This study was previously presented in abstract form at the 23rd Annual Meeting of the American Society for Bone and Mineral Research, Phoenix, Arizona, USA, October 12–16, 2001
Changes in the level of biochemical markers of bone resorption with risedronate treatment for osteoporosis were examined as a surrogate for the decrease in fracture risk. Greater decreases in bone resorption markers were associated with greater decreases in vertebral (and nonvertebral) fractures.
Antifracture efficacy of antiresorptive therapies is only partially explained by increases in bone mineral density. Early decreases in bone resorption may also play a role. We tested this hypothesis by measuring two bone resorption markers, the C-telopeptide of type I collagen (CTX) and the N-telopeptide of type I collagen (NTX), in osteoporotic patients in risedronate vertebral fracture trials. We studied 693 women with at least one vertebral deformity (mean age, 69 ± 7 years) who received calcium (and vitamin D if required) and placebo or risedronate 5 mg daily for 3 years. The reductions in urinary CTX (median, 60%) and NTX (51%) at 3-6 months with risedronate therapy were significantly associated (p < 0.05) with the reduction in vertebral fracture risk (75% over 1 year and 50% over 3 years). The changes in both CTX and NTX accounted for approximately one-half (CTX, 55%; NTX, 49%) of risedronate's effect in reducing the risk of vertebral fractures in the first year and approximately two-thirds (CTX, 67%; NTX, 66%) over 3 years compared with placebo. The changes in CTX and NTX accounted for 77% and 54%, respectively, of risedronate's effect in reducing the risk of nonvertebral fractures over 3 years compared with placebo. The relationships between vertebral fracture risk and changes from baseline in CTX and NTX were not linear (p < 0.05). There was little further improvement in fracture benefit below a decrease of 55-60% for CTX and 35-40% for NTX. The decrease in bone resorption in patients taking risedronate accounts for a large proportion of the reduction in fracture risk. There may be a level of bone resorption reduction below which there is no further fracture benefit.
ANTIRESORPTIVE THERAPIES MAY reduce fracture risk by increasing bone mineral density (BMD), decreasing bone turnover, changing bone microarchitecture, or all of these effects.(1) Although one study found that an increase in hip BMD with alendronate was associated with a decreased risk of vertebral fractures,(2) only a weak relationship between changes in BMD and fracture risk reduction has been reported by others.(3, 4) It has been estimated that change in BMD explains only 4–28% of the reduction in vertebral fracture risk observed with antiresorptive agents.(5–7)
Risedronate reduces the risk of vertebral fracture in the first year of treatment.(8, 9) This effect occurs too rapidly to be solely a result of BMD changes, which are maximal by about the third year. A reduction in bone turnover, as reflected in the biochemical markers, may better explain the early fracture response, because the decrease in bone resorption is maximal at about 3 months.
High levels of bone turnover are associated with increased fracture risk in postmenopausal women.(10–12) Reductions in bone turnover are associated with a reduction in vertebral fracture risk in women treated with hormone replacement therapy(13) and raloxifene.(14) In patients treated with bisphosphonates, the early decrease in bone turnover is related to the long-term increase in BMD,(15) but a potential association with vertebral fracture risk has not been tested.
We have examined the relationship between changes in bone resorption markers and vertebral fracture risk reduction in two randomized trials of the effects of risedronate on vertebral fracture (the VERT studies).(8, 9) These studies included postmenopausal women with severe osteoporosis who had a high rate of incident vertebral deformities and who received either placebo or risedronate 5 mg daily for 3 years.
The objective of our work was to examine the relationship between change in bone resorption (as measured by changes in biochemical markers) and vertebral fracture risk reduction with risedronate. We also examined the relationship between a number of baseline characteristics, including the level of bone turnover markers, and subsequent risk of vertebral fracture.
MATERIALS AND METHODS
In the VERT studies, 2442 postmenopausal women (more than 3 years since menopause and less than 85 years of age) with two vertebral fractures or one vertebral fracture and a low BMD T-score (less than −2) were randomized to receive placebo or risedronate 5 mg daily for 3 years. They were recruited from 80 centers in Europe and Australia and 110 sites in North America. All subjects also received elemental calcium (1000 mg/day), and up to 500 IU/day vitamin D if baseline levels were low. Detailed descriptions of the study populations and results can be found elsewhere.(8, 9)
A subset of this population agreed to take part in a bone marker protocol; 693 women had both one baseline and one or more follow-up bone marker measurements and were included in the present analysis.
