Osteoporosis risk assessment

The management of osteoporosis has been transformed over the past quarter century. The advent of non-invasive methods for assessing bone mineral density (BMD), coupled with the emergence of agents that retard involutional bone loss and prevent fracture, have promoted a disorder that was thought to be an inevitable consequence of ageing to the ranks of those amenable to effective targeted intervention. The linked paper by Hippisley-Cox and Coupland (doi:10.1136/bmj.e3427) evaluates a new tool for assessing the risk of osteoporotic fracture.1 What role do such tools play in helping prevent osteoporosis and its consequences?

In 1994 the World Health Organization generated a clinical definition of osteoporosis based on a BMD T score of −2.5 or lower.2 Although this definition performs well at a population level, the multifactorial nature of fracture pathogenesis and the fact that many people with fractures have a T score above −2.5 meant that initial algorithms for targeting treatment (such as those developed by the Royal College of Physicians) indicated intervention for some patients at relatively low risk of fracture. A major step forward was the publication in 2008 of a WHO algorithm (FRAX),3 which estimated the 10 year probability of fracture on the basis of clinical risk factors and BMD. This risk calculator was developed using primary data from a worldwide collection of nine cohort studies (and validated in a further 11) with more than a million patient years of follow-up. Potential risk factors were systematically meta-analysed and the 10 year fracture risk modelled according to nine clinical variables, all of which contributed independently to fracture risk. The risk factor score was designed to accommodate a super added BMD value. The tool has been extensively validated and is now used worldwide as part of the investigation of clinical osteoporosis.

Hippisley-Cox and Coupland present a different approach to estimating 10 year fracture risk using data from UK general practice.1 Their original QFracture tool, published in 2009,4 incorporated several risk factors for which information is readily available in general practice. In response to the 2012 National Institute for Health and Clinical Excellence (NICE) draft guidance on risk assessment for osteoporosis,5 the group has updated the tool, adding further risk factors such as epilepsy or use of anticonvulsants, ethnic group, type 1 diabetes, and previous fracture, which, despite being a major predictor of incident fracture, had not been included in the initial algorithm. General practices within the QResearch database, which covers more than 13 million patients registered at more than 620 general practices, were randomly allocated to the derivation or validation cohort. Cox proportional hazards models were used in the derivation dataset to investigate the usefulness of 30 different clinical risk factors. The resulting prediction model was then tested in the validation dataset. Because the two datasets were randomly allocated from the same population (resulting in very similar baseline characteristics), the model proved, unsurprisingly, highly successful in the validation subset.

The approach of using an algorithm based on clinical risk factors derived from a UK general practice database is not new. Van Staa and colleagues performed a similar procedure in 2006,6 using the health improvement network database as the derivation cohort and the general practice research database for validation. This yielded a relatively simple algorithm, similar to that proposed by Hippisley-Cox and Coupland; the tool was presented as a proof of concept that such an approach might be useful in risk assessment, but its authors emphasised two caveats, which were important considerations in the subsequent development of the FRAX calculator.

Firstly, there was no capacity to combine BMD assessment with clinical risk factors obtained from data routinely collected from general practice; the ability of FRAX to integrate BMD assessment with the predictive capacity of BMD independent clinical risk factors represents a clear advantage. Secondly, the drugs currently approved for preventing and treating osteoporosis have been evaluated in patients with low BMD, with or without previous fracture; the effectiveness of these agents in reducing fracture risk in people selected on the basis of risk factors that act independently of bone strength is uncertain. Ultimately, doctors want to identify patients at increased risk of fracture in whom currently available treatments have demonstrable benefit. These two caveats provide strong reasons why, on current evidence, QFracture is unlikely to supersede FRAX as the risk prediction algorithm of choice in the management of osteoporosis. Other advantages of FRAX include the recent development of simple modifications to the resulting individualised 10 year probability of fracture to account for dose relations in risk factors such as use of glucocorticoids,7 and the capacity to account for the competing hazard of death, which enables fracture probability estimates to be corrected in patients with a life expectancy of less than 10 years.

Whatever approach is taken to risk assessment, the clinician is left with the decision of how to treat. This is something that QFracture does not deal with, and existing guidance from NICE has been complicated, subject to recent legal challenge, and difficult to put into practice.8 The National Osteoporosis Guideline Group has filled this void by providing intervention thresholds for the treatment of osteoporosis that are easily applied in primary care using age and individualised 10 year probability of fracture through an accessible web based system.9 The widespread availability of these complementary clinical decision aids, their international validation, and their biological coherence have already begun to transform the clinical management of osteoporosis in primary and secondary care.