Comparing non-vertebral fracture risk reduction with osteoporosis therapies: looking beneath the surface
- First Online:
- Cite this article as:
- Sebba, A. Osteoporos Int (2009) 20: 675. doi:10.1007/s00198-008-0802-0
- 97 Views
Data from pivotal trials of pharmacologic agents used to treat osteoporosis differ, suggesting that these agents vary in ability to reduce the risk of non-vertebral fractures (NVFs). However, variability among clinical trials in inclusion criteria, baseline characteristics, and definition of NVFs may account for many of these apparent differences.
Data from pivotal trials of individual pharmacologic agents for osteoporosis differ, and suggest that differences may exist between anti-resorptive agents in their ability to reduce the risk of NVFs. Careful examination of these trials’ inclusion criteria and patient characteristics indicates substantial differences between patient populations with respect to the baseline risk of NVFs. When baseline fracture risk is lower, the ability to produce a statistically significant reduction in fracture risk over the course of a clinical trial is reduced.
Analysis of clinical trials reveals that the number and type of baseline vertebral fractures and also baseline bone mineral density, all associated with the risk of vertebral fracture, vary.
Discussion and conclusion
The propensity to fall and patient frailty are additional factors associated with fracture risk that may influence study outcomes. One of the most significant variables, which also often differs considerably between trials, is the definition of an NVF. Variability between clinical trials in inclusion criteria, patients’ baseline characteristics, and how NVFs are defined may account for much of the apparent difference between agents in their ability to reduce NVF risk.
Fragility fractures in women with postmenopausal osteoporosis are common and associated with a significantly greater mortality risk compared with postmenopausal women without fractures [1, 2]. Many clinical trials have been conducted to address the efficacy of pharmaceutical agents for the treatment and prevention of osteoporosis, but almost all of these have had as their primary end-point, the treatment effect on vertebral fractures rather than non-vertebral fractures (NVFs). The primacy of vertebral fractures over NVFs in clinical trials is in large measure, a consequence of Food and Drug Administration (FDA) guidelines, which recommend that the treatment efficacy of osteoporosis agents be demonstrated with reductions in vertebral fracture risk over a period of 3 years . Furthermore, vertebral fractures are usually the earliest clinical indication of osteoporosis and are common; indeed, they are the most common fracture type at any single anatomical site [4–7].
In order for an osteoporotic agent to demonstrate efficacy, it is necessary to evaluate it in a population that is at increased risk. In most clinical trials, selection of patients has tended to favor women at a higher risk for vertebral rather than NVFs. This is, in part, a result of excluding patients with prior hip fractures from placebo-controlled trials, as is commonly practiced for reasons of safety; at the same time, patients with one or two non-severe vertebral fractures are typically considered eligible for inclusion. Because the presence of a prior vertebral fracture substantially increases the likelihood of future fracture, and because of the relatively greater ease identifying morphometric vertebral fractures, women with prevalent vertebral fractures are most commonly enrolled in clinical trials . NVFs will also occur (although the population enrolled is not selected for this group) and are generally dealt with as secondary end points.
NVFs are common, with an annual incidence of approximately 1% in the overall population . The lifetime risk of both wrist and hip fractures in postmenopausal women is estimated to be 18% and 16%, respectively . These fractures are also important, with substantial morbidity, particularly with respect to hip fractures. NVFs frequently require hospitalization and surgery, as well as long-term rehabilitation . Hip fractures alone account for approximately 300,000 hospitalizations per year.
The frequency of NVFs relative to vertebral fractures should not be underestimated. The vertebral fracture arm of the Fracture Intervention Trial (FIT) included more than 2,000 U.S. postmenopausal women with a prior vertebral fracture. In this group, among patients receiving placebo, 15% experienced a vertebral fracture over the course of 3 years, while 14.7% experienced at least one non-vertebral fracture . The overall incidence of osteoporotic NVFs in general is greater than that of vertebral fractures; the National Osteoporosis Foundation estimates that osteoporosis accounts for approximately 1.5 million NVFs in the United States annually, compared with 547,000 vertebral fractures .
Since nearly all fracture trials report the occurrence of NVFs, there is a temptation to use these fracture reduction figures from different trials in order to compare the relative efficacy of NVF reduction. (It should be noted that, to date, no weekly or monthly therapy for the prevention of NVFs has been studied for efficacy with the exception of zoledronic acid. Thus, except for zoledronic acid, all data in this review are based on earlier once-daily therapy.) The apparent similarities among the trials make this a seductive approach. The present review is intended to describe the variations and subtle variables that exist between clinical trials undertaken to assess NVF risk reduction, and examine the factors that affect NVF risk, evaluating how these factors may affect the outcomes of clinical trials designed to measure the efficacy of pharmacologic agents.
