Relationships Between Bone Mineral Density and Incident Vertebral Fracture Risk with Raloxifene Therapy

Authors

  • Somnath Sarkar Ph.D.,

    Corresponding author
    1. Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, Indiana, USA
    • Eli Lilly and Company Lilly Corporate Center Indianapolis, IN 46285, USA
    Search for more papers by this author
    • Dr. Somnath Sarkar, Dr. Bruce H. Mitlak, Dr. Mayme Wong, Dr. John L. Stock, and Dr. Kristine D. Harper have financial interests in the forms of corporate appointments and stock ownership in Eli Lilly and Company. Dr. Dennis M. Black does consultant work for Eli Lilly and Company.

  • Bruce H. Mitlak,

    1. Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, Indiana, USA
    Search for more papers by this author
  • Mayme Wong,

    1. Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, Indiana, USA
    Search for more papers by this author
  • John L. Stock,

    1. Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, Indiana, USA
    Search for more papers by this author
  • Dennis M. Black,

    1. Department of Epidemiology and Biostatistics, University of California San Francisco, San Francisco, California, USA
    Search for more papers by this author
  • Kristine D. Harper

    1. Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, Indiana, USA
    Search for more papers by this author

  • Portions of this work were presented in part at the meeting of the 1999 American Society for Bone and Mineral Research, 1999; the meeting of the World Congress of Osteoporosis, 2000; and the meeting of the American College of Rheumatology, 2000.

Abstract

Although low absolute values of bone mineral density (BMD) predict increased fracture risk in osteoporosis, it is not certain how well increases in BMD with antiresorptive therapy predict observed reductions in fracture risk. This work examines the relationships between changes in BMD after 1 year or 3 years of raloxifene or placebo therapy and the risk for new vertebral fractures at 3 years. In the Multiple Outcomes of Raloxifene Evaluation (MORE) trial, 7705 postmenopausal women with osteoporosis were randomized to placebo or raloxifene 60 mg/day or 120 mg/day. Relationships between baseline BMD and changes in BMD from baseline with the risk of new vertebral fractures were analyzed in this cohort using logistic regression models with the raloxifene doses pooled. As has been observed in other populations, women with the lowest baseline lumbar spine or femoral neck BMD in the MORE cohort had the greatest risk for vertebral fractures. Furthermore, for any percentage change, either increase or decrease in femoral neck or lumbar spine BMD at 1 year or 3 years, raloxifene-treated patients had a statistically significantly lower vertebral fracture risk compared with placebo-treated patients. The decrease in fracture risk with raloxifene was similar across the range of percentage change in femoral neck BMD observed at 3 years; patients receiving raloxifene had a 36% lower risk of vertebral fracture compared with those receiving placebo. At any percentage change in femoral neck and lumbar spine BMD observed at 1 year, raloxifene treatment decreased the risks of new vertebral fractures at 3 years by 38% and 41%, respectively. The logistic regression model showed that the percentage changes in BMD with raloxifene treatment accounted for 4% of the observed vertebral fracture risk reduction, and the other 96% of the risk reduction remains unexplained. The present data show that the measured BMD changes observed with raloxifene therapy are poor predictors of vertebral fracture risk reduction with raloxifene therapy.

INTRODUCTION

The worldHealth Organization's (WHO) definition of osteoporosis is based on the relationship between low bone mineral density (BMD) and “consequent increase in bone fragility and susceptibility to fracture.”(1) An increased risk of a new vertebral fracture in postmenopausal women is predicted by a low lumbar spine BMD,(2) an increase in age,(3) and the existence of prevalent vertebral fractures.(4, 5) Even after adjustment for age, both low baseline BMD and prevalent vertebral fractures are predictive of an increased risk of subsequent vertebral fractures.(3, 6) Baseline BMD is useful for determining if patients are candidates for therapeutic intervention(1) because it predicts the future risk of an osteoporotic fracture, the ultimate clinical outcome.

