Although there is strong evidence that bisphosphonates prevent certain types of osteoporotic fractures, there are concerns that these medications may be associated with rare atypical femoral fractures (AFF). Recent published studies examining this potential association are conflicting regarding the existence and strength of this association. We conducted a systematic review and meta-analysis of published studies examining the association of bisphosphonates with subtrochanteric, femoral shaft, and AFF. The random-effects model was used to calculate the pooled estimates of adjusted risk ratios (RR). Subgroup analysis was performed by study design, for studies that used validated outcome definitions for AFF, and for studies reporting on duration of bisphosphonate use. Eleven studies were included in the meta-analysis: five case-control and six cohort studies. Bisphosphonate exposure was associated with an increased risk of subtrochanteric, femoral shaft, and AFF, with adjusted RR of 1.70 (95% confidence interval [CI], 1.22–2.37). Subgroup analysis of studies using the American Society for Bone and Mineral Research criteria to define AFF suggests a higher risk of AFF, with bisphosphonate use with RR of 11.78 (95% CI, 0.39–359.69) as compared to studies using mainly diagnosis codes (RR, 1.62; 95% CI, 1.18–2.22), although there is a wide confidence interval and severe heterogeneity (I2 = 96.15%) in this subgroup analysis. Subgroup analysis of studies examining at least 5 years of bisphosphonate use showed adjusted RR of 1.62 (95% CI, 1.29–2.04). This meta-analysis suggests there is an increased risk of subtrochanteric, femoral shaft, and AFF among bisphosphonate users. Further research examining the risk of AFF with long-term use of bisphosphonates is indicated as there was limited data in this subgroup. The public health implication of this observed increase in AFF risk is not clear.
It is estimated that by 2020 approximately one-half of the United States population greater than age 50 years will have osteoporosis or low bone mass, and the prevalence is expected to grow with the aging population. Medications to prevent and treat osteoporosis, such as bisphosphonates, are therefore increasingly used. Bisphosphonates prevent common osteoporotic fractures in patients with osteoporosis, particularly in the hip and vertebrae. Bisphosphonates remain among the most commonly used medications for prevention of fracture in osteoporotic patients, with a recent study estimating that over 4 million women in the United States over 45 years of age were taking bisphosphonates in 2008.
Although bisphosphonates prevent typical osteoporotic fractures, prior studies have raised concern that these agents may actually cause atypical femoral fractures (AFF). Although the mechanism of this is unknown, studies suggest that bisphosphonates may negatively affect bone remodeling and lead to increased microdamage.[4, 5] Unlike typical hip fractures found in the femoral neck, trochanteric, and intertrochanteric regions of the femur, AFF occur in the subtrochanteric and femoral shaft regions. Fractures in the subtrochanteric or femoral shaft regions are rare, with some studies estimating 3 fractures per 10,000 person-years in certain populations. Hip fractures are more common, with an estimated incidence of 103 fractures per 10,000 person-years. However, although the incidence of hip fractures is decreasing in the United States, the incidence of subtrochanteric and femoral shaft fractures is stable or possibly increasing.[7, 8]
One potential reason for this rise in incidence may be the growing use of bisphosphonates. Several epidemiologic studies demonstrated an increased risk of subtrochanteric and femoral shaft fractures with use of bisphosphonates, particularly with long-term use of these medications.[9-11] However, other observational studies and reanalysis of randomized controlled trials of bisphosphonates found no increased risk of subtrochanteric and femoral fractures or AFF, although this may be a result of limited statistical power.[12-14]
This study aimed to systematically review all published studies that examined the potential association of AFF with bisphosphonate use. We did not impose age, gender, or geographic limitation for included studies. We summarized this published data in a meta-analysis, to determine the risk of subtrochanteric, femoral shaft, and AFF with use of bisphosphonates, particularly with long-term use.
A full study protocol is available online, titled “Study protocol for Bisphosphonates and Risk of Atypical Femur Fracture: A Meta-analysis” (see Supporting Information). We used the Preferred Reporting Items for Systematic Reviews and Meta-analyses statement to guide our methods for this study.
