Change in Bone Turnover and Hip, Non-Spine, and Vertebral Fracture in Alendronate-Treated Women: The Fracture Intervention Trial

Abstract

We used data from the Fracture Intervention Trial to assess the relationship change in bone turnover after 1 year of alendronate or placebo treatment and subsequent hip, non-spine, and spine fracture risk among 6186 postmenopausal women. In the alendronate group (n = 3105), greater reductions in one or more biochemical marker were associated with a lower risk of fracture.

Introduction: There are few data on the relationship between short-term change in biochemical markers of bone turnover and non-spine fracture risk among bisphosphonate-treated women, and the clinical use of such measurements is unknown.

Materials and Methods: We measured biochemical markers of bone turnover (bone-specific alkaline phosphatase [bone ALP], intact N-terminal propeptide of type I collagen, and C-terminal crosslinked telopeptide of type 1 collagen) and BMD of the spine and hip at baseline and after 1 year of alendronate or placebo. During a mean follow-up of 3.6 years, 72 hip, 786 non-spine, and 336 vertebral fractures were documented.

Results and Conclusions: Each 1 SD reduction in 1-year change in bone ALP was associated with fewer spine (odds ratio = 0.74; CI: 0.63, 0.87), non-spine (relative hazard [RH] = 0.89; CI: 0.78, 1.00; p < 0.050), and hip fractures (RH = 0.61; CI: 0.46, 0.78). Alendronate-treated women with at least a 30% reduction in bone ALP had a lower risk of non-spine (RH = 0.72; CI: 0.55, 0.92) and hip fractures (RH = 0.26; CI: 0.08, 0.83) relative to those with reductions <30%. We conclude that greater reductions in bone turnover with alendronate therapy are associated with fewer hip, non-spine, and vertebral fractures, and the effect is at least as strong as that observed with 1-year change in BMD.

INTRODUCTION

A NUMBER OF effective osteoporosis treatments are now available, and recent guidelines have emphasized the need to appropriately identify and treat individuals at high risk of fracture.¹ Despite the increasing number of individuals receiving treatment, there is disagreement about the use of monitoring during therapy. Although low bone mass is a strong risk factor for fracture in untreated populations,^{2, 3} the usefulness of serial bone mass measurements during treatment is uncertain.^4-6 Among women treated for osteoporosis the strength of the relationship between change in bone mass and subsequent fracture varies considerably,^7-10 suggesting that other factors may be important or that techniques for assessing changes in bone mass lack the precision required to quantify this relationship accurately.

The antifracture efficacy of osteoporosis treatments that reduce bone turnover and the observation that accelerated bone turnover itself may be an independent risk factor for fracture^11-14 have led some to advocate the use of biochemical markers of bone turnover to monitor the response to antiresorptive therapy. The rationale for such measurements is that substantial changes in bone turnover occur rapidly, typically within a few months of initiation of therapy, whereas treatment-induced changes in bone mass are smaller and typically not evident for 1-2 years.⁴ Although several studies of antiresorptive treatments have found that short-term reductions in bone turnover, as assessed by biochemical markers, are associated with larger long-term increases in bone mass^{15, 16} and a lower risk of vertebral fracture,^{17, 18} there are few prospective data directly relating change in bone turnover to subsequent non-spine fracture outcomes. Recent reviews have commented on the need to study the role of monitoring bone turnover during treatment and the need for prospective studies relating change in bone turnover to fracture risk.¹⁵

The Fracture Intervention Trial (FIT), a 2.5- to 4.5-year study of 6459 postmenopausal women, showed that alendronate therapy significantly reduced the risk of hip, spine, and other nonvertebral fractures.^19-21 We measured biochemical markers of bone turnover using archived serum specimens from baseline and after 1 year of alendronate or placebo treatment to examine the effect of baseline levels of bone turnover, and changes in bone turnover after 1 year of therapy on the risk of fracture.

