Dr. Abrahamsen receives consultancy fees from Nycomed and Novartis, research grants from Roche, and speaker's fees from Servier, Eli Lilly, and MSD. Dr. Eiken receives speaker fees from Nycomed, Roche, and Servier. Dr. Eastell receives research funding or consultation honoraria from Amgen, AstraZeneca, Aventis, Eli Lilly, GlaxoSmithKline, Hologic, Interleukin Genetics, Kyphon, Lilly, Maxygen, Nastech Pharmaceuticals, Nestle Research Center, New Zealand Milk Limited, Novartis, Novo-Nordisk, Ono Pharma, Organon, Osteologix, Paraxel, Pfizer, Procter & Gamble, Roche Diagnostics, Sanofi-aventis, Servier, Shire, Transpharma Medical Limited, Unilever, and Unipath.
Published online on December 29, 2009
Alendronate (aln) is a potent bisphosphonate with a prolonged duration of action. Recent reports have found long-term aln use to be common in patients with subtrochanteric or proximal diaphyseal femur fracture, raising concerns that these fractures could be a consequence of excessive suppression of bone turnover. Two national observational register-based studies were performed: (1) cross-sectional study (N = 11,944) comparing age distribution, exposure, and trauma mechanisms between different types of proximal femur fractures and (2) matched cohort study in patients with prior nonhip fractures (N = 5187 + 10,374), testing the hypothesis that the increase in the risk of subsequent atypical femur fractures exceeded the increase in typical hip fractures. We also sought evidence of a dose-response relationship, where high adherence to or long-term use of aln led to more atypical femur fractures. We found that 7% of patients with atypical fractures were aln exposed, and the same was found for typical hip fractures. In the cohort study, the HR for subtrochanteric/diaphyseal fracture with aln was 1.46 (0.91–2.35, p = 0·12) compared with 1.45 (1.21–1.74, p < 0·001) for hip fracture after adjustment for comorbidity and co-medications. The risk was reduced by high adherence, and the ratio between hip and subtrochanteric/diaphyseal femur fractures was identical in aln-treated patients and the control cohort even in the limited number of patients who received long-term treatment. Subtrochanteric/diaphyseal femur fractures share the epidemiology and treatment response of classical hip fractures and are best classified as osteoporotic fractures.
Alendronate is a potent oral bisphosphonate with a prolonged duration of action. The pharmacokinetics of alendronate allows for a once weekly regimen and leads to continued maintenance of BMD in months to years after discontinuation.(1) Whereas alendronate has sufficient antiresorptive potential to initially blunt the response to teriparatide,(2) long-term extensions of the primary licensing trials was associated with no signs of increased susceptibility to fractures with alendronate therapy of up to 10 yr of duration.(3) The optimum duration of treatment remains to be firmly established. Recent case reports(4–7) have found alendronate use to be common in patients with subtrochanteric or proximal diaphyseal fracture of the femur. This is a serious concern because alendronate is one of the most widely used antiosteoporotic therapies because of both vertebral and peripheral antifracture efficacy(8) and lately to the availability of generic formulations. Kwek et al.(7) retrospectively reviewed 17 patients with low-energy subtrochanteric femur fractures who were on alendronate therapy with a duration of treatment ranging from 2 (three patients) to 8 yr (one patient). The study included no comparator groups such as long-term alendronate-exposed patients without incident fracture or patients with subtrochanteric fractures unexposed to alendronate, and it is unclear how large a fracture population cases were drawn from. A larger case series been published by Neviaser et al.(5) as a retrospective review of all 70 cases of low-energy femoral shaft fractures—consisting of 20 diaphyseal and 50 subtrochanteric femur fractures—admitted to their trauma center in a 5-yr period. Alendronate was used by 25 (35%), with a mean duration of 6·2 yr. Although it is not known if these fractures are the result of excessive suppression of bone turnover or if they are osteoporotic by nature, the challenge to the clinical community is how to avoid exposing patients to undue risk of long-term side effects while at the same time targeting treatments to high-risk patients to reduce their risk of fractures. For any adverse events attributable to antiosteopororotic medications, we have to know not only the time and dose relationship but also the absolute incidence rates, so that any such risks can be weighed against the consequences of osteoporotic fractures.
