Effect of Osteoporosis Treatment on Mortality: A Meta-Analysis
Journal of Clinical Endocrinology & Metabolism March 2010|
Mark J. Bolland, Andrew B. Grey, Greg D. Gamble and Ian R. Reid
Department of Medicine, University of Auckland, 1142 Auckland, New Zealand
"This analysis of more than 1,400 deaths in approximately 40,000 subjects has shown that treatment of osteoporosis with agents with proven vertebral and nonvertebral fracture efficacy reduces mortality by approximately 10%. The mortality reduction was mainly observed in studies of older, frailer individuals at high risk of fracture....In summary, effective osteoporosis treatments that prevent vertebral and nonvertebral fractures reduce mortality by approximately 10% in older, frailer individuals with osteoporosis who are at high risk of fractures. Fragility fractures are known to cause substantial morbidity and mortality, and effective osteoporosis treatment prevents both vertebral and nonvertebral fractures. The reduction in mortality with effective osteoporosis treatment, when added to the established reduction in morbidity from such treatment, provides another reason for vigorously intervening in frail elderly patients with osteoporosis at high risk of fracture. Mortality should be included as a prespecified endpoint in future studies of osteoporosis therapies."
Context: Fragility fractures cause significant morbidity and mortality. Effective osteoporosis treatment can reduce fracture incidence, but it is not known whether it reduces mortality.
Objective: The aim of the study was to determine whether effective osteoporosis treatment reduces mortality.
Data Sources: We searched Medline and the Cochrane Central Register of Trials prior to September 2008, as well as 2000-2008 American Society for Bone and Mineral Research conference abstracts.
Study Selection: Eligible studies were randomized placebo-controlled trials of approved doses of medications with proven efficacy in preventing both vertebral and nonvertebral fractures, in which the study duration was longer than 12 months and there were more than 10 deaths. Trials of estrogen and selective estrogen receptor modulators were specifically excluded.
Data Extraction: Data were extracted from the text of the retrieved articles, published meta-analyses, or the Food and Drug Administration web site.
Data Synthesis: Eight eligible studies of four agents (risedronate, strontium ranelate, zoledronic acid, and denosumab) were included in the primary analysis. During two alendronate studies, the treatment dose changed, and those studies were only included in secondary analyses. In the primary analysis, treatment was associated with an 11% reduction in mortality (relative risk, 0.89; 95% confidence interval, 0.80-0.99; P = 0.036). In the secondary analysis, the results were similar (relative risk, 0.90; 95% confidence interval, 0.81-1.0; P = 0.044). Mortality reduction was not related to age or incidence of hip or nonvertebral fracture, but was greatest in trials conducted in populations with higher mortality rates.
Conclusions: Treatments for osteoporosis with established vertebral and nonvertebral fracture efficacy reduce mortality in older, frailer individuals with osteoporosis who are at high risk of fracture.
Fragility fractures associated with osteoporosis cause significant costs, morbidity, and mortality (1). Between 10 and 20% of people who sustain hip fractures die within 1 yr (1), and increased mortality is also observed after vertebral and other nonvertebral fractures (2). In the last 30 yr, a number of treatments have been developed that effectively prevent fragility fractures. Recent systematic reviews and meta-analyses have concluded that osteoporosis treatment can significantly reduce the incidence of vertebral and nonvertebral fractures (3). Despite clear evidence of fracture prevention, it is not known whether effective osteoporosis treatment reduces mortality. In a recent study, annual zoledronic acid first administered within 3 months after hip fracture reduced the incidence of a second hip fracture by 30% and of nonvertebral fracture by 27% (4). In a secondary analysis, zoledronic acid also reduced all-cause mortality by 28% (4). This is the only large randomized controlled trial of osteoporosis treatment to report a statistically significant survival benefit, although most trials were not powered to detect mortality differences. Therefore, we carried out a meta-analysis of existing placebo-controlled randomized trials to determine whether effective osteoporosis treatment reduces mortality.
