.5 and two or more risk factors.⁸ The FIT investigators did not attempt to stratify patients by overall risk, and they have not published any subgroup analyses that compare the benefit of treatment in patients who had very high overall fracture risk but relatively high bone density for their age. None of the other studies reviewed has done this, either.

How Well do the Different Bone Measurement Tests at Different Sites Predict Fractures?

The literature describing the performance of bone measurement tests to predict fractures is extensive. We focused our review on a meta-analysis published in 1996 that provided a summary of older studies, and on prospective cohort studies that were not included in the meta-analysis or that were published since 1996. We did not review case-control studies because previous systematic reviews found that the results of case-control studies of fracture risk vary widely⁴⁹, and that at least some of the variation is related to the method used to select cases and controls.²⁵² Case-control studies do indicate whether, on average, mean values of bone density measured by a particular technique differ for patients who have and have not had a fracture, but they do not measure how well the test discriminates between high-risk and low-risk individuals.

The meta-analysis⁴⁹ that assessed 23 publications from 11 separate prospective cohort studies published before 1996 pooled studies to estimate the age-adjusted relative risk of various types of fractures for a one standard deviation decrease in bone density. Nearly all available data included in these studies were from women in their late 60s or older and all tests used densitometry techniques. Data from these studies were insufficient to estimate age-specific relative risks or to assess the relationship between length of followup and relative risk.

Results of the meta-analysis indicated that measurements made at the hip predicted hip fracture better than measurements made at other sites.^44,284,285 For bone density measurements made at the femoral neck, the pooled relative risk per one standard deviation decrease in bone density was 2.6 (CI 2.0-3.5).

This estimate was based almost entirely on one high-quality report from the SOF cohort using data from 8,134 women over 65 years of age after 1.8 years of followup.⁴⁴ In that study, the relative risk of hip fracture per one standard deviation decrease was 2.7 (CI 2.0-3.6) for measurements made at the proximal femur, versus 2.0 (CI 1.5-2.7) for measurements made at the heel, 1.6 (CI 1.2-2.1) for the distal radius, and 1.6 (CI 1.2-2.2) for the lumbar spine. Because of the clinical importance of hip fracture, these findings established DXA of the hip and, in particular, the femoral neck as the standard test for measuring bone density in epidemiologic studies.

For measures made in the forearm (proximal radius), the pooled relative risk for one standard deviation decrease in bone density was 2.1 (CI 1.6-2.7) for hip fractures.^189,286-289 The individual studies contributing to the estimate of the relative risk of forearm bone density for hip fracture included a series of 135 nursing home residents (RR 1.9; CI 1.4-2.7), a population-based study of 1,076 postmenopausal women (RR 2.5; CI 1.3-4.8), and a report of 9,704 subjects in the SOF cohort (RR 1.4; CI 1.1-1.8).

For studies using densitometry methods, calcaneal bone density was associated with forearm (RR 1.6; CI 1.4-1.8), hip (RR 2.0; CI 1.7-2.4), and vertebral fractures⁴⁹ (RR 2.4; CI 1.8-3.2). In two reports from the SOF cohort, the relative risk of heel bone density for hip fracture was 2.0 (CI 1.5-2.7) and 2.7 (CI 1.8-4.0). No other cohort study reported about prediction of hip fracture from calcaneal densitometry measurement.

For measurement of bone density at the forearm, the relative risk was 2.2 (CI 1.7-2.6) for vertebral fractures, and 1.5 (CI 1.3-1.6) for fractures at all sites. For forearm fractures, measurement at the proximal radius (RR 1.8; CI 1.5-2.1) and distal radius (RR 1.7; CI 1.4-2.0) were slightly better than measurement at the heel (RR 1.6; CI 1.4-1.8), lumbar spine (RR 1.5; CI 1.3-1.8), and hip (RR 1.4; CI 1.4-1.6). For vertebral fractures, measurement at the lumbar spine (RR 2.3; CI 1.9-2.8), heel (RR 2.4; CI 1.8-3.2), and proximal radius (RR 2.2; CI 1.7-2.6) had similar results.

Table 9. Prospective studies of densitometry

Table 9. Prospective studies of densitometry

Study	Population/setting	Exclusions from population	N	Age	Proportion of original cohort in sample	Site, machine, and measure
Study of Osteoporotic Fractures (SOF)44,285,290,297	Community-dwelling white women from four U.S. areas recruited from lists (e.g., jury selection, drivers' licenses, HMO memberships)	Black women	8,134	>65 years	97% returned at 5 years	Heel QUS: UBA 575; BUA
		Women unable to walk without assistance, or with bilateral hip replacements			72% of original cohort known to be alive	Spine, Hip DXA: Hologic QDR 1000; BMD, BMC
					Fracture followup 99% complete
Epidemiologie de l'Osteoporose (EPIDOS)292-294,298	Women from five cities in France recruited from voting lists and health insurance companies	Black women	7,575	>75 years	2% lost to followup	Heel QUS: Lunar Achilles; BUA, SOS
		Women with bilateral hip replacement, previous hip fracture, Paget's disease of bone, malignant bone disease, renal failure, hyperthyroidism, or treated hypothyroidism				Hip DXA: Lunar DPX Plus; BMD
						Whole body DXA: Lunar DPX Plus; BMD, BMC
Hawaii Osteoporosis Study (HOS)45	Women of Japanese ancestry, recruited from a 30% random sample of wives of participants in the Honolulu Heart Program	N/A	560	Mean 73.7 years	Not reported	Heel QUS: Walker-Sonix; BUA
						Hand absorptiometry: OsteoGram Phalangeal; BMD Bonalyzer Metacarpal; BMD
Diagnostisch Onderzoek Mammacarcinoom (DOM)300	Ongoing screening program for breast cancer (every woman born between 1911 and 1941 living in the vicinity of Utrecht, the Netherlands)	None	440	Mean 57 years	82% participation	Hand quantitative microdensitometry: machine not specified
	10 women for each age year from DOM were asked to participate in the osteoporosis study
Osteoporosis Risk Factor and Prevention Study (OSTPRE)295	Random stratified sample of 14,220 women in Kuopio Province, Finland	Women with spinal osteophytes or deformations, hip deformations, or incorrect measures were excluded after densitometry	3,222	Mean 53.4 years	Of 3,222 undergoing densitometry, 97.5% followup	Hip, spine DXA: Lunar DPX, BMD
Aberdeen - Porter57	Women aged >70 years in residential care within the catchment areas of Doncaster and Hull Royal Infirmaries, Aberdeen	Women not able to walk with assistance and those with bilateral hip surgery	1,414	>70 years	100%	Heel QUS: Osteosonics Ultrasonic bone analyzer 1001; BUA
				Mean 84 years
Aberdeen - Stewart299	Women from 45 to 49 years old recruited from population-based register within 20 mile radius of Aberdeen	Not specified	790	45-49 years	79% of 1,000 consenting included	Heel QUS: Walker-Sonix UBA 575; BUA
						Hip, spine DXA: Norland XR-26; BMD, BMC, or both (not reported)
Rotterdam291	People >55 years living in Ommoord (Rotterdam)	None	3,078 women	>55 years	Complete followup on 7,046 (88%); 4,268 women	Hip DXA: Lunar DPX-L; BMD
					Complete data on 5,304 (66%); 3,078 women
Modena296	Consecutive admissions over 8 months to Department of Orthopedic Surgery, Modena, Italy	Patients with diseases of bone metabolism, rheumatoid arthritis, under treatment with specified drugs, or HRT were excluded	211	Mean 52 years	84% (211/250)	Hand QUS: Sonic 1200; SOS of phalanx

BMC = bone mineral content; BMD = bone mineral density; BUA = broadband ultrasound attenuation; DPX = dual-photon X-ray; DXA = dual-energy X-ray absorptiometry; HMO = health maintenance organization; HRT = hormone replacement therapy; QDR = quantitative digital radiography; QUS = quantitative ultrasound; UBA = ultrasound bone analyzer

Table 10. Quality grading -- prospective studies of (more...)

Table 10. Quality grading -- prospective studies of bone measurement techniques and fracture risk

Cohort	Publication		Grade
EPIDOS
		Good	Fair	Poor
	Duboeuf (1997)292	X
	Schott (1998)298	X
	Garnero (1998)293		Xa
	Hans (1996)294		Xa
SOF
	Black (1992)285	X
	Cummings (1993)44	X
	Nevitt (1994)297	X
	Bauer (1997)290		Xb
	Porter (1990)57	X
	de Laet (1998)291		Xc
	Huang (1998)45		Xd
	Kroger (1995)295		Xd
	Mele (1997)296		Xd
	Stewart (1996)299			Xe
	Vecht-Hart (1997)300			Xe

a

75% of initial cohort received all tests used in the analyses, and differences between those included and not included were not reported.

b

64% of initial cohort received all tests used in the analyses.

c

48% of initial cohort had followup measures used in the analyses.

d

Self-reported fractures confirmed from medical records and radiographs.

e

Some or all of self-reported fracture outcomes were not verified.