We conducted two national register-based analyses to address atypical femur fractures as a potential adverse effect of alendronate. First, we performed a cross-sectional study to compare age-specific fracture rates, exposure, and trauma mechanisms between the different types of proximal femur fractures. Second, we conducted a register-based matched cohort analysis to test the hypothesis that the increase in the risk of these “atypical” femur fractures in patients treated with alendronate exceeded the increase in their risk of typical osteoporotic femur fractures. We also specifically sought evidence of a dose-response relationship, in which patients with high adherence to or long-term use of alendronate experienced more subtrochanteric or diaphyseal fractures than patients with lower exposure.
MATERIALS AND METHODS
National Hospital Discharge Register
The National Hospital Discharge Register(9) maintains a record of all hospitalizations and hospital outpatient appointments in the country including data on the discharge diagnoses linked with civil registry numbers.
National Prescription Database
All prescriptions dispensed from Danish pharmacies are captured in the National Prescription Database. This database is maintained by the Danish Medicines Agency and contains civil registry numbers, WHO ATC codes, strength, number of tablets, and date of redemption. Repeatable prescriptions generate a record each time the medication is dispensed. We obtained information about all prescriptions redeemed in the full study period.
Subtrochanteric/diaphyseal femur and hip fracture rates in 2005 for persons ≥60 yr of age with fractures categorized as low energy (code EUS0: “fall to same level” with absence of motor vehicle code EUK1), high energy (other EUS codes and/or presence of motor vehicle code EUK1), or unknown (EUS and EUK codes not recorded). Mechanism codes were available for the 2005 dataset only.
Cohort study in patients with baseline fractures
This approach reproduces—within an observational setting—the patient selection process and intention to treat (ITT) principle known from clinical trials.(10) We made the decision not to include patients with hip fractures in the cohorts to avoid the risk of misclassifying admissions because of fracture complications as new incident fractures. The postfracture cohorts were set up to monitor the risk of a first proximal femoral fracture in a tertiary prevention setting. We retrieved information on all patients born 1945 and earlier who sustained hospital treated fractures, except hip fractures, in Denmark between January 1, 1997 and December 31, 2005. The dataset contained 160,565 first fracture patients. All patients who began alendronate treatment and stayed on therapy for at least 6 mo (N = 5,187) after index fracture were entered into the treated cohort (Fig. 1). For each alendronate-exposed patient, we assigned two randomly selected bisphosphonate-unexposed fracture controls fulfilling the following criteria: (1) same sex, (2) born in the same year, (3) fracture at same location (spine, shoulder, forearm, or “other”), (4) sustained index fracture before the date of matched case redeeming first prescription, and (5) alive on the date of matched case redeeming first prescription. After first randomizing the sequence of cases and of potential controls, a procedure of assignment and testing was automatically reiterated until matching had been achieved. ICD-10–modified Charlson comorbidity index was calculated from hospital contacts in the year preceding the first alendronate prescription. A supplementary analysis was conducted for etidronate, using the same procedure. We identified 3428 patients(mean age, 70.6 ± 9.0 [SE] yr; 91% women) who began etidronate and persisted for at least 6 mo and assigned 6856 matched, untreated fracture controls. The calendar year 1996 was used as a run-in year providing information on prior prescriptions and comorbidity for patients entering the cohorts in 1997.
Data on outcome
The endpoint was incident fracture of the hip (femoral neck, ICD-10 code S72.0; pertrochanteric femur, S72.1), subtrochanteric femur (S72.2), and the femoral diaphysis (S72.3). In both cohorts, only fractures occurring 6 mo or later after initiation of therapy (or matching) qualified as endpoints.