MATERIALS AND METHODS

Study design and subjects

We analyzed data from the FIT, a randomized, double-blind clinical trial of 6459 women recruited at 11 U.S. clinical centers.²² Subjects were postmenopausal women between the ages of 55 and 80 years with femoral neck BMD ≤0.68 g/cm² (Hologic QDR-2000). FIT had two arms: the Vertebral Fracture Arm,¹⁸ which included 2027 women who had vertebral fractures identified on radiographs at baseline, and the Clinical Fracture Arm, which included 4432 women without baseline vertebral fracture, but with a baseline femoral T score ≤ −1.6. Women in the Vertebral Fracture Study were followed for an average of 2.9 years (range, 2.5-3 years), whereas women in the Clinical Fracture Study were followed for an average of 4.3 years (range, 4.0-4.5 years). Both arms of FIT are analyzed together in this analysis. All women provided written informed consent, and the protocol was approved by the appropriate institutional review boards.

Subjects were randomly allocated to daily alendronate (ALN) or placebo (PBO). The dose of alendronate was initially 5 mg/day for 2 years but was increased to 10 mg/day (Merck & Co., Whitehouse Station, NJ, USA) at the second annual visit because other trials suggested that 10 mg had greater effects on BMD. Eighty-two percent of participants in each treatment group had dietary calcium intakes at baseline <1000 mg/day; they were asked to take a daily supplement (OsCal) containing 500 mg of elemental calcium and 250 IU of vitamin D.

Measurements

Biochemical markers of bone turnover:

Participants provided a serum specimen before randomization and at each subsequent annual visit: these were frozen and transported to a central repository. Twenty percent of randomized subjects, selected at random, were fasting during the baseline visit, but all follow-up specimens were nonfasting. All of the nonfasting specimens were stored at −20°C for ∼3 years before they were transferred to −70°C freezers for long-term storage. Paired baseline and 1-year follow-up specimens were thawed and assayed for biochemical markers of bone turnover in 2001 after a mean storage of 8.7 years for the baseline specimens. Approximately 14% of the cohort did not have a 1-year specimen; for these individuals, the specimen collected at the second annual visit was used.

Serum was analyzed for biochemical markers of bone turnover. Bone-specific alkaline phosphatase (bone ALP), a marker of bone formation, was measured with a standard assay (Tandem-R Ostase; Hybritech, San Diego, CA, USA). This IRMA uses two monoclonal antibodies directed against the human bone isoenzyme purified from human SAOS-2 osteosarcoma cells as a standard. This assay has a 16% cross-reactivity with the circulating liver isoenzyme,²³ and the intra- and interassay CVs are 7% and 12%, respectively. C-terminal crosslinked telopeptide of type 1 collagen (S-CTX), a marker of bone resorption, was measured using a one-step ELISA assay with reported intra- and interassay CVs of 5% and 8%, respectively.²⁴ Serum intact N-terminal propeptide of type I collagen (P1NP), another marker of bone formation, was measured by radioimmunoassay (Intact P1NP; Orion Diagnostica, Espoo, Finland).²⁵ The intra- and interassay CVs of the P1NP immunoassay are <5% and 8%, respectively. The P1NP assays were performed by the manufacturer, and the bone ALP and S-CTX assays were performed at a central facility (Synarc, Lyon, France).

BMD:

BMD was measured at the hip and posterior-anterior (PA) spine on all participants using Hologic QDR 2000 densitometers (Hologic, Waltham, MA, USA) and were repeated annually. Detailed descriptions of these procedures have been previously published.²⁶

Fracture outcomes:

A clinical fracture was defined as one diagnosed by a physician. Self-reports of fractures were confirmed by written reports of radiographs or other tests. Pathological fractures or fractures caused by trauma sufficient to fracture a normal bone in most young adults were excluded by a blinded Endpoints Adjudication Committee. Facial and skull fractures were excluded because they are not associated with osteoporosis or low BMD. Lateral spine radiographs were obtained according to published guidelines²⁷ at baseline and ∼3 (vertebral fracture arm) and 4 years (clinical fracture arm) after randomization. The assessment of morphometric vertebral fractures at baseline has been previously described.²⁶ A new vertebral fracture was defined as a decrease of 20% and at least 4 mm in any vertebral height from baseline to the end of the study. All assessments were blinded to treatment allocation.