EPIDOS = Epidemiologie de L'Osteoporose; SOF = Study of Osteoporotic Fractures

Quantitative Ultrasound of the Heel

Six publications evaluated the association of QUS measures of the heel with subsequent fractures: one from SOF,²⁹⁰ two from EPIDOS,^293,294 one from HOS,⁴⁵ and two from Aberdeen.^57,299 All six studies evaluated hip fractures, but one⁴⁵ included vertebral fractures and another²⁹⁹ reported fractures of the wrist, toes, ribs, foot, and nose, as well as of the hip. The baseline probability of hip fracture over 2 to 3 years ranged from 1 to 5.2 percent (Evidence Table 5).

Three reports directly compared the performance of QUS of the heel with that of DXA of the hip. In a study of a subset of the SOF cohort, the overall probability of a hip fracture was 0.9 percent.²⁹⁰ The relative risk of DXA of the hip for hip fracture was 2.6 (CI 1.9-3.8) versus 2.0 (CI 1.5-2.7) for QUS. QUS was still associated with an increased risk for hip fracture after adjustment for the hip DXA result. As shown in Evidence Table 5, a "high risk" ultrasound result increased the probability of a hip fracture from 0.9 to 1.8 percent, while a "low risk" ultrasound result decreased the probability of a fracture from 0.9 to 0.6 percent. As mentioned earlier, for DXA of the hip the comparable probabilities are 2.3 percent for a "high risk" test result and 0.5 percent for a "low risk" result.

Table 11. Probability of fracture in subgroups of women defined (more...)

Table 11. Probability of fracture in subgroups of women defined by QUS and DXA of the hip results^a

QUS cut point	Ultrasound result b	DXA of the hip result c	Probability of fracture
85	high	high	.0220 (35/1594)
	high	low	.0053 (19/3613)
	low	low	0 (0/694)
	low	high	0 (0/92)
80	high	high	.0227 (35/1541)
	high	low	.0055 (18/3271)
	low	low	.0010 (1/1036)
	low	high	0 (0/145)
75	high	high	.0233 (34/1458)
	high	low	.0050 (14/2814)
	low	low	.0013 (5/1493)
	low	high	.0044 (1/228)

a

Bauer DC⁴³⁵

b

A QUS result lower than the cut point in the first column is classified as "high risk," and a QUS result higher than the cut point is classified as "low risk."

c

Dual-energy X-ray absorptiometry of the femoral neck using a T-score = -2.5 as the cutoff value for "high" versus "low" risk.

DXA = Dual-energy X-ray absorptiometry; QUS = Quantitative ultrasound

Table 11 shows the probability of fracture for subgroups of women in the SOF defined by having "low risk" or "high risk" QUS and hip DXA results. Using a cutoff value of 85 (dB/MHz), 786 women had a "low risk" result, and none of them had a fracture. Using a cutoff value of 80 (dB/MHz), 4,182 women had a "high risk" QUS, and 1,081 had a "low risk" QUS. Those who had a "low risk" QUS result had a low risk of fracture, regardless of their hip DXA result (.001 if the hip DXA was "high risk" and 0 if it was "low risk").

These subsets of the SOF excluded women who had a fracture before QUS measurements were done. This bias is potentially important because women who had fractures early in the study may have differed in risk factors and in the predictability of their fractures from women who fractured later.

In the second study, from the EPIDOS cohort,²⁹⁴ the relative risk for hip fracture per standard deviation unit was 2.0 (CI 1.6-2.4) for QUS versus 1.9 (CI 1.6-2.4) for DXA of the hip. In the third study,²⁹⁹ the relative risks for DXA of the femoral neck and QUS for all nonspine fractures were nearly equal (1.4; CI 1.3-2.4 versus 1.43; CI 1.2-2.4). Absolute risk differences for QUS ranged from 0.007 to 0.017, compared with 0.007 to 0.025 for DXA of the hip. The number of women with low bone density who subsequently had any nonspine fracture was similar for QUS (between 2 and 8 in 100) and hip DXA (between 2 and 7 in 100), as was the number of women with normal bone density who had a fracture. Three studies^45,296,299 included vertebral fractures in their evaluation of the association of QUS and fracture risk. The odds ratio for a vertebral fracture, adjusted for age and calculated for one standard deviation decrease in calcaneal BUA, was 1.50 (CI 1.05-2.16) in one study⁴⁵ in which 39 patients of a sample of 560 sustained a vertebral fracture. For the other two studies, no vertebral fractures occurred in the samples.

What Factors Related to Bone Testing Influence Diagnosis?

How do Bone Measurement Test Results Affect Patients' and Physicians' Decisions and Actions?

Many clinicians believe that the results of bone measurement tests motivate patients to exercise, adhere to medication regimens, and change other behaviors to reduce their risk of fracture. Of 1,335 women from the SOF cohort who completed a questionnaire about estrogen therapy and were taking estrogen, 33.6 percent reported their primary reason was prevention or treatment of osteoporosis.³⁵² Furthermore, test results appear to influence physicians' decisions to prescribe HRT.³⁵³ On the other hand, the psychological effect of receiving a result indicating osteoporosis could produce anxiety and perceived vulnerability³⁵⁴ that may be unwarranted. We reviewed four publications that address the influence of test results on women's decisions about treatment and health-related behaviors.^355-358 One publication evaluated the effect of bone measurements on practice patterns of physicians and also evaluated their level of understanding the reports they receive.³⁵⁹

Are Bone Measurement Tests Effective for Monitoring Response to Therapy and for Guiding Decisions about Changes in Management?

To be effective in monitoring, a test needs to be sufficiently accurate to detect differences in bone density over time. These differences must predict fracture risk, and changes in therapy made because of these differences must reduce the risk of fracture. To determine if bone measurement tests are effective for monitoring, we reviewed studies related to:

• The reliability of densitometry in detecting changes in bone density in patients on therapy for osteoporosis.

• How well changes in densitometry predict fracture outcomes in patients on therapy.

• How well an "adequate" response to therapy can be distinguished from an "inadequate" response.

• If adjustment of therapy based on changes in bone density leads to reduction in fracture risk.

Table 12. Interval between measurements required to (more...)

Table 12. Interval between measurements required to reliably detect bone loss^a

Precision error	Estimated bone loss	Difference in measurements b	Time frame for a reliable bone mass measurement followup
CV%	%	%	Years
1	1	2.77	2.77
1	3	2.77	0.92
2	1	5.54	5.54
2	3	5.54	1.85
3	1	8.32	8.32
3	3	8.32	2.77
4	1	11.08	11.08
4	3	11.08	3.70
5	1	13.30	13.30
5	3	13.30	4.43
6	1	16.63	16.63
6	3	16.63	5.54

a

Agency for Health Care Policy and Research⁴³⁶

b

Two scans (measurements) would have to differ by more than this amount to be confident that a real change had occurred with 95% confidence that the detected losses are real.

CV = Coefficient of variation

Bone density tests are not precise enough to reliably measure short-term bone loss in postmenopausal women who are not on treatment for osteoporosis. In population-based studies, postmenopausal women lose bone (femoral neck) at a rate of 0.45 to 0.6 percent per year.^302,360-362 In one of these studies, which stratified by age, the rate was 0.6 percent per year in women age 60 to 69, 1.1 percent per year in women age 70 to 79, and 2.1 percent per year in women over 80 years of age.³⁶⁰ Higher rates of bone loss have been reported in case series and in the control groups of therapeutic trials performed in patients recruited from specialty referral clinics.^363,364

In randomized trials of treatment for osteoporosis, the relationship between changes in bone density and fracture rates is inconsistent.³⁶⁵ For some therapies, however, such as fluoride, increases in bone density have been associated with unchanged or even increased fragility. Other treatments, such as osteocalcin, reduced fracture rates without an appreciable effect on bone density. Because of these results, there is a consensus that while reduction or reversal of bone loss is an important intermediate outcome, it is not an adequate substitute for direct evidence that a treatment prevents fractures.⁸

One report from the FIT showed that, on average, women who had larger increases (3 percent or more) in total hip bone density after taking alendronate for 2 years had a lower risk of vertebral fracture than women who had unchanged or lost bone density.³⁶⁶ The differences between patients who had a small (3 percent or less) increase and those who lost bone were not statistically significant, and the results did not change when the analysis was limited to patients with high adherence, which was measured by pill counts. These findings do not support the view that monitoring bone density changes can help detect poor compliance.

As the authors pointed out, the findings "do not necessarily suggest that measurement of change in bone density in individual patients can predict their specific risk for vertebral fractures."³⁶⁶ Most importantly, the analysis did not address whether the women who lost bone or had small increases in bone density benefited less from therapy than women who had a greater response. These women may have been at higher risk of fracture in the first place, and might have had even worse results had they been taking placebo. At present there is no way to know whether increasing the dosage of the medication might improve outcomes.

