Adjuvant bisphosphonate treatment in early breast cancer: meta-analyses of individual patient data from randomised trials

Summary Background Bisphosphonates have profound eﬀ ects on bone physiology, and could modify the process of metastasis. We undertook collaborative meta-analyses to clarify the risks and beneﬁ ts of adjuvant bisphosphonate treatment in breast cancer. Methods We sought individual patient data from all unconfounded trials in early breast cancer that randomised between bisphosphonate and control. Primary outcomes were recurrence, distant recurrence, and breast cancer mortality. Primary subgroup investigations were site of ﬁ rst distant recurrence (bone or other), menopausal status (postmenopausal [combining natural and artiﬁ cial] or not), and bisphosphonate class (aminobisphosphonate [eg, zoledronic acid, ibandronate, pamidronate] or other [ie, clodronate]). Intention-to-treat log-rank methods yielded bisphosphonate versus control ﬁ rst-event rate ratios (RRs). Findings We received data on 18 766 women (18 206 [97%] in trials of 2–5 years of bisphosphonate) with median follow-up 5·6 woman-years, 3453 ﬁ rst recurrences, and 2106 subsequent deaths. Overall, the reductions in recurrence (RR 0·94, 95% CI 0·87–1·01; 2p=0·08), distant recurrence (0·92, 0·85–0·99; 2p=0·03), and breast cancer mortality (0·91, 0·83–0·99; 2p=0·04) were of only borderline signiﬁ cance, but the reduction in bone recurrence was more deﬁ nite (0·83, 0·73–0·94; 2p=0·004). Among premenopausal women, treatment had no apparent eﬀ ect on any outcome, but among 11 767 postmenopausal women it produced highly signiﬁ cant reductions in recurrence (RR 0·86, 95% CI 0·78–0·94; 2p=0·002), distant recurrence (0·82, 0·74–0·92; 2p=0·0003), bone recurrence (0·72, 0·60–0·86; 2p=0·0002), and breast cancer mortality (0·82, 0·73–0·93; 2p=0·002). Even for bone recurrence, however, the heterogeneity of beneﬁ t was barely signiﬁ cant by menopausal status (2p=0·06 for trend with menopausal status) or age (2p=0·03), and it was non-signiﬁ cant by bisphosphonate class, treatment schedule, oestrogen receptor status, nodes, tumour grade, or concomitant chemotherapy. No diﬀ erences were seen in non-breast cancer mortality. Bone fractures were reduced (RR 0·85, 95% CI 0·75–0·97; 2p=0·02).


Introduction
Circulating tumour cells can be attracted to surfaces within the bone where they can displace haemopoietic stem cells and bind to the osteoblastic niche. 1 These disseminated malignant cells can remain quiescent for years.Then, for reasons that are not well understood, they can exit this dormant state, start to proliferate, and establish macrometastases in the bone or elsewhere. 2,3][6] For this reason, and because bisphosphonates can be added to the aromatase inhibitor treatment of post menopausal breast cancer to restrict adverse skeletal eff ects of oestrogen deprivation, reliable evidence is needed about the eff ects of bisphosphonates on breast cancer outcomes.
Improvements in bone-metastasis-free survival, diseasefree survival, and overall survival in women with early breast cancer have been reported in some adjuvant trials of oral clodronate 7,8 or of intravenous zoledronic acid. 9,10owever, in other trials of adjuvant bisphos phonates no signifi cant benefi ts were seen in analyses that included all randomised patients, although both planned and exploratory subset analyses suggested benefi ts either in postmenopausal women 11 or in older women. 12,13][16] To help clarify whether adjuvant bisphosphonates reduce the risk of bone and other metastases, and whether menopausal status aff ects effi cacy, we undertook collab orative meta-analyses of all unconfounded randomised trials that compared breast cancer outcomes in those allocated adjuvant bisphosphonate versus those who were not.