Data are shown as mean ± SD or median [range]. Demographics were compared by t-tests and χ2 analysis. We used Cox proportional hazards models incorporating time to fracture, death, or end of study (12/31/2005) to obtain crude and adjusted HRs. Medication possession ratio (MPR) was calculated by dividing the number of defined daily doses redeemed within the at-risk period with the length of this period. Adjustment was made for age, sex, and exposure to oral glucocorticoids. In the fully adjusted analysis, the number of co-medications and Charlson index were also incorporated to further control for comorbidity. A subanalysis for long-time adherers (>6 yr with MPR > 80%) was prespecified. If alendronate use causes subtrochanteric fractures, we expected to observe (1) higher event rates in patients highly adherent to treatment and (2) a shift of fracture events from typical osteoporotic hip fractures to atypical proximal femur fractures, because alendronate reduces the risk of osteoporotic hip fractures.(11) The first analysis used MPR as a time-dependent covariate in the Cox proportional hazards model in patients treated with alendronate and tested for effects of MPR on risk and the second analysis entailed comparing the distribution of incident fractures by type in the treated cohort with that observed in the untreated cohort. There were no missing values in the cohort analyses and no patients were lost to follow-up. Analyses were done using SPSS version 14.0.
The use of data from the hospital discharge register and the prescriptions database required specific permission from the National Board of Health, and data access itself was subject to approval from the National Bureau of Statistics pursuant to guidelines approved by the Data Protection Agency. The study was not a clinical trial. All required permissions were obtained. We did not have access to patient notes, X-rays, patient names, addresses, or civil registry numbers.
Fracture rates and trauma mechanisms (cross-sectional study 2005)
Subtrochanteric and diaphyseal femur fractures occurred with an age and sex distribution matching that of typical hip fractures (Table 1), and the ratio between high- and low-energy trauma was similar. Only 7% of patients with atypical fractures were alendronate users, and the same percentage was found among patients with hip fractures. In contrast, patients with subtrochanteric fractures were almost twice as likely to be recent or current oral glucocorticoid users.
Table Table 1.. Proximal Femur and Hip Fracture Incidence Rates, Mechanism Codes, and Recent Alendronate or Glucocorticoid Use for 2005 in Denmark for Persons ≥60 yr of Age
Alendronate use and risk of typical and nontypical femur fractures (cohort study)
Patients who began alendronate received a significantly larger number of baseline co-medications, had a greater median Charlson comorbidity score, and a significantly larger proportion had been exposed to oral glucocorticoids within the past year (19.2% versus 4.6%; Table 2). We identified 506 new hip fractures and 76 subtrochanteric or diaphyseal femur fractures (Fig. 2; Table 3). Alendronate use was associated with a significantly increased HR for diaphyseal/subtrochanteric fracture (HR = 1.64; 95% CI = 1.05–2.58) but also for hip fracture (HR = 1.50; 95% CI = 1.26–1.79). After adjustment for baseline comorbidity (number of co-medications, glucocorticoid use, and Charlson index), the HR for subtrochanteric/diaphyseal fracture was 1.46 (0.91–2.35, p = 0.12; Fig. 3) compared with an HR for hip fracture of 1.45 (1.21–1.74, p < 0.001). The association was not significantly modified by prior (p = 0.27) or current glucocorticoid (p = 0.23) use, and interaction tests were negative (p = 0.30). The distribution between typical and atypical hip fractures was identical in the alendronate cohort and the untreated cohort (14% atypical in alendronate cohort, 13% in control cohort; p = 0.70).
Table Table 2.. Baseline Demographics for the Matched Postfracture Cohorts
Relationship between refill compliance and fracture risk
The risk of both typical and atypical femur fractures was significantly associated with alendronate refill compliance (Table 4). Thus, in the 5187 alendronate-treated patients, higher MPR was strongly associated with significantly lower risk of fracture both at the hip (HR = 0.47; 95% CI = 0.34–0.65; p < 0.001) and at the atypical sites (HR = 0.28, 95% = CI 0.12–0.63; p < 0.01). This remained significant after adjustment for baseline comorbidity and co-medications.
Table Table 4.. Matched Cohorts Analysis in Alendronate-Exposed Patients (N = 5187): Relationship With Medication Possession Ratio
Long-term alendronate use and fracture risk
The subpopulation who remained (>6 yr) highly refill compliant long term (MPR > 80%) was limited to 178 patients (Table 5). Only 39 hip fractures and 5 atypical fractures (in alendronate-exposed patients and their 356 controls) were available for a limited subanalysis. Long-term alendronate use was associated with parallel, nonsignificant increases in the risk of hip fracture (HR = 1.24, 95% CI = 0.66–2.34; p = 0·52) and the risk of atypical femur fracture (HR = 1.37, 95% CI = 0.22–8.62; p = 0·74; Fig. 3). We observed no shift from typical to atypical femur fractures with prolonged alendronate use. The distribution of femoral fracture types was similar between the two cohorts, with 10.0% of fractures in the long-term alendronate cohort being diaphyseal or subtrochanteric compared with 12.5% in the control cohort.
