Bone‐modifying agents for reducing bone loss in women with early and locally advanced breast cancer: a network meta‐analysis

Abstract

Background

Bisphosphonates and receptor activator of nuclear factor‐kappa B ligand (RANKL)‐inhibitors are amongst the bone‐modifying agents used as supportive treatment in women with breast cancer who do not have bone metastases. These agents aim to reduce bone loss and the risk of fractures. Bisphosphonates have demonstrated survival benefits, particularly in postmenopausal women.

Objectives

To assess and compare the effects of different bone‐modifying agents as supportive treatment to reduce bone mineral density loss and osteoporotic fractures in women with breast cancer without bone metastases and generate a ranking of treatment options using network meta‐analyses (NMAs).

Search methods

We identified studies by electronically searching CENTRAL, MEDLINE and Embase until January 2023. We searched various trial registries and screened abstracts of conference proceedings and reference lists of identified trials.

Selection criteria

We included randomised controlled trials comparing different bisphosphonates and RANKL‐inihibitors with each other or against no further treatment or placebo for women with breast cancer without bone metastases.

Data collection and analysis

Two review authors independently extracted data and assessed the risk of bias of included studies and certainty of evidence using GRADE. Outcomes were bone mineral density, quality of life, overall fractures, overall survival and adverse events. We conducted NMAs and generated treatment rankings.

Main results

Forty‐seven trials (35,163 participants) fulfilled our inclusion criteria; 34 trials (33,793 participants) could be considered in the NMA (8 different treatment options).

Bone mineral density

We estimated that the bone mineral density of participants with no treatment/placebo measured as total T‐score was ‐1.34. Evidence from the NMA (9 trials; 1166 participants) suggests that treatment with ibandronate (T‐score ‐0.77; MD 0.57, 95% CI ‐0.05 to 1.19) may slightly increase bone mineral density (low certainty) and treatment with zoledronic acid (T‐score ‐0.45; MD 0.89, 95% CI 0.62 to 1.16) probably slightly increases bone mineral density compared to no treatment/placebo (moderate certainty). Risedronate (T‐score ‐1.08; MD 0.26, 95% CI ‐0.32 to 0.84) may result in little to no difference compared to no treatment/placebo (low certainty). We are uncertain whether alendronate (T‐score 2.36; MD 3.70, 95% CI ‐2.01 to 9.41) increases bone mineral density compared to no treatment/placebo (very low certainty).

Quality of life

No quantitative analyses could be performed for quality of life, as only three studies reported this outcome. All three studies showed only minimal differences between the respective interventions examined.

Overall fracture rate

We estimated that 70 of 1000 participants with no treatment/placebo had fractures. Evidence from the NMA (16 trials; 19,492 participants) indicates that treatment with clodronate or ibandronate (42 of 1000; RR 0.60, 95% CI 0.39 to 0.92; 40 of 1000; RR 0.57, 95% CI 0.38 to 0.86, respectively) decreases the number of fractures compared to no treatment/placebo (high certainty). Denosumab or zoledronic acid (51 of 1000; RR 0.73, 95% CI 0.52 to 1.01; 55 of 1000; RR 0.79, 95% CI 0.56 to 1.11, respectively) probably slightly decreases the number of fractures; and risedronate (39 of 1000; RR 0.56, 95% CI 0.15 to 2.16) probably decreases the number of fractures compared to no treatment/placebo (moderate certainty). Pamidronate (106 of 1000; RR 1.52, 95% CI 0.75 to 3.06) probably increases the number of fractures compared to no treatment/placebo (moderate certainty).

Overall survival

We estimated that 920 of 1000 participants with no treatment/placebo survived overall. Evidence from the NMA (17 trials; 30,991 participants) suggests that clodronate (924 of 1000; HR 0.95, 95% CI 0.77 to 1.17), denosumab (927 of 1000; HR 0.91, 95% CI 0.69 to 1.21), ibandronate (915 of 1000; HR 1.06, 95% CI 0.83 to 1.34) and zoledronic acid (925 of 1000; HR 0.93, 95% CI 0.76 to 1.14) may result in little to no difference regarding overall survival compared to no treatment/placebo (low certainty). Additionally, we are uncertain whether pamidronate (905 of 1000; HR 1.20, 95% CI 0.81 to 1.78) decreases overall survival compared to no treatment/placebo (very low certainty).

Osteonecrosis of the jaw

We estimated that 1 of 1000 participants with no treatment/placebo developed osteonecrosis of the jaw. Evidence from the NMA (12 trials; 23,527 participants) suggests that denosumab (25 of 1000; RR 24.70, 95% CI 9.56 to 63.83), ibandronate (6 of 1000; RR 5.77, 95% CI 2.04 to 16.35) and zoledronic acid (9 of 1000; RR 9.41, 95% CI 3.54 to 24.99) probably increases the occurrence of osteonecrosis of the jaw compared to no treatment/placebo (moderate certainty). Additionally, clodronate (3 of 1000; RR 2.65, 95% CI 0.83 to 8.50) may increase the occurrence of osteonecrosis of the jaw compared to no treatment/placebo (low certainty).

Renal impairment

We estimated that 14 of 1000 participants with no treatment/placebo developed renal impairment. Evidence from the NMA (12 trials; 22,469 participants) suggests that ibandronate (28 of 1000; RR 1.98, 95% CI 1.01 to 3.88) probably increases the occurrence of renal impairment compared to no treatment/placebo (moderate certainty). Zoledronic acid (21 of 1000; RR 1.49, 95% CI 0.87 to 2.58) probably increases the occurrence of renal impairment while clodronate (12 of 1000; RR 0.88, 95% CI 0.55 to 1.39) and denosumab (11 of 1000; RR 0.80, 95% CI 0.54 to 1.19) probably results in little to no difference regarding the occurrence of renal impairment compared to no treatment/placebo (moderate certainty).

Authors' conclusions

When considering bone‐modifying agents for managing bone loss in women with early or locally advanced breast cancer, one has to balance between efficacy and safety. Our findings suggest that bisphosphonates (excluding alendronate and pamidronate) or denosumab compared to no treatment or placebo likely results in increased bone mineral density and reduced fracture rates. Our survival analysis that included pre and postmenopausal women showed little to no difference regarding overall survival. These treatments may lead to more adverse events. Therefore, forming an overall judgement of the best ranked bone‐modifying agent is challenging. More head‐to‐head comparisons, especially comparing denosumab with any bisphosphonate, are needed to address gaps and validate the findings of this review.

Plain language summary

Do bone‐modifying medicines help reduce bone loss in women with early or locally advanced breast cancer?

Key messages

• Giving medicines that slow down bone breakdown (bone‐modifying medicines) likely helps make bones stronger and lowers the chance of fractures, but it might also lead to more unwanted effects. 
• Because each treatment has its own benefits and drawbacks, it is difficult to determine which medicine is the best option.
• We need more studies where these medicines are compared directly to each other.

What is bone loss and why do people with cancer develop it?

Bone loss means that more old bone is broken down than new bone is formed, causing an imbalance. Bone‐modifying medicines called bisphosphonates or denosumab can help by slowing down the breakdown of bone. Women with breast cancer are especially prone to bone loss because cancer treatments can weaken bones, making them more susceptible to fractures.

What did we want to find out?

We wanted to identify which medicines work best for reducing bone loss in women with breast cancer and if they have any unwanted effects. We wanted to know whether these medicines:

• make bones stronger ('bone density');
• improve how women feel and function in their daily lives;
• lower the chance of having fractures;
• increase how long women live; and
• have unwanted effects such as damage to the jaw bone ('osteonecrosis of the jaw') or kidney issues.

What did we do?

We looked for studies that compared different bone‐modifying medicines for reducing bone loss resulting from early or locally advanced breast cancer. We compared their findings, summarised the results, and assessed how confident we were in the evidence, based on how the studies were done and how many people took part. We used statistics to compare multiple treatments against each other and rank them based on how well they worked and any unwanted effects.

What did we find?

We found 47 studies involving 35,163 people at different ages and with different stages of breast cancer. They received either bone‐modifying medicines (bisphosphonates or denosumab) or placebo/no treatment, alongside cancer treatment. Cancer treatment may include chemotherapy, endocrine therapy and/or radiotherapy.

Thirty‐four studies involving 33,793 people reported data that could be included in this review. These studies compared eight different bone‐modifying agents to reduce bone loss.

Most bone‐modifying medicines (except alendronate and pamidronate) seem to improve bone density (9 trials; 1166 people) and lower the risk of fractures (16 trials; 19,492 people).

Bone‐modifying medicines do not seem to have an impact on quality of life (only three studies; no quantitative analysis).

This review suggests that these medicines might not affect survival when including both pre and postmenopausal women (17 trials; 30,991 people).

These medicines can cause more unwanted effects such as osteonecrosis of the jaw (12 trials; 23,527 people). The bisphosphonate, clodronate, and denosumab, might not affect kidney function, while ibandronate and zoledronic acid could increase the risk of kidney problems compared to no treatment or placebo (12 trials; 22,469 people).

What are the limitations of the evidence?

Overall, we are moderately confident in the evidence that one treatment is better or worse than another. Our confidence is limited because we sometimes observe conflicting results for the same treatments.

We did not have enough evidence to reach a firm conclusion on which treatments are the best. This is because not all the studies provided the information we needed, so we could not compare every treatment for each outcome.

How up‐to‐date is the evidence?

The evidence is up‐to‐date until January 2023.

Summary of findings

Summary of findings 1. Summary of findings.

Patient or population: adult women (18 years of age and older) with a confirmed diagnosis of early or locally advanced breast cancer Intervention: bone‐modifying agents (alendronate, clodronate, denosumab, ibandronate, pamidronate, risedronate, zoledronic acid) Comparison: no treatment/placebo Settings: inpatient and outpatient care
Outcomes	Anticipated absolute effects (95% CI)1		Relative effects (95% CI)2	Certainty of the evidence (GRADE)3	Interpretation of findings
	Comparator	Intervention
Bone mineral density4 (Subnet based on 9 studies including 1166 participants; outcome measured for a follow‐up between one and five years)	No treatment/placebo Total T‐score: ‐1.34	Alendronate 2.36 (‐3.35 to 8.07)	MD 3.70 (‐2.01 to 9.41)	⊕⊝⊝⊝ very lowbc	We are uncertain whether alendronate increases bone mineral density compared to no treatment/placebo.
		Clodronate n.r.	‐	‐	‐
		Denosumab n.r.	‐	‐	‐
		Ibandronate ‐0.77 (‐1.39 to ‐0.15)	MD 0.57 (‐0.05 to 1.19)	⊕⊕⊝⊝ lowab	Treatment with ibandronate may slightly increase bone mineral density compared to no treatment/placebo.
		Pamidronate n.r.	‐	‐	‐
		Risedronate ‐1.08 (‐1.66 to ‐0.50)	MD 0.26 (‐0.32 to 0.84)	⊕⊕⊝⊝ lowab	Treatment with risedronate may result in little to no difference regarding bone mineral density compared to no treatment/placebo.
		Zoledronic acid ‐0.45 (‐0.72 to ‐0.18)	MD 0.89 (0.62 to 1.16)	⊕⊕⊕⊝ moderateb	Treatment with zoledronic acid probably slightly increases bone mineral density compared to no treatment/placebo.
Quality of life5 (3 studies including 1032 participants; only narrative)	See comment	See comment	Not estimable	See comment	All three studies showed only minimal differences in quality of life between the respective interventions examined.
Fracture rate (Subnet based on 16 studies including 19,492 participants; outcome measured for a follow‐up between one and ten years)	No treatment/placebo 70 per 1000	Alendronate n.r.	‐	‐	‐
		Clodronate 42 per 1000 (27 to 64)	RR 0.60 (0.39 to 0.92)	⊕⊕⊕⊕ high	Treatment with clodronate decreases the number of fractures compared to no treatment/placebo.
		Denosumab 51 per 1000 (36 to 71)	RR 0.73 (0.52 to 1.01)	⊕⊕⊕⊝ moderatea	Treatment with denosumab probably slightly decreases the number of fractures compared to no treatment/placebo.
		Ibandronate 40 per 1000 (27 to 60)	RR 0.57 (0.38 to 0.86)	⊕⊕⊕⊕ high	Treatment with ibandronate decreases the number of fractures compared to no treatment/placebo.
		Pamidronate 106 per 1000 (53 to 214)	RR 1.52 (0.75 to 3.06)	⊕⊕⊕⊝ moderatea	Treatment with pamidronate probably increases the number of fractures compared to no treatment/placebo.
		Risedronate 39 per 1000 (11 to 151)	RR 0.56 (0.15 to 2.16)	⊕⊕⊝⊝ lowc	Treatment with risedronate may decrease or increase the number of fractures compared to no treatment/placebo.
		Zoledronic acid 55 per 1000 (39 to 78)	RR 0.79 (0.56 to 1.11)	⊕⊕⊕⊝ moderatea	Treatment with zoledronic acid probably slightly decreases the number of fractures compared to no treatment/placebo.
Overall survival (Subnet based on 17 studies including 30,991 participants; outcome measured for a follow‐up between two and ten years)	No treatment/placebo 920 per 1000	Alendronate n.r.	‐	‐	‐
		Clodronate 924 per 1000 (907 to 938)	HR 0.95 (0.77 to 1.17)	⊕⊕⊝⊝ lowab	Treatment with clodronate may result in little to no difference regarding overall survival compared to no treatment/placebo.
		Denosumab 927 per 1000 (904 to 944)	HR 0.91 (0.69 to 1.21)	⊕⊕⊝⊝ lowab	Treatment with denosumab may result in little to no difference regarding overall survival compared to no treatment/placebo.
		Ibandronate 915 per 1000 (894 to 933)	HR 1.06 (0.83 to 1.34)	⊕⊕⊝⊝ lowab	Treatment with ibandronate may result in little to no difference regarding overall survival compared to no treatment/placebo.
		Pamidronate 905 per 1000 (862 to 935)	HR 1.20 (0.81 to 1.78)	⊕⊝⊝⊝ very lowbc	We are uncertain whether treatment with pamidronate decreases overall survival compared to no treatment/placebo.
		Risedronate n.r.	‐
		Zoledronic acid 925 per 1000 (909 to 939)	HR 0.93 (0.76 to 1.14)	⊕⊕⊝⊝ lowab	Treatment with zoledronic acid may result in little to no difference regarding overall survival compared to no treatment/placebo.
Adverse event: Osteonecrosis of the jaw (Subnet based on 12 studies including 23,527 participants; outcome measured for a follow‐up between one and eight years)	No treatment/placebo 1 per 1000	Alendronate n.r.	‐	‐	‐
		Clodronate 3 per 1000 (1 to 9)	RR 2.65 (0.83 to 8.50)	⊕⊕⊝⊝ lowc	Treatment with clodronate may increase the occurrence of osteonecrosis of the jaw compared to no treatment/placebo.
		Denosumab 25 per 1000 (10 to 64)	RR 24.70 (9.56 to 63.83)	⊕⊕⊕⊝ moderated	Treatment with denosumab probably increases the occurrence of osteonecrosis of the jaw compared to no treatment/placebo.
		Ibandronate 6 per 1000 (2 to 16)	RR 5.77 (2.04 to 16.35)	⊕⊕⊕⊝ moderated	Treatment with ibandronate probably increases the occurrence of osteonecrosis of the jaw compared to no treatment/placebo.
		Pamidronate n.r.	‐	‐	‐
		Risedronate n.r.	‐	‐	‐
		Zoledronic acid 9 per 1000 (4 to 25)	RR 9.41 (3.54 to 24.99)	⊕⊕⊕⊝ moderated	Treatment with zoledronic acid probably increases the occurrence of osteonecrosis of the jaw compared to no treatment/placebo.
Adverse event: renal impairment (Subnet based on 12 studies including 22,469 participants; outcome measured for a follow‐up between one and eight years)	No treatment/placebo 14 per 1000	Alendronate n.r.	‐	‐	‐
		Clodronate 12 per 1000 (8 to 19)	RR 0.88 (0.55 to 1.39)	⊕⊕⊕⊝ moderatea	Treatment with clodronate probably results in little to no difference regarding occurrence of renal impairment compared to no treatment/placebo.
		Denosumab 11 per 1000 (8 to 17)	RR 0.80 (0.54 to 1.19)	⊕⊕⊕⊝ moderatea	Treatment with denosumab probably results in little to no difference regarding occurrence of renal impairment compared to no treatment/placebo.
		Ibandronate 28 per 1000 (15 to 54)	RR 1.98 (1.01 to 3.88)	⊕⊕⊕⊝ moderated	Treatment with ibandronate probably increases occurrence of renal impairment compared to no treatment/placebo.
		Pamidronate n.r.	‐	‐	‐
		Risedronate n.r.	‐	‐	‐
		Zoledronic acid 21 per 1000 (12 to 36)	RR 1.49 (0.87 to 2.58)	⊕⊕⊕⊝ moderatea	Treatment with zoledronic acid probably increases occurrence of renal impairment compared to no treatment/placebo.
1 Baseline risks obtained from the respective study population. For bone mineral density the average total T‐Score from all control groups is used. For overall survival the 5‐year‐survival from Team IIB 2006 is used. 2 Results from network meta‐analysis (random‐effects model). Network estimates are reported as risk ratio, hazard ratio or mean difference with corresponding 95% confidence interval. 3 We would have downgraded for risk of bias, only if the sensitivity analysis excluding studies with high risk of bias yielded different results than the main analysis. 4 Bone mineral density measured as T‐score. The T‐score is a standard deviation and measures how much the bone mineral density differs from that of an average healthy young woman. A T‐score between +1 and ‐1 ist considered normal or healthy. The greater the negative T‐score, the lower the bone mass; the higher the positive T‐Score, the higher the bone mass. A T‐score of ‐2.5 or lower indicates osteoporosis. 5 Quality of life was reported poorly in the included studies. Only three studies reported results, so no quantitative analysis could be performed. a Downgraded one level for imprecision since 95% CI is wide and/or crosses unity b Downgraded one level for inconsistency (heterogeneity) c Downgraded two levels for imprecision since 95% CI is very wide and crosses unity d Downgraded one level for imprecision, although 95% CI is very wide, but it does not cross unity and shows a very strong effect CI: confidence interval; HR: hazard ratio; MD: mean difference; n.r.: not reported; RR: risk ratio
GRADE Working Group grades of evidence (or certainty in the evidence) High quality: We are very confident that the true effect lies close to that of the estimate of the effect Moderate quality: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low quality: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low quality: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

Background

Description of the condition

Breast cancer remains the most common cancer amongst women, accounting for 11.6% of all cancer occurrences (Bray 2018). In 2018, there were more than 2 million new cases of breast cancer worldwide. In the same year 626,679 women died from the disease (Bray 2018).

While long‐term outcomes are improving for women with breast cancer, rates of recurrence and death are still significantly high (Aft 2012; Dhesy‐Thind 2017). Anti‐cancer therapy can lead to a reduction in bone density and a higher risk of osteoporosis (Gnant 2009; Gnant 2011; Hadji 2014). Based on a guideline for supportive care in oncology, the incidence rate of breast cancer patients developing osteoporosis is 18.07 per 1000 person‐years (meaning, on average, 18.07 patients develop osteoporosis in 1000 breast cancer patients observed for one year; Khan 2011), which significantly enhances the risk of fractures. The incidence rate is largely influenced by factors such as therapy‐induced menopause as well as oestrogen suppression therapy. Supportive treatments (e.g. corticosteroids) may also harm the bone (Greep 2003; Hadji 2009). Osteoporosis may lead to bone fracture and this, in turn, may require surgery. Bone fracture is associated with an increased mortality rate; for example, a study by Kanis and colleagues found that men and women aged 50 years or older who had suffered a hip fracture had a much higher mortality rate soon after the event (Kanis 2003). In the same study, the five‐year mortality rates increased in all age groups, compared to the general population in Sweden (Kanis 2003). That is why it is important to reduce the loss of bone density in breast cancer patients (Kanis 2008).

Premenopausal women with hormone receptor‐positive breast cancer receive treatment that involves tamoxifen alone, or suppression of ovarian function in combination with tamoxifen or aromatase inhibitors, all of which can lead to loss of bone density and an increased risk of osteoporosis (Gnant 2009; Gnant 2011; Hadji 2014; Hadji 2017). Postmenopausal women treated with aromatase inhibitors are also affected and show an increased fracture rate compared to women treated with tamoxifen (Gnant 2015; Hadji 2011; Kalder 2014; Rabaglio 2009). Chemotherapy also leads to loss of bone density (Greep 2003; Hadji 2009).

Description of the intervention

Bone‐targeted treatment is important for breast cancer patients without metastases (metastasis means cancer that has spread from a primary site, e.g. from the breast to the bone). Bone‐modifying agents like bisphosphates or receptor activator of nuclear factor‐kappa B ligand (RANKL)‐inhibitors may reduce cancer treatment‐induced bone loss (CTIBL). In patients with postmenopausal status (those who are either postmenopausal, or premenopausal with suppressed ovarian function), bone‐modifying agents may also prevent bone metastases and prolong overall survival (EBCTCG 2015; Gnant 2015). Bisphosphonates of interest are alendronate, clodronate, ibandronate, pamidronate, risedronate and zoledronate; the RANKL‐inhibitor of interest is denosumab.

How the intervention might work

To accomplish its functions, bone undergoes continuous destruction (resorption) carried out by osteoclasts, and formation by so‐called osteoblasts (Rodan 1998). Bisphosphonates are analogues of pyrophosphate that target osteoclastic cells, and are grouped into amino‐bisphosphonates or non‐amino‐bisphosphonates. Amino‐bisphosphonates affect the osteoclast metabolism by targeting the farnesyl diphosphate synthase, which is responsible for post‐translational modification of guanosine‐5'‐triphosphate‐binding proteins. Non‐amino‐bisphosphonates function by forming an analogue of adenosine triphosphate. The resulting metabolite has toxic properties and induces apoptosis (programmed cell death) of osteoclasts (Reyes 2016). Bisphosphonates therefore suppress bone resorption by promoting the apoptosis of osteoclasts. Bisphosphonates vary in terms of route of administration (oral or intravenous), dose, and frequency and duration of administration.

Denosumab, a fully humanised monoclonal antibody, functions by targeting and neutralising RANKL, which has been found to be a major contributor to the progression of bone metastases (Hanley 2012). RANKL is expressed by osteoblastic stromal cells and binds to RANK, thereby mediating osteoclast differentiation, activation and survival. RANKL is responsible for osteoclast‐mediated bone resorption (Hsu 1999; Yasuda 1998). Denosumab binds to RANKL with high affinity and blocks the interaction of RANKL and RANK. This action decreases bone resorption since the route mediating differentiation, activation and survival of bone resorptive osteoclasts is blocked (Bekker 2004). Denosumab is given subcutaneously at a fixed dose.

Since bisphosphonates and denosumab influence bone metabolism, there could be adverse events such as osteonecrosis of the jaw (a severe bone disease affecting the jaw bone, characterised by delayed wound healing after invasive procedures, as well as infection and death of the bone tissue) and hypocalcaemia (blood calcium levels under the normal range of 2.1 millimoles per litre (mmol/L) to 2.6 mmol/L, which may lead to further complications) (Tesfamariam 2019). Bisphosphonates are also known to have the ability to cause renal complications (Edwards 2013).

Why it is important to do this review

Bisphosphonates are recommended as an addition to usual treatment for postmenopausal women with breast cancer, since they may reduce the risk of fractures and bone recurrence (meaning the development of bone metastases), and prolong overall survival; they may also reduce therapy‐induced bone loss in pre and postmenopausal patients (Dhesy‐Thind 2017; EBCTCG 2015; Gnant 2015; National Guideline Alliance 2018; O'Carrigan 2017). Denosumab has also been found to reduce fractures, but research on long‐term survival is still ongoing (Dhesy‐Thind 2017). An overall ranking of all different treatment options, focusing on different patient‐relevant outcomes, is still missing. Therefore, a comparison is urgently needed to inform recommendations in national and international guidelines (Dhesy‐Thind 2017), and to help patients and healthcare providers with decision‐making.

Objectives

To assess and compare the effects of different bone‐modifying agents as supportive treatment to reduce bone mineral density loss and osteoporotic fractures in women with breast cancer without bone metastases and generate a ranking of treatment options using network meta‐analyses (NMAs).

Methods

Criteria for considering studies for this review

Types of studies

We included studies if they were randomised controlled trials (RCTs). We included both full‐text and abstract publications if sufficient information on study design, characteristics of participants (women with early or locally advanced breast cancer) and interventions (adjuvant bisphosphonates or RANKL‐inhibitors) were provided. We included trials that included participants receiving these bone‐modifying agents in at least one treatment arm. In the case of cross‐over trials, only the first period of the trial has been analysed. There were no limitations with respect to length of follow‐up.

We excluded studies that were non‐randomised, case reports, and clinical observations.

Types of participants

We included trials involving adult women (18 years of age and older) with a confirmed diagnosis of early or locally advanced breast cancer, meaning all stages without metastases, defined by the TNM Classification of Malignant Tumors (TNM) staging system showing any stages for T, any for N and 0 for M. In this staging system 'T' refers to the size and extent of the main (primary) tumour, 'N' refers to the number of nearby lymph nodes that have cancer and 'M' refers to whether the cancer has metastasised, which means that the cancer has spread from the primary tumour to other parts of the body (NCI 2015). By including M0, only studies with non‐metastasised participants have been included in this analysis. We included both pre and postmenopausal participants. We assumed that participants who fulfilled the inclusion criteria were equally eligible to be randomised to any of the interventions we compared.

Types of interventions

We included trials comparing different bone‐modifying agents with each other and with control regimens (placebo or no treatment) as adjuvant therapy for early or locally advanced breast cancer. We considered any type of bisphosphonate or RANKL‐inhibitor, apart from radioactive bisphosphonates. We did not impose any restriction on the dose, route, frequency or duration of treatment with bone‐modifying agents. We investigated the following comparisons. In order to establish fair comparisons, concomitant treatments did not differ between study arms.

Interventions

• Bisphosphonates (alendronate, clodronate, ibandronate, pamidronate, risedronate and zoledronate)

• RANKL‐inhibitors (denosumab)

• Placebo/no treatment

Comparisons

• Bisphosphonates versus placebo/no treatment

• RANKL‐inhibitors versus placebo/no treatment

• Bisphosphonates versus RANKL‐inhibitors

• One type of bisphosphonate versus another type of bisphosphonate

The options, bisphosphonates and RANKL‐inhibitors, are recommended in clinical guidelines to reduce therapy‐induced bone events in breast cancer patients in early or locally advanced stages. We grouped interventions by evaluating different drug doses together as one drug of interest, according to the product characteristics.

For the patient population described above, all interventions and combinations of interventions have been analysed within a full network (for ideal network see Figure 1). We included all RCTs comparing at least two study arms with one intervention of interest. When no direct evidence from RCTs existed and the trials were considered sufficiently similar with respect to the patient population, indirect estimates of intervention effects have been obtained by means of network calculations.

1.

Figure 1: Ideal network with all possible comparisons of all treatment options

Types of outcome measures

We included all trials fulfilling the inclusion criteria mentioned above, irrespective of reported outcomes. We estimated the relative ranking of the competing interventions according to each of the following outcomes. To make sure outcomes were patient‐relevant, we held a meeting with patients and patient representatives during the development of the protocol, and discussed the relevance and order of proposed outcomes.

Primary outcomes

• Bone density, defined as the amount of minerals (mostly calcium and phosphorous) contained in a certain volume of bone. Analysis of this outcome is intended to determine the potential prevention of cancer treatment‐induced bone loss (CTIBL).

• Quality of life.

Secondary outcomes

• Overall fracture rate, defined as the number of bone fractures of all kinds occurring during and after treatment with bone‐modifying agents. If possible, we subclassified this outcome by site of fracture (vertebral and non‐vertebral).

• Overall survival (time‐to‐event outcome) or all‐cause mortality.

• Disease‐free survival (time‐to‐event outcome), defined as the length of time from diagnosis to the patient surviving without any signs or symptoms (distant, locoregional, or new primary symptoms in the contralateral breast, or as defined in the trial).

• Adverse events:

  • osteonecrosis of the jaw;

  • renal (we considered all trials reporting renal adverse events; as bone‐modifying agents have been associated with renal toxicity with variable expression, we considered renal adverse events to be clinically significant and requiring treatment or hospital admission);

  • bone pain;

  • hypocalcaemia.

• Any bone recurrence, meaning the development of metastasis to the bone.

Method and timing of outcome measurement

• Bone density (or bone mineral density): assessed by using validated techniques, e.g. dual‐energy X‐ray absorptiometry, and given as mass of mineral per volume of bone. We analysed this outcome at the longest reported follow‐up. Unfortunately, we were not able to analyse this outcome at six months, one year and two years separately, since most studies did not report bone density at different time points.

• Quality of life: assessed using validated generic and disease‐specific questionnaires. This outcome should be analysed at six months, one year, two years, or at the longest reported follow‐up. Nevertheless, we were not able to analyse this outcome quantitatively.

• Overall fracture rate: assessed by radiographic imaging and, if indicated, computed tomography or magnetic resonance imaging. We analysed fractures occurring at any time after participants were randomised to intervention or comparator groups.

• Adverse events (osteonecrosis of the jaw, renal adverse events, hypocalcaemia, bone pain). We measured this outcome at any time after participants were randomised to intervention or comparator groups.

• Overall survival or all‐cause mortality: defined as the time from randomisation to the date of death. If we were unable to retrieve the necessary information to analyse time‐to‐event outcomes, we assessed the number of events per treatment group for dichotomised outcomes. We paid special attention regarding the transitivity assumption when it came to the patient population and different treatments (see Assessment of heterogeneity), e.g. we set date limits in order to compare more recent studies with each other rather than against older ones, since treatment of breast cancer has changed and older regimens might not have been comparable with newer regimens. We analysed this outcome at the longest reported follow‐up. Unfortunately, we were not able to analyse this outcome at six months, one year and two years separately since most studies did not report overall survival at different time points.

Search methods for identification of studies

Electronic searches

We conducted systematic searches for the following databases from inception until 17 January 2023 without any language restrictions.

• The Cochrane Breast Cancer Group's (CBCG's) Specialised Register. The process of identifying studies and coding references is outlined on the CBCG's website (breastcancer.cochrane.org/sites/breastcancer.cochrane.org/files/public/uploads/specialised_register_details.pdf). Trials with the keywords 'early and locally advanced stages', 'bone‐modifying agents', 'bisphosphonates', 'RANKL‐inhibitors' and 'bone density' have been extracted and considered for inclusion in the review;

• CENTRAL (the Cochrane Library, issue 01, 2023);

• MEDLINE (via OvidSP) from 1946 to 17 January 2023;

• Embase (via Embase.com) from 1947 to 17 January 2023;

• The WHO International Clinical Trials Registry Platform (ICTRP) (who.int/trialsearch/);

• ClinicalTrials.gov (clinicaltrials.gov/);

• WHO ICTRP and ClinicalTrials.gov have been searched for all prospectively registered and ongoing trials. All search strategies are included in Appendix 1.

Searching other resources

We identified further studies from reference lists of identified relevant trials or reviews. A copy of the full article for each reference reporting a potentially eligible trial has been obtained. Where this was not possible, attempts have been made to contact trial authors to provide additional information.

Data collection and analysis

Selection of studies

We applied Cochrane’s Screen4Me workflow to help assess the search results. Screen4Me comprises three components:

• known assessments, a service that matches records in the search results to records that have already been screened in Cochrane Crowd (Cochrane’s citizen science platform where the Crowd help to identify and describe health evidence) and labelled as 'RCT' or 'not an RCT';

• the RCT classifier, a machine‐learning model that distinguishes RCTs from non‐RCTs;

• Cochrane Crowd, if appropriate (crowd.cochrane.org).

More detailed information regarding evaluations of the Screen4Me components can be found in the following publications: Marshall 2018; McDonald 2017; Noel‐Storr 2018; Thomas 2017.

Two out of four review authors (TJ, ME, AA, AH) each independently screened the results for eligibility for this review by reading the abstracts using Covidence software (Covidence). We coded the abstracts as either 'retrieve' or 'do not retrieve'. In the case of disagreement, or if it was unclear whether we should retrieve the abstract or not, we obtained the full‐text publication for further discussion. Independent review authors eliminated studies that clearly did not satisfy the inclusion criteria, and obtained full‐text copies of the remaining studies. Two out of four review authors (TJ, ME, AA, AH) each read these studies independently to select relevant studies; in the event of disagreement, a third author adjudicated (NS or AW). We did not anonymise the studies before the assessment. We included a PRISMA flow diagram in the full review that shows the status of identified studies (Moher 2009), as recommended in Section 11.2.1 of the Cochrane Handbook for Systematic Reviews of Interventions (Schuenemann 2021). We included studies in the Characteristics of included studies table irrespective of whether measured outcome data were reported in a ‘usable’ way.

There were no language restrictions and articles were translated if they were not published in English. We recorded the details of excluded studies in the Characteristics of excluded studies table.

Data extraction and management

Two out of five review authors (TJ, JCV, MK, AA, AH) each extracted data using a standardised data extraction form. If the authors were unable to reach a consensus, we consulted another review author (NS or AW) for a final decision. If required, we contacted the authors of specific studies for supplementary information as recommended in Chapter five of the Cochrane Handbook for Systematic Reviews of Interventions (Li 2021). After agreement was reached, we entered data into Review Manager Web (RevMan Web 2023). We extracted the following information.

• General information: author, title, source, publication date, country, language, duplicate publications;

• Quality assessment: sequence generation, allocation concealment, blinding (participants, personnel, outcome assessors), incomplete outcome data, selective outcome reporting, other sources of bias;

• Study characteristics: trial design, aims, setting and dates, source of participants, inclusion/exclusion criteria, comparability of groups, statistical methods, power calculations, subgroup analysis, treatment cross‐overs, compliance with assigned treatment, time point of randomisation, length of follow‐up;

• Participant characteristics: participant details, baseline demographics, age, ethnicity, number of participants recruited/allocated/evaluated, participants lost to follow‐up, cancer type and stage, hormone receptor status, HER2 status, additional diagnoses, type and intensity of pain, menopausal status;

• Interventions: type, dose, route, frequency and cycles of treatment with bone‐modifying agents, producer information, prescription and administration of vitamin D and calcium, main cancer treatment (endocrine therapy, chemotherapy) and its details;

• Outcomes: bone density and measuring instrument, quality‐of‐life and measuring instrument, fracture rate (vertebral or non‐vertebral), overall survival or all‐cause mortality, disease‐free survival, adverse events (osteonecrosis of the jaw, renal, bone pain, hypocalcaemia), bone recurrence;

• Notes: sponsorship/funding for trials and notable conflicts of interest of authors' trial registry record information (e.g. national clinical trial numbers).

We collated multiple reports of the same study, so that each study rather than each report was the unit of interest in the review. We collected characteristics of the included studies in sufficient detail to populate a table of Characteristics of included studies in the full review.

Assessment of risk of bias in included studies

We completed a Risk of bias table for each included study, using Review Manager Web (RevMan Web 2023).

Two out of five review authors (TJ, JCV, MK, AA, AH) each independently assessed the risk of bias for each study and, if they were unable to reach a consensus, we consulted a third review author (NS or AW) for a final decision. We assessed the following criteria, as outlined in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).

• sequence generation;

• allocation concealment;

• blinding (participants and personnel);

• blinding (outcome assessors);

• incomplete outcome data;

• selective outcome reporting;

• other sources of bias.

For each of the domains listed above, we judged each study to be at either high, low or unclear risk of bias. We presented these judgements in the Risk of bias tables, along with justification for our decision.

For performance bias (blinding of participants and personnel) and detection bias (blinding of outcome assessment), we evaluated the risk of bias separately for each outcome. We grouped these outcomes according to whether they were measured subjectively or objectively when reporting our findings in the Risk of bias tables.

We defined the outcome 'overall survival' as not being influenced by blinding of patients or outcome assessors (objective outcome). We defined the following outcomes as not being influenced by blinding of patients (objective outcomes):

• bone density;

• overall fracture rate;

• disease‐free survival;

• adverse event: osteonecrosis of the jaw;

• adverse event: renal;

• adverse event: hypocalcaemia;

• any bone recurrence.

We defined the following outcomes as subjective outcomes:

• quality of life;

• adverse event: bone pain.

Measures of treatment effect

Relative treatment effect

We used intention‐to‐treat data. For binary outcomes, we extracted the number of patients and number of events per arm and calculated risk ratios (RRs) with 95% confidence intervals (CIs) for each trial. This applied to the outcomes: overall fracture rate (fractures might be defined as evident fractures with associated symptoms or as defined in the trials themselves, while vertebral fractures might be defined as 20% to 25% or greater height reduction, measured by radiographs), bone density (assessed with dual‐energy X‐ray absorptiometry scans, or as reported in trials if different), and adverse events like osteonecrosis of the jaw and renal adverse events.

We calculated continuous outcomes as mean differences (MDs) when assessed with the same instruments; otherwise we calculated standardised mean differences (SMDs) with 95% CIs.

For time‐to‐event outcomes, we extracted hazard ratios (HRs) from published data, according to Parmar 1998 and Tierney 2007. This applied to the outcomes, overall and disease‐free survival.

Since we conducted a network meta‐analysis, we defined the direction for every RR or HR we were reporting and added 'RR or HR smaller than 1.0 favours...' when reporting results. We did not report pairwise meta‐analysis results since these have been shown elsewhere (O'Carrigan 2017).

Relative treatment ranking

We obtained a treatment hierarchy using P‐scores (Rücker 2015). P‐scores allow ranking of treatments on a continuous zero‐to‐one scale in a frequentist network meta‐analysis.

Unit of analysis issues

Studies with multiple treatment groups

As recommended in Chapter 23.3.4 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021), for studies with multiple treatment groups, we would have combined arms as long as they could have been regarded as subtypes of the same intervention.

When arms could not be pooled this way, we included multi‐armed trials using a network meta‐analysis approach that accounted for the within‐study correlation between the effect sizes by re‐weighting all comparisons of each multi‐armed study (Rücker 2012; Rücker 2014). For pairwise meta‐analysis, we treated multi‐armed studies as multiple independent comparisons and did not combine these data in any analysis.

Dealing with missing data

As suggested in Chapter 10.12 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021), we took the following steps to deal with missing data.

If the number of participants evaluated for a given outcome was not reported, we used the number of participants randomised per treatment arm as the denominator. If only percentages but no absolute number of events were reported for binary outcomes, we calculated numerators using percentages. If estimates for mean and standard deviations were missing, we calculated these statistics from reported data whenever possible, using approaches described in Chapter 5.6 of the Cochrane Handbook for Systematic Reviews of Interventions (Li 2021). If standard deviations were missing, and we were not able to calculate them from reported data, we calculated the values according to a validated imputation method (Furukawa 2006). If data were not reported numerically but graphically, we estimated missing data from figures. We performed sensitivity analyses to assess how sensitive results were to imputing data in some way. We addressed the potential impact of missing data on the findings of the review in the Discussion section.

Assessment of heterogeneity

Assessment of clinical and methodological heterogeneity within treatment comparisons

To evaluate the presence of clinical heterogeneity, we generated summary statistics for the important clinical and methodological characteristics across all included studies. Within each pairwise comparison, we assessed the presence of clinical heterogeneity by visually inspecting the similarity of these characteristics.

Assessment of transitivity across treatment comparisons

To infer the assumption of transitivity, we assessed whether the included interventions were similar when they were evaluated in RCTs with different designs; for example, whether combinations of two drugs are administered the same way in studies comparing them to other combinations of two drugs and in those comparing combinations of two drugs to combinations of three drugs. Furthermore, we compared the distribution of the potential effect modifiers across the different pairwise comparisons.

Assessment of statistical heterogeneity and inconsistency

To evaluate the presence of heterogeneity and inconsistency in the entire network, we used the generalised heterogeneity statistic Q_total and the generalised I² statistic, as described in Schwarzer 2015. We used the decomp.design command in the R package netmeta 1.0‐1 (R 2021; netmeta 2023) for decomposition of the heterogeneity statistic into a Q statistic for assessing the heterogeneity between studies with the same design and a Q statistic for assessing the design inconsistency to identify the amount of heterogeneity/inconsistency within, as well as between, designs.

To evaluate the presence of inconsistency locally, we compared direct and indirect treatment estimates of each treatment comparison. This served as a check for consistency of a network meta‐analysis (Dias 2010). For this purpose, we used the netsplit command in the R package netmeta 1.0‐1, which enabled the splitting of the network evidence into direct and indirect contributions (R 2021; netmeta 2023). For each treatment comparison, we presented direct and indirect treatment estimates plus the network estimate using forest plots. In addition, for each comparison we gave the z value and P value of the test for disagreement (direct versus indirect). It should be noted that in a network of evidence there may be many loops, and with multiple testing there is an increased likelihood that we might find an inconsistent loop by chance. Therefore, we have been cautious when deriving conclusions from this approach.

If we found substantive heterogeneity or inconsistency (or both), we explored possible sources by performing prespecified sensitivity and subgroup analyses (see below). In addition, we reviewed the evidence base, reconsidered inclusion criteria and discussed the potential role of unmeasured effect modifiers to identify further sources.

We interpreted I² values according to Chapter 10.10.2 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021), as follows.

• 0% to 40%, might not be important.

• 30% to 60% may represent moderate heterogeneity.

• 50% to 90% may represent substantial heterogeneity.

• 75% to 100% represents considerable heterogeneity.

We used the P value of the Chi² test only for describing the extent of heterogeneity and not for determining statistical significance. In addition, we reported Tau², the between‐study variance in random‐effects meta‐analysis. In the event of excessive heterogeneity that was unexplained by subgroup analyses, we did not report outcome results as the pooled effect estimate of the network meta‐analysis, but provided a narrative description of the results of each study.

Assessment of reporting biases

In pairwise comparisons with at least 10 trials, we examined the presence of small‐study effects graphically by generating funnel plots. We used linear regression tests (Egger 1997) to test for funnel plot asymmetry. A P value less than 0.1 has been considered significant for this test (Sterne 2011). We additionally conducted comparison‐adjusted funnel plots (Chaimani 2012) and the accompanying regression test to assess selection bias. We examined the presence of small‐study effects for the primary outcome only. Moreover, we searched study registries to identify trials that were completed but not published.

Data synthesis

Methods for direct treatment comparisons

Pairwise comparisons are part of the network meta‐analysis, thus we did not perform additional pairwise meta‐analyses. In order to outline available direct evidence, we provided forest plots for pairwise comparisons, without giving an overall estimate. Only in the case where data were not sufficient to be combined in a network meta‐analysis, e.g. in the case of inconsistency, we performed pairwise meta‐analyses according to recommendations provided in Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021). We used the random‐effects model. We used the R package meta for statistical analyses (Balduzzi 2019; R 2021). When trials were too clinically heterogenous to be combined, we performed only subgroup analyses without calculating an overall estimate.

Methods for indirect and mixed comparisons

When the data had been considered sufficiently similar to be combined, we performed a network meta‐analysis using the frequentist weighted least squared approach described by Rücker 2012. We used a random‐effects model, taking into account the correlated treatment effects in multi‐arm studies. We assumed a common estimate for the heterogeneity variance across the different comparisons. To evaluate the extent to which treatments were connected, we undertook a network plot for our primary and secondary outcomes. For each comparison, we gave the estimated treatment effect along with its 95% CI. We presented the results graphically using forest plots, with placebo/no treatment as the reference treatment. We used the R package netmeta for statistical analyses (R 2021; netmeta 2023).

Certainty in the evidence

Two review authors (AA, NS) independently rated the certainty of the evidence for each outcome. We used the GRADE system to rank the certainty in the evidence, using the guidelines provided in Chapter 14.2 of the CochraneHandbook for Systematic Reviews of Interventions (Schuenemann 2021) and specifically for network meta‐analyses (Puhan 2014).

The GRADE approach to assess the certainty in the body of evidence for each outcome by performing network meta‐analysis uses five domains: study limitations (risk of bias of included studies); indirectness (relevance to the review question); inconsistency (assessment of heterogeneity and incoherence); imprecision (e.g. confidence intervals); and publication bias.

The GRADE assessment of the evidence for each outcome results in one of the four following categories.

• High certainty: we are very confident that the true effect lies close to that of the effect estimate.

• Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the effect estimate, but there is a possibility that it is substantially different.

• Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the effect estimate.

• Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the effect estimate.

The GRADE system uses the following criteria for assigning a GRADE level to a body of evidence (Schuenemann 2021).

• High quality: randomised trials; or double‐upgraded observational studies.

• Moderate quality: downgraded randomised trials; or upgraded observational studies.

• Low quality: double‐downgraded randomised trials; or observational studies.

• Very low quality: triple‐downgraded randomised trials; or downgraded observational studies; or case series/case reports.

We downgraded our assessment of the evidence by one or two points if there was:

• limitation to study quality;

• inconsistency;

• uncertainty about directness;

• imprecise or sparse data;

• high probability of reporting bias.

Subgroup analysis and investigation of heterogeneity

The following subgroup analysis for network meta‐analysis was conducted on all efficacy and safety outcomes.

• Premenopausal participants versus postmenopausal participants (premenopausal patients with ovarian suppression therapy (gonadotropin‐releasing hormone antagonist (GnHR) analogues, ovariectomy, radiomenolysis) were still considered as premenopausal).

The following subgroup analyses for network meta‐analysis were planned to be conducted on all efficacy and safety outcomes, but unfortunately data were not available to perform them.

• Participants receiving endocrine therapy versus those not receiving endocrine therapy, or hormone receptor (HR)‐positive versus HR‐negative, also compared to human epidermal growth factor receptor 2 (HER2)‐positive. Because of lack of data availability, this subgroup analysis could not be performed.

• Type of endocrine therapy (e.g. tamoxifen alone versus aromatase inhibitor alone versus ovarian function suppression (OFS) in combination with tamoxifen versus OFS in combination with aromatase inhibitor). Because of lack of data availability, this subgroup analysis could not be performed.

• Type of bone‐modifying agent (bisphosphonate versus RANKL inhibitor). Since this corresponds to the main analysis using network meta‐analysis, no subgroup analysis was carried out.

• Bisphosphonates of the first (non‐amino bisphosphonates: etidronate, clodronate) and second generation (amino‐bisphosphonates: alendronate, risedronate, pamidronate, ibandronate, zoledronate), independently. Since this would only exclude clodronate from the network, we did not conduct this analysis.

• Duration of bone‐modifying interventions: one year versus two‐to‐five years. Since, in most studies, the duration of bone‐modifying agents was longer than one year, we did not conduct this subgroup analysis.

• Participants with high risk of relapse (defined as receiving chemotherapy additionally to endocrine therapy) versus participants only receiving endocrine therapy. Because of lack of data availability, this subgroup analysis could not be performed.

• Participants with status N1, N2, N3 versus status N0. Because of lack of data availability, this subgroup analysis could not be performed.

Sensitivity analysis

We performed a sensitivity analysis to test the robustness of our results by analysing studies with low risk of bias only. We judged studies as being at high risk of bias overall if they were at high risk for two or more of the Risk of bias domains.

Summary of findings and assessment of the certainty of the evidence

We included a summary of findings table to present the main findings in a transparent and simple tabular format. In particular, we included key information concerning the certainty in the evidence, the magnitude of effect of the interventions examined, and the sum of available data on the outcomes defined under 'Outcomes to be included in the Summary of findings table'.

Outcomes to be included in the Summary of findings table

• bone density;

• quality of life;

• fracture rate;

• overall survival;

• adverse event: osteonecrosis of the jaw;

• adverse event: renal.

Results

Description of studies

Results of the search

We identified 5143 potentially relevant publications through database searches as well as reference lists. After having removed duplicates, we screened 3849 records for relevancy with respect to our research question. Out of 3849 records, 100 studies appeared to be relevant to the present research question and were then screened in a full‐text and abstract screening. We excluded 41 studies for reasons specified in advance, amongst which wrong comparison was the most frequent one. Other reasons for exclusion included, for instance, wrong study design or early termination of the trial. Additionally, we identified two ongoing studies and ten studies awaiting classification.

Finally, we identified 47 studies relevant to our research question, involving a total number of 35,163 participants, in the qualitative analysis. The overall numbers of references screened, identified, selected, excluded and included are documented according to the PRISMA flow diagram (Figure 2).

2.

Included studies

See also Characteristics of included studies tables.

Forty‐seven studies were included in narrative analysis (ABCSG‐12 2011; ABCSG‐18 2019; Aft 2012; ANZAC 2009; ARBI 2009; ARIBON 2012; AZURE 2018; BONADIUV 2019; Bundred 2009; Cohen 2008; D‐CARE 2013; Delmas 1997; Diel 1998; Ellis 2008; EXPAND 2011; FEMZONE 2014; GAIN 2013; GeparX 2016; Hershman 2008; H‐FEAT 2011; HOBOE 2019; JONIE 2017; Kanis 1996; Kristensen 2008; Mardiak 2000; Monda 2017; N02C1 2009; NATAN 2016; NEOZOL 2018; NEO‐ZOTAC BOOG 2010; Novartis I 2006; NSABP B‐34 2012; Powles 2006; ProBONE I 2005; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; Saarto 2004; SABRE 2010; Safra 2011; Saito 2015; Solomayer 2012; Sun 2016; SWOG S0307 2019; Team IIB 2006; Tevaarwerk 2007); 34 of them were also analysed quantitatively (ABCSG‐12 2011; ABCSG‐18 2019; Aft 2012; ARIBON 2012; AZURE 2018; BONADIUV 2019; Cohen 2008; D‐CARE 2013; Ellis 2008; EXPAND 2011; FEMZONE 2014; GAIN 2013; GeparX 2016Hershman 2008; HOBOE 2019; JONIE 2017; Kristensen 2008; Mardiak 2000; Monda 2017; NATAN 2016; NEOZOL 2018; NEO‐ZOTAC BOOG 2010; Novartis I 2006; NSABP B‐34 2012; Powles 2006; ProBONE II 2015; REBBeCA 2008; Saarto 2004; SABRE 2010; Safra 2011; Solomayer 2012; Sun 2016; SWOG S0307 2019; Team IIB 2006). The 13 remaining studies were excluded from quantitative analysis for the following reasons: four studies did not report any outcome of interest (ANZAC 2009; Bundred 2009; H‐FEAT 2011; ProBONE I 2005); another six studies only reported the results in a way that we were not able to add to the analysis (e.g. narratively) or, if no events occurred in both study arms, we were not able to add study data (ARBI 2009; N02C1 2009; REBBeCA II 2016; Rhee 2013; Saito 2015; Tevaarwerk 2007); we judged another three studies as too old to be comparable in the quantitative analyses with the other included studies in terms of anti‐cancer treatment regimens (Delmas 1997; Diel 1998; Kanis 1996).

