FRAX Facts

Man is a tool-using animal…. Without tools he is nothing, with tools he is all. Thomas Carlyle, Sartor Resartus, bk. 1, ch. 5

A tool knows exactly how it is meant to be handled, while the user of the tool can only have an approximate idea. Milan Kundera, the Book of Laughter and Forgetting, pt. 7, ch. 8 (1978, trans. 1980)

Osteoporosis is a serious public health problem. Worldwide, an estimated 200 million adults have osteoporosis; current projections indicate that, by 2010, 40 million Americans will have osteopenia and 12 million will have osteoporosis. Although there have been many advances in the diagnosis and treatment of osteoporosis in the past two decades, it remains an underdiagnosed and undertreated disease. In the United States, only 20% of patients at risk for osteoporosis are evaluated or treated.⁽¹⁾ This is caused in part by user-unfriendly nomenclature and reporting (T-scores in negative numbers, BMD given to three decimal places) and expressions of relative rather than absolute risk (“You have a 2-fold increased risk of fracture” or “This treatment will reduce your risk of fracture by 50%”—but compared with what?). Identifying a high-risk population using BMD measurement (T-score of −2.5 or below) is a start, but one half or more osteoporosis-related fractures occur in patients with T-scores better than −2.5, which are in the “osteopenic” or normal range.^(2–4) Most of the guidance from professional organizations has focused on postmenopausal white women, with limited or no recommendations for osteoporosis care in nonwhite women or in men.

Although low BMD is a strong predictor of future fracture risk, other factors are also important. Age is a powerful independent risk factor that has largely been ignored in clinical guidelines. Family history of osteoporosis, personal history of fragility fracture as an adult, low body weight, and cigarette smoking are frequently mentioned as factors to consider in deciding whom to test and whom to treat. Specialists may use this information in an intuitive fashion, but clear guidance for primary care physicians has not been provided. Thus, there is a need for a systematic approach to risk assessment.

Although other fracture risk assessment tools have been proposed,^(5,6) they have not been widely adopted. In February 2008, a long-awaited tool was released by the World Health Organization.⁽⁷⁾ Known as FRAX, it was developed and validated under the direction of Prof. John Kanis with the support of many individuals and organizations including the American Society for Bone and Mineral Research, the National Osteoporosis Foundation, International Society for Clinical Densitometry, and the International Osteoporosis Foundation. FRAX is currently available online at http://www.shef.ac.uk/FRAX. At some point, it is likely to be incorporated into the reporting software for bone densitometry equipment and physicians will need to know how to best use this information for patient care.

When choosing risk factors to be included in the model, the developers of FRAX limited the number and complexity to allow ease of input, selected only well-recognized, independent contributors to fracture risk, and, of necessity, did not include many risk factors that did not meet their criteria or for which information was not available. The current list of factors and their definitions are shown in Table 1. Table 2 shows the increase in fracture probability associated with the various risk factors in a sample patient.

Table Table 1.. Risk Factors Considered in FRAX

Table Table 2.. Relative to No Risk Factors, Ratios of 10-yr Probability for Hip Fracture and Any One of Four Major Osteoporotic Fractures for a 62-yr-Old Woman With a BMI of 25 kg/m² and Femoral Neck T-Score of −1.8 (Derived From the FRAX Tool Online January 2009)

BMD can be included, but the model also works without it. The initial model required a T-score calculated using NHANES III normal values for white women, which was not necessarily the same as the output from commercial DXA equipment, especially for men. A “FRAX patch” was created (http://www.oregonosteoporosis.com/fraxpatch.htm) that converts femoral neck BMD into the appropriate T-score. In February 2009, FRAX was changed so that users could enter either correct T-scores or select the manufacturer of the densitometry equipment (GE-Lunar, Hologic, or Norland) and enter the femoral neck BMD in grams per square centimeter, bypassing the need for the FRAX patch. The data entry screen for U.S. whites as of March 1, 2009, is shown in Fig. 1.