We searched MEDLINE and EMBASE databases (January 1, 1990 to October 19, 2012) for studies examining the association of bisphosphonate use and AFF, as well as subtrochanteric and femoral shaft fractures. These dates were chosen because the initial reports of the association of AFF with bisphosphonates occurred after the 1990 s. We performed the search using the following keywords and Medical Subject Headings: [“diphosphonates” (this term includes alendronate, clodronic acid, and etidronic acid) OR “bisphosphonate” OR “ibandronic acid” OR “pamidronate” OR “zoledronic acid” OR “risedronic acid”] AND [“femoral fractures” OR “femur fracture” OR “hip fractures” OR “diaphyseal AND femoral fracture” OR “atypical AND femoral fractures”] AND [“subtrochanteric” OR “diaphyseal” OR “midshaft” or “atypical”]. We included only studies published in English. We also searched available abstracts from the American College of Rheumatology (ACR), American Society of Bone and Mineral Research (ASBMR), and American Academy of Orthopaedic Surgeons (AAOS) annual meetings from 2006 to 2011. If potential studies were found in abstract form, we then searched MEDLINE and EMBASE to determine whether the abstracts were published in full manuscript form. Only fully published data were used for the meta-analysis.
Study eligibility and selection
We included prospective or retrospective cohort studies, case-control studies, and secondary analysis of randomized controlled trials in our review. Eligible studies compared bisphosphonate use to: (1) no bisphosphonate use; (2) placebo; or (3) a nonbisphosphonate osteoporosis medication. We used subtrochanteric, femoral shaft, or AFF as the reported outcomes required for inclusion into the study. We allowed for heterogeneity in outcomes because the types of reports available for each potential study were variable; certain studies only had access to International Classification of Diseases (ICD) codes to identify fracture, whereas other studies identified fractures based on imaging and could identify atypical features. Case reports, case series, studies on animal models, review articles, letters, or editorials were excluded from the study. Studies examining bisphosphonate use in patients with malignancies or metabolic syndromes such as Paget's disease were also excluded. Studies reporting only crude estimates of association between bisphosphonate use and subtrochanteric fracture, femoral shaft fracture, or AFF were not included in the study, because we believed important potential confounders such as age and sex should be controlled. Two authors (LG and SCK) independently reviewed abstracts and/or full texts to determine study eligibility using a study screening form the authors created to determine eligibility for inclusion in our analyses (see Supporting Information, eFig. 1). If there was disagreement on eligibility between the two authors this was resolved by author consensus.
Data abstraction and quality assessment
Independent data collection forms were used by two authors (LG and SCK) to extract data from each eligible study. The Newcastle-Ottawa Quality Assessment Scale was then used independently by the two authors (LG and SCK) to help determine quality of the studies included in the meta-analysis. This commonly used quality assessment scale gives points to nonrandomized studies on quality of definition of selection groups (0–4 points), ascertainment of exposures and outcomes (0–3 points), as well as comparability of the two populations to be studied (0–2 points). The Newcastle-Ottawa Quality Assessment scale determines that a larger total number of points attributed to a study correlates with increased quality of the study. The scale can be modified a priori in order to be tailored to the proposed meta-analysis. We determined a priori that studies radiologically confirming AFF and reporting on duration of bisphosphonate treatment received increased number of points in the ascertainment of exposures and outcomes (see Supporting Information, eTable 1).
The DerSimonian and Laird random-effects model was used to calculate the pooled estimates of adjusted risk ratios (RR); we assumed that either odds ratio or hazard ratio was equivalent to RR because the probability of the outcome, AFF, was expected to be very low.[17, 18] This statistical technique weights individual studies by sample size and variance (both within-study and between-study variance) and yields a pooled point estimate of RR and a 95% confidence interval (CI). The DerSimonian and Laird technique was considered an appropriate pooling technique because of the relative heterogeneity of the source population in each study. Subgroup analysis was performed by study design (case-control or cohort design), for studies that used validated outcome definitions, and for studies reporting on duration of bisphosphonate use. The I2 statistic was calculated to determine the amount of variability in the study results that is attributed to the heterogeneity of included studies. To assess the potential for publication bias, we performed Begg's test and the Egger's test and constructed funnel plots to visualize possible asymmetry. Due to the limited number of studies, meta-regression was not performed. All statistical analyses were performed in Comprehensive Meta Analysis version 2.2 (www.meta-analysis.com) and Stata 10 (Stata Corp, College Station, TX, USA). We followed the Meta-analyses of Observational Studies in Epidemiology guidelines in the report of this meta-analysis.