Factors affecting reductions achieved in non-vertebral fracture risk
Ideally, head-to-head clinical trials would be conducted so as to provide more reliable comparison data for efficacy. In the osteoporosis field, such studies have been done, but most depended on surrogate markers instead of fracture outcomes [12–14] or used retrospective database analysis, a technique subject to channeling (i.e., the tendency for patients with different baseline characteristics to be treated with different medications within the same class, as when a newer medication with specific claims of efficacy is given to a subgroup of patients with unique risk factors) and possibly other biases . In the absence of head-to-head trials, clinicians may attempt to compare different clinical trials in order to arrive at some understanding of relative treatment efficacy. Regardless of the condition being treated or the agent being evaluated, these kinds of comparisons are challenging and should be discouraged. Differences in study designs, patient populations, and dosing equivalency, among many other considerations, can make it difficult to reliably compare results from different trials. This is certainly the case when comparing the relative antifracture efficacies of bisphosphonates and other agents in osteoporosis patients, particularly with regard to NVF efficacy.
Fracture efficacy in clinical trials of osteoporotic agents
Characteristics and fracture rates of placebo groups in major clinical trials for osteoporotic fracture risk reduction (in order of decreasing non-vertebral fracture rate)
≥1 Baseline vertebral fracture (%)
Mean age (years)
Non-vertebral fracture rate (%)
Vertebral fracture rate (%)
Risedronate (5.0 mg/day)
Risedronate (5.0 mg/day)
Risedronate (2.5 or 5.0 mg/day)
Group I: 39 Group II: 45
Group I: 74 Group II: 83
Alendronate (10 mg/day)
Alendronate (10 mg/day)
FIT w/o VF20
Ibandronate (2.5 mg/day)
Zoledronic acid (5 mg/y)
Teriparatide (20 μg/day)
Raloxifene (60 mg/day)
Calcitonin (100 IU/day or 200 IU/day)
100 IU/day: 10b
100 IU/day: 22
200 IU/day: 15
200 IU/day: 18b
This review will focus mainly on bisphosphonates for NVF reduction, but it is worth noting that, in total, only three osteoporosis agents studied in phase III trials have demonstrated NVF risk reduction in the overall population. The North American Vertebral Efficacy with Risedronate Therapy (VERT-NA) study—the only pivotal study that had NVF reduction as a primary end point—found a significant 39% reduction in NVF in patients treated with 5 mg daily oral risedronate (p < 0.02) . The multinational VERT study (VERT-MN) did not show a significant reduction . The Fracture Prevention Trial (FPT), which evaluated teriparatide, observed a significant reduction in NVFs of 53% (p < 0.05) , and the Health Outcomes and Reduced Incidence with Zoledronic Acid Once Yearly (HORIZON) fracture trial observed a significant 25% reduction in risk of NVFs in patients treated with an annual 5 mg dose of IV zoledronic acid (p < 0.001) .
Key factors in the conduct of clinical trials
Defining non-vertebral fractures
Non-vertebral fracture exclusion criteria in major clinical trials for osteoporotic fracture risk reduction
Exclusion criteria for non-vertebral fractures
A 3-year (2 years for the 2.5 mg/day group) study of 1,226 postmenopausal women with ≥2 prevalent VFs, comparing risedronate (2.5 or 5 mg/day) with placebo
Fractures other than those to the wrist, leg, humerus, hip, pelvis, or clavicle
A 3-year (1 year for the 2.5 mg/day group) study of 2,458 postmenopausal women with ≥1 prevalent VF, comparing risedronate (2.5 or 5 mg/day) with placebo
Fractures other than those to the wrist, leg, humerus, hip, pelvis, or clavicle
A 3-year study comparing risedronate (2.5 or 5 mg/day) with placebo, consisting of two groups: 1) 5,445 postmenopausal women (age 70–79 years) with either a T-score ≤−4 or a T-score ≤−3 plus a nonskeletal risk factor for hip fracture, and 2) 3,886 postmenopausal women (age ≥80 years) with ≥1 skeletal risk factor or a T-score ≤−4, or a T-score ≤−3 plus a hip-axis length of ≥11.1 cm
Fractures other than those to the wrist, leg, humerus, hip, pelvis, or clavicle
A 3-year study of 2,027 postmenopausal women with an FN BMD ≥0.68 g/cm2 and ≥1 VF, comparing alendronate (5 mg/day for first 2 years and 10 mg/day thereafter) with placebo
Pathological fractures, excessive trauma fractures, facial and skull fractures
FIT w/o VF20
A 4.5-year study of 4,432 postmenopausal women with an FN BMD ≥0.68 g/cm2 and no VFs, comparing alendronate (5 mg/day for first 2 years and 10 mg/day thereafter) with placebo
Pathological fractures, excessive trauma fractures, facial and skull fractures
A 3-year study of 881 postmenopausal women with a T-score ≤−2.5 comparing alendronate (5 or 10 mg/day) with placebo
A 3-year study of 2,946 postmenopausal women with a T-score ≤−2.0 and 1–4 VFs, comparing ibandronate (2.5 mg/day or intermittent dosing [i.e., 20 mg every other day for 12 doses, every 3 months]) with placebo