High-quality, indirect evidence about the potential impact of adjusting therapy based on the results of bone density tests comes from a very recent report of the FIT and Multiple Outcomes of Raloxifene Evaluation Trial.³⁶⁷ Patients in the treatment arms of the two trials (n=6,588) had DXA of the hip and spine at baseline and at 12 and 24 months. During the first year of alendronate therapy, a small group of patients (1.4 percent) appeared to have a decline of 4 percent or more in bone density (these were the worst responders). These patients had a 92 percent chance of gaining bone density in the second year of treatment, and their average gain (4.8 percent) was higher than that of patients who had seemed to respond to alendronate in the first year. Conversely, patients who gained the most bone density in the first year (8 percent or more) lost bone density on average in the second year of treatment. A similar pattern was observed for raloxifene. The investigators attributed this pattern to the statistical phenomena known as "regression to the mean."³⁶⁷

Can Markers Be Used Instead of Bone Measurement Tests to Identify Women Who Have Low Bone Density?

Two good-quality studies^380,382 measured urinary NTX and found that between 51 and 61 percent of women with osteoporosis had higher levels, and that 50 to 82 percent of those without osteoporosis had lower levels. Overall, only 22% of women with high levels of NTX had osteoporosis. High levels of BALP correctly identified 70% of patients with osteoporosis, but 42% of women without osteoporosis also had high levels, and overall only 21% of women with high levels had osteoporosis. The predictive values of osteocalcin and PICP levels (15%-23%) were similar to those for NTX and BALP, while those for PYR and D-PYR levels were worse (12%-13%).³⁸⁰

The other three studies were small (n= 34 to 78) and did not reflect the mix of patients seen in general clinical practice.^378,379,381 Across all studies, for all markers measured, to correctly identify 80 percent of women with low bone density, fewer than 30 percent of those withnormal bone density would be classified correctly. In one small study in which women with osteoporosis were identified before enrollment, CTX was more accurate. Seventy-five percent of women with osteoporosis had an elevated CTX level, and 85 percent of normal controls had lower CTX levels.³⁷⁹ At this cutoff, 25 percent of women with osteoporosis would be incorrectly classified as having normal bone density. Lowering the cutoff for a positive test to result in no false-negative results would lead to 73 percent of women with normal bone density to be classified as osteoporotic.

How Well do Markers Predict Fractures?

In a large, good-quality, population-based study of 305 Swedish women aged 40 to 80, those with decreased levels (1 standard deviation decrease) of ICTP were 1.9 times more likely to have a fracture of any type 5 years later, and those with decreased levels (1 standard deviation decrease) of PICP were 1.8 times more likely to fracture.³⁸⁴ This study sample was randomly selected from population files of one city, making it similar to a general clinical population. However, there was a 63 percent participation rate, suggesting that the sample may not be representative.

In the EPIDOS cohort,³⁸⁶ women with levels of CTX 2 standard deviations above the premenopausal mean were 2.2 times more likely to have a hip fracture within 22 months, and those with free D-PYR above this range were 1.9 times more likely to fracture. Other markers measured (BALP, osteocalcin, NTX) were not significant predictors of fracture.

In the Rotterdam study, increased levels of free, but not total, PYR and D-PYR
were associated with increased risk of hip fracture.³⁸⁷ Women with higher levels of free D-PYR were 1.4 times more likely to fracture at a mean follow-up of 22 months. BALP and NTX were not predictive of fracture in this and another study.³⁸⁶ An increased level of ALP was associated with risk of fracture in one of two studies.³⁸⁵

In the Rotterdam study, women with osteocalcin levels above the median were 3.1 times more likely to have a hip fracture at a median follow-up of 2.4 years, but this difference was not statistically significant (95 percent CI 1.0-9.2). In both reports from EPIDOS, there was no association between elevated total osteocalcin levels and the subsequent risk of fracture.^384,386 However, increased levels of undercarboxylated osteocalcin, were associated with an increased risk of hip fracture (OR = 2.0, 95 percent CI = 1.2-3.2).³⁷⁶ There was also a strong association between increased levels of undercarboxylated osteocalcin and fracture in a group of elderly women living in a nursing facility who had vitamin D deficiency.³⁸³ Those with increased levels of undercarboxylated osteocalcin at baseline were 5.9 times more likely to sustain a hip or other non-vertebral fracture within 18 months. While the results of these two analyses need to be confirmed in other prospective studies, they have generated intense interest. Investigators are examining whether these results reflect decreased intake or deficiency of vitamin K, which mediates carboxylation of osteocalcin.

Results of well-done studies in which markers were combined with other risk factors to enhance prediction of fracture are inconsistent. One good-quality study found a greater risk of fracture in women with both a low bone density and either high CTX or high free D-PYR levels than in women with either risk factor alone.³⁸⁶ Other studies did not find the same increased predictive value when bone density and markers were combined. In the Rotterdam study, adjustment for bone density did not affect risk estimates for the markers, but adjustment for disability (ability to rise, walk, bend) did.³⁸⁷ According to a Hawaiian study, adjustment for baseline calcaneous bone density did not affect the relationship between marker levels and risk of vertebral fracture at three year follow-up.³⁸⁵ In the Swedish study, combining markers with bone density did not increase ability to predict fracture.³⁸⁴

In a nested case-control analysis of 109 elderly women (more than 74 years old) with hip fractures and 292 without fractures in the EPIDOS study, CTX and D-PYR had similar accuracy to predict hip fracture at a mean follow up period of 22 months.³⁸⁶ Garnero defined a "positive" test for CTX or D-PYR as one that was two standard deviations or more above the mean for premenopausal women. Of the women who had a fracture within 22 months of follow up, 36% had a CTX above this value, as did 19% of those who did not have a fracture. For D-PYR, 36% of the women who had a fracture had a "positive" test result, as did 19% of those who did not have a fracture. In the study of institutionalized women with vitamin D deficiency,³⁸³ 53 percent of those who fractured had an elevated undercarboxylated osteocalcin level at baseline, and 79 percent of those without an elevated level did not fracture. Almost half of the women who fractured were not identified as at-risk by their marker result. Follow-up time was short in both of these studies (22 months in EPIDOS and 18 months in the institutionalized women with vitamin D deficiency); accuracy of the tests might have been different with a longer period of observation.

Can Markers Help Select Patients for Treatment?

If markers can accurately predict which patients are at risk for rapid bone loss, they could help clinicians select patients who would benefit most from treatment, potentially leading to initiation of treatment in time to prevent further bone loss and eventual fracture. However, we identified no studies in which treatment decisions were made on the basis of marker test results.

In this multicenter study of 295 women aged 67 to 89 participating in the Study of Osteoporotic Fractures,³⁹⁷ osteocalcin, PYR, NTX, and CTX levels measured at baseline correlated significantly with annual percent change in bone density by hip DXA measured an average of 3.8 years later. BALP, however, was not significantly correlated with DXA.

In the OFELY study, a large, population-based study of determinants of bone loss in postmenopausal women in France,³⁹⁹ osteocalcin, NTX, and CTX were significantly associated with forearm bone loss. This study had the longest follow-up period, 4 years.

There was no clear trend across all the studies regarding the correlation between marker results and bone loss. It is difficult to compare these studies because of differences in populations, markers measured, and site of bone density measures. For example, one of six studies of ALP,³⁹² one of three studies of BALP,³⁹³ and one of five studies of urinary calcium found a significant correlation with bone loss.³⁹⁵ Two of six studies found a significant correlation^393,397 between hip DXA and OC. Six studies measured HPR, and two found a significant correlation.^389,393

When a regression equation with osteocalcin, ALP, CA, and HPR was used to predict bone loss at different sites, these markers were correlated with bone density at the proximal forearm and ultradistal forearm, but not at the spine or total body.³⁹⁸

In the remaining studies, markers had high sensitivity and low specificity for predicting which women would lose bone rapidly,³⁹⁵ or had poor sensitivity.³⁹² These studies were small and were conducted at single institutions. One study³⁹⁴ measured the accuracy of six markers to predict bone loss by spine and hip densitometry 4 years later in 141 volunteers participating in an ovarian screening program in the United Kingdom. Sensitivity of osteocalcin, ALP, CA, HPR, PYR, and D-PYR ranged from 33 to 58 percent, and specificity ranged from 50 to 72 percent.

Three longitudinal studies provided information that made it possible to calculate test characteristics when a combination of two or more markers and/or other risk factors was used to predict bone loss.^388,390,393 Predictive value was improved when results of several markers were combined, or when markers were combined with risk factors such as initial bone density,³⁹⁰ body mass index,³⁹³ fat mass,³⁸⁸ or calcium intake.³⁹³

A cross-sectional study of 320 teachers in one Italian town⁴⁰⁰ examined the relationship of markers to other risk factors for osteoporosis. Only age and time since menopause were significantly associated with PYR; age, number of cigarettes smoked per day, and body mass index were associated with serum osteocalcin. There was no relationship between these markers and bone density measured at the distal forearm. In another study of 249 premenopausal volunteers,⁴⁰¹ osteocalcin and PYR were not associated with height, weight, calcium intake, vitamin D supplementation, caffeine intake, or bone density measured at the lumbar spine, distal forearm, and proximal femur.

Can Markers Predict Response to Therapy?

There is no direct evidence about whether the use of markers to monitor treatment in women with osteoporosis leads to a reduction in fracture risk. The best evidence for this would be a trial in which women on treatment are randomized to monitoring with markers or usual care, and then prospectively followed for incidence of fracture. We found no studies that used this design. Also, no studies measured outcomes, either fracture or bone density, after changes were made in treatment based on the results of marker tests.