Study characteristics and quality
The electronic literature search resulted in 478 potential studies examining bisphosphonates and subtrochanteric, femoral shaft, and atypical femoral fracture. Further search of abstracts from the ACR, ASBMR, and AAOS meetings from 2006 to 2012 yielded an additional 24 abstracts for review. After review of study titles and abstracts from these studies, the exclusion criteria listed in the Methods section were applied. A total of 29 studies remained and underwent more detailed examination to determine whether they met study criteria. Duplicate studies and studies for which fully published study data was not obtainable were excluded. Studies that met all requirements except used only crude estimates of association were excluded from the analysis,[22, 23] as were studies that did not present data that allowed for calculation of adjusted relative risk ratio. Eleven studies, composed of five case-control and six cohort studies, including one study presenting results from three randomized controlled trials, were used in the final analysis. Figure 1 illustrates our study selection process. Four studies were included in analyses examining long-term bisphosphonate use, which was defined for this meta-analysis as greater than 5 years.
Descriptions of the included studies are displayed in Tables 1 and 2. The study populations were from the United States, Canada, Europe, and Taiwan. The number of study participants ranged from 477 to 1,521,131 participants. Many studies only included women.[9, 10, 12, 14, 25] Mean ages in the study populations ranged from 68 to 84 years. Most studies examined use of oral bisphosphonates; however, three studies also described intravenous use of zoledronic acid.[11, 12, 26] Most studies compared fractures in users of bisphosphonates to nonusers of bisphosphonates, although three studies also compared fracture rates in users of non-bisphosphonate osteoporosis medications.[11, 13, 14] All studies used medical record review and/or ICD codes to find the main outcome of interest, subtrochanteric or femoral shaft fractures. Several studies confirmed features of atypical fractures with radiologic review[9, 10, 12, 25-27] and two of these studies used the ASBMR criteria for definition of AFF.[26, 27] Several studies also examined long-term use of bisphosphonates in a subset of patients.[9, 10, 13, 14, 26, 28, 29] Long-term use was defined differently in each study, varying from greater than 1 year of bisphosphonate use to greater than 6 years of bisphosphonate use. Only studies that presented adjusted estimates of association for long-term use were included in the subgroup analysis of long-term bisphosphonate use, which for this study was defined as bisphosphonate use of 5 years or greater.[9, 13, 28, 29] Based on the Newcastle-Ottawa Quality Assessment scale, all the included studies were of moderate to high quality.
|Study, year, country||Subjects||Type of bisphosphonate||Outcome definition and number of events||Variables matched and/or controlled||Quality assessment scoresa|
|Abrahamsen and colleagues, 2009, Denmarkb||15,561 women with history of non-hip fracture in a national patient registry (1997–2005)||Alendronate||ST/FS fractures, 76||Matched on: sex, birth year, baseline fracture location; adjusted for: age, sex, and comorbidities||s3/c2/o2|
|Abrahamsen and colleagues, 2010, Denmarkb||197,835 patients without previous hip fracture in a national patient registry (1996–2005)||Alendronate||ST/FS fractures, 1049||Matched on: age, sex, index year; adjusted for: sex, Charlson index, hormone therapy, steroid use, fracture history, and number of co-medications||s4/c2/o1|
|Black and colleagues, 2010, USA||14,195 women enrolled in three randomized controlled trials examining bisphosphonate use||Alendronate; zoledronic acid||ST/FS fractures, 12||Placebo-controlled||s3/c2/o2|
|Hsiao and colleagues, 2011, Taiwan||11,278 women with osteoporosis and prior vertebral/hip fracture in a national insurance database (2001–2007)||Alendronate||ST/FS fractures, 61||Adjusted for: age, fracture history, comorbidities, and co-medications||s3/c2/o1|