Analysis

We analyzed the ALN and PBO groups separately. Baseline characteristics of ALN- and PBO-treated women were compared with χ² and t-tests. Results were similar with or without log-transformation of marker levels; only untransformed data are presented. For quintile analyses, we determined separate cutpoints for the ALN and PBO groups.

Our primary analysis assessed the association of baseline marker, or percent change in marker between baseline and 1 year, and documented spine, non-spine and hip fractures. Parallel analyses were performed using change in BMD (spine and femoral neck) between baseline and 1-year. Hip and non-spine fracture outcomes were examined in age-adjusted proportional hazards models, using time to event after the second marker or BMD measurement. Hip fractures were included in the non-spine fracture outcome. Fracture events reported before the follow-up marker or BMD measurement were excluded, but women with excluded fractures were still eligible for other endpoint analyses. Incident vertebral fractures detected at the end of follow-up were examined in age-adjusted logistic regression models. Unlike hip and other non-spine fractures, the exact date of most new vertebral fracture events was unknown, and no incident spine fractures were excluded from our analyses.

To assess the continuous relationship between change in marker and fracture risk, we performed an exploratory analysis using logistic regression, with change in marker flexibly modeled using splines. The Splus implementation of the spline fit decomposes the effect of marker change into linear and nonlinear components and provides a test for departures from linearity on the log-odds scale. Finding no evidence for nonlinearity in these exploratory analyses (p > 0.05), we used standard logistic models to plot the fitted probabilities of hip, spine, and non-spine fracture according to 1-year change in marker, stratified by treatment.

Secondary analyses were performed in subgroups to determine if our results were robust. Such analyses included limiting the analyses to osteoporotic women (baseline femoral neck BMD T score < −2.5 or existing vertebral fracture) and limiting the analysis to compliant women (defined as >75% self-reported compliance and >70% compliance using pill counts). We repeated our analyses including hip and non-spine fractures occurring before the follow-up marker or BMD measurement.

We also examined the effect of specific marker cutpoints selected a priori (at least a 15%, 30%, or 50% reduction in bone ALP or at least a 30%, 50%, or 70% reduction in P1NP) based on previous studies.^{16, 28} The 30% cutpoint approximates the least significant change for bone ALP, defined as the minimum difference between two successive measurements in an individual that likely represents a true biological change.²⁹

RESULTS

Baseline and follow-up characteristics

Baseline levels of biochemical markers were similar to those reported for other cohorts of similar age.¹⁴ Among the 3105 women randomized to the ALN group (Table 1), mean values fell by 31-59% over the first year of alendronate therapy. Spine BMD increased by 4.1% over the first year of ALN therapy and increased a total of 8.4% during the entire study. Among the 3081 women randomized to PBO (Table 1), mean values of bone turnover fell 8-31% over the first year, presumably reflecting the effect of calcium and vitamin D supplementation. Spine BMD increased 0.74% during the first year in the PBO group and increased a total of 1.6% during the entire study. S-CTX results in the fasting subset were similar to the overall group (Table 1).

Table Table 1.. Characteristics of FIT Population at Baseline and After 1 Year of ALN or PBO Therapy

Fracture outcomes

In the PBO group, there were 217 women with spine fractures, 426 with non-spine fractures, and 46 with hip fractures during the entire follow-up period. Among these fractures, 137 non-spine fractures and 15 hip fractures occurred before the follow-up marker or BMD measurement; these fractures were excluded from the primary change in marker and change in BMD analyses. In the ALN group, there were 119 women with spine fractures, 360 with non-spine fractures, and 26 with hip fractures during the entire follow-up period. Among these fractures, 134 non-spine and 11 hip fractures occurred before the follow-up marker or BMD measurement; these fractures were also excluded from the change in marker and change in BMD analyses.

