Osteoporosis-related fractures are associated with significant morbidity, mortality, and health care expenditure worldwide. The low sensitivity of bone density testing alone to predict fractures has led to the development of a variety of fracture assessment tools that use the combination of bone density and clinical risk factors to improve the prediction of low-trauma fractures. These fracture assessment tools quantitatively predict the 10-year probability of hip and major osteoporosis-related fractures, and can be used with various intervention strategies to effectively intervene with cost-effective therapies to prevent future fractures.
Osteoporotic fractures often result in significant disability, increased morbidity and mortality, and significant psychological and financial burden to the affected individuals, their families, and society. Incident osteoporosis-related fractures in the United States are expected to increase from 2 million fractures in 2005 to a projected 3 million fractures per annum by 2025. The associated direct and indirect fracture-related costs to care for these patients are expected to increase from $17 billion to $25 billion by 2025. There is a general awareness by both physicians and the lay public of the overall fracture burden of osteoporosis on society, the presence of proven therapeutic interventions approved by the US Food and Drug Administration (FDA) for the prevention and treatment of osteoporotic fractures and the necessity to effectively predict fracture events.
The diagnosis of osteoporosis is derived from the World Health Organization (WHO) central dual-energy X-ray absorptiometry (DXA) diagnostic criteria of a T-score of −2.5 or less performed at the lumbar spine, femoral neck, total hip, or one-third radius sites. DXA bone mineral density (BMD) testing provides the clinician with an estimate of fracture risk in terms of a continuum of risk rather than a specific cut point below which most patients will fracture. It is used to calculate a patient’s relative risk for any T-score or Z-score. The accuracy of BMD measurements using central DXA to predict osteoporotic fractures is comparable to the use of blood pressure measurement for prediction of stroke and is considerably superior to serum cholesterol as a predictor of myocardial infarction. Risk stratification using BMD has been delineated by the large Marshall meta-analysis, which showed that risk of fracture increases by 1.4-fold–2.6-fold for every standard deviation (SD) decrease in BMD compared with the reference population used in the calculation (applies to T-scores or Z-scores). Prediction of fracture is enhanced when using site-specific measurements such as spine BMD to predict spine fractures and femoral neck BMD to predict hip fractures. Thus, for every 1 SD (equivalent to 1 T-score) decrease in spine BMD there is a 1.8-fold increased risk of spine fractures, and for every 1 SD decrease in hip BMD there is a 2.6-fold increased risk of hip fractures. These measures of increased risk per SD decrease in BMD T-scores are referred to as gradients of risk. Calculation of an individual’s relative risk of fracture can be performed by taking the gradient of risk at the site measured to the power of the T-score or Z-score (GR T/Z ). An individual with a T-score of −2.0 SD at the femoral neck has a relative risk of 2.6 2 or a sevenfold increased risk compared with an individual with a T-score of 0. Substitution of the Z-score for T-score provides a relative risk of an individual compared with an age, gender, and race-matched or ethnicity-matched individual with a Z-score of 0.
Large epidemiologic studies in the United States, Europe, and Australia have established that greater than 50% of osteoporosis-related fractures (low-trauma, fragility fractures) occur in patients who have a central DXA test result consistent with the WHO’s criteria of osteopenia (low bone density), T-scores between −1.0 and −2.5, rather than osteoporosis, T-score less than or equal to −2.5. The Study of Osteoporotic Fractures (SOF) found that 54% of women who sustained a hip fracture during the 5 years of the study had either osteopenia or normal bone density. Together, these population-based studies indicate that incidence rates of osteoporosis-related fractures increase with increasing age and decreasing bone density, and the greatest number of low-trauma fractures occur in patients diagnosed with osteopenia.
The low sensitivity and low positive predictive values using BMD testing alone for prediction of fracture risk has led to the development of new strategies that are inclusive of risk stratification algorithms that consider independent clinical risk factors for fracture. Age, personal history of fracture, prevalent fractures, parental history of osteoporosis-related fractures, glucocorticoid use, and so forth add to the gradient of risk and more accurately predict future fracture risk in the individual patient. As an example, compare fracture prediction measures using BMD alone and relative risk versus calculation of absolute risk probabilities for 50-year-old and 80-year-old white women from the United States with identical T-scores of −3 but having no additional clinical risk factors for fracture other than their age disparity. Both women have the diagnosis of osteoporosis by WHO diagnostic criteria and identical 2.6 3 or an 18-fold increased relative risk of fracture compared with an individual with a T-score of 0. If both women’s future fracture probability is calculated using one of the available fracture risk calculators, the risk of sustaining a fracture at the hip, spine, forearm, and humerus within the next 10 years is approximately 8% in the 50-year-old woman and 23% in the 80-year-old woman. Inclusion of age as a known independent clinical risk factor results in a threefold greater risk of fracture in the older individual that is not appreciated by BMD testing alone or use of relative risk.