Fractures of the hands, feet, face, and skull
A 3-year study of 7,765 postmenopausal women with a T-score ≤−2.5 or a T-score ≤−1.5 plus ≥2 mild or 1 moderate VF, comparing zoledronic acid (5 mg 1 infusion/y) with placebo
Excessive trauma fractures; fractures of the fingers, toes, and facial bone
A 21-month study of 1,637 postmenopausal women with ≥2 mild or 1 moderate VF (plus a T-score ≤−1 in patients with <2 moderate VFs), comparing teriparatide (20 or 40 μg/day) with placebo
Excessive trauma fractures; fractures of the skull, face, hands, fingers, and toes
A 3-year study comparing raloxifene (60 or 120 mg/day ) with placebo in 7,705 postmenopausal women divided into two groups: 1) women with a T-score ≤−2.5, and 2) women with a T-score ≤−2.5 plus ≥1 moderate/severe VF or ≥2 mild VFs or ≥2 moderate VFs
Pathological fractures; fractures of the fingers, toes, and skull; fractures resulting from a traffic collision, beating, or as a result of being struck by a moving or falling object
In the case of the FIT trial, which evaluated alendronate, facial and skull fractures were excluded, as were pathological and excessive trauma fractures; however, fractures of the digits were included [11, 20]. The oral iBandronate Osteoporosis vertebral fracture trial in North America and Europe (BONE) excluded facial and skull fractures as well as digital fractures but not pathological or excessive trauma fractures . The HORIZON trial excluded digital and facial fractures and fractures resulting from excessive trauma but did not exclude pathological fractures [22, 26]
In contrast to these trials, the VERT-NA, VERT-MN, and Hip Intervention Program (HIP) trials of risedronate defined the sites constituting NVFs in a completely different way. For the purposes of assessing the efficacy of risedronate in NVF reduction, NVFs were defined as fractures occurring in one or more of six specific sites—clavicle, humerus, wrist, pelvis, hip, and leg—regardless of whether or not these fractures were associated with trauma [16–18]. Fractures from all six sites were combined into a composite NVF end point. Using this approach, both the VERT-NA and one arm of the HIP trial (the group of women 70–79 years of age enrolled with a low bone mineral density [BMD]) demonstrated a significant reduction in NVF with risedronate [17, 18].
Among the outliers of NVF definitions, it is worth noting the phase III trial conducted by Liberman et al. , an earlier trial of alendronate in which no limitations on NVFs were given either for location of fracture or by degree of trauma. The Multiple Outcomes of Raloxifene Evaluation (MORE) trial, on the other hand, excluded not only digital and skull fractures as well as pathological fractures but also “fractures resulting from a traffic collision, beating, or having been struck by a falling or moving object” . The inclusion of fractures that are not osteoporotic in nature could confound the outcomes in a clinical trial that is assessing the efficacy of an anti-osteoporotic agent.
The effect of this definitional issue can be seen in the vertebral fracture arm of the FIT, which studied osteoporotic women with preexisting vertebral fractures (a separate arm of the FIT trial studied women with low BMD but no preexisting vertebral fractures ). In the original reporting of the trial, patients receiving active treatment experienced a non-significant reduction in NVFs compared with placebo-treated patients (risk reduction of 20%; 95% confidence interval [CI]: 63, 101) . The failure of this analysis to show a reduction in NVFs with alendronate was addressed in a subsequent retrospective analysis in which “nonvertebral osteoporotic fractures” were evaluated. This group was now defined in the same way as were NVFs in the VERT, HORIZON, and HIP trials (i.e., as fractures of the clavicle, humerus, wrist, pelvis, hip, or leg) . When defined in this way, patients with existing vertebral fractures receiving alendronate experienced a statistically significant 32% risk reduction for non-vertebral osteoporotic fractures during the course of the trial (95% CI: 0.49, 0.92), a rate similar to that achieved with risedronate . This comparison provides strong evidence of the importance of restricting the definition of fractures in clinical trials to relevant osteoporotic sites, and also illustrates the pitfalls for the non-critical reader of clinical trials in which different investigators use the same term (NVF) but with different meanings.
The question of dosing in evaluating NVF risk is contingent on the relative potency and duration of action of the various osteoporotic agents. Among bisphosphonates, potency is at least partially related to a given agent’s ability to inhibit farnesyl diphosphate (FPP) as well as its affinity for bone mineral . Binding affinities for hydroxyapatite, for example, appear different between bisphosphonates, and this may play a role in the variable persistence of efficacy between these agents.