Identifi cation of studies and collection of data
8][19] Trials were eligible if they began before 2008 and randomly assigned women between a bisphosphonate of any type, dose, and schedule versus a control group (open label or placebo) with no bisphosphonate, all other treatments being similar in both groups.Information was sought during 2012-14 for each individual patient on date of randomisation, allocated treatment, age, menopausal status, tumour diameter, grade, spread to locoregional lymph nodes, HER2 and oestrogen and progesterone receptor (ER/PR) status, dates and sites of any breast cancer recurrence, other second pri mary cancer, bone fracture, and the date and cause of death.
The main defi nitions and analysis methods are those used in previous EBCTCG reports, [17][18][19] but with some amendments that refl ect the potential eff ect of bisphosphonates on bone metastases (appendix).

Outcomes
The pre-defi ned coprimary endpoints were any recurrence of breast cancer (distant, locoregional, or new primary in the contralateral breast); distant recurrence, ignoring any previous locoregional or contralateral recurrence; and breast cancer mortality (estimated by log-rank subtraction, as in previous EBCTCG reports 18,19 ).
Secondary outcomes were all-cause mortality; death without recurrence; bone recurrence as the fi rst distant recurrence (with or without concurrent other recurrence); other fi rst (extraskeletal) distant recurrence (with all analyses of distant recurrence ignoring any previous locoregional or contralateral recurrence); locoregional recurrence as fi rst event (ipsilateral breast, chest wall, or locoregional lymph nodes); contralateral new primary breast cancer as fi rst event; and any bone fractures.

Statistical analyses
Time-to-event analyses were stratifi ed by age, ER status, nodal status, and trial.Within each stratum, they compared all those allocated bisphosphonate versus all those allocated control, regardless of treatment compliance (yielding intention-to-treat analyses).Log-rank statistics were used to assess the eff ects (bisphosphonate vs control) on various outcomes, and, for each, to estimate fi rst-event rate ratios (RRs) and their CIs.We did statistical analyses using EBCTCG in-house Fortran programs.
Pre-specifi ed primary subgroup investigations were of site of fi rst distant recurrence (bone, other), menopausal status (premenopausal, perimenopausal, post menopausal [natural or induced, either potentially reversibly, using luteinising hormone-releasing hormone analogues, or permanently by oophorectomy] or, if menopausal status was unavailable, years of age, grouped as <45, 45-54, ≥55 years), and class of bisphosphonate (amino bisphosphonate [zoledronic acid, ibandronate, pamidronate, risedronate, alendronate], other [clodronate]).Exploratory investigations were undertaken of potential interactions between treatment effi cacy and ER status, nodal status, histological grade, use or not of adjuvant chemotherapy, and follow-up period.If appropriate, tests comparing eff ects in diff erent subgroups were for trend rather than heterogeneity.
We pre-specifi ed that comparisons of treatment effi cacy within subgroups would exclude local and contralateral recurrence if the prior hypothesis that bisphosphonates would reduce distant but not local or contralateral recurrence was established from analyses of the overall results in all randomised patients.As bone recurrence was the only type of recurrence signifi cantly reduced by bisphosphonates we used this instead as the primary endpoint for subgroup comparisons, but the appendix includes subgroup analyses for any distant recurrence.Because the ABCSG-12 9 and AZURE 11,14 trials had helped generate the hypothesis of the relevance of menopausal status to the eff ects of treatment, we provide sensitivity analyses of this hypothesis that treated these trials as hypothesis-generating, with the remaining trials hypothesis-testing.The policy on data sharing from this study is available online.

Role of the funding source
The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.The writing committee had full access to all the data in the study and had fi nal responsibility for the decision to submit for publication.