This analysis of more than 1,400 deaths in approximately 40,000 subjects has shown that treatment of osteoporosis with agents with proven vertebral and nonvertebral fracture efficacy reduces mortality by approximately 10%. The mortality reduction was mainly observed in studies of older, frailer individuals at high risk of fracture. When applied to the trials in this analysis, a 10% RR reduction would correspond to an absolute mortality benefit ranging from 0.4-7 deaths prevented per 1000 patient-years of treatment. The mortality benefit appeared similar across the different classes of agents in this analysis (bisphosphonates, strontium ranelate, and denosumab), although data from studies of non-bisphosphonates are fewer. These agents are reported to reduce the risk of vertebral fracture by up to 77% and to reduce the risk of nonvertebral fracture by up to 49% (3). A reduction in mortality of 10% provides a further reason to treat older, frailer individuals with osteoporosis at high risk of fracture with agents with proven vertebral and nonvertebral efficacy.
The mechanism by which deaths are prevented with effective osteoporosis treatment is not clear. Prevention of fractures is one possible explanation. Vertebral, hip, and other nonvertebral fractures are associated with 2- to 3-fold increased mortality (2, 31, 32). However, much of that increased risk does not appear to be related to the fracture event. In the Study of Osteoporotic Fractures, hip and pelvic fractures were associated with a RR of death of 2.4, but only 14% of deaths were attributable to the fracture (33). Kanis et al. (34, 35) analyzed the patient register from Sweden and concluded that after hip fracture, 17-32% of deaths (depending on age) were causally related to the fracture, and after vertebral fracture requiring hospitalization, 28% of deaths were causally related to the fracture. Based upon these data, if approximately 20% of people die as a consequence of their hip fracture (1), 4-6% of deaths would be attributable to the fracture. If effective treatment for osteoporosis reduces fractures by approximately 50%, an approximate reduction in mortality from hip fracture of 2-3% could be predicted. In the study of zoledronic acid after hip fracture, only 8% of the observed mortality benefit was explained by prevention of subsequent fractures (36). Thus, fracture prevention seems likely to explain only a small proportion of the mortality reduction we observed.
Only one study in this analysis reported data on cause of deaths (4, 36). In that study, there was a similar rate of occurrence of pneumonia, cancer, and cardiovascular disease in patients treated with placebo or zoledronic acid, but death rates from these conditions tended to be decreased in those treated with zoledronic acid. This suggests that treating osteoporosis might improve the ability of an individual to cope with, and recover from, an acute illness, possibly by maintaining physiological reserve and preventing frailty (36). This notion is supported by the fact that most of the studies analyzed were of bisphosphonates, a class of agents that are specifically targeted to bone and generally have minimal activity in nonskeletal tissues. It is further supported by our finding that mortality reduction by osteoporosis therapy is greater in populations with higher baseline risk of death (i.e. in more frail populations). In less frail populations that are less likely to experience fragility fractures or die as a consequence of these fractures, any benefit on mortality from treating osteoporosis might be small and potentially obscured by other, more common causes of death.
However, it is likely that preventing fractures and frailty is not the only mechanism for the observed mortality reduction. Bisphosphonates may be taken up by calcified blood vessels because of their high affinity for mineralized tissue, and inhibition of the mevalonate pathway in vessel walls has been shown to impact on local nitric oxide generation (37). This, in turn, impacts on several components of the atherogenic process including monocyte adhesion to the endothelial surface, platelet aggregation, vascular smooth muscle cell proliferation, and vasoconstriction (38). Thus, a statin-like effect of bisphosphonates on vessel walls might account for some of their effect on mortality. Because there are limited clinical data available to support or refute this hypothesis, it remains speculative but merits further study. Another possibility is that low-grade inflammation after infusion of iv bisphosphonates might promote changes at a cellular level that ultimately enhance the ability to cope with stressors thereby promoting survival (39, 40).