Understanding the importance of dosing in the context of clinical trials has been confounded by the fact that in two pivotal fracture trials the dose was much less than that currently used. In FIT, the dosage of alendronate for the first two of the 3–4.5 years of the trial was 5 mg/day, compared with the approved 10 mg/day or 70 mg/week for the treatment of osteoporotic women. In BONE, the dosage of ibandronate used was 2.5 mg/day, in contrast to the currently approved 150 mg/month (double the dose) or 3 mg every 3 months given intravenously (i.v.). The lower dose used may well have contributed to the apparent lack of effect in these trials. It should be noted that at doses above 50 mg, ibandronate absorption is nonlinear relative to dose . Still, the fact that ibandronate is the only bisphosphonate in which superiority of the less frequent dose was demonstrated in a trial using BMD as an endpoint suggests that NVF reduction may occur in clinical practice even though the BONE trial did not demonstrate such a reduction. It may also be the case that NVF reduction is more sensitive to dosing than are vertebral fractures. This is supported by a dose-related reduction in NVFs in an earlier trial looking at alendronate dosed at 1, 2.5, and 5 mg/day  and by data from meta-analyses of these two agents (see below).
Baseline differences in clinical risk factors for non-vertebral fracture
One method of evaluating studies is to review the naturally occurring fracture rate in untreated patients. It is evident from examining the NVF and vertebral fracture rates of the placebo groups of the various bisphosphonate clinical trials that substantial differences exist in the fracture risk among the various trial populations (Table 1). This variability may be seen simply by noting the differences between the VERT-MN trial, in which the placebo group experienced an NVF rate of 16%, and the BONE trial, in which the NVF rate for placebo-treated patients was about half (8.2%) [16, 21].
It is also instructive to compare the differing outcomes of the VERT-NA and VERT-MN trials and the respective fracture risks of their patient populations. While it is desirable in a study to have as many fractures as possible, it is tempting to assume that this alone will demonstrate efficacy. In VERT-NA, the NVF rate for placebo-treated patients was 8.4%, and patients receiving active treatment did experience a significant reduction in NVFs (risk reduction of 39%; 95% CI: 6, 61; p = 0.02) . In VERT-MN, where the NVF rate for placebo-treated patients was, at 16%, nearly double that of VERT-NA, NVF reduction in the active treatment group did not reach statistical significance (risk reduction of 33%; 95% CI: 44, 104; p = 0.063) . Thus, successful NVF reduction is clearly not purely contingent upon the event rate.
These differences in patient populations are, not surprisingly, also reflected in the vertebral fracture rates of the placebo-treated groups. For example, in VERT-NA and VERT-MN, the 1-year placebo vertebral fracture rates were 6.4% and 13%, respectively [16, 17]. The FPT trial, which evaluated teriparatide versus placebo, observed a cumulative vertebral fracture rate of 14% in the placebo arm at approximately 18 months . In contrast, in the pivotal trial of ibandronate (BONE), which studied approximately 3,000 women over 3 years, an incident vertebral fracture rate in the placebo group of 1.8% after 1 year and 6% after 2 years was observed . The difference between these latter studies does not appear to be baseline BMD at the spine. The mean T-score at the lumbar spine was similar (−2.8 in BONE and −2.6 in FPT), but the incidence of vertebral fractures was substantially different [21, 23].
No early spine radiographs were obtained in FIT at these early time points.
Baseline T-score and vertebral fracture criteria for inclusion in major clinical trials for osteoporotic fracture risk reduction
BMD T-score and/or baseline fracture inclusion criteria
≥2 vertebral fractures
≥2 vertebral fractures or 1 vertebral fracture and LS T-score ≤−2
Group I (age 70–79 years): FN T-score <−4.0 or FN T-score <−3.0 and an extra risk factor for hip fracture
Group II (age ≥80 years): at least one nonskeletal risk factor, FN T-score <−4.0 or FN T-score <−3.0 + a hip-axis length ≥11.1 cm
FN T-score ≤−2 and ≥1 vertebral fracture
FIT w/o VF20
FN T-score ≤−2 and no vertebral fractures
LS T-score ≤−2.5
LS T-score of −2 to −5 in at least 1 vertebra; 1–4 vertebral fractures
FN T-score ≤−2.5, or FN T-score ≤−1.5 with ≥2 mild or ≥1 moderate VF
≥1 moderate or 2 mild atraumatic thoracic or lumbar vertebral fractures; for women with only 1 moderate vertebral fracture, LS or FN T-score <−1a
FN or LS T-score <−2.5 and/or ≥2 mild or ≥1 moderate or severe vertebral fracturea
1–5 LS or thoracic vertebral compression fracture and LS T-score ≤−2, and no prior hip fracture
Baseline fracture status
It is known that patients with a prior vertebral fracture are at increased risk for subsequent vertebral fracture. A meta-analysis of 34 osteoporosis fracture studies reporting risk associated with subsequent fracture found that the relative risk of a future vertebral fracture among perimenopausal and postmenopausal women who have experienced a prior vertebral fracture is 4.4 (95% CI: 3.6, 5.4) . Of particular interest, though, is the fact that prior vertebral fractures are associated not only with vertebral fracture risk but also with increased risk for NVFs. The same meta-analysis found that women who had previously experienced vertebral fracture also had an increased risk of sustaining a subsequent wrist fracture of 1.4 (95% CI: 1.2, 1.7) and a relative risk of hip fracture of 2.3 (95% CI: 2.0, 2.8) .