As the number of therapeutic choices available for prevention and treatment of osteoporosis has increased, there is a growing need to understand the relationship between treatment-induced changes in BMD and fracture risk reduction. However, it is not known how well treatment-induced changes in BMD are correlated with or are predictive of the reductions in fracture risk observed with therapy. Although increases in BMD resulting from antiresorptive therapies differ greatly in magnitude, reductions in vertebral fracture risk are quite similar. For example, in postmenopausal women without prevalent vertebral fractures, raloxifene, 60 mg/day for 3 years,(7) and alendronate, given for an average of 4.2 years,(8) increased lumbar spine BMD by 2.8% and 6.8%, respectively, compared with placebo. In these same studies, raloxifene, 60 mg/day, decreased the absolute risk of new vertebral fractures by 2.4%, and the relative risk by 55% (95% CI, 29% and 71%),(9) and alendronate decreased the absolute risk of new vertebral fractures by 1.7% and the relative risk by 44% (95% CI, 20% and 61%).(8) The relationship between changes in BMD and fracture risk may be quantitatively different for the various therapeutic agents evaluated, because these agents act to inhibit bone resorption through different mechanisms.(10)

For assessing the effects of a therapeutic intervention in patient management and to reduce the cost and length of clinical trials, biological markers often are used as surrogates for those actual clinical outcomes that are observed in long trials with a large number of subjects. The effect of a therapeutic intervention on a surrogate endpoint must reliably predict the effect on the actual clinical outcome in order for the surrogate to be considered an acceptable substitute.(11) It is uncertain whether monitoring the changes in BMD observed with antiresorptive therapy can be used as reliable surrogates for predicting fracture efficacy.(12–14) However, several reports suggest that the increases in BMD during antiresorptive therapy may be poor estimates of the resultant fracture risk reduction observed with bisphosphonates, estrogen, and raloxifene.(12–16)

The Multiple Outcomes of Raloxifene Evaluation (MORE) trial evaluated the effects of raloxifene, a selective estrogen receptor modulator, on the risk of new vertebral fractures and changes in BMD in 7705 postmenopausal women with osteoporosis.(7) Because of the large size of the MORE study cohort, it is possible to analyze the relationships between BMD and vertebral fracture risk. These analyses will determine if the relationships between baseline BMD and vertebral fracture risk in the MORE cohort are similar to those reported in other study populations. Other analyses will examine whether increases in BMD with raloxifene can predict vertebral fracture risk reduction and if vertebral fracture risk reduction observed with raloxifene can be attributed to changes in BMD. The primary overall objective was to determine whether changes in BMD could be used as possible surrogate endpoints to monitor clinical response to raloxifene therapy.

MATERIALS AND METHODS

The design of the 3-year MORE trial was described previously.(7) A total of 7705 women with osteoporosis, who were at least 2 years postmenopausal, were randomized to either placebo or raloxifene, 60 mg/day or 120 mg/day. All women received daily calcium (500 mg) and vitamin D (400-600 IU) supplements. Women were excluded if they had bone disease other than osteoporosis, substantial postmenopausal symptoms, abnormal uterine bleeding, or endometrial carcinoma. Additionally, women who had pathological fractures, women for whom satisfactory thoracic and lumbar spine radiographs could not be obtained, and women with fewer than two lumbar vertebrae that were evaluable were excluded. Complete inclusion and exclusion criteria and procedures for subject recruitment and follow-up were described previously.(7)

BMD was assessed in the femoral neck and lumbar spine at baseline and at yearly intervals using dual-energy X-ray absorptiometry (DXA). A central reading facility (Osteoporosis and Arthritis Research Group [OARG], University of California, San Francisco, CA, USA) performed quality assurance of the longitudinal and cross-calibration adjustments to the DXA machines and patients' BMD measurement data and provided correction factors to adjust for differences between sites and performance of densitometers over time. To assess vertebral fractures, spinal radiographs obtained at baseline, 2 years and 3 years, were analyzed using a combination of semiquantitative (SQ) readings and quantitative morphometric (QM) analysis. A fracture identified by SQ assessment was followed with another independent binary SQ assessment and a QM measurement. An adjudicated fracture was reported if it was confirmed by at least two of the three determinations. The criterion for diagnosis of a new vertebral fracture was based on a reduction in the anterior, middle, and/or posterior vertebral height of >20% and at least 4 mm, compared with the baseline radiograph. All vertebral radiographs were assessed at a central quality assurance center (OARG) by radiologists blinded to treatment group assignment but not to the temporal sequence of the radiographs. Because the BMD values obtained from fractured vertebrae would be artificially high, all previous BMD values from vertebrae with visually observable fractures on the DXA screen were excluded from the analyses.