All trials included in this analysis reported statistically significant reductions in vertebral and nonvertebral fractures, either individually or in pooled analyses. Despite this clear evidence of fracture prevention, there was a statistically significant reduction in mortality with treatment in only one individual trial (4), and in three trials there were more deaths in the treatment group than in the placebo group (17, 19, 20), although none of the increases was statistically significant. The absence of apparent mortality benefit in the individual studies might be because of chance, because the studies were underpowered to detect an effect on mortality, because the reduction in mortality occurs by mechanisms other than fracture prevention, or because a suboptimal drug dose was used during part of the trial (for the alendronate trials). It is interesting to note that in the risedronate Hip Intervention Program trial (16), there was a nonsignificant increase in deaths in the risedronate 2.5 mg daily group, a dose now also considered suboptimal for fracture prevention. Another noteworthy feature highlighted in our analysis is the heterogeneity in the results for the two zoledronic acid trials (4, 19). Although the CIs for the results from the trials overlap, the results from these trials appear to differ. Both trials involved large numbers of participants, with 242 deaths in each trial suggesting that chance is unlikely to explain the differences. Colon-Emeric and their colleagues (41) suggested that the difference between trials was likely due to the difference in baseline risk of death because the study by Black et al. (19) had a 4-fold greater number of participants.
There are limitations to our analyses. There were only 10 studies that met the criteria for inclusion in our review. However, the trials that were included all had fracture prevention as a primary endpoint, and such trials are likely to have the largest sample size, the longest duration, and the most complete ascertainment of adverse events. Nine of the 10 studies were carried out in postmenopausal women with low bone density, a history of fractures, or both, and there was no significant reduction in mortality in these studies; one study was carried out in men or women with a hip fracture with a significant reduction in mortality rate. It is unclear whether the result in the hip fracture trial will be generalizable to women with osteoporosis or osteopenia who have not had a hip fracture. No studies of teriparatide met the inclusion and exclusion criteria for our analysis, and thus the data are currently insufficient to determine whether teriparatide affects mortality or whether antiresorptive and anabolic therapies for osteoporosis differ in their effect on mortality. Data on mortality were not available for 10 studies. However, the size of those studies, together with the baseline age and comorbidities of participants, suggest that there would have been few deaths in those studies. Therefore, we believe it unlikely that the missing data from those studies affected the results of these analyses.
In summary, effective osteoporosis treatments that prevent vertebral and nonvertebral fractures reduce mortality by approximately 10% in older, frailer individuals with osteoporosis who are at high risk of fractures. Fragility fractures are known to cause substantial morbidity and mortality, and effective osteoporosis treatment prevents both vertebral and nonvertebral fractures. The reduction in mortality with effective osteoporosis treatment, when added to the established reduction in morbidity from such treatment, provides another reason for vigorously intervening in frail elderly patients with osteoporosis at high risk of fracture. Mortality should be included as a prespecified endpoint in future studies of osteoporosis therapies.
Figure 1 shows the results of our search. A total of 367 potentially relevant articles were identified from the initial searches, but only eight studies met the inclusion and exclusion criteria (4, 10, 11, 14, 15, 16, 17, 18, 19). In two alendronate studies, treatment was 5 mg daily for the first 2 yr of the trial, and thereafter 10 mg daily (20, 21). Because the approved dose for treatment of osteoporosis is 10 mg daily, these studies were only included in secondary analyses. All 10 studies are described in Table 1, and they collectively reported 1,417 deaths in 39,549 participants.
Figure 2 shows that osteoporosis treatment in eight studies was associated with an 11% reduction in mortality [relative risk (RR), 0.89; 95% confidence interval (CI), 0.80-0.99; P = 0.036]. When the two alendronate studies were included, the analysis produced a similar result (RR, 0.90; 95% CI, 0.81-1.00; P = 0.044) (Fig. 3). When the analysis was restricted to the five studies of bisphosphonates at their approved dose, the RR was 0.89 (95% CI, 0.71-1.10; P = 0.29; I2 = 50.4, random effects model applied). Including the two alendronate studies in this analysis produced a similar result (RR, 0.91; 95% CI, 0.80-1.03; P = 0.13; I2 = 35.4, fixed effects model applied). There was no evidence of publication bias for any of the analyses performed (Fig. 4).
