Introduction

Breast cancer is one of the most common cancer in women worldwide [1]. In the US, 279,100 people are estimated to be diagnosed with breast cancer, and 42,170 patients die of breast cancer in 2020 [2]. About 80% of all breast cancer is hormone receptor-positive [2], and adjuvant endocrine therapy including aromatase inhibitor (AI) is recommended to prevent recurring or developing new breast cancer in postmenopausal women with hormone-sensitive breast cancer without metastasis [3].

One of the major side effects of AI is a rapid bone loss by depleting residual estrogen in postmenopausal patients [4]. While postmenopausal women, generally speaking, lose bone mineral density (BMD) by about 1% per year [5], patients with AI use lose up to ~ 5% of BMD per year [6]. As expected, the risk of fracture almost doubles in those patients [7, 8]. Therefore, the current guidelines support using bone modifying agents (BMAs) such as bisphosphonates or denosumab to prevent the fracture [4].

However, head-to-head comparison of those medications in terms of the anti-fracture efficacy has not been conducted. Our aim of this study is to compare skeletal effect of each BMA based on BMD change and the incidence of fracture by a network meta-analysis.

Methods

All the studies investigating the skeletal impact of BMAs on the BMD changes or the risk of fracture in patients on AI were identified using a two-level search strategy. First, databases including PubMed, Cochrane library, and EMBASE were searched through August 11, 2019, using web-based search engines, as shown in Fig. 1. Second, relevant studies were identified through a manual search of secondary sources, including references of initially identified articles, reviews, and commentaries. All references were downloaded for consolidation, elimination of duplicates, and further analyses. Search terms included breast cancer or breast carcinoma or breast neoplasms; aromatase inhibitors or anastrozole or letrozole or exemestane; bisphosphonate or alendronate or risedronate or zoledronate or denosumab or diphosphonates; randomized. Two independent and blinded authors (HM and SS) reviewed the search results separately to select the studies based on inclusion and exclusion criteria. If there was lack of a consensus, a third author (TK) was consulted for final decision [9]. There was no language restriction. The references included in the studies were reviewed to minimize missing relevant studies. The study was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines [10].

Fig. 1
figure 1

Flow diagram for identification of relevant studies

Full size image

The included studies met the following criteria: the study was peer-reviewed by journals, the design was a randomized controlled trial of patients with different preventive strategies for bone loss and fracture (risedronate, zoledronate, denosumab), or no upfront treatment, the study had an end-result of at least one of the followings: BMD change from the baseline (ΔBMD,%) of lumbar spine (LS) or total hip (TH) at 1 year or 2 years, or the risk of fracture. We used the Cochrane risk of bias assessment to explore sources of bias [11]. According to this scale, we evaluated the risk of bias based on random sequence generation, allocation concealment, blinding of participants and researchers, blinding of outcome assessment, selective reporting and incomplete outcome data, and we categorized them as high, low or unclear. To assess the risk of publication bias, a visual estimation of the funnel plot was used. For each study, data regarding BMD and the incidence of fracture in each cohort were abstracted. We used mean difference of BMD adjusted with baseline factors (e.g., age, baseline T-score, or chemotherapy use) if they were reported. If the number of patients in each cohort were reported, we calculated the unadjusted difference of BMD. We performed network meta-analysis using “netmeta” 1.1–0 package (R Foundation for Statistical Computing, Vienna, Austria). We used the random effect model for the analysis. Within the framework, I2 statistics, which represent the proportion of total variation in study estimates due to heterogeneity, were used to quantify heterogeneity [12]. The I2 statistics represent the proportion of variability that is not attributable to chance. Lastly, we conducted sensitivity analyses by excluding a study with the largest number of patients and a study with the most heterogeneity.

Results

We identified 16 eligible studies [13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28], enrolling a total of 7699 patients receiving AI after surgical treatment for breast cancer [The number of patients: Risedronate (n = 312), Zoledronate (n = 1708), Denosumab (n = 1838), No upfront treatment (n = 3841)]. All 16 studies were randomized controlled trials comparing one of the medications with no upfront treatment. The analyzed studies are summarized in Table 1. The median age of the studies was similar. Of note, the studies of risedronate include exclusively patients with baseline T-score below − 1.0, while the studies of the other BMAs included all patients regardless of baseline BMD. Thirty-three percent of patients in the studies of zoledronic acid and 49% of patients in the studies of denosumab had baseline T-score less than − 1.0.