Although study subjects included in many of the clinical trials evaluating osteoporotic fracture had to have had a prior fracture, the particular type and number of these fractures can differ, as may be seen in Table 3. Inclusion criteria for prevalent vertebral fractures range from no fractures (as in the clinical fracture arm of FIT) to one or more. In VERT-MN, the mean number of prevalent vertebral fractures was four, and in VERT-NA, the mean number was 2.5 [16, 17]. Similarly, the FPT and BONE each required at least one vertebral fracture to enroll. Because each additional vertebral fracture increases the risk of subsequent fracture, even within the first year subsequent to sustaining a fracture, the heterogeneity in prevalent fractures in the populations studied may have influenced their respective efficacy outcomes .
It is, moreover, not only the presence or number of prior vertebral fractures that determines NVF risk; prevalent NVFs are themselves associated with the occurrence of subsequent NVFs . Data from the FPT also suggest that this relationship exists for prevalent NVFs: among placebo patients with 0, 1, or ≥2 baseline NVFs, the incidence of subsequent NVFs was 3.6%, 8.2%, and 18%, respectively .
The accepted standard for determining osteoporosis, endorsed by the World Health Organization (WHO), is a BMD ≥2.5 standard deviations below the young female adult mean (i.e., a T-score ≤ −2.5), based upon data from Caucasian women aged 20–29 years participating in the National Health and Nutrition Examination Survey . The reference standard for BMD is at the femoral neck, although measurements may be taken at the lumbar spine (using preferentially L1–L4), the total hip, or femoral neck .
It is interesting to note that in the case of the BONE trial (in which no NVF reduction was demonstrated in the overall enrolled population), the inclusion criteria specified that study patients have between one and four prevalent vertebral fractures as well as a BMD T-score of −2.0 to −5.0. Most studies that use BMD at the spine as an entry criterion stipulate a T-score of −2 or less. It may be of note that for determining osteoporosis, BMD of the spine should be based on measurements taken at L1–L4 inclusive, preferably using all four vertebrae . In the BONE study, BMD had to be less than −2 in one or more lumbar spine vertebrae , i.e., patients might have been included in the study with a T-score of less than −2 in only a single vertebra. While these patients clearly had osteoporosis (based on the presence of a prevalent vertebral fracture), they would also have been at a very different overall fracture risk compared to patients with lumbar BMD that was more homogeneously low . This may partly provide an explanation as to why, in the placebo group, there was a very low vertebral fracture rate after 1 year (1.8%) and a low overall NVF rate after 3 years (8.2%; data on file [reference # 161–233], Hoffmann-La Roche Inc., Nutley, NJ 07110, received May 15, 2008) . Furthermore, this provides an example of the potential influence of a rather nuanced factor influencing study outcomes.
Looking at the issue of lumbar spine versus femoral neck BMD measurements more closely, a Canadian study sought to determine the impact of several potential risk factors upon fracture rates. In this prospective population-based cohort study of 5,143 postmenopausal women, the authors observed that for each standard deviation decrease in lumbar spine BMD, the relative risk of an NVF was 1.34 . A decrease in femoral neck BMD of one standard deviation was associated with a relative risk for NVF of 1.41. Thus, it can be seen from these data that NVFs become more common as T-scores become lower. These data, furthermore, give credence to the speculation that low femoral neck BMD is associated with a greater risk for NVF than is low lumbar spine BMD.
As may be seen in Table 3, the use of entry femoral neck BMD measurement criteria are by no means universal among osteoporosis clinical trials. In fact, several studies, including the VERT-MN and VERT-NA trials and FPT, allowed study subjects to enter with no minimum BMD measurement of any kind as long as they had greater than or equal to two prevalent fractures. In addition, several pivotal studies (FIT, VERT, BONE, FPT, and Prevent Recurrence of Osteoporotic Fractures [PROOF]) included BMD thresholds higher than the WHO criteria for osteoporosis diagnosis [11, 16, 17, 21, 23, 25]. If this were the only risk factor, it seems likely that patients without BMD evidence of osteoporosis would have been at lower risk for NVFs than patients with T-scores greater than −2.5.