|Kim and colleagues, 2011, USAb||33,815 new-users of bisphosphonates, raloxifene, or calcitonin enrolled in Medicare (1996–2006)||Alendronate; etidronate; risedronate||ST/FS fractures, 104||Matched on: propensity score (encompassing age, race, sex, health care utilization, comorbidities, and co-medications)||s3/c2/o1|
|Vestergaard and colleagues, 2011, Denmark||414,245 patients from a national patient registry (1996–2006)||Alendronate; etidronate; clodronate; ibandronate; pamidronate; risedronate; zoledronic acid||ST/FS fractures, 4179||Matched on: age, sex; adjusted for: steroid use, hormone therapy, alcoholism, fracture history||s3/c2/o2|
|Study, year, country||Subjects||Type of bisphosphonate||Case definition and number of events||Control definition and number of events||Variables matched and/or controlled||Quality assessment scoresa|
|Lenart and colleagues, 2009, USA||Postmenopausal women at one institution (2000–2007)||Any||ST/FS fractures, 41||IT/FN fractures, 82||Matched on: age, race, body mass index||s2/c2/e3|
|Park-Wyllie and colleagues, 2011, Canadaa||Women using bisphosphonates in a prescription database (2002–2008)||Alendronate; risedronate; etidronate||ST/FS fractures, 716||No ST/FS fractures, 3580||Matched on: age, cohort entry period; adjusted for: SES, co-medications, comorbidities, resource utilization, BMD||s2/c2/e3|
|Schilcher and colleagues, 2011, Sweden||Women with age greater than 55 in a national patient registry (2008)||Alendronate; etidronate; ibandronate; risedronate||Atypical ST/FS fractures, 59||Typical ST/FS fractures, 263||Matched on: none; adjusted for: age, steroid use, Charlson index, hormone therapy, antiepileptic use, antidepressant use, PPI use||s3/c2/e3|
|Feldstein and colleagues, 2012, USAb||Patients in a managed care database (1996–2009)||Any oral bisphosphonate||Atypical FS fractures, 75||Typical FS fractures, 122||Matched on: none; adjusted for: age, sex, steroid use, number of co-medications||s4/c2/e3|
|Meier and colleagues, 2012, Switzerlandb||Patients admitted with ST or FS fractures at one institution (1999–2010)||Alendronate; risedronate; pamidronate; ibandronate; etidronate; zoledronic acid||AFF, 39||Typical ST/FS fractures, 438||Matched on: none; adjusted for: age, sex, vitamin D, steroid, and PPI use||s2/c2/e3|
Bisphosphonates and atypical femur fracture
The overall pooled estimate of adjusted RR for AFF associated with bisphosphonates using data from the five case-control and six cohort studies was 1.70 (95% CI, 1.22–2.37) (Fig. 2). A large amount of heterogeneity was noted (I2 = 89.19%, p < 0.05). Exclusion of individual studies that appeared to have much higher RR as compared to the rest of the included studies [10, 25, 26] did not significantly change the overall pooled adjusted RR. The pooled adjusted RR based on five case-control studies[9, 10, 25-27] was 11.12 (95% CI, 2.68–46.18) with severe heterogeneity (I2 = 91.13%, p < 0.05) and the pooled adjusted RR using only the cohort studies[11-14, 28, 29] was 1.52 (95% CI, 1.08–2.15) with moderate heterogeneity (I2 = 68.91%, p < 0.05).
Subgroup analysis was then performed stratifying by outcome definition. The outcome was defined by either (1) no X-ray confirmation; (2) X-ray confirmation; or (3) ASBMR-pattern confirmation (Fig. 3). The pooled adjusted RR based on the two studies that used ASBMR-defined AFF[25, 26] was 11.78, with a wide 95% CI (0.39–359.69) and severe heterogeneity (I2 = 96.15%, p < 0.05). The pooled adjusted RR based on the studies that used X-ray confirmation of fractures (not using ASBMR-defined criteria) as part of their definition for AFF[10, 24] was 28.16 (95% CI, 13.30–59.59, I2 = 0% with p = 0.40). Last, the pooled adjusted RR based on the remaining studies that used primarily ICD codes to define subtrochanteric and femoral shaft fractures[9, 11-14, 27, 28] was lower, at 1.62 (95% CI, 1.18–2.22; I2 = 65.92%, p < 0.05).
Another subgroup analysis was performed examining long-term use of bisphosphonates (Fig. 4). When examining the subset of studies reporting 5 years or greater of bisphosphonate use,[9, 13, 28, 29] the pooled adjusted RR was 1.62 (95% CI, 1.29–2.04; I2 = 0%, p = 0.575). None of these studies examining long-term use used the ASBMR criteria for defining AFF.