Baseline biochemical markers and fracture risk

Analyses of baseline biochemical markers and any subsequent spine, hip, or non-spine fractures did not reveal any consistent trends in either the ALN or PBO groups (Table 2). Plots of fracture incidence by quintile of baseline marker revealed no evidence of nonlinear relationships (data not shown). The results were similar to the overall group among the subset of participants who provided fasting baseline S-CTX specimens (Table 2).

Table Table 2.. Baseline Marker Levels, Baseline BMD, and Fracture Risk

Change in biochemical markers and fracture risk

Analyses of change in biochemical marker after baseline revealed that greater reductions in one or more markers in the ALN group were associated with greater reductions in subsequent spine, hip, and non-spine fracture. When the predicted risk of hip, spine, or non-spine fracture from regression models were plotted by percent change of bone ALP, women in the ALN group with the greatest percent reduction in marker had the lowest risk of fracture, while those with the smallest reduction in bone ALP had the highest risk of fracture (Figs. 1, 2, and 3). There was no evidence that the relationships between change in marker and fracture risk were nonlinear (p > 0.05).

Figure FIG. 1.

One-year change in bone ALP and non-spine fracture risk among ALN-treated women. Percent change in bone ALP and predicted risk (log OR) of non-spine fracture (solid line) and 95% CI (dotted lines) from logistic regression model. Individual data points represented on the x axis. Departure from linearity p value = 0.16.

Figure FIG. 2.

One-year change in bone ALP and hip fracture risk among ALN-treated women. Percent change in bone ALP and predicted risk (log OR) of hip fracture (solid line) and 95% CI (dotted lines) from logistic regression model. Individual data points represented on the x axis. Departure from linearity p value = 0.33.

Figure FIG. 3.

One-year change in bone ALP and spine fracture risk among ALN-treated women. Percent change in bone ALP and predicted risk (log OR) of spine fracture (solid line) and 95% CI (dotted lines) from logistic regression model. Individual data points represented on the x axis. Departure from linearity p value = 0.34.

The significant relationships between change in marker and fracture risk among ALN-treated women were confirmed and quantified in an age-adjusted logistic regression with spine fractures as the outcome and with proportional hazard models with non-spine and hip fracture outcomes (Table 3). For example, each SD reduction in the 1-year percent change in bone ALP was associated with a 26% reduction in spine fractures (CI: 13%, 37%). Furthermore, we found that greater reductions in bone ALP were associated with fewer hip and non-spine fractures: each SD reduction in change in bone ALP was associated with an 11% reduction in the risk of non-spine fracture (CI: 0%, 22%), and a 39% reduction in the risk of hip fracture (CI: 22%, 54%). One-year changes in P1NP and S-CTX were also associated with a statistically significant reduction in spine fracture (Table 3). The relationships between 1-year change in PINP or S-CTX and subsequent hip and spine fracture closely resembled those observed with bone ALP, but the associations with PINP and S-CTX did not reach statistical significance. The relationships between change in S-CTX and spine and non-spine fracture were similar between the entire cohort and the fasting subset. There were too few women with hip fractures among fasting women to analyze separately.

Table Table 3.. Fracture Risk per SD of 1-Year Decrease in Marker or 1-Year Increase in BMD Among Alendronate-Treated Women

Relationships between change in marker and fracture outcomes were similar in models that further adjusted for change in spine BMD. For example, after adjusting for 1-year change in spine BMD, each SD reduction in percent change in bone ALP was associated with a 26% reduction in spine fractures (CI: 12%, 37%), a 14% reduction in non-spine fracture (CI: 2%, 24%), and a 40% reduction in hip fracture (CI: 21%, 55%). Results using absolute reduction in marker or absolute reduction in BMD did not differ from the overall results (data not shown).

We found no relationship between change in marker and subsequent fracture in the PBO group (data not shown). For example, among women in the PBO group, the relative hazard (CI) for non-spine fracture was 1.04 (CI: 0.92, 1.18) per SD reduction in bone ALP.