Table Table 5.. Matched Cohorts Analysis in Fracture Patients With Long-Term Exposure to Alendronate (>6 yr, MPR > 80%, N = 178) and Their Untreated Fracture Controls (N = 356)
Etidronate use and fracture risk
The combined incidence rate for diaphyseal and subtrochanteric fractures was 3.40 per 1000 patient-years in fracture patients receiving etidronate compared with 2.54/1000 in untreated fracture patients, based on 49 fractures in the treated group and 73 fractures in the untreated group (HR = 1.13, 95% CI = 0.78–1.64; p = 0.51) adjusted for age, sex, number of different drugs (ATC codes) redeemed in past year, oral glucocorticoids use in past year (0/1), and Charlson comorbidity index. Hip fractures occurred at a rate of 18.89/1000 patient-years in etidronate users (272 incident fractures) and 12.92/1000 patient-years in untreated patients (372 incident fractures; adjusted HR = 1.32; 95% CI = 1.13–1.55; p < 0.01).
This study showed that subtrochanteric and diaphyseal fractures show an age, sex, and trauma mechanism pattern that is similar to that of classical osteoporotic hip fractures. Patients with these atypical femur fractures were no more likely to be on alendronate treatment than patients with hip fractures, but oral glucocorticoid use was more prevalent. Furthermore, the risk of all three types of fracture was attenuated by a high degree of adherence to alendronate, and the ratio between classical hip fractures and subtrochanteric/diaphyseal femur fractures was identical in alendronate-treated fracture patients and their matched untreated fracture controls even in the small number of long-term, highly adherent patients. Interestingly, almost identical results were seen in etidronate users, where the incidence of subtrochanteric and diaphyseal fractures was slightly higher than observed in alendronate users. Taken together, these results suggest that the increased risk of subtrochanteric and diaphyseal femur fractures with alendronate use is more likely to be caused by osteoporosis than by alendronate therapy. Our findings did not confirm the hypothesis that subtrochanteric or diaphyseal fractures with alendronate treatment occurred with a frequency over and above that expected for a typical osteoporotic fracture. We observed a similarly increased apparent risk of typical hip fracture in alendronate users, although solid clinical evidence exists to the effect that alendronate treatment reduces hip fracture risk.(8) Bisphosphonates are preferentially prescribed to fracture patients at high risk of proximal femur fractures—probably as a consequence of femoral BMD being used in clinical decision-making—and this increased risk is incompletely offset by therapy.
Initial concerns of excessive suppression of bone turnover were raised by a report of spontaneous nonspinal fractures during long-term (3–8 yr) alendronate treatment. Bone histomorphometry was indicative of markedly suppressed bone resorption, especially at the endocortical surfaces in all nine patients and delayed fracture healing in four patients.(12) In the subsequent subtrochanteric fracture case series by Kwek et al.,(7) a characteristic X-ray pattern was observed, and contralateral stress reactions were common. Interestingly, some patients had experienced pain and radiological stress reactions at the fracture site before development of the low-energy fracture. Only one patient received prednisolone, whereas two patients had known malignant disease, of these one had panhypopituitarism after radiotherapy for nasopharyngeal carcinoma and probably received hydrocortisone substitution. In the larger series of Neviaser et al.(5) assessing 70 patients with subtrochanteric or diaphyseal femur fractures,(5) 35% were alendronate users compared with only 7% in our national dataset. Part of the explanation for this discrepancy may lie in the denominator because they included only low-energy fractures, where this study had to include both high- and low-energy fractures because of missing or incomplete trauma mechanism codes. However, neither the series of Neviaser et al. nor the study by Kwek et al.(7) stated the total number (i.e., both high and low energy) of subtrochanteric or diaphyseal fractures from which the cases were drawn, making direct comparisons impossible. Danish national databases estimate that two of three subtrochanteric/diaphyseal fractures result from low-energy trauma, indicating by extension that only ∼10% of patients with low-energy subtrochanteric/diaphyseal femur fractures in Denmark were alendronate users. The difference may reflect ascertainment bias or simply national differences in prescribing practice.
