Abstract
Increases in lumbar spine BMD account for 30–41% of the vertebral fracture risk reduction with teriparatide treatment. The remaining fracture risk reduction is caused by improvements in non‐BMD determinants of bone strength.
Introduction: Changes in BMD account for a small percentage of the fracture risk reduction seen in patients treated with antiresorptive drugs. The relationship between changes in lumbar spine BMD and vertebral fracture risk reduction with teriparatide treatment has not been assessed.
Materials and Methods: The relationship between spine BMD and the risk of new vertebral fractures after teriparatide treatment was assessed using data from the Fracture Prevention Trial. Postmenopausal women with osteoporosis (n = 1637) were randomized to placebo or teriparatide 20 or 40 μg/day for a median of 19 months. Spine BMD was assessed at baseline and 18 months. Vertebrae whose fracture status changed during the trial were removed from the calculation of BMD. Baseline and endpoint lateral spine radiographs were assessed using a visual semiquantitative technique.
Results: Both the baseline and change in spine BMD were contributors to vertebral fracture risk. The mean spine BMD increase in teriparatide‐treated patients was 0.09 g/cm2 across tertiles of baseline spine BMD. Compared with placebo, teriparatide significantly reduced the risk of new vertebral fracture for all endpoint BMD values. Teriparatide‐mediated increases in spine BMD accounted for 30% (in the low baseline spine BMD tertile) to 41% (in the high baseline spine BMD tertile) of the reduction in vertebral fracture risk.
Conclusions: Increases in BMD account for approximately one third of the vertebral fracture risk reduction seen with teriparatide. The majority of the risk reduction, however, results from improvements in non‐BMD determinants of bone strength.
INTRODUCTION
The primary goal of osteoporosis therapy is to reduce fracture risk, but because fractures occur infrequently, clinicians must rely on surrogates to assess response to therapy.(1) The most widely used surrogate is BMD. BMD measured by DXA offers sufficient precision to follow disease progression(2) and is the best‐validated technology for showing a response to therapy.(3)
Low BMD is consistently correlated with increased fracture risk; however, antiresorptive‐mediated increases in BMD seem to account for a small percentage of the fracture risk reduction based on individual patient data.(4–7) Cummings et al.(6) attributed 16% of alendronate‐mediated vertebral fracture risk reduction to increases in lumbar spine BMD. Similarly, Watts et al.(7) concluded that 18% of the vertebral fracture risk reduction with risedronate could be attributed to increases in spine BMD, whereas Li et al.(5) placed this value at 28%. Sarkar et al.(4) found that only 4% of the fracture risk reduction seen with raloxifene was attributable to changes in femoral neck BMD. Importantly, the statistical methodologies used by Cummings et al. and Sarkar et al. differed from those used by Li et al. and Watts et al. The former analyses used BMD change as a covariate for fracture risk, whereas the latter used the BMD endpoint value as a covariate. Most recently, Watts et al.(8) reported that 12% and 7% of the effect of risedronate on reducing nonvertebral fractures was attributable to changes in spine and femoral neck BMD, respectively. This latter analysis also used the methodology of Li et al. In contrast, meta‐analyses relying on group means rather than individual patient data have suggested that a large percentage of the antifracture efficacy observed with antiresorptive drugs can be explained by changes in BMD.(9,10) However, because use of group means dampens inherent response variability, individual patient data provides the most comprehensive information and thus should be used when exploring the relationship between BMD change and fracture risk.(11–13)
Teriparatide is a bone formation agent that has been shown to significantly increase spine BMD and reduce vertebral fracture risk(14) through a mechanism of action on bone remodeling that is essentially opposite from that of bisphosphonates.(15) In this analysis, we assessed the vertebral fracture risk reduction attributable to teriparatide‐mediated changes in spine BMD. To accomplish this, we explored whether percent change in BMD or the absolute endpoint spine BMD is the appropriate covariate for fracture risk. First, we examined the relationship between percent change in spine BMD and vertebral fracture risk. We examined the effect of baseline BMD on the relationship between percent change in spine BMD and vertebral fracture risk. Finally, we examined the relationship between the absolute endpoint spine BMD value—a combination of baseline spine BMD and postbaseline spine BMD increases—and vertebral fracture risk.