Design

Except for one study, all included trials compared two treatment options with each other (ABCSG‐12 2011; ABCSG‐18 2019; ANZAC 2009; ARBI 2009; ARIBON 2012; AZURE 2018; Aft 2012; BONADIUV 2019; Bundred 2009; Cohen 2008; D‐CARE 2013; Delmas 1997; Diel 1998; EXPAND 2011; Ellis 2008; FEMZONE 2014; GAIN 2013; GeparX 2016; H‐FEAT 2011; Hershman 2008; HOBOE 2019; JONIE 2017; Kanis 1996; Kristensen 2008; Mardiak 2000; Monda 2017; N02C1 2009; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; NSABP B‐34 2012; Novartis I 2006; Powles 2006; ProBONE I 2005; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010; Saarto 2004; Safra 2011; Saito 2015; Solomayer 2012; Sun 2016; Team IIB 2006; Tevaarwerk 2007). In SWOG S0307 2019, three different treatment options were compared with each other (three‐arm study comparing zoledronic acid vs. clodronate vs. ibandronate).

Sample sizes

The 47 studies reported on 35,163 participants. The smallest trials included 11 participants (Cohen 2008; ProBONE I 2005), and the largest trial randomised 6097 participants (SWOG S0307 2019).

Setting

The included trials were performed by a range of research groups and in different countries. Thirty studies took place in a single country: the USA (Aft 2012; Cohen 2008; NSABP B‐34 2012; REBBeCA 2008; REBBeCA II 2016; SWOG S0307 2019; Tevaarwerk 2007), Greece (ARBI 2009), the UK (ARIBON 2012), Italy (BONADIUV 2019; HOBOE 2019), Germany (Diel 1998; EXPAND 2011; FEMZONE 2014; GAIN 2013; GeparX 2016; Novartis I 2006; ProBONE I 2005; ProBONE II 2015; Solomayer 2012), Japan (H‐FEAT 2011; JONIE 2017; Saito 2015), France (NEOZOL 2018), Netherlands (NEO‐ZOTAC BOOG 2010; Team IIB 2006), Israel (Safra 2011), Korea (Rhee 2013), Finland (Saarto 2004), and China (Sun 2016). Six trials took place in a continental setting: Europe (ABCSG‐12 2011; NATAN 2016, Austria and Germany; ABCSG‐18 2019, Austria and Sweden; Kristensen 2008, Denmark, Sweden and Iceland); and North America (Ellis 2008; N02C1 2009, USA and Canada). Four trials were conducted in an intercontinental setting: Kanis 1996 and Powles 2006 (North America and Europe), SABRE 2010 (North America, Europe and Africa), D‐CARE 2013 (39 countries worldwide). There was no precise information regarding the country in which the study had been conducted for seven trials (ANZAC 2009; AZURE 2018; Bundred 2009; Delmas 1997; Hershman 2008; Mardiak 2000; Monda 2017).

Participants

All participants had a confirmed diagnosis of early or locally advanced breast cancer without bone metastases.

Hormone receptor status

Seventeen studies included only hormone receptor‐positive participants (ABCSG‐18 2019; ARBI 2009; ARIBON 2012; BONADIUV 2019; EXPAND 2011; Ellis 2008; H‐FEAT 2011; HOBOE 2019; Monda 2017; Novartis I 2006; ProBONE II 2015; REBBeCA II 2016; Rhee 2013; SABRE 2010; Safra 2011; Sun 2016; Team IIB 2006). Twenty‐two studies included both hormone receptor‐positive and ‐negative participants (ABCSG‐12 2011; ANZAC 2009; AZURE 2018; Aft 2012; D‐CARE 2013; Diel 1998; FEMZONE 2014; GAIN 2013; GeparX 2016; Hershman 2008; JONIE 2017; Kanis 1996; Kristensen 2008; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; NSABP B‐34 2012; Powles 2006; SWOG S0307 2019; Saarto 2004; Solomayer 2012; Tevaarwerk 2007). ProBONE I 2005 included only hormone receptor negative participants. Seven trials did not give any information about the hormone receptor status (Bundred 2009; Cohen 2008; Delmas 1997; Mardiak 2000; N02C1 2009; REBBeCA 2008; Saito 2015).

HER2‐status

ABCSG‐12 2011 included only participants who were HER2‐positive. Three studies included only participants with negative HER2 status (JONIE 2017; NEO‐ZOTAC BOOG 2010; NEOZOL 2018). Thirteen studies evaluated participants with both HER2 status positive and negative (ABCSG‐18 2019; ANZAC 2009; AZURE 2018; Aft 2012; BONADIUV 2019; D‐CARE 2013; GAIN 2013; GeparX 2016HOBOE 2019; NATAN 2016; SWOG S0307 2019; Solomayer 2012; Tevaarwerk 2007). There was no information regarding the HER2 status for the remaining thirty trials (ARBI 2009; ARIBON 2012; Bundred 2009; Cohen 2008; Delmas 1997; Diel 1998; EXPAND 2011; Ellis 2008; FEMZONE 2014; Hershman 2008; H‐FEAT 2011; Kanis 1996; Kristensen 2008; Mardiak 2000; Monda 2017; N02C1 2009; NSABP B‐34 2012; Novartis I 2006; Powles 2006; ProBONE I 2005; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010; Saarto 2004; Safra 2011; Saito 2015; Sun 2016; Team IIB 2006).

Nodal status

Two studies included only node‐positive participants with a nodal status N1‐N3 (GAIN 2013; Saarto 2004), whereas twenty‐four studies included both node‐negative and node‐positive participants (ABCSG‐12 2011; ABCSG‐18 2019; AZURE 2018; BONADIUV 2019; D‐CARE 2013; Diel 1998; EXPAND 2011; FEMZONE 2014; GeparX 2016; HOBOE 2019; Hershman 2008; JONIE 2017; Kristensen 2008; NATAN 2016; NEO‐ZOTAC BOOG 2010; NSABP B‐34 2012; Powles 2006; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; SABRE 2010; SWOG S0307 2019; Solomayer 2012; Tevaarwerk 2007). The remaining twenty‐one studies did not give any precise information about the nodal status (ANZAC 2009; ARBI 2009; ARIBON 2012; Aft 2012; Bundred 2009; Cohen 2008; Delmas 1997; Ellis 2008; H‐FEAT 2011; Kanis 1996; Mardiak 2000; Monda 2017; N02C1 2009; NEOZOL 2018; Novartis I 2006; ProBONE I 2005; Rhee 2013; Safra 2011; Saito 2015; Sun 2016; Team IIB 2006).

Menopausal status

Twenty studies included participants with postmenopausal status only (ABCSG‐18 2019; ARBI 2009; ARIBON 2012; BONADIUV 2019; Cohen 2008; Delmas 1997; EXPAND 2011; FEMZONE 2014; H‐FEAT 2011; Monda 2017; Novartis I 2006; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010; Safra 2011; Saito 2015; Sun 2016; Team IIB 2006; Tevaarwerk 2007), whereas five studies included participants with premenopausal status only (ABCSG‐12 2011; Hershman 2008; HOBOE 2019; N02C1 2009; ProBONE II 2015). Additionally, sixteen studies included both participants with pre and postmenopausal status (Aft 2012; ANZAC 2009; AZURE 2018; D‐CARE 2013; Diel 1998; Ellis 2008; GAIN 2013; JONIE 2017; Kanis 1996; Kristensen 2008; NATAN 2016; NEOZOL 2018; NEO‐ZOTAC BOOG 2010; Powles 2006; Saarto 2004; Solomayer 2012). The remaining six studies did not give any precise information about the menopausal status (Bundred 2009; GeparX 2016; Mardiak 2000; NSABP B‐34 2012; ProBONE I 2005; SWOG S0307 2019).

Interventions

Bisphosphonates and receptor activator of nuclear factor‐kappa B ligand (RANKL)‐inhibitors

For an overview of all seven bone‐modifying agents and the main comparator, no treatment/placebo, see the ideal network diagram in Figure 1.

Twenty trials compared zoledronic acid with the main comparator (ABCSG‐12 2011; ANZAC 2009; AZURE 2018; Aft 2012; Bundred 2009; EXPAND 2011; FEMZONE 2014; HOBOE 2019; Hershman 2008; JONIE 2017; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; Novartis I 2006; ProBONE I 2005; ProBONE II 2015; Safra 2011; Solomayer 2012; Sun 2016; Tevaarwerk 2007), but the studies had different treatment intervals, almost every three or four weeks. The duration of treatment differed as well, from less than six months to five years of treatment.

Eight trials used risedronate (ARBI 2009; Delmas 1997; H‐FEAT 2011; Monda 2017; N02C1 2009; REBBeCA 2008; REBBeCA II 2016; SABRE 2010). Risedronate was mostly taken weekly for two years.

Four trials tested ibandronate against the main comparator (ARIBON 2012; BONADIUV 2019; GAIN 2013; Team IIB 2006). It was given mostly for two years; only Team IIB 2006 gave it for three years.

Six trials used clodronate (Diel 1998; Kanis 1996; Mardiak 2000; NSABP B‐34 2012; Powles 2006; Saarto 2004). In all studies, clodronate was given daily for two or three years.

Three trials tested alendronate (Cohen 2008; Rhee 2013; Saito 2015). The duration of treatment varied: Rhee 2013 gave alendronate daily for six months; whereas Cohen 2008 gave it weekly for one year and Saito 2015 weekly for two years.

One trial compared the effects of pamidronate with no treatment for four years (Kristensen 2008).

Four trials tested denosumab against placebo (ABCSG‐18 2019; D‐CARE 2013; Ellis 2008; GeparX 2016). Treatment in ABCSG‐18 2019 and D‐CARE 2013 continued treatment for up to five years, whereas Ellis 2008 tested it for two years and GeparX 2016 for six months.

One three‐armed trial explored the effect of zoledronic acid for 2.5 years in comparison with clodronate or ibandronate for three years (SWOG S0307 2019).

Anti‐cancer treatment

Surgery

Twenty‐seven studies included surgeries as a possible treatment (ABCSG‐12 2011; Aft 2012; BONADIUV 2019; Bundred 2009; D‐CARE 2013; Delmas 1997; FEMZONE 2014; GAIN 2013; GeparX 2016; Hershman 2008; HOBOE 2019; JONIE 2017; Kristensen 2008; Mardiak 2000; Monda 2017; N02C1 2009; NATAN 2016; NEOZOL 2018; Novartis I 2006; Powles 2006; REBBeCA 2008; REBBeCA II 2016; Saarto 2004; SABRE 2010; Solomayer 2012; Sun 2016; Tevaarwerk 2007). Some of the participants in SABRE 2010 were previously treated with a mastectomy. Some participants in D‐CARE 2013 got an ovarian ablation. Participants in FEMZONE 2014 were treated either with radical mastectomy, modified radical mastectomy, lumpectomy or quadrantectomy. Two studies reported that some of the participants received a mastectomy (BONADIUV 2019; Tevaarwerk 2007). In Hershman 2008 and N02C1 2009, some participants were treated with prior hysterectomy. Two studies treated some of their participants with either lumpectomy or mastectomy (JONIE 2017; Kristensen 2008; REBBeCA 2008). Participants in REBBeCA II 2016 received mastectomy, lumpectomy or axillary node removal. In Powles 2006, participants were treated with either resection with wide excision, segmental mastectomy with or without axillary dissection, or mastectomy according to the protocols of the participating centres. Participants in Saarto 2004 received an axillary evacuation and breast‐conserving resection or total mastectomy. All participants in Solomayer 2012 received surgery, which was either breast‐conserving or ablative surgery. In Sun 2016 and GAIN 2013, all participants received either radical mastectomy or breast‐conserving surgery. Participants in Novartis I 2006 were treated with tumour resection. Most participants in GeparX 2016 received breast‐conserving surgery. In Bundred 2009, participants received surgery at the end of the study, but there was no further information given about the type of surgery. Nine studies gave surgery before treatment, which was not further specified (ABCSG‐12 2011; Aft 2012; Delmas 1997; HOBOE 2019; Kanis 1996; Mardiak 2000; Monda 2017; NATAN 2016; NEOZOL 2018).

Endocrine therapy

In four studies, some of the participants received only selective oestrogen receptor modulators as endocrine therapy (Aft 2012; Diel 1998; Mardiak 2000; Saarto 2004).

Aromatase inhibitors as the only endocrine treatment were used in sixteen studies. Four trials chose anastrozole, which was given to all participants (ARBI 2009; ARIBON 2012; Monda 2017; SABRE 2010). Six trials treated all participants with letrozole (Bundred 2009; EXPAND 2011; H‐FEAT 2011; Novartis I 2006; Safra 2011; Sun 2016). One trial gave anastrozole or letrozole to all participants (Rhee 2013) and, in four trials, all participants were treated with either letrozole, anastrozole or exemestane (BONADIUV 2019; FEMZONE 2014; REBBeCA II 2016; Saito 2015). Two studies also used aromatase inhibitors as the only endocrine therapy, but did not give further information about which one was used (ABCSG‐18 2019; AZURE 2018). While in ABCSG‐18 2019, all patients were treated with aromatase inhibitors, only some of the participants of AZURE 2018 received aromatase inhibitors.

One study used ovarian function suppression as the only type of endocrine treatment (Diel 1998). In this trial, some of the participants received goserelin.

Four studies combined aromatase inhibitors and selective oestrogen receptor modulators as endocrine therapy that was given to some of the participants (GAIN 2013; HOBOE 2019; REBBeCA 2008; Tevaarwerk 2007), while one trial treated all participants with this combination (Team IIB 2006).

Six trials used a combination of aromatase inhibitors, selective oestrogen receptor modulators and ovarian function suppression (ABCSG‐12 2011; Aft 2012; D‐CARE 2013; Ellis 2008; Hershman 2008; ProBONE II 2015). ABCSG‐12 2011 treated all their participants with goserelin, but only some of their participants with anastrozole or tamoxifen; the remaining five treated only some of their participants with the three types of endocrine therapy.

In addition, GAIN 2013 treated some of their participants with luteinising hormone‐releasing hormones.

Chemotherapy

Thirty‐four trials used chemotherapy as an anti‐cancer treatment (ABCSG‐12 2011; ABCSG‐18 2019; ANZAC 2009; AZURE 2018; Aft 2012; D‐CARE 2013; Delmas 1997; Diel 1998; Ellis 2008; GAIN 2013; GeparX 2016; Hershman 2008; JONIE 2017; Kanis 1996; Kristensen 2008; Mardiak 2000; Monda 2017; N02C1 2009; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; NSABP B‐34 2012; Powles 2006; ProBONE I 2005; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SWOG S0307 2019; Saarto 2004; Solomayer 2012; Sun 2016; Team IIB 2006; Tevaarwerk 2007). 
The treatment methods were slightly different in all trials, as listed below:

The participants in Aft 2012 received a combination of epirubicin and docetaxel, while ANZAC 2009 treated all their participants with CEF, a combination of cyclophosphamide, epirubicin and fluorouracil. In GAIN 2013, epirubicin, paclitaxel and cyclophosphamide were given. Some of the participants in Saarto 2004 were treated with CMF, a combination of cyclophosphamide, methotrexate and fluorouracil, whereas Sun 2016 used ACT, a combination of doxorubicin and cyclophosphamide, followed by paclitaxel. NEO‐ZOTAC BOOG 2010 and NEOZOL 2018 used TAC, which includes docetaxel, doxorubicin and cyclophosphamide as a treatment. Participants in Diel 1998 received either CMF (cyclophosphamide, methotrexate and fluorouracil), EC (epirubicin and cyclophosphamide) or FEC (fluorouracil, epirubicin and cyclophosphamide). In Hershman 2008, AC (doxorubicin and cyclophosphamide), T (paclitaxel), a combination of AC and T, CMF (cyclophosphamide, methotrexate, fluorouracil) or CAF (cyclophosphamide, doxorubicin and fluorouracil) were given. JONIE 2017 treated all their participants with FEC (fluorouracil, epirubicin and cyclophosphamide), followed by paclitaxel. Participants in Kristensen 2008 were treated with either CMF (cyclophosphamide, methotrexate and fluorouracil) or CEF (cyclophosphamide, epirubicin and fluorouracil). Mardiak 2000 used CMF (cyclophosphamide, methotrexate and fluorouracil), CMFVP (cyclophosphamide, methotrexate, fluorouracil, vincristine and prednisone), FAC (fluorouracil, doxorubicin and cyclophosphamide) or FEC (fluorouracil, epirubicin and cyclophosphamide). Participants in Powles 2006 received either 2M (mitoxantrone and methotrexate), 3M (mitoxantrone, methotrexate and mitomycin), CMF (cyclophosphamide, methotrexate and fluorouracil), AC (doxorubicin and cyclophosphamide), FEC (fluorouracil, epirubicin and cyclophosphamide), ECF (epirubicin, cisplatin and fluorouracil) or other chemotherapy regimens. ProBONE II 2015 treated their participants with cyclophosphamide, fluorouracil, anthracyclines and taxanes. Six studies also used taxanes or anthracyclines, but did not specify which ones (D‐CARE 2013; GeparX 2016; N02C1 2009; NATAN 2016; Team IIB 2006; Tevaarwerk 2007). Eleven trials reported, that they used chemotherapy, but did not give further information about the type of chemotherapy (ABCSG‐12 2011; AZURE 2018; Delmas 1997; Kanis 1996; Monda 2017; NSABP B‐34 2012; ProBONE I 2005; REBBeCA 2008; REBBeCA II 2016; SWOG S0307 2019; Solomayer 2012). In three studies, some participants received chemotherapy prior to randomisation (ABCSG‐18 2019; Ellis 2008; Rhee 2013).

Radiotherapy

Seventeen trials reported on the use of radiotherapy (ABCSG‐12 2011; Aft 2012; Delmas 1997; Diel 1998; Ellis 2008; GAIN 2013; Kanis 1996; Kristensen 2008; Mardiak 2000; NATAN 2016; NSABP B‐34 2012; Powles 2006; ProBONE II 2015; REBBeCA 2008; Saarto 2004; Sun 2016; Tevaarwerk 2007). Diel 1998 used a dose of 50 Gy to the breast, Saarto 2004 reported a total dose of 50 Gy in 25 fractions to regional lymph nodes and the operation scar, and Sun 2016 used a dose of 50 Gy in 25 fractions, but added an extra 10‐16 Gy to the tumour bed. The remaining trials did not specify the total amount of Gy given as radiotherapy. Participants in GAIN 2013 and Saarto 2004 all received radiotherapy, while the participants of the other studies only received it when it was administered. Mardiak 2000 treated all participants with stage III breast cancer who received primary surgery with radiotherapy.

Supplemental therapy

In twenty‐two trials, vitamin D and calcium were supplemented (ABCSG‐18 2019; ARBI 2009; ARIBON 2012; AZURE 2018; Aft 2012; BONADIUV 2019; Cohen 2008; D‐CARE 2013; Ellis 2008; Hershman 2008; Monda 2017; N02C1 2009; NEO‐ZOTAC BOOG 2010; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010; SWOG S0307 2019; Solomayer 2012; Sun 2016). One trial used daily alfacalcidol as supplemental therapy (Saito 2015). HOBOE 2019 prescribed calcium and vitamin D to patients with osteoporosis or severe osteopenia, but their use was not dictated by the protocol and data were not collected.

Other

Six trials treated their participants with anti‐HER2 therapy (Aft 2012; D‐CARE 2013; HOBOE 2019; NATAN 2016; NEOZOL 2018; Tevaarwerk 2007). Five studies gave trastuzumab as the treatment drug (Aft 2012; HOBOE 2019; NATAN 2016; NEOZOL 2018; Tevaarwerk 2007), while D‐CARE 2013 did not specify the treatment drug.

Excluded studies

We excluded 41 studies, which are presented in the Characteristics of excluded studies, for the following reasons:

• inadequate randomisation and allocation concealment (Fuleihan 2005);

• no subgroup contained only BC (Gucalp 1994);

• subgroup BC not separately reported (Purohit 1995);

• no comparison (only single‐arm study) (Hines 2010);

• only abstract given, not clear if subgroup BC reported (Vinholes 1995);

• only trial registry entries, study terminated and no results published (IBIS 3 FEASIBILITY; JPRN‐UMIN000004375; NCT00247650; NCT00324714; PERIDENO);

• study withdrawn (NCT00873808);

• population with bone metastases (Lipton 1999; Smith 1999);

• population without bone metastases not reported (Ralston 1997);

• population of healthy women at high risk of BC (IBIS II 2003;

• metastases after treatment reported (Rizzoli 1996);

• wrong comparison (analysed the effect of capecitabine rather than ibandronate) (NCT00196859);

• wrong comparison (compared 2 different doses of bisphosphonates) (Gessner 2000; NCT03664687);

• wrong comparison (compared 2 different doses of denosumab) (NCT02051218);

• wrong comparison (bisphosphonates vs. physical activity) (NCT00202059);

• wrong comparison (2 years vs 5 years of bisphosphonate use) (BATMAN 2005);

• wrong comparison (3 years vs 5 years of bisphosphonate use) (SUCCESS);

• wrong comparison (upfront vs delayed bisphosphonate use) (Ahn 2009; CALGB 79809; N03CC; NCT05164952; Takahashi 2012; Z‐FAST 2012; ZO‐FAST 2013; ZOLMENO 2017);

• wrong comparison (concomitant medication not identical) (NCT03358017);

• non‐randomised study design (Ahmad 2007; Chen 2011; Ciardo 2020; Lee 2011; NCT00295867; Nakatsukasa 2019; Toulis 2016);

• summary of two studies plus non‐randomised study design (Vriens 2017);

• non‐randomised study design for bisphosphonate use (Van Hellemond 2019).

Risk of bias in included studies

See the Risk of bias tables in the Characteristics of included studies table. The risk of bias is summarised in Figure 3, which presents our judgements for each study in a cross‐tabulation. In summary, we considered the quality of included trials to be moderate.

3.

Allocation

Random sequence generation

Twenty‐four trials described a random component in the sequence generation process and were at low risk of selection bias (ABCSG‐12 2011; ABCSG‐18 2019; AZURE 2018; Aft 2012; BONADIUV 2019; D‐CARE 2013; Ellis 2008; GAIN 2013; GeparX 2016; H‐FEAT 2011; HOBOE 2019; Hershman 2008; Kanis 1996; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; NSABP B‐34 2012; Powles 2006; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010; SWOG S0307 2019; Tevaarwerk 2007). The other 23 trials were randomised studies, but without any further report on the sequence generation process (ANZAC 2009; ARBI 2009; ARIBON 2012; Bundred 2009; Cohen 2008; Delmas 1997; Diel 1998; EXPAND 2011; FEMZONE 2014; JONIE 2017; Kristensen 2008; Mardiak 2000; Monda 2017; N02C1 2009; Novartis I 2006; ProBONE I 2005; ProBONE II 2015; Saarto 2004; Safra 2011; Saito 2015; Solomayer 2012; Sun 2016; Team IIB 2006); hence we judged the risk of selection bias for these studies as unclear.

Allocation concealment

Twenty‐one studies reported on the method to conceal allocation and were at low risk of selection bias (ABCSG‐12 2011; ABCSG‐18 2019; AZURE 2018; Aft 2012; BONADIUV 2019; D‐CARE 2013; Ellis 2008; GAIN 2013; GeparX 2016; H‐FEAT 2011; HOBOE 2019; Hershman 2008; Kanis 1996; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; NSABP B‐34 2012; Powles 2006; REBBeCA 2008; REBBeCA II 2016; SABRE 2010). Twenty‐six trials provided no further information addressing allocation concealment and were considered to be at unclear risk of selection bias (ANZAC 2009; ARBI 2009; ARIBON 2012; Bundred 2009; Cohen 2008; Delmas 1997; Diel 1998; EXPAND 2011; FEMZONE 2014; JONIE 2017; Kristensen 2008; Mardiak 2000; Monda 2017; N02C1 2009; Novartis I 2006; ProBONE I 2005; ProBONE II 2015; Rhee 2013; SWOG S0307 2019; Saarto 2004; Safra 2011; Saito 2015; Solomayer 2012; Sun 2016; Team IIB 2006; Tevaarwerk 2007).

Blinding

Blinding of participants

Nineteen trials described some type of blinding of participants and were at low risk of performance bias (ABCSG‐18 2019; ARIBON 2012; BONADIUV 2019; Cohen 2008; D‐CARE 2013; Delmas 1997; Ellis 2008; Hershman 2008; Kanis 1996; Mardiak 2000; N02C1 2009; NSABP B‐34 2012; Powles 2006; ProBONE I 2005; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010). Four trials provided no information and were therefore at unclear risk of performance bias (ANZAC 2009; Bundred 2009; Monda 2017; Team IIB 2006). Twenty‐four trials were designed as open‐label studies and were at high risk of bias (ABCSG‐12 2011; ARBI 2009; AZURE 2018; Aft 2012; Diel 1998; EXPAND 2011; FEMZONE 2014; GAIN 2013; GeparX 2016; H‐FEAT 2011; HOBOE 2019; JONIE 2017; Kristensen 2008; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; Novartis I 2006; SWOG S0307 2019; Saarto 2004; Safra 2011; Saito 2015; Solomayer 2012; Sun 2016; Tevaarwerk 2007).

Blinding of personnel

Nineteen trials described some type of blinding of personnel and were at low risk of performance bias (ABCSG‐18 2019; ARIBON 2012; BONADIUV 2019; Cohen 2008; D‐CARE 2013; Delmas 1997; Ellis 2008; Hershman 2008; Kanis 1996; Mardiak 2000; N02C1 2009; NSABP B‐34 2012; Powles 2006; ProBONE I 2005; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010). Four trials provided no information and were at unclear risk of performance bias (ANZAC 2009; Bundred 2009; Monda 2017; Team IIB 2006). Twenty‐four trials were designed as open‐label studies and were at high risk of bias (ABCSG‐12 2011; ARBI 2009; AZURE 2018; Aft 2012; Diel 1998; EXPAND 2011; FEMZONE 2014; GAIN 2013; GeparX 2016; H‐FEAT 2011; HOBOE 2019; JONIE 2017; Kristensen 2008; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; Novartis I 2006; SWOG S0307 2019; Saarto 2004; Safra 2011; Saito 2015; Solomayer 2012; Sun 2016; Tevaarwerk 2007).

Blinding of outcome assessment: subjective outcomes

Nineteen trials reported blinding of outcome assessment for subjective outcomes and were at low risk of detection bias (ABCSG‐18 2019; ARIBON 2012; BONADIUV 2019; Cohen 2008; D‐CARE 2013; Delmas 1997; Ellis 2008; Hershman 2008; Kanis 1996; Mardiak 2000; N02C1 2009; NSABP B‐34 2012; Powles 2006; ProBONE I 2005; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010). Four trials provided insufficient information and were therefore judged as at unclear risk of bias (ANZAC 2009; Bundred 2009; Monda 2017; Team IIB 2006). Twenty‐four trials were open‐label studies, which we judged as at high risk of bias (ABCSG‐12 2011; ARBI 2009; AZURE 2018; Aft 2012; Diel 1998; EXPAND 2011; FEMZONE 2014; GAIN 2013; GeparX 2016; H‐FEAT 2011; HOBOE 2019; JONIE 2017; Kristensen 2008; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; Novartis I 2006; SWOG S0307 2019; Saarto 2004; Safra 2011; Saito 2015; Solomayer 2012; Sun 2016; Tevaarwerk 2007).

Blinding of outcome assessment: objective outcomes

Due to the nature of objective outcomes being unlikely to be vulnerable to bias, 46 studies were judged as at low risk of detection bias (ABCSG‐12 2011; ABCSG‐18 2019; ANZAC 2009; ARBI 2009; ARIBON 2012; AZURE 2018; Aft 2012; BONADIUV 2019; Cohen 2008; D‐CARE 2013; Delmas 1997; Diel 1998; EXPAND 2011; Ellis 2008; FEMZONE 2014; GAIN 2013; GeparX 2016; H‐FEAT 2011; HOBOE 2019; Hershman 2008; JONIE 2017; Kanis 1996; Kristensen 2008; Mardiak 2000; Monda 2017; N02C1 2009; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; NSABP B‐34 2012; Novartis I 2006; Powles 2006; ProBONE I 2005; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010; SWOG S0307 2019; Saarto 2004; Safra 2011; Saito 2015; Solomayer 2012; Sun 2016; Team IIB 2006; Tevaarwerk 2007). One trial had only an abstract published and was therefore judged as at unclear risk of bias (Bundred 2009).

Incomplete outcome data

Thirty‐nine trials addressed incomplete outcome data adequately, describing reasons for missing data or including all randomised participants in the statistical analysis; we assessed these studies as at low risk of attrition bias (ABCSG‐12 2011; ABCSG‐18 2019; ARIBON 2012; AZURE 2018; Aft 2012; BONADIUV 2019; Cohen 2008; D‐CARE 2013; Delmas 1997; Diel 1998; EXPAND 2011; Ellis 2008; FEMZONE 2014; GAIN 2013; GeparX 2016; HOBOE 2019; Hershman 2008; JONIE 2017; Kanis 1996; Kristensen 2008; N02C1 2009; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; NSABP B‐34 2012; Novartis I 2006; Powles 2006; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; Rhee 2013; SABRE 2010; SWOG S0307 2019; Saarto 2004; Safra 2011; Saito 2015; Solomayer 2012; Team IIB 2006; Tevaarwerk 2007). Five studies provided insufficient information and were at unclear risk of attrition bias (ANZAC 2009; Bundred 2009; H‐FEAT 2011; Monda 2017; Sun 2016). ARBI 2009 did not analyse data from 23 of 70 participants and no reason was given for 12 of these participants. In addition, they did not have ITT analysis. Therefore, it was judged as being at high risk of bias. Mardiak 2000 did not evaluate 10 of 72 participants because of a "short duration of therapy"; we therefore judged the risk of bias as high. ProBONE I 2005 was terminated, and no data were published. The attrition bias was therefore judged as high.

Selective reporting

Twenty‐two trials published a study protocol or included all expected outcomes and were at low risk of reporting bias (ABCSG‐12 2011; ABCSG‐18 2019; AZURE 2018; Aft 2012; BONADIUV 2019; D‐CARE 2013; Diel 1998; GAIN 2013; HOBOE 2019; JONIE 2017; Kanis 1996; Mardiak 2000; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; NSABP B‐34 2012; Powles 2006; ProBONE II 2015; REBBeCA 2008; REBBeCA II 2016; SWOG S0307 2019; Saarto 2004). Nineteen trials provided little information on primary or secondary outcomes and their definition and were therefore judged as at unclear risk for reporting bias (ANZAC 2009; ARBI 2009; ARIBON 2012; Bundred 2009; Cohen 2008; Delmas 1997; Ellis 2008; FEMZONE 2014; GeparX 2016; H‐FEAT 2011; Kristensen 2008; Monda 2017; N02C1 2009; Rhee 2013; SABRE 2010; Saito 2015; Sun 2016; Team IIB 2006; Tevaarwerk 2007). EXPAND 2011 and Novartis I 2006 did not analyse or present any participants for DFS, although it was a prespecified outcome. Hershman 2008 reported recurrence, although it was not actually a prespecified endpoint. ProBONE I 2005 did not publish data because the study was terminated. Safra 2011 did not report all time points as predefined and, it seems, according to fig. 3, that time points favouring zoledronic acid were chosen for the reporting of the main outcome, BMD. Solomayer 2012 did not report results for BMD in the full‐text reference, but it was planned as an outcome as described in the trial registry. Hence, we judged the risk of bias for these six studies as high.

Other potential sources of bias

Aft 2012 chose a difficult endpoint (DTCs), which may or may not correlate directly with clinically evident bone metastases; Bundred 2009, H‐FEAT 2011 and Team IIB 2006 had only the abstract information available; FEMZONE 2014 was terminated; Novartis I 2006 never published their study results; ProBONE I 2005 was terminated, and no data were published. We therefore judged these seven studies as at unclear risk of other bias.

EXPAND 2011 reported, that "there IS an agreement between Principal Investigators and the Sponsor (or its agents) that restricts the PI's rights to discuss or publish trial results after the trial is completed."; Kristensen 2008 had patients with ER‐positive BC status, but did not allow endocrine therapy, which could have led to an incorrect treatment. These two studies were therefore judged as being at high risk of other bias.

Effects of interventions

See: Table 1

We present our main findings from the network meta‐analyses (NMAs) for each pairwise comparison of each intervention with no treatment/placebo in Table 1. Results for other comparisons are reported in the text below and in additional tables and figures.

Pairwise comparisons

Pairwise comparisons are part of the NMAs, and we did not perform additional pairwise meta‐analyses. The direct effect estimates for all pairwise comparisons are presented in the upper triangle of each league table (see Additional Tables). For descriptive presentation only, we created forest plots with direct pairwise comparisons.

Transitivity

The included trials were similar in clinical and methodological characteristics that could potentially affect the relative treatment effects, thus we assumed the transitivity assumption held. Distributions of potential effect modifiers across the different pairwise comparisons are displayed in Appendix 2.

Bone density

Changes in bone mineral density were reported in different units such as g, g/cm², g/cm³, T‐Score and Z‐score, which are not comparable in quantitative analysis. Also, bone mineral density was measured in different parts of the body, described as lumbar spine, femoral neck, total hip, distal tibia, total body, proximal femur, phalanges, forearm or wrist site. We extracted data for the most common site, 'lumbar spine', with the most common unit, 'T‐score', measured with dual‐energy X‐ray absorptiometry to collect data which would be comparable in a quantitative analysis.

Thirteen studies (ABCSG‐12 2011; ABCSG‐18 2019; ARBI 2009; ARIBON 2012; BONADIUV 2019; Cohen 2008; Ellis 2008; EXPAND 2011; HOBOE 2019; Monda 2017; Novartis I 2006; ProBONE II 2015; Safra 2011) reported changes in BMD as a T‐score. Nine studies including 1166 participants were examined in the statistical analysis (ABCSG‐12 2011; BONADIUV 2019; Cohen 2008; EXPAND 2011; HOBOE 2019; Monda 2017; Novartis I 2006; ProBONE II 2015; Safra 2011). One of these reported results at six months follow‐up (Safra 2011). All other studies reported results for a follow‐up between one and five years.

Two studies reported their results as medians and not means, and could not be included in the network meta‐analysis (ARBI 2009; ARIBON 2012). Two studies reported their results as percentages (Ellis 2008) or percentage change (ABCSG‐18 2019) and could not be included in the network meta‐analysis.

Of the studies included in the statistical analysis, three studies had mixed populations with pre and postmenopausal women. Six studies included only postmenopausal women. The network diagram is presented in Figure 4. The star‐shaped network, based on nine pairwise comparisons, was connected and compared five different treatment options (alendronate, ibandronate, risedronate, zoledronic acid and no treatment/placebo). There were no closed loops in the network. The forest plot, including all pairwise comparisons, is shown in Figure 5.

4.

Network diagram for outcome: bone mineral density (T‐score). Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients.

5.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, bone mineral density (T‐score)

Network meta‐analysis

The evidence suggests that zoledronic acid is likely to slightly increase bone mineral density compared to no treatment/placebo (MD 0.89, 95% CI 0.62 to 1.16; moderate certainty) (Figure 6; Table 2). No other comparison showed meaningful results as indicated by the low to very low certainty of evidence judged for the other bisphosphonates (described below).

6.

Forest plot for outcome bone mineral density (T‐Score): 
Random‐effects model
Reference treatment: No treatment/placebo
Treatments are ordered by P‐Score (descending).

Abbreviations:
CI: confidence interval
MD: mean difference

1. League table: bone mineral density (T‐score).

Alendronate	.	.	.	3.70 [‐2.01, 9.41]
2.81 [‐2.90, 8.53]	Zoledronic acid	.	.	0.89 [0.62, 1.16]
3.13 [‐2.61, 8.87]	0.32 [‐0.36, 0.99]	Ibandronate	.	0.57 [‐0.05, 1.19]
3.44 [‐2.30, 9.18]	0.63 [‐0.01, 1.27]	0.31 [‐0.54, 1.16]	Risedronate	0.26 [‐0.32, 0.84]
3.70 [‐2.01, 9.41]	0.89 [0.62, 1.16]	0.57 [‐0.05, 1.19]	0.26 [‐0.32, 0.84]	No treatment/placebo

Results of network meta‐analysis for outcome bone mineral density (T‐Score). Only subnets with > 1 designs. Treatments are ordered by P‐Score (ascending). Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as mean differences (MD) with a corresponding 95% confidence interval. For the network estimates in the lower triangle, an MD higher than 0.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an MD higher than 0.0 favours the row‐defining treatment (higher bone mineral density). To obtain MDs for comparisons in the opposing direction, signs must be reversed.

No. of studies: 9
No. of treatments: 5
No. of pairwise comparisons: 9
No. of designs: 4

Heterogeneity/Inconsistency: Q = 18.89, df = 5, P = 0.002; I² = 73.5%, Tau² = 0.0732

Ibandronate may slightly increase bone density compared to no treatment/placebo (MD 0.57, 95% CI ‐0.05 to 1.19; low certainty). Risedronate may result in little to no difference regarding bone mineral density compared to no treatment/placebo (MD 0.26, 95% CI ‐0.32 to 0.84; low certainty). We are uncertain whether alendronate increases bone mineral density compared to no treatment/placebo (MD 3.70, 95% CI ‐2.01 to 9.41; very low certainty). Our main reasons for downgrading the evidence were imprecision and inconsistency. Reasons for downgrading are provided in Table 1. It was not possible to rate the certainty of the evidence for clodronate, denosumab and pamidronate because they were not included in the network.

The rankings according to the P‐score presented in Figure 6 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Although alendronate was ranked first (P‐Score: 0.87), the confidence interval was very wide and the estimate was based on one study involving 11 participants (MD 3.70, 95% CI ‐2.01 to 9.41; very low certainty).

The fixed‐effect model showed little to no difference compared to the random‐effects model (not shown). In the entire network, Cochran's Q test and generalised I² statistics showed substantial heterogeneity between studies (Q_total = 18.89, P = 0.002; I² = 73.5%, Tau² = 0.0732). Since there were no closed loops, inconsistencies could not be analysed.

None of the pairwise comparisons consisted of ten or more studies, so a funnel plot could not be constructed. Instead, a comparison‐adjusted funnel plot was created (Figure 7). The funnel plot consisted of nine studies, so the number of studies was too small to test for small‐study effects. Visual inspection of the funnel plot showed no evidence of small‐study effects.

7.

Comparison‐adjusted funnel plot for the outcome, bone mineral density (T‐Score)

Subgroup analysis

For studies with a mixture of pre and menopausal populations, no subgroup analysis was performed because the network consisted of only one pairwise comparison.

The subgroup analysis including studies with postmenopausal women only indicated zoledronic acid (MD 1.25, 95% CI 0.97 to 1.53), ibandronate (MD 0.57, 95% CI 0.25 to 0.89) and risedronate (MD 0.26, 95% CI 0.03 to 0.49) probably slightly increase bone mineral density compared with no treatment/placebo. Additionally, zoledronic acid probably slightly increases bone mineral density compared to ibandronate (MD 0.68, 95% CI 0.26 to 1.10) and risedronate (MD 0.99, 95% CI 0.63 to 1.35). Nevertheless, confidence intervals were overlapping with that of the main analysis and the ranking of treatments did not change. This cluster included six studies (data not shown).

Sensitivity analysis

For the sensitivity analysis, three studies were excluded because they were considered to be at a high risk of bias (EXPAND 2011; Novartis I 2006; Safra 2011). The results of the overall sensitivity analysis, including six studies, showed similar results to the main analysis (data not shown).

The results of the sensitivity analysis, including studies with postmenopausal women only, showed similar results, except that ibandronate (MD 0.57, 95% CI 0.25 to 0.89) and risedronate (MD 0.26, 95% CI 0.03 to 0.49) now probably slightly increase bone mineral density compared to no treatment/placebo, but confidence intervals were overlapping. This analysis included three studies (BONADIUV 2019; Cohen 2008; Monda 2017) comparing alendronate, ibandronate, risedronate and no treatment/placebo (data not shown).

Quality of life

Network meta‐analysis

Due to insufficient reporting, no network meta‐analysis was possible for quality of life. Three studies reported quality of life (FEMZONE 2014; GeparX 2016; Monda 2017).

FEMZONE 2014 included 168 postmenopausal participants and compared zoledronic acid to no treatment/placebo. The study was considered to be at a high risk of bias. Quality of life was measured with the FACT‐B questionnaire after six months. The mean score of participants treated with zoledronic acid was 108.2, while participants with no treatment/placebo had a slightly higher mean score of 112.

GeparX 2016 included 780 pre and postmenopausal participants and compared denosumab to no treatment/placebo. The study was considered to be at a high risk of bias. Quality of life was measured with the FACT‐Taxane questionnaire, FACT‐Taxane Trial Outcome Index (TOI), FACT‐G total score, and FACT‐Taxane total score scales at baseline, after nab‐paclitaxel, at end of treatment and 90 days post‐surgery. The addition of denosumab did not change the quality of life scores at any time point.

Monda 2017 included 84 postmenopausal participants and compared risedronate to no treatment/placebo. The study was considered to be at a low risk of bias. Quality of life was measured with the SF‐36 questionnaire after two years. The mean score of participants treated with risedronate was 48.6 (standard deviation: 7.3) and participants with no treatment/placebo had a slightly lower mean score of 44.8 (standard deviation: 6.4).

Subgroup analysis

Due to the low number of studies reporting quality of life, no subgroup analysis could be performed.

Sensitivity analysis

Due to the low number of studies reporting quality of life, no sensitivity analysis could be performed.

Overall fracture rate

Twenty‐four studies reported the outcome overall fracture rate (ABCSG‐12 2011; ABCSG‐18 2019; ARBI 2009; ARIBON 2012; AZURE 2018; D‐CARE 2013; EXPAND 2011; Ellis 2008; FEMZONE 2014; HOBOE 2019; Hershman 2008; Kanis 1996; Kristensen 2008; Monda 2017; Novartis I 2006; Powles 2006; ProBONE II 2015; REBBeCA 2008; Rhee 2013; SABRE 2010; SWOG S0307 2019; Safra 2011; Sun 2016; Team IIB 2006). Sixteen studies including 19,492 participants were examined in the statistical analysis (ABCSG‐12 2011; ABCSG‐18 2019; ARIBON 2012; AZURE 2018; D‐CARE 2013; Ellis 2008; FEMZONE 2014; Kristensen 2008; Monda 2017; Powles 2006; ProBONE II 2015; REBBeCA 2008; SABRE 2010; SWOG S0307 2019; Sun 2016; Team IIB 2006). Eight studies reported the overall fracture rate in a format where it was not possible to add the data to the statistical analysis (ARBI 2009; EXPAND 2011; HOBOE 2019; Hershman 2008; Kanis 1996; Novartis I 2006; Rhee 2013; Safra 2011). In seven of these studies, no fractures occurred in either treatment arm:

• ARBI 2009 compared risedronate to no treatment; there were 70 participants with a follow‐up time of two years.

• EXPAND 2011 compared zoledronic acid to no treatment; there were 81 participants with a follow‐up time of three years.

• Hershman 2008 compared zoledronic acid to placebo; there were 114 participants with a follow‐up time of two years.

• HOBOE 2019 compared zoledronic acid to no treatment; there were 1065 participants with a median follow‐up time of 64 months.

• Novartis I 2006 compared zoledronic acid to no treatment; there were 83 participants with a follow‐up time of three years.

• Rhee 2013 compared alendronate to placebo; there were 98 participants and the duration of the study was six months.

• Safra 2011 compared zoledronic acid to no treatment; there were 90 participants with a follow‐up time of five years.

Of the studies included in the statistical analysis, 11 studies included both pre and postmenopausal women. Five studies included only postmenopausal women. The network diagram is presented in Figure 8. The network, based on eighteen pairwise comparisons, was connected and compared seven different treatment options (clodronate, denosumab, ibandronate, pamidronate, risedronate, zoledronic acid and no treatment/placebo). There were four closed loops in the network. The forest plot, including all pairwise comparisons, is shown in Figure 9.

8.

Network diagram for outcome: fractures. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients.

9.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome fractures

Network meta‐analysis

Ibandronate and clodronate decrease the number of fractures compared to no treatment/placebo (RR 0.57, 95% CI 0.38 to 0.86, RR 0.60, 95% CI 0.39 to 0.92, respectively, high certainty). Ibandronate and clodronate are likely to reduce fracture rates compared to pamidronate (RR 0.38, 95% CI 0.17 to 0.85, RR 0.40, 95% CI 0.18 to 0.90, respectively, moderate certainty; Figure 10; Table 3).

10.

Forest plot for outcome, fractures: 
Random‐effects model
Reference treatment: no treatment/placebo. 
Treatments are ordered by P‐Score (descending). 
Abbreviations:
CI: confidence interval
RR: risk ratio

2. League table: fractures.

Ibandronate	0.79 [0.51, 1.23]	.	.	0.62 [0.40, 0.96]	0.89 [0.48, 1.63]	.
0.95 [0.64, 1.42]	Clodronate	.	.	0.78 [0.51, 1.18]	0.41 [0.19, 0.87]	.
1.02 [0.25, 4.18]	1.07 [0.26, 4.40]	Risedronate	.	.	0.56 [0.15, 2.16]	.
0.79 [0.47, 1.33]	0.83 [0.48, 1.42]	0.77 [0.19, 3.09]	Denosumab	.	0.73 [0.52, 1.01]	.
0.73 [0.50, 1.06]	0.76 [0.52, 1.11]	0.71 [0.18, 2.86]	0.92 [0.57, 1.49]	Zoledronic acid	0.72 [0.48, 1.10]	.
0.57 [0.38, 0.86]	0.60 [0.39, 0.92]	0.56 [0.15, 2.16]	0.73 [0.52, 1.01]	0.79 [0.56, 1.11]	No treatment/placebo	0.66 [0.33, 1.33]
0.38 [0.17, 0.85]	0.40 [0.18, 0.90]	0.37 [0.08, 1.69]	0.48 [0.22, 1.04]	0.52 [0.24, 1.14]	0.66 [0.33, 1.33]	Pamidronate

Results of network meta‐analysis for outcome fractures. Treatments are ordered by P‐Score (ascending). Only subnets with > 1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as risk ratios (RR) with a corresponding 95% confidence interval. For the network estimates in the lower triangle, an RR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an RR below 1.0 favours the row‐defining treatment (fewer events of fractures). To obtain RRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 15
No. of treatments: 7
No. of pairwise comparisons: 17
No. of designs: 7

Q_total = 14.64, df = 11, P = 0.20/Q_within = 8.55, df = 9, P = 0.48/Q_between = 6.09, df = 2, P = 0.048; I² = 24.8%, Tau² = 0.0380

Denosumab and zoledronic acid probably slightly decrease the number of fractures compared to no treatment/placebo (RR 0.73, 95% CI 0.52 to 1.01, RR 0.79, 95% CI 0.56 to 1.11, respectively; moderate certainty). Risedronate may decrease or increase the number of fractures compared to no treatment/placebo (RR 0.56, 95% CI 0.15 to 2.16, low certainty). Pamidronate probably increases the number of fractures compared to no treatment/placebo (RR 1.52, 95% CI 0.75 to 3.06; moderate certainty). Our main reason for downgrading the evidence was imprecision. Reasons for downgrading are provided in Table 1. It was not possible to rate the certainty of the evidence for alendronate as this treatment is not included in our network.

The rankings according to the P‐score presented in Figure 10 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Ibandronate was ranked first (P‐Score: 0.81) followed by clodronate (P‐Score: 0.75). No treatment/placebo (P‐Score: 0.20) and pamidronate (P‐Score: 0.05) were ranked lowest.

The fixed‐effect model showed slight differences compared to the random‐effects model (not shown). In the entire network, Cochran's Q test and generalised I² statistics showed no important heterogeneity between studies (Q_total = 14.64, P = 0.20 / Q_within = 8.55, P = 0.48 / Q_between = 6.09, P = 0.05; I² = 24.8 %, Tau² = 0.0380). Since there were closed loops in the network, inconsistencies could be analysed. The test for inconsistencies in closed loops indicated no disagreements between direct and indirect estimates (Figure 11; Table 4).

11.

Comparison of direct and indirect evidence (in closed loops) for outcome, fractures. 
Abbreviations:
CI: confidence interval
RR: risk ratio

3. Comparison of direct and indirect evidence (in closed loops) for outcome, fractures.

Comparison	No. of studies	Network estimate	Direct estimate	Indirect estimate	Test for disagreement
Clodronate vs. ibandronate	1	1.05 [0.70; 1.56]	1.26 [0.81; 1.95]	0.44 [0.17; 1.15]	0.0523
Clodronate vs. no treatment/placebo	1	0.60 [0.39; 0.92]	0.41 [0.19; 0.87]	0.72 [0.43; 1.19]	0.2249
Clodronate vs. zoledronic acid	1	0.76 [0.52; 1.11]	0.78 [0.51; 1.18]	0.71 [0.31; 1.64]	0.8471
Ibandronate vs. no treatment/placebo	2	0.57 [0.38; 0.86]	0.89 [0.48; 1.63]	0.41 [0.24; 0.70]	0.0625
Ibandronate vs. zoledronic acid	1	0.73 [0.50; 1.06]	0.62 [0.40; 0.96]	1.16 [0.56; 2.41]	0.1494
Zoledronic acid vs. no treatment/placebo	5	0.79 [0.56; 1.11]	0.72 [0.48; 1.10]	0.94 [0.52; 1.73]	0.4797

Estimates are reported as risk ratios with corresponding 95% confidence intervals. The results of a test for disagreement between direct and indirect evidence reported as a P value. Only comparisons for which both direct and indirect evidence exists are shown.

Subgroup analysis

When analyses were restricted to studies with mixed populations, excluding five studies (ARIBON 2012; FEMZONE 2014; Monda 2017; SABRE 2010; Team IIB 2006), the results showed slight differences. The network diagram showed three closed loops instead of four, because the connection was missing for ibandronate and no treatment/placebo. In the ranking of treatments, risedronate was now ranked lower than no treatment/placebo, but the confidence intervals were very large and overlapping. Zoledronic acid and denosumab resulted in a slightly reduced fracture rate compared to no treatment/placebo (RR 0.71, 95% CI 0.57 to 0.89; RR 0.77, 95% CI 0.64 to 0.92, respectively). Denosumab probably resulted in a slightly decreased fracture rate compared to pamidronate (RR 0.47, 95% CI 0.25 to 0.88; RR 0.51, 95% CI 0.27 to 0.94, respectively). Additionally, ibandronate resulted in a reduced and clodronate in a slightly reduced fracture rate compared to zoledronic acid (RR 0.61, 95% CI 0.50 to 0.75; RR 0.76, 95% CI 0.64 to 0.90, respectively) and denosumab (RR 0.57, 95% CI 0.40 to 0.81; RR 0.71, 95% CI 0.51 to 0.98, respectively) but confidence intervals were overlapping (data not shown).