In meta-regression analyses, mortality reduction was not related to mean age (P = 0.41), incidence of hip (P = 0.19), or nonvertebral fracture (P = 0.50) in the placebo group, or vertebral (P = 0.28) or nonvertebral (P = 0.26) fracture risk reduction, but it was related to the mortality rate in the placebo group (P = 0.030). In the four studies in which the mortality rate was greater than 10 per 1000 patient-years (range, 13.9-70.2 deaths per 1000 patient-years) (4, 15, 16, 18), there was a significant reduction in mortality (RR, 0.83; 95% CI, 0.72-0.94; P = 0.0052). By contrast, in the six studies with mortality rate below 10 per 1000 patient-years (range, 4.3-9.7 deaths per 1000 patient-years) (10, 11, 14, 17, 19, 20, 21), there was no reduction in mortality (RR, 1.01; 95% CI, 0.87-1.19; P = 0.86).
Finally, we performed a series of sensitivity analyses. Only one study was carried out after hip fracture (4). When we excluded this trial from the primary analysis, the overall result was not significant (RR, 0.94; 95% CI, 0.84-1.06; P = 0.31). The pooled result for the primary analysis was also not significant when any one of four other trials (10, 11, 15, 16, 18) was excluded. Trials of ibandronate were not eligible for this analysis because ibandronate was not associated with nonvertebral fracture efficacy in the Agency for Healthcare Research and Quality systematic review (3). However, a more recent meta-analysis suggested that higher doses of ibandronate do reduce the incidence of nonvertebral fractures when compared with lower doses of ibandronate (22). None of the studies of higher doses of ibandronate were placebo-controlled and so are not comparable to the studies in our analysis. In the two studies of lower doses of ibandronate that were placebo-controlled and lasted more than 12 months, the RRs of mortality in participants allocated to ibandronate were 0.88 (95% CI, 0.45-1.73; P = 0.71) (23) and 0.95 (95% CI, 0.44-2.03; P = 0.89) (24). Including these two studies in our secondary analysis did not change the overall result (RR, 0.90; 95% CI, 0.82-1.00; P = 0.041). No studies of teriparatide met the inclusion and exclusion criteria for our analysis, with the largest study reporting six deaths in the treatment group and four deaths in the control group (RR, 1.51; 95% CI, 0.43-5.3; P = 0.52) (25). Including this study in our secondary analysis did not change the overall result (RR, 0.90; 95% CI, 0.82-1.00; P = 0.050). Finally, six studies involving 2202 participants had 10 deaths or less (25, 26, 27, 28, 29, 30). When we included these six studies in our secondary analysis, the overall result did not change (RR, 0.90; 95% CI, 0.81-1.00; P = 0.045). Mortality data were unavailable for 10 studies involving 2650 participants.
All analyses were repeated with a random effects model, and the results are shown in Supplemental Table 1 (published as supplemental data on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org). Applying a random effects model did not change the nature or magnitude of the effects in the primary or secondary analyses, but they were no longer statistically significant.
Materials and Methods
We followed the QUOROM guidelines for meta-analysis of randomized controlled trials (5). MEDLINE and the Cochrane Central Register of Trials were searched before September 2008 for randomized placebo-controlled trials of osteoporosis treatments using the terms "osteoporosis," "randomized controlled trial," and "placebo" as keywords and the corresponding MeSH terms. We also searched for any additional studies in the reference lists of recent meta-analyses of treatment for osteoporosis, in the retrieved trials, and in the conference abstracts of 2000-2008 American Society for Bone and Mineral Research annual scientific meetings.