Table 1 Main characteristics and outcomes of the trials included in the analysis
Full size table

The characteristics of the network are shown in Fig. 2. Briefly, ten, eight, ten comparisons were included in the analysis for ΔBMD at 1 year, ΔBMD at 2 years and the incidence of fracture, respectively. The main outcomes are summarized in Table 1. The forest plots show the results of the meta-analysis of ΔBMD of LS at 1 year (Fig. 3). There was a significant heterogeneity in the analysis (I2 = 63.5%, p = 0.008). All treatment groups increased BMD significantly compared with no upfront treatment group at 1 year. At 2-year follow-up, BMDs at LS and TH further increased with all BMAs. Zoledronate and denosumab similarly increased BMD at LS (5.45% and 5.64% at 1 year, and 7.26% and 7.97% at 2 years, respectively), and at TH (3.34% and 4.65% at 1 year, and 3.75% and 5.31% at 2 years, respectively) (Supplementary Table 1, Supplementary Fig. 1A–C). The BMD gains by those two drugs were comparable except that denosumab gained significantly higher BMDs at TH at 2 years compared with zoledronate (0.66% [0.11– 1.21%]) (Supplementary Fig. 1C). In comparison with risedronate, zoledronate and denosumab showed significantly higher BMD increment at LS at 1 year (3.10% [2.23–3.98%] and 3.29% [2.29–4.29%], respectively), as shown in Fig. 3.

Fig. 2
figure 2

Network of bone modifying agents and no upfront treatment for all endpoints. The thickness of lines is proportional to the number of direct comparisons in studies

Full size image
Fig. 3
figure 3

Forrest plot comparing the BMD change (%, lumbar spine) with each BMA at 1 year (MD mean difference, CI confidence interval)

Full size image

We performed a sensitivity analysis excluding a study with the most heterogeneity [17]. The analysis of ΔBMD of LS and TH showed consistent findings with the original analysis (Supplementary Fig. 2). The sensitivity analysis excluding a study with the largest number of patients [13] also showed similar results of BMD changes (Supplementary Fig. 3). However, the superior total hip BMD gain with denosumab at 2 years was not observed in the sensitivity analyses (Supplementary Figs. 2 and 3).

A network meta-analysis of the risk of fracture was also performed, as shown in Fig. 4. There was no significant heterogeneity (I2 = 0%, p = 0.784). Among all BMAs, denosumab and risedronate reduced the incidence of fracture significantly compared with no upfront treatment group (RR 0.51 [0.38–0.67] and RR 0.54 [0.35–0.83], respectively). In particular, denosumab was associated with a lower incidence of fracture compared with all the bisphosphonates including zoledronate (RR 0.60 [0.38–0.94]). Interestingly, zoledronate did not show significant anti-fracture efficacy, although it increased BMDs more than any other bisphosphonates.

Fig. 4
figure 4

Forrest plot comparing the risk of fracture of each BMA (RR relative risk of fracture, CI confidence interval)

Full size image

We performed a subgroup analysis in the patients with a low baseline T-score (T-score < − 1.0). There was no heterogeneity in the studies (I2 = 0%, p = 0.525). Again, risedronate and denosumab use showed a significantly lower risk of fracture compared with no upfront treatment group (RR [95% CI] 0.54 [0.35–0.83] and 0.58 [0.42–0.81], respectively) (Supplementary Fig. 4).

Most studies in this analysis had a low risk of selective reporting, incomplete outcome data, and other risks. However, about half of the RCTs were open-label, which precludes the blinding of participants and researchers. Some RCTs did not specify the randomization protocol, which can lead to a bias in random sequence generation and allocation concealment (Fig. 5). Figure 6 shows the funnel plots for the outcomes that were assessed in this analysis. There was an obvious heterogeneity among the studies using zoledronate with BMD change (LS) as an end-result. Additionally, in the assessment of fracture risk, the studies on risedronate showed asymmetric distribution; the studies with large standard errors were associated with greater efficacy, suggesting the presence of publication bias.

Fig. 5
figure 5

Risk of bias assessment chart

Full size image
Fig. 6
figure 6

Funnel plots for each endpoint

Full size image

Discussion

Our study is the first to analyze the skeletal effect of the different bisphosphonates and denosumab in postmenopausal patients receiving aromatase inhibitor as an adjuvant therapy for breast cancer The salient findings of our study are the following: (1) Our study showed that BMAs significantly increased the BMD of LS and TH at 1 and 2 years. (2) Zoledronate and denosumab were associated with significantly higher BMD gain among all the treatments. (3) Denosumab increased BMDs at cortical-bone-rich skeleton such as hip more than bisphosphonates. (4) Denosumab and risedronate demonstrated a significant fracture risk reduction. (5) Although zoledronate seemed as potent as denosumab in terms of increasing BMDs, the anti-fracture efficacy of zoledronate was unclear.