MATERIALS AND METHODS
Study participants
These posthoc analyses were conducted using data from postmenopausal women with osteoporosis who participated in the Fracture Prevention Trial. Methods and results of this trial have been previously published.(14) Briefly, 1637 ambulatory, postmenopausal women ranging in age from 42 to 86 years were randomized in a double‐blinded manner to receive daily, self‐administered, subcutaneous injections of placebo (n = 544), teriparatide 20 μg/day (n = 541), or teriparatide 40 μg/day (n = 552) with daily oral calcium (1000 mg) and vitamin D (400–1200 IU) supplementation. The median duration of exposure to teriparatide was 19 months, and the median duration of observation was 21 months.
Lumbar spine BMD assessment
In the Fracture Prevention Trial, BMD was assessed by DXA using Hologic (Hologic Corp., Bedford, MA, USA), Norland (Norland Corp., Ft Atkinson, WI, USA), and Lunar equipment (Lunar Corp., Madison, WI, USA). Measurements were performed at baseline, 12 months, 18 months, and study endpoint. To eliminate differences attributable to densitometer manufacturer, spine BMD values were converted to standardized units (mg/cm2).(16) Vertebrae whose fracture status changed during the trial as assessed by lateral spine radiography were removed from the calculation of both baseline and follow‐up lumbar spine BMD. At least two evaluable vertebrae were required for assessment of spine BMD; this was calculated from the sum of the BMCs and the sum of the areas of all evaluable spine vertebrae.
Lateral radiography and fracture assessment
Lateral spine radiographs were performed at baseline and study endpoint. Radiographs were assessed at a central site (Osteoporosis and Arthritis Research Center, University of California San Francisco, San Francisco, CA, USA) by expert radiologists blinded to treatment groups. The radiographic assessment was performed using a visual semiquantitative technique developed by Genant et al.,(17) whereby a new fracture was reported if a greater than ∼20% height loss was observed in a vertebrae that was not fractured at baseline.
Statistical analysis
Subjects who had spine BMD measurements at baseline and 18 months and for whom evaluable lateral spine radiographs were available at baseline and study endpoint were included. Changes in spine BMD were calculated after imputing missing values by the last observation carried forward method. Because of a limited number of new vertebral fractures, data from the teriparatide 20 and 40 μg/day treatment groups were pooled. The relationship between the percent change in spine BMD and the risk of new vertebral fracture was tested by logistic regression. The presence of any new vertebral fracture was considered binomial (yes/no), and spine BMD variables (either percent change or endpoint BMD) were continuous covariates. In these models, treatment, BMD variables, and the treatment effect‐by‐BMD variables interaction were fixed effects. The interaction was tested at the 10% significance level. The models estimate the absolute risk of a new vertebral fracture for each treatment group over the range of spine BMD values. The relationships between baseline spine BMD and the percent change in spine BMD were evaluated using Spearman correlation. To evaluate the effect of baseline spine BMD on the relationship between percent change in BMD and vertebral fracture risk, logistic regression analysis was used to model vertebral fracture risk as a function of baseline and percent change in BMD for each treatment group. To further support the use of endpoint spine BMD as opposed to the change in BMD, we stratified the placebo group into subjects with lumbar spine BMD changes greater than or equal to the median change and those with BMD changes less than the median change. We assessed the difference in vertebral fracture risk between the two subgroups given the same endpoint spine BMD.
The percentage of the treatment effect of teriparatide (defined as the difference between the teriparatide and placebo groups) on vertebral fracture risk explained by changes in spine BMD was estimated using a modified method as proposed by Li et al.(5) Specifically, the analysis by Li et al. used ORs, whereas we used absolute risk reduction because this is how fracture risk is usually expressed. Because the absolute fracture risk was relatively low, the results of the two methods would be similar. The overall absolute fracture risk reduction caused by teriparatide treatment and the vertebral fracture risk reduction caused by spine BMD increases alone were estimated within one model using logistic regression.
RESULTS
Demographics
The baseline characteristics of the placebo and pooled teriparatide groups were not significantly different (Table 1).