Inclusion criteria were: 1) randomized, double-blind, placebo-controlled trials analyzed by intention-to-treat; 2) trials of agents with proven vertebral and nonvertebral antifracture efficacy; 3) trials using agents at the currently approved dosage for treatment of osteoporosis; 4) mean age of trial participants at baseline above 50 yr; 5) number of deaths in study greater than 10; 6) trial population of men, women, or both; and 7) trial duration longer than 1 yr. We prespecified that we would not include studies with 10 deaths or less because such studies were likely to have small random differences in the numbers of deaths between groups that would produce improbable estimates of the treatment effect. A threshold of more than 10 deaths provides a reasonable balance between including as many studies as possible and excluding small studies with improbable results.
Exclusion criteria were: 1) trials of estrogen and selective estrogen receptor modulators (such as raloxifene, bazedoxifene, and lasofoxifene) because of their effects on multiple systems and outcomes such as heart attack, stroke, colon cancer, breast cancer, and venous thromboembolism (6, 7, 8) that might obscure any effects on mortality from effective osteoporosis treatment; 2) duplicate publications, with the largest study otherwise conforming to the inclusion and exclusion criteria included in analyses; 3) trials in which most subjects had a major systemic pathology other than osteoporosis, such as malignancy; 4) trials in which subjects had secondary osteoporosis such as glucocorticoid-induced osteoporosis; and 5) trials in which subjects did not have osteoporosis. We prespecified the exclusion of studies of glucocorticoid-induced osteoporosis because these studies generally included adults of all ages, with a variety of other major morbidities, but with the majority of participants at low mortality risk.
We prespecified that only agents with proven vertebral and nonvertebral fracture efficacy would be included in the analyses because these agents are preferentially prescribed to patients and including weaker agents that do not prevent nonvertebral fractures might obscure any effect on mortality from more potent agents. We used the Agency for Healthcare Research and Quality comprehensive systematic review as the basis for our criteria of treatment efficacy (3). This report concluded that treatment with alendronate, risedronate, zoledronic acid, estrogen, and teriparatide reduced the risk of vertebral and nonvertebral fractures. Strontium ranelate was not assessed in this review, but a Cochrane review concluded that strontium reduced both vertebral and nonvertebral fractures (9). In addition, data from a phase III trial of denosumab that fulfills criteria for inclusion were recently presented (10, 11), and similar data showing reductions in vertebral and nonvertebral fractures exist for clodronate (12, 13). Therefore, trials of strontium ranelate, denosumab, and clodronate were included in our analyses.
In summary, trials of alendronate 10 mg daily, risedronate 5 mg daily, zoledronic acid 5 mg annually, teriparatide 20 mg daily, strontium ranelate 2 g daily, denosumab 60 mg 6-monthly, and clodronate 800 mg daily were eligible for inclusion in our analyses, whereas trials of these agents at other doses and trials of etidronate, ibandronate, pamidronate, calcitonin, calcium, vitamin D, or PTH 1-84 were not eligible.
Mortality data were initially abstracted from the text of the article. Where these data were not available, we searched company web sites, published meta-analyses, and the Food and Drug Administration web site.
Studies were pooled using a fixed effects model. In the presence of significant heterogeneity (assessed using Cochran's Q statistic and the I2 statistic; I2 >50% was used as a threshold indicating significant heterogeneity), a random effects model was used. Meta-regression using the covariables age, hip and nonvertebral fracture rate in the placebo group, vertebral and nonvertebral fracture risk reduction, and mortality rate in the placebo group was undertaken to establish whether these indices of frailty and disease severity modified the impact of osteoporosis treatment on mortality. Publication bias was assessed using Funnel plots and Egger's regression model. All tests were two-tailed. Results were considered statistically significant for P < 0.05. All analyses were undertaken using Comprehensive Meta-analysis Version 2 (Biostat, Englewood NJ).