Women were considered to have osteoporosis if they had a baseline femoral neck or lumbar spine BMD of >2.5 SD below the normal, peak bone mass of premenopausal women (T score < −2.5), or had at least one moderate or two mild prevalent vertebral fractures irrespective of baseline BMD. The results from both doses of raloxifene were pooled to increase the statistical power of the present analyses.

Baseline BMD and vertebral fracture risk

Using logistic regression analysis methods, we examined the relationship between baseline femoral neck and lumbar spine BMD and the risk of having at least one new vertebral fracture over 3 years. The presence of any incident vertebral fracture was considered a binomial (yes/no) response and baseline BMD was considered a continuous covariate. The model included treatment (pooled raloxifene and placebo), baseline BMD, and treatment effect-by-baseline BMD interaction as fixed effects. If the interaction term is significant, this indicates that the treatment effect (fracture risk) is dependent on the baseline BMD, whereas a nonsignificant interaction term indicates that the observed fracture risk is independent of baseline BMD. This model estimated the risk of a new vertebral fracture for each treatment group (pooled raloxifene or placebo) over the broad range of baseline BMD values. The absolute risk of a new vertebral fracture for the raloxifene and placebo groups and the relative risk between groups, along with the 95% CIs, were calculated from this model.

Percentage change in BMD and vertebral fracture risk

The relationships between the percentage change in lumbar spine or femoral neck BMD from baseline to the 1-year or 3-year endpoint and the risk of having at least one new vertebral fracture at 3 years were tested by logistic regression analysis. The presence of any new vertebral fracture was considered a binomial (yes/no) response and changes in BMD were considered as continuous covariates. In these models, treatment (pooled raloxifene and placebo), percentage change in BMD, and the treatment effect-by-percentage change in BMD interaction were fixed effects. The models estimated the risk of a new vertebral fracture for each treatment group (pooled raloxifene or placebo) over the broad range of percentage changes in BMD. Additionally, the models estimated the contribution of a change in BMD to the reduced risk of incident vertebral fractures observed with raloxifene (treatment effect). The absolute risk of a new vertebral fracture for the raloxifene and placebo groups and the relative risk between groups, along with the 95% CI, were calculated from the fitted models.

These analysis models also were used to examine the relationships between the risk of having at least one new vertebral fracture at 3 years with the absolute endpoint BMD values and the absolute BMD changes in the lumbar spine and femoral neck at 3 years.

Statistical analyses

A generalized linear modeling technique was used to relate the risk (binomial regression with logit link) of a new vertebral fracture with either baseline BMD or percentage change in BMD for each patient.(17) These logistic regression models included a treatment dummy variable (pooled raloxifene groups and placebo), with baseline BMD (or percentage change in BMD), as continuous variables. Changes in BMD from baseline to 3 years or from baseline to 1 year were calculated after imputing missing values by the last observation carried forward (LOCF) method. Although logistic regression models are used commonly to estimate odds ratios (ORs), this method can be used to calculate absolute and relative risks for vertebral fractures. The 95% CI for absolute and relative risks for vertebral fractures were calculated using the Mantel Haenszel large sample method. To quantify the amount of fracture risk reduction attributed to raloxifene that is explained by a change in BMD, the relative risk reduction adjusted for the change in BMD was compared with the actual observed relative risk reduction, that is, before adjustment for change in BMD. This method, similar to that of Freedman,(18) cannot be used effectively if there is a significant interaction between the percentage change in BMD and treatment. According to the study protocol, the interaction was tested at the 10% significance level in this model. A nonsignificant interaction term indicates that the effects of treatment and percentage change in BMD on the observed fracture risk are independent, whereas a significant interaction term indicates that the treatment effect (fracture risk) is dependent on the percentage change in BMD. In cases in which the interaction term was significant, the treatment effect (fracture risk) cannot be quantified as a single value.