Sensitivity analysis

For the sensitivity analysis, one study was excluded due to a high risk of bias (Kristensen 2008). Given that the excluded study was the only study including pamidronate, the results of the sensitivity analysis were the same as in the main network meta‐analysis (data not shown).

Vertebral fractures

Nine studies reported vertebral fractures (ABCSG‐18 2019; Ellis 2008; Kanis 1996; Kristensen 2008; Monda 2017; Powles 2006; Rhee 2013; SABRE 2010; Sun 2016). Five studies including 3610 participants were examined in the statistical analysis (ABCSG‐18 2019; Kristensen 2008; Monda 2017; Powles 2006; Sun 2016). One study was excluded from the quantitative analyses because the study was considered too old to be comparable with the other studies (Kanis 1996) and three studies reported vertebral fractures in a format where it was not possible to add the data to the statistical analysis (Ellis 2008; Rhee 2013; SABRE 2010). In these three studies, there were no cases of vertebral fractures occurring in either treatment arm:

• Ellis 2008 compared denosumab to placebo; there were 252 participants with a follow‐up time of four years.

• Rhee 2013 compared alendronate to placebo; there were 98 participants and the duration of the study was six months.

• SABRE 2010 compared risedronate to placebo; there were 154 participants with a follow‐up time of two years.

Of the studies included in the statistical analysis, four studies included a mixed population of pre and postmenopausal women. One study included postmenopausal women only. The network diagram is presented in Figure 12. The star‐shaped network, based on five pairwise comparisons, was connected and compared six different treatment options (clodronate, denosumab, pamidronate, risedronate, zoledronic acid and no treatment/placebo). There were no closed loops in the network. A forest plot, including all pairwise comparisons, is shown in Figure 13.

12.

Network diagram for outcome: vertebral fractures. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients

13.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, vertebral fractures

Network meta‐analysis

Denosumab reduces the occurrence of vertebral fractures compared to no treatment/placebo (RR 0.55, 95% CI 0.36 to 0.84; high certainty) (Figure 14; Table 5). No other comparison showed meaningful results.

14.

Forest plot for outcome vertebral fractures: Random effects model. Reference treatment: No treatment/placebo. Treatments are ordered by P‐Score (descending). RR: risk ratio. CI: confidence interval.

4. League table: vertebral fractures.

Risedronate	.	.	.	0.14 [0.01, 2.59]	.
0.32 [0.01, 6.91]	Clodronate	.	.	0.44 [0.17, 1.13]	.
0.25 [0.01, 4.90]	0.80 [0.28, 2.26	Denosumab	.	0.55 [0.36, 0.84]	.
0.21 [0.01, 6.30]	0.65 [0.09, 4.77]	0.82 [0.14, 4.94]	Zoledronic acid	0.67 [0.12, 3.82]	.
0.14 [0.01, 2.59]	0.44 [0.17, 1.13]	0.55 [0.36, 0.84]	0.67 [0.12, 3.82]	No treatment/Placebo	0.80 [0.29, 2.19]
0.11 [0.01, 2.46]	0.35 [0.09, 1.39]	0.44 [0.15, 1.30]	0.53 [0.07, 4.00]	0.80 [0.29, 2.19]	Pamidronate

Results of network meta‐analysis for the outcome of vertebral fractures. Treatments are ordered by P‐Score (ascending). Only subnets with > 1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as risk ratios (RR) with corresponding 95% confidence intervals. For the network estimates in the lower triangle, an RR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an RR below 1.0 favours the row‐defining treatment (fewer events of vertebral fractures). To obtain RRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 5
No. of treatments: 6
No. of pairwise comparisons: 5
No. of designs: 5

Heterogeneity/inconsistency: Q = 0, df = 0, P = n.a.; I² = n.a., Tau² = n.a.

The rankings according to the P‐score presented in Figure 14 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Although risedronate was ranked first (P‐Score: 0.85), the confidence interval was very wide and the estimate was based on one study involving 71 participants (RR 0.14, 95% CI 0.01 to 2.59). No treatment/placebo (P‐Score: 0.23) and pamidronate (P‐Score: 0.16) were ranked lowest.

The fixed‐effect model did not show any differences compared to the random‐effects model (not shown). Since no pairwise comparison contained more than one study and there were no closed loops in the network, heterogeneity in the entire network and inconsistencies between direct and indirect estimates could not be tested (Q_total = 0, P = not applicable, Q_within = 0, P = not applicable, Q_between = 0, P = not applicable, I² = not applicable, Tau² = not applicable).

Subgroup analysis

One study included postmenopausal women only (Monda 2017). Therefore, the network diagram did not differ compared to the main analysis except that risedronate was missing. The ranking according to P‐scores also did not change.

Sensitivity analysis

No sensitivity analysis was performed.

Non‐vertebral fractures

Eight studies reported non‐vertebral fractures (ARIBON 2012; Ellis 2008; FEMZONE 2014; Kanis 1996; Kristensen 2008; Powles 2006; ProBONE II 2015; SABRE 2010). Seven studies including 2381 participants were examined in the statistical analysis (ARIBON 2012; Ellis 2008; FEMZONE 2014; Kristensen 2008; Powles 2006; ProBONE II 2015; SABRE 2010). As mentioned above, one study was excluded from the quantitative analyses because the study was considered too old to be comparable with the other studies (Kanis 1996).

Of the studies included in the statistical analysis, three included postmenopausal women only. Four of the studies included in the statistical analysis included mixed populations with pre and postmenopausal women. The network diagram is presented in Figure 15. The star‐shaped network, based on seven pairwise comparisons, was connected and compared seven different treatment options (clodronate, denosumab, ibandronate, pamidronate, risedronate, zoledronic acid and no treatment/placebo). There were no closed loops in the network. The forest plot, including all pairwise comparisons, is shown in Figure 16.

15.

Network diagram for outcome: non‐vertebral fractures. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients

16.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, non‐vertebral fractures

Network meta‐analysis

Evidence suggests that clodronate and pamidronate are likely to reduce the occurrence of non‐vertebral fractures compared to no treatment/placebo (RR 0.38, 95% CI 0.15 to 0.97, RR 0.23, 95% CI 0.06 to 0.84, respectively, moderate certainty; Figure 17; Table 6). No other comparison showed meaningful results.

17.

Forest plot for outcome, non‐vertebral fractures: 
Random‐effects model
Reference treatment: no treatment/placebo
Treatments are ordered by P‐Score (descending).

Abbreviations:
CI: confidence interval
RR: risk ratio

5. League table: non‐vertebral fractures.

Risedronate	.	.	.	0.11 [0.01, 2.03]	.
0.29 [0.01, 6.15]	Clodronate	.	.	0.38 [0.15, 0.97]	.
0.14 [0.01, 3.55]	0.47 [0.08, 2.68]	Zoledronic acid	.	0.81 [0.19, 3.53]	.
0.12 [0.01, 2.52]	0.41 [0.11, 1.53]	0.87 [0.15, 4.97]	Denosumab	0.93 [0.36, 2.39]	.
0.11 [0.01, 2.03]	0.38 [0.15, 0.97]	0.81 [0.19, 3.53]	0.93 [0.36, 2.39]	No treatment/placebo	0.83 [0.21, 3.24]	0.61 [0.25, 1.48]
0.09 [0.00, 2.28]	0.32 [0.06, 1.65]	0.67 [0.09, 4.99]	0.77 [0.15, 4.05]	0.83 [0.21, 3.24]	Ibandronate
0.07 [0.00, 1.41]	0.23 [0.06, 0.84]	0.49 [0.09, 2.75]	0.57 [0.16, 2.07]	0.61 [0.25, 1.48]	0.74 [0.15, 3.73]	Pamidronate

Results of network meta‐analysis for outcome of non‐vertebral fractures. Treatments are ordered by P‐Score (ascending). Only subnets with > 1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as risk ratios (RR) with corresponding 95% confidence intervals. For the network estimates in the lower triangle, an RR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an RR below 1.0 favours the row‐defining treatment (fewer events of non‐vertebral fractures). To obtain RRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 7
No. of treatments: 7
No. of pairwise comparisons: 7
No. of designs: 6

Heterogeneity/inconsistency: Q = 0.96, df = 1, P = 0.33; I² = 0.0%, Tau² = 0.0

The rankings according to the P‐score presented in Figure 17 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Although risedronate was ranked first (P‐Score: 0.90), the confidence interval was very wide and the estimate was based on one study involving 154 participants (RR 0.11, 95% CI 0.01 to 2.03). Pamidronate (P‐Score: 0.16) was ranked lowest.

The fixed‐effect model did not show any differences compared to the random‐effects model (not shown). In the entire network, Cochran's Q test and generalised I² statistics showed no important heterogeneity between studies (Q_total = 0.96, P = 0.33; I² = 0, Tau² = 0.0). Since there were no closed loops, inconsistencies could not be analysed.

Subgroup analysis

When the analysis was confined to postmenopausal women (from 4 studies), the results were very similar to those obtained from the main analysis (data not shown). Four treatments remained in the network (ibandronate, risedronate, zoledronic acid and no treatment/placebo).

For those studies with mixed populations, ibandronate and risedronate dropped out of the network. The analysis showed similar results to the main analysis, except that zoledronic acid was now ranked lowest in the ranking according to P‐score, but confidence intervals were overlapping.

Sensitivity analysis

No sensitivity analysis was performed.

Overall survival

Twenty studies reported overall survival (ABCSG‐12 2011; ABCSG‐18 2019; AZURE 2018; Aft 2012; BONADIUV 2019; D‐CARE 2013; Diel 1998; FEMZONE 2014; GAIN 2013; HOBOE 2019; Kristensen 2008; Mardiak 2000; NATAN 2016; NEO‐ZOTAC BOOG 2010; NSABP B‐34 2012; Powles 2006; SWOG S0307 2019; Saarto 2004; Solomayer 2012; Team IIB 2006). Seventeen studies including 30,991 participants were examined in the statistical analysis (ABCSG‐12 2011; ABCSG‐18 2019; AZURE 2018; BONADIUV 2019; D‐CARE 2013; GAIN 2013; HOBOE 2019; Kristensen 2008; Mardiak 2000; NATAN 2016; NEO‐ZOTAC BOOG 2010; NSABP B‐34 2012; Powles 2006; Saarto 2004; Solomayer 2012; SWOG S0307 2019; Team IIB 2006). Three studies reported overall survival in a format where it was not possible to add the data to the statistical analysis (Aft 2012; Diel 1998; FEMZONE 2014).

Of the studies included in the statistical analysis, 15 included pre and postmenopausal women. Two studies included postmenopausal women only (BONADIUV 2019; Team IIB 2006). The network diagram is presented in Figure 18. The network, based on 19 pairwise comparisons, was connected and compared six different treatment options (clodronate, denosumab, ibandronate, pamidronate, zoledronic acid and no treatment/placebo). There were four closed loops in the network. The forest plot, including all pairwise comparisons, is shown in Figure 19.

18.

Network diagram for outcome: overall survival. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients

19.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, overall survival

Network meta‐analysis

Evidence from the network meta‐analysis showed no meaningful results in comparisons of overall survival between all treatment options (Figure 20; Table 7). This is because the evidence on survival was judged to be of low or very low certainty overall as indicated below.

20.

Forest plot for outcome, overall survival: 
Random‐effects model
Reference treatment: no treatment/placebo
Treatments are ordered by P‐Score (descending)

Abbreviations:
CI: confidence interval
HR: hazard ratio

6. League table: overall survival.

Denosumab	.	.	0.91 [0.69, 1.21]	.	.
0.98 [0.69, 1.39]	Zoledronic acid	0.95 [0.65, 1.38]	0.92 [0.72, 1.17]	0.94 [0.64, 1.37]	.
0.96 [0.68, 1.37]	0.98 [0.76, 1.26]	Clodronate	0.92 [0.72, 1.19]	0.95 [0.65, 1.38]	.
0.91 [0.69, 1.21]	0.93 [0.76, 1.14]	0.95 [0.77, 1.17]	No treatment/placebo	0.90 [0.65, 1.23]	0.83 [0.56, 1.23]
0.86 [0.60, 1.26]	0.88 [0.67, 1.16]	0.90 [0.68, 1.18]	0.95 [0.74, 1.21]	Ibandronate	.
0.76 [0.47, 1.23]	0.77 [0.50, 1.20]	0.79 [0.51, 1.23]	0.83 [0.56, 1.23]	0.88 [0.56, 1.39]	Pamidronate

Results of network meta‐analysis for outcome, overall survival. Treatments are ordered by P‐Score (ascending). Only subnets with > 1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as hazard ratios (HR) with corresponding 95% confidence intervals. For the network estimates in the lower triangle, an HR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an HR below 1.0 favours the row‐defining treatment (longer overall survival). To obtain HRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 17
No. of treatments: 6
No. of pairwise comparisons: 19
No. of designs: 6

Q_total = 29.34, df = 13, P = 0.006/Q_within = 27.89, df = 11, P = 0.003/Q_between = 1.28, df = 2, P = 0.53; I² = 55.7%, Tau² = 0.0305

Clodronate (HR 0.95, 95% CI 0.77 to 1.17), denosumab (HR 0.91, 95% CI 0.69 to 1.21), ibandronate (HR 1.06, 95% CI 0.83 to 1.34) and zoledronic acid (HR 0.93, 95% CI 0.76 to 1.14) may result in little to no difference regarding overall survival compared to no treatment/placebo (low certainty). We are uncertain whether treatment with pamidronate (HR 1.20, 95% CI 0.81 to 1.78) decreases overall survival compared to no treatment/placebo (very low certainty). Our main reasons for downgrading the evidence were imprecision and inconsistency; these are provided in Table 1. It was not possible to rate the certainty of the evidence for alendronate and risedronate because they were not included in the network.

The rankings according to the P‐score presented in Figure 20 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Although denosumab was ranked first (P‐Score: 0.70), zoledronic acid (P‐Score: 0.69) and clodronate (P‐Score: 0.63) showed similar P‐Scores and estimates, and the confidence intervals were overlapping. Pamidronate (P‐Score: 0.18) was ranked lowest.

The fixed‐effect model showed slight differences compared to the random‐effects model, but confidence intervals were very wide and overlapping (not shown). In the entire network, Cochran's Q test and generalised I² statistics showed moderate to substantial heterogeneity between studies (Q_total = 29.34, P = 0.006/Q_within = 27.89, P = 0.003/Q_between = 1.28, P = 0.53; I² = 55.7 %, Tau² = 0.0305). Since there were closed loops in the network, inconsistencies could be analysed. Tests for inconsistencies in closed loops indicated no disagreements between direct and indirect estimates (Figure 21; Table 8).

21.

Comparison of direct and indirect evidence (in closed loops) for outcome overall survival. HR: hazard ratio. CI: confidence interval.

7. Comparison of direct and indirect evidence (in closed loops) for outcome, overall survival.

Comparison	No. of studies	Network estimate	Direct estimate	Indirect estimate	Test for disagreement
Clodronate vs. ibandronate	1	0.90 [0.68; 1.18]	0.95 [0.65; 1.38]	0.84 [0.57; 1.26]	0.6737
Clodronate vs. no treatment/placebo	4	0.95 [0.77; 1.17]	0.92 [0.72; 1.19]	1.01 [0.69; 1.47]	0.7158
Clodronate vs. zoledronic acid	1	1.02 [0.79; 1.32]	1.05 [0.73; 1.53]	0.99 [0.70; 1.41]	0.8245
Ibandronate vs. no treatment/placebo	3	1.06 [0.83; 1.34]	1.12 [0.81; 1.53]	0.98 [0.67; 1.42]	0.5944
Ibandronate vs. zoledronic acid	1	1.14 [0.86; 1.49]	1.06 [0.73; 1.55]	1.22 [0.82; 1.81]	0.6264
Zoledronic acid vs. no treatment/placebo	6	0.93 [0.76; 1.14]	0.92 [0.72; 1.17]	0.95 [0.65; 1.39]	0.8973

Estimates are reported as hazard ratios with corresponding 95% confidence intervals. Result of tests for disagreement between direct and indirect evidence reported as P values. Only comparisons for which both direct and indirect evidence exists are shown.

Subgroup analysis

When the analysis was confined to a mixed population of pre and postmenopausal women (from 15 studies), the results were very similar to those obtained from the main analysis. The network diagram did not show any differences at all. The ranking according to the P‐score differed slightly, with denosumab and zoledronic acid swapping their ranks as well as ibandronate and no treatment/placebo, but confidence intervals were overlapping (data not shown). There were insufficient studies in this Cochrane review to restrict the analysis to postmenopausal women.

Sensitivity analysis

Two of the 17 included studies were excluded for the sensitivity analysis due to a high risk of bias. The results did not differ, except that pamidronate dropped out of the network because one of the excluded studies used pamidronate as an intervention.

Disease‐free survival

Disease‐free survival was defined as the length of time from diagnosis to the patient surviving without any signs or symptoms (distant, loco regional, or new primary symptoms in the contralateral breast, or as defined in the trial). Seventeen studies reported disease‐free survival (ABCSG‐12 2011; ABCSG‐18 2019; Aft 2012; AZURE 2018; BONADIUV 2019; D‐CARE 2013; GAIN 2013; HOBOE 2019; JONIE 2017; NATAN 2016; NEO‐ZOTAC BOOG 2010; Novartis I 2006; NSABP B‐34 2012; Saarto 2004; SWOG S0307 2019; Team IIB 2006; Tevaarwerk 2007). Fourteen studies including 29,024 participants were examined in the statistical analysis (ABCSG‐12 2011; ABCSG‐18 2019; AZURE 2018; BONADIUV 2019; D‐CARE 2013; GAIN 2013; HOBOE 2019; JONIE 2017; NATAN 2016; NEO‐ZOTAC BOOG 2010; NSABP B‐34 2012; Saarto 2004; SWOG S0307 2019; Team IIB 2006). Three studies reported disease‐free survival in a format where it was not possible to add the data to the statistical analysis (Aft 2012; Novartis I 2006; Tevaarwerk 2007).

Of the studies included in the statistical analysis, 12 included pre and postmenopausal women. Two studies included postmenopausal women only (BONADIUV 2019; Team IIB 2006). The network diagram is presented in Figure 22. The network, based on sixteen pairwise comparisons, was connected and compared five different treatment options (clodronate, denosumab, ibandronate, zoledronic acid and no treatment/placebo). There were four closed loops in the network. The forest plot, including all pairwise comparisons, is shown in Figure 23.

22.

Network diagram for outcome: disease‐free survival. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients

23.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, disease‐free survival

Network meta‐analysis

Zoledronic acid prolongs disease‐free survival compared to no treatment/placebo (HR 0.88, 95% CI 0.81 to 0.96; high certainty; Figure 24; Table 9). No other comparison showed meaningful results.

24.

Forest plot for outcome, disease‐free survival: 
Random‐effects model
Reference treatment: no treatment/placebo
Treatments are ordered by P‐Score (descending)

Abbreviations:
CI: confidence interval
HR: hazard ratio

8. League table: disease‐free survival.

Zoledronic acid	0.92 [0.79, 1.06]	.	0.94 [0.80, 1.11]	0.90 [0.81, 0.99]
0.96 [0.86, 1.08]	Clodronate	.	0.95 [0.81, 1.11]	0.89 [0.77, 1.04]
0.94 [0.82, 1.07]	0.97 [0.84, 1.13]	Denosumab	.	0.94 [0.85, 1.05]
0.93 [0.83, 1.05]	0.97 [0.85, 1.10]	0.99 [0.85, 1.16]	Ibandronate	0.95 [0.81, 1.11]
0.88 [0.81, 0.96]	0.92 [0.82, 1.02]	0.94 [0.85, 1.05]	0.95 [0.85, 1.06]	No treatment/placebo

Results of network meta‐analysis for outcome disease‐free survival. Treatments are ordered by P‐Score (ascending). Only subnets with > 1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as hazard ratios (HR) with corresponding 95% confidence intervals. For the network estimates in the lower triangle, an HR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an HR below 1.0 favours the row‐defining treatment (longer disease‐free survival). To obtain HRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 14
No. of treatments: 5
No. of pairwise comparisons: 16
No. of designs: 5

Q_total = 11.07, df = 11, P = 0.44/Q_within = 10.52, df = 9, P = 0.31/Q_between = 0.74, df = 2, P = 0.69; I² = 0.6%, Tau² < 0.0001

The rankings according to the P‐score presented in Figure 24 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Zoledronic acid was ranked first (P‐Score: 0.86). The main comparator, no treatment/placebo, was ranked lowest (P‐Score: 0.10).

The fixed‐effect model showed similar results compared to the random‐effects model (not shown). In the entire network, Cochran's Q test and generalised I² statistics showed moderate heterogeneity between studies (Q_total = 11.07, P = 0.44 / Q_within = 10.52, P = 0.31 / Q_between = 0.74, P = 0.69; I² = 0.6 %, Tau² < 0.0001). Since there were closed loops in the network, inconsistencies could be analysed. Tests for inconsistencies in closed loops indicated no disagreements between direct and indirect estimates (Figure 25; Table 10).

25.

Comparison of direct and indirect evidence (in closed loops) for outcome, disease‐free survival

Abbreviations:
CI: confidence interval
HR: hazard ratio

9. Comparison of direct and indirect evidence (in closed loops) for outcome, disease‐free survival.

Comparison	No. of studies	Network estimate	Direct estimate	Indirect estimate	Test for disagreement
Clodronate vs. ibandronate	1	0.97 [0.85; 1.10]	0.95 [0.81; 1.11]	1.00 [0.80; 1.24]	0.7146
Clodronate vs. no treatment/placebo	2	0.92 [0.82; 1.02]	0.89 [0.77; 1.04]	0.94 [0.81; 1.10]	0.6168
Clodronate vs. zoledronic acid	1	1.04 [0.93; 1.16]	1.09 [0.94; 1.26]	0.97 [0.81; 1.16]	0.3142
Ibandronate vs. no treatment/placebo	3	0.95 [0.85; 1.06]	0.95 [0.81; 1.11]	0.95 [0.81; 1.13]	0.9501
Ibandronate vs. zoledronic acid	1	1.07 [0.95; 1.21]	1.06 [0.90; 1.25]	1.09 [0.91; 1.31]	0.8116
Zoledronic acid vs. no treatment/placebo	6	0.88 [0.81; 0.96]	0.90 [0.81; 0.99]	0.85 [0.72; 1.01]	0.6115

Estimates are reported as hazard ratios with corresponding 95% confidence intervals. Results of tests for disagreement between direct and indirect evidence reported as P values. Only comparisons for which both direct and indirect evidence exists are shown.

Subgroup analysis

A subgroup analysis involving a mixed population of pre and postmenopausal women only (from 12 studies) showed negligible differences in the results. There were no differences in the network diagram or the ranking according to the P‐score.

Sensitivity analysis

One of the 13 included studies was excluded due to a high risk of bias. Neither the network diagram nor the ranking according to the P‐score showed different results.

Adverse events

Osteonecrosis of the jaw

Twenty‐seven studies reported osteonecrosis of the jaw as an adverse event (ABCSG‐12 2011; ABCSG‐18 2019; Aft 2012; ARBI 2009; ARIBON 2012; AZURE 2018; BONADIUV 2019; D‐CARE 2013; FEMZONE 2014; GAIN 2013; GeparX 2016; HOBOE 2019; Hershman 2008; Monda 2017; NATAN 2016; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; NSABP B‐34 2012; Powles 2006; ProBONE II 2015; Rhee 2013; SWOG S0307 2019; Safra 2011; Solomayer 2012; Sun 2016; Team IIB 2006; Tevaarwerk 2007). Twelve studies including 23,527 participants were examined in the statistical analysis (AZURE 2018; Aft 2012; D‐CARE 2013; GAIN 2013; GeparX 2016; HOBOE 2019; NATAN 2016; NSABP B‐34 2012; ProBONE II 2015; SWOG S0307 2019; Solomayer 2012; Team IIB 2006). Fifteen studies reported osteonecrosis of the jaw in a format where the data could not be added to the statistical analysis (ABCSG‐12 2011; ABCSG‐18 2019; ARBI 2009; ARIBON 2012; BONADIUV 2019; FEMZONE 2014; Hershman 2008; Monda 2017; NEO‐ZOTAC BOOG 2010; NEOZOL 2018; Powles 2006; Rhee 2013; Safra 2011; Sun 2016; Tevaarwerk 2007). In these 15 studies, there were no cases of osteonecrosis of the jaw in either treatment arm.

• ABCSG‐12 2011 compared zoledronic acid to no treatment; there were 1803 participants with a median follow‐up time of 62 months.

• ABCSG‐18 2019 compared denosumab to placebo; there were 3425 participants with a median follow‐up time of 73 months.

• ARBI 2009 compared risedronate to no treatment; there were 70 participants with a follow‐up time of two years.

• ARIBON 2012 compared ibandronate to placebo; there were 50 participants with a follow‐up time of five years.

• BONADIUV 2019 compared ibandronate to placebo; there were 171 participants with a follow‐up time of five years.

• FEMZONE 2014 compared zoledronic acid to no treatment; there were 168 participants with a follow‐up time of five years.

• Hershman 2008 compared zoledronic acid to placebo; there were 114 participants with a follow‐up time of two years.

• Monda 2017 compared risedronate to no treatment; there were 84 participants with a follow‐up time of two years.

• NEO‐ZOTAC BOOG 2010 compared zoledronic acid to placebo; there were 250 participants with a follow‐up time of 60 months.

• NEOZOL 2018 compared zoledronic acid to no treatment; there were 53 participants with a median follow‐up time of 5.7 and 5.4 months respectively for each treatment arm.

• Powles 2006 compared clodronate to placebo; there were 1069 participants with a median follow‐up time of 5.6 years.

• Rhee 2013 compared alendronate to placebo; there were 98 participants and the duration of the study was six months.

• Safra 2011 compared zoledronic acid to no treatment; there were 90 participants with a follow‐up time of five years.

• Sun 2016 compared zoledronic acid to no treatment; there were 120 participants with a follow‐up time of one year.

• Tevaarwerk 2007 compared zoledronic acid to no treatment; there were 68 participants with a follow‐up time of up to ten years.

Of the studies included in the statistical analysis, 11 studies had mixed populations of pre and postmenopausal women and one study included postmenopausal women only (Team IIB 2006). The network diagram is presented in Figure 26. The network, based on 14 pairwise comparisons, was connected and compared five different treatment options (clodronate, denosumab, ibandronate, zoledronic acid and no treatment/placebo). There were four closed loops in the network. The forest plot, including all pairwise comparisons, is shown in Figure 27.

26.

Network diagram for outcome, adverse event: osteonecrosis of the jaw. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients

27.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, adverse event: osteonecrosis of the jaw

Network meta‐analysis

Evidence suggests that denosumab (RR 24.70, 95% CI 9.56 to 63.83; moderate certainty), ibandronate (RR 5.77, 95% CI 2.04 to 16.35; moderate certainty) and zoledronic acid (RR 9.41, 95% CI 3.54 to 24.99; moderate certainty) are likely to increase the occurrence of osteonecrosis of the jaw compared to no treatment/placebo. Clodronate is likely to reduce the occurrence of osteonecrosis of the jaw compared to zoledronic acid (RR 0.28, 95% CI 0.13 to 0.61; moderate certainty) and denosumab (RR 0.11, 95% CI 0.02 to 0.48; moderate certainty). Ibandronate is likely to reduce the occurrence of osteonecrosis of the jaw compared to denosumab (RR 0.23, 95% CI 0.06 to 0.96; moderate certainty) (Figure 28; Table 11). Our main reason for downgrading the evidence was imprecision. Reasons for downgrading are provided in the Table 1. It was not possible to rate the certainty of the evidence for alendronate, pamidronate and risedronate because they were not included in the network.

28.

Forest plot for outcome, adverse event: osteonecrosis of the jaw: 
Random‐effects model
Reference treatment: no treatment/placebo
Treatments are ordered by P‐Score (descending).

Abbreviations:
CI: confidence interval
RR: risk ratio

10. League table AE: osteonecrosis of the jaw.

No treatment/Placebo	0.33 [0.01, 8.17]	0.15 [0.03, 0.80]	0.12 [0.03, 0.40]	0.04 [0.02, 0.10]
0.38 [ 0.12, 1.21]	Clodronate	0.47 [0.19, 1.14]	0.28 [0.13, 0.61]	.
0.17 [ 0.06, 0.49]	0.46 [ 0.19, 1.09]	Ibandronate	0.60 [0.30, 1.17]	.
0.11 [ 0.04, 0.28]	0.28 [ 0.13, 0.61]	0.61 [ 0.32, 1.16]	Zoledronic acid	.
0.04 [ 0.02, 0.10]	0.11 [ 0.02, 0.48]	0.23 [ 0.06, 0.96]	0.38 [ 0.10, 1.49]	Denosumab

Results of network meta‐analysis for outcome adverse event: osteonecrosis of the jaw. Treatments are ordered by P‐Score (ascending). Only subnets with >1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as risk ratios (RR) with corresponding 95% confidence interval. For the network estimates in the lower triangle an RR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle an RR below 1.0 favours the row‐defining treatment (less events of osteonecrosis of the jaw). To obtain RRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 12. No. of treatments: 5. No. of pairwise comparisons: 14. No. of designs: 5

Q_total=5.19, df=9, p=0.82 / Q_within=5.12, df=7, p=0.65 / Q_between=0.08, df=2, p=0.96; I²=0.0%, Tau²=0.0

The rankings according to the P‐score presented in Figure 28 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. No treatment/placebo was ranked first (P‐Score: 0.99), followed by clodronate (P‐Score: 0.75). Denosumab was ranked lowest (P‐Score: 0.03).

The fixed‐effect model did not show any differences compared to the random‐effects model (not shown). In the entire network, Cochran's Q test and generalised I² statistics showed no important heterogeneity between studies (Q_total = 5.19, P = 0.82 / Q_within = 5.12, P = 0.65 / Q_between = 0.08, P = 0.96; I² = 0.0%, Tau² = 0.0). Since there were closed loops in the network, inconsistencies could be analysed. Tests for inconsistencies in closed loops indicated no disagreements between direct and indirect estimates (Figure 29; Table 12).

29.

Comparison of direct and indirect evidence (in closed loops) for outcome adverse event: osteonecrosis of the jaw

Abbreviations:
CI: confidence interval
RR: risk ratio

11. Comparison of direct and indirect evidence (in closed loops) for outcome, adverse event: osteonecrosis of the jaw.

Comparison	No. of studies	Network estimate	Direct estimate	Indirect estimate	Test for disagreement
Clodronate vs. ibandronate	1	0.46 [0.19; 1.09]	0.47 [0.19; 1.14]	0.38 [0.01; 11.96]	0.9119
Clodronate vs. no treatment/placebo	1	2.65 [0.83; 8.50]	3.00 [0.12; 73.63]	2.60 [0.75; 9.09]	0.9354
Clodronate vs. zoledronic acid	1	0.28 [0.13; 0.61]	0.28 [0.13; 0.61]	0.38 [0.01; 11.14]	0.8656
Ibandronate vs. no treatment/placebo	2	5.77 [2.04; 16.35]	6.81 [1.25; 37.11]	5.21 [1.39; 19.53]	0.8074
Ibandronate vs. zoledronic acid	1	0.61 [0.32; 1.16]	0.60 [0.30; 1.17]	0.80 [0.10; 6.50]	0.7911
Zoledronic acid vs. no treatment/placebo	6	9.41 [3.54; 24.99]	8.47 [2.48; 28.92]	11.28 [2.25; 56.49]	0.7813

Estimates are reported as risk ratios with corresponding 95% confidence intervals. Results of tests for disagreement between direct and indirect evidence reported as P values. Only comparisons for which both direct and indirect evidence exists are shown.

Subgroup analysis

When the analysis was confined to a mixed population of pre and postmenopausal women (from 11 studies), the results were very similar to those obtained from the main analysis. There were no differences in the network diagram or the ranking according to the P‐score (data not shown).

Sensitivity analysis

One of the 12 studies was considered to be at a high risk of bias and was excluded in the sensitivity analysis. The results showed negligible differences, except that ibandronate no longer resulted in fewer events of osteonecrosis of the jaw compared to denosumab (RR 0.25, 95% CI 0.06 to 1.07), but confidence intervals were overlapping. There were no differences in the network diagram or the ranking according to the P‐score (data not shown).

Renal impairment

Eighteen studies reported renal impairment as an adverse event (ABCSG‐12 2011; ABCSG‐18 2019; AZURE 2018; Ellis 2008; FEMZONE 2014; GAIN 2013; Hershman 2008; HOBOE 2019; NATAN 2016; NEOZOL 2018; NSABP B‐34 2012; Powles 2006; Saarto 2004; Safra 2011; Sun 2016; SWOG S0307 2019; Team IIB 2006; Tevaarwerk 2007). Twelve studies including 22,469 participants were examined in the statistical analysis (ABCSG‐18 2019; AZURE 2018; FEMZONE 2014; GAIN 2013; HOBOE 2019; NATAN 2016; NEOZOL 2018; NSABP B‐34 2012; Powles 2006; Sun 2016; SWOG S0307 2019; Team IIB 2006). In six studies, there were no cases of renal impairment in either treatment arm (ABCSG‐12 2011; Ellis 2008; Hershman 2008; Saarto 2004; Safra 2011; Tevaarwerk 2007):

• ABCSG‐12 2011 compared zoledronic acid to no treatment; there were 1803 participants with a median follow‐up time of 62 months.

• Ellis 2008 compared denosumab to placebo; there were 252 participants with a follow‐up time of four years.

• Hershman 2008 compared zoledronic acid to placebo; there were 114 participants with a follow‐up time of two years.

• Saarto 2004 compared clodronate to no treatment; there were 299 participants with a follow‐up time of ten years.

• Safra 2011 compared zoledronic acid to no treatment; there were 90 participants with a follow‐up time of five years.

• Tevaarwerk 2007 compared zoledronic acid to no treatment; there were 68 participants with a follow‐up time of up to ten years.

Of the studies included in the statistical analysis, 10 included pre and postmenopausal women, while two studies included postmenopausal women only (FEMZONE 2014; Team IIB 2006). The network diagram is presented in Figure 30. The network, based on 14 pairwise comparisons, was connected and compared five different treatment options (clodronate, denosumab, ibandronate, zoledronic acid and no treatment/placebo). There were four closed loops in the network. A forest plot, including all pairwise comparisons, is shown in Figure 31.

30.

Network diagram for outcome adverse event: renal impairment. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients

31.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, adverse event: renal impairment

Network meta‐analysis

Evidence suggests denosumab (RR 0.40, 95% CI 0.19 to 0.88; moderate certainty), clodronate (RR 0.44, 95% CI 0.20 to 0.96; moderate certainty) and no treatment/placebo (RR 0.50, 95% CI 0.26 to 0.99; moderate certainty) are likely to reduce the occurrence of renal impairment compared to ibandronate (Figure 32; Table 13). Ibandronate increases the occurrence of renal impairment compared to no treatment/placebo (RR 1.98, 95% CI 1.01 to 3.88; moderate certainty). Zoledronic acid probably increases the occurrence of renal impairment compared to no treatment/placebo (RR 1.49, 95% CI 0.87 to 2.58; moderate certainty) while clodronate and denosumab probably result in little to no difference regarding the occurrence of renal impairment compared to no treatment/placebo (RR 0.88, 95% CI 0.55 to 1.39, RR 0.80, 95% CI 0.54 to 1.19, respectively, moderate certainty). Reasons for downgrading the evidence are provided in Table 1. It was not possible to rate the certainty of the evidence for alendronate, pamidronate and risedronate because they were not included in the network.

32.

Forest plot for outcome, adverse event: renal impairment: 
Random‐effects model
Reference treatment: no treatment/placebo
Treatments are ordered by P‐Score (descending).

Abbreviations:
CI: confidence interval
RR: risk ratio

12. League table: AE: renal impairment.

Denosumab	.	0.80 [0.54, 1.19]	.	.
0.92 [0.50, 1.68]	Clodronate	0.92 [0.57, 1.49]	0.65 [0.11, 3.87]	0.20 [0.04, 0.96]
0.80 [0.54, 1.19]	0.88 [0.55, 1.39]	No treatment/Placebo	0.58 [0.32, 1.04]	0.77 [0.34, 1.72]
0.54 [0.27, 1.05]	0.58 [0.29, 1.18]	0.67 [0.39, 1.16]	Zoledronic acid	0.31 [0.08, 1.19]
0.40 [0.19, 0.88]	0.44 [0.20, 0.96]	0.50 [0.26, 0.99]	0.75 [0.34, 1.66]	Ibandronate

Results of network meta‐analysis for outcome, adverse event: renal. Treatments are ordered by P‐Score (ascending). Only subnets with > 1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as risk ratios (RR) with corresponding 95% confidence intervals. For the network estimates in the lower triangle, an RR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an RR below 1.0 favours the row‐defining treatment (fewer events of renal impairment). To obtain RRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 12
No. of treatments: 5
No. of pairwise comparisons: 14
No. of designs: 5

Q_total = 5.40, df = 9, P = 0.80/Q_within = 1.92, df = 7, P = 0.96/Q_between = 3.48, df = 2, P = 0.18; I² = 0.0%, Tau² = 0.0

The rankings according to the P‐score presented in Figure 32 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Although denosumab was ranked first (P‐Score: 0.86), clodronate (P‐Score: 0.75) resulted in a similar estimate, and the confidence intervals were overlapping. Ibandronate was ranked lowest (P‐Score: 0.07).

The fixed‐effect model did not show any differences compared to the random‐effects model (data not shown). In the entire network, Cochran's Q test and generalised I² statistics showed no important heterogeneity between studies (Q_total = 5.40, P = 0.80 / Q_within = 1.92, P = 0.96 / Q_between = 3.48, P = 0.18; I² = 0.0 %, Tau² = 0.0). Since there were closed loops in the network, inconsistencies could be analysed. Tests for inconsistencies in closed loops indicated no disagreements between direct and indirect estimates (Figure 33; Table 14).

33.

Comparison of direct and indirect evidence (in closed loops) for outcome, adverse event: renal impairment

Abbreviations:
CI: confidence interval
RR: risk ratio

13. Comparison of direct and indirect evidence (in closed loops) for outcome, adverse event: renal impairment.

Comparsison	No. of studies	Network estimate	Direct estimate	Indirect estimate	Test for disagreement
Clodronate vs. ibandronate	1	0.44 [0.20; 0.96]	0.20 [0.04; 0.96]	0.57 [0.23; 1.39]	0.2543
Clodronate vs. no treatment/placebo	2	0.88 [0.55; 1.39]	0.92 [0.57; 1.49]	0.48 [0.10; 2.43]	0.4500
Clodronate vs. zoledronic acid	1	0.59 [0.29; 1.18]	0.65 [0.11; 3.87]	0.58 [0.27; 1.23]	0.9059
Ibandronate vs. no treatment/placebo	2	1.98 [1.01; 3.88]	1.30 [0.58; 2.91]	5.16 [1.53; 17.37]	0.0636
Ibandronate vs. zoledronic acid	1	1.33 [0.60; 2.93]	3.25 [0.84; 12.53]	0.83 [0.31; 2.21]	0.1098
Zoledronic acid vs. no treatment/placebo	6	1.49 [0.87; 2.58]	1.73 [0.96; 3.12]	0.63 [0.15; 2.64]	0.2017

Estimates are reported as risk ratios with corresponding 95% confidence intervals. Results of tests for disagreement between direct and indirect evidence reported as P values. Only comparisons for which both direct and indirect evidence exists are shown.

Subgroup analysis

Two of the twelve included studies consisted of postmenopausal women only. When the subgroup analysis was confined to a mixed population of pre and postmenopausal women (from 10 studies), the results were very similar to those obtained from the main analysis except that clodronate (RR 0.43, 95% CI 0.18 to 1.02) and no treatment/placebo (RR 0.49, 95% CI 0.22 to 1.08) no longer resulted in fewer events of renal impairment compared to denosumab, but confidence intervals were overlapping. There were no differences in the network diagram or the ranking of treatments (data not shown).

Sensitivity analysis

None of the twelve studies had a high risk of bias, therefore no sensitivity analysis was performed.

Bone pain

Eighteen studies reported bone pain as an adverse event (ABCSG‐12 2011; ABCSG‐18 2019; Aft 2012; Delmas 1997; Ellis 2008; EXPAND 2011; FEMZONE 2014; Hershman 2008; HOBOE 2019; NATAN 2016; NEO‐ZOTAC BOOG 2010; Novartis I 2006; NSABP B‐34 2012; ProBONE II 2015; SABRE 2010; Solomayer 2012; Sun 2016; SWOG S0307 2019). Seventeen studies including 17,059 participants were examined in the statistical analysis (ABCSG‐12 2011; ABCSG‐18 2019; Aft 2012; Ellis 2008; EXPAND 2011; FEMZONE 2014; Hershman 2008; HOBOE 2019; NATAN 2016; NEO‐ZOTAC BOOG 2010; Novartis I 2006; NSABP B‐34 2012; ProBONE II 2015; SABRE 2010; Solomayer 2012; Sun 2016; SWOG S0307 2019). One study reported bone pain in a format that prevented inclusion of its data in the statistical analysis (Delmas 1997). There were no occurrences of bone pain in either treatment arm.

Of the 17 studies included in the statistical analysis, 13 studies included pre and postmenopausal women. Four studies included postmenopausal women only. The network diagram is presented in Figure 34. The network, based on nineteen pairwise comparisons, was connected and compared six different treatment options (clodronate, denosumab, ibandronate, risedronate, zoledronic acid and no treatment/placebo). There were three closed loops in the network. A forest plot, including all pairwise comparisons, is shown in Figure 35.

34.

Network diagram for outcome, adverse event: bone pain. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients

35.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, adverse event: bone pain

Network meta‐analysis

Evidence suggests that clodronate (RR 0.62, 95% CI 0.40 to 0.98; moderate certainty) and no treatment/placebo (RR 0.80, 95% CI 0.66 to 0.98; moderate certainty) are likely to reduce the occurrence of bone pain compared to zoledronic acid. Additionally, clodronate is likely to reduce the occurrence of bone pain compared to ibandronate (RR 0.57, 95% CI 0.35 to 0.94; moderate certainty) (Figure 36; Table 15).

36.

Forest plot for outcome, adverse event: bone pain:
Random‐effects model
Reference treatment: no treatment/placebo
Treatments are ordered by P‐Score (descending).

Abbreviations:
CI: confidence interval
RR: risk ratio

14. League table: AE: bone pain.

Clodronate	0.76 [0.25, 2.32]	.	.	0.63 [0.38, 1.03]	0.57 [0.35, 0.95]
0.78 [0.48, 1.25]	No treatment/placebo	1.00 [0.06, 16.11]	0.87 [0.56, 1.34]	0.80 [0.66, 0.98]	.
0.78 [0.05, 13.03]	1.00 [0.06, 16.11]	Risedronate	.	.	.
0.67 [0.35, 1.29]	0.87 [0.56, 1.34]	0.87 [0.05, 14.44]	Denosumab	.	.
0.62 [0.40, 0.98]	0.80 [0.66, 0.98]	0.80 [0.05, 13.03]	0.93 [0.57, 1.50]	Zoledronic acid	0.92 [0.57, 1.48]
0.57 [0.35, 0.94]	0.74 [0.45, 1.22]	0.74 [0.04, 12.44]	0.85 [0.44, 1.66]	0.92 [0.57, 1.47]	Ibandronate

Results of network meta‐analysis for outcome, adverse event: bone pain. Treatments are ordered by P‐Score (ascending). Only subnets with > 1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as risk ratios (RR) with corresponding 95% confidence intervals. For the network estimates in the lower triangle, an RR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an RR below 1.0 favours the row‐defining treatment (fewer events of bone pain). To obtain RRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 17
No. of treatments: 6
No. of pairwise comparisons: 19
No. of designs: 5

Q_total = 21.49, df = 13, P = 0.064/Q_within = 21.48, df = 12, P = 0.044/Q_between = 0.01, df = 1, P = 0.92; I² = 39.5%, Tau² = 0.0366

The rankings according to the P‐score presented in Figure 36 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Clodronate was ranked first (P‐Score: 0.85), followed by no treatment/placebo (P‐Score: 0.65). Zoledronic acid (P‐Score: 0.30) and ibandronate (P‐Score: 0.25) were ranked the lowest.

The fixed‐effect model showed slight differences compared to the random‐effects model (data not shown). In the entire network, Cochran's Q test and generalised I² statistics showed moderate heterogeneity between studies (Q_total = 21.49, P = 0.064 / Q_within = 21.48, P = 0.044 / Q_between = 0.01, P = 0.92; I² = 39.5 %, Tau² = 0.0366). Since there were closed loops in the network, inconsistencies could be analysed. Tests for inconsistencies in closed loops indicated no disagreements between direct and indirect estimates (Figure 37; Table 16).

37.

Comparison of direct and indirect evidence (in closed loops) for outcome, adverse event: bone pain

Abbreviations:
CI: confidence interval
RR: risk ratio

15. Comparison of direct and indirect evidence (in closed loops) for outcome, adverse event: bone pain.

Comparison	No. of studies	Network estimate	Direct estimate	Indirect estimate	Test for disagreement
Clodronate vs. ibandronate	1	0.57 [0.35; 0.94]	0.57 [0.35; 0.95]	0.54 [0.05; 5.44]	0.9576
Clodronate vs. no treatment/placebo	1	0.78 [0.48; 1.25]	0.76 [0.25; 2.32]	0.78 [0.46; 1.33]	0.9576
Clodronate vs. zoledronic acid	1	0.62 [0.40; 0.98]	0.63 [0.38; 1.03]	0.61 [0.19; 1.89]	0.9576
Ibandronate vs. zoledronic acid	1	1.09 [0.68; 1.75]	1.09 [0.68; 1.76]	1.02 [0.08; 13.11]	0.9576
Zoledronic acid vs. no treatment/placebo	12	1.24 [1.02; 1.52]	1.25 [1.02; 1.52]	1.20 [0.35; 4.09]	0.9576

Estimates are reported as risk ratios with corresponding 95% confidence intervals. Results of tests for disagreement between direct and indirect evidence reported as P values. Only comparisons for which both direct and indirect evidence exists are shown.

Subgroup analysis

When the subgroup analysis was confined to the 13 studies with both pre and postmenopausal women, the network diagram retained two closed loops and risedronate was absent as a treatment option. In this cluster analysis, the results were similar to the main analysis, except that clodronate (RR 0.63, 95% CI 0.39 to 1.01) and no treatment/placebo (RR 0.82, 95% CI 0.65 to 1.04) no longer resulted in fewer events of bone pain compared with zoledronic acid. The ranking according to the P‐score did not show any differences compared to the main analysis (data not shown).

When the subgroup analysis was confined to studies including postmenopausal only (4 studies), the network consisted of three different treatments (zoledronic acid, risedronate and no treatment/placebo). In this cluster analysis, the results were similar to those in the main analysis, except that no treatment/placebo no longer resulted in fewer events of bone pain compared with zoledronic acid (RR 0.73, 95% CI 0.47 to 1.13). The ranking according to the P‐score did not differ compared to the ranking in the main analysis (data not shown).

Sensitivity analysis

Of the 17 included studies, 11 were considered to be at a high risk of bias and were therefore excluded in the sensitivity analysis. Two different sensitivity analyses were conducted ‐ one including all participants and one including studies with a mixed population exclusively. In the analysis including all participants, ibandronate dropped out of the network. In general, the results showed similar effect estimates, but larger confidence intervals. Clodronate (RR 0.75, 95% CI 0.24 to 2.41) and no treatment/placebo (RR 1.00, 95% CI 0.66 to 1.51) no longer showed fewer events of bone pain compared to zoledronic acid. The ranking of treatments differed, with denosumab now ranked lowest, but confidence intervals were overlapping. In the analysis including studies with a mixed population, only risedronate additionally dropped out of the network. The results were similar to that of the first sensitivity analysis (data not shown).

Hypocalcaemia

Ten studies reported hypocalcaemia as an adverse event (ABCSG‐18 2019; D‐CARE 2013; Ellis 2008; GeparX 2016; HOBOE 2019; Kanis 1996; NEO‐ZOTAC BOOG 2010; NSABP B‐34 2012; SWOG S0307 2019; Tevaarwerk 2007). Seven studies including 18,628 participants were examined in the statistical analysis (ABCSG‐18 2019; D‐CARE 2013; GeparX 2016; HOBOE 2019; NEO‐ZOTAC BOOG 2010; NSABP B‐34 2012; SWOG S0307 2019). Three studies reported hypocalcaemia in a format where it was not possible to add the data to the statistical analysis (Ellis 2008; Kanis 1996; Tevaarwerk 2007). In these three studies, there were no cases of hypocalcaemia reported in either treatment arm:

• Ellis 2008 compared denosumab to placebo; there were 252 participants with a follow‐up time of four years.

• Kanis 1996 compared clodronate to placebo; there were 133 participants and the duration of the study was three years.

• Tevaarwerk 2007 compared zoledronic acid to no treatment; there were 68 participants with a follow‐up time of up to ten years.

Of the studies included in the statistical analysis, all studies had mixed populations of pre and postmenopausal women. The network diagram is presented in Figure 38. The network, based on nine pairwise comparisons, was connected and compared five different treatment options (clodronate, denosumab, ibandronate, zoledronic acid and no treatment/placebo). There were two closed loops in the network. A forest plot, including all pairwise comparisons, is shown in Figure 39.

38.

Network diagram for outcome, adverse event: hypocalcaemia. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients

39.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, adverse event: hypocalcaemia

Network meta‐analysis

Evidence suggests that no treatment/placebo reduces the occurrence of hypocalcaemia compared to denosumab (RR 0.55, 95% CI 0.48 to 0.64; high certainty) (Figure 40; Table 17). No other comparison showed meaningful results because the evidence for hypocalcaemia was judged to be low certainty.

40.

Forest plot for outcome, adverse event: hypocalcaemia: 
Random‐effects model
Reference treatment: no treatment/placebo
Treatments are ordered by P‐Score (descending).