Figure Figure 1. FRAX: WHO fracture risk assessment tool.

Given the same patient characteristics, fracture risk is known to vary widely by country.⁽⁸⁾ FRAX allows results to be calculated for countries where the incidences of both fracture and mortality are known. Currently, the model gives results for Austria, China, France, Germany, Italy, Japan, Spain, Sweden, Turkey, the United Kingdom, and the United States—and within the United States, by race/ethnicity (white, black, Hispanic, and Asian). Countries not currently included in FRAX are encouraged to submit country-specific data to the FRAX organizers or use the country in FRAX for which the epidemiology of osteoporosis most closely approximates their own. The model uses worldwide-derived risk ratios but country-specific fracture and mortality rates.

The output of FRAX is 10-yr probability for hip fracture and 10-yr probability for any one of four “major osteoporotic fractures” (four-fracture; including hip fracture but also proximal humerus, wrist, and clinical vertebral fracture).

Finding accurate rates for hip fracture is relatively easy, because virtually all are admitted to a hospital, and such data are nationally collected. In contrast, patients with humerus or wrist fractures are usually treated outside of hospital and therefore are undercounted. Capturing the rate of clinical vertebral fractures is even more daunting because definitions vary, and it is difficult to distinguish painful incident vertebral fractures from the more common radiologic vertebral deformities detected in patients with back pain. The best epidemiological data on the risk of any one of four major osteoporotic fractures comes from Malmö, Sweden.⁽⁹⁾ Most of the countries incorporated into FRAX have provided hip fracture data only, requiring the Sheffield WHO Group to estimate the four-fracture rate using Malmö's age-specific ratios of four-fracture to hip fractures. This was not the case for the United States, where Olmstead County, MN, incidence data were available for each of the four fracture types.⁽¹⁰⁾

FRAX U.S. must be considered a work in progress. The WHO-Sheffield Group clearly stated that the Web-based tool posted on the Sheffield website in February 2008 was a beta test version. Since that time, several programming corrections have been made. A major change was made in October 2008, when a correction for four-fractures in the United States was implemented, because the algorithm had double-counted hip fractures, resulting in overestimation of the four-fracture risk by ∼15% in women younger than 70 yr and ∼45% in women over age 70 yr. Currently, the National Osteoporosis Foundation has recommended further updating of the U.S. fracture incidence and mortality data to be used in FRAX. Hip fracture rates have fallen in the United States since 1989–1991,⁽¹¹⁾ the years that incidences were reported from Olmstead County. More recent U.S. hip fracture incidence rates are now available through the National Inpatient Survey (NIS), a dataset developed under the Agency for Health Care Research and Quality (AHRQ). Thus, modifications of U.S. FRAX can be expected in coming months. These changes will generally cause reductions in fracture estimates, especially among people <65 yr of age.

There are important limitations of FRAX, in large part because of lack of complete information in the WHO databases. A number of these are pointed out on the FRAX website, both scrolling down from the data entry screen and as “FAQs.” Some important risk factors for fractures are not included in the model. Examples include vitamin D deficiency; falls; physical activity; bone turnover markers; previous treatment for osteoporosis; medications such as antiepilepsy drugs, aromatase inhibitors, and androgen deprivation therapy; and others. In the FRAX calculator, saying “yes” to “secondary osteoporosis” does not change fracture risk if BMD is also entered; this is explained by BMD subsuming the risk in the model's prediction equations. The model's calculator does not permit combinations of secondary risk factors; for example, a patient with both hyperthyroidism and malabsorption receives the same relative risk as either one alone. The true prevalence of secondary causes of osteoporosis is not known, but among women likely to be evaluated by the FRAX tool, 40–65% may have one or more such conditions.^(12,13) FRAX does not consider low spine BMD (only femoral neck BMD is used) and does not consider the considerably higher risk of spine fracture among those with prevalent spinal fractures. Although fracture risk increases with the number and severity of fractures, the dose and duration of exposure to corticosteroid therapy, tobacco, and alcohol, the model assumes an average exposure and does not account for dose or duration. Thinness, a well-recognized risk for fracture, does not contribute to risk if BMD is known, because its association with fracture is largely through BMD; however, fracture risk in patients with known BMD paradoxically decreases as BMI decreases, apparently because of increased mortality associated with low BMI.