Some researchers have advocated using markers as a way of monitoring compliance with treatment. We identified no studies of the use of markers to increase compliance. It is not clear whether this is a more effective method than directly asking the patient about compliance, side effects, or difficulty with following the regimen.

Evidence Tables 15 and 16 describe studies in which markers were used to monitor or predict response to therapy in women already on treatment for osteoporosis.

What Diagnostic or Laboratory Tests are Appropriate for Evaluating Patients with Osteoporosis for Secondary Causes?

Osteoporosis can be a sign of systemic disease such as hyperparathyroidism, multiple myeloma, vitamin D deficiency, hyperthyroidism, liver disease, renal failure, and rare disorders such as Menkes syndrome, Marfan syndrome, Ehlers-Danlos syndrome, and homocystinuria due to cystathione deficiency, among others. While some factors, such as corticosteroid and excessive alcohol use, can be ascertained in a routine history and physical examination, diagnosis of many of the other secondary causes of osteoporosis requires laboratory testing.

We addressed the question of what is the appropriate evaluation for secondary causes for women with osteoporosis by conducting a search of the medical literature and a review of expert guidelines and recommendations. Because of the lack of evidence for this question, we also performed a supplemental analysis of physician practice patterns.

Our literature search found no data about the prevalence of secondary causes of osteoporosis in women found to have low bone density or an atraumatic fracture in population-based studies or in the primary care setting. There also is no evidence available to judge when a search for secondary causes is needed, which secondary causes should be sought, or the probability of a secondary cause in women diagnosed with osteoporosis.

Some literature, however, does report either bone density, measures of bone turnover, or the occurrence of osteoporosis in individuals with diseases or disorders known to be secondary causes of osteoporosis, such as hyperthyroidism, hyperparathyroidism, or Cushing's syndrome. These reports were not reviewed because they did not add information about the occurrence of the disease or disorder in women with osteoporosis, and because of methodologic limitations of these studies such as referral bias and small sample sizes. Several small case series report the occurrence of secondary osteoporosis in men referred with severe osteoporosis or vertebral fracture.^418-420 The majority of men in these series (54 to 78 percent) were found to have secondary causes of osteoporosis, although it is unclear whether these diagnoses were preestablished or whether osteoporosis was diagnosed first and the secondary causes detected by a subsequent laboratory evaluation. The distribution of secondary causes attributable in women may be markedly different, however. Regardless, because these series include men from tertiary referral centers, they are unlikely to represent the larger community population with osteoporosis. Only a prospective, systematically evaluated cohort of women with newly diagnosed osteoporosis would provide data to appropriately answer this question.

Guidelines and recommendations about laboratory testing to identify secondary causes of osteoporosis were obtained from expert sources published within the last 10 years (Appendix D). We considered three sources: published practice guidelines, review articles on diagnosis and management of osteoporosis, and chapters published in textbooks likely to be consulted by primary care practitioners or endocrinologists. We attempted to delineate the practice base and audience for which the recommendations were written, the level of evidence review undertaken in preparation of the review, the secondary disorders that should be considered, and the specific laboratory tests to be used routinely or as indicated by history or examination.

Of the 10 practice guidelines we reviewed, only 3 appeared to be based on a formal evidence review; the rest were based on expert opinion. Only one of the four review articles and none of the textbook chapters were based on an evidence review. Despite this, even within an evidence-based review, many of the recommendations for laboratory testing were stated as being empirically-based rather than evidence based. This is consistent with the lack of direct data on the prevalence of secondary causes of osteoporosis or on the outcomes of diagnostic algorithms. Slightly greater weight could be given to practice guidelines, however, because these frequently represent the considered opinion of a group of experts and undergo peer review. Expert opinion is usually written by a single subspecialist whose opinion may be based on the management of complicated osteoporosis, and may not be representative of the larger population.

Without exception, all practice guidelines suggest a laboratory evaluation, at least in some patients with established osteoporosis, to exclude secondary causes. The scope of this workup, however, varies among guidelines. Because osteoporosis is often a "diagnosis of exclusion," most guidelines and opinions affirm that the diagnosis of primary osteoporosis cannot be made until coexistent disease or disorders are ruled out. In the first two practice guidelines in Appendix D from osteoporosis foundations, the statements of philosophy strongly contrast. The National Osteoporosis Foundation cautions practitioners that "limited biochemical testing may be required in some cases," whereas the Foundation for Osteoporosis Research and Education states that "the workup of secondary causes can be extensive."

The majority of the 10 practice guidelines suggest that the routine evaluation of a woman with osteoporosis would include measurement of serum calcium, phosphorus, ALP, liver enzymes, and creatinine, all part of a metabolic panel, as well as a complete blood count. Most guidelines suggest additional tests, depending on indications of individual patients. These might include serum protein electrophoresis, thyroid function tests, 24-hour urinary calcium, parathyroid hormone, and 25-OH vitamin D levels.

The majority of expert opinions cited in textbooks describe a more extensive routine evaluation: serum calcium, phosphorus, ALP, and creatinine, or, alternatively, a metabolic panel plus thyroid function tests, complete blood count, and serum or urine protein electrophoresis. If indicated by history and physical examination, additional testing would include measurement of parathyroid hormone and 25-OH vitamin D levels. At least half of the review articles would include a metabolic panel and complete blood count, as well as thyroid function tests. Additional testing, if indicated, would include protein electrophoresis, follicle-stimulating hormone, and a bone biopsy.

Review of these guidelines indicates a lack of agreement in the medical community about the appropriate evaluation for secondary causes. Without evidence of the prevalence of secondary disorders in a population tested by bone density measurement, subsequently diagnosed with osteoporosis, and then evaluated for secondary disorders, it is unlikely that this can be resolved. As a first step, identification of the prevalences of specific secondary causes of osteoporosis in a population of women identified with low bone density would be invaluable. An evaluation could then be constructed based on the probability of disease.

To obtain more information about how physicians evaluate women with osteoporosis for secondary causes, we performed a supplemental analysis about physician practice patterns based on responses to a questionnaire. Details of this analysis are presented in a subsequent section (see Supplemental Analysis 1: Practice Variation in the Evaluation of Secondary Causes). The results of this pilot study indicated that physician practice patterns vary in the community. Physician test-ordering behavior differed among primary care practitioners and specialists. Family practitioners less frequently recommended bone density tests for postmenopausal women, obtained bone density tests, and undertook evaluations for secondary causes compared with internists, endocrinologists, and obstetricians/gynecologists. Endocrinologists more frequently evaluated patients for secondary causes than primary care practitioners. TSH, chemistry battery, and complete blood count were the most frequently ordered tests cited by respondents. These were also the most frequently recommended tests in our review of expert guidelines discussed above.

What are the Costs and Cost-Effectiveness of Diagnostic Strategies for Identifying Women with Osteoporosis?

We took two approaches to address the question about the cost-effectiveness of diagnostic strategies. First, we conducted a literature review and critically appraised relevant articles. Finding this literature limited, we performed a supplemental analysis by constructing our own models of cost-effectiveness. We compared various combinations of QUS measurements of the heel and DXA measurements of the hip for hip fracture outcomes using data from two large prospective cohort studies of osteoporosis. For a full description of this analysis, see Supplemental Analysis 2: Model-Based Economic Comparison of Diagnostic Alternatives.

Another study compared the costs of diagnosing osteoporosis using SCORE with different combinations of DXA of the hip, spine, and forearm radius in 392 women who were retirees or employees of a corporation in northern California.¹² One hundred of these women (26 percent) had a T-score under -2.5 with DXA at one or more of the three sites (used as the gold standard in this study). The cost of performing all three tests on every patient was $235. The authors suggested a sequential strategy in which all women first underwent an evaluation using SCORE. Women determined to be at higher risk then received densitometry (e.g., DXA at the radius followed by DXA at the hip and spine for those not assessed as osteoporotic at the radius). Assuming SCORE costs $5 per patient to administer, the example sequential strategy would identify 98 of the 100 osteoporotic women, at a cost of $170 per case. With this strategy, all but 84 of the 392 women would require densitometry, and the majority of women with osteoporosis would undergo DXA at three sites (hip, spine, and radius).

Two other studies reviewed data on women age 60 to 69 years⁴²¹ and women age 50 to 54 years⁴²² to compare using QUS and/or a series of clinical questions to determine who to refer for assessment with DXA. These authors used DXA of the femoral neck and/or the lumbar spine as the gold standard for comparison. Their referral criteria (relevant to these analyses) included (1) women suspected to be osteoporotic from radiological and clinical findings, (2) women with a medical condition predisposing them to osteoporosis, (3) women on corticosteroids, (4) women who underwent menopause before age 45, and (5) women with a family history of osteoporosis.

In the older cohort (age 60 to 69 years), the sensitivity and specificity (73 percent and 81 percent, respectively) were better and the cost per osteoporotic case identified was lower for QUS compared with the referral criteria and with DXA for all women. For women age 50 to 54, when disease prevalence is considered, QUS also had the lowest cost per osteoporotic case identified. The authors used sensitivity analyses to determine that if DXA costs were less than 3.5 times those of QUS, then DXA for all women was more cost-effective. However, when both osteoporotic and osteopenic cases were targeted, the authors found the referral criteria to have lower cost per case identified than QUS, and sensitivity analyses found this pattern to be consistent across ratios of DXA to QUS costs. An additional clinical criterion -- estrogen-deficient women who would undergo or continue with treatment if found to be osteoporotic or osteopenic -- also was included with the other clinical criteria in the younger cohort. However, the cost per case detected was higher for this group than for the five-question clinical criteria group.

