Effect of medications on prevention of secondary osteoporotic vertebral compression fracture, non-vertebral fracture, and discontinuation due to adverse events: a meta-analysis of randomized controlled trials

Background Bone loss with aging and menopause increases the risk of fragile vertebral fracture, osteoporotic vertebral compression fracture (OVCF). The fracture causes severe pain, impedes respiratory function, lower the quality of life, and increases the risk of new fractures and deaths. Various medications have been prescribed to prevent a secondary fracture, but few study summarized their effects. Therefore, we investigated their effects on preventing subsequent OVCF via meta-analyses of randomized controlled trials. Methods Electronic databases, including MEDLINE, EMBASE, CENTRAL, and Web of Science were searched for published randomized controlled trials from June 2015 to June 2019. The trials that recruited participants with at least one OVCF were included. We assessed the risk of bias of every study, estimated relative risk ratio of secondary OVCF, non-vertebral fracture, gastrointestinal complaints and discontinuation due to adverse events. Finally, we evaluated the quality of evidence. Results Forty-one articles were included. Moderate to high quality evidence proved the effectiveness of zoledronate (Relative Risk, RR: 0.34; 95% CI, 0.17–0.69, p = 0.003), alendronate (RR: 0.54; 95% CI: 0.43–0.68; p < 0.0001), risedronate (RR: 0.61; 95% CI: 0.51–0.73; p < 0.0001), etidronate (RR, 0.50; 95% CI, 0.29–0.87, p < 0.01), ibandronate (RR: 0.52; 95% CI: 0.38–0.71; p < 0.0001), parathyroid hormone (RR: 0.31; 95% CI: 0.23–0.41; p < 0.0001), denosumab (RR, 0.41; 95% CI, 0.29–0.57; p < 0.0001) and selective estrogen receptor modulators (Raloxifene, RR: 0.58; 95% CI: 0.44–0.76; p < 0.0001; Bazedoxifene, RR: 0.66; 95% CI: 0.53–0.82; p = 0.0002) in preventing secondary fractures. Moderate quality evidence proved romosozumab had better effect than alendronate (Romosozumab vs. alendronate, RR: 0.64; 95% CI: 0.49–0.84; p = 0.001) and high quality evidence proved that teriparatide had better effect than risedronate (risedronate vs. teriparatide, RR: 1.98; 95% CI: 1.44–2.70; p < 0.0001). Conclusion Zoledronate, alendronate, risedronate, etidronate, ibandronate, parathyroid hormone, denosumab and selective estrogen receptor modulators had significant secondary prevention effects on OVCF. Moderate quality evidence proved romosozumab had better effect than alendronate. High quality evidence proved PTH had better effect than risedronate, but with higher risk of adverse events. Electronic supplementary material The online version of this article (10.1186/s12891-019-2769-8) contains supplementary material, which is available to authorized users.


Background
Osteoporotic vertebral compression fracture (OVCF) is one of the most common fragile fractures, with a prevalence of 30 to 50% in people over 50 years of age [1]. It causes severe pain and disability, raises the risk of secondary fracture more than 4-fold [2,3], and increases the risk of mortality [4]. Therefore, secondary prevention of OVCF was critical and should be emphasized to improve patients' quality of life. However, though the primary prevention efficacy of medications have been well summarized [5][6][7][8][9][10], only one systematic review targeted on their secondary prevention effects [11]. Therefore, to investigate the efficacy of current medication therapies on preventing secondary OVCF, we conducted this study through systematically literature review and meta-analyses of randomized controlled trials (RCTs).

Search for studies
Four major electronic databases (MEDLINE, EMBASE, CENTRAL, and Web of Science) were searched with a developed search strategy that consisted of keywords "controlled trials", "osteoporotic fracture", "bisphosphonate", "parathyroid hormone", "denosumab" "calcitonin", "Raloxifene", "Bazedoxifene" "hormone replacement", "romosozumab", "abaloparatide", etc., and others (Additional file 1). The search spanned the period from June 2015 to June 2019, with weekly alerts of updated published trials. Reference lists from other reviews and studies were also checked for relevant articles. The references were managed with Endnote X7 (Clarivate Analytics).

