Should There Be a Fracas Over FRAX and Other Fracture Prediction Tools?

Clinical practice guidelines regarding when to initiate osteoporosis treatment have evolved from the use of bone mineral density (BMD) thresholds to a more complex consideration of the patient’s 10-year absolute fracture risk.¹ There are sound reasons for this shift: most fractures occur in patients with BMD T scores above −2.5, and other clinical risk factors, especially age, affect fracture risk.¹ Assuming equal relative treatment efficacy, prescribing therapy for a patient with a higher 10-year fracture risk will result in a greater absolute fracture reduction and be more cost-effective for society.

The World Health Organization’s FRAX tool was developed to help clinicians calculate the 10-year fracture risk for individual patients. In the United Kingdom, fracture risk thresholds identify those who should be referred for BMD testing and those in whom treatment decisions can be made without using BMD testing.² In the United States, where universal screening of older women is recommended, National Osteoporosis Foundation guidelines suggest using the 10-year fracture risk to identify patients with osteopenia whose high fracture risk warrants treatment.

In this issue of Archives, Ensrud et al used the Study of Osteoporotic Fractures cohort to ask whether the FRAX model, with its 9 risk factors and femoral neck BMD, can predict fracture better than simpler models using, for example, only age and BMD. They found that the simple models performed as well as FRAX when they were evaluated using receiver operating characteristic curves or according to the proportion of patients in each quartile of risk who later experience a fracture. Similarly, age and history of fracture performed similarly to FRAX models without BMD. All of the clinical prediction models performed best for hip fractures (area under the curve, 0.74–0.75) but had only modest discriminative ability for major osteoporotic fracture and clinical fracture (area under the curve, 0.63–0.68). Only women were used in Ensrud and colleagues’ analysis, but other investigators have reported that fracture prediction models have less discriminative ability in men.³ Ensrud and coauthors’ analysis is a useful addition to the developing science of fracture prediction. We clinicians, however, believe that the fracas should not be about which fracture prediction model to use, but rather how the concept of the 10-year fracture risk should be used in clinical practice.

There is little argument that the 10-year fracture risk is invaluable in counseling individual patients, and, as Ensrud et al point out, different models sometimes give widely varying estimates of fracture risk. For example, an average 70-year-old white woman in the United States with a T score of −1.5 has a 10-year hip fracture risk of approximately 1.5% if, as in Ensrud and colleagues’ model, no other risk factors are considered. Adding a previous vertebral fracture and parental hip fracture history moves her risk estimate to 5%. While at a population level both FRAX and the simpler models will correctly classify a similar proportion of individuals who will and will not experience fracture, clinicians will probably prefer to use the more specific estimate from FRAX to counsel these individuals. Patients with T scores lower than −2.5 or prior low trauma fracture have been clearly shown to benefit from treatment, yet only 21.8% of older women with a prior fracture are tested or treated for osteoporosis.⁴ Perhaps a fracas is needed to convince clinicians and patients to use these proven therapies. Using validated fracture risk models to raise awareness is a welcome advance.

It is less clear how to use the 10-year fracture risk to select patients for BMD testing or treatment. Prospective studies demonstrating that this approach affects patient outcomes are currently lacking. Testing and treatment thresholds have been selected based on careful cost-effectiveness analyses,⁵ but they depend on the unproven assumption that available treatments are equally effective across all levels of BMD. Of particular debate is the National Osteoporosis Foundation guidelines’ use of 10-year fracture risk thresholds (3% risk of hip fracture or 20% risk of major fracture) to select patients with osteopenia for treatment. Several post hoc analyses of clinical trials have reported significant decreases in fracture risk for this group^6,7; however, these analyses are susceptible to selection bias, and not all have shown benefit.^8,9 Prospective data are conflicting. A randomized trial of clodronate in unselected older women was effective in reducing clinical fractures regardless of BMD or, for that matter, 10-year fracture risk,¹⁰ but a trial of risedronate in older women selected on the basis of fracture risk factors was effective only in those with osteoporotic T scores.¹¹ This issue has enormous clinical implications. It is estimated that 93% of US white women older than 75 years would be eligible for treatment using current National Osteoporosis Foundation guidelines,¹² a seemingly unrealistic figure given our low current treatment of the smaller population of patients with osteoporosis or prior fracture.

Because regulatory bodies are now requiring pharmaceutical companies to use the 10-year fracture risk in subject selection for osteoporosis trials, clarification of this fracas should occur within the next several years. It will then be possible to determine how well FRAX successfully reclassifies patients over an effective treatment threshold compared with the simple models proposed by Ensrud et al and others.³ Until then, we believe that clinicians should prioritize improving care in patients who are known to benefit from osteoporosis therapy and use fracture risk models to select other patients with careful clinical judgment.