Baseline femoral neck T-scores were in fact quite heterogeneous. In the case of the VERT-NA trial, despite the fact that no inclusion criteria for femoral neck BMD were specified, the mean baseline femoral neck T-score was −2.7 for patients receiving 5 mg risedronate and −2.6 for patients receiving placebo , and most patients in the HORIZON trial had femoral neck T-scores less than −2.5, including 70.8% in the placebo group and 72.6% in the zoledronic acid group .
This issue was, however, of particular concern for the FIT. When this study was initiated, the original intent was to enroll patients with a T-score of −2.0 or lower. This was calculated as requiring a femoral neck BMD measurement of ≤0.68 g/cm2, which was thought to correspond to a T-score of −2. Subsequent to this determination, data from National Health and Nutrition Examination Survey III were released showing that a BMD value of 0.68 g/cm2 corresponded to a T-score of −1.6, not the previously assumed −2. Patients recruited for the study were thus at a lower risk for fracture than anticipated. Approximately one third of the patients in the study turned out to have a BMD of −1.5 to −2, and another third had a T-score of −2 to −2.5. Both groups would be regarded as having low bone mass (osteopenia) rather than osteoporosis, another possible contributory reason that the study may not have demonstrated NVF reduction in the original analysis .
The influence of inclusion criteria in clinical trials measuring fracture risk can be seen in an analysis of fracture risk in the FIT broken down by baseline T-scores. This analysis, applied to the clinical fracture arm (in which there was no vertebral fracture requirement for entry into the trial), was preplanned by the study authors and went some way in addressing the problem of the unintended −1.6 T-score entry threshold . This analysis revealed that the efficacy of alendronate for NVF reduction was strongly associated with femoral neck T-score. It was shown that alendronate did not have any positive effect on clinical fractures in patients with a femoral neck T-score between −1.6 and −2.0 (relative hazard = 1.14; 95% CI: 0.82, 1.60) or in patients with a T-score between −2.0 and −2.5 (relative hazard = 1.03; 95% CI: 0.77, 1.39) compared with placebo . In patients with a T-score less than −2.5, however, a significant clinical fracture risk reduction of 36% (relative hazard = 0.64; 95% CI: 0.50, 0.82) was observed.
A similar pattern was seen in the placebo-controlled Fosamax International Trial (FOSIT) of alendronate treatment in 1,908 otherwise healthy postmenopausal women with low BMD; inclusion criteria for this 1-year multicenter trial were based on lumbar spine BMD (T-score ≤ −2.0) . After 1 year, patients receiving active treatment experienced a 47% reduction in relative risk of NVFs (95% CI: 10, 70; p = 0.021) . A retrospective analysis was undertaken to determine the relationship between baseline hip BMD and the degree of NVF reduction, indicating that patients with a hip T-score greater than −2.0 had a relative risk for NVF of 1.2 (95% CI: 0.5, 2.9); those with hip BMD from −2.0 to −2.5 had an NVF relative risk of 0.32 (95% CI: 0.7, 1.5); and patients with a hip T-score less than −2.5 had a relative risk of 0.26 (95% CI: 0.1, 0.7) .
Similar results were found in the BONE trial. Minimal entry criteria included a mean femoral neck T-score of −2.0, and thus, patients in BONE, like those in the FIT, appear to have been at low risk for NVFs . Results from the overall studied group did not show a decrease in NVFs among those patients who received 2.5 mg daily of oral ibandronate. As in FIT, this raised questions about the efficacy of ibandronate with respect to NVF reduction, but a similar pattern of a relationship to entry femoral neck BMD emerges. In a subanalysis of the data, those patients who had a femoral neck T-score of less than −3.0 experienced a relative risk reduction for NVFs of 69% (p = 0.013) .
Possible further evidence of the importance of baseline femoral neck BMD can be seen in the HIP trial of risedronate versus placebo, in which patients were stratified into two age groups: 70–79 years and ≥80 years. Patients in the 70–79 years group, who had known mean T-scores between −2.9 and −2.7, experienced a 20% reduction in NVF (p = 0.03) . Mean femoral neck BMD was unknown in the 80 years and older group; no significant improvement in NVF fracture risk was seen in this group (58% of these older patients were included in the study based on the presence of nonskeletal risk factors, and only 16% were included as a consequence of low femoral neck BMD) .
Although some are retrospective analyses, there is a high degree of consistency in the data from the FIT, FOSIT, BONE, and HIP trials. For femoral neck BMD T-scores, a pattern emerges, which suggests that baseline T-score influences the degree of NVF risk reduction, with subgroups of patients with a lower T-score more readily demonstrating evidence of NVF reduction. These BMDs ranged from the highest BMD (a mean femoral neck T-score of −2) in the BONE trial to the first arm of HIP, in which the mean femoral neck T-score was −3.7 , the lowest reported femoral neck BMD in these studies.