Selection of studies
Three authors (YZJ, BX, and MJC) independently screened the titles and abstracts of studies and evaluated their relevance to our study. A study was included if it involved patients with osteoporosis. A subsequent fulltext assessment was done by three authors (YZJ, BX, and JHL) independently. Randomized controlled trials (RCTs) published in English that investigated the efficacy of currently approved medications for patients with OVCF were included. The studies that included osteoporosis patients without distinguishing their fracture history were included if the data of the participants with prevalent fractures was adequately presented. Studies that recruited patients with traumatic vertebral fracture, secondary osteoporosis, or did not report results in dichotomous data (i.e., patient-years, etc.), were excluded. The included medications were the approved ones, including zoledronate, alendronate, risedronate, etidronate, ibandronate, minodronate, pamidronate, calcitonin, hormone replacement therapy, parathyroid hormone, denosumab, romosozumab, raloxifene, and bazedoxifene [12][13][14]. Post hoc analyzed RCTs were also included, with taking care of duplicated data input. Disagreements between reviewers were resolved by discussion or, if unresolved, by consultation with consultation with librarians and a statistic professor from SMG-SNU Boramae Medical Center.

Data extraction and risk of bias
Basic characteristics of each study were independently extracted by YZJ, BX, and JHL with a designed table that contains the number of participants, interventions, comparisons, and outcomes. The primary outcome of this study was the vertebral fracture ratio in the final visit, and the secondary outcomes were gastrointestinal (GI) complaints of bisphosphonates, discontinuation due to adverse events (AEs), and non-vertebral fracture ratio.
The risk of bias was measured independently by YZJ, BX, and JHL with the tool recommended in updated guidelines of Cochrane Back and Neck Group [15]. The detection bias was rated for main result (vertebral fracture). The loss ratio was acceptable for a middle-or long-term trial (observational period > 1 year), if that was not exceeded 30%. The risk of other sources of bias was rated as low risk if the article stated both conflict of interest and sponsor of the trial and no other serious risk of bias was reported.

Data analysis and quality of evidence
Relative risk (RR) and its 95% confidence intervals (CIs) were used to estimate the effect of interventions, with pvalues < 0.05 considered significant. The overall effect size was calculated with a random effects model [15]. Heterogeneity between studies was identified and measured with p-value and I 2 value from Chi-squared test, p-value < 0.1 was identified as significant, and I 2 < 40% was considered as not important, I 2 between 40 and 74% indicated moderate to substantial, I 2 > 75% was identified as a considerable magnitude [15]. In studies with more than two arms, intervention groups were input into each subgroup and the data in the control groups were separated equally and then were compared to their counterparts. Sensitivity analyses were used to explore the interference from a study by excluding it from syntheses and the impact from loss to follow-up population by compositing the missing events according to event ratio in control groups [16]. The data was analyzed by two authors (YZJ and JHL) with RevMan 5.3.3 (Cochrane).
We evaluated five factors of the results to determine the quality of evidence, including study limitation, imprecision, indirectness, inconsistent and publication bias, followed the GRADE approach. The criteria for downregulating the level referred to the handbook of GRADE and guidelines from Cochrane Back and Neck Group [15,16]. In the case that an outcome included one trial with no unclear or high risk of bias, the study limitation item was rated as not serious if its result remained same direction and signficancy with the pooled result.
Approximately half of the biases were rated as unclear risk (Additional file 2). Risk of other sources of bias was rated as high in one study because the criteria used in its    two clinical centers were different [21]. Performance bias was rated as high risk in 6 trials for significantly different compliance between groups [26,46,52] and the openlabel study design used in 4 trials [20,27,37,40]. Fujita et al. treated a teriparatide 1.4 μg/week group as a placebo group, and therefore, we followed their grouping and classified their data of teriparatide 1.2 μg/week group as the control group [45]. On the other side, since the control groups in other studies all received placebo, which was different from Fujita et al.'s study, it might affect the final result. Therefore, we performed a sensitivity analysis about this result, with excluding the Fujita el al.'s study from the original analysis and then compared the results from original analysis and sensitivity analysis (Table 1). Sorensen et al. reported a 2-year extension trial [23] of a 3-year original trial [22]. In the entension trial, the authors treated the initial time point of the extention trial as baseline. Therefore, while synthesizing the data, we deemed the data from the two studies were not duplicated and synthesized the data as from two studies. However, because the participants in experimental group and control group in the extension study had different medication history, the risk of selection bias of the extended trial was rated as high (Additional file 2).