Abbreviations:
CI: confidence interval
RR: risk ratio

16. League table: AE: hypocalcaemia.

Ibandronate	0.72 [0.06, 7.88]	0.46 [0.05, 4.45]	.	.
0.66 [0.06, 6.77]	Clodronate	0.65 [0.11, 3.87]	0.50 [0.05, 5.55]	.
0.49 [0.05, 4.55]	0.74 [0.16, 3.34]	Zoledronic acid	0.49 [0.12, 1.99]	.
0.26 [0.02, 3.09]	0.39 [0.08, 2.06]	0.53 [0.15, 1.90]	No treatment/placebo	0.55 [0.48, 0.64]
0.14 [0.01, 1.72]	0.22 [0.04, 1.15]	0.29 [0.08, 1.06]	0.55 [0.48, 0.64]	Denosumab

Results of network meta‐analysis for outcome, adverse event: hypocalcaemia. Treatments are ordered by P‐Score (ascending). Only subnets with > 1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as risk ratios (RR) with corresponding 95% confidence intervals. For the network estimates in the lower triangle, an RR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an RR below 1.0 favours the row‐defining treatment (fewer events of hypocalcaemia). To obtain RRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 7
No. of treatments: 5
No. of pairwise comparisons: 9
No. of designs: 4

Q_total = 1.86, df = 4, P = 0.76/Q_within = 1.78, df = 3, P = 0.62/Q_between = 0.07, df = 1, P = 0.78; I² = 0.0%, Tau² = 0.0

The rankings according to the P‐score presented in Figure 40 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Although ibandronate was ranked first (P‐Score: 0.79), the confidence interval was very wide and the estimate was based on one study involving 3651 participants (RR 0.26, 95% CI 0.02 to 3.09). Denosumab was ranked lowest (P‐Score: 0.03).

The fixed‐effect model did not show any differences compared to the random‐effects model. In the entire network, Cochran's Q test and generalised I² statistics showed no important heterogeneity between studies (Q_total = 1.86, P = 0.76 / Q_within = 1.78, P = 0.62 / Q_between = 0.07, P = 0.78; I² = 0.0 %, Tau² = 0.0). Since there were closed loops in the network, inconsistencies could be analysed. Tests for inconsistencies in closed loops indicated no disagreements between direct and indirect estimates (Figure 41; Table 18).

41.

Comparison of direct and indirect evidence (in closed loops) for outcome, adverse event: hypocalcaemia

Abbreviations:
CI: confidence interval
RR: risk ratio

17. Comparison of direct and indirect evidence (in closed loops) for outcome, adverse event: hypocalcaemia.

Comparison	No. of studies	Network estimate	Direct estimate	Indirect estimate	Test for disagreement
Clodronate vs. ibandronate	1	1.52 [0.15; 15.55]	1.40 [0.13; 15.40]	5.59 [0.00; 84531.23]	0.7842
Clodronate vs. no treatment/placebo	1	0.39 [0.08; 2.06]	0.50 [0.05; 5.55]	0.32 [0.03; 3.08]	0.7842
Clodronate vs. zoledronic acid	1	0.74 [0.16; 3.34]	0.65 [0.11; 3.87]	1.03 [0.06; 16.58]	0.7842
Ibandronate vs. zoledronic acid	1	0.49 [0.05; 4.55]	0.46 [0.05; 4.45]	2.94 [0.00; 1349690.87]	0.7842
Zoledronic acid vs. no treatment/placebo	2	0.53 [0.15; 1.90]	0.49 [0.12; 1.99]	0.78 [0.04; 15.49]	0.7842

Estimates are reported as risk ratios with corresponding 95% confidence intervals. Results of tests for disagreement between direct and indirect evidence reported as P values. Only comparisons for which both direct and indirect evidence exists are shown.

Subgroup analysis

All studies reporting hypocalcaemia consisted of mixed populations. Therefore, no subgroup analysis, separating studies with only pre or postmenopausal patients, was performed.

Sensitivity analysis

None of the included studies was considered to be at a high risk of bias. Therefore, no sensitivity analysis was performed.

Bone recurrence

Twenty‐one studies reported bone recurrence as an adverse event (ABCSG‐12 2011; ARIBON 2012; AZURE 2018; D‐CARE 2013; Diel 1998; Ellis 2008; EXPAND 2011; GAIN 2013; HOBOE 2019; JONIE 2017; Kanis 1996; Kristensen 2008; Mardiak 2000; NATAN 2016; Novartis I 2006; NSABP B‐34 2012; Powles 2006; Saarto 2004; Solomayer 2012; SWOG S0307 2019; Team IIB 2006). Eighteen studies including 27,087 participants were examined in the statistical analysis (ABCSG‐12 2011; ARIBON 2012; AZURE 2018; D‐CARE 2013; Ellis 2008; EXPAND 2011; GAIN 2013; HOBOE 2019; JONIE 2017; Kristensen 2008; Mardiak 2000; NATAN 2016; Novartis I 2006; NSABP B‐34 2012; Powles 2006; Saarto 2004; SWOG S0307 2019; Team IIB 2006). Three studies reported bone recurrence in a format where it was not possible to add the data to the statistical analysis (Diel 1998; Kanis 1996; Solomayer 2012). In one study, there were no cases of bone recurrence reported in either treatment arm:

• Solomayer 2012 compared zoledronic acid to no treatment/placebo; there were 96 participants with a median follow‐up time of 88 months.

Of the studies included in the statistical analysis, 14 studies had mixed populations of pre and postmenopausal women. Four of the studies included postmenopausal women only. The network diagram is presented in Figure 42. The network, based on 20 pairwise comparisons, was connected and compared six different treatment options (clodronate, denosumab, ibandronate, pamidronate, zoledronic acid and no treatment/placebo). There were four closed loops in the network. A forest plot, including all pairwise comparisons, is shown in Figure 43.

42.

Network diagram for outcome: occurrence of bone metastases. Any two treatments are connected by a line when there is at least one study comparing the two treatments. Line width: number of studies and node size: number of patients

43.

Forest plot of pairwise comparisons for the descriptive presentation of studies for outcome, occurrence of bone metastases

Network meta‐analysis

Denosumab (RR 0.75, 95% CI 0.59 to 0.95; high certainty), clodronate (RR 0.82, 95% CI 0.70 to 0.95; high certainty), ibandronate (RR 0.83, 95% CI 0.75 to 0.91; high certainty) and zoledronic acid (RR 0.83, 95% CI 0.73 to 0.94; high certainty) reduce the occurrence of bone metastasis compared to no treatment/placebo. Additionally, denosumab (RR 0.66, 95% CI 0.47 to 0.95; high certainty), clodronate (RR 0.73, 95% CI 0.54 to 0.99; high certainty), ibandronate (RR 0.73, 95% CI 0.56 to 0.97; high certainty) and zoledronic acid (RR 0.74, 95% CI 0.55 to 0.98; high certainty) are even likely to reduce the occurrence of bone metastasis compared to pamidronate (Figure 44; Table 19).

44.

Forest plot for outcome occurrence of bone metastases: Random effects model. Reference treatment: No treatment/placebo. Treatments are ordered by P‐Score (descending). RR: risk ratio. CI: confidence interval.

18. League table: occurrence of bone metastases.

Denosumab	.	.	.	0.75 [0.59, 0.95]	.
0.91 [0.69, 1.21]	Clodronate	0.91 [0.69, 1.21]	0.98 [0.76, 1.27]	0.84 [0.70, 1.02]	.
0.91 [0.70, 1.17]	0.99 [0.84, 1.17]	Ibandronate	1.07 [0.81, 1.42]	0.82 [0.74, 0.90]	.
0.90 [0.69, 1.18]	0.99 [0.83, 1.18]	1.00 [0.86, 1.16]	Zoledronic acid	0.84 [0.72, 0.97]	.
0.75 [0.59, 0.95]	0.82 [0.70, 0.95]	0.83 [0.75, 0.91]	0.83 [0.73, 0.94]	No treatment/placebo	0.89 [0.68, 1.15]
0.66 [0.47, 0.95]	0.73 [0.54, 0.99]	0.73 [0.56, 0.97]	0.74 [0.55, 0.98]	0.89 [0.68, 1.15]	Pamidronate

Results of network meta‐analysis for outcome, occurrence of bone metastases. Treatments are ordered by P‐Score (ascending). Only subnets with > 1 designs. Upper triangle: direct estimates; lower triangle: network estimates. Comparisons should be read from left to right, and the estimate is in the cell in common between the column‐defining treatment and the row‐defining treatment. Effect estimates are presented as risk ratios (RR) with corresponding 95% confidence intervals. For the network estimates in the lower triangle, an RR below 1.0 favours the column‐defining treatment and for the direct estimates in the upper triangle, an RR below 1.0 favours the row‐defining treatment (fewer events of bone metastasis). To obtain RRs for comparisons in the opposing direction, reciprocals should be taken.

No. of studies: 18
No. of treatments: 6
No. of pairwise comparisons: 20
No. of designs: 6

Q_total = 10.51, df = 14, P = 0.72/Q_within = 9.96, df = 12, P = 0.62/Q_between = 0.54, df = 2, P = 0.76; I² = 0.0%, Tau² = 0.0

The rankings according to the P‐score presented in Figure 44 need to be interpreted in conjunction with the associated confidence intervals, certainty of estimates and caution. Denosumab was ranked first (P‐Score: 0.85). Pamidronate was ranked lowest.

The fixed‐effect model did not show any differences compared to the random‐effects model (data not shown). In the entire network, Cochran's Q test and generalised I² statistics showed no important heterogeneity between studies (Q_total = 10.51, P = 0.72 / Q_within = 9.96, P = 0.62 /Q_between = 0.54, P = 0.76; I² = 0.0 %, Tau² = 0.0). Since there were closed loops in the network, inconsistencies could be analysed. Tests for inconsistencies in closed loops indicated no disagreements between direct and indirect estimates (Figure 45; Table 20).

45.

Comparison of direct and indirect evidence (in closed loops) for outcome, occurrence of bone metastases

Abbreviations:
CI: confidence interval
RR: risk ratio

19. Comparison of direct and indirect evidence (in closed loops) for outcome, occurrence of bone metastases.

Comparison	No. of studies	Network estimate	Direct estimate	Indirect estimate	Test for disagreement
Clodronate vs. ibandronate	1	0.99 [0.84; 1.17]	0.91 [0.69; 1.21]	1.04 [0.84; 1.28]	0.4711
Clodronate vs. no treatment/placebo	4	0.82 [0.70; 0.95]	0.84 [0.70; 1.02]	0.78 [0.61; 1.00]	0.6310
Clodronate vs. zoledronic acid	1	0.99 [0.83; 1.18]	0.98 [0.76; 1.27]	1.00 [0.79; 1.27]	0.9116
Ibandronate vs. no treatment/placebo	3	0.83 [0.75; 0.91]	0.82 [0.74; 0.90]	0.91 [0.69; 1.19]	0.4718
Ibandronate vs. zoledronic acid	1	1.00 [0.86; 1.16]	1.07 [0.81; 1.42]	0.97 [0.82; 1.15]	0.5520
Zoledronic acid vs. no treatment/placebo	7	0.83 [0.73; 0.94]	0.84 [0.72; 0.97]	0.80 [0.62; 1.03]	0.7862

Estimates are reported as risk ratios with corresponding 95% confidence intervals. Results of tests for disagreement between direct and indirect evidence reported as P values. Only comparisons for which both direct and indirect evidence exists are shown.

Subgroup analysis

When analyses were restricted to studies with mixed populations only (14 studies), the analysis showed very similar results compared to the main analysis. Neither the effect estimates nor the ranking of treatments differed and there were negligible differences in some confidence intervals (data not shown).

In the cluster including studies with postmenopausal women only, four studies were included. The remaining treatments were ibandronate, zoledronic acid and no treatment/placebo. Compared to no treatment/placebo, ibandronate (RR 0.80, 95% CI 0.53 to 1.20) and zoledronic acid (RR 0.33, 95% CI 0.03 to 3.06) no longer resulted in fewer events of bone metastasis. The ranking according to the P‐score also differed slightly, with zoledronic acid now being ranked higher than ibandronate, but overall confidence intervals were very wide and overlapping (data not shown).

Sensitivity analysis

Four studies were considered to be at a high risk of bias and were therefore excluded from the sensitivity analysis. Two different sensitivity analyses were conducted ‐ one including all participants and one including studies with a mixed population exclusively. In the analysis involving all participants, pamidronate dropped out of the network. The results were very similar to the main analysis. Neither the effect estimates nor the ranking of treatments differed and there were negligible differences in some confidence intervals. In the analysis involving a mixed population, the results indicated the same treatment ranking and negligible differences in effect estimates and confidence intervals (data not shown).

Discussion

Summary of main results

The objectives of this review were to systematically evaluate the effect of different bone‐modifying agents for women with breast cancer in early or locally advanced stages, and to collect further information on the safety and efficacy of these interventions. We identified 47 studies involving 35,163 participants and investigating six bisphosphonates, one RANKL‐inhibitor ‐ denosumab, and no treatment/placebo. From the 47 studies, 34 studies, including 33,793 participants, were considered in quantitative analyses and 13 studies could not be included in the analyses. For each outcome, the network was connected, so a ranking of all treatment options was possible; however, a clear winner could not be determined. See Table 1.

Bone mineral density: zoledronic acid leads to increased bone mineral density compared to no treatment/placebo. Ibandronate may slightly increase bone mineral density compared to no treatment/placebo. Risedronate may result in little to no difference regarding bone mineral density compared to no treatment/placebo and it is uncertain whether alendronate increases bone mineral density compared to no treatment/placebo.

Quality of life: three studies reported quality of life; all three studies showed minimal differences in quality of life between the respective interventions examined (zoledronic acid, denosumab, risedronate and no treatment/placebo).

Fracture rate: clodronate and ibandronate decrease the number of fractures compared to no treatment/placebo and pamidronate. Denosumab and zoledronic acid probably slightly decrease the number of fractures compared to no treatment/placebo while pamidronate probably increases the number of fractures compared to no treatment/placebo. Finally, risedronate may decrease or increase the number of fractures compared to no treatment/placebo.

Overall survival: evidence from network meta‐analyses showed no meaningful results in comparisons of overall survival between all treatment options. Clodronate, denosumab, ibandronate and zoledronic acid may result in little to no difference regarding overall survival compared to no treatment/placebo. Contrary to these results, a comprehensive review using individual participant data has shown that, in postmenopausal women, there is a survival benefit from bisphosphonates (EBCTCG 2015). The difference in results between this Cochrane review and the EBCTCG 2015 review is that the latter review combines individual participant data rather than aggregate data, and important analyses by menopausal status were possible.

Osteonecrosis of the jaw: clodronate resulted in fewer events of osteonecrosis of the jaw compared to zoledronic acid, and ibandronate leads to a reduced risk for osteonecrosis of the jaw compared to denosumab. Ibandronte, zoledronic acid and denosumab lead to a higher risk for osteonecrosis of the jaw compared to no treatment/placebo.

Renal impairment: clodronate, denosumab and no treatment/placebo lead to a decrease in the number of renal impairments compared to ibandronate. Ibandronate and zoledronic acid likely increase the occurrence of renal impairment compared to no treatment/placebo, while clodronate and denosumab show minimal difference in the occurrence of renal impairment compared to no treatment/placebo.

Overall completeness and applicability of evidence

We were able to compare all eight treatment options presented in our ideal network, which consisted of different bisphosphonates, the RANKL‐inhibitor denosumab and no treatment/placebo for the prevention of bone loss in women with early or locally advanced breast cancer.

Not all studies reported on key outcomes, resulting in very different graphical networks for each outcome. Our main outcome, bone mineral density, was measured differently across studies, so information from a subset of studies could be used for our analysis.

For each outcome, a connected network could be analysed, so all respective treatment options could be compared for each outcome.

We detected substantial inconsistency in the outcome of bone mineral density, moderate‐to‐substantial inconsistency for overall survival, and moderate inconsistency within the network for bone pain, indicating differences within pairwise comparisons. We found no signs of inconsistencies between direct and indirect evidence. However, this inconsistency within pairwise comparisons can not be statistically explained or resolved in sensitivity and subgroup analyses. It probably originates from the interplay of some effect modifiers, in which our included studies slightly differ (e.g. cancer stages, study start date, and regions). These are only minor differences. From a clinical point of view, our included studies, therefore, remain largely comparable.

In addition to the studies included in this review, we are aware of a further 12 studies which may be eligible for inclusion in our review. Of these, ten studies are still awaiting assessment as no results were available, and two are still ongoing. These studies may alter our results if included in our analyses.

Despite these limitations, we were able to identify an extensive number of studies. We were able to consider the experience of almost 35,000 participants, emphasising the overall completeness and applicability of our findings.

Quality of the evidence

Risk of bias

We rated the risk of bias for each study. We took into consideration if outcomes were objective or subjective to participants and outcome assessors. Overall, nine studies showed a high risk of bias in two or more domains. The risk of bias was mostly related to blinding and reporting bias. The reasons why the risk of bias was unclear were often due to insufficient available information to make a judgement.

Certainty of the evidence

The certainty of the evidence for most outcomes was assessed as moderate. This included fracture rate, osteonecrosis of the jaw and renal impairment as they showed imprecision (mostly downgraded one point). Bone mineral density and overall survival were assessed as being at very low to low certainty. We downgraded one point for inconsistency and one to two points due to imprecision, since 95% confidence intervals were wide and/or crossed unity.

Potential biases in the review process

One review author, an information specialist experienced in medical terminology, developed the sensitive search strategy. We searched all relevant databases, trial registries, conference proceedings, and reference lists and are, therefore, confident that we have identified all relevant studies.

To minimise potential biases in the review process, we conducted the selection of studies, data extraction, risk of bias assessment and GRADE assessment in duplicate by two independent review authors and consulted a third review author in case no consensus could be reached. We collated multiple reports of the same study, so that each study rather than each report was the unit of interest in the review. However, comprehensive reporting of identified records was partially scarce, which complicated correct allocation of the reports. In case we were uncertain whether two reports belonged to the same trial, we considered them as individual trials.

We decided to combine no treatment and placebo. By doing so, we gained networks, which were more connected and were able to compare all of our included treatment options directly.

For our primary outcome, we had planned to create funnel plots for comparisons when including at least 10 studies. Unfortunately, there were not enough studies to create a funnel plot for any comparison. Instead, we created a comparison‐adjusted funnel plot, but we were not able to test for small‐study effects as the number of studies was too limited. Nevertheless, visual inspection of the funnel plot did not show evidence of small‐study effects.

For a more comprehensive presentation of results, we estimated absolute treatment effects using the actual reported event rates for our chosen main comparator (no treatment/placebo). However, if we choose another comparator to estimate absolute event rates, these effects could all change. This applies to bone mineral density, where we estimated the average T‐score from all control groups, and for overall survival, where we used the five‐year‐survival from Team IIB 2006 to calculate anticipated absolute effects. Thus, when interpreting the results of our network meta‐analysis, it must be considered that the reported absolute event rates are for illustrative purposes and do not reflect anticipated real‐life event rates.

The Summary of findings table is not ideal for presenting the results of such extensive analysis. Also, we surmise that the overall judgement of the risk of bias and the certainty in the evidence could diverge between different author teams. The risk of bias tool and GRADE approach are sensitive to subjective assessments and can be done more or less stringently.

Agreements and disagreements with other studies or reviews

To our knowledge, this is the first comprehensive review with network meta‐analysis comparing different bisphosphonates and the RANKL inhibitor, denosumab, for women with breast cancer in early or locally advanced stages.

Our review showed similar results to the systematic review and meta‐analysis by Valachis 2011, comparing zoledronic acid versus no/delayed use in primary breast cancer, for fracture rate and disease‐free survival. Valachis 2011 showed a benefit for the use of zoledronic acid for overall survival, whereas our review resulted in little to no difference. Our review showed similar results to the review evaluating bisphosphonates in women with early breast cancer (Vidal 2012) where there were significant beneficial effects on survival for bisphosphonate use compared to no use (Vidal 2012). For overall survival, however, the effects of ibandronate and pamidronate pointed in the opposite direction in our review.

A systematic review and network meta‐analysis evaluating the effect of denosumab compared to bisphosphonates, selective oestrogen receptor modulators and placebo in women with hormone‐sensitive cancer receiving endocrine therapy showed increased bone mineral density for denosumab, ibandronate and risedronate compared to placebo (Nicolopoulos 2023). In general, our review showed similar results yielding small beneficial effects for bisphosphonate use compared to no treatment/placebo.

Our review showed similar results regarding disease‐free survival, fracture rate and bone metastases to the systematic review and meta‐analysis evaluating the effects of bisphosphonates and denosumab in women with early breast cancer (O'Carrigan 2017). The review showed a survival benefit of bone agents when analyses were confined to postmenopausal women, so too did the comprehensive review including individual participant data (EBCTCG 2015).

Authors' conclusions

Implications for practice.

The findings of our systematic review and network meta‐analyses may assist clinicians and patients in making decisions regarding the use of bone‐modifying agents for women with breast cancer in early or locally advanced stages. Our results provide a comprehensive overview of different bone‐modifying agents, including a treatment ranking for each outcome. However, our analysis did not identify a clear frontrunner when considering a range of outcomes. It is important to interpret the rankings and all considered outcomes cautiously before reaching a decision. Due to missing data from the included studies, not all treatments could be compared to each other for each outcome. More trials with head‐to‐head comparisons, encompassing all potential agents, are needed to provide a complete picture and validate the results of this analysis.

When interpreting the results of this systematic review, it is crucial to understand that network meta‐analyses are not a substitute for direct head‐to‐head comparisons. Also, it is important to acknowledge that the results of this network meta‐analysis do not necessarily rule out differences which could be clinically relevant for certain individuals.

Implications for research.

Despite sufficient direct and indirect comparisons of various treatment options in the network meta‐analysis, head‐to‐head trials are required to offer clear recommendations. Future trials should consistently report all patient‐relevant outcomes. Most studies have primarily focussed on comparing a single bone‐modifying agent to either no treatment or placebo. Only one three‐arm study (SWOG S0307 2019) examined different bisphosphonates against each other. No other study compared different bisphosphonates or a bisphosphonate against the RANKL‐inhibitor denosumab. Hence, studies comparing different bisphosphonates or comparing a bisphosphonate against denosumab would be highly beneficial.

History

Protocol first published: Issue 10, 2019

Appendices

Appendix 1. Search strategies

CENTRAL (via Cochrane Library)

ID Search

#1 MeSH descriptor: [Breast Neoplasms] explode all trees

#2 breast near cancer*

#3 breast near neoplasm*

#4 breast near carcinom*

#5 breast near tumour*

#6 breast bear tumor*

#7 breast near malignan*

#8 #1 or #2 or #3 or #4 or #5 or #6 or #7

#9 MeSH descriptor: [Diphosphonates] explode all trees

#10 (diphosphonate* or diphosph*nate*)

#11 (bisphosph*nate* or biphosph*nate*)

#12 (diphosphonic* or bisphosphonic*)

#13 #9 or #10 or #11 or #12

#14 MeSH descriptor: [Alendronate] explode all trees

#15 (alendronat* or aledronic*)

#16 (fosamax* or binosto* or adronat* or alendros* or onclast*)

#17 #14 or #15 or #16

#18 MeSH descriptor: [Clodronic Acid] explode all trees

#19 (clodronic* or clodronat*)

#20 (bonefos* or clasteon* or difosfonal* or ossiten* or mebonat* or loron*)

#21 Cl2MDP

#22 #18 or #19 or #20 or #21

#23 MeSH descriptor: [Etidronic Acid] explode all trees

#24 (etidronic* or etidronat*)

#25 (didronel* or xidifon* or dicalcium or xidiphon*)

#26 (HEDP or EHDP)

#27 #23 or #24 or #25 or #26

#28 MeSH descriptor: [Technetium Tc 99m Medronate] explode all trees

#29 (medronat* or medronic*)

#30 (Technetium near/2 Tc 99m near/2 Medronat*)

#31 (Tc‐99m‐MDP or Tc‐MDP)

#32 #28 or #29 or #30 or #31

#33 (ibandronic* or ibandrovic* or ibandronat*)

#34 (bon*iva* or bondronat* or adronil*)

#35 (RPR102289A or RPR‐102289A)

#36 (BM210955 or BM‐210955)

#37 #33 or #34 or #35 or #36

#38 (pamidronat* or pamidronic* or amidronat*)

#39 MeSH descriptor: [Risedronic Acid] explode all trees

#40 (risedronic* or risedronat*)

#41 (actonel* or atelvia* or benet* or optinate*)

#42 (NE58095 or NE‐58095)

#43 #38 or #39 or #40 or #41 or #42

#44 (zoledronic* or zoledronat*)

#45 (zometa* or zomera* or aclasta* or reclast* or aredia* or orazol*)

#46 (m05BA08 or CGP‐42446$ or CGP42446$ or zol‐446 or zol446)

#47 #44 or #45 or #46

#48 (neridronat* or neridronic*)

#49 ("AHHexBP" or "6AHHDP" or "6‐AHHDP" or nerixia)

#50 #48 or #49

#51 (tiludronat* or tiludronic*)

#52 (skelid* or tildren* or sr 42329 or sr42329 or sr 41319b or sr41319b)

#53 #51 or #52

#54 (incadronat* or incadronic*)

#55 (cimadronat* or cimadronic*)

#56 (bisphonal* or YM175 or YM 175)

#57 #54 or #55 or #56

#58 MeSH descriptor: [RANK Ligand] explode all trees

#59 (rank* near/3 ligand*)

#60 RANK ligand inhibitor*

#61 (protein* near/2 RANKL) or (protein* near/2 TRANCE)

#62 (osteoclast* near/2 differentiation factor*)

#63 (osteoclast* near/2 ligand*)

#64 Tumor Necrosis Factor‐Related Activation‐Induced Cytokin*

#65 #58 or #59 or #60 or #61 or #62 or #63 or #64

#66 MeSH descriptor: [Receptor Activator of Nuclear Factor‐kappa B] explode all trees

#67 ((receptor activator* near/3 nf‐kappab) or (receptor activator* near/3 nuclear factor kappab))

#68 ((receptor activator* near/3 nf‐kappa) or (receptor activator* near/3 nuclear factor kappa))

#69 tnfrsf11a

#70 (trance r or trance receptor*)

#71 #66 or #67 or #68 or #69 or #70

#72 MeSH descriptor: [Denosumab] explode all trees

#73 denosumab*

#74 (xgeva* or prolia*)

#75 (AMG162 or AMG‐162)

#76 #72 or #73 or #74 or #75

#77 romosozumab*

#78 (AMG 785 or AMG785)

#79 (cdp 7851 or cdp7851)

#80 evenity*

#81 #77 or #78 or #79 or #80

#82 blosozumab*

#83 (Ly2541546 or Ly 2541546)

#84 #82 or #83

#85 #13 or #17 or #22 or #27 or #32 or #37 #43 or #47 or #50 or #53 or #57 or #65 or #71 or #76 or #81 or #84

#86 #8 and #85

MEDLINE (via Ovid)

# Searches

1 exp BREAST NEOPLASMS/

2 (breast adj6 cancer*).tw.

3 (breast adj6 neoplasm*).tw.

4 (breast adj6 carcinoma*).tw.

5 (breast adj6 tumo?r*).tw.

6 or/1‐5

7 exp DIPHOSPHONATES/

8 (disphosphonate* or diphosph#nate*).tw,kf,ot,nm.

9 (bisphosph#nate* or biphosph#nate*).tw,kf,ot,nm.

10 (diphosphonic* or bisphosphonic*).tw,kw,ot,nm.

11 or/7‐10

12 ALENDRONATE/

13 (alendronat* or aledronic*).tw,kf,ot,nm.

14 (fosamax* or binosto* or adronat* or alendros* or onclast*).tw,kf,ot,nm.

15 or/12‐14

16 CLODRONIC ACID/

17 (clodronic* or clodronat*).tw,kf,ot,nm.

18 (bonefos* or clasteon* or difosfonal* or ossiten* or mebonat* or loron*).tw,kf,ot,nm.

19 Cl2MDP.tw,kf,ot,nm.

20 or/16‐19

21 ETIDRONIC ACID/

22 (etidronic* or etidronat*).tw,kf,ot,nm.

23 (didronel* or xidifon* or dicalcium or xidiphon*).tw,kf,ot.

24 (HEDP or EHDP).tw,kf,ot.

25 or/21‐24

26 TECHNETIUM TC 99M MEDRONATE/

27 (medronat* or medronic*).tw,kf,ot,nm.

28 (Technetium adj2 Tc 99m adj2 Medronat*).tw,kf,ot,nm.

29 or/26‐28

30 IBANDRONIC ACID/

31 (ibandronic* or ibandrovic* or ibandronat*).tw,kf,ot,nm.

32 (bon?iva* or bondronat* or adronil*).tw,kf,ot,nm.

33 (RPR102289A or RPR‐102289A).tw,kf,ot,nm.

34 (BM210955 or BM‐210955).tw,kf,ot,nm.

35 or/30‐34

36 PAMIDRONATE/

37 (pamidronat* or pamidronic* or amidronat*).tw,kf,ot,nm.

38 or/36‐37

39 RISEDRONIC ACID/

40 (risedronic* or risedronat*).tw,kf,ot,nm.

41 (actonel* or atelvia* or benet* or optinate*).tw,kf,ot,nm.

42 (NE58095 or NE‐58095).tw,kf,ot,nm.

43 or/39‐42

44 ZOLEDRONIC ACID/

45 (zoledronic* or zoledronat*).tw,kf,ot,nm.

46 (zometa* or zomera* or aclasta* or reclast* or aredia* or orazol*).tw,kf,ot,nm.

47 (m05BA08 or CGP‐42446$ or CGP42446$ or zol‐446 or zol446).tw,kf,ot,nm.

48 or/44‐47

49 (neridronat* or neridronic*).tw,kf,ot,nm.

50 (AHHexBP or 6AHHDP or 6‐AHHDP or nerixia).tw,kf,ot,nm.

51 or/49‐50

52 (tiludronat* or tiludronic*).tw,kf,ot,nm.

53 (skelid* or tildren* or sr 42329 or sr42329 or sr 41319b or sr41319b).tw,kf,ot,nm.

54 or/52‐53

55 (incadronat* or incadronic*).tw,kf,ot,nm.

56 (cimadronat* or cimadronic*).tw,kf,ot,nm.

57 (bisphonal* or YM175 or YM 175).tw,kf,ot,nm.

58 or/55‐57

59 (olpadronat* or olpadronic*).tw,kf,ot,nm.

60 (ig 8801 or ig8801).tw,kf,ot,nm.

61 or/59‐60

62 RANK LIGAND/

63 (rank* adj3 ligand*).tw,kf,ot,nm.

64 ((protein* adj2 RANKL) or (protein* adj2 TRANCE)).tw,kf,ot,nm.

65 (osteoclast* adj2 differentiation factor*).tw,kf,ot,nm.

66 (osteoclast* adj2 ligand*).tw,kf,ot,nm.

67 tumor necrosis factor related activation induced cytokine.tw,kf,ot,nm.

68 or/62‐67

69 RECEPTOR ACTIVATOR OF NUCLEAR FACTOR‐KAPPA B/

70 ((receptor activator* adj3 nf‐kappab) or (receptor activator* adj3 nuclear factor kappab)).tw,kf,ot,nm.

71 ((receptor activator* adj3 nf‐kappa) or (receptor activator* adj3 nuclear factor kappa)).tw,kf,ot,nm.

72 tnfrsf11a.tw.

73 (trance r or trance receptor*).tw,kf,ot,nm.

74 or/69‐73

75 DENOSUMAB/

76 denosumab*.tw,kf,ot,nm.

77 (xgeva* or prolia*).tw,kf,ot,nm.

78 (AMG162 or AMG‐162).tw,kf,ot,nm.

79 or/75‐78

80 romosozumab*.tw,kf,ot,nm.

81 (cdp 7851 or cdp7851).tw,kf,ot,nm.

82 (AMG 785 or AMG785).tw,kf,ot,nm.

83 evenity*.tw,kf,ot,nm.

84 or/80‐83

85 blosozumab*.tw,kf,ot,nm.

86 (Ly2541546 or Ly 2541546).tw,kf,ot,nm.

87 or/85‐86

88 15 or 20 or 25 or 29 or 35 or 38 or 43 or 48 or 51 or 54 or 58 or 61 or 68 or 74 or 79 or 84 or 87

89 randomized controlled trial.pt.

90 controlled clinical trial.pt.

91 randomi?ed.ab.

92 placebo.ab.

93 drug therapy.fs.

94 randomly.ab.

95 trial.ab.

96 groups.ab.

97 or/89‐96

98 CLINICAL TRIAL, PHASE III/

99 ("Phase 3" or "phase3" or "phase III" or P3 or "PIII").ti,ab,kw.

100 98 or 99

101 exp ANIMALS/ not HUMANS/

102 (97 or 100) not 101

103 6 and 88 and 102

Embase (via Ovid)

# Searches

1 exp BREAST/

2 exp BREAST DISEASE/

3 (1 or 2) and exp NEOPLASM/

4 exp BREAST TUMOR/

5 exp BREAST CANCER/

6 exp BREAST CARCINOMA/

7 (breast* adj5 (neoplas* or cancer* or carcin* or tumo* or metasta* or malig*)).ti,ab.

8 or/3‐7

9 exp BISPHOSPHONIC ACID DERIVATIVE/

10 (diphosphonate* or diphosph#nate*).tw,kw,ot.

11 (bisphosph#nate* or biphosph#nate*).tw,kw,ot.

12 (diphosphonic* or bisphosphonic*).tw,kw,ot.

13 or/9‐12

14 ALENDRONIC ACID/

15 (alendronat* or aledronic*).tw,kw,ot.

16 (fosamax* or binosto* or adronat* or alendros* or onclast* or alend*).tw,kw.

17 or/14‐16

18 CLODRONIC ACID/

19 (clodronic* or clodronat*).tw,kw.

20 (bonefos* or clasteon* or difosfonal* or ossiten* or mebonat* or loron*).tw,kw.

21 Cl2MDP.tw,kw.

22 or/18‐21

23 ETIDRONIC ACID/

24 (etidronic* or etidronat*).tw.

25 (didronel* or xidifon* or dicalcium or xidiphon*).tw,kw.

26 (HEDP or EHDP).tw,kw.

27 or/23‐26

28 MEDRONATE TECHNETIUM TC 99M/

29 (medronat* or medronic*).tw,kw.

30 (Technetium adj2 Tc 99m adj2 Medronat*).tw,kw.

31 (Tc 99m MDP or Tc MDP).tw,kw.

32 or/28‐31

33 IBANDRONIC ACID/

34 (ibandronic* or ibandrovic* or ibandronat*).tw,kw.

35 (bon?iva* or bondronat* or adronil*).tw,kw.

36 (RPR102289A or RPR‐102289A).tw,kw.

37 (BM210955 or BM‐210955).tw,kw.

38 or/33‐37

39 PAMIDRONIC ACID/

40 (pamidro* or pamidronic* or amidronat*).tw,kw.

41 or/39‐40

42 RISEDRONIC ACID/

43 (risedronic* or risedronat*).tw,kw.

44 (actonel* or atelvia* or benet* or optinate*).tw,kw.

45 (NE58095 or NE‐58095).tw,kw.

46 (cgp 23339 or cgp23339 or cgp 23339a or cgp 23339a).tw,kw.

47 or/42‐46

48 ZOLEDRONIC ACID/

49 (zoledronic* or zoledronat*).tw,kw.

50 (zometa* or zomera* or aclasta* or reclast* or aredia* or orazol*).tw,kw.

51 (m05BA08 or CGP‐42446 or CGP42446 or CGP‐42446a or CGP42446a or zol‐446 or zol446).tw,kw.

52 or/48‐51

53 NERIDRONIC ACID/

54 (neridronat* or neridronic*).tw,kw.

55 (AHHexBP or 6AHHDP or 6‐AHHDP or nerixia).tw,kw.

56 or/53‐55

57 TILUDRONIC ACID/

58 (tiludronat* or tiludronic*).tw,kw.

59 (skelid* or tildren* or sr 42329 or sr42329 or sr 41319b or sr41319b).tw,kw.

60 or/57‐59

61 INCADRONIC ACID/

62 (incadronat* or incadronic*).tw,kw.

63 (cimadronat* or cimadronic*).tw,kw.

64 (bisphonal* or YM 175 or YM175).tw,kw.

65 or/61‐64

66 OLPADRONIC ACID/

67 (olpadronat* or olpadronic*).tw,kw.

68 (ig 8801 or ig8801).tw,kw.

69 or/66‐68

70 OSTEOCLAST DIFFERENTIATION FACTOR/

71 (osteoclast* adj2 differentiation factor*).tw,kw.

72 (osteoclast* adj2 ligand*).tw,kw.

73 (rank* adj3 ligand*).tw,kw.

74 ((protein* adj2 RANKL) or (protein* adj2 TRANCE)).tw,kw.

75 tumor necrosis factor related activation induced cytokine.tw,kw.

76 or/70‐75

77 "RECEPTOR ACTIVATOR OF NUCLEAR FACTOR KAPPA B"/

78 ((receptor activator* adj3 nf‐kappab) or (receptor activator* adj3 nuclear factor kappab)).tw,kw.

79 ((receptor activator* adj3 nf‐kappa) or (receptor activator* adj3 nuclear factor kappa)).tw,kw.

80 tnfrsf11a.tw.

81 (trance r or trance receptor*).tw,kw.

82 or/77‐81

83 DENOSUMAB/

84 denosumab*.tw,kw.

85 (xgeva* or prolia*).tw,kw.

86 (AMG162 or AMG‐162).tw,kw.

87 or/83‐86

88 ROMOSOZUMAB/

89 (cdp 7851 or cdp7851).tw,kw.

90 (AMG 785 or AMG785).tw,kw.

91 evenity*.tw,kw.

92 or/88‐91

93 BLOSOZUMAB/

94 blosozumab*.tw,kw.

95 (Ly2541546 or Ly 2541546).tw,kw.

96 93 or 94 or 95

97 13 or 17 or 22 or 27 or 32 or 38 or 41 or 47 or 52 or 56 or 60 or 65 or 69 or 76 or 82 or 87 or 92 or 96

98 8 and 97

99 RANDOMIZED CONTROLLED TRIAL/

100 CONTROLLED CLINICAL STUDY/

101 Random*.ti,ab.

102 RANDOMIZATION/

103 INTERMETHOD COMPARISON/

104 placebo.ti,ab.

105 (compare or compared or comparison).ti.

106 (open adj label).ti,ab.

107 ((double or single or doubly or singly) adj (blind or blinded or blindly)).ti,ab.

108 DOUBLE BLIND PROCEDURE/

109 parallel group$1.ti,ab.

110 (crossover or cross over).ti,ab.

111 ((assign$ or match or matched or allocation) adj5 (alternate or group$1 or intervention$1 or patient$1 or subject$1 or participant$1)).ti,ab.

112 (controlled adj7 (study or design or trial)).ti,ab.

113 (volunteer or volunteers).ti,ab.

114 trial.ti.

115 or/99‐114

116 8 and 97 and 115

ClinicalTrials.gov

expert search: (breast cancer OR breast neoplasm* OR breast carcinoma*) AND (bisphosphonates OR diphosphonates OR zoledronate OR "zoledronic acid" OR clodronate OR "clodronic acid" OR "etidronic acid" OR etidronate OR medronate OR medronic OR alendronate OR aledronic OR ibandronate OR ibandronic OR pamidronate OR pamidronic OR risedronate OR risedronic OR tiludronate OR tiludronic OR incadronate OR incadronic OR olpadronate OR olpadronic OR neridronate OR neridronic OR rank OR osteoclast OR denosumab OR romosozumab OR blosozumab)

WHO ICTRP

Three searches with following deduplication

#1

"breast cancer" AND bisphosphonates OR "breast cancer" AND diphosphonates OR "breast cancer" AND zoledronate OR "breast cancer" AND "zoledronic acid" OR "breast cancer" AND clodronate OR "breast cancer" AND "clodronic acid" OR breast cancer AND "etidronic acid" OR breast cancer AND etidronate OR breast cancer AND medronate OR "breast cancer" AND medronic OR "breast cancer" AND alendronate OR "breast cancer" AND aledronic OR "breast cancer" AND ibandronate OR "breast cancer" AND ibandronic OR "breast cancer" AND pamidronate OR "breast cancer" AND pamidronic OR "breast cancer" AND risedronate OR "breast cancer" AND risedronic OR "breast cancer" AND tiludronate OR "breast cancer" AND tiludronic OR "breast cancer" AND incadronate OR "breast cancer" AND incadronic OR "breast cancer" AND olpadronate OR "breast cancer" AND olpadronic OR "breast cancer" AND neridronate OR "breast cancer" AND neridronic OR "breast cancer" AND rank OR "breast cancer" AND osteoclast OR "breast cancer" AND denosumab OR "breast cancer" AND romosozumab OR "breast cancer "AND blosozumab

#2

"breast neoplasm" AND bisphosphonates OR "breast neoplasm" AND diphosphonates OR "breast neoplasm" AND zoledronate OR "breast neoplasm" AND "zoledronic acid" OR "breast neoplasm" AND clodronate OR "breast neoplasm" AND "clodronic acid" OR "breast neoplasm" AND "etidronic acid" OR "breast neoplasm" AND etidronate OR "breast neoplasm" AND medronate OR "breast neoplasm" AND medronic OR "breast neoplasm" AND alendronate OR "breast neoplasm" AND aledronic OR "breast neoplasm" AND ibandronate OR "breast neoplasm" AND ibandronic OR "breast neoplasm" AND pamidronate OR "breast neoplasm" AND pamidronic OR "breast neoplasm" AND risedronate OR "breast neoplasm" AND risedronic OR "breast neoplasm" AND tiludronate OR "breast neoplasm" AND tiludronic OR "breast neoplasm" AND incadronate OR "breast neoplasm" AND incadronic OR "breast neoplasm" AND olpadronate OR "breast neoplasm" AND olpadronic OR "breast neoplasm" AND neridronate OR "breast neoplasm" AND neridronic OR "breast neoplasm" AND rank OR "breast neoplasm" AND osteoclast OR "breast neoplasm" AND denosumab OR "breast neoplasm" AND romosozumab OR "breast neoplasm" AND blosozumab

#3

"breast carcinoma" AND bisphosphonates OR "breast carcinoma" AND diphosphonates OR "breast carcinoma" AND zoledronate OR "breast carcinoma" AND "zoledronic acid" OR "breast carcinoma" AND clodronate OR "breast carcinoma" AND "clodronic acid" OR "breast carcinoma" AND "etidronic acid" OR "breast carcinoma" AND etidronate OR "breast carcinoma" AND medronate OR "breast carcinoma" AND medronic OR "breast carcinoma" AND alendronate OR "breast carcinoma" AND aledronic OR "breast carcinoma" AND ibandronate OR "breast carcinoma" AND ibandronic OR "breast carcinoma" AND pamidronate OR "breast carcinoma" AND pamidronic OR "breast carcinoma" AND risedronate OR "breast carcinoma" AND risedronic OR "breast carcinoma" AND tiludronate OR "breast carcinoma" AND tiludronic OR "breast carcinoma" AND incadronate OR "breast carcinoma" AND incadronic OR "breast carcinoma" AND olpadronate OR "breast carcinoma" AND olpadronic OR "breast carcinoma" AND neridronate OR "breast carcinoma" AND neridronic OR "breast carcinoma" AND rank OR "breast carcinoma" AND osteoclast OR "breast carcinoma" AND denosumab OR "breast carcinoma" AND romosozumab OR "breast cancer" AND blosozumab

Appendix 2. Study characteristics per pairwise comparison

Alendronate vs. No treatment/Placebo

Study

N

Year

Length of follow‐up

Stage of disease

Menopausal status

Endocrine therapy

Type of endocrine therapy

Hormon receptor status

Human epidermal growth factor receptor 2

Duration of bone‐modifying intervention

Mean Age N1

Mean Age N2

Cohen 2008

11

2008

‐

‐

only post

‐

tamoxifen

‐

‐

1 year

‐

‐

Clodronate vs. No treatment/Placebo

Study

N

Year

Length of follow‐up

Stage of disease

Menopausal status

Endocrine therapy

Type of endocrine therapy

Hormon receptor status

Human epidermal growth factor receptor 2

Duration of bone‐modifying intervention

Mean Age N1

Mean Age N2

Mardiak 2000

73

2000

84 monthsa

mainly stage III

‐

some pts ET

‐

‐

‐

2‐5 years

55.0a

54.0a

NSABP‐34 2012

3323

2012

90.8 monthsa

stage I‐III

‐

some pts ET

mostly tamoxifen

both

‐

2‐5 years

‐

‐

Powles 2006

1069

2006

5.6 yearsa

mainly stage I‐II

pre and post

some pts ET

tamoxifen

both

‐

2‐5 years

52.8

52.7

Saarto 2004

299

2004

10 years

mainly T1‐T3

pre and post

some pts ET

tamoxifen and toremifene

both

‐

‐

52.0

52.0

amedian was reported instead

Denosumab vs. No treatment/Placebo

Study

N

Year

Length of follow‐up

Stage of disease

Menopausal status

Endocrine therapy

Type of endocrine therapy

Hormon receptor status

Human epidermal growth factor receptor 2

Duration of bone‐modifying intervention

Mean Age N1

Mean Age N2

ABCSG‐18 2019

3425

2019

66 months

early stage

only post

some pts ET

‐

HR+

both

2‐5 years

64.3

64.0

D‐CARE 2013

4509

2013

5 years

stage II‐III

pre and post

some pts ET

‐

both

both

2‐5 years

50.0a

51.0a

Ellis 2008

252

2008

4 years

early stage

pre and post

all pts ET

‐

HR+

‐

2‐5 years

59.2

59.7

GeparX 2016

780

2016

‐

T1‐T4

pre and post

‐

‐

HR+

both

< 1 year

49.0a

48.5a

amedian was reported instead

Ibandronate vs. No treatment/Placebo

Study

N

Year

Length of follow‐up

Stage of disease

Menopausal status

Endocrine therapy

Type of endocrine therapy

Hormon receptor status

Human epidermal growth factor receptor 2

Duration of bone‐modifying intervention

Mean Age N1

Mean Age N2

ARIBON 2012

50

2012

5 years

early stage

only post

all pts ET

aromatase inhibitor

HR+

‐

2‐5 years

67.8

67.5

BONADIUV 2019

171

2019

5 years

early stage

only post

all pts ET

aromatase inhibitor

HR+

both

2‐5 years

60.5

59.6

GAIN 2013

2994

2013

38.7 monthsa

early stage

pre and post

some pts ET

OFS with aromatase inhibitor

both

both

2‐5 years

49.0a

50.0a

TEAM IIB 2006

1116

2006

4.6 yearsa

stage I‐III

only post

all pts ET

tamoxifen followed by aromatase inhibitor

HR+

‐

2‐5 years

‐

‐

amedian was reported instead

Pamironate vs. No treatment/Placebo

Study

N

Year

Length of follow‐up

Stage of disease

Menopausal status

Endocrine therapy

Type of endocrine therapy

Hormon receptor status

Human epidermal growth factor receptor 2

Duration of bone‐modifying intervention

Mean Age N1

Mean Age N2

Kristensen 2008

953

2008

10 years

stage I‐III

pre and post

no pts ET

‐

both

‐

2‐5 years

‐

‐

Risedronate vs. No treatment/Placebo

Study

N

Year

Length of follow‐up

Stage of disease

Menopausal status

Endocrine therapy

Type of endocrine therapy

Hormon receptor status

Human epidermal growth factor receptor 2

Duration of bone‐modifying intervention

Mean Age N1

Mean Age N2

Monda 2017

84

2017

2 years

‐

only post

all pts ET

aromatase inhibitor

HR+

‐

2‐5 years

55.7

56.1

REBBeCA 2008

87

2008

2 years

mainly stage I‐II

only post

some pts ET

‐

‐

‐

2‐5 years

50.1

49.0

SABRE 2010

154

2010

2 years

‐

only post

all pts ET

aromatase inhibitor

HR+

‐

2‐5 years

63.8

64.8

Zoledronic acid vs. No treatment/Placebo

Study

N

Year

Length of follow‐up

Stage of disease

Menopausal status

Endocrine therapy

Type of endocrine therapy

Hormon receptor status

Human epidermal growth factor receptor 2

Duration of bone‐modifying intervention

Mean Age N1

Mean Age N2

ABCSG‐12 2011

1803

2011

62 monthsa

early stage

only pre

all pts ET

OFS with tamoxifen

both

‐

2‐5 years

‐

‐

Aft 2012

120

2012

61.9

stage II‐III

pre and post

some pts ET

‐

both

both

1 year

49.0a

47.0a

AZURE 2018

3360

2018

10 years

stage II‐III

pre and post

some pts ET

‐

both

both

2‐5 years

47.0a

48.0a

EXPAND 2011

81

2011

3 years

‐

only post

all pts ET

aromatase inhibitor

HR+

‐

2‐5 years

58.4

61.3

FEMZONE 2014

168

2014

5 years

T1‐T4

only post

all pts ET

aromatase inhibitor

both

‐

< 1 year

71.1

70.6

Hershman 2008

114

2008

1 year

mainly stage I‐II

only pre

‐

‐

‐

‐

1 year

43.0

42.0

HOBOE 2019

1065

2019

64 monthsa

early stage

only pre

‐

‐

‐

‐

‐

45.2a

44.9a

JONIE 2017

180

2017

3 years

stage IIA‐IIB

pre and post

some pts ET

‐

both

HER2‐

2‐5 years

49.5

49.0

NATAN 2016

693

2016

54.7 monthsa

early stage

pre and post

some pts ET

‐

both

both

2‐5 years

‐

‐

NEOZOL 2018

53

2018

5.7 monthsa

stage II‐III

pre and post

‐

‐

both

HER2‐

< 1 year

51.2

50.5

NEO‐ZOTAC BOOG 2010

250

2010

60 months

stage II‐III

pre and post

no pts ET

‐

both

HER2‐

< 1 year

49.5

48.9

Novartis I 2006

83

2006

3 years

‐

only post

all pts ET

‐

HR+

‐

2‐5 years

58.4

61.3

ProBONE II 2016

70

2016

60 months

stage I‐III

only pre

some pts ET

‐

HR+

‐

2‐5 years

43.2

42.8

Safra 2011

90

2011

5 years

stage I‐III

only post

all pts ET

‐

HR+

‐

2‐5 years

59.0

61.1

Solomayer 2012

96

2012

88 monthsa

mainly stage I‐II

pre and post

some pts ET

‐

both

both

2‐5 years

54.0a

54.0a

Sun 2016

120

2016

1 year

stage I‐IIIA

only post

all pts ET

aromatase inhibitor

HR+

‐

1 year

58.0a

56.0a

amedian was reported instead

Zoledronic acid vs. Clodronate vs. Ibandronate

Study

N

Year

Length of follow‐up

Stage of disease

Menopausal status

Endocrine therapy

Type of endocrine therapy

Hormon receptor status

Human epidermal growth factor receptor 2

Duration of bone‐modifying intervention

Mean Age N1

Mean Age N2

Mean Age N3

SWOG S0307 2019

6097

2019

10 years

stage I‐III

‐

some pts ET

‐

‐

‐

2‐5 years

53a

52.6a

52.7a

amedian was reported instead

Characteristics of studies

Characteristics of included studies [ordered by study ID]

ABCSG‐12 2011.