In a graphical presentation of the relationship, the continuous values of fracture risks observed at the continuous percentage changes in BMD were plotted as separate curves with 95% CI for each treatment group. If the placebo and raloxifene curves with their respective 95% CI bands were superimposed on each other, it was inferred that the percentage change in BMD explained the observed reduction in fracture risk. On the other hand, if the placebo and raloxifene curves with their respective 95% CI bands did not overlap in the graph, it was inferred that the percentage change in BMD did not explain the decrease in fracture risk observed with raloxifene therapy. From these graphs, the percentage of fracture risk reduction between the raloxifene and placebo groups was estimated at the lower 25th and median 50th percentiles and the upper 75th percentile of changes in BMD and these discrete data were shown separately in tables.

RESULTS

The baseline characteristics of the women in the MORE trial were previously reported in detail(7) and summarized briefly in Table 1. The women in the study population had a mean age of 66.5 years, a mean body mass index (BMI) of 25.2 kg/m2, and were a mean of 18.7 years postmenopausal. Prevalent vertebral fractures were present in 37.3% of the total study population. At randomization, the lumbar spine and femoral neck BMD were 0.82 ± 0.13 g/cm2 and 0.62 ± 0.08 g/cm2 (mean ± SD), respectively. There were no statistically significant differences in the baseline characteristics between the placebo and the raloxifene groups.

Table Table 1. Baseline Characteristics of 6828 Postmenopausal Women With Osteoporosis in the MORE Triala
original image

Percentage changes in BMD at the lumbar spine or femoral neck compared with placebo were not significantly different between the two doses of raloxifene at any time point in the MORE study. The incidence of new vertebral fractures and the relative risk of a new vertebral fracture compared with placebo were not significantly different between the raloxifene groups at 3 years.

Baseline BMD

Women in the MORE cohort with the lowest baseline femoral neck or lumbar spine BMD had the greatest risk of a new vertebral fracture (Fig. 1). Baseline BMD measurements at both the femoral neck and the lumbar spine were significantly (p < 0.001) related to the risk of a new vertebral fracture at 3 years, as determined by logistic regression analysis (Fig. 1). Among women in the placebo group, a 1 SD decrease in baseline femoral neck BMD and in baseline lumbar spine BMD significantly increased the risk of a new vertebral fracture 1.5-fold and 2-fold, respectively, at 3 years.

Figure FIG. 1..

(A) Logistic regression analysis curves of baseline femoral neck BMD and the risk of new vertebral fractures with 95% CIs for the pooled raloxifene (circles) and placebo (stars) groups. (B) Logistic regression analysis curves of baseline lumbar spine BMD and the risk of new vertebral fractures with 95% CIs for the pooled raloxifene (circles) and placebo (stars) groups.

The efficacy of raloxifene in decreasing fracture risk was similar across the broad range of baseline femoral neck BMD (interaction, p = 0.76; Fig. 1A). Raloxifene decreased the risk of incident vertebral fractures by approximately 41% compared with the placebo group at every baseline value of femoral neck BMD ranging from 0.50 to 0.75 g/cm2, which represents the 5th-95th percentiles of the population. In contrast, the effect of raloxifene treatment on fracture risk reduction was different at each percentile of baseline lumbar spine BMD (interaction, p = 0.08; Fig. 1B). Women in the lower percentiles for baseline lumbar spine BMD had greater decreases in fracture risk with raloxifene treatment compared with women in the upper percentiles (Fig. 1B).

Percentage changes in BMD at 3 years

At 3 years, raloxifene increased femoral neck and lumbar spine BMD by 2.2% and 2.7%, respectively, compared with placebo. These increases in BMD were relatively modest in comparison with the robust 40% reduction in vertebral fracture risk at 3 years. To explain this disparity between the BMD increases and fracture risk reduction and to test the hypothesis that changes in BMD can predict fracture risk reduction, the relationships between percentage changes in femoral neck and lumbar spine BMD and the observed fracture risk reduction were assessed using logistic regression models.