A previous meta-analysis showed that bisphosphonates (ibandronate, alendronate, risedronate and zoledronate) were associated with an increase in BMD in postmenopausal women receiving adjuvant AI for breast cancer [29]. However, they did not compare the skeletal effect of each medication. Among the bisphosphonates, our analysis found that zoledronate use was associated with significantly more BMD gain than the others. In patients with Crohn disease, zoledronate showed better efficacy as well [30]. However, a meta-analysis of patients with primary osteoporosis did not find a significant superiority of zoledronate in terms of BMD gain compared with the other bisphosphonates [31]. The distinct characteristics of the study subjects may explain this difference. The patients on aromatase inhibitors, like patients with Crohn disease, lose bone at a much higher rate than postmenopausal osteoporosis [6]. The efficacy of zoledronate, which is the most potent bisphosphonate, might appear more prominent in the setting of rapid bone decline.

In terms of fracture risk reduction, no previous study has shown a direct anti-fracture efficacy of bisphosphonates in this specific patient group. Although a large meta-analysis by Early Breast Cancer Trialists' Collaborative Group revealed that adjuvant bisphosphonates reduce fracture risk (RR 0·85 [0·75–0·97]) in patients with early breast cancer, the study group included both patients with and without receiving aromatase inhibitors, and the authors did not compare the anti-fracture efficacy of each bisphosphonate [32]. Another meta-analysis by O’Carrigan et al. analyzed 44 randomized controlled trials and found that bisphosphonates did not lower the risk of fracture regardless of the kind of bisphosphonates, the timing of bisphosphonate use (immediate vs. delayed) or patients’ AI use [33].

In our analysis, we found that risedronate significantly decreased the risk of fracture in breast cancer patients receiving AIs compared with no upfront treatment group. When we analyzed patients with low BMD (T-score < − 1.0), risedronate still demonstrated a significant fracture risk reduction. However, the most potent zoledronate with a superior BMD gain over risedronate, did not show expected anti-fracture efficacy. This might be a reflection of the trials of zoledronate in our analysis, which included patients with all BMDs, whereas the trials of risedronate exclusively studied patients with low BMDs (T-score < − 1.0). Only 33% of patients receiving zoledronate had T-score < − 1.0 at the baseline. Therefore, the low incidence of fracture in studies of zoledronate might lead to an insufficient power to detect anti-fracture efficacy.

Denosumab is a relatively new anti-resorptive agent, and so far, there have been two meta-analyses, which examined the effect of denosumab in patients with osteoporosis [34] and postmenopausal women with a high risk of fracture [35]. Both studies showed that denosumab increased BMD significantly more than the other BMAs. In our study, we also observed consistent findings in this specific group of patients. Of note, a recent meta-analysis compared the effect of denosumab and zoledronate in the same patient group as ours, using both fixed and random effect models. In this analysis, different models resulted in inconsistent findings where a fixed effect model showed a significantly better fracture risk reduction at 36 months with denosumab, but a random effect model did not show any difference in two treatment groups [36]. With a larger number of studies in our analysis, we noted that denosumab use was associated with a significant lower risk of fracture in a random effect model as well.

In terms of potential side effects of long-term use of anti-resorptives, atypical femur fracture (AFF) is particularly concerning as anti-resorptives tend to accumulate in active resorptive areas like skeletal metastasis. The study from MD Anderson reported a very low incidence of AFF (0.05 cases per 100,000 person-year) with bisphosphonate use [37]. In terms of denosumab, a retrospective study reported a higher incidence of AFF (5 cases in 277 patients with a median of 10 doses), but the study was very small and 4 out of 5 cases had been exposed to bisphosphonate prior [38].

To our knowledge, this is the first network meta-analysis comparing different treatments for bone loss in patients with adjuvant AI for breast cancer, comprising of all the currently available studies. The larger number of patients were included (n = 7699). The consistent results from sensitivity analyses and previous meta-analyses support the validity of the findings. The effects of BMAs on BMD were analyzed on the different axial skeletal sites (LS and TH) and different time points (1 year and 2 years). Additionally, our analysis is the first to report the anti-fracture efficacy of risedronate in this specific patient population. The limitation of our study is the heterogeneous characteristics of patients at baseline among the trials, which may impact the effect size of each treatment.

Conclusion

Currently, we do not have a specific recommendation for a particular BMA for postmenopausal breast cancer patients receiving an AI. Our finding revealed that BMAs, especially denosumab and zoledronate, significantly increased the BMD in LS and TH. Notably, denosumab might generate more BMD gain in cortical-bone-rich hip bone than zoledronate. Both risedronate and denosumab reduced the risk of fracture, but the anti-fracture efficacy of zoledronate, the most potent bisphosphonate, remains uncertain. In conclusion, our finding suggests denosumab might be a better option for preserving BMD and preventing fracture in postmenopausal patients receiving AI, as adjuvant therapy for breast cancer.