The other studies compared universal HRT with selective therapy based on the results of bone density tests.^423,424 These analyses did not provide information about the choice of diagnostic test.

What Risk Factors Predict Bone Density, Bone Loss, and Fractures?

Several factors are consistently associated with increased risks of low bone density and fractures in postmenopausal women, including increasing age, white race, low weight or weight loss, nonuse of estrogen replacement, history of previous fracture, family history of fracture, history of falls, and low scores on one or more measures of physical activity or function. Other factors are less consistent predictors across studies, but also have significant associations with bone density and fractures. These include smoking, alcohol use, caffeine use, low calcium and vitamin D intake, and use of certain drugs. Additional risk factors are important in specific studies. The relative risks of several risk factors are comparable to that of a 1 standard deviation difference in bone density. Predictors for low bone density are similar to those for fracture except for those specifically related to falls. Most of the strongest risk factors are consistently related to outcomes in different racial and ethnic populations. Risk factors are generally similar for the various fracture sites except that fractures related to falls have additional functional risk factors.

Are Tools for Assessing Risk Factors Accurate in Identifying Women at Risk for Fractures?

Few studies report results of the performance of tools for assessing clinical risk factors for osteoporosis. Our evidence review suggests that specific tools to assess risk have inadequate sensitivity and specificity and have not yet been widely tested. Generalizability of these tools to the clinical setting is therefore currently limited. However, some tools -- especially those developed in large community populations and containing variables known to be strong predictors -- may ultimately be applicable to the clinical setting once they are tested there.

Although this approach may be useful in the general population, translating the results of population studies to the care and management of individuals requires additional considerations. Risk factors that are not common may be dominant in individual patients. Also, some risk factors may be important for different types of fractures and for different age groups, for example. It is not yet known how tools for risk assessment would perform under these varying circumstances. In addition, clinicians may find it difficult to selectively exclude low-risk women from diagnostic protocols because a substantial number of women without specific risk factors also experience fractures. Patients may have similar concerns.

Are Risk Factors Useful in Treatment Decisions?

Identification of important risk factors could be useful in making treatment decisions. Although models have been proposed, they have not been prospectively tested in large trials and are not used by clinicians. Clinical trials have not focused on what should be done for patients who have several risk factors and a high overall risk of fracture but who do not meet the WHO definition for osteoporosis. A few risk factors -- sex, race, age, menopause, a history of previous fracture, and low bone density -- have been primary selection criteria for randomized trials of bisphosphonates and provide clues to the validity of the risk-based approach.

How Well do the Different Bone Measurement Tests at Different Sites Predict Fractures?

Among different bone density tests measured at various sites, bone density measured at the femoral neck by DXA is the best predictor of hip fracture and is comparable to forearm measurements for predicting fracture at other sites.

In recent prospective studies, QUS measured at the heel predicted hip fracture and all nonspine fractures as well or nearly as well as DXA measured at the femoral neck. For both tests, a result in the osteoporotic range is associated with an increased short-term probability of hip fracture. However, clinical trials of recent pharmacologic therapies have used a low hip DXA, rather than QUS, as a criterion for entry. Physicians who use QUS must consider whether the results of these trials are generalizable to patients identified by QUS to have a high risk of fracture.

QUS of the heel and DXA of the hip provide independent information about fracture risk. Individuals who have low scores by one of these tests, but not the other, have a greater risk of fracture than those who have higher scores by both tests, and a lower risk of fracture than those whose results on both tests are low.

Both of these tests predict hip fracture better than DXA of the lumbar spine. RA or QMD of the hand can predict the risk of nonspine fractures in general, many of which are in the forearm, but there are no recent data about the ability of hand measurement to predict hip fracture. Correlations between different bone measurement tests are generally too low to be accepted as evidence that one test will identify patients at similar risk to those identified by another test. While peripheral measures may approach hip DXA in predicting hip fracture, there are no recent prospective studies that directly compare prediction of hip fracture by these tests with DXA of the hip.

What Factors Related to Bone Testing Influence Diagnosis?

The likelihood of being diagnosed with osteoporosis varies greatly, depending on the site and type of bone measurement test, on the brand of densitometer, and on the relevance of the reference range in the local population. The variation between techniques, along with the lack of methods to integrate bone density results with clinical predictors, makes it difficult for clinicians to provide accurate information to patients about their test results.

The likelihood of being diagnosed with osteoporosis also depends on the number of sites tested. Testing in the forearm, hip, spine, or heel will generally identify different groups of patients. A physician cannot say, based only on a forearm test, that the patient "does not have osteoporosis." Conversely, although the results of a test at any site are associated to some degree with fractures at other sites, the physician may not be able to assess whether the patient who has a low T-score on a hand or forearm test has significant bone loss at other sites.

Because of these limitations, some experts question whether there is added benefit to the patient from using T-scores to diagnose osteoporosis, as opposed to reporting test results as continuous values and assessing overall risk. Others propose that results using various techniques and sites be "calibrated" to the results of DXA of the hip and reported as a "T-score equivalent." The success of this approach may be limited by differences in precision and low correlation among different techniques.

How do Bone Measurement Test Results Affect Patients' and Physicians' Decisions and Actions?

One randomized trial suggests that women who undergo densitometry are more likely to start HRT than women who do not. In a randomized trial and a large uncontrolled case series, women who had densitometry and were told they had osteoporosis were more likely to start or continue HRT than women who were told they had normal bone density. In one randomized trial, physicians found densitometry reports confusing and were not confident that their interpretations of T-scores were correct. While overall the reporting format did not influence treatment decisions, gynecologists who better understood the report were more likely to prescribe changes in HRT than those who considered the report confusing. These results suggest that the behaviors of both patients and physicians are influenced by the results of bone measurement tests.

Are Bone Measurement Tests Effective for Monitoring Response to Therapy and for Guiding Decisions about Changes in Management?

Currently, the weight of evidence is against repeating bone density tests within the first year of treatment. There is insufficient evidence to determine whether repeating bone density tests 2 years after starting therapy is useful. There also are no studies about the effect of monitoring responses to therapy using densitometry, or of the choice of test, on the outcome of therapy.

Can Markers be Used Instead of Bone Measurement Tests in Identifying Women at Risk for Osteoporosis?

No single marker or cluster of markers accurately predicted the results of densitometry in individuals. Densitometry measures current bone status, whereas markers measure the process of bone turnover.

How Well do Markers Predict Fractures?

No marker was associated with increased fracture risk consistently across all studies. The EPIDOS study provides evidence that using markers in conjunction with densitometry may increase predictability, but this result has not been confirmed in other studies.

Can Markers Help Select Patients for Treatment?

Studies correlating marker results and bone loss indicated no clear trend. Sensitivity and specificity were too low to be useful for the purpose of selecting patients for treatment. Some studies found better test accuracy when a combination of two or more markers and/or other risk factors was used to predict bone loss.

What are the Adverse Effects of Using Markers to Identify Women at Risk for Osteoporosis?

The primary adverse effect of biochemical markers is the potential for false-positive and false-negative results. Rates of false-positive and false-negative test results vary widely. If markers were used to select women for treatment, a false-positive test could lead to the initiation of unnecessary treatment. A false-negative result could lead to more serious consequences if markers were to be used as an initial test to select women for further testing.

Can Markers Predict Response to Therapy?

There is a small correlation between response to therapy as measured by densitometry and marker results, but no marker is accurate enough to reliably identify nonresponders to treatment. The best results, from the EPIDOS study, have not been replicated in other studies. The use of bone density at 2 years as the gold standard for measuring response to therapy has been called into question by a recent study that found that bone density at 2 years is not a reliable predictor of subsequent response to therapy.

What Diagnostic or Laboratory Tests are Appropriate for Evaluating Patients with Osteoporosis for Secondary Causes?

There is no evidence on which to base recommendations for a strategy of testing to determine secondary causes of osteoporosis. The accumulated literature suggests only the presence of low bone density in certain disorders such as hyperparathyroidism, multiple myeloma, and hyperthyroidism, but not the prevalence of these disorders given a diagnosis of osteoporosis.

Despite this lack of evidence, expert opinion and practice guidelines provide many suggestions for diagnostic workups. Some support extensive testing to rule out major concomitant disease, while others suggest a limited testing strategy based on findings in the history and physical examination. Because the diagnosis of primary osteoporosis is often seen as a diagnosis of exclusion, the pattern of diagnostic testing may continue to be costly until the diagnostic yield is fully demonstrated.

Within the community, the pattern of testing varies widely, particularly by specialty type. Our supplemental analysis suggested that assumptions about the probability of a secondary disease or disorder to explain the occurrence of osteoporosis vary by practice type and specialty. TSH, chemistry battery, and complete blood count were the most frequently ordered tests cited by respondents. These also were the most frequently recommended tests in our review of expert guidelines.