Baseline BMD, change in BMD, and fracture risk

Baseline hip BMD was associated with the risk of spine, non-spine, and hip fracture in both the ALN and PBO groups (Table 2). However, baseline spine BMD was not significantly associated with the risk of non-spine fracture in the ALN group or with the risk of hip fracture in the PBO group. Among ALN-treated women, the independent effects of baseline spine BMD and percent change in bone ALP on the risk of spine and non-spine fracture are shown in Figs. FIG. 4., FIG. 5.. These figures show that greater reductions in bone turnover were associated with fewer spine and non-spine fractures among women in the lowest and middle tertile of baseline spine BMD. The effects of greater reductions in bone turnover were inconsistent among women in the highest tertile of baseline spine BMD.

Figure FIG. 4.

Baseline BMD, change in bone ALP, and spine fracture risk. Age-adjusted risk of morphometric spine fracture by tertile of baseline spine BMD and tertile of percent change in bone ALP after 1 year of ALN therapy. Referent group (OR = 1.0) is those women in both the tertile with lowest baseline BMD and tertile with least reduction in bone ALP.

Figure FIG. 5.

Baseline BMD, change in bone ALP, and non-spine fracture risk. Age-adjusted risk of non-spine fracture by tertile of baseline spine BMD and tertile of percent change in bone ALP after 1 year of ALN therapy. Referent group (RH = 1.0) is those women in both the tertile with lowest baseline BMD and tertile with least reduction in bone ALP.

Table Table 5.. Fracture Incidence in PBO Group and in ALN Group With and Without ≥30% 1-Year Reduction in Bone ALP

In models designed in parallel with the change in marker analyses, we examined the effect of 1-year change in spine and hip BMD on fracture risk, again excluding hip and non-spine fractures that occurred before the follow-up BMD measurement (Table 3). Quintile analyses did not reveal evidence of nonlinear relationships (data not shown). In age-adjusted logistic regression and proportional hazard models in the ALN group, change in spine BMD was not associated with the risk of spine, non-spine, or hip fracture. Greater increases in hip BMD were not associated with fewer hip or non-spine fractures, but increases in hip BMD were associated with a reduction in spine fractures.

Other analyses among alendronate-treated women

We performed a variety of additional analyses to determine if our change in marker analyses differed among selected subgroups of women or with specific fracture outcomes. For example, results were unchanged in analyses restricted to those subjects with both baseline and 1-year serum specimens. For example, among the 2648 alendronate-treated women with both baseline and 1-year serum specimens, each SD reduction in bone ALP was associated with a 24% reduction in spine fracture (95% CI: 11%, 36%) and a 38% reduction in hip fracture (95% CI: 13%, 55%). The results were also similar to the overall results when analyses were limited to ALN-treated women with existing vertebral fracture or baseline hip BMD T score < −2.5 (N = 1764): each SD reduction in change in bone ALP in this group was associated with a 21% reduction in spine fracture (CI: 5%, 34%) and a 42% reduction in hip fracture (CI: 21%, 58%). Similarly, when analyses were limited to highly compliant subjects, the relationship between change in bone ALP (per SD reduction) and fracture among ALN-treated women did not change (spine fracture OR = 0.80; CI: 0.66, 0.96, and non-spine fracture RH = 0.84, CI: 0.74, 0.95). Finally, inclusion of hip and non-spine fractures occurring before the follow-up marker or BMD measurement or controlling our analyses for the presence or absence of vertebral fracture at baseline had little effect on the results (data not shown).

Clinical use of specific change in bone turnover cutpoints among alendronate-treated women

We performed exploratory analyses with the aim of developing clinically useful change in marker cutpoints for predicting fracture risk. We calculated odd ratios or relative hazard ratios for spine, non-spine, and hip fracture using specific change in marker thresholds selected a priori. Table 4 contains the risk of fracture among ALN-treated women who did not show a reduction in bone ALP of at least 15%, 30%, or 50%, and the risk of fracture among ALN-treated women who did not show a reduction in P1NP of at least 30%, 50%, or 70%. For example, among women with at least a 30% reduction in bone ALP (56% of alendronate-treated subjects), the hazard ratio for hip fracture was 0.26 (CI: 0.08, 0.83) compared with those with 1-year bone ALP reductions of <30%.