Sample collection and laboratory analyses
Morning second void urine samples were collected at baseline, 3 months, and 6 months, and stored at −20°C until analyzed. All samples from an individual were analyzed in one batch. Cross-linked N-telopeptides of type I collagen (NTX) were measured by automated analyser (Vitros ECi; Ortho Clinical Inc., Rochester, NY, USA) (interassay CV < 6.7%). Cross-linked C-telopeptides of type I collagen (CTX) were measured using an automated analyser (Enzymun ES 600; Roche Diagnostics, Penzberg, Germany) (interassay CV < 4.9%). Measurements were corrected for creatinine using a dry chemistry method. The 3- to 6-month value used in the analyses was the average of the measurements taken at 3 and 6 months.
BMD at the lumbar spine (L1-L4, LS BMD) and femoral neck (FN BMD) was measured by DXA using Hologic and Lunar densitometers (Hologic, Bedford, MA, USA, and Lunar Inc., Madison, WI, USA). Measurements were calculated as standardized BMD using published equations,(16) and the T-scores for femoral neck BMD were based on data from the Third National Health and Nutrition Survey (NHANES III) survey.(17)
Identification of fractures
Lateral thoracolumbar (T4-L4) radiographs were obtained at baseline and annually during the studies. Prevalent and incident vertebral fractures were diagnosed quantitatively and semiquantitatively, as previously described.(8, 9, 18, 19)
Incident nonvertebral osteoporosis-related fractures (prospectively defined as clavicle, hip, humerus, leg, pelvis, and wrist) were collected as adverse events and confirmed by radiograph.
The primary population of interest was the subset of 693 patients who had both radiographic assessments for vertebral fracture and baseline and follow-up urinary bone marker measurements. Demographic and baseline characteristics were summarized across treatment groups using descriptive statistics. Baseline lumbar spine data were summarized using the gender-specific T-score for patients with all four vertebrae (L1-L4) deemed intact. Post-baseline data were summarized within and between treatment groups. Percent change from baseline at 3–6 months for urine bone resorption markers were summarized using nonparametric statistics (median, interquartile range, and Wilcoxon signed-rank and rank sum tests) because the data were not normally distributed. The incidence of new vertebral fractures was estimated using time-to-first fracture methodology, consistent with previous analyses.(8, 9)
Cox regression was used to explore the relationship between fracture incidence and selected baseline measures. Univariate models were constructed to examine the simple relation between baseline measures and fracture incidence, ignoring all the other measures. A multiple regression model was consequently constructed, comprising baseline measures that were statistically significantly associated (p < 0.05) with fracture incidence.
To visualize the association between fracture incidence and early changes in bone turnover makers, the probability of sustaining a fracture was plotted against the 3- to 6-month bone turnover maker data. Empirical displays of the incidence were constructed using a smoothing curve. Because these displays were not model-dependent, no confidence intervals were constructed. Cox regression polynomial models were formed to compare the fit of the data when using linear, quadratic, and cubic functions. These models were statistically compared using the likelihood-ratio κ2 test.
A method based on the Cox regression model proposed by Li et al.(7) was used to estimate the overall treatment effect in reducing the risk of incident fracture and the effect explained by bone resorption within a single model. The time-to-first incident fracture was modeled with treatment group and bone resorption marker as covariates. As a contrast, the model proposed by Freedman et al.(20) was also used. For this methodology, the estimate of the treatment effect explained by bone resorption and the overall fracture risk reduction was obtained in two different models. Adjustment for prognostically important baseline risk factors (e.g., age, baseline BMD, and prevalent vertebral fractures) was performed and did not change the results of the main model.
The relationship between levels of CTX at 3–6 months and risk of vertebral fracture was examined using a T-score, which was calculated using reference data from 159 premenopausal women. Reference values were transformed (on a square root scale for CTX and a log scale for NTX) to obtain a normal distribution. Month 3–6 values were transformed, and a T-score was calculated for each patient.