In FRAX, mortality is included in the Poisson regression that is used to calculate fracture probability. Although mortality is important in calculating population risks and societal impact, applying a “mortality offset” to the 10-yr risk of a patient seen in the office after BMD testing seems counterintuitive. It might be more reasonable, in this setting, to assume that the patient would live 10 yr.

There is, of course, uncertainty around almost all of the factors considered in FRAX, so there must be some confidence interval around these numbers, but none is provided. If FRAX is recalculated for an individual, the magnitude of difference that would indicate a true change in fracture probability is not known. Small differences in BMD or T-score within the measurement error result in large differences in fracture risk estimates. For example, a 70-yr-old woman with a “true” T-score of −2.0 could have a result of −1.8 (2% 10-yr hip fracture risk) or −2.2 (3% 10-yr hip fracture risk).

Perhaps the greatest current limitation of FRAX is that few use it. Busy clinicians cannot be expected to go online during patient visits and enter the required information. A recent experiment showed that providers, even those well versed in osteoporosis care, were unlikely to access a Web-based fracture risk assessment tool; only 1 in 20 providers who referred patients for bone densitometry testing responded to a mailed invitation to access a tool quite similar to FRAX.⁽¹⁴⁾ An answer to the accessibility problem lies in including FRAX in densitometer reports. Although densitometry companies are ready to provide software upgrades that allow FRAX integration into their machines' reports, it may be some time until the software containing the updated version of U.S. FRAX is reapproved by the FDA (densitometers are medical devices, and as such require FDA approval for both safety and the accuracy of their reports).

Even if FRAX is incorporated into BMD reporting software, it may not be applicable to all circumstances and may cause confusion in some. FRAX is not intended to be used in patients receiving pharmacologic treatment for osteoporosis. In the United States, the National Osteoporosis Foundation specifically recommends using FRAX to calculate risk for patients who have T-scores between −1.0 and −2.5 in the spine, femoral neck, or total hip (i.e., FRAX could be unwanted and potentially confusing in patients who are on treatment, those with hip or vertebral fractures, those with T-scores −2.5 or below [especially those who are low in the spine and not in the hip], and those with T-scores better than −1.0 [where treatment has not been shown to be effective]).

Applications of FRAX include providing guidance to individual patients, selecting subjects for clinical trials of osteoporosis therapy, and health economic projections, including establishing intervention thresholds based on local conditions. In the United States, new recommendations for intervention have been provided by the National Osteoporosis Foundation.⁽¹⁵⁾

Appropriate use of FRAX for providing guidance to individual patients can be considered in three steps: first, presenting the data in ways that allow the patient and healthcare provider to achieve an adequate and comfortable level of health numeracy (i.e., “having a good grasp of the numbers”); second, clarifying the potential benefits and risks of treatment to allow the patient to consider whether treatment is worth it; third, providing a fair answer to the usual question that follows the first two, “doctor, what do you think?”

Because patients understand risk differently, it may be helpful to frame risk estimates in different ways and encourage patients to get a feel for these risk numbers and how they apply to their medical care.⁽¹⁶⁾ FRAX provides two probabilities as 10-yr percentages: risk of hip fracture and risk of any one of four fractures (hip, wrist, humerus, or painful vertebral fracture)—all based on the assumption of no treatment. Whereas providers are comfortable interpreting percentages, the patient might better understand odds (e.g., 1 in 5 instead of 20%). It is important to understand that the mix of fractures in the four-fracture (major) number changes with age; for the woman in her 50s and 60s, these are mostly wrist fractures; in contrast, hip and vertebral fractures predominate among women in their 70s and 80s. This step concludes by asking “how does living with this risk feel?”, thus encouraging patients to judge the importance of this risk in their lives and giving a clue as to their optimism or pessimism.