Study characteristics
Methods	Setting: phase III, open‐label, multicentre, randomised Recruitment period: June 1999‐May 2006 Length of study: 3 years Length of follow‐up: median 62 months
Participants	Eligibility criteria: Premenopausal, hormone receptor‐positive patient; Histologically verified (minimally) invasive breast cancer, local radical treatment; 0‐9 involved axillary lymph nodes (≥ 10 histologically examined nodes); Tumour stage: pT1b‐3, yT0 or yT1a. Exclusion criteria: T1a, T4d, yT4; M1; Previous breast tumour irradiation; Previous or concurrent chemotherapy (except for preoperative chemotherapy); Serum creatinine > 1.5 x UNL or creatinine clearance < 60 mL/min. Stage of disease: stage I/II ≤ 10 axillary lymph nodes > stage II (21.7% zoledronic acid, 21.2% no zoledronic acid) node‐positive (30.4% zoledronic acid, 30.4% no zoledronic acid) TNM staging system: T1: zoledronic acid with tamoxifen: 75% zoledronic acid with anastrozole: 76% tamoxifen alone: 76% anastrozole alone: 78% T2: zoledronic acid with tamoxifen: 22% zoledronic acid with anastrozole: 22% tamoxifen alone: 22% anastrozole alone: 21% N0 zoledronic acid with tamoxifen: 66% zoledronic acid with anastrozole: 68% tamoxifen alone: 68% anastrozole alone: 67% node‐positive zoledronic acid with tamoxifen: 31% zoledronic acid with anastrozole: 30% tamoxifen alone: 30% anastrozole alone: 31% Mean age: intervention 1: < 40 years: 18%, > 40 years: 82% intervention 2: < 40 years: 19%, > 40 years: 81% Menopausal status: 100% premenopausal RANKL status: NR Hormone receptor status: 93% (intervention 1) vs. 94% (intervention 2) HR+ Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 1803 Country of participants: NR
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: zoledronic acid (4 mg, intravenously, every 6 months, 3 years); intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: SERM (tamoxifen), aromatase inhibitor (anastrozole), ovarian function suppression (goselerin, monthly), preoperative chemotherapy was allowed but no patients received adjuvant chemotherapy, postoperative radiotherapy was administered according to guidelines from local institutions)
Outcomes	disease‐free survival; relapse‐free survival; overall survival; bone mineral density; bone metastasis‐free survival; safety.
Notes	Funding sources: AstraZeneca, Novartis Conflicts of interest: MG has received research support from and has served as a consultant for AstraZeneca, Novartis, and Pfizer, and has received lecture fees and honoraria for participation on advisory boards from AstraZeneca, Novartis, Sanofi‐Aventis, Roche, Schering, Amgen, and Pfizer. GL‐E has received lecture fees from AstraZeneca and Novartis. RJ has served as a consultant for and received honoraria for participation on advisory boards from AstraZeneca, Roche, and Sanofi‐Aventis, and has received lecture fees from AstraZeneca, Roche, and Sanofi‐Aventis. MS has received lecture fees from AstraZeneca and Novartis. GP has received travel grants and lecture fees from AstraZeneca, Novartis, Roche, and GlaxoSmithKline. HE has received honoraria for participation on advisory boards and lecture fees from AstraZeneca and Novartis. WE has received consultancy fees and travel support from Novartis and AstraZeneca and lecture fees from Novartis, Sanofi‐Aventis, and Roche. GS has received lecture fees from AstraZeneca, Novartis, Roche, and Amgen. PD has received consultancy fees from Novartis and Genomic Health, lecture fees and payment for development of educational presentations from Novartis and Pfizer, and travel expenses from AstraZeneca, Novartis, Roche, and Pfizer. GH has received travel expenses from Novartis. RG has served as a consultant for and received honoraria for participation on advisory boards from Novartis and AstraZeneca. All other authors declare that they have no conflicts of interest.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Computer‐generated adaptive randomisation method"
Allocation concealment (selection bias)	Low risk	"Assign treatment groups via an automated telephone service"
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective bias
Incomplete outcome data (attrition bias) All outcomes	Low risk	Intention‐to‐treat analysis (ITT) and no missing outcome data
Selective reporting (reporting bias)	Low risk	All predefined outcomes reported
Other bias	Low risk	Study appeared to be free of other sources of bias.

ABCSG‐18 2019.

Study characteristics
Methods	Setting: multicentre, phase III, prospective, randomised, double‐blind, placebo‐controlled, parallel assignment Recruitment period: 12/2006 to 07/2013 Length of study: 5 years (planned) Length of follow‐up: 66 months (planned)
Participants	Eligibility criteria: histologically or cytologically confirmed adenocarcinoma of the breast; female subjects with non‐metastatic disease who are oestrogen receptor (ER) and/or progesterone receptor (PR)‐positive, and who have completed their treatment pathway (surgery, chemotherapy); subjects who are currently on, or will initiate an approved nonsteroidal aromatase inhibitor therapy (e.g. anastrazole) in the adjuvant setting; postmenopausal woman defined as: age ≥ 60 years, having undergone a bilateral oophorectomy; age < 60 years meeting the following requirements: FSH and oestradiol in the postmenopausal range, and having a negative pregnancy test within 7 days prior to randomisation with the exception of women who have undergone a hysterectomy; ECOG performance score of 0 or 1; before any study‐specific procedure is performed, a signed and dated written informed consent must be obtained. Exclusion criteria: aromatase inhibitor therapy for more than 24 months; prior or concurrent treatment with Selective Oestrogen Receptor Modulators (SERMs) (e.g. tamoxifen); evidence of metastatic disease; current or prior to IV bisphosphonate administration. Oral bisphosphonate treatment if taken 3 or more years continuously, or if taken longer than 3 months but less than 3 years unless subject has had a washout period of at least 1 year prior to randomisation, or any use during the 3‐month period prior to randomisation. Prior administration of denosumab; known liver or renal disease as determined by the investigator and indicated by the following criteria: AST ≥ 2·5 x ULN, ALT ≥ 2·5 x ULN, serum creatinine ≥ 2 x ULN; recurrence of the primary malignancy (e.g. during the allowed interval of pre‐treatment with AI); diagnosis of any second non‐breast malignancy within the last 5 years, except for adequately treated basal cell or squamous cell skin cancer, or for in situ carcinoma of the cervix uteri; known history of: Paget’s disease (bone), Cushing’s disease, hyperprolactinemia or other active metabolic bone disease; hypercalcaemia or hypocalcaemia. Stage of disease: NR TNM staging system: 27% (denosumab) vs. 29.6% (placebo) node positive Mean age: intervention 1: 64.0 ± 7.9 intervention 2: 64.6 ± 8.0 Menopausal status:100% postmenopausal RANKL status: NR Hormone receptor status: ER+: 98.8% denosumab, 99.1% placebo PR+: 83.2% denosumab, 84.8% placebo Human epidermal growth factor receptor 2 status: 6% (denosumab) vs. 6.6% (placebo), HER2 positive Participants randomised: total: 3425 Country of participants: Austria, Sweden
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: denosumab (60 mg, subcutaneously, every 6 months, 5 years); intervention 2: placebo (subcutaneously, every 6 months, 5 years); supplemental: Vitamin D (at least 400 IU, daily, over the complete study period), Calcium (500 mg, daily, over the complete study period); other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy as indicated, nonsteroidal aromatase inhibitors (anastrozole, letrozole) started with or before treatment, adjuvant and neoadjuvant chemotherapy, radiotherapy possible before randomisation
Outcomes	time to first clinical fracture; per cent change in BMD (lumbar spine, total hip, femoral neck); new vertebral fractures; new or worsening vertebral fractures; DFS; bone metastases‐free survival (BMFS); OS.
Notes	Funding sources: Amgen was the legal sponsor of the study and had a role in protocol and study design, and reviewed the manuscript, but was not involved in data collection, data interpretation, or writing of the manuscript. CF and SF had access to the raw data. The corresponding author had full access to all of the data and the final responsibility to submit for publication. Conflicts of interest: MG reports personal fees and non‐financial support from Amgen, Celgene, and Eli Lilly; grants, personal fees, and non‐financial support from AstraZeneca and Novartis; personal fees from NanoString Technology; grants and personal fees from Roche; other support from Accelsoir; grants and non‐financial support from Pfizer; and non‐financial support from Ipsen, all outside of the submitted work. GP reports personal fees from Novartis, Amgen, Pfizer, AstraZeneca, Accord, Bondimed, Roche, and Eli Lilly outside of the submitted work. GGS reports personal fees from Amgen, Celgene, Eli Lilly, AstraZeneca, Pfizer, and Novartis, and grants and personal fees from Roche outside of the submitted work. DE reports personal fees and non‐financial support from Roche, Pfizer, and Novartis, and non‐financial support from Amgen outside of the submitted work. RG reports grants, personal fees, and non‐financial support from Amgen outside of the submitted work. FF reports other support from Springer, and grants and non‐financial support from Comesa, Bondimed, Novartis, Roche, AstraZeneca, and Pfizer outside of the submitted work. MB reports grants and personal fees from Amgen; grants, personal fees, and non‐financial support from Celgene; personal fees and non‐financial support from Roche and Pfizer; and personal fees from Novartis and Pierre Fabre outside of the submitted work. FH reports personal fees and non‐financial support from Roche and Celgene; and personal fees from Eli Lilly, Pfizer, Novartis, and Amgen outside of the submitted work. VB‐R reports personal fees from Eli Lilly and grants from Roche outside of the submitted work. PS reports grants and personal fees from Amgen, and personal fees from Roche and Pfizer outside of the submitted work. BM reports personal fees from Roche and Celgene outside of the submitted work. CF and SF report receiving research funding from Amgen during the conduct of the study, awarded to their institution. CFS reports grants, personal fees, and non‐financial support from Amgen and AstraZeneca; grants and personal fees from Novartis and Roche; and other support from Tesaro and Pfizer outside of the submitted work. VW, EM‐Z, RJ, CM, and RE declare no competing interests.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomly permuted block design with block sizes 2 and 4
Allocation concealment (selection bias)	Low risk	Assigned by an interactive voice response system
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Low number lost to follow‐up; intention‐to‐treat analysis
Selective reporting (reporting bias)	Low risk	All prespecified endpoints were addressed.
Other bias	Low risk	Study appeared to be free of other sources of bias.

Aft 2012.

Study characteristics
Methods	Setting: single‐centre, phase II Recruitment period: March 2003 to May 2006 Length of study: 1 year Length of follow‐up: 61.9 months
Participants	Eligibility criteria: • stage II‐III (> T2 and/or N1) newly diagnosed BC; • ECOG 0‐1, with no evidence of distant metastases; • normal cardiac, renal and liver function. Exclusion criteria: • evidence of distant metastasis by CT scan of the chest, abdomen, pelvis or bone scan; • prior malignancies, serious functional disorders of the heart, liver, or kidneys; • pregnancy, women below 18 years of age. Stage of disease: stage II‐III (≥ T2 and/or ≥ N1) ductal carcinoma (47% intervention 1, 49% intervention 2) lobular carcinoma (7% intervention 1, 7% intervention 2) mean tumour size: 3.81 cm (intervention 1), 3.56 cm (intervention 2) TNM staging system: ≥ T2 and/or ≥ N1 Mean age: intervention 1: median of 49 (range: 29‐67) intervention 2: median of 47 (range: 31‐68) Menopausal status: premenopausal: 51.7 % (intervention 1), 55.9 % (intervention 2) postmenopausal: 48.3 % (intervention 1), 44.1 % (intervention 2) RANKL status: NR Hormone receptor status: • ER+: 53.3% intervention 1, 59.3% intervention 2 • PR*: 40% intervention 1, 52.5% intervention 2 Human epidermal growth factor receptor 2 status: 21.6% intervention 1, 16.9% intervention 2 Participants randomised: total: 120 Country of participants: USA
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 3 weeks for 1 year; intervention 2: no treatment; supplemental: 800 IU of vitamin D and 1000 mg of calcium daily; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: All women received four cycles of intravenous neoadjuvant epirubicin (75 mg/m²) plus docetaxel (75 mg/m²) every 3 weeks, with granulocyte stimulating factor support and oral dexamethasone premedication (20 mg), followed by surgery and two cycles of adjuvant epirubicin plus docetaxel administered every 3 weeks. Adjuvant radiation, endocrine, and trastuzumab therapies were administered when indicated. 24 postmenopausal women with HR+ breast cancer received an aromatase inhibitor; 11 premenopausal women with OR+ received an aromatase inhibitor. 3 of the postmenopausal women with HR+ breast cancer received tamoxifen; 24 of the premenopausal women with OR+ received tamoxifen. 3 postmenopausal women with HR+ received no endocrine therapy; 2 premenopausal women with OR+ received no endocrine therapy.
Outcomes	primary endpoint was the number of patients with detectable DTCs at 3 months. secondary endpoints were changes in bone‐turnover markers (N‐telopeptide [NTx], osteocalcin, serum bone‐specific alkaline phosphatase [BSAP]), measured at baseline, 3 months and 12 months, and in bone‐mineral density, measured at baseline and 12 months.
Notes	Funding sources: Novartis Pharmaceuticals and Pfizer Inc. Conflicts of interest: •RA and KW: Novartis •MN: Novartis, Sanofi‐Aventis, Pfizer •ME: Novartis •KD: Novartis •WS: received funds for portions of the statistical analysis •All other authors declared no conflicts of interest.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Block randomisation by formal probability model and implemented with SAS process plan generated by statistician
Allocation concealment (selection bias)	Low risk	Allocation placed in sequentially numbered, opaque envelopes in locked cabinet, only accessible to study's patient co‐ordinator after enrolment
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Low number of withdrawals for the primary outcome. A negative outcome was assigned to participants with missing data points. For all other outcomes of interest in this review, there were no significant differences in attrition between the groups and reasons for any withdrawal were provided.
Selective reporting (reporting bias)	Low risk	All predefined outcomes were reported.
Other bias	Unclear risk	DTCs is a difficult endpoint, which may or may not correlate directly with clinically evident bone metastases.

ANZAC 2009.

Study characteristics
Methods	Setting: phase II Recruitment period: July 2007 and July 2009 Length of study: 18 weeks (during chemotherapy before surgery, surgery marked end of study) Length of follow‐up: standard follow‐up
Participants	Eligibility criteria: • female with a histologic diagnosis of invasive breast cancer; • scheduled to receive neoadjuvant anthracycline‐based chemotherapy • 18 years or older; • WHO performance status of 0 to 2; • T2 to T4 tumour with no evidence of metastatic disease; • prepared to undergo additional tumour biopsies for research. Exclusion criteria: • concurrent treatment with tamoxifen or an aromatase inhibitor • need for oral anticoagulants; • exposure to bisphosphonates in the last year; • active dental problems including dental abscess or osteonecrosis of the jaw; • insufficient renal function (creatinine clearance < 40 mL/min). Stage of disease: T2‐T3: 33 patients T4: 2 patients T4d: 5 patients TNM staging system: NR Mean age: intervention 1: 51 (46‐56) intervention 2: 49 (46‐57) Menopausal status: 55% premenopausal 45% postmenopausal RANKL status: NR Hormone receptor status: intervention 1: 16 (80 %) ER+ intervention 2: 14 (70 %) ER+ Human epidermal growth factor receptor 2 status: intervention 1: 9 (45 %) HER2‐positive intervention 2: 8 (40 %) HER2‐positive Participants randomised: total: 40 Country of participants: NR
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: single intravenous infusion of 4 mg zoledronic acid 24 hours after the first cycle of FEC; intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: neoadjuvant chemotherapy with 5‐fluouracil 500 mg/m², epirubicin 100 mg/m², cyclophosphamide 500 mg/m² (FEC) every 3 weeks, followed by 3 cycles of docetaxel 100 mg/m²
Outcomes	• change in apoptotic index; • evaluation of changes in Ki67 proliferation from baseline to subsequent time points; • changes in growth index reflecting a combined contribution of proliferation and apoptosis to changes in growth.
Notes	Funding sources: "The authors thank Janet Horsman, Cancer Clinical Trials Centre, University of Sheffield; Yvonne Stephenson and team, Medical School, University of Sheffield; Simone Detre, Institute of Cancer Research, Royal Marsden, London; Rosie Taylor, School of Health and Related Research, University of Sheffield; and Weston Park Hospital Cancer Charity and Experimental Cancer Medicine Centre Network for part funding of this research." Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"patients were randomised in a 1:1 ratio".
Allocation concealment (selection bias)	Unclear risk	No further information given
Blinding of participants (performance bias)	Unclear risk	No further information given
Blinding of personnel (performance bias)	Unclear risk	No further information given
Blinding of outcome assessment (detection bias) subjective outcomes	Unclear risk	No further information given
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	"all sections were counted blinded to treatment allocation and time point" "Areas of invasive carcinoma on the digital image of the section were selected at random by a special pathologist blinded to treatment allocation and timing of sample"
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Not enough information given
Selective reporting (reporting bias)	Unclear risk	Study protocol not accessible
Other bias	Low risk	Study appeared to be free of other sources of bias.

ARBI 2009.

Study characteristics
Methods	Setting: multicentre, phase III Recruitment period: February 2005 and February 2007 Length of study: 2 years Length of follow‐up: 2 years
Participants	Eligibility criteria: • postmenopausal; • histologically confirmed hormone‐receptor positive breast cancer; • completed primary surgery and chemotherapy; • scheduled to receive anastrozole. Exclusion criteria: • menopause induced by prior chemotherapy or any other drug therapy; • evidence of metastatic bone disease from bone scans; • previous hip fractures or protheses; • known bone metabolism disorder; • non‐treated hypocalcemia; • previous treatment with selective oestrogen receptor modulators (SERMs); • hormone replacement therapy (HRT); • previous treatment with BPs; • liver or renal dysfunction. Stage of disease: early breast cancer TNM staging system: NR Mean age: intervention 1: 62.6 (SD 8.5) intervention 2: 64.5 (SD 9.2) Menopausal status: 100 % postmenopausal RANKL status: NR Hormone receptor status: ER+ intervention 1: 97.3 % intervention 2: 90.9 % PR+ intervention 1: 70.3 % intervention 2: 72.7 % Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 70 Country of participants: Greece
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 35 mg of oral risedronate weekly for 2 years (first thing in the morning with water); intervention 2: no treatment; supplemental: 400 IU vitamin D and 1000 mg of calcium per day for 2 years; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: anastrozole
Outcomes	• the effect of risedronate in patients with mild osteopenia receiving anastrozole therapy, measured in lumbar spine and hip at 12 and 24 months; • BMD in patients with normal BMD before starting treatment; • the effect of risedronate on patients receiving anastrozole therapy who have BMD in the region of severe osteopenia or osteoporosis; • adverse event; • ONJ; • (fragility) fractures.
Notes	Funding sources: "CM has received educational grants and lecture honoraria from AstraZeneca, Novartis (Basel, Switzerland), and Pfizer Inc (New York, NY, USA). ET, AP, BV, GX, JP, VZ, JM, KK, DK, ZA, and HG have received unrestricted educational grants from AstraZeneca, Novartis, and Pfizer Inc. The other authors declare that they have no competing interests." Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomly assigned"
Allocation concealment (selection bias)	Unclear risk	No further information given
Blinding of participants (performance bias)	High risk	"open‐label clinical trial"
Blinding of personnel (performance bias)	High risk	"open‐label clinical trial"
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	"open‐label clinical trial"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	High risk	Data from 23 of 70 patients were not analysed; no reason given for 12 of these 23 people, no ITT analysis
Selective reporting (reporting bias)	Unclear risk	Study protocol accessible but no outcomes listed
Other bias	Low risk	Study appeared to be free of other sources of bias.

ARIBON 2012.

Study characteristics
Methods	Setting: multicentre, double‐blind Recruitment period: December 2003‐October 2005 Length of study: 2 years Length of follow‐up: 5 years
Participants	Eligibility criteria: postmenopausal women with a histologically confirmed diagnosis of oestrogen‐receptor positive breast cancer Exclusion criteria: menopause induced by either prior chemotherapy or drug therapy; concurrent administration of medication(s) with effects on bone such as bisphosphonates or hormone replacement therapy; abnormal renal function; disorders of bone metabolism; previous bilateral hip fractures or bilateral hip prostheses that would have made BMD assessment impossible. Stage of disease: early stage TNM staging system: NR Mean age: intervention 1: median 67.8 (range 58.9 to 73.4) intervention 2: median 67.5 (range 63.6 to 71.0) Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: 100% ER+ Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 50 Country of participants: UK
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: ibandronate (150 mg, oral, every 28 days, 2 years); intervention 2: placebo (oral, every 28 days, 2 years); supplemental: Vitamin D (400 IU, daily) Calcium (500 mg, daily); other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: aromatase inhibitor (anastrozole, 1 mg/day)
Outcomes	change in BMD at the lumbar spine and hip; changes in bone resorption and formation markers over time; any fracture; adverse events.
Notes	Funding sources: The study was sponsored by the University of Sheffield, and funded through unrestricted academic grants from AstraZeneca and Roche, who also supplied the trial medications of anastrozole and ibandronate, respectively. Both companies were kept informed of the progress of the study but had no involvement in the analysis or interpretation of the results presented. Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"were randomized"
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes, unlikely to be vulnerable to bias
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of patients that withdrew from each arm was almost equal (4 from arm 1 and 6 from arm 2)
Selective reporting (reporting bias)	Unclear risk	Study protocol was not accessible.
Other bias	Low risk	Study appeared to be free of other sources of bias.

AZURE 2018.

Study characteristics
Methods	Setting: multicentre, phase III Recruitment period: September 2003 to February 2006 Length of study: 5 years Length of follow‐up: 10 years
Participants	Eligibility criteria: • at least 18 years of age; • Karnofsky performance status of at least 80; • histologically confirmed breast cancer with axillary lymph‐node metastasis (N1) or a T3‐T4 primary tumour; • complete primary tumour resection was mandated or intended after neoadjuvant therapy; • patients who were eligible for completion surgery (margin excision, mastectomy, or axillary lymph‐node dissection) after completion of adjuvant chemotherapy could be included; • serum creatinine level less than 1.5 times the upper limit of the normal range. Exclusion criteria: • clinical or imaging evidence of distant metastases; • if complete treatment of the primary breast tumour and regional lymph nodes was not possible; • cancer diagnosis within the preceding 5 years; • use of bisphosphonates during the previous year; • diagnosis of osteoporosis or other bone disease likely to require bone‐targeted treatment; • patients with clinically significant, active dental problems or planned jaw surgery. Stage of disease: axillary lymph nodes: 0: 1.7 % (intervention 1), 1.9 % (intervention 2) 1‐3: 61.9 % (intervention 1), 61.5 % (intervention 2) ≥ 4: 35.9% (intervention 1), 36.2% (intervention 2) TNM staging system: • T1: 32.1% (intervention 1), 31.2% (intervention 2) • T2: 50.6% (intervention 1), 51.7% (intervention 2) • T3: 13.5% (intervention 1), 13.6% (intervention 2) • T4: 3.5% (intervention 1), 3.5% (intervention 2) Lymph nodes: • 1‐3 involved: 59.1% intervention 1 vs. 61.5% intervention 2 • ≥ 4 involved: 38.2% intervention 1 vs. 36.8% intervention 2 Mean age: intervention 1: median of 47 (range: 42‐57) intervention 2: median of 48 (range: 41‐54) Menopausal status: premenopausal: 44.7% (intervention 1), 44.8 % (intervention 2) postmenopausal ≤ 5 years: 14.7% (intervention 1), 14.5 % (intervention 2) postmenopausal > 5 years: 30.8% (intervention 1), 31.3 % (intervention 2) RANKL status: NR Hormone receptor status: Human epidermal growth factor receptor 2 status: Participants: total: 3360 Country of participants: NR
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 3 to 4 weeks for 6 doses, then every 3 months for 8 doses, then every 6 months for 5 doses; intervention 2: no treatment; supplemental: 200‐500 IU of oral vitamin D and 400‐1000 mg of oral calcium, both only in the first 6 months; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: chemotherapy alone: 21.5% (intervention 1), 21.5% (intervention 2) endocrine therapy plus chemotherapy: 73.9% (intervention 1), 74.1% (intervention 2) type of chemotherapy: either anthracyclines or taxanes radical radiotherapy with curative intent within 6 months of starting therapy "Less than half of the HER2 positive patients included in AZURE received trastuzumab and only around two thirds of women received an aromatase inhibitor, with most use being after a few years of tamoxifen rather than as initial endocrine treatment."
Outcomes	• disease‐free survival as assessed annually for 10 years; • time to bone metastases as first recurrence assessed annually for 10 years; • time to bone metastases per se as assessed annually for 10 years; • time to distant metastases as assessed annually for 10 years; • overall survival as assessed by final analysis at 10 years; • skeletal‐related events prior to development of bone metastases as assessed annually for 10 years; • skeletal‐related events following development of bone metastases as assessed annually for 10 years; • safety and toxicity of zoledronic acid as assessed annually for 10 years; • evaluation of the influence of prognostic factors (e.g. oestrogen receptor or progesterone receptor [ER/PR] status, TNM stage, tumour grade, HER2/neu, and menopausal status) on treatment outcome; • analysis of tumour‐specific mutations, proteomics and gene expression changes in tumour cells.
Notes	Funding sources: Novartis Pharmaceuticals and the National Cancer Research Network Conflicts of interest: RE Coleman: Novartis, Amgen, Roche, Pfizer D Cameron: Novartis, Roche D Dodwell: Roche, AstraZeneca, Novartis R Burkinshaw: Novartis Pharmaceuticals
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"randomly assigned (1:1) by a central automated 24‐h computer‐generated telephone minimisation system"
Allocation concealment (selection bias)	Low risk	"randomly assigned (1:1) by a central automated 24‐h computer‐generated telephone minimisation system"
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT analysis
Selective reporting (reporting bias)	Low risk	All endpoints were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

BONADIUV 2019.

Study characteristics
Methods	Setting: single‐centre, phase II, single‐blind Recruitment period: 2011‐2014 Length of study: 2 years Length of follow‐up: 5 years
Participants	Eligibility criteria: diagnosis of hormone receptor‐positive early breast cancer; postmenopausal status; age < 75 years. Exclusion criteria: premenopausal status at time of randomisation to zoledronic acid; disorders of bone metabolism; body mass index (BMI) < 18 kg/m²; chronic use of steroids; use of bisphosphonates at time of randomisation to zoledronic acid; renal disorder; previous bilateral hip fractures or bilateral hip protheses; psychiatric disorders or any condition preventing participants from taking oral drugs; patients with normal BMD (T score more than ‐1 at both LS and Th) were excluded from the study. Stage of disease: NR TNM staging system: T1‐2: 61.8% ibandronate vs. 69.5% placebo N0: 58.4% ibandronate vs. 64.6% placebo Mean age: intervention 1: 60.5 intervention 2: 59.6 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: 100% HR+ Human epidermal growth factor receptor 2 status: 20.2% (intervention 1) vs. 14.6% (intervention 2) Participants: total: 171 Country of participants: Italy
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: ibandronate (150 mg, oral, every 28 days, 2 years); intervention 2: placebo (oral, every 28 days, 2 years); supplemental: Vitamin D (4000 IU, weekly, 5 years) Calcium (500 mg, daily, 5 years); other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: aromatase inhibitor (70.8% letrozole, 20.5% anastrozole, 8.7% exemestane)
Outcomes	lumbar spine and trochanter T‐score mean difference; number of patients with adverse events as a measure of safety and tolerability; overall survival; disease‐free survival.
Notes	Funding sources: This research did not receive any specific grant from funding agencies in the public, commercial, or not‐for‐profit sectors. Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Allocation of the patients in the trial arms was conducted by a stratified randomisation, using computer‐generated randomised permuted blocks within defined post‐menopausal and BMI strata."
Allocation concealment (selection bias)	Low risk	"Allocation of the patients in the trial arms was conducted by a stratified randomisation, using computer‐generated randomised permuted blocks within defined post‐menopausal and BMI strata."
Blinding of participants (performance bias)	Low risk	Single‐blind
Blinding of personnel (performance bias)	Low risk	Single‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Single‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes, unlikely to be vulnerable to bias
Incomplete outcome data (attrition bias) All outcomes	Low risk	A total of 27 patients didn't complete the study (17 in arm 1 and 10 in arm 2); intention‐to‐treat (ITT) analysis
Selective reporting (reporting bias)	Low risk	Protocol was available and all prespecified outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

Bundred 2009.

Study characteristics
Methods	Setting: phase IV Recruitment period: NR Length of study: 14 days Length of follow‐up: NR
Participants	Eligibility criteria: oestrogen receptor positive (> 10%) invasive breast cancer; signed written informed consent; postmenopausal status as defined by one of the following: age greater than 55 years with cessation of menses; age less than 55 but no spontaneous menses for at least one year; bilateral oophorectomy; age less than 55 and spontaneous menses within the last year with postmenopausal gonadotrophin levels (LH and FSH levels greater than 40 IU/L); or postmenopausal oestradiol levels (less than 5 ng/dL); hormone receptor‐positive tumours as defined by ER and PR greater than 10% of tumour cells positive by immunohistochemical staining (or Allred score > 2); patients must have surgically operable breast cancer, which has not received prior systemic (i.e. adjuvant chemotherapy) or local (radiotherapy) treatments according to best practice; patients must consent to or have undergone a core biopsy for the diagnosis of their breast cancer. Exclusion criteria: oestrogen receptor negative (< 10%) (or Allred score 0 or 1) or ER unknown invasive breast cancer patients on hormone replacement therapy at diagnosis patients who have received prior treatment with oral or intravenous bisphosphonates within the last twelve months known hypersensitivity to Zometa (zoledronic acid) or other bisphosphonates abnormal renal function as defined by serum creatinine and equal to or greater than 3mg/dL (265.2mmol/L). Stage of disease: NR TNM staging system: NR Mean age: overall 50‐75 intervention 1: NR intervention 2: NR Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: 100% ER+ Human epidermal growth factor receptor 2 status: NR Participants: total: 109 Country of participants: NR
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: zoledronic acid (4 mg, intravenously, single‐dose, 2‐4 days before definitive surgical excision) intervention 2: placebo (oral, daily) intervention 3: no treatment supplemental: NR other: surgery at the end of the study Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy, aromatase inhibitor (letrozole)
Outcomes	NR
Notes	Funding sources: NR Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"were randomised"
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	Unclear risk	Only abstract
Blinding of personnel (performance bias)	Unclear risk	Only abstract
Blinding of outcome assessment (detection bias) subjective outcomes	Unclear risk	Only abstract
Blinding of outcome assessment (detection bias) objective outcomes	Unclear risk	Only abstract
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Only abstract
Selective reporting (reporting bias)	Unclear risk	Only abstract
Other bias	Unclear risk	Only abstract

Cohen 2008.

Study characteristics
Methods	Setting: multicentre, randomised, double‐blind, placebo‐controlled Recruitment period: NR Length of study: 1 year Length of follow‐up: NR
Participants	Eligibility criteria: eligible patients had completed at least 2 years of tamoxifen therapy and were withdrawn from this treatment as part of their routine clinical care; postmenopausal. Exclusion criteria: premenopausal state; treatment with an aromatase inhibitor; known or suspected metastatic disease; concomitant hyperthyroidism, liver disease, acromegaly, Cushing syndrome, rheumatoid arthritis, multiple myeloma, Paget disease of bone, renal osteodystrophy, or osteomalacia; treatment within 3 months before enrolment with androgen, oestrogen, calcitonin, systemic corticosteroids, fluoride, bisphosphonates, vitamin D in dosages exceeding 800 IU/day, vitamin D metabolites, thiazide diuretics, or lithium; adjustments in thyroid hormone dosages of more than 25 μg within 6 months before enrolment; impaired renal function; documented gastro‐oesophageal reflux disease or peptic ulcer disease; allergy or intolerance to bisphosphonates. Stage of disease: NR TNM staging system: NR Mean age: overall 58 ± 2 intervention 1: NR intervention 2: NR Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: NR Human epidermal growth factor receptor 2 status: NR Participants randomised: 11 Country of participants: NR
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: alendronate (70 mg, oral, weekly, 1 year); intervention 2: placebo (oral, weekly, 1 year); supplemental: Vitamin D (400 IU, daily) Calcium (1200 mg, daily); other: NR. Previous bone‐modifying interventions: study started within 3 months of tamoxifen discontinuation Cancer treatment during study period: NR
Outcomes	BMD in lumbar spine, hip and forearm; bone turnover.
Notes	Funding sources: This study was funded by a grant from the Avon Foundation and by grant K24 DK074457 from the National Institutes of Health. Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomized"
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	"no patient withdrew from the study".
Selective reporting (reporting bias)	Unclear risk	Study protocol not accessible
Other bias	Low risk	Study appeared to be free of other sources of bias.

D‐CARE 2013.

Study characteristics
Methods	Setting: multicentre, double‐blind, randomised, placebo‐controlled, phase III Recruitment period: June 2010‐August 2012 Length of study: 5 years Length of follow‐up: 5 years
Participants	Eligibility criteria: histologically confirmed, American Joint Committee on Cancer (AJCC) stage II or III breast cancer; high risk of breast cancer recurrence, defined as documented evidence of one or more of the following criteria: i) biopsy evidence of breast cancer in regional lymph node(s) (LN) (node‐positive disease); nodal micrometastases only are not considered node‐positive ii) tumour size > 5 cm (T3) or locally advanced disease (T4); documented pathological evaluation of breast cancer for hormone receptor (oestrogen receptor [ER] and progesterone receptor [PR]) status and HER‐2 status; subjects must be receiving or be scheduled to receive standard of care systemic adjuvant or neoadjuvant chemotherapy and/or endocrine therapy and/or HER‐2 targeted therapy; for subjects receiving adjuvant therapy only: subjects must have undergone complete resection of the primary tumour with clean surgical margins, or subjects must have undergone resection of the primary tumour and be scheduled for further treatment of the primary tumour with curative intent. Definitive treatment must be planned to be completed within approximately 9 months of randomisation to zoledronic acid time between definitive surgery and randomisation to zoledronic acid must be ≤ 12 weeks. Definitive surgery may include secondary interventions (e.g. to clear inadequate surgical margins) subjects with node‐positive disease must have undergone treatment of axillary LN with curative intent, or subjects must be scheduled for further treatment of regional lymph nodes with curative intent. definitive treatment must be planned to be completed within approximately 9 months of randomisation to zoledronic acid subjects must not have received prior neoadjuvant treatment. Endocrine treatment for less than 30 days prior to surgery is not considered prior neoadjuvant treatment; for subjects receiving neoadjuvant therapy only: time between start of neoadjuvant treatment and randomisation to zoledronic acid must be ≤ 8 weeks and subjects must be scheduled to undergo definitive treatment (including surgery and/or radiotherapy) with curative intent within approximately 9 months of starting neoadjuvant treatment female subjects with age ≥ 18 years subjects with reproductive potential must have a negative pregnancy test within 14 days before randomisation to zoledronic acid serum calcium or albumin‐adjusted serum calcium ≥ 2.0 mmol/L (8.0 mg/dL) and ≤ 2.9 mmol/L (11.5 mg/dL) Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1. Exclusion criteria: prior or current evidence of any metastatic involvement of any distant site; history of breast cancer (other than ductal carcinoma in situ [DCIS] or lobular carcinoma in situ [LCIS]) prior to the current diagnosis; osteoporosis requiring treatment at the time of randomisation to zoledronic acid or treatment considered likely to become necessary within the subsequent six months; any prior or synchronous malignancy (other than breast cancer), except i) malignancy treated with curative intent and with no evidence of disease for ≥ 5 years prior to enrolment and considered to be at low risk for recurrence by the treating physician; ii) adequately treated non‐melanoma skin cancer or lentigo maligna without evidence of disease; active infection with Hepatitis B virus or Hepatitis C virus; known infection with human immunodeficiency virus (HIV); prior history or current evidence of osteomyelitis/osteonecrosis of the jaw; active dental or jaw condition which requires oral surgery; planned invasive dental procedure for the course of the study; non‐healed dental or oral surgery; use of oral bisphosphonates within the past 1 year; prior or current IV bisphosphonate administration; prior administration of denosumab; subject is currently enroled in or has not yet completed at least 30 days since ending other investigational device or investigational drug study(s), or subject is receiving other investigational agent(s); subject is pregnant or breastfeeding, or planning to become pregnant within 7 months after the end of treatment; subject is of childbearing potential and is not willing to use, in combination with her partner, 2 highly effective methods of contraception or abstinence during treatment and for 5 months after the end of treatment; subject has known sensitivity to any of the products to be administered during the study (e.g. mammalian derived products, calcium, or vitamin D); subject has any kind of disorder that compromises the ability of the subject to give written informed consent and/or to comply with study procedures; any major medical or psychiatric disorder that, in the opinion of the investigator, prevents the subject from completing the study or interferes with the interpretation of the study results. Stage of disease: stage II or III TNM staging system: N0: 131 (6%, intervention 1) vs. 130 (6%, intervention 2) N1: 1381 (61%, intervention 1) vs. 1369 (61%, intervention 2) N2: 505 (22%, intervention 1) vs. 506 (22%, intervention 2) N3: 224 (10%, intervention 1) vs. 230 (10%, intervention 2) Mean age: intervention 1: median 50 (range 44‐59) intervention 2: median 51 (range 44‐59) Menopausal status: premenopausal: 1195 (53%, intervention 1) vs. 1165 (52%, intervention 2) postmenopausal: 1061 (47%, intervention 1) vs. 1088 (48%, intervention 2) RANKL status: NR Hormone receptor status: 1744 (77%, intervention 1) vs 1748 (78%, intervention 2) ER+ and/or PR+ Human epidermal growth factor receptor 2 status: 454 (intervention 1) vs. 451 (intervention 2) Participants randomised: total: 4509 Country of participants: 39 different countries
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: denosumab (120 mg, subcutaneously, every 4 weeks for the first 6 months, then every 3 months, 5 years); intervention 2: placebo (120 mg, subcutaneously, every 4 weeks for the first 6 months, then every 3 months, 5 years); supplemental: Vitamin D (400 IU, daily) Calcium (500 mg, daily); other: Anti‐HER2 therapy: 362 (intervention 1) 359 (intervention 2) Ovarian ablation: 199 (intervention 1) 193 (intervention 2). Previous bone‐modifying interventions: NR Cancer treatment during study period: Endocrine therapy SERM (tamoxifen): 991 (intervention 1) vs. 970 (intervention 2) aromatase inhibitor (exemestane, letrozole, anastrozole): 934 (intervention 1) vs. 964 (intervention 2) ovarian function suppression (goselerin, leuprolide): 199 (intervention 1) vs. 193 (intervention 2) Chemotherapy total: 2166 (intervention 1) vs. 2155 (intervention 2) taxanes: 1833 (intervention 1) vs. 1853 (intervention 2) anthracyclines: 1888 (intervention 1) vs. 1904 (intervention 2)
Outcomes	Bone Metastasis‐free Survival (BMFS); Disease‐free Survival (DFS); Disease‐free Survival (DFS) in the postmenopausal subset; Overall Survival; Distant Recurrence‐free Survival.
Notes	Funding sources: Amgen Conflicts of interest: "RC has received steering committee fees and travel support from Amgen; has received funding for IME lectures from Eisai, Amgen, and Genomic Health; has received consulting fees from Astellas and Scancell; and has a patent for biomarker testing pending (Inbiomotion). CB has received clinical trial grants from Amgen; has received research grants from AbbVie, Astellas Pharma, AstraZeneca, Bristol‐Myers Squibb, Celgene, Covance, Lilly, Medivation, Merck Serono, Merck Sharp and Dohme, Novartis, Pfizer, PharmaMar, and Roche/Genentech; and has received honoraria for presentations and consulting from AstraZeneca, Bristol‐Myers Squibb, Lilly, Merck Sharp and Dohme, Novartis, Pfizer, and Roche/Genentech. MM has received grants from Roche and Novartis; and has received personal fees from Roche, Novartis, AstraZeneca, Amgen, Taiho, PharaMar, Lilly, Puma, and Pfizer. HI has received honoraria and consulting fees from Daiichi‐Sankyo, F Hoffmann‐La Roche via Chugai, Novartis, AstraZeneca, Pfizer, and Lilly, and was an uncompensated member of the steering committee for this D‐CARE study. ID reports an institutional educational grant from Amgen. JC has received institutional research funding from Roche, Eisai, Puma Biotechnologies, and Boerhinger Ingelheim; has served on advisory boards for Amgen, Eisai, Pfizer, Vertex Pharmaceuticals, Puma Biotechnologies, Pierre Fabre, Seattle Genetics, Boehringer Ingelheim, Genomic Health, and AstraZeneca; has served as a meeting chair for Roche Products Ireland and MSD Ireland; has received conference travel and accommodation funding from Bayer, Pfizer, Roche, Merck Sharpe and Dohme, AstraZeneca, and AbbVie, and is an employee of and stockholder in Oncomark Ltd. SD has received institutional grants from Amgen. TD, DJ, and YZ are employees of Amgen; and DJ is a stockholder in Amgen. DMF, RH, JG, JS, AC, AMP, and KT declare no competing interests."
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"On confirmation of eligibility by local investigator, investigators at each site telephoned an interactive voice response system to centrally randomise patients (1:1) based on a fixed stratified permuted block randomisation list (block size 4) [...] The randomisation list was generated and maintained by an Amgen‐based independent internal randomisation group not involved in the implementation of the study using a fully validated randomisation system."
Allocation concealment (selection bias)	Low risk	"On confirmation of eligibility by local investigator, investigators at each site telephoned an interactive voice response system to centrally randomise patients (1:1) based on a fixed stratified permuted block randomisation list (block size 4) [...] The randomisation list was generated and maintained by an Amgen‐based independent internal randomisation group not involved in the implementation of the study using a fully validated randomisation system."
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT analysis
Selective reporting (reporting bias)	Low risk	All prespecified outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

Delmas 1997.

Study characteristics
Methods	Setting: single‐centre, double‐blind, placebo‐controlled, randomised, oral‐dose, phase II Recruitment period: NR Length of study: 2 years Length of follow‐up: 3 years
Participants	Eligibility criteria: all were postmenopausal for ≥ 6 months due to chemotherapy or radiotherapy following breast cancer surgery (follicle‐stimulating hormone [FSH] level > 15 mU/mL and oestradiol level c 30 pg/mL) Exclusion criteria: severe allergies or abnormal reactions to any drugs; history of chronic alcohol or drug abuse; significant psychiatric or organic disease; clinically significant abnormal laboratory values (unless related to breast cancer or adjuvant chemotherapy); evidence of osteoporosis; hyperparathyroidism; Paget's disease of bone; renal osteodystrophy; treatment with bisphosphonates, fluoride, corticosteroids, anabolic drugs or vitamin D (> 400 IU/d) within the preceding 6 months; treatment with calcitonin within the previous month. Stage of disease: NR TNM staging system: NR Mean age: intervention 1: 45.7 ± 4.0 intervention 2: 46.6 ± 4.6 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: NR Human epidermal growth factor receptor 2 status: NR Participants randomised: 53 Country of participants: NR
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: risedronate (30 mg, oral, daily for 2 weeks and then no medication for 10 weeks, eight 12‐week cycles, 2 years; intervention 2: placebo (oral, daily for 2 weeks and then no medication for 10 weeks, eight 12‐week cycles, 2 years); supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: 36 of 53 patients received tamoxifen; chemotherapy
Outcomes	changes in lumbar spine and proximal femur bone mineral density; biochemical markers of bone turnover.
Notes	Funding sources: NR Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomly assigned"
Allocation concealment (selection bias)	Unclear risk	No information
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	"The group of primary interest was the intent‐to‐treat (ITT) population, which includes all available data from patients randomized onto the study."
Selective reporting (reporting bias)	Unclear risk	Study protocol not accessible
Other bias	Low risk	Study appeared to be free of other sources of bias.

Diel 1998.

Study characteristics
Methods	Setting: single‐centre Recruitment period: 1990‐1995 Length of study: 2 years Length of follow‐up: median follow‐up of 8.5 years
Participants	Eligibility criteria: T1‐T4; N0‐N2; primary breast cancer with positive immunocytochemical detection of tumour cells in bone marrow. Exclusion criteria: confirmed distant metastasis; prior or simultaneous secondary malignant disease; skeletal disease; hepatic or renal dysfunction pregnancy; history of neoadjuvant chemotherapy or hormonal therapy. Stage of disease: T3 and T4 (17% intervention 1), 16% (intervention 2) node‐positive disease: 51% (intervention 1), 54% (intervention 2) TNM staging system: T1: 38% (intervention 1), 37% (intervention 2) T2: 45% (intervention 1), 46% (intervention 2) T3 and T4: 17% (intervention 1), 16% (intervention 2) Mean age: intervention 1: NR ("well balanced") intervention 2: NR ("well balanced") Menopausal status: intervention 1: 64% postmenopausal intervention 2: 61% postmenopausal RANKL status: NR Hormone receptor status: ER+: intervention 1: 75% intervention 2: 71% PR+: intervention 1: 62% intervention 2: 63% Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 302 Country of participants: Germany
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 1600 mg of oral clodronate daily for 2 years; intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy alone: 41% (intervention 1), 38% (intervention 2) SERM: 31% (intervention 1), 30% (intervention 2) goselerin: 10% (intervention 1), 8% (intervention 2) chemotherapy alone: 25% (intervention 1), 28% (intervention 2) all patients received standard surgical treatment and customary adjuvant endocrine therapy or chemotherapy ± radiotherapy combination therapy: 16% (intervention 1), 15% (intervention 2) + standard surgical treatment
Outcomes	• incidence and the number of new bone and visceral metastases, as well as the length of time to their appearance and OS; • DFS.
Notes	Funding sources: Boehringer Mannheim (Roche Pharma) Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomly assigned"
Allocation concealment (selection bias)	Unclear risk	No information
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	High ‐ open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT analysis
Selective reporting (reporting bias)	Low risk	All endpoints were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

Ellis 2008.

Study characteristics
Methods	Setting: multicentre, phase III Recruitment period: NR Length of study: 2 years Length of follow‐up: 4 years
Participants	Eligibility criteria: histologically or cytologically confirmed adenocarcinoma of the breast; early‐stage disease; oestrogen receptor‐positive; completed their treatment pathway (surgery, chemotherapy, radiation, and/or hormone therapy); currently on or will initiate aromatase inhibitor therapy, and are expected to stay on aromatase inhibitor therapy for the duration of the 24‐ month study; all treatment pathway must be completed 4 weeks prior to study entry, and all acute toxic effects of any above therapy must be resolved to Grade 1 by National Cancer Institute (NCI) Common terminology criteria for adverse events (CTCAE); female, 18 or older; ECOG performance status 0 and 1; lumbar spine, total hip or femoral neck BMD equivalent to a t‐score classification of ‐1.0 to ‐2.5. Exclusion criteria: osteoporosis; prior vertebral fracture; current use of bisphosphonates; use of any antineoplastic therapy apart from aromatase inhibitors. Stage of disease: early stage breast‐cancer TNM staging system: NR Mean age: intervention 1: 59.2 (range 38‐84) intervention 2: 59.7 (range 35‐81) Menopausal status: 50 % were postmenopausal with more than 10 years since their last menstrual period. 23 % were postmenopausal with 5 to 10 years since their last menstrual period. RANKL status: NR Hormone receptor status: 100% of the women were ER+ Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 252 Country of participants: Canada, USA
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 60 mg of subcutaneous denosumab every 6 months for 24 months; intervention 2: subcutaneous placebo every 6 months for 24 months; supplemental: > 400 IU of vitamin D and 1 g of calcium per day; other: NR. Previous bone‐modifying interventions: 3 patients in arm 1 and 7 patients in arm 2 have previously received oral bisphosphonates. Cancer treatment during study period: 53 patients in arm 1 and 58 patients in arm 2 received prior tamoxifen therapy.
Outcomes	percentage change in lumbar spine BMD from baseline to 12 months compared with placebo; percentage changes from baseline in lumbar spine BDM at 6 months and total hip and femoral neck BMD at 6 and 12 months; exploratory endpoints included percentage changes from baseline in lumbar spine, total hip, and femoral neck BMD at 1, 3, and 24 months; the percentage of patients with BMD gains at the lumbar spine; percentage changes from baseline in total body and on‐third radius BMD at 12 and 24 months; percentage changes in trochanter BMD at months 1, 3, 6, 12 and 24 months; percentage changes from baseline in markers of bone remodeling sCTx and PINP at 1, 6, 12 and 24 months; incidence of vertebral and nonvertebral fractures; overall survival.
Notes	Funding sources: Georgiana K. Ellis, Amgen; Henry G. Bone, Amgen; Rowan Chlebowski, Lilly, Amgen Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"interactive voice response system was used to randomly assign patients 1:1[…]"
Allocation concealment (selection bias)	Low risk	"interactive voice response system was used to randomly assign patients 1:1[…]"
Blinding of participants (performance bias)	Low risk	"Masking: Triple (Participant, Care Provider, Investigator)"
Blinding of personnel (performance bias)	Low risk	"Masking: Triple (Participant, Care Provider, Investigator)"
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	"Masking: Triple (Participant, Care Provider, Investigator)"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of patients that withdrew from the study was pretty similar and without big differences in the reasons why they withdrew from the study.
Selective reporting (reporting bias)	Unclear risk	Study protocol was available but no outcomes were prespecified.
Other bias	Low risk	Study appeared to be free of other sources of bias.

EXPAND 2011.