Publication bias assessment
There was no statistical evidence of publication bias among the included studies by using Egger's (p = 0.4) and Begg's (p = 0.5) tests. However, the funnel plot shows minor asymmetry suggesting a small potential publication bias (Supporting Information, eFig. 2).
Considering the widespread use of bisphosphonates in the United States, even rare events that could be associated with these medications raise concern. Atypical femur fractures are one such rare event that is associated with bisphosphonates in several studies. This meta-analysis summarized the available data in published literature exploring a possible association of bisphosphonates and risk of subtrochanteric and femoral shaft fractures and AFF, and found an increased risk of these fractures with use of bisphosphonates.
We found that this increased risk persists for long-term users of bisphosphonates in our meta-analysis. However, these data should be interpreted with caution, as not only were there a small number of studies included in this subgroup analysis, but it is possible that the risk we found was underestimated, considering that the studies included in the long-term analysis did not radiologically confirm atypical fractures.
In our analyses, we noted that three of the case-control studies appeared to have higher RR as compared to the rest of the studies. One potential explanation for this observation is the selection of the controls, which in all these studies were “typical” fractures. In choosing controls that had an outcome for which bisphosphonates are known to have a beneficial effect, these studies could exaggerate any potential harmful association of “atypical” fractures attributable to bisphosphonates (see Supporting Information, eFig. 3). In addition, only one of these studies used the ASBMR criteria for AFF; therefore, the other two studies may overestimate the number of atypical fractures seen in their cohorts. Aspenberg and colleagues recently presented an abstract reanalyzing the Schilcher and colleagues published study data using ASBMR criteria for defining AFF. This study found the RR of AFF with bisphosphonate decreased by more than one-half using the ASBMR criteria for AFF as compared to their prior published RR. Similarly, the included article by Feldstein and colleagues, which in general used ASBMR criteria to define AFF, also included a small number cases which did not have the classic “transverse” or “short oblique” fracture angle required for the ASBMR definition; this may be why the RR is lower than the other study in our meta-analysis which used ASBMR criteria for the outcome definition. Last, it is also possible that part of the discrepancy lies in the studies that used solely ICD codes to diagnosis subtrochanteric and femoral shaft fractures, which may include a significant portion of fractures that are not AFF and therefore show falsely low RR associated with use of bisphosphonates.
A main limitation in our study is the varying definition of AFF used in the included studies. A task force of the ASBMR suggests that in addition to the subtrochanteric or femoral shaft location, AFF should be defined by additional radiologic and histologic features, such as non-comminuted fractures that occur in a short oblique or transverse configuration with minimal or no trauma. Some studies suggest that only a subset, 17% to 29%, of subtrochanteric and femoral shaft fractures meet these criteria for AFF. However, due to inability to radiologically confirm several of these features, many epidemiologic studies have used the location of the fracture, subtrochanteric or femoral shaft, as a marker for AFF, while other studies were able to use the more precise criteria. Nevertheless, including the studies with the broader definition of AFF would likely bias our study towards the null, and despite this we have found a significant risk of AFF with use of bisphosphonates. When only examining the subset of studies that used ASBMR-criteria to confirm AFF we still found a significant association of AFF with bisphosphonates; however, caution should be used in the interpretation of this subanalysis considering the small amount of studies using ASBMR criteria, leading to a wide confidence interval and severe heterogeneity.
As is common with meta-analyses, we were also limited by the significant heterogeneity of the studies, with some I2 values greater than 90%. Potential causes of heterogeneity between the studies include variations in the study design and size, patient characteristics (age, gender, race, geographic location, underlying comorbidities, and use of other drugs), type of bisphosphonates, type of comparison group (nonusers or nonbisphosphonate osteoporosis medications), outcome definition or ascertainment methods, choice of control fracture, statistical methods, and the overall study quality. We used a random effects model to account for both within-study and between-study variance for our analysis. We also performed a subgroup analysis by study design, outcome definition, and long-term bisphosphonate use to further explore a potential source of heterogeneity.