In absolute fracture prediction, improvement in accuracy requires an increased gradient of risk to change the performance characteristics of the test. In addition to age, gender, weight, and height (calculated body mass index [BMI]), other easily identifiable clinical risk factors (CRFs) that contribute independently to fracture risk include a personal or family history of fragility fractures, use of glucocorticoid medication, rheumatoid arthritis, cigarette smoking, excessive alcohol intake, and a variety of secondary conditions that contribute or cause fragility fractures that are inclusive of falls. The addition of non-BMD CRFs increases the performance characteristics of fracture prediction and calculation of quantitative output measures of future fracture risk. Available fracture prediction algorithms such as the WHO Fracture Risk Assessment Tool (FRAX), the Garvan Institute fracture risk calculator (Garvan), and QFractureScores (QFracture) can be seen in Figs. 1 and 2 .
The WHO Collaborating Center for Metabolic Bone Diseases developed FRAX as a robust computer-based algorithm that calculates the 10-year probability of (1) hip fractures and (2) major osteoporosis-related fractures (hip, clinical spine, humerus, and forearm), with or without inclusion of femoral neck BMD. FRAX calculates both fracture probabilities from easily obtained CRFs in both men and women and in the United States for Asian, black, white, and Hispanic people. Using local mortality and hip fracture rates, FRAX has been calibrated for use in more than 30 other countries. FRAX calculations are based on the following variables: age, BMI, parental history of hip fracture, personal history of fragility fracture, current tobacco smoking, excessive alcohol intake, ever use of oral glucocorticoid medication, rheumatoid arthritis, and other secondary causes for osteoporosis, and takes into account the risk of death. The information necessary to determine the interactions between CRFs with and without BMD has been derived from a meta-analysis of 60,000 patients in 9 prospective population-based cohorts from North America, Europe, Asia, and Australia and has been validated in an additional 11 prospective cohorts involving 230,000 patients. This paradigm permits the determination of the predictive power of each CRF and interaction between CRFs by using multivariate analysis, thus optimizing the accuracy of major osteoporosis-related and hip fracture probabilities.
As in the case of DXA BMD testing, FRAX 10-year probabilities have been included in many clinical guidelines and should be considered in clinical medicine as a reference point to assist in the determination of overall osteoporosis fracture risk. FRAX is FDA approved for incorporation into DXA machines’ standardized printouts to provide a patient’s 10-year fracture probabilities in addition to BMD and vertebral fracture assessment (VFA) as unique point-of-service diagnostic testing. DXA-based VFA is also an important adjunctive imaging tool in patients who meet specific criteria for this test and permits the diagnosis of previously undiagnosed vertebral fractures. The diagnosis of an unsuspected low-trauma vertebral fracture is consistent with the clinical diagnosis of osteoporosis that is independent of the additive effect of a personal history of fracture in FRAX and may result in a potential reevaluation of pharmacologic intervention.
Patient demographics and CRF screening can be inserted into the FRAX algorithm by the technologist or interpreting clinician using a FRAX questionnaire. The questionnaire can be filled out by the patient before or at the time of bone density testing and VFA. The tool can be used online by anyone with Internet access, downloaded to an i-phone and i-pad, or used with downloaded hand-held charts. Although not as accurate as FRAX with inclusion of BMD ( Fig. 3 ), the algorithm is able to calculate fracture probabilities without BMD ( Fig. 4 ) and is used to screen patients for further evaluation and therapeutic intervention by clinicians as part of a case-finding strategy.
FRAX predicts hip-related and major osteoporosis-related fractures using BMD and 7 CRFs that contribute independently of BMD to fracture risk, and has been endorsed by the National Osteoporosis Foundation (NOF), International Society for Clinical Densitometry (ISCD), and International Osteoporosis Foundation (IOF). FRAX performance characteristics have been validated in 2 long-term observational studies. The Manitoba Bone Density Program is a long-term observational study that, independently of the FRAX cohorts, tested the performance of FRAX in 36,730 women and 2873 men using data linkage between various Canadian provincial health care databases. This method permitted a direct comparison of fracture risk estimates by the Canadian FRAX tool with fractures observed during 10 years. Ten-year estimates using Kaplan-Meir curves for hip fractures in women were 2.7% (95% CI 2.1%–3.4%) compared with those predicted by the Canadian FRAX tool of 2.8%, which included BMD. In men, the observed hip fracture risk was 3.5% (95% CI 0.8%–6.2%) compared with 2.9% predicted from the Canadian FRAX tool. The observed major osteoporosis-related fractures for all women during 10 years was 12% (95% CI 10.8%–13.4%) compared with 11.1% predicted with inclusion of BMD in Canadian FRAX, whereas for all men the observed incidence of major osteoporosis-related fractures was 10.7% (95% CI 6.6%–14.9%) compared with a predicted 8.4%. Receiver operating curve analysis of the data was 0.830 (95% CI 0.815–0.846) for hip fractures and 0.694 (95% CI 0.684–0.705) for major osteoporosis-related fractures. Canadian FRAX with BMD had better performance characteristics for predicting fracture during 10 years of observation than Canadian FRAX without BMD (CRFs) or BMD alone.