The second step, considering treatment effects, also involves numeracy and framing. Risk reduction must always be expressed in absolute terms, again in at least two ways and again asking for the patient's considered judgments. Assuming a 35% relative fracture risk reduction with bisphosphonate treatment for a patient whose 10-yr risk is 20%, we might frame the estimate as follows: “by taking drug treatment for the next 5–10 yr, you can expect to reduce your risk from 20% to 13%, or from 1 in 5 to 1 in 8; now, does that seem worthwhile to you?” Promoting patient autonomy by sharing the decision process is key to making any decision fair and ethical.

The final step is answering the almost inevitable question posed after this kind of counseling, “what is your opinion, doctor?” Whereas the provider may be aware of details of cost-effectiveness studies and numbers needed to treat, these should not be an explicit part of the counseling process. However, some general statements can be provided based on national guidelines. For example, “The NOF considers a four-fracture risk of >20% to be high enough to warrant treatment;” or “Osteoporosis Canada considers fracture risk >20% to be high, risk of 10–20% to be moderate, and <10% to be low.” Leaving room for patient preferences that differ from the provider's judgments is also important, whether cost-effectiveness thresholds are crossed. This is also a good time to review the individual risk factors that have contributed to the patient's risk estimate and, if available, to provide comparison with peer risk rates. Finally, giving permission to delay the treatment decision promotes patient autonomy; we should counsel, on one hand, that the current risk can be expected to double in the next 8 yr, but, on the other hand, we can expect that progress in medical science will likely provide additional information on fracture risk and new options for treatment in the future.

FRAX can also be used dynamically. For example, in counseling a woman age 62, free of osteoporosis risk factors (BMI = 25 kg/m², T-score = −1.8), she can be reassured to know that her 10-yr risk is 1.4% for hip fracture and 13% for four-fracture (based on current FRAX calculations). However, 10 yr in the future, with bone loss of 0.5%/yr (age = 72 yr, BMI = 25 kg/m², and T-score = −2.3), her 10-yr risk would be 4.4% for hip fracture and 21% for 4-fracture. Should she acquire one of the key osteoporosis risk factors in this interval, her risk could increase by as much as 90%, depending on the risk factor. For a patient with intermittent use of systemic glucocorticoids, risk might be calculated with and without the “glucocorticoid” risk, with that patient's risk being somewhere in between. Likewise, for a woman whose mother had severe disabling spine fractures but no hip fracture, risk might be calculated with and without “parent fractured hip,” with the patient's risk being somewhere in between.

The development of the WHO FRAX calculator represents a major achievement. Based on large datasets and well validated in test populations, it currently is the gold standard among fracture models. Whereas more parsimonious models may seem to work as well in some situations (e.g., for elderly women, when BMD is known, for certain fracture types),⁽¹⁷⁾ only FRAX, derived from sufficiently extensive databases, can provide validated results for men and women, over a large age range, with or without BMD results. It is also the one most likely to find wide acceptance because it has been selected to be the report used by densitometer manufacturers. However, it has limitations, so clinical judgment is still important. Physicians need to review clinically relevant risk factors such as medical conditions or therapies that are associated with increased fracture risk, or a low spine BMD, which are not accounted for in this tool. How an absolute fracture risk is presented to patients is critical for optimal patient care. Acceptance and use of the WHO FRAX calculator worldwide holds the promise that more men and women at risk for osteoporotic fractures will be identified and treated, but implementing the tool and gaining widespread use continues to be a difficult and drawn-out process.