Table Table 4.. Fracture Risk With Specific Change in Bone ALP, Change in P1NP, and Change in BMD Cutpoints Among Alendronate-Treated Women

To show the effect of specific reductions in bone ALP on absolute fracture rates, Table 5 shows the proportion of women with spine, non-spine, and hip fractures during FIT among three groups: PBO-treated women, ALN-treated women with <30% reduction in bone ALP, and ALN-treated women with at least a 30% reduction in bone ALP. The probability of non-spine fracture in the placebo-treated group was 9.8% compared with 8.7% in ALN-treated women with <30% reduction in bone ALP and 6.8% in ALN-treated women with at least a 30% reduction in bone ALP. Similarly, 1.0% of PBO-treated women suffered a hip fracture during FIT, whereas in the ALN-treated group, the probability of hip fracture was 0.8% among those without a 30% reduction in bone ALP and 0.2% among those with at least a 30% reduction in bone ALP.

DISCUSSION

The FIT is the largest placebo-controlled study to date with concurrent measurements of bone mass, bone turnover, and fracture outcomes. We measured two serum markers of bone formation (bone ALP and P1NP) and one serum marker of bone resorption (S-CTX) and observed substantial reductions after 1 year of ALN therapy. During nearly 4 years of follow-up, we found that ALN-treated women with greater 1-year reductions in one or more markers of bone turnover had fewer subsequent spine, hip, and other non-spine fractures. Adjustment for baseline BMD had little effect on these relationships. These results reinforce the importance of bone turnover and highlight the mechanism by which bisphosphonates reduce fracture risk. These results further suggest that serial measurement of markers of bone turnover, particularly bone ALP, could provide useful clinical information to help guide bisphosphonate therapy.

The relationships between 1-year change in all three markers of bone turnover and incident spine fracture were all statistically significant. One-year change in bone ALP was also significantly associated with subsequent non-spine and hip fractures, but similar analyses with change in P1NP and S-CTX did not reach statistical significance. Our S-CTX results must be interpreted with caution because recent studies have found that serum S-CTX levels follow a dramatic circadian rhythm that is blunted by fasting.^{30, 31} In this study, 80% of the baseline samples and virtually all follow-up specimens were obtained in the nonfasting state. Analyses restricted to the subset who were fasting at baseline were similar to those observed in the overall cohort, but the interpretation of these results is limited by the small sample size and lack of fasting samples after baseline.

To provide some guidance on the clinical use of serial marker measurements among ALN-treated women and the use for an individual patient, we calculated risk ratios and the absolute risk of fracture for several specific changes in bone ALP and P1NP cutpoints. These results indicated that change in marker was significantly associated with the probability of fracture. For example, compared with ALN-treated women with a <30% reduction in serial bone ALP measurements, the risk of hip fracture was reduced 74% among those with bone ALP reductions of at least 30%. Furthermore, our analyses found that, compared with ALN-treated women with at least a 30% reduction in bone ALP, the risk of hip and non-spine fracture among those with a <30% reduction approached the average observed in the PBO group. However, there was little difference in spine fracture incidence between the bone ALP cutpoint categories in the ALN group, and both categories had a lower incidence than the PBO group. Additional studies of bisphosphonate-treated women with suboptimal reductions in bone markers are needed to confirm these findings and to determine if subsequent interventions, such as reassessing compliance, altering the dose, or changing to another therapy, result in fewer fractures.

Previous studies have examined the relationship between change in marker and fracture risk in raloxifene-¹⁷ and risedronate-¹⁸ treated women. Bjarnason et al.¹⁷ found that, among raloxifene-treated women enrolled in the Multiple Outcomes of Raloxifene Evaluation (MORE), greater 1-year reductions in bone ALP and osteocalcin (a marker of bone formation) were associated with fewer incident vertebral fractures. For example, among raloxifene-treated women, each SD reduction in the 1-year change in bone ALP was associated with a 25% reduction in vertebral fracture (CI: 8%, 38%). These analyses were performed in a subset of raloxifene-treated subjects, and other fracture outcomes were not reported. Eastell et al.¹⁸ found that, among risedronate-treated women, greater reductions in bone turnover were associated with fewer spine and non-spine fractures, but non-spine fractures occurring before the follow-up marker measurement were not excluded. As acute fractures are known to increase markers of bone turnover, at least temporarily,¹⁵ the inclusion of such fractures in the risedronate analysis may have overestimated the use of serial marker measurements. The relationship between change in marker and hip fracture risk was not reported in that study. Neither the raloxifene nor the risedronate study reported the risk of fracture using specific change in marker cutpoints.