The raloxifene group had a statistically significantly lower vertebral fracture risk compared with the placebo group at any percentage change in femoral neck BMD from −7.3 to 10.1%, which represents the 5th-95th percentiles of the population. This relationship is shown graphically in Fig. 2A, as the raloxifene curve is below the placebo curve at all values of femoral neck BMD. Because there was no statistically significant interaction between treatment and percentage change in femoral neck BMD, the interaction effect was dropped from the logistic regression model (interaction, p = 0.81).

Figure FIG. 2..

(A) Logistic regression analysis curves of the percentage change in femoral neck BMD at 3 years and the risk of new vertebral fractures with 95% CIs for the pooled raloxifene (circles) and placebo (stars) groups. (B) Logistic regression analysis curves of the percentage change in lumbar spine BMD at 3 years and the risk of new vertebral fractures with 95% CIs for the pooled raloxifene (circles) and placebo (stars) groups.

The relative risk of vertebral fracture with raloxifene was decreased by approximately 36%, at the lower 25th and median 50th percentiles and the upper 75th percentile of the percentage change in femoral neck BMD (Table 2). For example, in the 75th percentile of the population, which corresponds to a 4.2% increase in femoral neck BMD from baseline, the relative risk of a new vertebral fracture was 0.64. This same 36% reduction in vertebral fracture risk also was observed in the 25th percentile of the population, which corresponds to a 1.9% decrease in femoral neck BMD (Table 2). The percentage of increase in femoral neck BMD at 3 years accounted for 4% of the total vertebral fracture risk reduction observed with raloxifene. This value was calculated by subtracting the relative risk reduction for vertebral fractures with raloxifene treatment before and after adjustment for the percentage change in femoral neck BMD.

Table Table 2. Relationship Between Percentage Change in Femoral Neck and Lumbar Spine BMD at 3 Years and Risks of New Vertebral Fractures
original image

The raloxifene group consistently had a lower fracture risk compared with the placebo group at any percentage change in lumbar spine BMD from −4.1% to 10.8%, which represents the 5th-95th percentiles of the population (Fig. 2B). However, the curves for the placebo and raloxifene groups have markedly different slopes, suggesting varying degrees of fracture risk reduction with different percentage changes in lumbar spine BMD. Indeed, the interaction effect between treatment and percentage change in lumbar spine BMD was statistically significant (p = 0.01). In the placebo group, the fracture risk increased as the lumbar spine BMD increased, and in the raloxifene group, the fracture risk changed very little over a wide range of percentage changes in lumbar spine BMD, resulting in a curve with a slope close to zero (0.0052).

At the 75th percentile of the population, which corresponds to a 6.1% increase in lumbar spine BMD, the risk of an incident vertebral fracture was decreased by 52% with raloxifene therapy (Table 2). In contrast, raloxifene decreased the risks of incident vertebral fractures by 39% and 46% at the 0.3% and 3.1% increases in lumbar spine BMD, which correspond to the 25th and 50th percentiles of the population, respectively. This logistic regression analysis showed that the percentage change in lumbar spine BMD accounts for a fraction of the total reduction in the relative risk of new vertebral fractures. However, this fraction cannot be quantified accurately because of the significant interaction effect.

The upward sloping curve for the placebo group (Fig. 2B) suggested that increased lumbar spine BMD was associated with an increased vertebral fracture risk. In the placebo group, women who had an increased number of new vertebral fractures also had a significant percentage increase in lumbar spine BMD (overall, p = 0.0001). The mean change in lumbar spine BMD was found to be significantly different between women with no new fractures (Fig. 3) compared with women who had one, two, or three or more new fractures (p < 0.05, all between-group comparisons). The relationship between the percentage change in femoral neck BMD and the number of new vertebral fractures was not statistically significant.

Figure FIG. 3..

Women in the placebo group who had an increasing number of new or worsening vertebral fractures also had an apparent increase in lumbar spine BMD at 3 years (overall, p = 0.001). This overall significance was not found for the percentage change in femoral neck BMD. The mean difference in lumbar spine BMD for women with no fractures also was significantly different from women with one, two, or three or more fractures (all comparisons between groups, p < 0.05). Numbers above the bars (n) indicate the number of women in that category. *Statistically significant change in BMD from baseline to 3 years.