Relationship between percent change in BMD and fracture risk
We first explored the simple relationship between change in spine BMD and fracture risk in the teriparatide and placebo groups (Fig. 1A), using the approach of Sarkar et al.(4) and Cummings et al.(6) Unexpectedly, this relationship differed markedly in the two groups, with the interaction between treatment and percent change in spine BMD being statistically significant (p = 0.01). In the placebo group, the fracture risk was inversely related to change in spine BMD, whereas in the teriparatide group, the fracture risk was positively associated with the percent change in spine BMD (p = 0.02). To explore this unexpected and unintuitive finding in the teriparatide group, we examined the relationship between the baseline spine BMD and the percent change in spine BMD and found that those subjects with lower baseline spine BMD had larger percent increases in BMD (Fig. 1B). The two variables were inversely correlated (r = −0.36; p < 0.01). Thus, in the teriparatide group, the impact of percent change in spine BMD on fracture risk was confounded by those patients with greater increases in spine BMD having lower baseline spine BMD. Subsequent analyses explored the relationship between these key variables—baseline BMD and change in BMD—with fracture risk.
(A) Relationships between percent changes in spine BMD and vertebral fracture risk. Data represents the 5th to 95th percentile changes in lumbar spine BMD for each treatment group. (B) Scatterplot of baseline spine BMD vs. percent change in spine BMD for subjects treated with TPTD. TPTD20 + 40, pooled teriparatide 20 and 40 μg/day groups.
Relationship between baseline BMD and change in BMD with fracture risk
In the placebo group, both the baseline spine BMD and the change in spine BMD were important determinants of vertebral fracture risk (Fig. 2A), with greater baseline spine BMD and greater increases in spine BMD associated with decreased risk. The importance of the baseline BMD as a determinant of fracture risk may be shown using an example. Subjects with a large increase in spine BMD but low baseline spine BMD (Fig. 2A, position A) had greater fracture risk than subjects with no increase in spine BMD but a high baseline spine BMD (Fig. 2A, position B). In the teriparatide group, neither baseline spine BMD nor the change in spine BMD was a significant determinant of vertebral fracture risk (Fig. 2B). In the median change analysis, vertebral fracture risk in subjects with lumbar spine BMD changes greater than or equal to the median change was similar to that in subjects with BMD changes less than the median change given the same endpoint spine BMD. Therefore, subjects with “large” changes in BMD had similar fracture risks to subjects with “small” changes in BMD provided the endpoint BMD was the same. This confirms that both the baseline spine BMD and change in spine BMD are determinants of vertebral fracture risk. Subsequent analyses focused on endpoint BMD, a measure that integrates both the baseline spine BMD and the change in spine BMD.
(A) Relationship between baseline lumbar spine BMD and the 18‐month endpoint percent change in lumbar spine BMD with vertebral fracture risk for the placebo group. Position “A” represents vertebral fracture risk in a subject with a relatively low baseline BMD and a relatively large increase in BMD. Position “B” represents vertebral fracture risk in a woman with a relatively high baseline BMD and little or no increase in BMD. (B) The relationship between baseline lumbar spine BMD and the 18‐month endpoint percent change in lumbar spine BMD with vertebral fracture risk for the teriparatide‐treated group. TPTD20 + 40, pooled teriparatide 20 and 40 μg/day groups; LS, lumbar spine.
Relationship between absolute endpoint BMD and fracture risk
Compared with placebo, vertebral fracture risk was significantly lower in the teriparatide group for all endpoint lumbar spine BMD values (Fig. 3). The mean absolute spine BMD increase in the pooled teriparatide group was 0.09 g/cm2 across tertiles of baseline spine BMD values. The approach to determining the contribution of change in spine BMD to teriparatide antifracture efficacy is shown in the example shown on Fig. 3. Position A represents the fracture risk associated with an endpoint spine BMD of ∼0.83 g/cm2. Point B represents the vertebral fracture risk associated with an endpoint spine BMD of 0.92 g/cm2 (i.e., 0.09 g/cm2 increase in spine BMD from position A). Point C represents the vertebral fracture risk after a teriparatide‐mediated increase in spine BMD of 0.09 g/cm2 over the course of the study. The total teriparatide‐mediated risk reduction associated with a 0.09 g/cm2 increase in spine BMD can be estimated by subtracting the risk associated with position C from the risk associated with position A (i.e., A – C). The total risk reduction associated with a 0.09 g/cm2 increase in spine BMD alone can be estimated by subtracting the risk associated with position B from the risk associated with position A (i.e., A – B). Thus, the percentage of TPTD‐mediated fracture risk reduction attributable to the increase in BMD can be estimated by dividing (A − B) by (A − C) [i.e., (A − B)/(A − C)].