Discussion
In this study, we focused on osteoporosis patients with a history of OVCF. We collected related RCTs, synthesized their results, and finally estimated the secondary prevention effects of the medications on OVCF. We found zoledronate, alendronate, risedronate, etidronate, ibandronate, minodronate, pamidronate, PTH, denosumab, romosozumab and SERMs had significant secondary prevention effect on OVCF. In the comparisons between the medications, teriparatide had a significantly superior effect to risedronate, and the quality of evidence was high. The effects of risedronate, ibandronate, PTH, and SERMs were supported by moderate quality evidence and the effects of alendronate, denosumab were supported by high quality evidence.
In the result of discontinuation due to adverse events, PTH was the only intervention that significantly elevated the ratio. None of the bisphosphonates increased the risk of GI complaints. Zoledronate, risedronate, and PTH had significant effect on preventing non-vertebral fracture in patients with prevalent OVCF. Most of widely used BPs, include zoledronate, alendronate, risedronate, etidronate and ibandronate, had significant effect, which were supported by moderate quality evidences. Among the medications, risedronate and ALN are first line osteoporosis medications, whose effects have been proved by substantial evidence [5,7]. Ibandronate is a nitrogen-containing BPs and IV injection of it allows for a dosing interval even longer than 2 months [59]. Zoledronate is another nitrogen-containing BPs that has the highest potency among clinical use BPs [60]. According to our result, 5 mg/year iv injection of zoledronate could significantly reduce the risk of secondary OVCF. The extremely low medication frequency could be its another advantage that might improve patients' compliance rate. Significantly elevated adverse events ratio or rare adverse events caused by BPs (e.g. osteonecrosis of jaw or atypical fracture, etc.) was not 7reported in any trial. Insignificant difference in GI complaints between BPs and control group indicated properly administrated BPs might help avoiding the risk of GI complaints, which was consistent with previous studies [61].
PTH is a bone anabolic medication that has significant efficacy against OVCF [62]. In this study, moderate quality evidence proved that the injection of PTH or teriparatide significantly reduce the risk of secondary OVCF and even the lowest dosage (28.2 μg/week) showed a significant effect. Compare with risedronate, teriparatide showed significantly better effect, which indicated PTH might have better effect on preventing secondary OVCF. It was consistent with previous studies that proved PTH had better effect on spine BMD compare with bisphosphonate [63,64]. But on the other side, the superiority of PTH over bisphosphonate on hip BMD remains controversial, and it has been showed the PTH had inferior effect on BMD of distal radius [63,64]. .Also, PTH treatment was the only medication that was associated with a series of adverse events that increased the risk of discontinuation. The most frequent adverse event was nausea, and other complaints included vomiting, headache, dizziness, and leg cramps [43,46].
SERMs included in the study were raloxifene and bazedoxifene. Both showed a significant effect in preventing secondary fracture. Raloxifene seemed to have a better effect when prescribed at a higher dosage, which was indicated by the significant and substantial heterogeneity between the two groups. Besides beneficial skeletal effects, SERMs reduce the risk of breast cancer [65]. However, an elevated risk of venous thromboembolic events due to raloxifene and bazedoxifene has been described [52]. Additionally, raloxifene significantly raises the risk of discontinuation [50]. Therefore, SERMs should be prescribed with an awareness of their risk of side effects.
Denosumab is a RANKL inhibitor that was proved to possess significant effect on preventing secondary OVCF. Side effects of it include skin rashes, infections, and osteonecrosis of the jaw [62], but presently, there was no significant difference in adverse events compared with control group. Additionally, Boonen et al. reported a significant reduction of fatal adverse events ratio with denosumab in patients with prevalent vertebral fracture [49]. One advantage of denosumab is its low dosing frequency, which might elevate compliance. Romosozumab is a sclerostin inhibitor that has been proved to have better effect on preventing secondary OVCF than alendronate. However, it should be noticed that the cardiac ischemic events and cerebrovascular events ratio were higher in romosozumab group. The role of sclerostin in vessels remains unclear, and the results from basic studies were controversial [66][67][68]. Therefore, further evaluation of safety profile of romosozumab is needed.