Percentage changes in BMD at 1 year

Because the increasing number of new vertebral fractures in the placebo group (Fig. 3) may have obscured the relationship between 3-year changes in lumbar spine BMD and relative risk of fracture (Fig. 2B), similar logistic regression analyses were performed using a 1-year change in lumbar spine BMD. At 1 year, raloxifene treatment increased femoral neck and lumbar spine BMD by 1.3% and 2.1%, respectively, compared with placebo. The interaction effect between treatment and 1-year percentage change in lumbar spine BMD was not statistically significant (p = 0.23) and thus was dropped from the logistic regression model.

The raloxifene group consistently had a lower fracture risk compared with the placebo group at any 1-year percentage change in femoral neck or lumbar spine BMD from the 5th-95th percentile (Fig. 4). The regression analysis curve for the relative risk of fracture with a 1-year percentage change in femoral neck BMD (Fig. 4A) was similar to that observed with a 3-year percentage change in femoral neck BMD (Fig. 2A). The relative risk of incident vertebral fractures was decreased by 38% at the 25th and 50th percentiles and the 75th percentile of a 1-year change in femoral neck BMD (Table 3). Thus, women at the 25th percentile, who had a 1.2% decrease in femoral neck BMD, had similar reduction in vertebral fracture risk as women at the 75th percentile, who had a 4.0% increase in femoral neck BMD. Lumbar spine BMD at 1 year was increased by 0.4% at the 25th percentile of the population and by 4.7% at the 75th percentile; yet, the risk of incident vertebral fractures was decreased by 41% with raloxifene treatment, irrespective of the percentage of increase in BMD (Table 3).

Table Table 3. Relationship Between Percentage Change in Femoral Neck and Lumbar Spine BMD at 1 Year and Risks of New Vertebral Fractures
original image
Figure FIG. 4..

(A) Logistic regression analysis curves of the percentage change in femoral neck BMD at 1 year and the risk of new vertebral fractures with 95% CIs for the pooled raloxifene (circles) and placebo (stars) groups. (B) Logistic regression analysis curves of the percentage change in lumbar spine BMD at 1 year and the risk of new vertebral fractures with 95% CIs for the pooled raloxifene (circles) and placebo (stars) groups.

All of these logistic regression analyses were performed with the baseline or percentage change in femoral neck and lumbar spine BMD with raloxifene, 60 mg/day, the currently marketed dose, and the statistical inferences were identical (data not shown). Logistic regression models were used to analyze the relationships between the observed reduction in fracture risk with raloxifene, with the absolute endpoint BMD values and the absolute changes in BMD at the lumbar spine and femoral neck after 3 years. The results on vertebral fracture risk reduction with absolute endpoint BMD values and absolute change in BMD (data not shown) were similar to that observed with percentage change in BMD.

DISCUSSION

The relationships between baseline BMD and vertebral fracture risk in placebo-treated women in the MORE cohort were consistent with the relationships observed in untreated women in prospective, observational studies.(19) For example, with a 1 SD decrease in femoral neck BMD, the relative risk of vertebral fracture was increased 1.5-fold in the MORE cohort and 1.8-fold in other studies.(2, 20) Logistic regression analysis confirmed that baseline femoral neck or lumbar spine BMD predicted vertebral fracture risk, and women with the lowest baseline BMD had the greatest fracture risk. In this study, raloxifene therapy consistently reduced vertebral fracture risk compared with placebo, irrespective of the baseline BMD value at the femoral neck or lumbar spine, as the regression curves for the raloxifene group were consistently below the curves for the placebo group.