Study characteristics
Methods	Setting: multicentre, phase III Recruitment period: NR Length of study: 3 years Length of follow‐up: 3 years
Participants	Eligibility criteria: • compliant postmenopausal women with primary operable breast cancer after 4 to 6 years of therapy with tamoxifen (end of tamoxifen therapy within last 6 months); • performance status 0‐2 (Eastern Cooperative Oncology Group); • patients without severe osteoporosis at study entry; • no evidence of relapse at the time of randomisation; • adequate function of bone marrow, kidney, and liver; • 18 years or older. Exclusion criteria: • oestrogen‐ and progesterone‐receptor status negative or unknown; • completion of adjuvant tamoxifen therapy more than 6 months prior to study start; • inflammatory breast cancer; • current/active dental problems including infection of the teeth or jawbone, dental or fixture trauma, or a current or prior diagnosis of osteonecrosis of the jaw, of exposed bone in the mouth, or of slow healing after dental procedures; • recent (within 6 months) or planned dental or jaw surgery; • history of diseases with an influence on bone metabolism such as Paget's disease and primary overactive parathyroid; • prior or concomitant therapies: chemotherapy within the last 12 months, intravenous or oral bisphosphonates, systemic corticosteroids, anabolic steroids or growth hormones, tibolone, parathyroid hormone, systemic sodium fluoride or any drugs known to affect the skeleton (such as calcitonin, mithramycin, or gallium nitrate); • patients with previous or concomitant cancers (not breast cancer) within the past 5 years EXCEPT adequately treated basal or squamous cell skin cancers or in situ cancer of the cervix. Patients with other previous cancer(s) must have been disease‐free for at least 5 years; • patients currently receiving oral bisphosphonates must discontinue these at least 3 weeks prior to study start. Stage of disease: NR TNM staging system: NR Mean age: intervention 1: 58.4 (SD: 7.3) intervention 2: 61.3 (SD: 7.3) Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: 100 % HR+ (no information about whether PR or ER) Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 81 Country of participants: Germany
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 6 months for 36 months in total; intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: participants in both arms received 2.5 mg of letrozole daily.
Outcomes	change in Bone Mineral Density (BMD) from baseline to month 36 [time frame: at 36 months as compared to baseline]; percent change in Bone Mineral Density (BMD) from baseline to month 36 [time frame: baseline, month 36]; change in T‐score from baseline to month 36 [time frame: baseline and month 36]; change in Z Score from baseline to month 36 [time frame: baseline, month 36]; change in bone mineral density from baseline to 12 months [time frame: baseline, 12 months]; number of participants with any kind of fractures, by visit [time frame: baseline, month 6, 12, 18, 24, 30 and 36]; median Disease‐Free Survival (DFS) [time frame: 36 months]; change in T‐Score from baseline to month 12 [time frame: baseline, month 12]; change in Z‐Score from baseline to month 12 [time frame: baseline, month 12].
Notes	Funding sources: Novartis Conflicts of interest: "There IS an agreement between Principal Investigators and the Sponsor (or its agents) that restricts the PI's rights to discuss or publish trial results after the trial is completed."
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Allocation: Randomized"
Allocation concealment (selection bias)	Unclear risk	No further information
Blinding of participants (performance bias)	High risk	"open‐label"
Blinding of personnel (performance bias)	High risk	"open‐label"
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	"open‐label"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of participants that withdrew from the study was pretty similar in both arms; ITT analysis was performed.
Selective reporting (reporting bias)	High risk	No participants were analysed for DFS although it was a prespecified outcome.
Other bias	High risk	"There IS an agreement between Principal Investigators and the Sponsor (or its agents) that restricts the PI's rights to discuss or publish trial results after the trial is completed."

FEMZONE 2014.

Study characteristics
Methods	Setting: multicentre, phase II Recruitment period: May 2010 to June 2010 Length of study: 6.5 months Length of follow‐up: 5 years
Participants	Eligibility criteria: postmenopausal women with primary invasive, histologically confirmed, ER‐ and/or PR‐positive breast cancer; clinical stage T1c (size ≥ 1.5 cm), T2, T3, T4a, b, c, N0 or N1, M0 (TNM classification). According to the modified RECIST criteria, tumours of size ≥ 1.5 cm are considered measurable by mammography and can be determined as target lesions; tumour measurable by mammography, sonography and clinical examination; adequate bone marrow, renal and hepatic function; good health status (ECOG performance status ≤ 2); adequate hepatic function; life expectancy of at least 12 months; 18 years or older; luteinising hormone and follicle‐stimulating hormone levels > 40 IU/L or oestradiol levels < 5 ng/dL. Exclusion criteria: prior treatment with letrozole or bisphosphonates; prior and concomitant anti‐breast‐cancer treatments such as chemotherapy, immunotherapy/biological response modifiers (BRMs), endocrine therapy other than letrozole (including steroids) and radiotherapy; patients who have received hormone replacement therapy (HRT) will NOT be excluded, provided that HRT is discontinued at least 2 weeks prior to entry into the study; patients with unstable angina, or uncontrolled cardiac disease or uncontrolled endocrine disorders; evidence of inflammatory breast cancer or distant metastasis; current active dental problems including infection of the teeth or jawbone (maxilla or mandibular); dental or fixture trauma, or a current or prior diagnosis of osteonecrosis of the jaw, of exposed bone in mouth, or of slow healing after dental procedures; recent (within 6 weeks) or planned dental or jaw surgery; history of diseases with an influence on bone metabolism, such as Paget's disease, osteogenesis imperfecta, and primary or secondary hyperthyroidism within the 12 months prior to study entry. Stage of disease: clinical stage T1c (size ≥ 1.5 cm), T2, T3, T4a, b, c TNM staging system: T1: 7.9% intervention 1 vs. 9.1% intervention 2 T2: 63.5% intervention 1 vs. 72.7% intervention 2 T3: 12.7% intervention 1 vs. 12.1% intervention 2 T4: 15.9% intervention 1 vs. 4.5% intervention 2 N0: 55.4% intervention 1 vs. 66.7% intervention 2 N1: 41.5% intervention 1 vs. 30.3% intervention 2 N2: 0% intervention 1 vs. 1.5% intervention 2 N3: 3.1% intervention 1 vs. 1.5% intervention 2 Mean age: intervention 1: 71.1 +/‐ 9.1 intervention 2: 70.6 +/‐ 8.3 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: ER+: intervention 1: 63 (96.9%) intervention 2: 64 (97.0 %) PR+: intervention 1: 59 (90.8 %) intervention 2: 59 (89.4 %) Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 168 Country of participants: Germany
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 4 weeks for 6 months; intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: 2.5 mg of letrozole per day surgery, radical mastectomy, modified radical mastectomy, lumpectomy/quadrantectomy
Outcomes	tumour response after 6 months of preoperative treatment in postmenopausal patients with primary breast cancer ("response" was defined as complete response (CR) or partial response (PR) based on MRI‐ or mammography and/or sonography according to modified 1 RECIST criteria); comparison of best response based on MRI‐ or mammography and/or sonography according to modified RECIST during 6 months of neoadjuvant treatment; comparison of the number of patients that qualify for breast‐conserving surgery rather than mastectomy; comparison of pathologic complete response (pCR); comparison of the change in tumour size; comparison of disease related‐death and overall survival; comparison of the "Quality of Life" using the FACT‐B questionnaire; description of the tumour response in subgroups of patients with polymorphism and special gene expression to compare the safety and tolerability of treatments.
Notes	Funding sources: "We are grateful to Novartis Germany for financial support for the trial and for supplying the medication." Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomized"
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	High risk	"open‐label"
Blinding of personnel (performance bias)	High risk	"open‐label"
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	"open‐label"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of patients that withdrew from the study was pretty similar in both arms; ITT analysis
Selective reporting (reporting bias)	Unclear risk	Study protocol was not accessible.
Other bias	Unclear risk	Study was prematurely terminated.

GAIN 2013.

Study characteristics
Methods	Setting: phase III Recruitment period: June 2004 to August 2008 Length of study: 2 years Length of follow‐up: median follow‐up of 38.7 months (range 0‐73)
Participants	Eligibility criteria: • histologically confirmed uni‐ or bilateral primary breast cancer; • adequate surgical treatment with histologic complete resection (R0) of the tumour and ≥ 10 resected axillary nodes with primary wound healing and no signs of infection; • stage of disease from pT1 to operable pT4a‐c with at least one pathologic involved axillary or internal mammary lymph node; • no evidence for distant metastasis after conventional diagnostic workup; • ECOG performance status < 2; • estimated life expectancy of at least 10 years, irrespective of the diagnosis of breast cancer. Exclusion criteria: • known hypersensitivity reaction to the compounds or incorporated substances or known dihydropyrimidine dehydrogenase deficiency; • inadequate organ function including eutrophils < 1.5 g/L, platelets < 100 g/L, aminotransferases, creatinine, or bilirubin > 1.25 x the upper limit of normal, alkaline phosphatase more than 3 x the upper limit of normal, creatinine clearance < 30 mL/min (if creatinine was above ULN); • insufficient or uncompensated cardiac function with left ventricular ejection fraction below the normal range of the institution; • history of severe heart disease, myocardial infarction within the last 6 months; • significant cardiac arrhythmias; • evidence for infection including wound infections and chronic infections; • secondary malignancy; • time since axillary dissection > 3 months (optimal < 1 month); • previously (neoadjuvant or adjuvant) treated invasive breast carcinoma; • previous or concurrent antitumour treatment for any reason; • simultaneous therapy with sorivudine or brivudine as virostatic; • immunosuppressive treatment or concurrent treatment with aminoglycosides; • pregnancy or lactation period; • no adequate nonhormonal contraception in premenopausal patients; concurrent treatment with other experimental drugs; • participation in another clinical trial with any investigational not‐marketed drug within 30 days before study entry. Stage of disease: intervention 1: pT1: 635 (31.9%) pT2: 1112 ((55.9%) pT3: 202 (10.2%) pT4: 41 (2.1%) TNM staging system: intervention 1: pN0: 0 pN1: 761 (38.1%) pN2: 696 (34.9%) pN3: 539 (27%) pT1: 635 (31.9%) pT2: 1112 (55.9%) pT3: 202 (10.2%) pT4: 41 (2.1%) intervention 2: pN0: 0 pN1: 370 (37.1%) pN2:362 (36.3%) pN3: 266 (26.7%) pT1: 320 (32.2%) pT2: 557 (56%) pT3: 103 (10.4%) pT4: 14 (1.4%) Mean age: intervention 1: < 40: 14.3% 40‐49: 36.0% 50‐59:32.4% ≥ 60: 17.3% median age: 49 intervention 2: < 40: 15.3% 40‐49:33.9% 50‐59: 32.2% ≥ 60: 18.6% median age: 50 Menopausal status: 48.4% (intervention 1) vs. 47.2% (intervention 2) pre or perimenopausal; 51.6% (intervention 1) vs. 52.8% (intervention 2) postmenopausal RANKL status: NR Hormone receptor status: 76.5% (intervention 1) vs. 77.7% (intervention 2) HR positive Human epidermal growth factor receptor 2 status: 22.1% (intervention 1) vs. 21.8% (intervention 2) HER2‐positive Participants randomised: total: 2994 Country of participants: Germany
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: ibandronate (50 mg, oral, daily, 2 years); intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy (tamoxifen + LHRH; 104 patients intervention 1 and 57 patients intervention 2), tamoxifen alone (695 patients intervention 1 and 350 patients intervention 2), tamoxifen + aromatase inhibitor (520 patients intervention 1 and 242 patients intervention 2), chemotherapy, radiotherapy
Outcomes	A: to compare the disease‐free survival after adjuvant chemotherapy with "ETC" (arm A1) or "EC‐TX" (arm A2) in patients with primary node‐positive breast cancer; B: to compare the disease‐free survival with (arm B1) or without ibandronate (arm B2) treatment for 2 years in patients with primary node‐positive breast cancer; to compare overall survival between arms A1 vs. A2 and B1 vs. B2; to evaluate the compliance in arms A1 vs. A2 and in B1; to compare the safety between arms A1 vs. A2 and B1 vs. B2; to assess the rate of responders to erythropoesis‐stimulating factors in arm A1 and A2; to compare the incidence of secondary primaries between arms A1 vs. A2; to compare the event‐free survival in subgroups of hormone sensitive and insensitive disease and in groups with 1‐3, 4‐9 or 10+ involved nodes between arms A1 vs. A2 and B1 vs. B2. Tertiary objectives: to determine prognostic factors like TS or TP and others on tumour tissue collected from primary surgery and to correlate them with study treatment effect.
Notes	Funding sources: Research Funding: Gunter von Minckwitz, Roche, Amgen, Novartis; Volker Mo¨bus, Amgen, Novartis, Roche, Johnson & Johnson; Volkmar Mu¨ller, Roche Conflicts of interest: Schneeweiß: Amgen Huober: Roche, Novartis, Amgen Jackisch: Amgen Diel: Amgen, Novartis, Roche Müller: Roche, Amgen Lück: Roche, GlaxoSmithKline, Eisai, Novartis Harbeck: Roche Loibl: Roche von Minckwitz: Roche, Amgen, Novartis Möbus: Amgen, GlaxoSmithKline, Pfizer, Roche, Celgene, Novartis, Johnson & Johnson Thomssen: Roche, Amgen Bauerfeind: Bristol‐Myers Squibb, Amgen Schmidt: Roche Clemes: Roche
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated permutated block randomisation
Allocation concealment (selection bias)	Low risk	Eligibility was centrally confirmed; computer‐generated permutated block randomisation
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Modified ITT analysis; very small number of participants did not commence treatment and were excluded from ITT analysis.
Selective reporting (reporting bias)	Low risk	All prespecified outcomes of ibandronate analysis reported
Other bias	Low risk	Study appeared to be free of other sources of bias.

GeparX 2016.

Study characteristics
Methods	Überdenken, ob so extrahieren Setting: multicentre, phase IIb, 2 x 2 randomised, open‐label Recruitment period: Length of study: 6 months Length of follow‐up: not reported/planned
Participants	Eligibility criteria: unilateral or bilateral primary carcinoma of the breast, confirmed histologically by core biopsy. Fine‐needle aspiration from the breast lesion alone is not sufficient. Incisional biopsy or axillary clearance is not allowed. In case of bilateral cancer, the investigator has to decide prospectively which side will be evaluated for the primary endpoint. tumour lesion in the breast with a palpable size of 2 cm or a sonographical size of 1 cm in maximum diameter. The lesion has to be measurable in two dimensions, preferably by sonography. If a tumour isn't measurable by sonography, then MRI or mammography is sufficient. In the case of inflammatory disease, the extent of inflammation can be used as a measurable lesion. patients must have stage cT1c ‐ cT4a‐d disease. inpatients with multifocal or multicentric breast cancer, the largest lesion should be measured. centrally confirmed ER‐negative and PR‐negative status. Central pathology also includes assessment of HER2, Ki‐67, TIL and RANK status on core biopsy. ER/PR negative is defined as ≤ 1% stained cells and HER2‐positive is defined as IHC 3+ or in‐situ hybridisation (ISH) according to ASCO‐CAP guidelines as of 2013. LPBC (lymphocyte predominant breast cancer) is defined as more than 50% stromal tumour infiltrating lymphocytes. Formalin‐fixed, paraffin‐embedded (FFPE) breast tissue from core biopsy therefore has to be sent to the GBG central pathology laboratory prior to randomisation. Exclusion criteria: patients with stages cT1a, cT1b, or any M1; prior chemotherapy for any malignancy; prior radiation therapy for breast cancer; History of disease with an influence on bone metabolism, such as osteoporosis, Paget's disease of bone, primary hyperparathyroidism requiring treatment at the time of randomisation or considered likely to become necessary within the subsequent six months; use of bisphosphonates or denosumab within the past 1 year; significant dental/oral disease, including prior history or current evidence of osteonecrosis/osteomyelitis of the jaw, active dental or jaw condition which requires oral surgery, non‐healed dental/oral surgery, planned invasive dental procedure for the course of the study; previous malignant diseases that have been disease‐free for less than 5 years (except CIS of the cervix and non‐melanomatous skin cancer); known or suspected congestive heart failure (> NYHA I) and/or coronary heart disease, angina pectoris requiring anti‐anginal medication, previous history of myocardial infarction, evidence of transmural infarction on ECG, uncontrolled or poorly controlled arterial hypertension (i.e. BP > 140/90 mm Hg under treatment with two antihypertensive drugs), rhythm abnormalities requiring permanent treatment, clinically significant valvular heart disease; currently, active infection; incomplete wound healing; definite contraindications for the use of corticosteroids. TNM staging system: N1: 23%, intervention 1 vs. 18.7%, intervention 2 T1: 135, intervention 1 vs. 156 intervention 2 T2: 22 intervention 1 vs. 213, intervention 2 T3: 18, intervention 1 vs. 11, intervention 2 T4: 11, intervention 1 vs. 6, intervention 2 Mean age: median 49 years intervention: median 49 years control: median 48.5 years Menopausal status: pre and perimenopausal: 55.9%, intervention 1 vs. 60.3%, intervention 2 postmenopausal: 43.8%, intervention 1 vs. 39.7%, intervention 2 RANKL status: not reported Hormone receptor status: HR+ 39.2% Human epidermal growth factor receptor 2 status: 19.7% (denosumab) vs. 19.5% (placebo) HER2 positive Participants randomised: total: 780 Country of participants: Germany
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: denosumab 120 mg every 4 weeks for 6 cycles; two different nab‐paclitaxel schedules; 2 x 2 factorial design; intervention 2: control, unclear (awaiting classification); supplemental: only in denosumab arm vitamin D (400 IU), calcium (500 mg); other: not reported. Previous bone‐modifying interventions: not reported Cancer treatment during study period: anthracycline/taxane‐based neoadjuvant chemotherapy
Outcomes	Primary: pCR rates of neoadjuvant treatment with or without denosumab in addition to nab‐paclitaxel and EC [time frame: 24 weeks]; pCR (ypT0 ypN0) rates of nab‐paclitaxel weekly for 12 weeks or 2 of 3 weeks for 12 weeks [time frame: 12 weeks]. Secondary: to test for interaction of denosumab treatment with RANK expression [time frame: 24 weeks]; to assess the pCR rates per arm for both randomisations separately for TNBC and HR‐/HER2+ tumours [time frame: 24 weeks].
Notes	Funding sources: not reported Conflicts of interest: Blohmer reported personal fees from Amgen during the conduct of the study; and personal fees from AstraZeneca, Merck Sharp & Dohme (MSD), Roche, and Seagen outside the submitted work. Link reported personal fees from Novartis, Pfizer, Roche, Tesaro, MSD, Amgen, Clovis, Lilly, Myriad, Eisai, GSK, and Gilead and nonfinancial support from MSD, Celgene, Clovis, and Daiichi Sankyo outside the submitted work. Reinisch reported personal fees from Roche, Lilly, AstraZeneca, Daiichi Sankyo, Pfizer, Seagen, Somatex, and Novartis and travel from Novartis and Celgene outside the submitted work. Untch reported personal fees (advisor; all fees to employer) from AbbVie, Eisai, and Seagen and personal fees (advisor, speaker; all fees to employer) from Amgen, AstraZeneca, Celgene, Daiichi Sankyo, Gilead, GlaxoSmithKline (GSK), Lilly, MSD Merck, Mylan, Myriad, Novartis, Pierre Fabre, Pfizer, Sanofi Aventis, and Roche outside the submitted work. Fasching reported grants from BioNTech and Cepheid and personal fees from Novartis, Pfizer, Daiichi Sankyo, AstraZeneca, Eisai, MSD, Lilly, Pierre Fabre, Seagen, Roche, Hexal, Agendia, Gilead, and Sanofi Aventis during the conduct of the study. Schneeweiss reported personal fees (honoraria) from Amgen, AstraZeneca, Celgene, Gilead, GSK, Lilly, MSD, Novartis, Pfizer, Pierre Fabre, Roche, Seagen, Teva, and Tesaro, and travel support from Celgene, Pfizer, and Roche outside the submitted work. Wimberger reported personal fees from Amgen, AstraZeneca, MSD, Novartis, Pfizer, Lilly, Roche, TEVA, Eisai, Clovis, and GSK outside the submitted work. Seiler reported travel costs from Novartis and personal fees from Roche, Mundipharma, Amgen, and AbbVie outside the submitted work. Huober reported grants from Hexal; grants and personal fees from Lilly; grants, personal fees, and travel expenses from Celgene; personal fees from Novartis, MSD, AstraZeneca, Gilead, Seagen, and Eisai; and personal fees and travel expenses from Roche, Pfizer, and Daiichi Sankyo outside the submitted work. Thill reported personal fees and nonfinancial support from Amgen; personal fees from AstraZeneca, Celgene, Pfizer, Roche, Hexal, Organon, Viatris, Vifor, Servier, Exact Sciences, Sysmex, Clearcut, PFM Medical, Daiichi Sankyo, Eisai, Gilead, Lilly, MSD, Novartis, Pierre Fabre, Clovis, and Seagen; and trial funding from Endomag and Exact Sciences outside the submitted work. Jackisch reported personal fees from Roche, AstraZeneca, Pfizer, and Amgen during the conduct of the study. Hanusch reported personal fees from Roche, Novartis, Pfizer, Celgene, Lilly, and AstraZeneca outside the submitted work. Denkert reported grants from the German Breast Group during the conduct of the study; personal fees from Novartis, Roche, MSD Oncology, Daiichi Sankyo, Molecular Health, AstraZeneca, Merck, and Lilly; grants from Myriad Genetics and Roche; and other (cofounder) from Sividon outside the submitted work; in addition, Denkert had a patent for Company: VMscope digital pathology software with royalties paid. Loibl reported honorarium for advisory boards (paid to institute) from AbbVie, Amgen, Bayer, Celgene, EirGenix, GSK, Lilly, Merck, Puma, Seagen; honorarium for advisory boards and lectures from AstraZeneca, Daiichi Sankyo, Novartis, Pierre Fabre, Prime/Medscape, and Pfizer; nonfinancial support and honorarium for advisory boards (paid to institute) from Bristol Myers Squibb; personal fees (lecture) from Chugai; other (paid to institute) from Ipsen; grants (honorarium for advisory boards and lectures from Roche; honorarium (lecture, paid to institute) from Samsung; nonfinancial support (paid to institute) from Vifor; and nonfinancial support (paid to institute/medical writing) from Gilead outside the submitted work; in addition, Loibl had a patent for EP14153692.0 pending, Immunsignature in TNBC; a patent for EP21152186.9 pending, CDK 4/6 Inhibitor Therapy; a patent for EP15702464.7 pending, predicting response to an anti‐HER2– containing therapy; and a patent for VMscope with royalties paid to VMscope GmbH. No other disclosures were reported.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Patients were first randomized centrally in a 1:1 ratio to either denosumab or no denosumab in addition to neoadjuvant chemotherapy (NACT), stratified by LPBC, BC subtype, and epirubicin and cyclophosphamide (EC) schedule."
Allocation concealment (selection bias)	Low risk	"patients were first randomized centrally [see above]"
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Intention‐to‐treat population
Selective reporting (reporting bias)	Unclear risk	Overall survival and disease‐free survival are not yet reported, but should be reported in the future.
Other bias	Low risk	None identified

H‐FEAT 2011.

Study characteristics
Methods	Setting : randomised, not blinded, phase II Recruitment period: 4 years (2011‐2015) Length of study: 12 months Length of follow‐up: not reported
Participants	Eligibility criteria: female, no age limits; postmenopausal women with hormone receptor‐positive early invasive breast cancer; patient who completed surgery, chemotherapy and radiotherapy as primary therapy within recent 12 weeks; eligible for letrozole as adjuvant therapy; T‐score is greater than ‐2.5 in the lumbar spine (L1‐L4); patient who has no fracture in the lumbar spine and hip; patient who is not administered any medication that alters bone mineral density; proved to be eligible for administration of risedronate after consultation of dentistry; adequate general condition; ECOG PS 0‐2; provided written Informed consent. Exclusion criteria: metastatic breast cancer; patient who has lumbar spine or hip fracture; previous history of endocrine therapy for breast cancer within 12 months; patient who is not in suitable condition for measurement of bone mineral density; patient who is planning to undergo dentistry surgery within 6 weeks; contraindication to study drugs. Mean age: not reported intervention: control: Menopausal status: 100% postmenopausal RANKL status: not reported Hormone receptor status: 100% HR+ Human epidermal growth factor receptor 2 status: not reported Participants randomised: not reported total: 103 Country of participants: Japan
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention: risedronate 17.5 mg/week for 12 months; control: no treatment; supplemental: letrozole 2.5mg/day; other. Previous bone‐modifying interventions: not reported Cancer treatment during study period: not reported
Outcomes	Primary: rate of change from baseline to 12 months in lumbar spine (L1‐L4) bone mineral density Secondary: rate of change from baseline to 6 months in lumbar spine (L1‐L4) bone mineral density; rate of change from baseline to 12 months in total hip bone mineral density; rate of change from baseline to 12 months in bone turnover markers.
Notes	Funding sources: Research Institute for Radiation Biology and Medicine, Surgical Oncology Department Conflicts of interest: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomised; institution was considered as a block; central registration
Allocation concealment (selection bias)	Low risk	Central registration
Blinding of participants (performance bias)	High risk	Open‐label ‐ no one was blinded.
Blinding of personnel (performance bias)	High risk	Open‐label ‐ no one was blinded.
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label ‐ no one was blinded.
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Only abstract available
Selective reporting (reporting bias)	Unclear risk	Only abstract available
Other bias	Unclear risk	Only abstract available

Hershman 2008.

Study characteristics
Methods	Setting: phase II, multicentre Recruitment period: NR Length of study: 1 year Length of follow‐up: 1 year
Participants	Eligibility criteria: • diagnosis of localised breast cancer, stage I or II (T1‐3, N0‐2, M0); • planned adjuvant chemotherapy (after surgery) of at least 6 months in duration; • hormone receptor status: oestrogen receptor and progesterone receptor status known; • age 18 to 50; • female; • premenopausal or perimenopausal; • creatinine less than 2 mg/dL; • at least 1 month since prior calcitonin; • at least 12 months since prior bisphosphonates given for more than 1 month duration; • no concurrent fluoride therapy (10 mg/day or more); • no concurrent enrolment in another experimental drug study. Exclusion criteria: • T score of less than 2.0 on bone mineral density (BMD); • fragility fracture; • lumbar spine anatomy that would preclude accurate BMD measurement of a minimum of 3 lumbar vertebrae; • pregnancy. Stage of disease: intervention 1: I: 13 (29 %) II: 30 (67 %) III: 2 (4 %) intervention 2 I: 19 (38 %) II: 27 (54 %) III: 4 (8 %) TNM staging system: T 1‐3 N0‐2 M0 Mean age: intervention 1: 43 intervention 2: 42 Menopausal status: 100% premenopausal RANKL status: NR Hormone receptor status: intervention 1: 37 (69 %) HR+ intervention 2: 37 (62 %) HR+ Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 114 Country of participants: USA
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 3 months for 12 months; intervention 2: placebo; supplemental: daily calcium and vitamin D supplements; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: 75 % (intervention 1) and 63 % (intervention 2) received tamoxifen therapy 25 % (intervention 1) and 27 % (intervention 2) received an aromatase inhibitor 4 cycles of chemotherapy: AC (doxorubicin + cyclophosphamide): 19% intervention 1, 19% intervention 2 T (paclitaxel): 4% intervention 1, 2% intervention 2 6‐8 cycles: ACT (doxorubicin + cycophosphamide + paclitaxel): 65% intervention 1, 70% intervention 2 CMF (cyclophosphamide + methotrexate + fluorouracil): 8% intervention 1, 5% intervention 2 CAF (cyclophosphamide + doxorubicin + fluorouracil): 10% intervention 1, 7% intervention 2
Outcomes	• per cent change in lumbar spine BMD at 6 months; • per cent change at any BMD site and markers of bone turnover at 12 months; • percentage change in BMD and bone turnover markers at 12 and 24 months (1 yr after last infusion).
Notes	Funding sources: Dawn L. Hershman, Novartis Pharmaceuticals; Elizabeth Shane, Novartis Pharmaceuticals; NIH Conflicts of interest: Hershman: Novartis Pharmaceuticals Shane: Novartis Pharmaceuticals
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"A separate restricted random assignment list was prepared for each stratum at each site, using random permuted blocks."
Allocation concealment (selection bias)	Low risk	"A separate restricted random assignment list was prepared for each stratum at each site, using random permuted blocks."
Blinding of participants (performance bias)	Low risk	"double‐blind"
Blinding of personnel (performance bias)	Low risk	"double‐blind"
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	"double‐blind"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT analysis
Selective reporting (reporting bias)	High risk	Recurrence was not actually an endpoint but was mentioned; study protocol stated, that QoL was measured but outcome was not reported in results.
Other bias	Low risk	Study appeared to be free of other sources of bias.

HOBOE 2019.

Study characteristics
Methods	Setting: phase III, multicentre Recruitment period: 2004‐2015 2004‐2009 post and premenopausal; 2009‐2015 only premenopausal Length of study: planned 5 years Length of follow‐up: median follow‐up of 64 months
Participants	Eligibility criteria: histological diagnosis of breast cancer; early stage breast cancer; surgical resection of breast cancer (breast conserving surgery or mastectomy); no evidence of disease; indication for adjuvant hormonal therapy (ER and/or PgR positive with immunohistochemistry exam in at least 1% of primary tumour cells, according to St. Gallen criteria); patient age at least 18 years; written informed consent; premenopausal status defined as last menstrual period within 12 months of randomisation (follicle‐stimulating hormone, luteinising hormone will not be considered as determinants of menopausal status due to chemotherapy induced reversible ovarian suppression); please note that patients who have received neoadjuvant or adjuvant chemotherapy and/or locoregional radiation therapy may be included in the study. Exclusion criteria: performance status (ECOG) > 2; previous or concomitant malignancy (with the exception of adequately treated non‐malignant skin cancer and carcinoma in situ of the uterine cervix; metastatic breast cancer; creatinine > 1.25 times the value of the upper normal limit; pregnant or lactating females; clinical or radiologic evidence of bone fractures; treatment with systemic cortisone therapy within 12 months prior to randomisation; treatment with drugs that could alter bone metabolism (calcitonin, mithramycin, gallium nitrate) within 2 weeks prior to randomisation; previous treatment with tamoxifen or aromatase inhibitors; aspartate transaminase and/or alanine transaminase > 3 times the value of the upper normal limit with clinical and laboratory findings that indicate a grade of hepatic insufficiency that could potentially increase the risk of assuming letrozole; any concomitant conditions that would, in the Investigator's opinion, contraindicate the use of any drugs used in this study; inability to provide informed consent; inability to comply with follow‐up; patient undergoing invasive dental work at time of baseline evaluation or foreseen during the course of adjuvant therapy. Stage of disease: early stage breast cancer TNM staging system: intervention 1: pT1: 67.3 % pT2: 26.8 % pT3: 2.8 % pT4: 0.6 % unknown: 2.5 % pN0: 54.6 % pN1: 31.0 % pN2: 9.9 % pN3: 4.5 % intervention 2: pT1: 67.1 % pT2: 27.8 % pT3: 2.8 % pT4: 0.8 % unknown: 1.4 % pN0: 55.1 % pN1: 30.6 % pN2: 10.7 % pN3: 3.7 % Mean age: intervention 1: median age of 45.2 intervention 2: median age of 44.9 Menopausal status: only long‐time follow‐up outcomes (at 5 years); only premenopausal included; earlier outcomes (especially bone health mixed) RANKL status: NR Hormone receptor status: PR+ intervention 1: 97.5 % intervention 2: 95.8 % Human epidermal growth factor receptor 2 status: intervention 1: 13.2 % HER2‐positive intervention 2: 11.8 % HER2‐positive Participants randomised: total: 1294 Country of participants: Italy
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 6 months for 5 years; intervention 2: no treatment; supplemental: no; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: letrozole 2.5 mg/day for 5 years; ovarian function suppression (OFS) with triptorelin: 3.75 mg at the start of treatment and then every 4 weeks, for 5 years or up to 55 years of age; previous adjuvant or neoadjuvant chemotherapy was allowed; radiotherapy was allowed; surgery before treatment; trastuzumab allowed in HER2+ patients
Outcomes	DFS in premenopausal patients 5 y BMD ‐ 1 y, subgroup menopausal status reported DFS postmenopausal patients OS ‐ 5 y mortality 5 y toxicity fracture (no cases 1 y) ONJ (no cases 1 y) distant metastases
Notes	Funding sources: National Cancer Institute, Naples, University of Campania "Luigi Vanvitelli" Conflicts of interest: F.P. reports nonfinancial support from Novartis, during the conduct of the study and personal fees from AstraZeneca, Bayer, Ipsen, Pierre Fabre, Incyte, Novartis, Celgene, Roche, BMS and Eli Lilly, outside the submitted work. M. De L. reports personal fees from Pfizer, Novartis, Roche, Celgene, AstraZeneca, Eisai and Eli Lilly, outside the submitted work. S. De P. reports personal fees from Pfizer, AstraZeneca and Novartis, during the conduct of the study and grants from Astra Zeneca, outside the submitted work. L.D.M. reports personal fees and non‐financial support from Roche and Pfizer and personal fees from Ipsen, Eli Lilly, Eisai, Novartis, Takeda and MSD, outside the submitted work. S.C. reports personal fees from Eli Lilly, outside the submitted work. M.C.P. reports personal fees from MSD, AstraZeneca, Bayer and Roche, outside the submitted work. N.N. reports grants, personal fees and non‐financial support from Merck Serono; grants, personal fees and non‐financial support from Thermofisher; personal fees from BMS, grants and personal fees from Qiagen; grants and personal fees from Roche; grants and personal fees from AstraZeneca; personal fees from Sanofi and personal fees from Boehringer Ingelheim, outside the submitted work. The other co‐authors have nothing to declare.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomisation was performed via a web‐based trial platform".
Allocation concealment (selection bias)	Low risk	"Independent Data Monitoring Committee"
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT analysis performed
Selective reporting (reporting bias)	Low risk	All predefined outcomes reported
Other bias	Low risk	Study appeared to be free of other sources of bias.

JONIE 2017.

Study characteristics
Methods	Setting: multicentre, phase II Recruitment period: March 2010 to June 2012 Length of study: 5 years Length of follow‐up: 8 years
Participants	Eligibility criteria: histologically proven invasive breast cancer of clinical stage IIA to IIIB (T ≥ 3.0 cm and node‐negative, or T ≥ 2.0 cm and cytologically or pathologically defined node‐positive); HER2‐negative (HER2 0 or 1+ on immunohistochemistry, or 2+ on immunohistochemistry and negative on fluorescence in situ hybridisation on zoledronic acid); 20 to 70 years of age; ECOG performance status of 0‐1; normal cardiac, renal and liver function; WBC more than 3000/mm3 and neutrophils more than 1500/mm3; haemoglobin more than 9 g/dL; platelet more than 100.000/mm³; AST and ALT less than 2.5 x upper limit of normal; serum creatinine less than 1.5 x upper limit of normal; left ventricular ejection fraction more than 50%. Exclusion criteria: bilateral breast cancer; inflammatory breast cancer; distant metastasis; history of chemotherapy, endocrine therapy, or radiotherapy for breast cancer; serious comorbidities such as heart failure, cardiac infarction, or serious disorders of infection; complicating dental or jaw infection or traumatic condition of teeth; history of treatment with a bisphosphonate within the previous 12 months; severe complication (infectious diseases, interstitial pneumonia, peripheral neuropathy, uncontrolled diabetes, bleeding tendency); those who are pregnant, potentially pregnant or breastfeeding; those known to have active Hepatitis B or C viral infection (HBs (+) or HCV (+)); patient judged inappropriate for this study by the physicians. Stage of disease: stage IIA to IIIB (T ≥ 3.0 cm and node negative, or T ≥ 2.0 cm and cytologically or pathologically defined node‐positive) TNM staging system: T ≥ 3 cm and T ≥ 2 cm node‐negative and node‐positive Mean age: intervention 1: 49.5 (range 34‐71) intervention 2: 49 (range 28‐70) Menopausal status: premenopausal: 56.8 % in intervention 1 57.6 % in intervention 2 postmenopausal: 43.2 % in intervention 1 42.4 % in intervention 2 RANKL status: NR Hormone receptor status: ER+: 80.7 % in intervention 1 81.5 % in intervention 2 PR+: 75 % in intervention 1 75 % in intervention 2 Human epidermal growth factor receptor 2 status: all patients were HER2‐negative Participants randomised: total: 180 Country of participants: Japan
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 3‐4 weeks for 6 doses, then every 3 months for 8 doses, then every 6 months for 5 doses for 5 years in total; intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy was prescribed for hormone receptor‐positive patients for at least 5 years chemotherapy: FEC100 (500/100/500 mg/m²) every 21 days for 4 cycles followed by weekly paclitaxel80 for 12 cycles (or weekly paclitaxel80 followed by FEC100) surgery after surgery, radiotherapy was mandatory in patients with breast‐conserving surgery; postoperative radiotherapy was performed according to the guidelines.
Outcomes	pathological complete response rate: pCR; clinical response rate; breast‐conserving rate; disease‐free survival; adverse events.
Notes	Funding sources: self‐funding Conflicts of interest: "The authors report no proprietary or commercial interest for any product or concept discussed in this article."
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomization was centralized at the data center of the JONIE trial group (Niigata University, Japan) and performed using the minimization method".
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of participants that withdrew from the study almost equal in both arms; ITT analysis was performed.
Selective reporting (reporting bias)	Low risk	Study protocol was available and all prespecified outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

Kanis 1996.

Study characteristics
Methods	Setting: multicentre, phase II Recruitment period: NR Length of study: 3 years Length of follow‐up: NR
Participants	Eligibility criteria: histologically proven breast cancer and recurrent disease diagnosed according to standard criteria Exclusion criteria: no information given Stage of disease: NR TNM staging system: NR Mean age: intervention 1: 59 intervention 2: 58 Menopausal status: premenopausal: 13% intervention 1 18% intervention 2 postmenopausal 87% intervention 1 82% intervention 2 RANKL status: NR Hormone receptor status: intervention 1: 65% ER+ intervention 2: 60% ER+ Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 133 Country of participants: UK, Canada
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 1600 mg of oral clodronate per day for 3 years; intervention 2: daily or placebo for 3 years; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy 37 patients in arm 1 and 24 patients in arm 2 chemotherapy 24 patients in arm 1 and 33 patients in arm 2 radiotherapy 1 patient in arm 2
Outcomes	effects of clodronate on the incidence and number of skeletal metastases; hypercalcaemia; bone pain; fractures; survival.
Notes	Funding sources: "The study was supported by Leiras Oy and Boehringer Mannheim GmbH" Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"The randomization was controlled at an independent center, using a prerandomization numbering system."
Allocation concealment (selection bias)	Low risk	"The randomization was controlled at an independent center, using a prerandomization numbering system."
Blinding of participants (performance bias)	Low risk	Double‐blind with an identical placebo
Blinding of personnel (performance bias)	Low risk	"All clinicians involved in the study, pharmacy staff and other personnel at each center were unaware of the treatment allocation."
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	"All clinicians involved in the study, pharmacy staff and other personnel at each center were unaware of the treatment allocation."
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	All randomised participants were analysed.
Selective reporting (reporting bias)	Low risk	All prespecified endpoints were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

Kristensen 2008.

Study characteristics
Methods	Setting: multicentre, phase III Recruitment period: January 1990 to January 1996 Length of study: 4 years Length of follow‐up: 10 years
Participants	Eligibility criteria: women with resectable adenocarcinoma of the breast and without signs of distant metastases; premenopausal women without lymph node metastases but with grade 2 or 3 malignancy and a primary tumour 55 cm in diameter independent of hormone receptor status; premenopausal women with negative or unknown hormone receptor status and with either axillary lymph node metastases or a primary tumour 5 cm in diameter; postmenopausal women with hormone receptor negative tumours and with either axillary lymph node metastases or a primary tumour 5 cm in diameter. Exclusion criteria: NR Stage of disease: 24% without lymph node activity; tumour sizes mostly 0‐50mm; malignancy grade 1‐3 TNM staging system: 24% without lymph node activity (= N0): 39% intervention 1 vs. 41% intervention 2 N1‐3: 32% (intervention 1 vs. 30% intervention 2 ≥ N4 primary tumour size: 0‐20 mm: 41% intervention 1 vs. 44% intervention 2 21‐50 mm: 50% both groups > 50 mm: 7% intervention 1 vs. 5% intervention 2 Mean age: intervention 1: > 39 years: 16% 40‐49 years: 45% 50‐59 years: 23% 60‐69 years: 15% intervention 2: > 39 years: 15% 40‐49 years: 48% 50‐59 years: 23% 60‐69 years: 14% Menopausal status: premenopausal: 67% intervention 1 66% intervention 2 postmenopausal: 33% intervention 1 34% intervention 2 RANKL status: NR Hormone receptor status: ER+: 13% intervention 1 17% intervention 2 PR+: 11% intervention 1 11% intervention 2 Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 953 Country of participants: Denmark, Sweden, Iceland
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 2 x 150 mg oral pamidronate per day for 4 years; intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: chemotherapy cyclophosphamide, methotrexate, and 5‐fluouracil (CMF) or cyclophosphamide, epirubicin, and 5‐fluouracil (CEF) loco‐regional radiotherapy was given according to guidelines at the participating centres
Outcomes	bone metastases; fractures; AEs; survival; BMD (in little subgroup).
Notes	Funding sources: "Supported in part by grants‐in‐aid from Pharmacia (now Pfizer) and Ciba‐Giegy (now Novartis)" Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No information given
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	High risk	"open‐label trial"
Blinding of personnel (performance bias)	High risk	"open‐label trial"
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	"open‐label trial"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	"intention‐to‐treat principle and a per‐protocol analysis were done".
Selective reporting (reporting bias)	Unclear risk	Study protocol not accessible
Other bias	High risk	Participants were not allowed to be on endocrine therapy. However, 17% of participants in control arm versus 13% in pamidronate arm were ER‐positive. This may potentially bias results against the control arm since these participants were not treated optimally.

Mardiak 2000.

Study characteristics
Methods	Setting: multicentre Recruitment period: 1990 to 1993 Length of study: 2 years Length of follow‐up: median follow‐up of 84 months
Participants	Eligibility criteria: women with BC with previously untreated locally advanced (pT ≥ 5 cm) disease or metastases but no bone or central nervous system metastases; age > 18 years; no serious functional disorders of the liver and kidney. Exclusion criteria: NR Stage of disease: intervention 1: stage III: 32 (87%) stage IV: 5 (13%) intervention 2: stage III: 34 (94%) stage IV: 2 (6%) TNM staging system: NR Mean age: intervention 1: median age of 55 (range: 29‐67) intervention 2: median age of 54 (range: 31‐79) Menopausal status: NR RANKL status: NR Hormone receptor status: NR Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 73 Country of participants: Slovak Republic
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 800 mg of oral clodronate twice per day for 2 years; intervention 2: oral placebo for 2 years; supplemental: NR; other: NR. Previous bone‐modifying interventions: Cancer treatment during study period: obligatory radiotherapy for patients in stage III chemotherapy SERM: 8 % (intervention 1), 19 % (intervention 2) unspecified endocrine therapy
Outcomes	overall survival; appearance of bone metastases; time to appearance of bone metastases.
Notes	Funding sources: Boehringer‐Mannheim Comp Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomised"
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	High risk	10/72 participants not evaluable because of "short duration of therapy"
Selective reporting (reporting bias)	Low risk	All prespecified outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

Monda 2017.

Study characteristics
Methods	Setting: prospective, randomised controlled trial Recruitment period: NR Length of study: 2 years Length of follow‐up: 2 years
Participants	Eligibility criteria: postmenopausal; hormone receptor‐positive breast cancer, completed primary surgery and chemotherapy; mild‐to‐moderate risk of fracture; advanced age, early menopause (age < 45 years); low body weight, current smoking habit. Exclusion criteria: menopause induced by prior chemotherapy or any other drug therapy; metastatic disease; recent hormonal treatment; previous hip fractures or protheses; known bone metabolism disorder; non‐treated hypocalcaemia or hypercalcaemia; previous treatment with selective oestrogen receptor modulators (SERMs); hormone‐replacement therapy (HRT) or BPs; liver or renal dysfunction. Stage of disease: NR TNM staging system: NR Mean age: intervention 1: 55.7 +/‐ 6.4 intervention 2: 56.1 +/‐ 6.3 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: all patients were hormone receptor‐positive Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 84 Country of participants: NR
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 35 mg of oral risedronate per week for 2 years; intervention 2: no treatment; supplemental: 800 IU of vitamin D and 1000 mg of calcium per day for 24 months each; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: anastrozole
Outcomes	changes in lumbar spine BMD at baseline and after 24 months; effect of risedronate on health‐related quality of life (HRQoL); vertebral‐fractures; AE ONJ.
Notes	Funding sources: "This study was supported by grants of Department of Biology, Universitá degli Studi di Napoli Federico II" Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"women were randomly assigned"
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	Unclear risk	No information given
Blinding of personnel (performance bias)	Unclear risk	No information given
Blinding of outcome assessment (detection bias) subjective outcomes	Unclear risk	No information given
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Not enough information given
Selective reporting (reporting bias)	Unclear risk	No study protocol
Other bias	Low risk	Study appeared to be free of other sources of bias.

N02C1 2009.

Study characteristics
Methods	Setting: phase III, multicentre Recruitment period: March 2003 to March 2006 Length of study: 1 year Length of follow‐up: 1 year
Participants	Eligibility criteria: premenopausal women scheduled to undergo adjuvant or neoadjuvant chemotherapy for primary breast cancer (stages I to IIIB); if a total abdominal hysterectomy had been performed, with at least one intact ovary, or if more than 3 months since the last menstrual period had lapsed, then patients must have had documented premenopausal oestrogen levels (≤ 1 month before study entry), understanding the limitation of this test; at least 18 years of age; ECOG performance status of 0 or 1. Exclusion criteria: hypercalcaemia, hypocalcaemia, an inability to stand or sit upright for at least 30 minutes, known swallowing disorder; BMD T‐score of ≤ ‐2.0 at the hip or LS; history of vertebral compression fracture; corticosteroid use at doses more than 5 mg/d of prednisone or equivalent for more than 2 weeks in the prior 6 months; previous treatment with bisphosphonates; disease‐affecting bone metabolism (hyperthyroidism, hyperparathyroidism, and hypercortisolism); serum creatinine more than 2.0; malabsorption syndrome, menopausal oestrogen therapy, oral contraceptive use, bilateral oophorectomy, pregnancy, active nursing, of childbearing potential, unwilling to employ adequate contraception, and having undergone dental extraction, root canal, or dental implants ≤ 3 months before registration. Stage of disease: stage I to IIIB TNM staging system: NR Mean age: intervention 1: 43.3 (SD: 5.36) intervention 2: 43.6 (SD: 6.09) Menopausal status: 100% premenopausal RANKL status: NR Hormone receptor status: NR Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 216 Country of participants: Canada, USA
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 35 mg of oral risedronate per week for 1 year; intervention 2: weekly oral placebo for 1 year; supplemental: 400 IU of oral vitamin D and 600 mg of oral calcium per day for 1 year; other. Previous bone‐modifying interventions: NR Cancer treatment during study period: tamoxifen was allowed during the study period Chemotherapy regimens included anthracyclines, taxanes, cyclophosphamide, methotrexate, or fluorouracil, with a mean duration of 146 days
Outcomes	average intra‐patient change in lumbar spine (L2‐L4, PA) bone mineral density (BMD) from baseline to one year after study entry [time frame: 1 year post study entry]; average intra‐patient change in femoral neck and total hip BMD from baseline to one year after study entry [time frame: 1 year post study entry]; incidence of osteopenia in the risedronate vs placebo groups at one year after study entry [time frame: 1 year post study entry]; incidence of osteoporosis in the risedronate vs placebo groups at one year after study entry [time frame: 1 year post study entry]; incidence of a 5% difference in intra‐patient BMD scores at baseline [time frame: baseline]; serum and urine N‐telopeptide and serum alkaline phosphatase at baseline and 6 months [time frame: up to 6 months]; frequency and severity of toxicity as measured by NCI CTC version 2.0 [time frame: up to 1 year post study treatment]; menopausal symptoms as measured by the Greene Climacteric Scale (GCS) at baseline, monthly during chemotherapy, at 6 months, 1 year, and 2 years after study entry [time frame: up to 2 years post study entry]; association of baseline serum oestradiol levels with permanent cessation of menses [time frame: baseline]; relationship between the subscales of the GCS (psychological, vasomotor, somatic and sexual) with type of chemotherapy, duration of chemotherapy, and menstrual cycle changes [time frame: up to 2 years post study entry].
Notes	Funding sources: Stephanie L. Hines, Aventis, Novartis; Charles L. Loprinzi, Aventis Study Sponsor: Alliance for Clinical Trials in Oncology "This study was funded by the NCI, with supplemental funding from Aventis." Conflicts of interest: Stephanie L. Hines, Aventis, Novartis; Charles L. Loprinzi, Aventis
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"were randomly assigned"
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	Low risk	"double‐blind"
Blinding of personnel (performance bias)	Low risk	"double‐blind"
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	"double‐blind"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of participants that withdrew from the study had fairly similar reasons for withdrawals.
Selective reporting (reporting bias)	Unclear risk	Not all outcomes listed in the protocol were reported in the study.
Other bias	Low risk	Study appeared to be free of other sources of bias.

NATAN 2016.