Thus, leaving aside age, fractures, and other considerations, there may have been a fourfold difference in fracture risk between these groups.
Baseline bone turnover markers
Much like baseline BMD and baseline fracture status, baseline levels of bone turnover markers (BTMs) may influence the outcomes of clinical trials. A post hoc analysis of data from the FIT demonstrates the potential role of BTMs in fracture risk. This analysis, by Bauer et al. , looked at the association between the incidence of NVFs in patients receiving active treatment and patients’ baseline levels of three BTMs: N-terminal propeptide of type I collagen (P1NP), bone-specific alkaline phosphatase (BSAP), and C-terminal cross-linked telopeptide of type I collagen (sCTx). Baseline levels of each of the three BTMs were stratified by tertile. Among osteoporotic women with the highest level of P1NP, active treatment was associated with a reduction of 46% in NVFs. Among women with the lowest baseline levels of P1NP, the NVF reduction was 12%, a difference that was statistically significant (p = 0.03). Reductions in fracture risk based on BSAP or sCTx were not significant. While these data are quite limited, one or more BTMs may ultimately prove to be predictors of NVF reduction.
Additional risk factors
Significant NVF reductions reported within the bisphosphonate class ranges from 25% (in the HORIZON recurrent fracture trial) to 39% (in the VERT-NA study) [17, 40]. Reduction in vertebral fractures in the bisphosphonate class of medications, however, is consistently reported in the range of 41–65%, suggesting that bisphosphonates have a lesser effect on reduction of NVFs than vertebral fractures [11, 16, 17, 21].
There are a number of factors, in addition to those already discussed, that determine whether a patient experiences an NVF as opposed to a vertebral fracture. Two notable factors are falls and frailty. Falls are associated far more with NVFs than with vertebral fractures, which in contrast, are more likely to be associated with low vertebral bone mass and compression fractures . A 3-year cohort study of 5,201 postmenopausal women found that a history of falls is an independent risk factor for NVFs .
Frailty is a less commonly used concept for evaluating fracture risk. Nevertheless, it provides a potentially useful means of describing a cluster of factors not in themselves sufficiently significant to measurably cause fractures, but which in combination may do so. Ensrud et al.  recently examined whether frailty, as a phenotype, conferred an independent risk for osteoporotic fracture. The definition of frailty consisted of the presence of three or more of the following factors: (1) unintentional weight loss of ≥5%, (2) weakness measured by grip strength, (3) self-reported low energy, (4) slow walking speed, and (5) low physical activity. In this cohort study of 6,724 women, frailty was, in fact, found to be an independent risk factor for NVF: the hazard ratio for NVF in frail patients was 1.25 (95% CI: 1.05, 1.49) . Although the propensity to fall and frailty status are relevant risk factors for NVFs, it is clearly the case that no trials have prioritized these factors for assessment.
Differences in bone tissue type
As discussed in the introduction (above), for the purposes of showing fracture reduction for registration, vertebral fractures are regarded as a single site. Since these fractures are frequent, this has led perforce to using vertebral fractures as the preferred single endpoint for demonstrating fracture reduction. Other fractures (i.e., fractures not occurring at the spine) have thus become “NVFs.” It is possible, but harder, to power a trial to show NVF reduction at one particular site (e.g., hip fracture reduction as a primary endpoint has been shown to be reduced in only one pivotal trial). If vertebral fractures represent one endpoint, then NVFs presumably can be considered as another. The notion of grouping all fractures that are not vertebral into one composite endpoint allows for a smaller and perhaps shorter trial that can still show reduction in the fracture rate in this cluster of endpoints. Although the classification of fractures into vertebral and non-vertebral types is useful in the clinical trial setting, it is a highly artificial distinction from the biological perspective, with very limited use to patients and physicians. From an anatomical perspective, there are two types of bone tissue that are structurally and functionally distinct: trabecular bone and cortical bone. Compared with cortical bone, trabecular bone has a greater surface area and is considerably more metabolically active . Trabecular bone is also distinguished from cortical bone in that it is found primarily in the vertebrae and the trochanter.
While it is the case that these two bone types possess structural as well as functional differences, it is possible that evaluating fractures by cortical and trabecular sites, rather than as vertebral fractures or NVFs, might yield useful information for the clinician and patient alike. Pending evaluation of this approach, it might be of some use to consider all fractures as a group when assessing fracture reduction and to look at the total sum of clinical fractures to evaluate fracture reduction efficacy, as was done in the FIT and HORIZON recurrent fracture trial [27, 40].