Results
Individual patient datasets were provided for 26 trials with 18 766 participants, 97% of all 19 291 women in the 32 completed trials that recorded recurrence data (table, appendix).In four other trials (620 women) recurrence was not recorded, and from the two ongoing trials (2116 women) outcome data cannot yet be provided.Mean scheduled treatment duration was 3•4 years; 18 206 (97%) of 18 766 participants were in trials of 2-5 years of treatment.Median follow-up was 5•6 woman-years (IQR 3•7-8•0).3453 women had a recurrence, after which 2106 died.
We did many subgroup analyses to investigate the eff ects of bisphosphonates on any recurrence, distant recurrence, bone recurrence, and breast cancer mortality (appendix).In the overall analyses, among all 18 766 women, the clearest evidence of eff ect of bis phosphonates was, as anticipated, on bone recurrence, so the most informative subgroup analyses should relate to this endpoint (fi gure 2).The effi cacy of bisphosphonates in reducing bone recurrence appeared to be greater in older women (2p=0•03 for trend with age in treatment eff ect) or, similarly, in postmenopausal women (2p=0•06 for trend with menopausal status).As menopausal status and age are closely correlated, we cannot determine reliably which is more relevant (appendix).Among the 4616 women younger than 45 years, bone recurrence appeared to be unaff ected by the treatment allocation (RR 1•00, 95% CI 0•79-1•26; 2p=0•97), but among the 7388 women 55 years or older there was a highly signifi cant treatment eff ect (RR 0•72, 0•59-0•88; 2p=0•002).Sensitivity analyses of the possible relevance of age and menopausal status that omitted the hypothesis-generating ABCSG-12 9 and AZURE 11,14 studies still showed signifi cant (2p=0•004) benefi t only in post menopausal women (appendix).As was the case for bone recurrence, the reductions in any distant recurrence with bisphosphonate were also signifi cantly greater in older women (2p=0•003 for trend with age) and post menopausal women (2p=0•01).
None of the other subgroup analyses of bone recurrence in fi gure 2 revealed any signifi cant evidence of heterogeneity of benefi t by tumour type (or for other breast cancer outcomes; appendix).Although the benefi t appeared somewhat larger in ER-negative than ER-positive disease and in node-positive than node-negative tumours, this apparent heterogeneity of treatment eff ect did not approach signifi cance and could be a chance fi nding.
Likewise, there was no signifi cant heterogeneity between the apparent eff ects on bone recurrence of the diff erent bisphosphonate regimens tested in these trials.For this outcome, the benefi ts of the nonamino bisphosphonate (clodronate, n=5053) and of the two most widely tested aminobisphosphonates (zoledronic acid, n=9290, and ibandronate, n=3072) appeared similar, but there was no apparent benefi t in the smaller oral pamidronate group (n=953).
For bone recurrence, the benefi ts appeared to be similar in trials of low-intensity anti-osteoporosis schedules (eg, 6-monthly intravenous zoledronic acid) and in trials of more intensive schedules such as those approved for use in metastatic bone disease (eg, monthly zoledronic acid, daily oral ibandronate, or daily oral clodronate).Likewise, the average eff ect appeared similar in trials that tested diff erent durations of treatment (trials of 2 years bisphosphonate vs none: RR 0•76, 95% CI 0•60-0•97; 2p=0•026; trials of 3-5 years bisphosphonate vs none: RR 0•85, 0•73-0•99; 2p=0•037; fi gure 2), and in the presence or absence of chemotherapy.There were signifi cant reductions in bone recurrence during years 0-1 and years 2-4 after randomisation but there appeared to be no further reduction thereafter.Again, though, this decrease in treatment eff ect over time was not signifi cant (trend 2p=0•11), perhaps because there is thus far only limited follow-up after the fi rst 5 years.
To enhance statistical power, the multiple subgroup analyses of bone recurrence (fi gure 2) and the corresponding analyses of other outcomes (appendix) can Error bars are SE .

F
all be restricted to postmenopausal women (appendix).In postmenopausal women, there was signifi cant (p=0•01) heterogeneity between agents in the reductions in bone recurrence, explained by the apparent lack of benefi t from pamidronate.The clodronate results did appear somewhat more promising than the aminobisphosphonate results but this diff erence was not signifi cant for distant recurrence or for bone recurrence, and was of only borderline signifi cance for breast cancer mortality, even though the reduction in postmenopausal breast cancer mortality was signifi cant with clodronate but not with the aggregate of all aminobisphosphonate regimens.Information on fractures was available from only 13 341 (71%) of 18 766 women.Among them, 422 (6•3%) of 6649 bisphosphonate-allocated patients had a fracture reported, as against 487 (7•3%) of 6692 control patients (RR 0•85, 95% CI 0•75-0•97; 2p=0•02; appendix), and the 5-year fracture risk was reduced from 6•3% to 5•1%, with little eff ect in years 0-1 and most of the gain in years 2-4.After year 5 there appeared to be little further gain, but in both groups the absolute rates after year 5 were lower than in years 0-4, perhaps refl ecting incomplete ascertainment.