Relationship between endpoint lumbar spine BMD and vertebral fracture risk. A − B and A − C represent fracture risk reductions associated with a 0.09 g/cm2 increase in BMD alone and in response to teriparatide, respectively. The percentage of TPTD‐mediated fracture risk reduction attributable to the increase in BMD can be estimated by (A − B)/(A − C). Error bands represent 95% CIs. TPTD20 + 40, pooled teriparatide 20 and 40 μg/day groups.
The total absolute risk reductions and absolute risk reductions caused by increases in BMD for different A values representing baseline spine BMD are shown in Table 2. As baseline BMD values increased, both the total absolute risk reduction and absolute risk reduction caused by increases in BMD decreased; however, the percent of absolute risk reduction caused by positive gains in spine BMD increased. The percentage of teriparatide‐mediated vertebral fracture risk reduction attributable to increase in spine BMD ranged from 30% to 41%.
Increases in Lumbar Spine BMD and the Percent of Total Risk Reduction Caused by BMD Increases Across Baseline BMD Values in Response to TPTD20 + 40
Increases in Lumbar Spine BMD and the Percent of Total Risk Reduction Caused by BMD Increases Across Baseline BMD Values in Response to TPTD20 + 40
DISCUSSION
Teriparatide‐mediated increases in lumbar spine BMD accounted for 30–41% of the reduction in vertebral fracture risk. The balance of fracture risk reduction was therefore attributable to improvements in non‐BMD determinants of bone strength. Improvements in non‐BMD determinants of bone strength accounted for 70% of the fracture risk reduction in subjects with a baseline lumbar spine BMD in the lowest tertile. In subjects with a baseline BMD in the highest tertile, improvements in non‐BMD determinants of bone strength accounted for 59% of the fracture risk reduction. Non‐BMD determinants of bone strength include several factors such as bone geometry, microarchitecture, remodeling rate, damage accumulation, and collagen/mineral matrix properties.(18) Teriparatide positively affects both BMD and non‐BMD determinants of bone strength.(19–26)
Both baseline and changes in spine BMD were significant determinants of fracture risk. Teriparatide‐treated subjects with lower baseline spine BMD had larger percent increases than those with higher baseline BMD (i.e., baseline spine BMD and the percent change in BMD were inversely correlated). Therefore, the fracture risk reduction associated with percent change in spine BMD needs to be considered in the context of the baseline value. We resolved this complexity by using the methodology of Li et al.,(5) using absolute endpoint spine BMD as the critical covariate; this variable accounts for both the baseline and change in spine BMD.
In summary, our findings suggest that absolute endpoint spine BMD should be used as the covariate when assessing the relationship between spine BMD changes and vertebral fracture risk. In the Fracture Prevention Trial, the mean teriparatide‐mediated absolute spine BMD increase was 0.09 g/cm2. The percentage of teriparatide‐mediated vertebral fracture risk reduction attributable to this increase in spine BMD ranged from ∼30% to 41% in postmenopausal women with osteoporosis. The remaining fracture risk reduction was because of improvements in other non‐BMD determinants of bone strength.
Acknowledgements
The authors thank the investigators of the Fracture Prevention Trial. This study was supported by Eli Lilly and Company. Data were analyzed at Lilly Research Laboratories, Eli Lilly and Company.
REFERENCES
Author notes
Parts of this manuscript were presented as an oral presentation at the 27th Annual Meeting of the American Society for Bone and Mineral Research in Nashville, TN, September 23–27, 2005 and at the Fifth European Congress on Clinical and Economic Aspects of Osteoporosis and Osteoarthritis, Vienna, Austria, March 15–18, 2006.
Dr Chen has received corporate appointments from Eli Lilly and Company. Drs Krege and Misurski are employees of and own stock in Eli Lilly and Company. Dr Miller has received funding from Amgen, Aventis Pharmaceuticals, Eli Lilly and Company, Merck & Co., Novartis, Pfizer, Pharmacia, Procter & Gamble, and Roche. He has served as a speaker, on advisory boards, and as a consultant for Amgen, Aventis Pharmaceuticals, Eli Lilly and Company, GlaxoSmithKline, Merck & Co., Novartis, NPS Pharmaceuticals, Procter & Gamble, and Roche. Dr Delmas has served as a consultant and speaker for Amgen, Aventis Pharmaceuticals, Eli Lilly and company, GlaxoSmithKline, Novartis, Nycomed, Procter & Gamble, Roche, and Sanofi‐Aventis.