In 2010, the NOF and ISCD published a FRAX Implementation Guide to ensure the appropriate use of FRAX in the United States. The NOF-ISCD FRAX Implementation Guide suggests that FRAX should be used in the following individuals:
Untreated postmenopausal women or men aged 50 years or older
Osteopenia or low bone mass (T-score between −1.0 and −2.5)
No prior hip or vertebral fracture (clinical or morphometric)
An evaluable hip BMD for inclusion in FRAX.
NOF-ISCD FRAX Implementation Guide software can be installed on all central DXA machines in the form of a default filter. The default filter is used to limit the use of FRAX to only individuals who meet the previously listed criteria (a–d). Central DXA testing facilities have the option to either use FRAX with the Implementation Guide FRAX filter in a default mode (filter always on), not use the filter (filter always off), or switch the filter on or off by intervening at the time of DXA testing.
The NOF-ISCD FRAX Implementation Guide includes a disclaimer of ‘This 10 year fracture risk estimate was calculated using FRAX version [X] and a “yes” response for the following FRAX risk factors in this individual: maternal/paternal history of hip fracture, tobacco use, etc.’ The NOF-ISCD Implementation Guide also provides guidance for determining when a previously treated patient can be considered as untreated for purposes of inclusion in the FRAX tool. Untreated patients include those who, in the past year, have not received estrogen, hormone treatment, a selective estrogen receptor modulator, teriparatide, and denosumab, or, in the past 2 years, a bisphosphonate unless taken orally for less than 2 months. Calcium and vitamin D supplementation are not considered treatment by the NOF-ISCD FRAX Implementation Guide .
There are important caveats of which DXA center interpreters of BMD and FRAX probabilities and treating clinicians should be aware when using the NOF-ISCD FRAX Implementation Guide default filter. The default filter does not permit use of FRAX in patients with normal or osteoporotic BMD, although no scientific data exist that substantiate a significant difference in fracture risk between T-scores of −1.0 and −1.1 or between T-scores of −2.4 and −2.5. Healthy younger men and women who have small skeletal structure (small bones) or less than average peak bone mass primarily on a genetic basis without other CRFs for fractures being present may have low FRAX probabilities but are presently being recommended for treatment by the 2008 NOF Clinician’s Guide for the Diagnosis and Treatment of Osteoporosis (T-score ≤−2.5). Epidemiologic data do not support treatment of these low-risk patients, and a major nuance of the Clinician’s Guide compared with previous 2005 NOF guidance supports the premise that fewer young patients at low risk should be treated. Patients with normal spine and hip BMD and no prevalent osteoporotic fractures but high FRAX probabilities would not be treated if the default filter was operational. For example, an 85-year-old white woman with specific CRFs (femoral neck T-score of −1.0, smokes, maternal hip fracture, rheumatoid arthritis, corticosteroid treatment) has major osteoporosis and hip fracture probabilities of 39% and 31% respectively, meets the Clinician’s Guide FRAX high-risk fracture probability thresholds of greater than or equal to 20% for major osteoporosis-related fractures and greater than or equal to 3% for hip fracture for initiation of treatment, but would not have a fracture probability calculated if the default filter was operational. Restricting FRAX to only patients with low bone mass assumes that prospective randomized control trial data are available that confirm the efficacy of pharmacologic treatment in patients with low bone mass but not with normal BMD, a highly controversial subject, whereas recent publications document pharmacologic benefit with or without BMD using FRAX for risk assessment with many, but not all, osteoporosis medications. Aligning FRAX to the restrictive randomization criteria used in clinical trials selects patients for treatment based on only clinical trial data and is analogous to the invalid argument of why not to monitor BMD that is primarily derived from similar data. In the National Osteoporosis Risk Assessment (NORA), patients with low-trauma fractures at nonhip and nonvertebral skeletal sites (eg, forearm and ribs) would be filtered out by the default filter with a BMD T-score greater than or equal to −1.0 (normal BMD) and not considered for treatment even though their future fracture risk is high. The default filter also excludes patients without an evaluable hip BMD. Although FRAX estimates without inclusion of femoral neck BMD are not as sensitive or predictive of future fracture risk as those inclusive of femoral neck BMD, FRAX without femoral neck BMD can still be useful in providing the clinician and patient with 10-year fracture probabilities that facilitate treatment decisions. Clinicians specializing in osteoporosis frequently use FRAX as part of their initial evaluation of patients who have been prescribed osteoporosis treatment by previous physicians. Although the patient’s calculated fracture probabilities are not reliable as a result of being on treatment, it is important for the treating clinician to appreciate, at the time of initial evaluation, a patient’s underlying hip and major osteoporosis-related fracture probabilities if they had never been treated with pharmacologic therapy. This additional piece of critical clinical information may be used by the clinician to continue or discontinue therapy or consider a temporary drug holiday. Excluding use of FRAX in these subsets of patients may deprive clinicians and patients’ of valuable information not otherwise available.