All meta-analyses are vulnerable to publication bias, but we attempted to minimize this bias by searching two major electronic databases and abstracts for major scientific meetings related to the topic. We also further examined for publication bias using three different statistical tests. In order to minimize confounding and selection bias that are inherent to observational studies, we objectively assessed the quality of individual studies and all the included studies were of medium to high quality. In addition, we only included studies in the analysis if they reported adjusted relative risks to minimize potential confounding. However, despite these efforts, potential confounding may remain, particularly in some of the included cohort studies that compare bisphosphonate users to those receiving no treatment at all - these studies could be subject to confounding by indication even with full adjustment.
Despite these limitations, this meta-analysis adds to the current literature by providing an in-depth summary and analysis of recent data describing the association of bisphosphonates with subtrochanteric, femoral shaft, and AFF. However, while the relative risk of these atypical fractures appears increased among bisphosphonate users in our study, the public health consequences are likely minimal considering the rarity of AFF as compared to typical osteoporotic fractures, especially in the first several years of bisphosphonate treatment. A recent study by Dell and colleagues that did not meet prespecified criteria to be included for this meta-analysis, found the estimated incidence of AFF among patients older than 45 years was 1.78/100,000 person-years in those taking bisphosphonates for less than 2 years. Comparatively, in this same population typical hip fracture incidence was 463/100,000 person-years in those taking bisphosphonates for 0 to 1 years. However, duration of bisphosphonate use increased the incidence of AFF: the incidence of AFF/100,000 person-years ranged from 38.9 for those taking bisphosphonates for 6 to 8 years, to 107.5 in those taking bisphosphonates for greater than 10 years.
Therefore, whereas the benefit of bisphosphonates likely outweighs the risk of the rare AFF early on in treatment, this benefit is less clear for long-term users. The U.S. Food and Drug Administration recently reviewed the limited published randomized control trial data examining long-term use of bisphosphonates, and found no clear benefit of long-term use (defined as greater than 5 years) of bisphosphonates in prevention of typical osteoporotic fractures. Black and colleagues, however, suggest that there may be benefit for the subgroup of patients with persistent T-scores less than –2.5 at the femoral neck after 3 to 5 years of bisphosphonate therapy or those with vertebral fracture.
Considering the above findings, the results of this meta-analysis suggest increased caution may be indicated for bisphosphonate use beyond 5 years, particularly in the subgroup of patients with no vertebral fractures or T-scores greater than –2.5, where there is unclear benefit of typical osteoporotic fracture prevention with bisphosphonates. Further research examining long-term use of bisphosphonates is indicated in order to better define the continued effectiveness of bisphosphonates in osteoporotic fracture prevention and the risk of associated rare side effects.
This meta-analysis shows an increased risk of subtrochanteric fracture, femoral shaft fracture, and AFF with bisphosphonate use. However, these atypical femur fractures are rare events even in bisphosphonate users, and their risk is likely to be outweighed by bisphosphonate's reduction of typical osteoporotic fractures in most patients. Increased caution may be indicated in long-term users where the benefit of bisphosphonates in prevention of typical osteoporotic fractures is less clear; however there is a paucity of data about AFF risk among long-term bisphosphonate users.
Author LG is supported by the NIH grant T32AR055885-04. Author DHS is supported by the NIH grants K24 AR055989, P60 AR047782, R21 DE018750, and R01 AR056215, and has received research support from Abbott Immunology, Amgen, and Lilly. Author DHS serves in unpaid roles on two Pfizer sponsored trials. Author SCK is supported by the NIH grant K23 AR059677, and received research support from Takeda Pharmaceuticals North America and Pfizer. Author SCK received tuition support for the Pharmacoepidemiology Program at the Harvard School of Public Health funded by Pfizer and Asisa.
This work was supported by grants from the NIH (T32AR055885-04 to LG; K23 AR059677 to SCK; K24 AR055989, P60 AR047782, R21 DE018750, and R01 AR056215 to DHS); SCK has also received research support from Takeda Pharmaceuticals North America and Pfizer; DHS has received research support from Abbott Immunology, Amgen, and Lilly.
Authors' roles: Study design: DHS, SCK. Data collection: LG, SCK. Data analysis and interpretations: LG, DHS, SCK. Drafting manuscript: LG. Revising manuscript: all authors. Approval of final manuscript: all authors. LG and SCK take responsibility for the integrity of the data analysis.