Atypical femur fractures are a growing clinical concern. Unfortunately, studies with access to a substantial number of femur X-rays from fracture patients have been limited by short-term drug history data. Although studies with long-term drug-dispensing information exist, they have had to rely on diagnosis codes, so atypical femur fractures could not be distinguished from other subtrochanteric or diaphyseal femur fractures.1, 2 The former type of study may show correct incidence rates yet provide odds ratios (OR) that are too high because of the undercapture of past drug exposure, whereas the latter type of study will provide incidence rates that are too high and odds ratios that are too low because they include fractures that may not be atypical.
Recent work using data from Kaiser Permanente Northwest and conducted in collaboration with Merck, published in this issue of JBMR,3 provides new information that helps fill this epidemiological gap. Although not a case-control study in a strict sense, this study provides interesting insights that add to those from the Swedish national register data published last year in the New England Journal of Medicine by Schilcher and colleagues.4
Briefly, the Swedish Schilcher study suggested a rapid, reversible, and above all very pronounced increase in atypical femur fractures among bisphosphonate (BP) users. Thus, the OR for these fractures was 1.3 (95% confidence interval [CI] 1.1–1.6) per 100 prescribed daily doses, corresponding to an almost 10-fold increase in risk within the first 2 years of treatment and a 120-fold increase in the first 5 years. If one compares this risk with the rate of hip fractures of 1510 per 100,000 BP users in the study, it seems that harm could outweigh benefit at the hip after approximately 7 years, but the confidence interval becomes very wide on extrapolation with the OR at 5 years ranging between 7 and 5300.
At first glance, the two studies have many features in common. The US study3 reviewed 864 X-rays from patients with fractures of the femoral shaft or subtrochanteric femur and found 75 atypical fractures, whereas the Swedish study reviewed 1234 X-rays and found 59 atypical fractures.4 The study populations differed slightly in demographics; the Swedish data set was for women aged 55+ years, whereas the US data set included men aged 65+ years and women aged 50+ years. The Swedish study used a national data set that covered a background population almost 20 times the size of the Kaiser Permanente Northwest HMO but was constrained by having access to less than 3 years of medication history. The lack of long-term exposure data when a long-term outcome is examined is troublesome, not the least when BP users often have substantial gaps in their treatment so that what would appear to be a new user with limited BP exposure could in fact be a patient who had used BPs or glucocorticoids for many years but done so at a time when prescriptions were not yet captured. A current BP user is far more likely to have used BP in the past than is a control subject drawn from the background population, and this could inflate risk estimates.
The good news is that the study by Feldstein and colleagues had access to 5+ years of medication history for more than 85% of fracture cases, and study power was boosted by including several years of fracture outcomes (January 1996 to June 2009).
Two findings from the Feldstein study are particularly interesting and may help advance our understanding of these fractures.
First, we may have to subdivide atypical femur fractures into those that display ASBMR major features5 only and into those that also satisfy one or more minor criteria. Only the latter subtype of fractures—22 fractures in the radiology review—appeared to show an increasing time trend and an association with BP use.3 This not only affects the incidence rate and the strength of the association but also potentially our understanding of the biomechanics and pathophysiology. The question of the significance of cortical thickening and how the mode of action of bisphosphonates can be reconciled with cortical expansion has attracted particular attention.5, 6
Second, even in the subgroup of patients who had fractures that satisfied one or more minor criteria, BP use was far from obligatory. The odds ratio of ever having used a BP in patients who had an atypical rather than a typical femoral shaft fracture was a modest 2.11 (95% CI 0.99–4.49). Where 78% of Swedish patients with atypical femur fractures had a history of BP use,4 the US study found this to be the case for 62% of patients with atypical fractures that also met ASBMR minor criteria but only 21% of those with atypical fractures that only fulfilled ASBMR major criteria.
Radiographically, these criteria consist of either a localized periosteal reaction on the lateral cortex (“beaking”) or a generalized increase in cortical thickness of the diaphysis.5 Bilaterality and the clinical features prodromal pain and delayed healing are also included among the minor ASBMR criteria. In the Feldstein study, the hallmark of bisphosphonate exposure in femoral shaft fractures was really the presence of a cortical reaction—at least patients who had short oblique fractures without a cortical reaction were often not BP users. Cortical thickening in atypical femur fractures is incompletely understood. Certainly, cortical thickening does not appear to be common in long-term bisphosphonate users, at least not to an extent that can be determined by measurements performed on dual-energy X-ray absorptiometry (DXA) scans.6 Also, in the short term, alendronate appears to have a very small if any effect on femoral cortical bone mass or density as assessed by quantitative computed tomography (QCT).7 It seems likely that cortical thickening would have to be a baseline patient characteristic that is unrelated to bisphosphonates but serves as a marker for propensity to suffer atypical femur fractures, a propensity that bisphosphonates could exacerbate in susceptible individuals, possibly through their effects on bone turnover. Or it may be that cortical thickening could be acquired as a reaction to a stress fracture brought about by the accumulation of hypermineralized, less tough older bone because of reduced bone remodeling. After all, atypical femur fractures are localized to the area of the lateral femur that surrounds the point carrying the highest tensile strain, so might cortical thickening represent an attempt at healing damage occurring as the result of osteon debonding?8 The Kaiser Permanente Northwest findings justify new interest in the cortical reaction seen in these fractures.
Atypical femur fractures remain rare. In the Kaiser Permanente Northwest HMO, the rate of atypical femur fractures with major and minor criteria could be shown to have increased from no events in the years 1996 to 1999 to around 5 per 100,000 patient-years in 2009. During the same period of time, classical hip fractures declined from 400 to 300 per 100,000 patient-years. Clearly, this decline may be owing in part to factors other than osteoporosis medications, but the numbers suggest that even if BP use leads to atypical femur fractures, the overall effect of the management strategy for hip fractures—which includes that use of bisphosphonates in patients at increased risk of fracture—has been associated with a decreasing societal fracture burden. Accordingly, the rate of atypical femur fractures in Sweden was 0.9 per 100,000 patient-years in nonbisphosphonate users and 55 per 100,000 in BP users. Although the latter is higher than in the Feldstein study, it is still 25 times lower than the rate of hip fractures found in Swedish BP users, highlighting that BP are targeted to patients at increased risk of osteoporotic fractures. The Kaiser Permanente Northwest atypical femur fracture rate is intermediate between that found in Swedish BP users and Swedish nonusers, which is what should be expected for a population where a subset of subjects receive BPs.
The low number of atypical femur fractures in these large study populations is reassuring, but it also makes it more difficult to draw firm conclusions about the relevance of subtypes of atypical femur fractures. The time-trend curve for atypical fractures in the Feldstein study may suggest that atypical femur fractures with minor criteria have gone from nonexistence in 1999 to account for the majority of atypical femur fractures in 2009. But this time trend is driven by 22 such fractures, however, and will need to be confirmed in other populations. Taken together, the Swedish and Kaiser Permanente Northwest studies confirm that atypical femur fractures are much rarer than the osteoporotic fractures that BP treatment prevents.
Both studies show that BP use is a risk factor for atypical femur fractures, although most BP users will not experience atypical fractures and many patients with atypical fractures have not used BPs. We need more long-term drug exposure data linked to radiographs to be able to make inferences about the optimum duration of antiresorptive treatment; it may be different for different bisphosphonates and different individuals. Also, more research is needed to clarify if these fractures are entirely unpredictable or if we can perhaps avoid them by developing a different treatment strategy for patients who are at risk of this type of fracture. The current article invites a second look at cortical thickening that may potentially unravel the mechanism.