Discussion
Taking all women together, regardless of menopausal status, this collaborative meta-analysis of individual patient data from 18 766 women randomised in trials of adjuvant bisphosphonates found a highly signifi cant reduction only in bone recurrence, and not in other breast cancer outcomes.Subgroup analyses suggested benefi t just in postmenopausal women, among whom there were highly signifi cant reductions not only in bone recurrence but also in any distant recurrence (bone or other), breast cancer mortality, and overall mortality.
Neither in the overall results nor in the results just among postmenopausal women, however, was there any signifi cant eff ect on distant recurrence at extra-osseous sites, on locoregional recurrence, or on the incidence of contralateral breast cancer.The lack of eff ect on new contralateral breast cancers is consistent with fi ndings of the large FIT and HORIZON-PFT fracture prevention trials, 20 but contrasts with reports from epidemiological studies that breast cancer incidence is reduced in postmenopausal women taking bisphosphonates for osteoporosis. 21,22Thus, the randomised evidence provides no support for the use of bisphosphonates as a breast cancer chemoprevention strategy.
Though the statistical signifi cance of the apparent interaction between menopausal status and treatment effi cacy is not extreme, greater benefi t for postmenopausal women had been hypothesised to explain the apparent discordance between the ABCSG-12 9 and AZURE 11,14 trial results.Sensitivity analyses that excluded these two hypothesis-generating datasets only marginally weakened the evidence of interaction with menopausal status, and the benefi t was still signifi cant in the remaining postmenopausal women.Moreover, there is some preclinical evidence that reproductive hormones can inhibit bisphosphonate effi cacy against cancer cells in the bone.The eff ects of zoledronic acid (100 μg/kg weekly) on the growth of disseminated MDA-231 breast cancer cells in bone were compared in ovariectomised mice (modelling the postmenopausal setting) and in sham-operated mice (modelling the premenopausal setting).Zoledronic acid decreased the number of detectable tumours in bone only in the ovariectomised animals. 23Likewise, in a prostate cancer mouse model the ability of disseminated tumour cells in the bone to form detectable tumours was inhibited by zoledronic acid only in castrated mice, not in shamoperated mice. 24he eff ects on bone recurrence emphasise the potential importance of host microenvironment factors to metastasis.Further studies are needed to clarify why menopausal status should importantly aff ect the response to bisphosphonates.The complex interactions between reproductive hormones, tumour biology, bone cell function, and bone marrow stem cells could well change as patients progress from the premenopausal setting, where oestradiol and inhibin are of major importance in bones, to the postmenopausal setting, where activin and other members of the TGF-β superfamily become the main regulators of bone cell metabolism. 25A clearer understanding of some of the other mechanisms involved in the development of bone metastasis is now emerging, although how these relate to menopausal status and reproductive hormones remains unknown. 26ther than the apparent eff ect of menopausal status or, similarly, age on treatment effi cacy, the proportional reductions in bone recurrence and breast cancer mortality with treatment did not depend signifi cantly on other patient or clinico pathological primary tumour characteristics, including ER status, axillary lymph node involvement, and tumour grade.Similar reductions were seen in the presence and absence of chemotherapy, suggesting that the benefi ts of bisphosphonates are approximately additive to those of chemotherapy, and vice versa.
As subgroup analyses can yield erratic results, it is diffi cult to determine from them whether diff erent bisphosphonate regimens have diff erent eff ects.The endpoint that should yield the most reliable subgroup analyses is bone recurrence.Both for all women and for postmenopausal women, subgroup analyses of bone recurrence suggested similar eff ects of oral clodronate and of the aggregate of all aminobisphosphonate regimens (mainly intravenous zoledronic acid).Likewise, they suggested no signifi cant heterogeneity in effi cacy between the diff erent aminobisphosphonates, though no benefi t was seen with oral pamidronate (which could be real, as oral pamidronate is poorly absorbed, has little eff ect on bone resorption biomarkers or the underlying metastatic bone disease, and failed to show effi cacy in myeloma 27,28 ).Numbers were insuffi cient to assess the effi cacy of the standard treatments for osteoporosis, oral risedronate or alendronate, as therapy for early breast cancer.Subgroup analyses based instead on breast cancer mortality suggested a greater eff ect with clodronate than with aminobisphosphonates.However, as the two drugs appeared to have similar eff ects on bone recurrence, their apparently diff erent eff ects on breast cancer mortality could be a chance fi nding.
Much more reliable comparisons of diff erent bisphosphonate regimens will emerge from ongoing trials that compare them directly.The SWOG0307 trial (NCT00127205) comparing clodronate versus zoledronic acid versus ibandronate in 5400 patients has completed recruit ment and addresses the choice of agent; the SUCCESS trial (NCT02181101) comparing 5 years versus 2 years of zoledronic acid in 3800 patients has also completed recruitment and addresses duration.Similarly, results from two ongoing trials (HOBOE-premenopausal [NCT00412022] and TEAM-IIb [ISRCTN17633610]), plus longer follow-up of the trials included in this metaanalysis, will eventually provide better evidence on any eff ect of bisphosphonates in premenopausal women, and will provide more stable estimates of the 10-year outcomes in postmenopausal women.
Consistent with the known eff ects on bone mineral density and quality, the use of adjuvant bisphosphonates was associated with a small reduction in fracture incidence.Although not highly signifi cant, it can be accepted as real because of evidence of fracture reduction in other types of patient.There was no apparent eff ect of adjuvant bisphosphonates on non-breast cancer mortality.Major adverse events with bisphosphonates are uncommon, but can include impaired renal function and osteonecrosis of the jaw.From the data provided, we were unable to assess the incidence of osteonecrosis of the jaw, but previous reports suggest it ranges from under 1% with clodronate, ibandronate, or 6-monthly zoledronic acid 12,13,29 to about 2% with more intensive zoledronic acid schedules. 30hese trials have shown that some years of adjuvant bisphosphonate treatment can reduce breast cancer recurrence rates in bone and improve breast cancer survival, but have provided clear evidence of benefi t only in women who are postmenopausal (natural or induced) at the time bisphosphonates are started.The use of bisphosphonates in breast cancer is mainly to reduce bone loss and risk of fracture in postmenopausal women with ER-positive disease treated with aromatase inhibitors.Our results show that such bisphosphonate treatment can, in addition, provide oncological benefi t, and suggest that adjuvant bis phosphonates should be considered in a broader range of postmenopausal women.