What are the Costs and Cost-Effectiveness of Diagnostic Strategies for Identifying Women with Osteoporosis?

In the literature we reviewed, economic assessments suggested that diagnosis and treatment of women at risk for osteoporosis would be more cost-effective by targeting treatment to those with the lowest bone measurement results. Inclusion of another assessment prior to the bone measurement test may improve the cost-effectiveness of diagnosis. The other assessment could be a risk profile (such as SCORE) or a less expensive, and perhaps more widely available, diagnostic test such as QUS of the heel.

We conducted cost-effectiveness analyses to estimate cost per hip fracture prevented using data from EPIDOS and SOF (see Supplemental Analysis 2). These analyses suggested that a sequential diagnostic approach was more cost-effective than using DXA of the femoral neck alone. The sequential approach we considered was QUS of the heel followed by DXA of the femoral neck only for those with low values of QUS/BUA. Diagnosis with DXA of the femoral neck alone prevented more fractures in most cases, but at an increased cost per fracture prevented compared with the sequential approach. Using SOF data, if the overall incidence of fractures were increased, that is, if the population were at a five- to tenfold higher risk of fracture than the SOF reported, the incremental cost per hip fracture prevented using diagnosis with QUS alone (compared with DXA alone) dropped. QUS alone may be a cost-effective option in high-risk populations. In senstitivity analyses, if treatment efficacy following diagnosis with QUS were 5 percent or 15 percent less than that following diagnosis with DXA, diagnosis with QUS alone would have higher costs to prevent fewer hip fractures than other diagnostic options. If reduction in fracture risk reduction is as little as 5 percent less than diagnosis with DXA, QUS alone may be dominated by other alternatives.

Methods

We designed a mailed, self-administered questionnaire to measure the following: (1) frequency of use of bone density testing in postmenopausal women, and characteristics influencing the decision to order it; (2) frequency of evaluation for secondary causes of osteoporosis for women with confirmed low bone density; (3) referral patterns for women diagnosed with osteoporosis; (4) principal secondary causes of osteoporosis considered in the evaluation; (5) factors influencing the decision to evaluate secondary causes; and (6) laboratory tests used routinely and selectively as indicated by history and physical examination in women with confirmed osteoporosis.

We used the American Medical Association (AMA) list of U.S. practitioners to identify physicians in active practice who are not in military service; both members and nonmembers of the AMA are contained in this list. To enable a comparison of primary care practitioners to specialists (endocrinologists), 500 physicians each in internal medicine, family practice, and obstetrics and gynecology were randomly selected using a nationwide sampling strategy. One thousand endocrinologists were similarly chosen nationwide. Questionnaires were mailed to subjects in January 2000. Because this was intended as a pilot project, additional mailings were not provided. Responses were accepted up until April 1, 2000. Analysis included descriptive frequencies only.

Results

A total of 367 of the 2,500 mailed questionnaires (15 percent) were completed and returned. The average age of physician respondents was 47.3±10.9 years. Thirty-two percent of the sample was female and 68 percent male. Ninety-three percent of respondents were in practice full-time and 7 percent part-time. Private or group practice types predominated in the sample (76 percent) compared with 10 percent who work in an academic medical center, 5 percent in an HMO, and 6 percent in other settings. Fifty-five percent of respondents practiced in a large city (>200,000 population), 35 percent in a midsized city (25,000-200,000 population), and 10 percent in a small town or rural environment.

Table S1-1. Proportion of postmenopausal women for (more...)

Table S1-1. Proportion of postmenopausal women for which bone density testing is recommended*

	Internists	Family Practitioners	Endocrinologists	Obstetricians/Gynecologists
>75%	22 (42%)	9 (20%)	80 (42%)	15 (23%)
25-75%	20 (38%)	19 (42%)	83 (43%)	25 (38%)
<25%	8 (15%)	16 (36%)	27 (14%)	26 (39%)
None	2 (4%)	1 (2%)	2 (1%)	0

*Percentages may not equal 100 because of rounding

Table S1-2. How often do you obtain bone density measurements (more...)

Table S1-2. How often do you obtain bone density measurements for your postmenopausal women patients?*

	Internists	Family Practitioners	Endocrinologists	Obstetricians/Gynecologists
Frequently (>1 patient/wk)	31 (60%)	15 (33%)	128 (66%)	35 (53%)
Occasionally (>1 patient/mo)	17 (33%)	20 (44%)	59 (31%)	22 (33%)
Rarely (>1 patient/yr)	3 (6%)	10 (22%)	4 (2%)	9 (14%)
Never	1 (2%)	0	2 (1%)	0

*Percentages may not equal 100 because of rounding

Table S1-3. How often do you evaluate a woman with (more...)

Table S1-3. How often do you evaluate a woman with osteoporosis for secondary causes?*

	Internists	Family Practitioners	Endocrinologists	Obstetricians/Gynecologists
Frequently (>1 patient/wk)	10 (20%)	4 (9%)	104 (54%)	12 (18%)
Occasionally (>1 patient/mo)	17 (33%)	11 (24%)	66 (35%)	10 (15%)
Rarely (>1 patient/yr)	22 (43%)	28 (62%)	20 (10%)	34 (52%)
Never	2 (4%)	2 (4%)	1 (1%)	9 (14%)

*Percentages may not equal 100 because of rounding

Three factors were cited as strongly influential in the decision to evaluate a woman for secondary causes by more than 90 percent of the responding physicians: no response to 1 year of treatment for osteoporosis, very low bone density, and osteoporosis in a premenopausal woman. When asked which age group they would most likely evaluate for secondary causes, respondents most often cited the youngest group, women 45-54 years of age. Concomitant serious medical illness, history of an atraumatic fracture, and unwillingness to undergo treatment for osteoporosis were modestly influential in the decision to evaluate osteoporosis in a woman, cited at least half the time by respondents.

Table S1-4. Laboratory tests ordered to assess secondary (more...)

Table S1-4. Laboratory tests ordered to assess secondary causes of osteoporosis

	Routinely	Only if indicated by history & exam
TSH - Thyroid-stimulating hormone	92%	6%
Chemistry battery	87%	9%
Complete blood count	69%	20%
24-hour urine calcium	44%	37%
Parathyroid hormone	43%	45%
Urinalysis	42%	33%
25-OH vitamin D	41%	40%
Serum protein electrophoresis	37%	48%
Ionized calcium	32%	43%
T3/T4 - Thyroxine or T3	31%	47%
Erythrocyte sedimentation rate	24%	50%
Bone-specific alkaline phosphatase	18%	52%
Serum or urinary collagen crosslinks	18%	50%
Urine pH	18%	50%
Cortisol	16%	63%
1,25(OH)2D3	16%	57%
24-hour creatinine clearance	15%	52%
Spinal X-ray or CT scan	10%	60%
Osteocalcin	7%	56%
Blood gas	0.6%	62%
Bone biopsy	0.3%	64%

Conclusions

The results of our analysis of physician practice pattern variation in the evaluation of secondary causes of osteoporosis in women indicated a lack of consensus. Physician test-ordering behavior differed among primary care practitioners and specialists. Family practitioners less frequently recommended bone density tests for postmenopausal women, obtained bone density tests, and underwent evaluations for secondary causes than did internists, endocrinologists, and obstetricians/gynecologists. Endocrinologists more frequently evaluated patients for secondary causes than did other practitioners. Thyroid-stimulating hormone, chemistry battery, and complete blood count were the most frequently ordered tests cited by respondents. These also were the most frequently recommended tests in our review of expert guidelines.

These data are limited by the low response rate and may not be representative of true practice variation that would be observable only within a large data set. Respondents likely represented a physician community most interested in osteoporosis, as evidenced by the higher response rate (21 percent) of endocrinologists, and may not reflect actual community practice.

Background

Our review of the literature on bone measurement tests indicates that dual-energy X-ray absorpiometry of the femoral neck (DXN-FN) or total hip is the best predictor of hip fractures (see section on Bone Measurement Tests, p. 48, above). Clinicians use this approach for making diagnosis and treatment decisions about osteoporosis. Women with T-scores less than -2.5 are considered osteoporotic and may be offered treatment.¹ Women with T-scores between -1 and -2.5 are considered ostepenic and may be monitored more closely for additional loss of bone density. Diagnosis with DXA-FN requires a relatively large and expensive instrument and a relatively high degree of operator skill to correctly position the target area. DXA is usually performed at a hospital or larger medical center and is often done on a referral basis.

Over the last few years, several newer diagnostic approaches, such as quantitative ultrasound (QUS) and peripheral DXA, have become more available. These approaches require less costly equipment and less space than DXA and are becoming more common in primary care offices and smaller clinics. As part of our evidence review, we met with a local technical expert panel (Appendix A). They suggested that a sequential diagnostic approach may become common in primary care. With this approach, a newer and more available diagnostic tool such as QUS is used in a primary care setting. If a woman is identified as being at higher risk of osteoporosis, she is then referred to DXA-FN for a definitive diagnosis. Also, as these diagnostic tools become more available in the primary care setting, clinicians may choose to use one of these newer methods alone.