Baseline characteristics and changes in bone resorption markers
The baseline characteristics were similar in the placebo and risedronate groups (Table 1). The risedronate group experienced a reduction in vertebral fracture risk of 75% at 1 year (p = 0.001, 33 and 9 women with fractures in placebo and risedronate groups, respectively) and 50% at 3 years (p = 0.002, 64 and 36 women with fractures in placebo group and risedronate groups, respectively); this magnitude of effect was similar to that reported elsewhere for the whole group.(8, 9) Bone resorption markers decreased in response to calcium in the placebo group (median decreases, 22% and 18% for NTX and CTX, respectively), but greater decreases (p = 0.001, Wilcoxon signed rank test) were observed in the risedronate group (51% and 60% for NTX and CTX, respectively)
Table Table 1.. Baseline Characteristics of the Subjects for Whom Bone Resorption Markers Were Available and of the Overall Study Population
Relationship between baseline characteristics and the risk of incident fracture
A number of baseline characteristics (levels of CTX and NTX, LS BMD and FN BMD, and the number of prevalent vertebral fractures) were significantly related to the subsequent risk of vertebral fracture in the placebo and risedronate groups at both 1 and 3 years (Table 2; Cox regression model adjusted for trial and treatment group). However, the relationship between vertebral fracture risk and bone resorption markers was no longer significant once either BMD or the number of prevalent vertebral fractures was entered into the model. This finding may be a result of associations between baseline bone resorption markers and BMD (NTX and CTX vs. lumbar spine BMD, r = −0.21 and −0.20; p < 0.001) and between baseline bone resorption markers and number of prevalent vertebral fractures (NTX and CTX, r = 0.19 and 0.09, p < 0.001 and p = 0.01, respectively).
Table Table 2.. Relationships (p Values) Between Baseline Variables and Fracture Incidence: Cox Regression Model Was Adjusted for Trial and Treatment Group
Relationship between change in bone resorption markers and the risk of incident fracture
Changes in NTX and CTX (at 3–6 months) were each statistically significantly associated with both 1- and 3-year vertebral fracture incidence (p < 0.001 for each). For 3-year nonvertebral osteoporosis-related fracture incidence, a similar association was observed (NTX: p = 0.031 and CTX: p = 0.027). Nonlinear functions were more appropriate than linear functions for modeling the relationship between 3- and 6-month changes in resorption markers and vertebral fracture incidence (NTX: p = 0.022 and CTX: p = 0.003). Below a certain value (a 35–40% decrease in NTX and a 55–60% decrease in CTX), there was no further decrease in vertebral fracture risk associated with further decreases in bone resorption markers (Fig. 1).
The change in bone resorption markers explained more than one-half of the treatment effect on vertebral fractures. The proportion of the vertebral fracture treatment effect explained by changes in bone resorption markers at 3–6 months for NTX and CTX was 49% (95% CI, 29–84%) and 55% (95% CI, 33–92%) at 1 year and 66% (95% CI, 24–100%) and 67% (95% CI, 35–100%) at 3 years, as determined by the method of Li et al.(7) The proportion of the vertebral fracture treatment effect explained by changes in bone resorption markers at 3–6 months for NTX and CTX was 29–34% at 1 year and 59–62% at 3 years, as determined by the method of Freedman et al.(20)Thus, the size of the estimates using the two methods was similar, but for the Freedman analysis, the 95% CIs were broad (0–100%).
The changes in bone resorption markers NTX and CTX explained between 54% and 77%, respectively, of the treatment effect on nonvertebral osteoporosis-related fractures over 3 years, similar in magnitude to that observed on vertebral fractures. Nonlinear functions were not more appropriate than linear functions in this model.
Calcium and vitamin D supplementation resulted in suppression in bone resorption markers. Furthermore, the greater the reduction in the bone resorption markers, the greater the reduction in the risk of vertebral fractures, and when analyzed alone, this association was significant (p < 0.01).
To further examine the relationship between levels of bone resorption and vertebral fracture risk, we compared the NTX and CTX T-scores and the incidence of vertebral fracture to determine the level of bone resorption associated with the lowest fracture risk. For vertebral fractures, the lowest fracture risk was associated with T-scores of −0.5 or below for CTX (3.2 nmol/nmol) and −1.5 or below for NTX (20.8 nmol BCE/mmol; Fig. 2). The relationship between T-score and vertebral fracture incidence was best described by a nonlinear rather than linear function (NTX, p = 0.047; CTX, p = 0.014).