Unlike the superior effects on OVCF of most medications, only zoledronate, risedronate, and PTH had a significant effect on preventing non-vertebral fractures in patients with prevalent OVCF. Combined with the effects of medications on OVCF, the findings might indicate zoledronate, risedronate, and PTH might be better options for patients with prevalent OVCF. Additionally, denosumab and alendronate showed marginally significant effects. The results might have less credibility than the main outcome because of missed information concerning the non-vertebral fracture status of the participants. But, the patients included in this study could still be considered as having a high risk of non-vertebral fracture because prevalent vertebral fracture and low bone mineral density are potential risk factors of non-vertebral fractures [69,70]. Therefore, the data might be instructive for clinical usage of the medications.
It must be noted that many phase 3 studies were excluded from this meta-analysis because the data of patients with prevalent fractures were not reported. The exclusion might cause an underestimation of the effects of some newly developed medications like denosumab and zoledronate. One limitation of this study include the absence of searching the gray literature, which might increase the risk of publication bias that might lead to an overestimation of the effect of newly developed medications like romosozumab and bazedoxifene. Also, we only included English written manuscript in this study. Though no solid evidence showed a bias caused by the language restriction [71][72][73], the manuscript written in other languages should be included in further studies for a more comprehensive understanding of the effects of the medications. The generalizability of results of GI complaints was limited, because most of the trials excluded patients with upper GI disease at baseline. Additionally, our criteria for assessing the risk of bias might be too stringent, which might underestimate the quality of evidence. Also, it should be noticed that the most common domains that downregulated the GRADE was the study limitation and imprecision. The same scenario has been reported in a review of systematic reviews, in which the authors indicated the need of high quality RCTs with large sample size for better clinical decisions [74]. In the aspect of study limitation, the two main categories of risk of bias that were rated as unclear to high risk of bias were performance bias and selection bias. For a higher quality of evidence, the report of study might better follow the guidance, like the CONSORT, and report the procedure of randomization and blinding could be great help. To decrease the impact from imprecision, RCTs with higher sample sizes were needed. Also, a report of a subgroup of population with prevalent fracture would help in expanding the sample size.
Most systematic reviews and meta-analyses included osteoporosis patients, regardless their fracture history that introduces indirectness in the results [5][6][7][8][9][10]. The results might be overestimated on patients had fracture history, and for optimized treatment, accurate analyses of OVCF patients is urged. However, only one systematic review satisfied the demand [11]. Compared with that, we included 14 more RCTs and new medicines such as romosozumab and abaloparatide that allowed for a more comprehensive review and comparisons between different medications. Also, our results included vertebral fracture, non-vertebral fracture, GI complaints of BPs and discontinuation due to AEs. In the end, we evaluated the quality of evidence. The updated information could offer more practical evidence for clinical use.
Our results are consistent with those from other systematic reviews about primary prevention of OVCF [5-7, 9, 75-77]. This could indicate that the medications have a consistent effect on osteoporosis patients, regardless their OVCF history. Also, medications used to prevent osteoporotic fracture had a low risk of severe adverse events in most of the 2-3 years follow-ups. Therefore, the benefits from reducing the risk of fracture, disability, and mortality very likely outweigh the disadvantages. But, careful evaluation of risk factors and arrangement of drug holidays are also necessary to minimize the risk of adverse events [78].
Lack of RCTs that compared interventions of secondary prevention effect limited our assessment of differences between interventions. Although indirect comparisons could be conducted through statistical analyses, high quality RCTs that provide direct evidence are necessary for a solid conclusion.