There are important CRFs that predict future fracture risk but are not incorporated into FRAX and, to variable degrees, in the other available fracture prediction tools, thus limiting each tool’s ability to refine a particular risk for the individual patient. An example of a limitation of FRAX is the age range used in the tool, which is between 40 to 90 years. Women who have had a surgical menopause at age 30 years and are not taking hormone treatment would be considered to be 40 years old with the attendant risk of a premenopausal woman. A similar argument has arisen when applying the same rule to a 95 year old who would be classified as being 90 years old. The FRAX CRFs that are treated as dichotomous or yes/no variables can be found in Table 1 . The inability to adjust the dose and, in many cases, the duration of exposure of the dichotomous CRFs may result in an underestimation or overestimation of hip-related and major osteoporosis-related fracture risk in the patient who is not within the range of CRF dose and duration of use that was initially used for the calculation of risk in the FRAX prospective observational studies. Fracture risk associated with the chronic use of high-dose glucocorticoids, exposure to alcohol and tobacco beyond the estimates in the base FRAX populations, a parental history of nonhip fragility fracture, and a personal history of multiple fragility fractures including multiple morphometric vertebral fractures (radiograph confirmed) will all exceed FRAX estimates and thus be underestimated in the calculation of hip-related and major osteoporosis-related 10-year fracture probabilities.
|Previous fracture||Spontaneous or low-trauma fracture in adult life (trauma that would not have resulted in fracture in a healthy person)|
|Parental history of hip fracture||History of hip fracture in patient’s mother or father|
|Current smoking||Presently smoking tobacco|
|Glucocorticoids (≥3 mo)||Present or past exposure to oral prednisone or its equivalent ≥5 mg/d for more than 3 mo|
|Rheumatoid arthritis||Confirmed diagnosis of rheumatoid arthritis|
|Secondary causes||Disease or medical conditions strongly associated with osteoporosis (diabetes mellitus, hypogonadism, premature menopause, osteogenesis imperfecta, untreated hyperthyroidism, malabsorptive diseases, chronic liver disease)|
|Alcohol||3 or more units per day (1 unit = standard glass of beer, single measure of spirits, medium glass of wine)|
Exposure to medications other than prednisone, such as anticonvulsants, lithium, and antiestrogenic and antiandrogenic medications that are known to adversely influence skeletal health are not included in FRAX. Secondary causes that are known to adversely affect skeletal health are only included in the calculation of fracture probability when BMD is not included in the risk calculation. Inclusion of BMD or T-scores into the FRAX tool is limited to the femoral neck because of having the highest gradient of risk of all potential measurement sites for hip fracture. However, it is common that clinicians encounter discordant DXA test results in which the lumbar spine BMD is 1 or more SDs lower than the femoral neck. Nevertheless, inclusion of spine BMD or the equivalent spine T-score instead of femoral neck BMD (or equivalent femoral neck T-score) in the FRAX tool will not generate accurate fracture probabilities for hip-related and major osteoporosis-related fractures.
Up to 30% of seniors living in the community fall each year, with 10% of the falls resulting in hip-related and other major osteoporosis-related fractures and, as such, should be considered in a comprehensive risk management strategy. However, falls are not included in FRAX, other than the assumption that fall risk is included, but are not acknowledged or recorded in the initial cohorts used to construct FRAX. Additional osteoporosis risk calculators are available, including the Garvan Institute fracture risk calculator (Garvan) and QFractureScores (QFracture) fracture prediction tools, which do incorporate falls. FRAX does not allow insertion of falls, whereas Garvan provides options for 0, 1, 2, and 3 or more falls in the past 12 months, and QFracture allows a dichotomous yes/no answer for falls within the past 12 months. There are significant differences in the manner in which the 3 calculators were constructed, CRFs entry data, and the definition of osteoporosis fracture probabilities that does not permit a valid comparison between the 3 tools. Despite these differences, it is worthy of note how falls may affect fracture probabilities using FRAX, Garvan, and QFracture, as seen in Fig. 5 . A summary of the strengths and limitations of FRAX is given in Table 2 .