National health databases are a source of long-term real-world data, providing information on how interventions perform in patients who would not have been able to enter the underlying clinical trials because of comorbidity or inability to adhere to the study protocol. Danish health registers have been shown to provide excellent capture of proximal femur fractures,(13) whereas the prescriptions database contained all redeemed prescriptions in the country since 1995, whether issued by hospitals, general practitioners, or private clinics. Because the registers are independent of health plans and insurance, patients do not leave health databases unless they die or emigrate.(9) These events are also captured. The main limitation of this study is that it is observational and health register based. For legal reasons, the identity of the patients could not be disclosed, and therefore, it was not possible to review X-rays or patient notes to confirm the diagnosis and grade the X-ray pattern.(14) The ICD-10 hierarchy does not differentiate between proximal and distal diaphyseal fractures, which may have diluted the strength of association at this fracture site. Furthermore, because trauma codes were not rigorously stated in the register even in the 2005 dataset, exclusion of fractures caused by high-impact trauma was not possible. It should also be borne in mind that the time span from introduction of alendronate on the Danish market to the end of observations was only 9 yr. Because of this and because of the relatively modest persistence with oral bisphosphonates,(15,16) only 178 alendronate-treated patients were long-term (6+ yr with MPR > 80%) highly adherent. Over a mean observation period of 2.5 yr, the risk of both hip and subtrochanteric/diaphyseal femur fractures was lower in patients with high MPR, exactly as previously reported for other osteoporotic fractures.(16) The ratio between classical hip fractures and nonclassical fractures remained constant, even in the limited number of long time adherers. It is important to bear in mind also that this dataset does not allow us to make inferences about subtype distributions of diaphyseal or subtrochanteric fractures. We are unable to address the possibility that alendronate could alter the ratio between spiral and transverse subtrochanteric fractures, possibly reducing the risk of the former while increasing the risk of the latter, because the ICD-10 coding system does not distinguish between these two subtypes. Although our results indicate that the total number of subtrochanteric fractures declines with adherence to alendronate treatment—an observation that provides essential reassurance to both the physician and to the patient—careful radiological and clinical review studies may be able to provide additional information specific to transverse subtrochanteric fractures.
Although in many ways more informative than case reports, observational cohort studies have obvious disadvantages compared with randomized clinical trials (RCTs). Importantly, treatment is not allocated randomly but agreed on by the treating physician and the patient. Unmeasured confounders can only be balanced out by randomization as in an RCT. We sought to limit confounding by establishing both cohorts from a population with recent fracture and by matching cohorts not only by age and sex but also by fracture site. Additional confounders were identified and entered into the subsequent analyses. Thus, it was clear from the cross-sectional study that oral glucocorticoid use was more prevalent in patients with subtrochanteric or diaphyseal fractures, so this was examined as a covariate. A tendency toward higher risk in glucocorticoid users was also seen in the cohort study (HR = 1.5, 95% CI = 0.7–2.8; p = 0.27), but glucocorticoid use did not modify the association between alendronate use and fracture risk. Indices of comorbidity were also used in the model; this included the ICD-10–modified Charlson index(17) and the number of co-medications.(18) Despite these precautions, residual confounding caused by unmeasured confounders will affect estimates. The decisions to prescribe and accept alendronate treatment is taken not only based on fracture history but also on risk factor profile and BMD. For this reason, the alendronate cohort almost certainly includes more patients with low BMD, physical inactivity, low BMI, current smoking, and other risk factors. It was therefore of particular relevance to study the relationship between degree of exposure and the incidence of fractures within the alendronate treated cohort itself and this analysis indicated an inverse relationship between MPR and risk.
This evidence base is split between observations indicating that low-energy subtrochanteric or diaphyseal femur fractures associated with long-term alendronate use have unusual radiological and clinical features, clinical trials indicating long-term safety, and epidemiological and pharmacoepidemiological observations indicating that these fracture types may share the epidemiology and treatment response characteristic of osteoporotic fractures. Further research, perhaps using advanced skelet al imaging in long-term bisphosphonate users, could provide important information on biological plausibility.
This study was funded by the Kaptajnløjtnant Harald Jensen og Hustrus Fond.