Study characteristics
Methods	Setting: phase III Recruitment period: between 01/2005 and 06/2009 Length of study: planned 5 years Length of follow‐up: median of 54.7 months
Participants	Eligibility criteria: ‐ complete baseline documentation sent to GBG; ‐ prior preoperative chemotherapy for at least 4 cycles, of which at least two must contain a taxane and an anthracycline; ‐ completely resected unilateral or bilateral primary carcinoma of the breast with histologically detectable tumour residuals (ypT1‐4) and/or histology confirmed involvement of axillary nodes (ypN1‐3). Sentinel node biopsy is allowed, but complete axillary clearance is mandatory in node‐positive cases; ‐ a maximum interval of 3 years from date of axillary surgery to entering this trial; ‐ Karnofsky index >= 70%; ‐ life expectancy of at least 10 years, disregarding the diagnosis of cancer; ‐ no clinical evidence of local recurrence or distant metastases; ‐ complete staging work‐up: All patients must have breast ultrasound, chest X‐ray, ultrasound or CT scan of the liver within 3 months prior to registration, as well as (bilateral) mammography or breast MRI and bone scan within 8 months prior to registration. In the case of a positive bone scan, a bone X‐ray is mandatory. Other tests may be performed as clinically indicated; ‐ adequate renal and hepatic function (serum creatinine, bilirubin, and transaminases within 1.5 × upper normal range); ‐ patients must be available and compliant for treatment and follow‐up. Patients registered on this trial must be treated and followed up at the participating centre. Exclusion criteria: ‐ known hypersensitivity reaction to the investigational compound; ‐ prior postoperative chemotherapy; ‐ prior treatment with bisphosphonates since breast cancer surgery; ‐ pregnant or lactating patients. Patients of childbearing potential must have a negative pregnancy test (urine or serum) within 14 days prior to registration and must implement adequate non‐hormonal contraceptive measures (barrier methods, intra‐uterine contraceptive devices, sterilisation) during study treatment ‐ history of diseases with influence on bone metabolism, such as Paget's disease of bone and primary hyperparathyroidism or osteoporosis requiring treatment at the time of study entry or considered likely to become necessary within the six months; ‐ other serious illnesses or medical conditions that may interfere with the understanding and giving of informed consent and the conduct of the study; ‐ prior or concomitant secondary malignancy (except non‐melanomatous skin cancer or carcinoma in situ of the uterine cervix); ‐ concurrent treatment with other experimental drugs or any other anti‐cancer therapy; ‐ abnormal renal function as evidenced by a calculated creatinine clearance of < 30 mL/minute; ‐ serum calcium concentration < 8.0 mg/dL (2.00 mmol/L) or > 12.0 mg/dL (3.00 mmol/L); ‐ concurrent treatment with sex hormones. Prior treatment must be stopped before study entry; ‐ current active dental problems including infection of the teeth or jawbone (maxilla or mandibular) dental or fixture trauma, or a current or prior diagnosis of osteonecrosis of the jaw, of exposed bone in the mouth, or of slow healing after dental procedures; ‐ recent (within 6 weeks) or planned dental or jaw surgery (e.g. extraction, implants). Stage of disease: early breast cancer with invasive residual disease TNM staging system: ypT0: 2.8% DCIS: 1.3% ypT1: 40‐6% ypT2: 39.9% ypT3: 11.2% ypT4: 4.3% ypN0: 28.1% ypN1: 44.8% ypN2: 20% ypN3: 7.1% Mean age: 66.4% < 55 y; 33.6% > 55 y intervention 1: 66.4% < 55 y; 33.6% > 55y intervention 2: 66% < 55 y; 34% > 55 y Menopausal status: 27.5% premenopausal 47.0% postmenopausal (difference changed from pre to postmenopausal during neoadjuvant chemotherapy) RANKL status: NR Hormone receptor status: ER+ and/or PgR+ 79.3% Human epidermal growth factor receptor 2 status: 17.4% HER2‐positive Participants randomised: total: 693 Country of participants: Germany, Austria
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 4 weeks for the first 6 months, every 3 months for the following 2 years, every 6 months for the last 2.5 years (5 years in total); intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: chemotherapy (anthracycline‐taxane‐containing neoadjuvant chemotherapy before surgery), radiotherapy, endocrine therapy, trastuzumab, surgery
Outcomes	‐ event‐free survival; ‐ OS; ‐ EFS with respect to the interval between surgery and randomisation; ‐ bone metastasis‐free survival; ‐ toxicity; ‐ compliance ; ‐ predictive value of primary breast tumour response on the effect of postoperative treatment; ‐ prognostic impact of chemotherapy‐induced amenorrhoea in premenopausal patients.
Notes	Funding sources: Novartis provided study medication as well as financial funding for the study but was not involved in the design, conduct or analysis and interpretation of the data. Conflicts of interest: Von Minckwitz G: institution received research grants from Roche and Novartis Tesch H: received honoraria from Roche and Novartis; has consultant/advisor role for Roche and Novartis Huober J: received honoraria from Novartis and Roche; has consultant/advisor role for Novartis and Amgen Dubsky P: received honoraria from Pfizer, AstraZeneca; has consultant/advisor role for Pfizer, Roche, Agendia, Sividon Blohmer J.U.: received honoraria from Roche, Amgen, Teva; has consultant/advisor role for Roche and Teva Hanusch C: has consultant/advisor role for Novartis Jackisch C: received honoraria from Amgen Ku¨mmel S: received honoraria from Roche, Celgene, Teva, Novartis; has consultant/advisor role for Roche; institution received research grants from Roche Fasching P: received honoraria from Amgen, Roche, Teva, Genomic Health, Novartis, Pfizer, GSK; has consultant/advisor role Roche, Pfizer, Genomic Health, Teva; institution received research grants from Novartis and Amgen Schneeweiss A: received honoraria from Roche and Celgene; has consultant/advisor role for Roche and Celgene Loibl S: received honoraria from Novartis, has consultant/advisor role for Novartis; institution received research grants from Roche, Novartis and SanofiAventis All other authors have declared no conflicts of interest.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Eligibility was centrally confirmed and block randomisation was used.
Allocation concealment (selection bias)	Low risk	Block randomisation
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Low loss to follow‐up; ITT analysis
Selective reporting (reporting bias)	Low risk	Outcomes listed in the prospectively registered trial were covered in the clinical trial report.
Other bias	Low risk	Study appeared to be free of other sources of bias.

NEO‐ZOTAC BOOG 2010.

Study characteristics
Methods	Setting: phase III, multicentre Recruitment period: July 2010 to April 2012 Length of study: 4.5 months Length of follow‐up: 60 months
Participants	Eligibility criteria: histologically confirmed breast cancer; large resectable or locally advanced disease T2 (≥ 2 cm and positive lymph nodes), T2 (≥ 3 cm), ≥ T3, T4, any N, M0 disease; measurable disease (breast and/or lymph nodes); HER2‐negative disease by core biopsy; no evidence of distant metastases (M1); no prior breast cancer; menopausal status unspecified; WHO performance status 0‐2; not pregnant or nursing; WBC ≥ 3.0 x 10^9/L; neutrophil count ≥ 1.5 x 10^9/L; platelet count ≥ 100 x 10^9/L; bilirubin ≤ 1.5 times upper limit of normal (UNL); ALT and/or AST ≤ 2.5 times UNL; alkaline phosphatase ≤ 5 times UNL; creatinine clearance ≥ 50 mL/min; accessible for treatment and follow‐up; no previous malignancy within the past 5 years except basal cell carcinoma of the skin or pre‐invasive carcinoma of the cervix; no peripheral neuropathy > grade 2 (of any cause); no other serious diseases including recent myocardial infarction, clinical signs of cardiac failure, or clinically significant arrhythmias; no poor dental health; no known hypersensitivity reaction to any of the components of the treatment; no medical or psychological condition that, in the opinion of the investigator, would not permit the patient to complete the study or sign meaningful informed consent; no prior breast surgery except for biopsy; no prior chemotherapy or radiotherapy; no prior bisphosphonates. Exclusion criteria: evidence of distant metastases (M1); history of breast cancer; prior breast surgery; prior chemo‐ or radiotherapy; previous malignancy within 5 years; prior BP usage; peripheral neuropathy > grade 2; current active dental problems. Stage of disease: stage II or III TNM staging system: T stage cT1: 0 (0 %) intervention 1, 2 (1.6%) intervention 2 cT2: 74 (60.2%) intervention 1, 69 (55.6%) intervention 2 cT3/T4+: 49 (39.8%) intervention 1, 53 (42.7%) intervention 2 N stage cN+: 68 (55.7%) intervention 1, 67 (54%) intervention 2 cN‐: 54 (44.3%) intervention 1, 57 (46%) intervention 2 Mean age: intervention 1: 49.5 (range: 34‐70) intervention 2: 48.9 (range: 29‐67) Menopausal status: premenopausal: intervention 1: 59%, intervention 2: 59.7% postmenopausal: intervention 1: 41%, intervention 2: 38.7% RANKL status: NR Hormone receptor status: ER+: intervention 1: 100 (82%), intervention 2: 103 (83.1%) PR+: intervention 1: 71 (58.2%), intervention 2: 83 (66.9%) Human epidermal growth factor receptor 2 status: HER2‐positive: 0% Participants randomised: total: 250 Country of participants: Netherlands
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 3 weeks for a total of 6 injections; intervention 2: no treatment; supplemental: 400 IU of vitamin D and 500 mg of calcium per day for the duration of the study; other. Previous bone‐modifying interventions: NR Cancer treatment during study period: TAC (docetaxel 75 mg/m2, adriamycin 50 mg/m2, and cyclophosphamide 500 mg/m2 IV) chemotherapy on day 1 followed by 6 mg SC granulocyte colony‐stimulating factor on day 2 once per cycle, every 21 days during six cycles
Outcomes	pCR (pathological complete response); disease‐free survival; overall survival; clinical response and treatment tolerability; heterogeneity of the ER/PR and HER2 measurements in core biopsy and the surgical specimen.
Notes	Funding sources: "This study was supported by grants from the Dutch Cancer Society (2010–4682), Amgen, Novartis and Sanofi. EudraCT number 2009‐016932‐11, NL30600.058.09." Conflicts of interest: The authors have declared no conflict of interest.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Patients were centrally randomized at the LUMC Datacenter of the Department of Surgery, through the online ALEA randomisation program. Randomization was done using Pocock’s minimizoledronic acidtion technique […]."
Allocation concealment (selection bias)	Low risk	"Patients were centrally randomized at the LUMC Datacenter of the Department of Surgery, through the online ALEA randomisation program. Randomization was done using Pocock’s minimizoledronic acidtion technique […]."
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of participants that withdrew from the study was small and equal in both arms; ITT‐analysis
Selective reporting (reporting bias)	Low risk	All predefined outcomes reported
Other bias	Low risk	Study appeared to be free of other sources of bias.

NEOZOL 2018.

Study characteristics
Methods	Setting: phase IIa, multicentre Recruitment period: April 2010 to October 2013 Length of study: 4.5 months Length of follow‐up: median follow‐up duration and the interval from chemotherapy to initiation of surgery was 5.7 and 5.4 months (arm A and arm B)
Participants	Eligibility criteria: women aged ≥ 18 years with newly diagnosed invasive breast cancer with a maximal diameter > 2 cm and Union of International Cancer Control stage IIa, IIb, and IIIa; absence of contraindication to treatment with Zometa: creatinine clearance greater than 30 mL/min (with Cockroft or MDRD method); ductal or lobular histological type of the breast tumour; WHO performance status 0‐2; patient who understands the French language; covered by, or having the right to Social Security; signed informed consent. Exclusion criteria: multifocal or multicentre tumours and inflammatory cancer; contraindication to ZOL; creatinine clearance < 30 mL/min using the Cockroft or Modification of Diet in Renal Disease method; women who are pregnant (positive pregnancy test) or breastfeeding, or absence of contraception in a woman who is able to become pregnant; patient with evolutionary dental problems, including dental infection or infection of the jaw, intrabuccal exposure of the jawbone, and history or current diagnosis of osteonecrosis of the jaw, requiring fast chirurgical care; noninvasive cancer; breast cancer of rare histological type (other than ductal and lobular); T4 breast tumour presence of organ, bone, or skin metastases (in the initial staging workup); history of breast cancer; other cancer currently in treatment (except carcinoma in situ); severe systemic disease potentially interfering with follow‐up; contraindication to injected products: known allergy to bisphosphonates, zoledronic acid or excipients, severe renal failure (creatinine clearance < 30 mL/min with Cockroft or MDRD method); prior treatment with bisphosphonates (either IV or oral); history of severe bone disease (severe osteoporosis with multiple skeletal‐related events). Stage of disease: stage II or III TNM staging system: IIa, IIb, IIIa (T2/T3) Mean age: intervention 1: 51.2 (range: 35‐68) intervention 2: 50.5 (range: 22‐72) Menopausal status: 22 women were postmenopausal 9 (37.5%) intervention 1 13 (50%) intervention 2 RANKL status: NR Hormone receptor status: ER+: 16 (66.7%) in intervention 1, 19 (73.1%) in intervention 2 PR+ : 11 (45.8%) in intervention 1, 11 (42.3%) in intervention 2 Human epidermal growth factor receptor 2 status: HER2‐positive: 2 (9.5%) in intervention 1, 8 (30.8%) in intervention 2 Participants randomised: total: 53 Country of participants: France
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 3 weeks for a total of 6 injections; intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: minimum of 6 courses (with a minimum of 3 cycles of anthracycline) but did not exceed 8 courses, interval between each cycle was 3 weeks 4 injections of doxorubicin (60 mg/m2) combined with cyclophosphamide (600 mg/m2) every 3 weeks, followed by 4 injections of docetaxel (100 mg/m2) every 3 weeks surgery
Outcomes	the primary endpoint was the evolution of serum VEGF (in pg/mL); secondary endpoints were the breast conservation rate, pathological complete response (pCR) after final surgery, change in the circulating tumour cell (CTC) levels from V0 to V final, and therapeutic complications.
Notes	Funding sources: The present study was supported in part by NOVARTIS Pharma S.A.S., which did not interfere with the production and reporting of the results. Conflicts of interest: The authors declare that they have no competing interest.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"the patients were randomly allocated [...] using an interactive web response system. The block randomisation method and stratification by the Scarff‐BloomRichardson grade were used to ensure the absence of a selection bias and to achieve balance in the allocation of the treatment arms".
Allocation concealment (selection bias)	Low risk	"the patients were randomly allocated [...] using an interactive web response system. The block randomisation method and stratification by the Scarff‐BloomRichardson grade were used to ensure the absence of a selection bias and to achieve balance in the allocation of the treatment arms".
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of participants that withdrew from the study after randomisation was very low.
Selective reporting (reporting bias)	Low risk	Study protocol available and prespecified outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

Novartis I 2006.

Study characteristics
Methods	Setting: multicentre, phase III, open‐label Recruitment period: NR Length of study: 3 years Length of follow‐up: 3 years
Participants	Eligibility criteria: compliant postmenopausal women with primary operable breast cancer after 4 to 6 years of therapy with tamoxifen (end of tamoxifen therapy within last 6 months); performance status 0‐2 (Eastern Cooperative Oncology Group); patients without severe osteoporosis at study entry; no evidence of relapse at the time of randomisation; adequate function of bone marrow, kidney, and liver. Exclusion criteria: oestrogen‐ and progesterone‐receptor status negative or unknown; completion of adjuvant tamoxifen therapy more than 6 months prior to study start; inflammatory breast cancer current/active dental problems including infection of the teeth or jawbone, dental or fixture trauma, or a current or prior diagnosis of osteonecrosis of the jaw, of exposed bone in the mouth, or of slow healing after dental procedures; recent (within 6 weeks) or planned dental or jaw surgery; history of diseases with an influence on bone metabolism such as Paget's disease and primary overactive parathyroid; prior or concomitant therapies: chemotherapy within the last 12 months, intravenous or oral bisphosphonates, systemic corticosteroids, anabolic steroids or growth hormones, Tibolone, parathyroid hormone, systemic sodium fluoride or any drugs known to affect the skeleton (such as calcitonin, mithramycin, or gallium nitrate); patients with previous or concomitant cancers (not breast cancer) within the past 5 years EXCEPT adequately treated basal or squamous cell skin cancers or in situ cancer of the cervix. Patients with other previous cancer(s) must have been disease‐free for at least 5 years; patients currently receiving oral bisphosphonates must discontinue these at least 3 weeks prior to study start. Stage of disease: NR TNM staging system: NR Mean age: intervention 1: 58.4 ± 7.3 intervention 2: 61.3 ± 7.3 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: NR Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 83 Country of participants: Germany
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: Zoledronic acid (4 mg, intravenously, every 6 months, 3 years); intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: Aromatase inhibitor (letrozole 2.5 mg/day for 3 years); nothing else reported
Outcomes	change in BMD from baseline to 36 months; per cent change in BMD from baseline to 36 months; change in T‐score from baseline to 36 months; change in Z‐score from baseline to 36 months; change in BMD from baseline to 12 months; number of participants with any kind of fracture from baseline to 6, 12, 18, 24, 30 and 36 months; median disease‐free survival; change in T‐score from baseline to 12 months; change in Z‐score from baseline to 12 months.
Notes	Funding sources: Novartis (sponsor) Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Randomised, no further information
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of participants that did not complete the study was fairly equal; ITT analysis
Selective reporting (reporting bias)	High risk	DFS was a prespecified outcome but data were not analysed/presented.
Other bias	Unclear risk	Study results were never published.

NSABP B‐34 2012.

Study characteristics
Methods	Setting: multicentre, placebo‐controlled, double‐blind, phase III Recruitment period: 2001‐2004 Length of study: 3 years Length of follow‐up: 90.7 months (median)
Participants	Eligibility criteria: total mastectomy or a lumpectomy with either an axillary dissection or sentinel node biopsy (If any sentinel node is histologically positive by H & E, or histologically suspicious on H & E and confirmed positive by immunohistochemistry, then the patient must have a completion axillary dissection); invasive adenocarcinoma on histologic examination with clinical assessment T1‐3, N0‐1, M0; not be participating in any other clinical trials; patients may participate in radiation therapy trials: (specified); analysis of both oestrogen and progesterone receptors on the primary tumour performed prior to randomisation (tumours will be defined as ER or PR positive if: 1) the Dextran‐coated charcoal or sucrose‐density gradient method shows them to have greater than or equal to 10 fmol/mg cytosol protein, or 2) if using individual laboratory criteria they can be shown to be positive by the enzyme immunoassay method or immunocytochemical assay. "Marginal or borderline," results (i.e. those not definitively negative) will also be considered positive; within the past 3 months: history and physical exam, a bone scan, thoracic and lumbar spine x‐rays, and a chest x‐ray; within the past 12 months: gynaecologic exam (for women who have a uterus and who will be taking tamoxifen) and a bilateral mammogram; postoperative absolute neutrophil count must be greater than or equal to 1500/mm3 (or less than 1500/mm3 if, in the opinion of the investigator, this represents an ethnic or racial variation of normal); postoperative platelet count must be greater than or equal to 100,000; postoperative evidence of adequate hepatic function, i.e. total bilirubin at or below the upper limit of normal (ULN) for the laboratory; and alkaline phosphatase less than 2.5 x the ULN; and the serum glutamate oxaloacetate transaminase/aspartate transaminase less than 1.5 x the ULN; postoperative evidence of adequate renal function (serum creatinine within or less than the laboratory's normal range); serum albumin and serum calcium must be within normal limits; patient with skeletal pain eligible for inclusion if bone scan and/or roentgenological examination fails to disclose metastatic disease (suspicious findings must be confirmed as benign by x‐ray, MRI, or biopsy); patients with prior non‐breast malignancies are eligible if they have been disease free for greater than or equal to 5 years and are deemed at low risk for recurrence by their treating physicians; patients with squamous or basal cell carcinoma of the skin that has been effectively treated, carcinoma in situ of the cervix that has been treated by surgery only, or lobular carcinoma in situ of the ipsilateral or contralateral breast treated by hormone therapy and/or surgery only are eligible, even if these were diagnosed within 5 years before randomisation; patients must have Zubrod performance status of 0, 1, or 2; special conditions for eligibility of lumpectomy patients: (specified); lumpectomy reserved for tumours less than 5 cm; margins of the resected specimen must be histologically free of invasive tumour and ductal carcinoma in situ; tumour present at the line of resection may need additional operative procedures to obtain clear margins (permissible even if axillary dissection has been performed). Exclusion criteria: significant non‐malignant bone disease that is likely to interfere with the interpretation of bone x‐rays; ulceration, erythema, infiltration of the skin or the underlying chest wall (complete fixation), peau d'orange, or skin oedema of any magnitude (specified); ipsilateral lymph nodes that on clinical examination are found to be fixed to one another or to other structures (cN2 disease); suspicious palpable nodes in the contralateral axilla or palpable supraclavicular or infraclavicular nodes, unless there is biopsy evidence that these are not involved with tumour; prior therapy for breast cancer, including irradiation, chemotherapy, biotherapy, and/or hormonal therapy, except for tamoxifen (tamoxifen may be given as adjuvant therapy before study entry, but only if it was started within 28 days before randomisation). Patients who started tamoxifen within 28 days before randomisation and who are being considered for chemotherapy must have their tamoxifen stopped at the start of chemotherapy); prior history of breast cancer, except lobular carcinoma in situ; any sex hormonal therapy, e.g. birth control pills, ovarian hormonal replacement therapy, etc. (these patients are eligible only if this therapy is discontinued prior to randomisation ‐ exceptions: patients may use low‐dose oestrogen vaginal creams or Estring® for symptomatic vaginal dryness, raloxifene (or other selective oestrogen receptor modulators) for the prevention of osteoporosis, and luteinising‐hormone‐releasing hormone agonists/antagonists for the purpose of medical ovarian ablation as a component of adjuvant therapy for the breast cancer); patients currently taking alendronate (Fosamax®) or other bisphosphonates or calcitonin to treat or prevent osteoporosis; non‐malignant systemic disease (cardiovascular, renal, hepatic, etc.) that would preclude a patient from being subjected to any of the treatment options or would prevent prolonged follow‐up; psychiatric or addictive disorders that would preclude obtaining informed consent; pregnancy or lactation at the time of proposed randomisation; bilateral malignancy or a mass or mammographic abnormality in the opposite breast suspicious for malignancy unless there is biopsy proof that the mass is not malignant; lumpectomy patients with irradiation and surgery; diffuse tumours (as demonstrated on mammography) that would not be considered surgically amenable to lumpectomy; patients treated with lumpectomy in whom there is another clinically dominant mass or mammographically suspicious abnormality within the ipsilateral breast remnant (such a mass must be biopsied and demonstrated to be histologically benign prior to randomisation or, if malignant, must be surgically removed with clear margins); patients in whom the margins of the resected specimen are involved with invasive tumour or ductal carcinoma in situ (additional surgical resections to obtain free margins are allowed); patients in whom the tumour is still present after the additional resection(s) must undergo mastectomy. Stage of disease: stage I‐III TNM staging system: 76% node negative (intervention 1) vs. 75% node negative (intervention 2) Mean age: intervention 1: < 49: 36%; > 50: 64% intervention 2: < 49: 65%, > 50: 65% Menopausal status: NR RANKL status: NR Hormone receptor status: 78% ER+ and/or PR+ in both interventions Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 3323 Country of participants: USA
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: clodronate (1600 mg, oral, daily, 3 years); intervention 2: placebo (oral, daily, 3 years); supplemental: (calcium and vitamin D); other. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy (5 years), SERM: mostly tamoxifen; chemotherapy when administered, radiotherapy, surgery (mastectomy or lumpectomy before randomisation)
Outcomes	DFS; incidence of skeletal metastases; OS; relapse‐free survival; incidence of non‐skeletal metastases; incidence of skeletal morbid events; osteonecrosis.
Notes	Funding sources: NSABP Foundation Inc, National Cancer Institute (NCI), Southwest Oncology Group, North Central Cancer Treatment Group, "The sponsors of the study had no role in study design, data collection, data analysis, data interpretation, or writing of this report, and had no access to the raw data." Conflicts of interest: AHGP received honoraria from Bayer, Novartis, Amgen, and Roche Diagnostics. SJA received travel costs for testimony about the trial to the US Food and Drug Administration in 2000 and 2004. All other authors declared that they had no conflicts of interest.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Stratified randomisation with a biased‐coin minimisation approach to generate a treatment assignment on entry
Allocation concealment (selection bias)	Low risk	Biased‐coin minimisation; "all patients, clinicians who treated and assessed patients, and protocol doctors were masked to treatment group assignment".
Blinding of participants (performance bias)	Low risk	"all patients, clinicians who treated and assessed patients, and protocol doctors were masked to treatment group assignment".
Blinding of personnel (performance bias)	Low risk	"all patients, clinicians who treated and assessed patients, and protocol doctors were masked to treatment group assignment".
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	"all patients, clinicians who treated and assessed patients, and protocol doctors were masked to treatment group assignment".
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT; very small number excluded as lost to follow‐up
Selective reporting (reporting bias)	Low risk	All predefined outcomes reported
Other bias	Low risk	Study appeared to be free of other sources of bias.

Powles 2006.

Study characteristics
Methods	Setting: multicentre, phase III Recruitment period: 1989 to 1995 Length of study: 2 years Length of follow‐up: median follow‐up of 5.6 years
Participants	Eligibility criteria: pre and post‐menopausal women with primary operable BC; no evidence of metastasis; psychologically and physically suitable for 2 years of treatment with oral clodronate or placebo. Exclusion criteria: significant renal, hepatic or non‐malignant bone disease or prior bisphosphonate use Stage of disease: intervention 1: I: 28.4% II: 56.3% other: 15.3% intervention 2: I: 27.3% II: 56.0% other: 16.7% TNM staging system: T1: 200 intervention 1 vs. 210 intervention 2 T2: 230 intervention 1 vs. 228 intervention 2 T3: 33 intervention 1 vs. 32 intervention 2 T4: 24 intervention 1 vs. 31 intervention 2 Node positive: 37% intervention 1 vs. 37.5% intervention 2 Mean age: intervention 1: 52.8 intervention 2: 52.7 Menopausal status: premenopausal: 50% (intervention 1), 49% (intervention 2) postmenopausal: 50% (intervention 1), 51% (intervention 2) RANKL status: NR Hormone receptor status: ER+: 46% intervention 1 45% intervention 2 Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 1069 Country of participants: UK, Canada, Norway, Finland
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 1600 mg of oral clodronate per day for 2 years; intervention 2: daily oral placebo for 2 years; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: chemotherapy endocrine therapy SERM radiotherapy
Outcomes	time to first bone metastases over five‐year study period; survival; changes in spine and hip BMD; effect on markers of bone turnover.
Notes	Funding sources: Bayer Schering Pharma Oy, Turku, Finland Conflicts of interest: EM: Bayer, Roche Diagnostics, Schering AHP: Bayer, Roche Diagnostics, Novartis, Berlex, Pfizer, Astrazeneca, Sopherion, Millenium JAK: Bayer BS: Berlex TP: Berlex PS: Berlex JK: Schering
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"randomised by means of numerically ordered and coded packages"
Allocation concealment (selection bias)	Low risk	Centralised blinded code
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT analysis; all participants included in the analysis and no missing outcome data
Selective reporting (reporting bias)	Low risk	All prespecified outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

ProBONE I 2005.

Study characteristics
Methods	Setting: randomised, double‐blind Recruitment period: NR Length of study: 2 years Length of follow‐up: NR (terminated due to rare patient population, planned number of patients could not be recruited in a reasonable time frame. Recruitment was stopped prematurely).
Participants	Eligibility criteria: female patients with histologically confirmed incident invasive breast cancer (T1‐4); no evidence of distant metastasis (M0); node‐negative (N0) and node‐positive (N1‐4) patients; patient has undergone complete primary tumour resection and axillary lymph node dissection less than 90 days before start of study drug treatment; hormone receptor status negative; age ≥ 18 years; patient is premenopausal at diagnosis of breast cancer (spontaneous and regular menses with premenopausal oestradiol levels (> 10 ng/mL); patient receives adjuvant standard chemotherapy with approved cytotoxic chemotherapeutic drugs (e.g. AC 4‐6 cycles) (prior neoadjuvant CT is allowed); bone density at study entry > ‐2.5 T‐Score; patient has given written informed consent prior to any study‐specific procedure; patient should be available for follow‐up. Exclusion criteria: history of treatment or disease affecting bone metabolism (e.g. Paget’s disease, primary hyperparathyroidism); known visceral metastasis or bone metastases; prior treatment with bisphosphonates; oestrogens or treatments for osteoporosis in addition to calcium and vitamin D; severe physical or psychological concomitant diseases that might impair compliance with the provisions of the study protocol or that might impair the assessment of drug or patient safety, e.g. clinically significant ascites, cardiac failure, NYHA III or IV, clinically relevant pathologic findings in ECG; other known concurrent, severe medical disorders jeopardising the life of the patient in the immediate future (e.g. myocardial infarction in previous six months, angina pectoris despite treatment, uncontrolled severe arterial hypertension, progressive cardiac or respiratory failure); known hypersensitivity to bisphosphonates; abnormal renal function as evidenced by a calculated creatinine clearance < 30 mL/minute. Creatinine clearance (CrCl) is calculated using the Cockcroft‐Gault formula: CrCl = [140‐age (years)] x weight (kg) {x 0.85 for female patients} [72 x serum creatinine (mg/dL)] current active dental problems, including infection of the teeth or jawbone (maxilla or mandibular); dental or fixture trauma, or a current or prior diagnosis of osteonecrosis of the jaw (ONJ), of exposed bone in the mouth, or of slow healing after dental procedures; recent (within 6 weeks) or planned dental or jaw surgery (e.g. extraction, implants); use of other investigational drugs (drugs not marketed for any indication) within 6 months before the start of study and participation in another clinical study; pregnancy or lactation; women of childbearing potential not applying a medically recognised form of contraception (i.e. oral contraceptives or implants, IUD, vaginal diaphragm or sponge, or condom with spermicide); known history or present abuse of alcohol or drugs (accepted social alcohol usage will not exclude the patient); history of noncompliance to medical regimens and patients who are considered potentially unreliable or incapable of giving informed consent as judged by the investigator. Stage of disease: NR TNM staging system: NR Mean age: intervention 1: 41.2 (± 6.2) intervention 2: 43.2 (± 2.6) Menopausal status: NR RANKL status: NR Hormone receptor status: NR Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 11 Country of participants: Germany
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: Zoledronic acid (4 mg, intravenously, every 3 months, 2 years); intervention 2: placebo (intravenously, every 3 months, 2 years); supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: NR
Outcomes	bone mineral density (BMD) measured by DXA at dual hips and os calcis after 24 months; bone mineral density (BMD) by QUS at os calcis and phalanges at 24 months; course of biochemical markers of bone turn over (FSH, oestradiol (E2), osteocalcin, PINP, procollagene‐I‐peptid, deoxypyridinoline in serum); pathologic fractures (proportion of patients with at least one fracture, type of fracture(s), number of fractures per patient, time to first pathologic fracture) during 24 months; development of metastases as assessed by X‐ray, CT, bone scan or MRI (proportion of patients developing metastases, number and localisation, metastases‐free survival) during 24 months and during 60 months; safety and tolerability (adverse events/serious adverse events; standard safety laboratory).
Notes	Funding sources: NR Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Randomised; no further information
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	High risk	Study was terminated and no data published.
Selective reporting (reporting bias)	High risk	Study was terminated and no data published.
Other bias	Unclear risk	No further information

ProBONE II 2015.

Study characteristics
Methods	Setting: prospective, single‐centre randomised, double‐blind, placebo‐controlled Recruitment period: 2005‐2009 Length of study: 2 years Length of follow‐up: 5 years
Participants	Eligibility criteria: > 18 years old; diagnosis of hormone receptor‐positive BC; histologically confirmed invasive BC without evidence of metastases (M0); receiving neoadjuvant or adjuvant chemotherapy and/or ET with approved standard regimens; T‐score at study entry > ‐2.5; according to the study protocol, patients receiving neoadjuvant treatment met the eligibility criteria if they had no nodal involvement, while patients receiving adjuvant treatment were required to have no more than 4 positive lymph nodes. Exclusion criteria: patients receiving any drug treatment at baseline, including bisphosphonates, steroids, calcitonin or suffering from any endocrine, renal or bone disease known to affect bone metabolism; patients with known visceral or bone metastases; patients with hypersensitivity to bisphosphonates, dental issues including osteonecrosis of the jaw (ONJ), planned or recent dental or jaw surgery, and renal impairment. Stage of disease: Tumour grade I‐III (I: 26.5% ZA, 22.2% placebo; II: 64.7% ZA, 58.3% placebo; III: 8.8% ZA, 19.4% placebo) TNM staging system: T1: 67.7% ZA, 80.6% placebo T2‐3: 32.4% ZA, 19.4% placebo N0: 85.3% ZA, 80.6% placebo N1‐2: 14.7% ZA, 19.4% placebo M0: 100% both groups Mean age: intervention 1: 43.2 intervention 2: 42.8 Menopausal status: 100% premenopausal RANKL status: NR Hormone receptor status: ER+ and PR+: 91.2% ZA, 88.9% placebo ER+ only: 2.9% ZA, 8.3% placebo PR+ only: 5.9% ZA, 2.8% placebo Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 70 Country of participants: Germany
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: zoledronic acid (4 mg, intravenously, every 3 months, 2 years); intervention 2: placebo (intravenously, every 3 months, 2 years); supplemental: vitamin D (400 IU, daily), calcium (500 mg, daily); other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: chemotherapy, radiotherapy (88.2% ZA, 91.7% placebo), endocrine therapy (GnRH analogs: 79.4% ZA, 88.9% placebo; raloxifen: 2.9% ZA, 0% placebo), SERM (94.1% ZA, 94.4% placebo), aromatase inhibitor (8.8% ZA, 13.9% placebo)
Outcomes	change in BMD at lumbar spine (L1‐L4) between baseline and 24 months; changes in femoral neck and total hip from baseline to 24 months; changes of BMD at lumbar spine, femoral neck and total hip from baseline to 60 months; pathologic fractures during the 24 months; development of metastases.
Notes	Funding sources: Novartis Germany Conflicts of interest: PH has received honoraria, unrestricted educational grants and research funding from the following companies: Amgen, Eli Lilly, Novartis, Roche, Sanofi Aventis and Wyeth. PHK has received honoraria, unrestricted educational grants and research funding from the following companies: Alexion, Amgen, Boehringer Ingelheim, Daiichi Sankyo, Eli Lilly, Glaxo‐Smith‐Kline, Ipsen, Jansen‐Cilag, Jenapharm, Kyowa Kirin, MSD, Novartis, Novo Nordisk, Nycomed, Roche, Pharmacia, Pfizer, Procter & Gamble, Sanofi Aventis, Shire, Takeda, Viropharma. FT has received honoraria, unrestricted educational grants and research funding from the following companies: Amgen, Eli Lilly, Novartis and Synexus. IK and OH declare that they have no conflict of interest.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomized"
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Only 5 patients were lost to follow‐up; ITT analysis
Selective reporting (reporting bias)	Low risk	Study protocol available and prespecified outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

REBBeCA 2008.

Study characteristics
Methods	Setting: single‐centre Recruitment period: May 2003 to July 2004 Length of study: 2 years Length of follow‐up: 2 years
Participants	Eligibility criteria: pre and newly postmenopausal (up to 8 years) women ages 18 and older; breast cancer treated with chemotherapeutic agents, with or without tamoxifen/aromatase inhibitors; negative pregnancy test. Exclusion criteria: stage 4 breast cancer; any illness or medication known to affect bone metabolism; history of osteoporosis or history of vertebral or hip fractures; kidney stones in the past 5 years; active peptic ulcer disease. Stage of disease: stage I: 28 women stage II: 51 women stage III: 5 women TNM staging system: NR Mean age: intervention 1: 50.1 +/‐ 5.1 intervention 2: 49 +/‐ 5.9 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: NR Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 87 Country of participants: USA
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 35 mg of oral risedronate per week for 2 years; intervention 2: weekly oral placebo for 2 years; supplemental: 200 IU of oral vitamin D and 500 mg of oral calcium per day for 2 years; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: lumpectomy 65.1% intervention 1, 61.4% intervention 2 mastectomy 48.8% intervention 1, 45.5% intervention 2 radiotherapy: 72.1% intervention 1, 72.7% intervention 2 at the end of the study 50% of the participants were on aromatase inhibitors Tamoxifen, toremifene citrate or fulvestrant: 74.4% intervention 1, 68.2% intervention 2
Outcomes	bone loss, as determined through BMD every six months; assessment whether it will be prevented at clinically relevant sites, such as the hip & spine, through the use of bisphosphonate therapy in study subjects. [time frame: 24 months]; correlation between biochemical markers of bone turnover and changes in BMD [time frame: 24 months]; changes in hip structural geometry in women following chemotherapy‐induced bone loss and determination if once weekly bisphosphonate, risedronate, would protect against these alterations in hip architecture, using the hip structural analysis (HSA) program of Beck; correlation of the bone mineral density assessed via standard DXA‐scan and HSA analysis; changes in body composition (total body mass, body fat, lean body mass); changes in gonadal hormones (testosterone, oestradiol, and sex hormone binding globulin (SHBG)).
Notes	Funding sources: "Funding for this study was provided to S.L.G. by the National Institutes of Health/National Institute of Diabetes and Digestive and Kidney Diseases (K24 DK062895‐03), a NCST from Procter and Gamble and the Alliance for Better Bone Health and to the General Clinical Research Center of the University of Pittsburgh by the National Institutes of Health/National Center for Research Resources (M01‐RR00056)." Conflicts of interest: "Although all authors completed the disclosure declaration, the following author(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a “U” are those for which no compensation was received; those relationships marked with a “C” were compensated. For a detailed description of the disclosure categories, or for more information about ASCO’s conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors. Employment or Leadership Position: None Consultant or Advisory Role: Susan L. Greenspan, Procter and Gamble (C); Victor G. Vogel, Procter and Gamble (C) Stock Ownership: None Honoraria: Susan L. Greenspan, Procter and Gamble; Rajib Bhattacharya, Procter and Gamble Research Funding: Susan L. Greenspan, Procter and Gamble; Rajib Bhattacharya, Procter and Gamble; Victor G. Vogel, Procter and Gamble Expert Testimony: None Other Remuneration: None"
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Patients were randomly assigned by computer generation[…]".
Allocation concealment (selection bias)	Low risk	"Patients were randomly assigned by computer generation[…]".
Blinding of participants (performance bias)	Low risk	"double‐blind"
Blinding of personnel (performance bias)	Low risk	"double‐blind"
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	"double‐blind"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT analysis
Selective reporting (reporting bias)	Low risk	Study protocol available and prespecified outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

REBBeCA II 2016.

Study characteristics
Methods	Setting: single‐centre, double‐blind, placebo‐controlled, randomised, phase 4 Recruitment period: 2008‐2013 Length of study: 2 years Length of follow‐up: 2 years
Participants	Eligibility criteria: postmenopausal women with hormone‐receptor‐positive breast cancer over age 55 years; currently receiving an AI including anastrozole, letrozole, or exemestane; low bone mass as classified by the World Health Organization (T‐score between −1.0 and −2.5 at the spine or hip); not treated with a bisphosphonate in the previous year, and had no illnesses or were on no other medications known to affect bone and mineral metabolism such as glucocorticoids or certain anti‐seizure medications; with no evidence of distant metastatic disease or osteoporosis (by BMD or clinical history); type of surgical procedure or addition of radiation therapy prior to this aromatase inhibitor therapy will not exclude patients; participants must provide voluntary, written informed consent to participate in the study, which includes understanding of the procedures, medications, and risks and benefits. Exclusion criteria: women with stage 4 breast cancer (presence of distant metastases); women with normal bone density by DXA (T‐score > ‐1.0 SD) bone density by DXA, except in the instance of a fragility fracture; women with history of any illness known to affect bone and mineral metabolism, such as renal failure (estimated GFR < 30), hepatic failure, malignancy (excluding breast cancer, treated superficial basal and squamous cell carcinoma and malignancies where the diagnosis itself or its treatment would not adversely affect bone metabolism), untreated primary hyperparathyroidism, and malabsorption; women being treated with oral glucocorticoid therapy > 3 months for suppression therapy, and certain anti‐seizure medications which may adversely affect bone metabolism (phenobarbital, phenytoin, carbamazepine); those with untreated active peptic ulcer disease; those with osteoporosis by BMD (T‐score ‐2.5 SD at the spine or total hip) or a history of fragility fracture as an adult. However, as discussed above, osteoporotic women may elect to enrol in the study; women treated with oral bisphosphonates or calcitonin for 3 months within the last year (3‐month washout period); men and children will be excluded because they do not get postmenopausal osteoporosis following treatment with an aromatase inhibitor; women with very poor dental hygiene (as assessed by the baseline dental exam) in need of dental extraction during the study; use of fluoride for more than 1 month ever (except for dental treatment); less than 2 evaluable vertebrae; distant metastatic disease. Stage of disease: stage I‐III TNM staging system: NR Mean age: intervention 1: 65 intervention 2: 64 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: 100% ER+ and/or PR+ Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 109 Country of participants: USA
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: risedronate (35 mg, oral, weekly, 2 years); intervention 2: placebo (oral, weekly, 2 years); supplemental: vitamin D (200 IU, daily, 2 years), calcium (500 mg, daily, 2 years); other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: radiotherapy (76% risedronate vs. 91% placebo), chemotherapy (32% risedronate vs. 50% placebo), aromatase inhibitor (anastrozole: 82% risedronate vs. 7% placebo; letrozole: 7% risedronate vs. 22% placebo; exemastane: 11% risedronate vs. 6% placebo), lumpectomy (81% risedronate vs. 79% placebo), mastectomy (35% risedronate vs. 32% placebo), axillary node removal (75% risedronate vs. 89% placebo)
Outcomes	changes in spine and hip BMD at 24 months; BMD changes at 12 months; bone turnover markers assessed by serum C‐telopeptide crosslinks type I collagen; bone formation, assessed by serum intact N‐terminal propeptide type I procollagen.
Notes	Funding sources: The study was funded by grant support from Procter and Gamble and the Alliance for Better Bone Health and Warner Chilcott who supplied the drug and matching placebo. Support was also provided by NIH grants K24DK062895, T32AG021885, and P30AG024827. Conflicts of interest: Susan L. Greenspan has the following grant funding but none present a conflict of interest: NIH, PCORI, Eli Lilly, Amgen. Karen T. Vujevich, Barry C. Lembersky, Shannon L. Puhalla, and Priya Rastogi have no conflict of interest to declare. Adam Brufsky has the following grant funding but none present a conflict of interest: NIH/NCI. G.J. van Londen has the following grant funding but none present a conflict of interest: NIH. Rachel C. Jankowitz has the following grant funding but none present a conflict of interest: NSABP Foundation, Komen Foundation. Subashan Perera has the following grant funding but none present a conflict of interest: NIH, PCORI.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"The study biostatistician randomized participants in a 1:1 ratio using random block sizes of 2 and 4".
Allocation concealment (selection bias)	Low risk	"The study biostatistician randomized participants in a 1:1 ratio using random block sizes of 2 and 4. The independent research pharmacist provided identically appearing active drug or placebo".
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Double‐blind
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of participants that withdrew from the study was almost equal in the study arms;, ITT analysis
Selective reporting (reporting bias)	Low risk	Study protocol available and prespecified outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

Rhee 2013.

Study characteristics
Methods	Setting: single‐centre, randomised, placebo‐controlled, double‐blind Recruitment period: NR Length of study: 6 months Length of follow‐up: 6 months
Participants	Eligibility criteria: HR‐positive early breast cancer treated with appropriate surgical treatment; ECOG performance status less than 2; postmenopausal; newly treated with any of the third‐generations AIs, either anastrozole or letrozole. Exclusion criteria: clinical or radiological evidence of distant metastasis; previous use of intravenous or oral bisphosphonate, any selective oestrogen receptor modulator or systemic corticosteroid within 12 months; continuous use of drugs affecting bone metabolism, use of drugs affecting the musculoskeletal system within 12 weeks of the study (oestrogen, progesterone, calcitonin, fluoride, calcitriol, calcitonin, mithramycin, gallium nitrate); oesophageal stricture or achalasia and hypersensitivity to alendronate or calcitriol; hypocalcaemia or hypercalcaemia; scheduled for surgery; judged to be inappropriate as clinical test subjects. Stage of disease: early stage breast cancer TNM staging system: NR Mean age: intervention 1: 57.1 ± 1.0 intervention 2: 58.5 ± 1.1 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: 100% HR+ and/or PR+ Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 98 Country of participants: Korea
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: alendronate (5 mg, oral, daily, 6 months); intervention 2: placebo (oral, daily, 6 months); supplemental: vitamin D (400 IU/day, 6 months), calcium (500 mg/day, 6 months); other. Previous bone‐modifying interventions: Cancer treatment during study period: aromatase inhibitor (anastrozole: 28 (alendronate) vs. 31 (placebo), letrozole (21 (alendronate) vs. 18 (placebo), calcitriol as part of the study drug in intervention 1; previous chemotherapy 46.9% (alendronate) vs. 53.1% (placebo)
Outcomes	changes in BMD in lumbar spine and total hip at 6 months; changes in serum levels of osteocalcin (OCN) and serum C‐telopeptide (CTX) at 6 months.
Notes	Funding sources: Yuyu Pharma Inc. Conflicts of interest: none declared
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Assignments to the Maxmarvil or placebo group were made using a list of randomly allocated treatment codes.
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	Low risk	Double‐blind
Blinding of personnel (performance bias)	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Double‐blind
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Number of participants that withdrew from the study was small in both arms.
Selective reporting (reporting bias)	Unclear risk	Study protocol not available
Other bias	Low risk	Study appeared to be free of other sources of bias.

Saarto 2004.

Study characteristics
Methods	Setting: monocentre Recruitment period: 1990‐1993 Length of study: 3 y Length of follow‐up: 10 y
Participants	Eligibility criteria: pre and postmenopausal women with operable breast cancer and histologically proven axillary metastases (T1 to T3, N1/2, M0) Exclusion criteria: age greater than 75 years; Karnofsky performance status of less than 70%; presence of other malignancies; peptic ulcer or its symptoms pregnancy; serum creatinine concentration greater than 150 µmol/L. Stage of disease: NR TNM staging system: T1: 49% T2: 44% T3: 6% unknown: 1% node positive: 100% Mean age: intervention 1: 52 intervention 2: 52 Menopausal status: 52% premenopausal, 48% postmenopausal RANKL status: NR Hormone receptor status: ER+: 64% ER‐: 29% ER unknown: 7% PR+: 55% PR‐: 38% PR unknown: 7% Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 299 Country of participants: Finland
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: clodronate (1600 mg, oral, daily, 3 years); intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy (postmenopausal), SERM (postmenopausal; tamoxifen 20 mg daily or toremifine 60 mg daily), chemotherapy (premenopausal; 6 cycles of cyclophosphamide 600 mg/m2, methotrexate 40 mg/m2 and fluorouracil 600 mg/m2, intravenously on day 1 and successive 3‐week intervals), radiotherapy (postoperative with megavoltage irradiation in a total dose of 50 Gy in 25 fractions to regional lymph nodes and the operation scar after mastectomy or residual breast tissue after breast‐conserving resection), surgery with axillary evacuation and total mastectomy or breast‐conserving resection
Outcomes	BMD lumbar spine and hip was measured before treatment and at 1, 2, 3, 5, and 10 years after therapy; occurrence of bone metastases; disease recurrence (any site); DFS.
Notes	Funding sources: Leiras Pharmaceutical Company Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Randomised, but no further information
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	"Clinical investigation and basic laboratory tests were repeated every 4 to 6 months with a radiologic examination if necessary. Investigators performing bone scans and radiologic examinations were blinded to treatment allocation".
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT; no missing data for the final population of 282
Selective reporting (reporting bias)	Low risk	Outcomes were not specified in methodology; however, all expected outcomes were reported.
Other bias	Low risk	Study appeared to be free of other sources of bias.

SABRE 2010.

Study characteristics
Methods	Setting: multicentre, phase III/IV Recruitment period: NR Length of study: 2 years Length of follow‐up: 2 years
Participants	Eligibility criteria: women defined as postmenopausal; histologically proven operable invasive breast cancer; hormone‐receptor‐positive breast cancer. Exclusion criteria: clinical evidence of metastatic disease; bilateral hip fractures or bilateral hip prosthesis; receiving or received in last 12 months hormonal therapy for breast cancer, bisphosphonate therapy, oestrogens; malabsorption syndrome. Stage of disease: NR TNM staging system: nodal negative: 64.4% (intervention 1), 72.7% (intervention 2) Mean age: intervention 1: 63.8 intervention 2: 64.8 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: 100% HR+ Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 154 Country of participants: USA, Canada, France, Greece, Netherlands, Spain, South Africa, UK
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: risedronate; intervention 2: placebo; supplemental: 200 IU of oral vitamin D and 500 mg of oral calcium twice daily, respectively for 2 years; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy, anastrozole
Outcomes	change from baseline in lumbar spine (L1‐L4) BMD 12 mo; change from baseline in total hip BMD; change from baseline in lumbar spine (L1‐L4) BMD 24 mo; change from baseline in bone formation markers; change from baseline in bone resorption and formation markers; change from baseline in LDL‐cholesterol; change from baseline in LDL‐cholesterol, HDL‐cholesterol, total cholesterol, and serum triglycerides.
Notes	Funding sources: AstraZeneca Conflicts of interest: Employment or Leadership Position: Glen Clack, AstraZeneca (C) Consultant or Advisory Role: Catherine Van Poznak, AstraZeneca (C), Novartis (C), Roche (C), Amgen (C); David Barlow, AstraZeneca (C); Andreas Makris, AstraZeneca (C); Richard Eastell, AstraZeneca (C), Sanofi Aventis (C) Stock Ownership: Glen Clack, AstraZeneca Honoraria: Rosemary A. Hannon, Pfizer; John R. Mackey, AstraZeneca; Andreas Makris, AstraZeneca; Richard Eastell, AstraZeneca, Sanofi Aventis Research Funding: Rosemary A. Hannon, AstraZeneca, Proctor & Gamble Pharmaceuticals; Andreas Makris, AstraZeneca; Richard Eastell, AstraZeneca, Sanofi Aventis Expert Testimony: No Other Remuneration: None (C = compensation)
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Random assignment was determined via a central scheme prepared by the biostatistics group at AstraZeneca, and investigators and trial monitors were unaware of each patient’s treatment assignment".
Allocation concealment (selection bias)	Low risk	"Random assignment was determined via a central scheme prepared by the biostatistics group at AstraZeneca, and investigators and trial monitors were unaware of each patient’s treatment assignment".
Blinding of participants (performance bias)	Low risk	Triple masked (participant, care provider, investigator)
Blinding of personnel (performance bias)	Low risk	Triple masked (participant, care provider, investigator)
Blinding of outcome assessment (detection bias) subjective outcomes	Low risk	Triple masked (participant, care provider, investigator)
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	No ITT, but dropouts balanced
Selective reporting (reporting bias)	Unclear risk	Not all predefined outcomes reported
Other bias	Low risk	Study appeared to be free of other sources of bias.

Safra 2011.