In addition to providing information about hip fracture risk and excluding hip and non-spine fractures before the follow-up marker measurement, our study had several other important strengths. Fracture outcomes were carefully collected and adjudicated in FIT, and we used markers of bone turnover that are widely available. The large size of our study allowed us to analyze clinically important subgroups, such as women with osteoporosis at baseline. Despite these strengths, our study did have several limitations. As noted above, the majority of serum specimens were collected in a non-fasting state, and recent studies have found that fasting significantly reduces circadian variability of serum CTX levels. Other studies have found that nonfasting serum CTX levels may be useful if collected in the early afternoon,³² but the time of collection was not available in our study. Storage of archived serum at −20°C may have adversely affected marker levels. Archived urine specimens were not available in the full cohort, and we were not able to analyze urine markers of bone resorption. After the second annual visit, the alendronate dose in FIT was increased from 5 to 10 mg/day, and the effect of this change on our results is unknown. Our results suggest that serial marker measurements may identify ALN-treated women who are likely to have a suboptimal response, but further studies are needed to determine the reasons for such a response and to evaluate whether interventions can further reduce fracture risk. Last, it is not practical to design a randomized clinical trial to prospectively determine the relationship between change in on-treatment variables and on-treatment outcomes. Our within-treatment group analyses have the same limitations of observational studies, including unmeasured confounding.

In analyses among ALN-treated women conducted in parallel with those for bone markers, we found that change in spine BMD was not related to the subsequent risk of spine, hip, or non-spine fractures, whereas change in hip BMD was associated with the risk of spine fracture. These results are consistent with previous analyses from the FIT that examined change in hip BMD and incident vertebral fractures.³³ Recent meta-analyses suggest that the treatment efficacy of antiresorptive therapy, including bisphosphonates, is not fully accounted for by change in BMD.^8-10 Others have hypothesized that reduced bone turnover is critically important during antiresorptive therapy.³⁴ Our results suggest that, among alendronate-treated women, short-term changes in bone turnover are as important, and perhaps more important, than changes in BMD.

Although not a primary aim of our study, we found no consistent association between the baseline levels of bone turnover and subsequent fractures in the ALN or PBO groups. Some, but not all, previous cohort studies¹⁵ have found that elevated levels of bone turnover, particularly markers of bone resorption, are associated with an increased risk of fracture. Two clinical trials have previously reported the relationship between baseline turnover and subsequent vertebral fracture among PBO-treated women; one trial found no relationship,³⁵ whereas the other found that elevated resorption markers were associated with an increased risk spine fracture after 1 year, but not after 3 years, of follow-up.³⁶ Differences in the population studied and the effects of calcium supplementation in the PBO group may account for these observed differences.

In summary, in this large placebo-controlled trial among postmenopausal women, we found that greater 1-year reductions in one or more biochemical markers in response to ALN treatment were associated with a lower risk of spine, hip, and non-spine fracture. These relationships were most consistent for bone ALP, and to a lesser extent, for P1NP and S-CTX. Furthermore, we identified clinically meaningful marker cutpoints, such as at least a 30% reduction in bone ALP, that identify a group of ALN-treated women with a 28% lower risk of non-spine fracture and a 74% lower risk of hip fracture. Serial measurements of bone turnover may identify ALN-treated women with suboptimal fracture protection, but further studies are needed to determine what interventions will further reduce fracture risk in such individuals.

Acknowledgements

This study was funded by Merck Research Laboratories.