Some meta-analyses have been performed in an attempt to more precisely quantify the efficacy of osteoporotic agents in reducing the risk of NVFs [45–47]. Perhaps the best known of these meta-analyses, conducted by the Osteoporosis Research Advisory Group (ORAG), strongly discourages a comparison of the relative efficacy of each osteoporotic therapy based upon the results of their analyses of the various treatments .
Among the bisphosphonates, risedronate ≤2.5 mg/day reduced NVF risk by 27% (95% CI: 0.61, 0.87). In this meta-analysis, alendronate, when dosed at 10–40 mg/day, resulted in an overall NVF reduction of 49% . In the case of the alendronate data, however, the ORAG authors were able to compare treatment efficacy among different definitions of NVFs (as discussed above) and found that when dosing was 10–40 mg/day, there was a significant reduction in osteoporotic NVFs (relative risk [RR] 54%; 95% CI: 0.32, 0.66; p < 0.01); however, the reduction in nonosteoporotic NVFs failed to reach significance (RR 43%; 95% CI: 0.32, 1.02; p < 0.06) . Dosing appears critical: there was no decrease in NVF risk with the 5 mg/day dose.
Evidence from a recent meta-analysis of four ibandronate clinical trials consisting of 8,710 patients assessing NVF reduction at different doses provides further evidence for the importance of dosing . This meta-analysis divided patients into three groups based on their calculated annual cumulative exposure to ibandronate. With respect to the composite endpoint of “key NVFs” (i.e., clavicle, humerus, wrist, pelvis, hip, and leg), patients with the lowest annual cumulative exposure, including those participating in the BONE trial, experienced a non-significant NVF reduction of 13% (p = 0.33) . In contrast, those with the highest annual cumulative exposure (which included patients receiving two currently FDA-approved doses of 150 mg/month or 3 mg IV every 3 months) experienced a 34% risk reduction (p = 0.032). To date, no meta-analyses have been performed for NVF risk reduction with either teriparatide or zoledronic acid.
In clinical practice, NVFs are more prevalent than vertebral fractures and carry a substantial burden for the patient. Pharmacologic agents approved for the treatment of osteoporosis have all demonstrated efficacy in reducing the risk of vertebral fracture, but some phase III studies (e.g., FIT, MORE, BONE) have not shown reductions in NVF risk. This may be a consequence of differences in how NVFs are defined in clinical trials. It may also be due to differences in the baseline characteristics of the patients recruited for these trials. Differences in the number of prevalent NVFs and vertebral fractures, as well as severity of prevalent fractures, may affect the ability to demonstrate both vertebral fracture and NVF reduction. In addition, an increase in NVF risk is likely to be observed in women with lower BMD T-scores (measured at hip and possibly spine sites) and in those with lower levels of certain BTMs. Further complicating the situation is the influence and unpredictability of falls and trauma on fracture risk, particularly the fact that falls and trauma have a differential effect on NVFs compared with vertebral fractures. As long as the issue of risk of falls remains unaddressed, the ability to reduce the incidence of NVFs will probably continue to be less than that of vertebral fractures. Age and height are, of course, also a critical issues in NVF risk evaluation, and differences between baseline ages and heights of the patient populations within given studies may exert additional confounding factors in comparing fracture risk. For example, whereas patients in the BONE and VERT-NA trials had a mean age of 69 years, the mean ages of participants in VERT-MN and FIT (vertebral fracture arm) was 71. [11, 16, 17, 21] The HIP study, meanwhile, divided patients into two age groups with mean ages of 74 years and 83 years .
Future osteoporosis trials will need to assess NVF risk as part of total fracture risk. It should be possible to recruit a study population at elevated risk specific for NVF. An assessment tool has been developed that could be used to determine the NVF risk in patients screened for clinical trials . Factors found to be of importance in determining NVF risk include age, height, femoral neck T-score, prevalent vertebral fractures, prior NVFs, and serum levels of vitamin D .
The results of clinical trials of osteoporosis agents have led to the common assumption that significant differences exist between these agents with regard to NVF risk reduction. However, pooled retrospective analyses of alendronate and ibandronate, for example, suggest that both agents are probably effective in reducing the risk of NVFs at the current U.S.-approved doses. What differences may exist between the agents’ relative efficacy has yet to be clearly established. Close scrutiny of clinical trials reveals that differences in the inclusion criteria and baseline patient characteristics of the study populations, as well as how NVFs are defined, may account for the apparent differences between agents in their ability to reduce NVF risk.
Editorial assistance was provided by Insight Medical Communications Inc., supported by a grant from Roche Laboratories, Inc., which did not participate in the preparation or writing of this manuscript, nor did they provide financial support to the author for the purpose of writing this manuscript.
Conflicts of interest
Dr. Sebba is a consultant for Roche, Novartis, GSK, and Merck. He is a speaker for Lilly, Novartis, Roche, GSK, and Merck.