We developed a simple economic model to compare both the sequential diagnostic approach and diagnosis with QUS alone to diagnosis with DXA-FN alone. This simple model is intended to evaluate potential differences in cost per fracture prevented among these three diagnostic approaches. These results may support a more detailed economic evaluation beyond the scope of this evidence report.

We chose QUS to be representative of the newer and more available diagnostic tools because there are published reports from two cohort studies²,³ that compare hip fracture rates following diagnosis with DXA and with QUS. In addition to these published reports, the investigators of the Study of Osteoporotic Fractures (SOF)⁴ provided data on women in the SOF cohort who received both DXA-FN and QUS and were followed for subsequent fractures.

The World Health Organization¹ and other groups recommend that osteoporotic women be defined as those with DXA T-scores less than -2.5. However, there are no such guidelines for QUS broadband ultrasound attenuation (QUS/BUA). We therefore performed a series of analyses using the SOF cohort data with various QUS/BUA cut points used to define women at high risk of fractures.

Objective

The objective of this economic evaluation is to use a relatively simple economic model to predict, for alternative diagnostic approaches, the total direct medical costs (from a payer/provider perspective) and number of hip fractures in a cohort of 1,000 older women. Data for this analysis was obtained from published studies of the Epidemiologie de l'Osteoporose (EPIDOS) cohort² and unpublished data from the SOF cohort.³ We developed this model to assess the potential for reduced cost per hip fracture prevented using either a sequential diagnostic approach (QUS followed by DXA-FN in those at higher risk of osteoporosis) or QUS alone relative to DXA alone.

Methods

Economic Model

We used a cost-effectiveness approach to estimate the cost per hip fracture prevented. We used a decision tree (Figure S2-1) to estimate costs per hip fracture prevented for three diagnostic approaches. The audience for this economic evaluation includes providers and payers of healthcare. Only direct medical costs (from a payer or provider perspective) are considered.

The target population for these analyses is the same as that for the EPIDOS² and SOF³ studies -- that is, older white women. The EPIDOS study consisted of 5,541 French women (out of 7,575 in the original cohort) with a mean age of 80 years. The original SOF cohort³ (U.S. women age 65 years or more) had been followed for about 5 years when the period for this study began. Of those alive at that time, 5,993 (72 percent) were included in this study, with a mean age of 76 years. In both studies, some centers lacked the QUS instruments at the time the diagnostic tests were taken; thus, not all women in the cohorts were included in these two studies. The fracture data in the reports on the EPIDOS and SOF cohorts had an average follow-up time of about 2 years. While this period is shorter than the period of most clinical trials of therapy to reduce fracture rates, we have used this 2-year time horizon for our analyses.

We considered three alternative diagnostic approaches: DXA-FN alone, QUS alone, and a sequential approach where QUS is used first and, for those at high risk, DXA-FN also is performed. Because there are no recommended cut points for QUS/BUA that define high risk of fracture, we compared the alternative diagnostic approaches across a series of QUS/BUA cut points (from 50 to 85 dB/MHz in steps of 5 dB/MHz). These approaches are summarized in Table S2-1 with respect to test(s) performed and subsequent treatment (based on the test results). Figure S2-2 illustrates which women would be treated given these alternatives based on whether test values are above or below the cut point for DXA-FN and/or for QUS. With each approach, all women diagnosed as osteoporotic (that is, at high risk of fracture) receive treatment. While treatment is not the focus of this report, differences among treatments with respect to reductions of risk of hip fracture and costs of the treatments may affect the cost per fracture prevented. The diagnostic approaches are compared within treatments but not across treatments. We restricted our economic evaluation to a one-time diagnostic test (no follow-up monitoring).

We used number of hip fractures as the only outcome of interest. We did not explicitly model sequelae of hip fractures (such as death or admission to a nursing home), although the costs of these sequelae were included in the cost of a hip fracture.

Results

Base Case

For treatment with alendronate, the results are summarized in Figure S2-3 and Appendix S2-1. In the top panel of Figure S2-3 (and subsequent Figures), the diagnostic alternatives are in the same order across QUS/BUA cut points. The alternative with the lowest incremental cost-effectiveness ratio (i.e., the most cost-effective) is compared to no diagnosis and no treatment. The alternative with the next higher incremental cost-effectiveness ratio is compared to the alternative with the next lower incremental cost-effectiveness ratio. For example, at 65 dB/MHz, the most cost-effective alternative is the sequential approach, DXA alone the next most cost-effective, and QUS alone the least cost-effective. Also, the incremental cost-effectiveness ratio is not included for alternatives that are dominated (including dominated by extended dominance), but the total direct medical costs and hip fractures prevented are displayed for all alternatives.

Appendix S2-1 summarizes results for all diagnostic alternatives, including those that are dominated. The sequential diagnostic approach has the lowest incremental cost per fracture prevented compared to no diagnosis and treatment. For QUS/BUA cut points of 70 dB/MHz or less, DXA is the next most cost-effective alternative, but the incremental cost per fracture prevented of DXA over the sequential option is substantial, at least 50 percent greater in relative terms or more than $85,500 per hip fracture prevented in absolute terms. At QUS/BUA cut points of 75 dB/MHz or greater, DXA alone is dominated. QUS alone is not a cost-effective alternative at cut points of 75 dB/MHz or greater. The cost per hip fracture prevented is around three times as great or greater as the cost per hip fracture for the sequential approach.

For HRT, the results are summarized in Figure S2-4 and Appendix S2-1. The sequential approach is again the most cost-effective option except at a QUS/BUA cut point of 85 dB/MHz, where DXA alone is the most cost-effective alternative. The next most cost-effective options, either DXA alone in four cases or QUS alone in three cases, have substantially higher costs per fracture prevented (at least 40 percent higher).

If calcium and vitamin D were used as a treatment, QUS alone is the most cost-effective alternative (cost per fracture prevented ranging from about $118,000 to $130,000 across the cut points). DXA alone is the only other nondominated alternative, but only for cut points of 60 dB/MHz or less, and the incremental cost per fracture prevented is about four times or more greater than for QUS alone. Results are summarized in Appendix S2-1.

Sensitivity Analyses

This summary will focus primarily on scenarios in which the order of the diagnostic alternatives changes from the base case. For scenarios not discussed, the magnitudes of the incremental cost-effectiveness ratios may differ across alternatives, but the order is the same as the base case. In some cases, the change in magnitude of the incremental cost-effectiveness ratios indicates that an alternative may become relatively cost-effective for certain sensitivity parameters. These cases also are cited.

The analyses are relatively insensitive to the cost of alendronate treatment. The only scenario in which the order differed from the base case is at 50 dB/MHz with the low cost of alendronate. In this case, QUS alone is not dominated, and the incremental cost per hip fracture prevented ($153,464) is intermediate between that of the sequential approach ($87,066) and that of DXA alone ($194,493). For sensitivity analyses on the cost of HRT, the results are summarized for low and high HRT costs for cut points of 50, 55, and 60 dB/MHz in Figure S2-5. At low HRT costs, QUS alone is the most cost-effective alternative at a cut point of 50 dB/MHz (the sequential approach is dominated), and QUS has only a slightly higher incremental cost-effectiveness ratio at 55 dB/MHz than the sequential approach, which is the most cost-effective alternative at this cost and cut point combination.

For sensitivity analyses for cost of DXA and treatment of osteoporotic women with alendronate, there is only one substantive deviation from the base case. At the high value of DXA cost ($183), the incremental cost per hip fracture prevented for QUS alone (about $230,716) is intermediate between the incremental cost-effectiveness ratio for the sequential approach ($123,400) and for DXA alone ($294,312). For treatment with HRT, there are several differences between the base case and the sensitivity scenarios at QUS/BUA cut points of 50, 55 or 60 dB/MHz. These are plotted in Figure S2-6 by QUS cut point and by cost of DXA. Finally, for high-cost DXA at QUS/BUA cut point of 85 dB/MHz, the sequential diagnostic approach replaces DXA alone as the most cost-effective alternative, and DXA alone is dominated for this scenario. Thus, the results are somewhat sensitive to DXA costs, especially the high cost for DXA, when HRT treatment is used for osteoporotic women.

Sensitivity analyses for cost of QUS showed no substantive changes for the analyses with alendronate treatment. For treatment with HRT, at QUS/BUA cut points of 70 and 80 dB/MHz, in the scenarios using the higher cost for QUS ($54), DXA alone dominated the sequential procedure. At QUS/BUA cut point of 85 dB/MHz, in the scenario of low cost for QUS ($30), the sequential diagnostic approach dominated DXA alone.

Finally, we performed sensitivity analyses on reduction in the degree of risk reduction achieved by therapy in women diagnosed as osteoporotic using QUS alone. For treatment with alendronate, QUS alone was the least cost-effective alternative at QUS/BUA cut points of 65 and 70 dB/MHz. With 5 percent reduction in risk reduction, QUS alone became dominated at 65 dB/MHz, and with 15 percent reduction, it was dominated at 70 dB/MHz. At QUS/BUA cut point of 75 dB/MHz, DXA was a more cost-effective alternative than QUS diagnosis (DXA was dominated in the base case) with 10 percent or greater reduction in risk reduction. At cut points of 80 and 85 dB/MHz, there was no substantive difference between sensitivity and base case scenarios up to a reduction in risk reduction of 30 percent.