Figure 1 :
Figure 1: Recurrence by site and breast cancer mortality in 24 trials of bisphosphonate versus no bisphosphonate (control) Kaplan-Meier graphs showing eff ects of treatment allocation on 10-year outcomes in all 18 766 patients.(A) Any recurrence.(B) Distant recurrence.(C) Bone recurrence.(D) Breast cancer mortality.O-E=observed minus expected.V=variance of O-E.RR=rate ratio (exp[{O-E}/V]).Error bars are SE.

AB 9 YearsC 9 YearsFigure 2 :
Figure 2: Multiple subgroup analyses of eff ects on bone recurrence in trials of bisphosphonate versus no bisphosphonate (control)Results are plotted as black squares with horizontal lines that denote 99% rather than 95% CIs to allow for multiple hypothesis testing.Total is plotted as a white diamond that denotes 95% CI.ER=oestrogen receptor.O-E=observed minus expected.

Figure 3 :
Figure 3: Main outcomes in premenopausal (excluding perimenopausal) and in postmenopausal women in trials of bisphosphonate versus no bisphosphonate (control) Kaplan-Meier graphs showing the eff ects of treatment allocation on 10-year breast cancer outcomes.(A) Premenopausal and (B) postmenopausal bone recurrence.(C) Premenopausal and (D) postmenopausal distant recurrence outside the bone.(E) Premenopausal and (F) postmenopausal breast cancer mortality.O-E=observed minus expected.V=variance of O-E.RR=rate ratio (exp[{O-E}/V]).Error bars are SE .