Although baseline BMD values were predictive of subsequent fracture risk in this study, the percentage changes in BMD after raloxifene therapy were found to be poorly predictive of the fracture risk reduction observed with raloxifene therapy, contrary to our hypothesis. Raloxifene-treated patients consistently showed lower fracture risk than placebo, irrespective of the percentage changes in femoral neck BMD observed at 1 or 3 years, as the raloxifene curves were consistently below the placebo curves, and the curves were not superimposed. However, the changes in femoral neck BMD at 1 year or 3 years did not predict the fracture benefit with raloxifene, because the observed fracture risk reduction was very similar over the range of BMD changes. In fact, the 1-year and 3-year changes in femoral neck BMD accounted for about 4% of the fracture risk reduction with raloxifene therapy. Similar results were observed with changes in lumbar spine BMD at 1 year. The relationships between the vertebral fracture risk reduction with raloxifene at 3 years and the absolute endpoint BMD values or the absolute BMD changes at 3 years were similar to that observed with the 3-year percentage changes in lumbar spine and femoral neck BMD. However, counterintuitive changes were observed in the lumbar spine BMD at 3 years. In the raloxifene group, the fracture risk was similar at all percentage changes in lumbar spine BMD at 3 years, as depicted by the flat curve, whereas in the placebo group, the fracture risk increased with increasing BMD. Therefore, lumbar spine BMD at 3 years was poorly predictive of the vertebral fracture risk at 3 years. Also, the 3-year change in lumbar spine BMD only accounted for a fraction of the observed vertebral fracture risk reduction with raloxifene, but this risk reduction could not be reliably calculated, because of the aberrant changes in the placebo group. Nevertheless, given the markedly different slopes for the change in lumbar spine BMD between the raloxifene and placebo groups, raloxifene-treated patients with the greatest increases in lumbar spine BMD had the greatest reduction in fracture risk.

Accuracy and reproducibility errors in BMD measurements may have contributed to the relationships observed between BMD and vertebral fracture risk in these analyses. In this study, the accuracy and reproducibility of the BMD measurements were verified according to established protocols.(21, 22) In cross-sectional studies of postmenopausal women of various age groups, the accuracy of lumbar spine BMD measurements decreases with increasing age because of spinal osteoarthritis in elderly populations.(23) The accuracy errors for posterior-anterior spine DXA can vary from 4% to 10%, and the accuracy error for proximal femur DXA measurements is estimated as 6%.(24) The short-term and long-term reproducibility errors are greater in the femoral neck than in the lumbar spine.(24, 25) Errors may result from differences between DXA machines, longitudinal drifts within machines, and placement of the patient on the machine for measurements. Even after statistical adjustments for reproducibility errors, changes in femoral neck BMD accounted for <10% of the observed vertebral fracture risk reduction after 3 years in the MORE trial.(26)

In addition, the amount of fat to lean body mass surrounding the tissues could affect the tissue density gradient assessed by DXA and influence the accuracy and reproducibility of BMD measurements. Postmenopausal women with greater body weight and fat or lean body mass were found to have higher lumbar spine, hip, and femoral neck BMD.(27, 28) In this study, the mean weight and mean BMI of women in the pooled raloxifene group were increased by 0.3 kg and 0.3 kg/m2, respectively, compared with the placebo group at 3 years (p < 0.001 for both endpoints). It is not known if these slight increases in BMI and weight observed with raloxifene contribute to the imprecision of the BMD measurements or observed decrease in vertebral fracture risk.

Lumbar vertebrae with fractures that were visually observable on the DXA screen were excluded from the BMD analyses in the MORE trial because accurate BMD measurements could not be obtained. It is possible that microarchitectural deformities in the vertebrae, which were not visually evident, could accumulate over time and contribute to the apparent increase in the overall lumbar spine BMD measurements and ultimately accumulate to a fracture.(29–31) In addition, the presence of degenerative conditions of the spine such as osteophytosis and end plate sclerosis could contribute to the variation in lumbar spine BMD measurements(32) and hence the accuracy of fracture risk prediction. However, the variations in BMD measurements caused by degenerative changes in the vertebrae would be expected to affect both treatment groups equally.