For treatment with HRT at QUS/BUA cut points of 50 and 65 dB/MHz, QUS alone -- not dominated in the corresponding base case scenarios -- is a dominated alternative with even a 5 percent risk reduction. At a cutpoint of 70 dB/MHz, QUS alone is dominated if the reduction in risk reduction is 15 percent. At a cut point of 75 dB/MHz, diagnosis with DXA alone, which is dominated in the corresponding base case scenario, is not dominated with even a 5 percent reduction in risk reduction and QUS alone is dominated with a 30 percent reduction in risk reduction. For all cut points except 85 dB/MHz, however, the sequential approach remained the most cost-effective alternative.

For treatment with calcium and vitamin D, QUS alone was the only nondominated alternative for QUS/BUA cut points of 65 dB/MHz or greater. If the reduction in risk reduction was at least 5 percent at 65 dB/MHz or at least 15 percent at 70 dB/MHz, then DXA alone is a nondominated alternative, although with a relatively high incremental cost-effectiveness ratio for most reductions in risk reduction. At QUS/BUA cut point of 75 dB/MHz, DXA alone is the only nondominated alternative for a 30 percent reduction in risk.

Sensitivity analyses for costs of calcium and vitamin D, risk reduction in response to treatment (alendronate, HRT, or calcium plus vitamin D), costs to treat fractures, and fracture rates did not show any changes in the order of the alternatives with respect to incremental cost-effectiveness ratios. The magnitude of cost-effectiveness ratios did change somewhat across these sensitivity analyses.

QUS alone in the base case is only cost-effective for treatment with calcium and vitamin D. As noted in the National Osteoporosis Foundation report,⁴ this is an alternative in which diagnosis may not be worth performing, because the cost and adverse experiences of both calcium and vitamin D are so low as to be relatively negligible. For treatment with alendronate, QUS alone is not dominated for cut points of 70 dB/MHz or higher. While diagnosis with QUS alone prevents the most fractures in these scenarios, the incremental cost-effectiveness ratios are all around $500,000 per fracture prevented or greater. One could ask whether, in a higher-risk population than the SOF cohort, diagnosis with QUS alone would be more cost-effective.

To address this question, we performed a series of analyses in which the risk of fractures in the population was increased by five-, seven- , and tenfold. In these scenarios, the order of alternatives is the same as in the base case. However, for higher-risk groups the incremental cost per hip fracture prevented drops for all alternatives. In particular, the incremental cost per hip fracture prevented for QUS alone is below $100,000 for several scenarios. For example, with alendronate treatment at 70 dB/MHz and a tenfold increase in fracture risk, the incremental cost per hip fracture prevented is $2,937 saved per hip fracture prevented for the sequential approach, $5,618 per hip fracture prevented for DXA alone, and $81,954 for QUS alone. QUS alone prevented 4.17 more fractures than DXA alone. Similarly, the incremental cost per hip fracture prevented following diagnosis with QUS alone at 75 dB/MHz given a 5-fold increase in fracture risk is $93,997 -- and for a 10-fold increase, $35,933. In these cases, almost 6 and more than 12 additional hip fractures are prevented, respectively, compared to the sequential approach. Similar results are observed for higher QUS/BUA cut points with alendronate and for treatment with HRT for QUS/BUA cut points of 70 dB/MHz or greater. Thus, QUS alone may be a cost-effective alternative in higher-risk populations.

Overall Conclusion

Overall, the sequential diagnostic approach may provide a cost-effective alternative to diagnosis with DXA-FN alone. Further, in a population with a much higher overall fracture risk than SOF, QUS alone may also be cost-effective. However, the results for diagnosis with QUS alone are sensitive to any reduction in efficacy following diagnosis with QUS. As diagnostic tools such as QUS become more readily available in the primary care setting, research that provides guidance to clinicians as to the best use of such tools is required. Based on the conclusion of the simple model presented here, use of QUS either for diagnosis of osteoporosis or to determine which women should receive DXA-FN may be appropriate and cost-effective.

Oregon Health & Science University Evidence-based Practice Center Osteoporosis Task Order Research Team

Local Panel

Risk Factors

1 exp osteoporosis
osteoporosis, postmenopausal

2 bone density

3 1 or 2

4 exp risk

5 3 and 4

6 exp transplantation

7 exp kidney failure

8 su.fs. (Surgery as a subheading anywhere in the article)

9 athletic injuries

10 exp sports

11 exp fractures/dt,su,th (limited to surgery and other therapies)

12 orthopedic$.mp. (As a textword anywhere)

13 6 or 7 or 8 or 9 or 10 or 11 or 12

14 5 not 13 (statements 6 through 12 were excluded from the study)

15 limit 13 to female

16 limit 14 to human

17 limit 15 to english language

18 looked at english abstracts of foreign articles

19 exp fractures or exp osteoporosis or bone density

20 exp risk (terms as in 4)

21 19 and 20

22 exp cohort studies
longitudional studies
follow-up studies
prospective studies

23 meta-analysis

24 exp case control studies
retrospective studies

25 predictive value of tests

26 evidence-based medicine

27 22 or 23 or 24 or 25 or 26

28 21 and 27

29 limit 28 to human

30 limit 29 to english language

Bone Density and Quantitative Ultrasound Testing

1 exp osteoporosis
osteoporosis, postmenopausal

2 bone density

3 1 or 2

4 densitometry, x-ray

5 exp ultrasonography

6 calcaneous/us (us = ultrasonics)

7 (dxa or sxa or bua or qct or qus or mxa or mrx or ra or dip or sos or ubps or spa or dpa).tw.

8 exp osteoporosis/us (us = ultrasonics)

9 4 or 5 or 6 or 7 or 8

10 3 and 9

11 limit 10 to human

12 limit 11 to english language

13 looked at english abstracts of foreign articles

Monitoring

1 exp osteoporosis/
osteoporosis, postmenopausal

2 bone density/

3 1 or 2

4 monitoring

5 densitometry, x-ray/

6 exp ultrasonography/

7 calcaneus/us

8 4 or 5 or 6 or 7

9 3 and 8

10 exp osteoporosis/dh,dt,th

11 9 and 10

12 limit 11 to human

Biochemical Markers

1 exp osteoporosis
osteoporosis, postmenopausal

2 bone density

3 1 or 2

4 exp biological markers
genetic markers

5 marker$.tw. (markers as a word in the title or abstract)

6 (bap or ctx or dpd or ntx or pyd).tw. (marker abbrev. in title and abstract)

7 4 or 5 or 6

8 3 and 7

9 limit 8 to human

10 limit 9 to english language

11 limit 10 to female

12 looked at english abstracts of foreign articles

Cost

1 exp osteoporosis

2 ec.fs. (ec=economics)

3 exp costs and cost analysis

4 exp economics

5 2 or 3 or 4

6 1 and 5

7 limit 6 to english language

8 looked at english abstracts of foreign articles

Introduction

The Methods Work Group for the Third U.S. Preventive Services Task Force (USPSTF) developed a set of criteria by which the quality of individual studies could be evaluated. At its September 1999 quarterly meetings, the USPSTF accepted the criteria, and definitions of quality categories relating to internal validity.

Design-Specific Criteria and Quality Category Definitions

Presented below are a set of minimal criteria for each study design and a general definition of three categories -- "good," "fair," and "poor." These specifications are not meant to be rigid rules but rather are intended to be general guidelines, and individual exceptions, when explicitly explained and justified, can be made. In general, a "good" study is one that meets all criteria well. A "fair" study is one that does not meet (or it is not clear that it meets) at least one criterion but has no major limitations. "Poor" studies have at least one major limitation.

Systematic Reviews

Criteria:

• Comprehensiveness of sources considered/search strategy used

• Standard appraisal of included studies

• Validity of conclusions

• Recency and relevance

Case Control Studies

Criteria:

• Accurate ascertainment of cases

• Nonbiased selection of cases/controls with exclusion criteria applied equally to both

• High response rate

• Diagnostic testing procedures applied equally to each group

• Measurement of exposure accurate and applied equally to each group

• Appropriate attention to potential confounding variable

Randomized Controlled Trials (RCTs) and Cohort Studies

Criteria:

• Initial assembly of comparable groups
  -- for RCTs: adequate randomization, including first concealment and whether potential confounders were distributed equally among groups
  -- for cohort studies: consideration of potential confounders with either restriction or measurement for adjustment in the analysis; consideration of inception cohorts

• Maintenance of comparable groups (includes attrition, crossovers, adherence, contamination)

• Important differential loss to followup or overall high loss to followup

• Measurements: equal, reliable, and valid (includes masking of outcome assessment)

• Clear definition of interventions

• Important outcomes considered

• Analysis: adjustment for potential confounders for cohort studies, or intention to treat analysis for RCTs.

Diagnostic Accuracy Studies

Criteria:

• Screening test relevant, available for primary care, adequately described

• Study uses a credible reference standard, performed regardless of test results

• Reference standard interpreted independently of screening test

• Handles indeterminate results in a reasonable manner

• Spectrum of patients included in study

• Large sample size

• Administration of reliable screening test