Our most important finding was the strong relationship between change in bone resorption markers and fracture risk on treatment. Thus, more than 50% of the risedronate-related fracture risk reduction was explained by the change in these markers. This observation would be consistent with the concept that high turnover is associated with an increased fracture risk. The mechanism for this could include the increased risk of microcracks in high-turnover bone as a result of the bone remodeling sites acting as stress risers and the deterioration of the microarchitecture of bone because of plate perforation(21); both of these effects might be prevented by using risedronate treatment to reduce the level of turnover.
Li et al.(7) have reported that the change in spine BMD explains about 28% of the reduction in vertebral fracture risk. The study of Li et al.(7) included the patients in this study (we did examine the relationship between change in BMD and fracture risk reduction and this was less than 10%; however the numbers for this analysis are too small to obtain a meaningful estimate). Thus, change in bone resorption markers are at least as good a surrogate for the fracture benefit of risedronate as BMD. It may be that BMD is misleading because it reflects both the reduction in bone turnover (and filling in of remodeling space) and the degree of secondary mineralization of bone.(22) The latter may have no beneficial effect on bone strength.
There seems to be a threshold for decrease in bone resorption (e.g., 35–40% as measured by NTX and 55–60% as measured by CTX) below which there is no further increase in fracture benefit. This ceiling could be the result of a trade-off of the beneficial effect of fewer stress risers being balanced by an increased fragility from inability to repair microdamage. Dogs given high doses of bisphosphonates have a marked decrease in bone turnover that is associated with accumulation of mirocracks and a decrease in the toughness of bone.(23) However, no threshold could be identified for nonvertebral fractures. This could represent a real difference between the relationship between change in bone resorption markers and these two types of fracture, or it could reflect the smaller numbers of patients with nonvertebral fractures (57) compared with incident vertebral fractures (100).
This study supports the use of bone turnover markers as suitable surrogates for vertebral fracture when monitoring the effects of antiresorptive therapy. This finding has implications both for dose ranging studies of future antiresorptive therapies and for the monitoring of the individual patient. For the latter, the least significant change approach has been used. This approach is statistically based and thresholds of 30–50% have been proposed as targets for response.(24) Our finding of no further benefit with further decreases in bone resorption below the level of 35–60% is consistent with that recommendation. Our analysis of vertebral fracture risk by CTX T-score indicates that patients with a CTX T-score higher than −0.5 are at highest risk and provides further guidance with regard to a target for treatment. Thus, the goal of treatment could be a decrease in bone resorption to a low T-score value, although we could not test this as the two estimates, percentage reduction and absolute values on treatment, are not independent.
Automated devices can measure bone turnover markers reliably; these tests are low in cost and widely available. Because of natural day-to-day variability in levels, the reliability of the measurements can be increased, when feasible, by obtaining duplicate measurements before and after start of treatment and by reducing other known sources of variability, such as sampling time and fasting state.(25)
There are limitations to this study. The confidence intervals of our estimates are broad because of the relatively small samples size and the variability of bone resorption markers. The confidence intervals from the Freedman et al.20 analysis are wider than those obtained using the methods of Li et al.7; these wider intervals are probably attributable to the larger variability generated when fitting two separate statistical models compared with one model. We cannot be certain that the lines are not straight, although the statistical significance of the quadratic term supports a nonlinear relationship. Finally, the analyses did not adjust for change in BMD.
We conclude that (1) the baseline level of bone resorption is related to subsequent fracture risk, (2) the reduction in bone resorption explains, in part, the reduction in the risk of fractures with risedronate, and (3) there is a level of bone turnover reduction below which no further fracture benefit is observed.
We would like to acknowledge the help of Dr Simon Pack and Lisa Bosch of Procter & Gamble Pharmaceuticals, as well as the help of Oldham Hospital Clinical Chemistry Department for measuring urinary CTX and creatinine. This study was supported by grants from Procter & Gamble Pharmaceuticals, Inc. (Cincinnati, OH) and Aventis Pharma, Bridgewater, NJ. Employees of Procter & Gamble Pharmaceuticals and Aventis Pharma participated in the design and execution of the study, the analysis of the data, and the preparation of the manuscript. All authors had full access to the data and analyses.