The presence of fewer microarchitectural deformities or degenerative changes at 1 year compared with 3 years may contribute to greater accuracy in predicting fracture risk reduction, making the percentage change in lumbar spine BMD at 1 year a better predictor, although still poor, than the change at 3 years. The relationship between BMD and fracture risk is confounded by other factors that contribute to the etiology of a vertebral fracture such as microarchitectural deterioration.(29, 30, 33, 34) It is not known to what extent, if any, these other factors explain the remaining 96% of the fracture risk reduction that is not explained by BMD. Although percentage change in femoral neck BMD at 1 year or 3 years was more predictive for fracture risk than the percentage change in lumbar spine BMD, changes in femoral neck BMD accounted for <10% of the fracture risk reduction with raloxifene because the slopes of the curves were shallow. Acceptable surrogate endpoints must reliably predict the clinical outcome.(11) Thus, because neither the percentage changes in femoral neck nor the lumbar spine BMD can reliably predict vertebral fracture risk reduction at 3 years, these measurements represent poor surrogates for the fracture outcome with raloxifene. There is some controversy as to whether BMD is an adequate surrogate endpoint to predict decreased fracture risk with other antiresorptive therapies and if such a relationship is theoretically valid and quantitatively similar for all osteoporosis therapies.(11)

Several mathematical models have tried to quantify the proportion of fracture risk reduction that can be attributed to BMD changes.(15, 16) Using the Freedman analysis method, endpoint lumbar spine and hip BMD values were estimated to contribute 14% and 9%, respectively, to the observed vertebral fracture risk reduction with raloxifene.(12) Two limitations of the Freedman model analyses must be considered when comparing these earlier results with the present regression analyses. The Freedman model did not account for any potential interaction effect, which was found to be significant for lumbar spine BMD in this analysis. Also, the Freedman model, as previously used,(12) did not assess the proportion of fracture risk reduction attributable to absolute change or percentage change in BMD, as was done in this analysis. Instead, the 1-year endpoint BMD, which is highly correlated with baseline BMD, was used in predicting fracture risk. Using the Freedman analysis model, the increase in spine BMD with alendronate therapy accounted for approximately 17% of the vertebral fracture risk.(13) However, it was not clear whether the alendronate data analysis took the interaction term into account because our analyses showed a significant interaction between treatment and spine BMD. In addition, the observed reduction in fracture risk after risedronate therapy was only partly explained by changes in BMD.(16) In contrast, other analyses have suggested that BMD measurements are appropriate surrogate endpoints to predict the relative risk of a future fracture.(35)

Despite the strong association between low baseline BMD and increased fracture risk in osteoporosis, BMD measurements cannot accurately predict individuals who will develop a future fracture.(20) In a recent study comparing the BMD results from the raloxifene and alendronate fracture trials, women in the active therapy groups who had decreases in BMD in the first year had increases in BMD in the second year, and vice versa, because of the principle of “regression to the mean.”(36) In the MORE study, even women who had a net decrease in BMD from baseline still had a significant 31% reduction in fracture risk (data on file, Lilly Research Laboratories, Indianapolis, IN, USA). Based on current evidence, we cannot assume that antiresorptive agents that elicit a greater BMD increase are more efficacious in decreasing vertebral fracture risk than those that have smaller effects on BMD in postmenopausal women.

This logistic regression analysis model shows that baseline BMD at either the femoral neck or the lumbar spine is a good predictor of fracture risk, because those patients with the lowest baseline BMD values have the greatest risk of future fracture. The logistic regression analysis model shows that changes in femoral neck BMD at 3 years may better predict fracture risk than changes in lumbar spine BMD, which are affected by inherent changes in the spine of the elderly and, perhaps, microarchitectural deformities. Regardless of the percentage change in femoral neck BMD, raloxifene-treated patients had greater reductions in fracture risk compared with placebo. However, the change in femoral neck BMD at either 1 year or 3 years could only explain a small fraction of the fracture risk reduction benefit. Similar to the relationships observed with other antiresorptive therapies, raloxifene-induced changes in BMD are correlated with fracture benefit but are poor surrogates for predicting the actual reductions in fracture risk with raloxifene treatment.

Acknowledgements

The authors are grateful for the assistance of Michele Hill in editing the article and preparing the graphs. The authors thank Hunter Heath III M.D., and Wentao Wu, M.S., of Lilly Research Laboratories for critically reviewing this article.

Ancillary