Study characteristics
Methods	Setting: phase II trial Recruitment period: NR Length of study: 2 y Length of follow‐up: 5 y
Participants	Eligibility criteria: postmenopausal women with histologically documented early (non‐metastatic) breast cancer; previous treatment with tamoxifen for at least 2.5 years and not more than 3.5 years; assigned to receive letrozole treatment; Karnofsky performance status ≥ 70; life expectancy ≥ 16 weeks; signed informed consent after full explanation of study by participating clinician and prior to any study‐specific procedures; adjuvant or neoadjuvant chemotherapy is allowed; no clinical and/or radiologic evidence of distant metastases; no prior treatment with an aromatase inhibitor; able to comply with treatment and scheduled follow‐up visits; age between 18 and 82 years. Exclusion criteria: pregnant or lactating women or women with childbearing potential; patients with other malignancies except adequately treated basal cell carcinoma of the skin or in‐situ cervix carcinoma; active infection or other serious underlying medical condition which would impair the ability of the patient to receive protocol treatment; clinical and/or radiological evidence of distant metastases; evidence of pathological fracture; prior treatment with an aromatase inhibitor; prior administration of any intravenous bisphosphonate during the last year; oral bisphosphonate must be discontinued within 4 weeks of enrolment; administration of long‐term systemic corticosteroids within the last 12 months (short‐term steroid treatment is allowed); prior use of parathyroid hormone treatment for more than 1 week; use of any drug known to affect the skeleton (calcitonin, mithramycin, gallium nitrate) within two weeks prior to enrolment; abnormal renal function: creatinine clearance must be above 30 mL/min (calculated by Cockroft formula); evidence of metabolic bone disease (Paget's, osteogenesis imperfecta, hyperparathyroidism within the 12 months prior to enrolment); baseline lumbar spine and or total hip bone mineral density T score below ‐2; known hypersensitivity to zoledronic acid; psychological, familial, sociologic, or geographic conditions which do not permit medical follow‐up and compliance with the study protocol; white blood cell ≤ 3.0 x 10exp9/L or granulocytes ≤ 1.5 x 10exp9/L or platelets ≤ 100 x 10exp9; total bilirubin > 1.5 x upper normal limit, SGOT and SGPT > 2.5 x upper normal limit; unable to undergo DXA bone density scanning (spine deformity, severe scoliosis, lumbar sacral spine surgery). Stage of disease: stage I‐III TNM staging system: NR Mean age: intervention 1: 59.0 ± 8.5 intervention 2: 61.1 ± 9.2 Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: 100% ER+ and/or PR+ Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 86 Country of participants: Israel
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: ZA (4 mg, intravenous, every 6 months, 2 years); intervention 2: no treatment; supplemental: vitamin D (200 IU, oral, twice daily, 2 years), calcium (600 mg, oral, twice daily, 2 years); other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy, SERM (tamoxifen treatment prior to study for 2.5 y inclusion criterium), aromatase inhibitor (letrozole 2.5 mg/day), chemotherapy allowed
Outcomes	BMD; fractures; OS; disease recurrence; renal function; liver function; safety.
Notes	Funding sources: sponsored by the Tel Aviv Sourasky Medical Center in collaboration with the Soroka University Medical Center. We thank Maria Soushko, PhD, of Phase Five Communications Inc., for medical editorial assistance with this manuscript. Financial support for this assistance and the study drug (zoledronic acid) were provided by Novartis Pharmaceuticals. Conflicts of interest: The authors declared no conflicts of interest.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No further information
Allocation concealment (selection bias)	Unclear risk	No further information
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	No ITT, but reasons given transparently and characteristics of patients discontinuing trial described as balanced between arms
Selective reporting (reporting bias)	High risk	Not all time points reported as predefined and it seemed, according to fig. 3, that time points favouring ZA were chosen for the reporting of the main outcome BMD.
Other bias	Low risk	Not identified

Saito 2015.

Study characteristics
Methods	Setting: NR Recruitment period: March 2008‐September 2010 Length of study: 2 years Length of follow‐up: 2 years
Participants	Eligibility criteria: postmenopausal breast cancer patients (stage I‐III); receiving any form of aromatase inhibitor (anastrozole (ANA), exemestane (EXE) or letrozole (LTZ)); bone mineral density as measured by DEXA was lower than the adult mean; written informed consent had been obtained. Exclusion criteria: NR Stage of disease: stage I‐III TNM staging system: NR Mean age: intervention 1: NR intervention 2: NR Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: NR Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 58 Country of participants: Japan
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 35 mg of oral alendronate weekly; intervention 2: no treatment; supplemental; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: aromatase inhibitor (any: ANA, EXE or LTZ)
Outcomes	BMD; measured levels of surrogate markers for bone health; adverse events associated with bone‐preserving therapies.
Notes	Funding sources: NR Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No further information
Allocation concealment (selection bias)	Unclear risk	No further information
Blinding of participants (performance bias)	High risk	"open‐label"
Blinding of personnel (performance bias)	High risk	"open‐label"
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	"open‐label"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	No ITT; reasons for dropouts transparently reported and balanced between arms
Selective reporting (reporting bias)	Unclear risk	No study registry entry available
Other bias	Low risk	Not identified

Solomayer 2012.

Study characteristics
Methods	Setting: multicentre, phase II Recruitment period: 2002‐2004 Length of study: 24 months Length of follow‐up: median 88 months
Participants	Eligibility criteria: patients, with primary breast cancer who had undergone complete primary tumour resection and axillary lymph node dissection; evidence of minimal residual disease (disseminated tumour cells in bone marrow); patients had to receive one of the following adjuvant therapy categories: chemotherapy ± hormonal therapy or hormonal therapy alone. Exclusion criteria: inflammatory, metastatic or recurrent breast cancer or a history of breast cancer prior to the currently diagnosed case; creatinine clearance < 30 mL/min, current active dental problems or trauma, or a current or prior diagnosis of osteonecrosis of the jaw; neoadjuvant chemotherapy, neoadjuvant hormonal therapy, or neoadjuvant radiotherapy; prior stem cell rescue/bone marrow transplant; history of other cancers aside from non‐melanomatous skin cancer or carcinoma in situ of the uterine cervix. Stage of disease: I: 57 IIa: 24 IIb: 11 IIIa: 3 IIIb: 1 TNM staging system: pT1: 51 pT2‐3: 29 nodal status negative: 70 nodal status positive: 10 Mean age: intervention 1: median: 54 (36‐71) intervention 2: median: 54 (37‐72) Menopausal status: 39% premenopausal, 61% postmenopausal RANKL status: NR Hormone receptor status: ER+: 81% ER‐: 19% PR+: 66% PR‐: 34% Human epidermal growth factor receptor 2 status: HER2/neu +: 26% HER2/neu ‐: 73% Participants randomised: total: 96 Country of participants: Germany
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 3‐4 mg of intravenous zoledronic acid every 4 weeks for 24 months; intervention 2: no treatment; supplemental: vitamin D and calcium; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy and/or chemotherapy
Outcomes	DTC counts in bone marrow; safety; changes in DTC counts versus baseline; bone‐metastasis–free survival, which included death from any cause or bone metastasis; disease‐free survival, which included death from any cause or disease recurrence at any site; BMD; occurrence of bone metastases.
Notes	Funding sources: Novartis Pharmaceuticals Conflicts of interest: "Dr EFS has received honoraria from Novartis, Roche, and AstraZeneca, and research grants from Roche. Dr WJ has received lecture honoraria and research grants from Novartis. Dr H‐JL has received lecture honoraria from Roche, Novartis, Sanofi, GlaxoSmithKline, and Pfizer, and is an advisory board member for Roche, Novartis, Pfizer, Fresenius, and BMS. Dr JH has received lecture honoraria from Novartis, and is an advisory board member for Roche, Novartis, and Amgen. Dr TF has received honoraria from Roche, Novartis, and AstraZeneca, and research grants from Roche. Dr BW is employed by Novartis. Dr DW has received research grants from Novartis and Roche. Drs GG, PH, SB, and BK have declared no conflict of interest".
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No further information
Allocation concealment (selection bias)	Unclear risk	No further information
Blinding of participants (performance bias)	High risk	"open‐label"
Blinding of personnel (performance bias)	High risk	"open‐label"
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	"open‐label"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Low ‐ ITT and per‐protocol analysis perfomed and reasons for discontinuation balanced between arms
Selective reporting (reporting bias)	High risk	BMD planned as an outcome as described in trial registry, but no results reported in the full‐text references
Other bias	Low risk	Not identified

Sun 2016.

Study characteristics
Methods	Setting: NR Recruitment period: January 2011‐February 2012 Length of study: 1 year Length of follow‐up: 1 year
Participants	Eligibility criteria: women > 60 years with cessation of menses, women ≤ 60 years with spontaneous cessation of menses > 12 months, women with bilateral oophorectomy, or women ≤ 60 years, with no spontaneous menses for 1 year but with postmenopausal oestradiol levels; histopathological or cytological diagnosis as invasive breast cancer; stage I, II, or III A breast cancer; oestrogen and/or progesterone receptor‐positive; no evidence of recurrent or metastatic disease; life expectancy of > 5 years; an Eastern Cooperative Oncology Group performance status of 0–2; the baseline total LS or FN BMD T‐score must have been < ‐2.0; haematology, liver, and kidney function are normal; good understanding and compliance by patients with the pilot programme and provision of informed consent. Exclusion criteria: patients with clinical or radiological evidence of distant metastases; patients with existing LS or total hip (TH) fracture, or a history of nontraumatic fractures or osteoporosis; patients who received recent treatment with any drugs known to affect the skeleton, prior treatment with intravenous bisphosphonates or AIs, prior exposure (within the past 6 months) to anabolic steroids or growth hormone; patients with diseases known to influence bone metabolism, other malignancy within 5 years (except adequately treated basal or squamous cell carcinoma of the skin and in situ carcinoma of the cervix), renal dysfunction, uncontrolled infections, diabetes mellitus, thyroid dysfunction, seizure disorders associated with falls, HIV, malabsorption syndrome, or mental illnesses; patients with a known hypersensitivity to zoledronic acid, other bisphosphonates, letrozole, calcium, or vitamin D; patients contraindicated for the dual X‐ray absorptiometry. Stage of disease: Arm A: stage I: 15, stage II: 25, stage IIIa: 10 Arm B: stage I: 13, stage II: 24, stage IIIa: 13 TNM staging system: NR Mean age: intervention 1: median: 58 (35‐83) intervention 2: median: 56 (33‐79) Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: 100% hormone receptor‐positive Human epidermal growth factor receptor 2 status: NR Participants randomised: total: 120 Country of participants: China
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: 4 mg of intravenous zoledronic acid every 6 months for 1 year; intervention 2: no treatment; supplemental: 400 IU of vitamin D and 500 mg of calcium daily; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: 43 in arm 1 and 45 in arm 2 received chemotherapy; all patients were assigned to take 2.5 mg letrozole per day.
Outcomes	intra‐patient percentage change in LS (L1–L5) BMD from baseline to month 12; percentage change in TH and FN BMD; incidence of osteoporosis; incidence of a clinically meaningful 5% decline in BMD at 1 year; change of serum N‐telopeptide of type 1 collagen (NTX); bone‐specific alkaline phosphatase (BSAP) concentrations.
Notes	Funding sources: "this research received no specific grant from any funding agency in the public, commercial, or not‐for‐profit sectors". Conflicts of interest: "The authors report no conflicts of interest in this work".
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomly assigned" with no further information
Allocation concealment (selection bias)	Unclear risk	No information given
Blinding of participants (performance bias)	High risk	"open‐label"
Blinding of personnel (performance bias)	High risk	"open‐label"
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	"open‐label"
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	No ITT and "patients who discontinued letrozole or ZA were withdrawn from study"; pts were also withdrawn for toxicity or refusal ‐ in total 20 pts; not clear which reason or in which arm
Selective reporting (reporting bias)	Unclear risk	No study registry entry available
Other bias	Low risk	Study appeared to be free of other sources of bias.

SWOG S0307 2019.

Study characteristics
Methods	Setting: multicentre, randomised Recruitment period: 2006‐2010 Length of study: 3 y Length of follow‐up: 10 y
Participants	Eligibility criteria: patients must have received, or have been planning to receive, systemic adjuvant therapy; neoadjuvant therapy permitted if enroled after surgery; radiation therapy allowed at any time; patients with previous bisphosphonate treatment for bone density eligible, if discontinued at registration; age 18 years and older; SWOG Performance Status 0–2; adequate local therapy; serum creatinine no more than 2 institutional upper limit of normal and calculated creatinine clearance no less than 30 mL/min; dental examination within 6 months of study initiation; negative pregnancy test in women with reproductive potential. Exclusion criteria: patients with renal failure or history of prior malignancy (other than specified in situ cancers or other cancers from which they were disease free for < 5 years) were excluded from participation. Stage of disease: I: 2000 II: 2640 III: 1236 TNM staging system: N0: 50.3% N1‐3: 31.5% N ≥ 4: 17.5% Mean age: intervention 1: median 53 intervention 2: median 52.6 intervention 3: median 52.7 Menopausal status: NR RANKL status: NR Hormone receptor status: 78.5% ER+ and/or PR+ Human epidermal growth factor receptor 2 status: 18.8% HER2+ Participants randomised: total: 6097 Country of participants: USA
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: ZA (4 mg, intravenously, monthly for 6 months and then every 3 months, 2.5 years); intervention 2: clodronate (1600 mg, oral, daily, 3 years); intervention 3: ibandronate (50 mg, oral, daily, 3 years); supplemental: vitamin D, calcium; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: endocrine therapy (75.2%), chemotherapy (79.6%)
Outcomes	DFS (defined as time from registration to first occurrence of disease recurrence (local, regional, distant), new breast primary, or death due to any cause); OS (defined as time from registration to death due to any cause); distributions of sites of first recurrence on the three arms; grade 3‐5 AEs related to study drug (CTCA Version 4.0).
Notes	Funding sources: National Cancer Institute of the National Institutes of Health, and in part by Breast Cancer Research Foundation, Susan G. Komen, Berlex Pharmaceuticals (Bayer), Roche/Genentech, and Novartis provided study drug and drug distribution support. Conflicts of interest: JRG has received fees for consulting or advising from Genentech/Roche, Genomic Health, Novartis, Pfizer, Merck, Immunomedics, AstraZeneca, Puma, Inbiomotion, Radius, Sandoz/Hexal AG. WEB has received research support to his institution from Merck, AstraZeneca. AHGP has received research support from Amgen, Novartis, Pfizer, honoraria from Pfizer, Novartis, Amgen, and fees for consulting or advising from Pfizer, Novartis, Amgen. ATS has received fees for consulting or advising from Amgen, Novartis, Biothera, Pfizer, AstraZeneca, and has received research support to her institution from Amgen and Seattle Genetics. DFH has received fees for consulting or advising from Cepheid, Freenome, Cellworkis, CVS Caremark Breast Panel, Agendia, Merrimack, Eli Lilly, Menarini Silon Biosystems, Puma, AstraZeneca, and has stock options with ONcimmune and InBiomotion. MMS has received fees for consulting or advising from Amgen. CHVP has received research support from Bayer and reports patents, royalties or other intellectual property from Now UpToDate. ECD has received fees for consulting or advising from STRAT, Novartis (spouse), and research support from Merck, Cerulean, Pfizer, Novartis, Lilly, Bayer. CIF has received honorarium from Biotheranostics, received fees for consulting or advising from Biotheranostics, received travel support from Biotheranostics, and received research support from Novartis, Eli Lilly, Oncothyreon, Genentech/Roche, Pfizer. ADE has stock or other ownership interest in Allergan, Abbott, Celgene, Biomarin, Abbvie, Agilent, Alexion, Lilly, Bristol Myers Squibb, Merck, Amgen, Incyte, Pfizer, Gilead, Tesaro, and has received research funding from Eisai, Immune Design, Genentech, Lilly, Innocrin, Astrellas. JHM has employment, leadership role, and stock or ownership to report for MHP. GNH received fees for consulting or advising from Novartis, and has received research funding from Novartis. All other authors declare no competing interests.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomised assignment generated by computer using the Oncology Patient Enrollment Network (OPEN) system maintained by NCI.
Allocation concealment (selection bias)	Unclear risk	No further information
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	Transparently described numbers of included patients for each outcome, balanced between 3 intervention arms
Selective reporting (reporting bias)	Low risk	All predefined outcomes reported
Other bias	Low risk	Not identified

Team IIB 2006.

Study characteristics
Methods	Setting: multicentre, prospective, phase III Recruitment period: February 2007‐May 2014 Length of study: 3 years Length of follow‐up: median 8.5 years
Participants	Eligibility criteria: histological confirmed invasive adenocarcinoma of the breast; stage I to III breast cancer; completed adequate surgical treatment; (neo)adjuvant chemotherapy, radiotherapy and/or trastuzumab are allowed; indication to receive adjuvant hormonal therapy according to most recent NABON guideline; ER and/or PgR receptor positive (ER expression more than or equal to 10% and/or PgR more than or equal to 10%); known human epidermal growth factor receptor 2 (HER2) status; adequate renal‐ and hepatic function as assessed by laboratory testing within four weeks prior to enrolment: renal function: creatinine less than or equal to 120 µmol/L. If limited values, the calculated creatinine clearance should be more than or equal to 30 mL/min with the Cockcroft and Gault‐formula hepatic function: total bilirubin less than or equal to 1.5 x upper normal limit (UNL); aspartate aminotransferase (AST) (serum glutamic oxaloacetic transaminase [SGOT]) and alanine transaminase (ALT) (serum glutamic pyruvic transaminase [SGPT]) less than or equal to 2.5 x UNL; alkaline phosphatase less than or equal to 2.5 x UNL; postmenopausal women: postmenopausal defined as: age more than or equal to 50 and amenorrhoea for more than one year bilateral surgical oophorectomy and no HRT (any age is acceptable) age less than 50 with natural amenorrhoea more than one year at breast cancer diagnosis (and uterus in situ) postmenopausal due to chemotherapy will be excluded in case of doubt about the menopausal status, assessment of FSH, LH and oestradiol has to be performed to define the menopausal status; WHO performance status zero or one; absence of any psychological, familial, sociological or geographical condition potentially hampering compliance with the study protocol and follow‐up schedule. Exclusion criteria: M1 disease by clinical examination according to the NABON guideline; bilateral invasive breast cancer (including CIS); patients having shown progressive disease in TEAM IIa (preoperative hormonal treatment with exemestane); one of the following diseases: uncontrolled cardiac disease psychiatric disorders preventing proper informed consent patients with untreated oesophagitis, gastric ulcers or irritable bowel disease (IBD) concomitant malignancies within the last five years, except for adequately treated carcinoma in situ of the uterine cervix or basal squamous cell carcinoma of the skin prior invasive breast cancer and/or CIS within the last 15 years other serious illnesses that may interfere with subject compliance, adequate informed consent or determination of causality of adverse events; history of disease with an influence on bone metabolism, including: Paget's disease of the bone primary hyperparathyroidism (patients cured by surgery may be included if interval more than or equal to one year); hormone replacement therapy during the last 12 months; current active dental problems including dental abscess or infection of the jawbone (maxilla or mandible), or a current or prior diagnosis of osteonecrosis of the jaw requiring maxillo‐facial surgery; recent (within four weeks of study entry) or planned dental or jaw surgery (e.g. extraction, implants). Recent dental fillings, teeth scaling and polishing or minor gingival surgery do not exclude the patient; concurrent participation in another clinical study that may interfere with the results of the trial involving investigational agents within thirty days of treatment from this study, unless this is agreed by the Study Coordinators; more than five weeks after final surgery or after end of adjuvant chemotherapy. Stage of disease: stage I‐III TNM staging System: NR Mean age: median 62 years intervention 1: median 62 years intervention 2: median 62 years Menopausal status: NR RANKL status: NR Hormone receptor status: 100% hormone receptor‐positive Human epidermal growth factor receptor 2 status: 9.5% HER2+ Participants randomised: total: 1116 Country of participants: Netherlands
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: oral ibandronate 50 mg/day for 3 years; intervention 2: no treatment; supplemental: NR; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: 56% of all patients received (neo)adjuvant chemotherapy (95% anthracyclines, 69% taxanes)
Outcomes	disease‐free survival; time to bone metastases; rate of bone metastases; other sites of recurrence overall survival; safety.
Notes	Funding sources: "Source(s) of Monetary Support: Pfizer, Roche Nederland BV, Leiden University Medical Centre" Conflicts of interest: Vincent O. Dezentje Consulting or Advisory Role: AstraZeneca Uncompensated Relationships: Novartis (Inst) Mathijs P. Hendriks Research Funding: Amgen (Inst), AstraZeneca (Inst), Bayer (Inst), Boehringer Ingelheim (Inst), Bristol Myers Squibb (Inst), Clovis Oncology (Inst), Eisai (Inst), Ipsen (Inst), Merck Sharp & Dohme (Inst), Novartis (Inst), Pfizer (Inst), Roche (Inst) Vivianne C.G. Tjan‐Heijnen Honoraria: Novartis, Roche, Lilly, AstraZeneca Consulting or Advisory Role:Pfizer, Lilly, Accord Healthcare, Novartis Research Funding:Roche (Inst), Eisai (Inst), Pfizer (Inst), Novartis (Inst), Lilly (Inst), Daiichi Sankyo/AstraZeneca (Inst), Gilead Sciences (Inst) Judith R. Kroep Consulting or Advisory Role: AstraZeneca (Inst), Novartis (Inst), Tesaro (Inst), Lilly (Inst), Vitroscan (Inst), MDS (Inst) Research Funding: AstraZeneca (Inst), Novartis (Inst), Philips Research (Inst) Sabine C. Linn Consulting or Advisory Role: Daiichi Sankyo (Inst) Research Funding: Genentech/Roche (Inst), AstraZeneca (Inst), Bristol Myers Squibb (Inst), Tesaro (Inst), Merck (Inst), Immunomedics (Inst), Eurocept Pharmaceuticals (Inst), Agendia (Inst), Novartis (Inst) Travel, Accommodations, Expenses: Daiichi Sankyo Europe GmbH (Inst) No other potential conflicts of interest were reported.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Patients were randomly assigned by a computer in a 1:1 ratio; stratification was performed according to Pocock’s minimisation strategy.
Allocation concealment (selection bias)	Low risk	Patients were centrally randomly assigned [see above].
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	"intention‐to‐treat population"
Selective reporting (reporting bias)	Low risk	All predefined outcomes reported
Other bias	Low risk	None identified

Tevaarwerk 2007.

Study characteristics
Methods	Setting: multicentre, clinical trial, open‐label, randomised Recruitment period: 2000‐2007 Length of study: 1 y Length of follow‐up: up to 10 y for survival data
Participants	Eligibility criteria: histologically‐confirmed T4 or node‐positive adenocarcinoma of the breast; diagnosis within five years of enrolment; postmenopausal women; menopause was conventionally defined as ≥ 1 year since the last menstrual period and no prior oophorectomy/hysterectomy or prior bilateral oophorectomy or previous hysterectomy, one or both ovaries intact, ≥ 60 years or FSH level in postmenopausal range; patients were required to have an ECOG performance status of 0 to 2; age > 18 years; adequate bone marrow reserve (absolute neutrophil count ≥ 1500/mm3, platelet count ≥ 100,000/mm3); adequate renal function (serum creatinine ≤ 1.5 mg/dL; BUN ≤ 30); adequate hepatic function (alkaline phosphatase and SGOT ≤ 1.5 times institutional upper limit of normal and bilirubin ≤ 1.5 mg/dL); normal calcium; no evidence of breast cancer recurrence, as demonstrated by normal complete blood cell count, liver function tests, chest X‐ray and bone scan within 28 days of enrolment; prior adjuvant chemotherapy was permitted. Exclusion criteria: participants with a history of second or other cancers were excluded, if the estimated risk of recurrence for the second malignancy was over 5%; concurrent bisphosphonate use; women not receiving tamoxifen were excluded for a T score of < −2.0 at the hip or spine. Stage of disease: stage II‐III (locally advanced disease) TNM staging system: Tumour size: < 2 cm: 23.5% 2.1‐5 cm: 42.6% > 5 cm: 26.5% Nodal status: negative: 2.9% positive: 97.1% Mean age: intervention 1: median age 54 y intervention 2: median age 50.5 y Menopausal status: 100% postmenopausal RANKL status: NR Hormone receptor status: ER+ and/or PR+: 58 ER‐ and/or PR‐: 10 Human epidermal growth factor receptor 2 status: HER2 amplified: 8 HER2 not amplified/unknown: 60 Participants randomised: total: 68 Country of participants: USA
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention 1: ZA (4 mg, intravenous, every 3 months, 1 year); intervention 2: no treatment; supplemental: "supplemental calcium and vitamin D was permitted at the discretion of the treating physician, but not routinely assessed or tracked"; other: NR. Previous bone‐modifying interventions: NR Cancer treatment during study period: chemotherapy (94%), endocrine therapy (83.8%), SERM (60.3%), aromatase inhibitor (9 women)
Outcomes	change in BMD at the lumbar spine and femoral neck at 1 year; change in calcaneal BMD; DFS; OS; toxicity; rates of metastases (bone, visceral and distant).
Notes	Funding sources: Novartis Pharmaceuticals Conflicts of interest: NR
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Participants were randomized using permuted blocks of varying sizes, stratified by the status of current endocrine therapy".
Allocation concealment (selection bias)	Unclear risk	No further information
Blinding of participants (performance bias)	High risk	Open‐label
Blinding of personnel (performance bias)	High risk	Open‐label
Blinding of outcome assessment (detection bias) subjective outcomes	High risk	Open‐label
Blinding of outcome assessment (detection bias) objective outcomes	Low risk	Due to nature of objective outcomes
Incomplete outcome data (attrition bias) All outcomes	Low risk	ITT for BMD and efficacy endpoints ‐ "Participants developing toxicities that failed to meet criteria for the next dose of ZA were taken off study" and "Participants removed from study were followed for toxicity until 4 weeks after the last dose of ZA and until death for survival data".
Selective reporting (reporting bias)	Unclear risk	Outcome rates of metastases not reported; in trial registry, investigators claimed they had not collected the data.
Other bias	Low risk	Not identified

ACT: doxorubicin, cycophosphamide and paclitaxel
AE: adverse event
AI: aromatase inhibitor
AJCC: American Joint Committee on Cancer
ALT: alanine transaminase
ANA: anastrozole
ASCO‐CAP: American Society of Clinical Oncology ‐ College of American Pathologists
AST: aspartate aminotransferase
BC: breast cancer
BMD: bone mineral density
BMFS: bone metastases‐free survival 
BP: bisphosphonate
BRM: biological response modifiers
BSAP: bone‐specific alkaline phosphatase
BUN: blood urea nitrogen
CAF: cyclophosphamide, doxorubicin and fluorouracil
CEF: cyclophosphamide, epirubicin and 5‐fluouracil
CIS: carcinoma in stitu
CMF: cyclophosphamide + methotrexate + fluorouracil
CR: complete response
CrCl: creatinine clearance
CT: chemotherapy 
CTC: circulating tumour cell 
CTCAE: common terminology criteria of adverse events
CTX: C‐telopeptide
DCIS: ductal carcinoma in situ
DFS: disease‐free survival
DTC: disseminated tumour cells
DXA: dual‐energy x‐ray absorptiometry 
EC: epirubicin/cyclophosphamid 
ECG: electrocardiogram
ECOG: Eastern Cooperative Oncology Group
EC‐TX: epirubicin, cyclophosphamide, paclitaxel and capecitabine
EFS: event‐free survival
ER: oestrogen receptor 
ETC: epirubicin, paclitaxel and cyclophosphamide 
EXE: exemestane
FEC: fluorouracil/epirubicin/cyclophosphamid
FFPE: formalin‐fixed, paraffin‐embedded
FN: femur neck
FSH: follicle‐stimulating hormone
GBG: German Breast Group
GCS: Greene Climacteric Scale
GFR: glomerular filtration rate
GnRH: gonadotropin‐releasing‐hormone
H & E: hematoxylin and eosin
HDL: high‐density lipoprotein 
HER2: human epidermal growth factor receptor
HIV: human immunodeficiency virus
HR: hormone receptor
HRQoL: health‐related quality of life
HRT: hormone replacement therapy
HSA: hip structural analysis
IBD: irritable bowel disease
ISH: in‐situ hybridisation
ITT: intention‐to‐treat
IUD: intrauterine device
IV: intravenous
LCIS: lobular carcinoma in situ
LDL: low‐density lipoprotein 
LH: luteinising hormone
LHRH: luteinising hormone‐releasing hormone
LN: lymph node
LPBC: lymphocyte predominant breast cancer
LS: lumbar spine
LTZ: letrozole
MDRD: modification of diet in renal disease
MRI: magnetic resonance imaging
NABON: Nationaal Borstkanker Overleg Nederland (National Breast Cancer Consultation Netherlands)
NACT: neoadjuvant chemotherapy
NCI‐CTC: National Cancer Institute ‐ common toxicity criteria
NR: not reported
NTX: N‐telopeptide of type 1 collagen
NYHA: New York Heart Association
OCN: osteocalcin
OFS: ovarian function suppression
ONJ: osteonecrosis of the jaw
OR: oestrogen receptor
OS: overall survival
PA: posteroanterior
pCR: pathological complete response
PgR: progesterone receptor
PINP: procollagen type I N‐terminal peptide 
PR: partial response
QUS: quantitative ultrasound
RANKL: receptor activator of nuclear factor kappa‐Β ligand
RECIST: response evaluation criteria in solid tumours
R0: histologic complete resection
SC: subcutaneous
SD: standard deviation
SERM: selective oestrogen receptor modulator
SGOT: serum glutamic oxaloacetic transaminase
SGPT: serum glutamate pyruvate transaminase
SHBG: sex hormone binding globulin
SWOG: Southwest Oncology Group
TH: total hip
TIL: tumor‐infiltrating lymphocyte
TNBC: triple‐negative breast cancer
TNM: tumor, node, metastasis
TP: thymidine phosphorylase
TS: thymidylate synthase
ULN: upper limit of normal
VEGF: vascular endothelial growth factor
WBC: white blood cells
WHO: World Health Organisation
ZA: zoledronic acid
ZOL: zoledronate

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Ahmad 2007	Non‐randomised study design
Ahn 2009	Wrong comparison (upfront vs delayed bisphosphonate use)
BATMAN 2005	Wrong comparison (2 years vs 5 years of bisphosphonate use)
CALGB 79809	Wrong comparison (upfront vs delayed bisphosphonate use)
Chen 2011	Asked for translation, non‐randomised study design
Ciardo 2020	Non‐randomised study design
Fuleihan 2005	Inadequate randomisation and allocation concealment
Gessner 2000	Wrong comparison (compares 2 different doses of bisphosphonate)
Gucalp 1994	No subgroup contains only BC
Hines 2010	No comparison (only single‐arm study)
IBIS 3 FEASIBILITY	Only trial registry entries; study terminated, no results published
IBIS II 2003	Wrong study population (healthy women at high risk of BC)
JPRN‐UMIN000004375	Only trial registry entries; study terminated, no results published
Lee 2011	Non‐randomised study design
Lipton 1999	Population with bone metastases
N03CC	Wrong comparison (upfront vs delayed bisphosphonate use)
Nakatsukasa 2019	Non‐randomised study design
NCT00196859	Wrong comparison (analyses the effect of capecitabine rather than ibrandronate)
NCT00202059	Wrong comparison (bisphosphonates vs. physical activity)
NCT00247650	Only trial registry entries; study terminated, no results published
NCT00295867	Non‐randomised study design
NCT00324714	Only trial registry entries; study terminated, no results published
NCT00873808	Study withdrawn
NCT02051218	Wrong comparison (compares 2 different doses of denosumab)
NCT03358017	Zoledronic acid + atorvastatin vs. no treatment: concomitant medication not identical
NCT03664687	Wrong comparison (compared 2 different doses of bisphosphonate)
NCT05164952	Wrong comparator: upfront vs. delayed bisphosphonate use
PERIDENO	Only trial registry entries; study terminated, no results published
Purohit 1995	Subgroup BC not separately reported
Ralston 1997	Population without bone metastases not reported
Rizzoli 1996	Metastases after treatment reported
Smith 1999	Population with bone metastases
SUCCESS	Wrong comparison (3 years vs. 5 years of bisphosphonate use)
Takahashi 2012	Wrong comparison (upfront vs delayed bisphosphonate use)
Toulis 2016	Non‐randomised study design
Van Hellemond 2019	Non‐randomised study design for bisphosphonate use
Vinholes 1995	Only abstract given; not clear if subgroup BC reported
Vriens 2017	Summary of 2 studies; non‐randomised study design
Z‐FAST 2012	Wrong comparison (upfront vs delayed bisphosphonate use)
ZO‐FAST 2013	Wrong comparison (upfront vs delayed bisphosphonate use)
ZOLMENO 2017	Wrong comparison (upfront vs delayed bisphosphonate use)

BC: breast cancer

Characteristics of studies awaiting classification [ordered by study ID]

ACTRN12616001051437.

Methods	Setting: randomised, double‐blinded, phase 3, 1:1 proportion, placebo‐controlled Recruitment period: since 09/2016 Length of study: 12‐month study period Length of follow‐up: participants will be followed‐up at 6 and 12 months after starting the trial drug/placebo with scans
Participants	Eligibility criteria: minimum age: 18 years; maximum age: 55 years; female; premenopausal women with oestrogen‐receptor positive, non‐metastatic breast cancer (TxNxM0) based on documented pathological and radiological evaluation. Menopausal status will be defined clinically at the diagnosis of breast cancer; premenopausal: a regular cycle in the last 3 months prior to diagnosis of breast cancer; perimenopausal: absent cycles for 3‐12 months; postmenopausal: absent cycles for 12 months or more; about to commence treatment with ovarian suppression with a view to subsequent aromatase inhibition as determined by the treating oncologist; endocrine therapy intended for at least 12 months; eastern Cooperative Oncology Group (ECOG) status 0 and 1; able to personally read and understand the participant information and consent form and provide written, signed and dated informed consent to participate in the study; able and willing to meet all protocol‐required procedures and visits. Exclusion criteria: bone mineral density T‐score at the lumbar spine/hip/femoral neck < ‐2.0 SD; pre‐existing minimal trauma fractures (excluding fractures of fingers, toes, hands, feet and skull); current evidence or prior history of any of the following: metabolic bone disorder(s) drugs for treatment of bone‐related disorders prolonged glucocorticoid use for 2 or more weeks continuously in the past 6 months significant inflammatory or malabsorptive condition(s) osteonecrosis or osteomyelitis of the jaw atypical femoral fracture(s) diabetes mellitus history of any solid organ or bone marrow transplant malignancy within the last 5 years (except breast cancer and non‐melanoma skin cancers) abnormalities of the following per central laboratory reference ranges: hypo/hypercalcaemia hypo/hyperparathyroidism renal impairment (eGFR < 45 mL/min/1.73m2) 25‐hydroxy vitamin D deficiency (< 12 nmol/L). Repletion will be allowed and participants may be re‐screened; self‐reported recreational drug use or alcohol dependence within 12 months prior to screening; pregnancy; history or evidence of any other clinically significant disorder, condition or disease that, in the opinion of the study investigator, would pose a risk to participant safety or interfere with the study evaluation, procedures or completion. Mean age: not reported Menopausal status: pre, peri and postmenopausal RANKL status: not reported Hormone receptor status: not reported Human epidermal growth factor receptor 2 status: not reported Participants randomised: not reported total: Country of participants: Australia
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention: denosumab 60 mg subcutaneous injection at 6‐month intervals for a 12‐month study duration administered by trial investigator; control: placebo ‐ normal saline subcutaneous injection at 6‐month intervals for a 12‐month study duration administered by trial investigator; supplemental; other. Previous bone‐modifying interventions: not reported Cancer treatment during study period: not reported
Outcomes	Primary: total volumetric BMD at the distal tibia Secondary: areal bone mineral density; bone remodelling markers; lean mass; homeostatic model assessment of insuline resistance; lipid levels; quality of life; visceral fat volume and distribution; subcutaneous fat volume and distribution; total volumetric BMD at the distal radius; cortical volumetric BMD at the distal radius and distal tibia; trabecular volumetric BMD at the distal radius and distal tibia; cortical porosity at the distal radius and distal tibia; trabecular number, thickness and separation at the distal radius and distal tibia; matrix mineral density at the distal radius and distal tibia.
Notes	Funding sources: University of Melbourne Conflicts of interest: not reported

ADAIDO.

Methods	Setting: ‐ controlled, randomised, open Recruitment period: ‐ not reported Length of study: ‐ not reported Length of follow‐up: ‐ not reported
Participants	Eligibility criteria: ‐ maximum age: 75 years; ‐ minimum age: 18; ‐ postmenopausal women; ‐ treated for breast cancer with AI (letrozole, anastrozole, examestane) for at least two years; ‐ denosumab stopped at least 4 months before ICF signature, after at least a 2‐year treatment duration to prevent/treat the CTIBL (in primary prevention); ‐ affected with osteopenia, diagnosed as femoral T‐scores by DXA performed within the last 36 months from the AI discontinuation, within the range ‐1.0 to ‐2.4; ‐ with low risk fracture, defined as a 10‐year predicted fracture risk < 20% for major osteoporotic fractures and < 3% for femur fractures; ‐ stopping AI treatment within 6 months from the last denosumab administration; ‐ current supplementation with calcium and vitamin D (according to clinical routine practice). Exclusion criteria: ‐ age > 75 years; ‐ BMI < 20 or > 35 kg/m2; ‐ osteoporosis diagnosed as femoral T‐score by DXA ‐2.5; ‐ clinical or morphometric fractures detected by thoracic and lumbar Rx; ‐ recent invasive dental surgery with no complete healing at the moment of inclusion; ‐ type 1 diabetes mellitus; ‐ poorly controlled type 2 diabetes mellitus (HbA1c > 7.5%, 58 mmol/mol); ‐ rheumatoid arthritis; ‐ current steroid or immunosuppressive therapies; ‐ active endocrinopathies (except hypothyroidism with good hormonal balance); ‐ chronic alcoholism; ‐ chronic kidney disease stages 4‐5 according to CKD‐EPI (eGFR < 30 mL/min) (test performed as routine); ‐ hepatic cirrhosis, HCV and HBV‐related chronic hepatitis, autoimmune hepatitis (autodeclaration); ‐ previous treatments with amino‐bisphosphonates (except previous treatment with clodronate); ‐ known history of reflux oesophagitis; ‐ other known contraindications to bisphosphonates. Mean age: ‐ not reported Menopausal status: ‐ not reported RANKL‐status: ‐ not reported Hormone receptor status: ‐ not reported Participants randomised: ‐ total: 190 Country of participants: Italy
Interventions	Bone‐modifying treatment (dose) ‐ Intervention: alendronate 70 mg; ‐ Control: no treatment. Previous bone‐modifying interventions: ‐ denosumab/+ AI
Outcomes	Primary: ‐ percent changes in BMD measured at lumbar and femoral sites at 12 months with respect to basal conditions Secondary: ‐ changes in serum bone‐specific alkaline phosphatase activity, CTX, P1NP and FGF23; ‐ changes in BMD measured at lumbar and femoral sites at 24 months with respect to basal conditions; ‐ occurrence of new fragility fractures at any sites anamnestically recorded at 3, 12 and 24 months visits; ‐ detection of new morphometric vertebral fractures by thoracic and lumbar Rx at 12 and 24 months of follow‐up; ‐ incidence of adverse events/clinical and morphometric fractures in treated and untreated patients.
Notes	Conflicts of interest: ‐ not reported

ChiCTR‐TRC‐09000408.

Methods	Setting: randomised, double‐blinded, phase 4, parallel‐group Recruitment period: from 2009/07 to 2011/01 Length of study: from 2009/07 to 2015/01 Length of follow‐up: not reported
Participants	Eligibility criteria: minimum age: 18 years; maximum age: 70 years; original breast cancer at II and III stages, with evidence of lymph node metastasis, no local or distance metastasis, with the original cancer to be resected; good body condition, ECOG 0 or 1 score; received or is receiving combination chemotherapy and/or combination endocrinology therapy or radiotherapy; received or is receiving surgery combined with new chemotherapy and/or new endocrinology therapy or radiotherapy. Exclusion criteria: metastasis or recurrence of breast cancer; received organ transplantation including autologous marrow self‐transplantation and periphery stem cell transplantation; received long‐term systematic steroid treatment; in past five years suffered other tumours, except original cervical cancer or cured skin basal cell carcinoma or squamous cell carcinoma; suffered other non‐vicious system diseases that would impact the long‐term follow‐up; involved in another study; pregnancy or intending to become pregnant. Mean age: not reported Menopausal status: pre, peri and postmenopausal RANKL status: not reported Hormone receptor status: not reported Human epidermal growth factor receptor 2 status: not reported Participants randomised: not reported total: 240 Country of participants: China
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention: additional to the basal chemotherapy, zoledronic acid 4 mg/time, intravenous drip, 3~4 weeks once; after three times changed to three months once; total six times; control: basal chemotherapy same as group A without zoledronic acid
Outcomes	Primary: process of disease; mortality; quality of life. Secondary: rate of bone metastasis; disease‐free survival (DFS).
Notes

El‐Ibrashi 2016.

Methods	Setting: ‐ randomised Recruitment period: ‐ April 2005 to March 2012 Length of study: ‐ not reported Length of follow‐up: ‐ 98.4 months (median)
Participants	Eligibility criteria: ‐ premenopausal females who had undergone primary surgery for stage I, II ER +ve and/or PR +ve breast cancer with < 10 positive lymph nodes Exclusion criteria: ‐ not reported Mean age: ‐ not reported Menopausal status: ‐ premenopausal RANKL‐status: ‐ not reported Hormone receptor status: ‐ not reported Participants randomised: ‐ total: 300 Country of participants: ‐ not reported
Interventions	Bone‐modifying treatment (dose) ‐ intervention: zoledronic acid 4 mg every 6 months; ‐ control: no treatment. Other treatment: ‐ standard tamoxifen 20 mg/day for five years plus goserelin 3.6 mg every 28 days
Outcomes	Primary: ‐ toxicity; ‐ disease‐free survival (DFS); Secondary: ‐ overall survival (OS)
Notes	Conflicts of interest: ‐ not reported

Gunmalm 2018.

Methods	Setting: randomised Recruitment period: not reported Length of study: 12‐month study period Length of follow‐up: not reported
Participants	Eligibility criteria: early breast cancer Exclusion criteria: not reported Mean age: 66.7 years Menopausal status: not reported RANKL status: not reported Hormone receptor status: not reported Human epidermal growth factor receptor 2 status: not reported Participants randomised: not reported total: 60 Country of participants: not reported
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention: denosumab (oral, 60 mg) every 6 months; control: zoledronic acid (IV, 4 mg) every 6 months. Cancer treatment during study period: aromatase inhibitors (letrozole 2.5 mg)
Outcomes	Primary: calcium metabolic measures; markers of bone turnover.
Notes	Not yet verified whether it is an RCT design or not

NCT02595138.

Methods	Setting: randomised, parallel, open‐label, phase 3 Recruitment period: 2015‐2018 (final data collection date for primary outcome measure) Length of study: not reported Length of follow‐up: not reported
Participants	Eligibility criteria: adult women (≥ 18 years of age) with early stage breast cancer (stage II‐III); histological confirmation of oestrogen and/or progesterone‐receptor negative (ER‐), human epidermal growth factor receptor 2 negative (HER2‐) breast cancer. ER/PR negative: nuclear reaction < 1%, HER2 negative: HER2; IHC 0, 1+ or FISH/CISH (‐) in case of IHC 2+; axillary lymph node positive with other high risk factors: LVI, T > 5 cm, grade III; patients finished standard chemotherapy and/or radiotherapy; ECOG performance status of 0, 1; adequate bone marrow, hepatic, and renal function; adequate bone marrow and coagulation function as shown by: absolute neutrophil count (ANC) ≥ 1.5 109/L; platelets > 100 x 109/L; haemoglobin (Hgb) > 9.0 g/dL INR < 2; adequate liver function as shown by: serum aspartate aminotransferase (AST) and alanine aminotransferase (ALT) < 2.5 x ULN; total serum bilirubin < 1.5 x ULN; adequate renal function as shown by: serum creatinine < 1.5 x ULN; fasting serum cholesterol < 300 mg/dL or 7.75 mmol/L and fasting triglycerides < 2.5 x ULN. In case one or both of these thresholds are exceeded, the patient can only be included after initiation of statin therapy and when the above mentioned values have been achieved; written informed consent. Exclusion Criteria: another malignancy within 5 years prior to enrolment except for adequately treated in‐situ carcinoma of the cervix, uteri, basal or squamous cell carcinoma or non‐melanomatous skin cancer; any severe and/or uncontrolled medical conditions, e.g. currently active infection; pregnant or lactating; patients unwilling to or unable to comply with the protocol. Mean age: not reported Menopausal status: not reported RANKL status: not reported Hormone receptor status: not reported Human epidermal growth factor receptor 2 status: not reported Participants randomised: total: 430 Country of participants: South Korea
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention: zoledronic acid; control: no intervention; supplemental; other. Previous bone‐modifying interventions: not reported Cancer treatment during study period: not reported
Outcomes	Primary: disease‐free survival (time frame: 3 years after last patient was enroled), the time from randomisation to the time of disease progression or relapse or death Secondary: overall survival [time frame: 5 years after last patient was enroled], the time from randomisation to the time of death; number of participants with adverse events as a measure of safety and tolerability (time frame: up to 28 days after last medication).
Notes

Rhee 2011.

Methods	Setting: longitudinal, prospective Recruitment period: not reported Length of study: 6 months Length of follow‐up: not reported
Participants	Eligibility criteria: postmenopausal women with endocrine‐responsive breast cancer Exclusion criteria: not reported Mean age: postmenopausal women 54.5 years Menopausal status: postmenopausal RANKL status: not reported Hormone receptor status: not reported Human epidermal growth factor receptor 2 status: not reported Participants randomised: total: 45 Country of participants: Korea
Interventions	Bone‐modifying treatment (dose, frequency, length) intervention: 5 mg of alendronate with 0.5 mg of calcitriol for 6 months; control: placebo.
Outcomes	Primary outcome: sclerostin levels; BMD.
Notes	It is not clear whether this study is actually a subpopulation of Rhee (2013).

RISAROS 2009.

Methods	Setting: randomised, parallel‐assignment, double‐blind, placebo‐controlled, phase 3 Recruitment period: October 2013 (study completion date) Length of study: not reported Length of follow‐up: not reported
Participants	Eligibility Criteria: 18 years or older; postmenopausal women (more than one year since last menstrual period or removal of ovarian function by surgical or radiotherapy means); operated on for an invasive breast cancer (histologically proven); surgical treatment completed and cycles of adjuvant chemotherapy (if necessary) completed; treated with aromatase inhibitor; osteopenic (‐2.5 < T score < ‐1) without osteoporotic fracture; with written informed consent signed; with social security. Exclusion Criteria: women presenting with a history of osteoporotic fracture or a T score less than ‐2.5 of at least one measurement site; women presenting with clinical signs of metastases; received other hormonal treatment in the last 3 months; received treatment with bisphosphonates, raloxifene, tamoxifen, parathormone, strontium ranelate, tibolone, calcitonin and corticosteroids at more than 5 mg/d for 3 months in the last year; presenting with a known and untreated hyperthyroid; presenting with a known hyperadrenocorticism; treated and followed for Paget's disease of bone; presenting with an untreated primary hyperparathyroid; presenting with an indication against risedronate (known hypersensibility to risedronate monosodium and/or one of its excipients, non‐corrected hypocalcemia, pregnancy or breastfeeding, severe renal insufficiency inferior to 30 mL/min); presenting with malabsorption syndrome for glucose/galactose; participating in another clinical trial concerning a medicine susceptible of influencing bone mass. Mean age: not reported Menopausal status: aged ≥ 55 years, without menstruation aged < 55 years, with amenorrhoea for ≥ 12 months, or those diagnosed with menopause by attending physicians based on the FSH and oestradiol levels; underwent bilateral oophorectomy RANKL status: not reported Hormone receptor status: not reported Human epidermal growth factor receptor 2 status: not reported Participants randomised: total: 20 Country of participants: France
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention: drug: risedronate injection 35 mg oral risedronate once per week for 24 months + calcium and vitamin D; control: patients receive placebo 35 mg once a week plus a calcium and vitamin D supplementation. Previous bone‐modifying interventions: not reported Cancer treatment during study period: not reported
Outcomes	Primary: evolution of the lumbar spine bone mineral density after one year of treatment [time frame: 1 year] Secondary: evolution of femoral BMD after one year of treatment (time frame: 1 year); evolution of lumbar spine and femoral BMD after two years of treatment (time frame: 2 years); evolution of bone resorption and formation markers (time frame: 2 years); proportion of fractures after two years of treatment (time frame: 2 years); evolution of oestradiol levels (time frame: 2 years).
Notes

Xu 2010.

Methods	Setting: clinical trial Recruitment period: not reported Length of study: 24 months Length of follow‐up: not reported
Participants	Eligibility criteria: not reported Mean age: not reported Menopausal status: pre, peri and postmenopausal RANKL status: not reported Hormone receptor status: not reported Human epidermal growth factor receptor 2 status: not reported Participants randomised total: 63
Interventions	Bone‐modifying treatment intervention: zoledronic acid + aromatase‐inhibitor; control: aromatase‐inhibitor. Previous bone‐modifying interventions: not reported Cancer treatment during study period: not reported
Outcomes	Primary endpoint: lumbar spine bone mineral density (LS BMD) Secondary endpoints: incidence of osteopenia and osteoporosis
Notes	Conflicts of interest: not reported

Yonehara 2007.

Methods	Setting: randomised Recruitment period: not reported Length of study: 6 months Length of follow‐up: not reported
Participants	Eligibility criteria: postmenopausal women with a histologically confirmed diagnosis of oestrogen‐receptor‐positive breast cancer Exclusion criteria: ovarian tumour; past history of malignancy; postmenopausal hormone replacement therapy either previously or currently; use of tamoxifen; excess alcohol drinking and excess tobacco use. Mean age: intervention: 65.0 ± 5.1 control: 63.3 ± 9.9 Menopausal status: 100% postmenopausal RANKL status: not reported Hormone receptor status: 100% ER+ Human epidermal growth factor receptor 2 status: not reported Participants randomised: total: 27 Country of participants: Norway
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention: drug: risedronate injection AI + bisphosphonate (risedronate sodium, 2.5 mg daily) for 6 months; control: nonsteroidal AI (anastrozole, 1 mg daily); supplemental: not reported; other: not reported. Previous bone‐modifying interventions: surgical resection between July 2005 and June 2006 Cancer treatment during study period: aromatase inhibitor (anastrozole, 1 mg/day)
Outcomes	Primary: per cent change in BMD Secondary: not reported
Notes

AI: aromatase inhibitor
ALT: alanine transaminase
ANC: absolute neutrophil count
AST: aspartate aminotransferase
BMD: bone mineral density
CISH: chromogenic in situ hybridisation
CKD‐EPI: Chronic Kidney Disease Epidemiology Collaboration
CTIBL: cancer treatment induced bone loss
CTX: C‐telopeptide
DFS: disease‐free survival
DXA: dual‐energy x‐ray absorptiometry 
ECOG: Eastern Cooperative Oncology Group
eGFR: estimated glomerular filtration rate
ER: oestrogen receptor 
FGF23: fibroblast growth factor 23
FISH: fluorescence in situ hybridisation
FSH: follicle‐stimulating hormone
HbAIc: glycated haemoglobin
HBV: hepatitis B virus
HCV: hepatitis C virus
HER2: human epidermal growth factor receptor
Hgb: haemoglobin
ICF: informed consent form
IHC: immunohistochemistry 
INR: international normalised ratio
IV: intravenous
LS: lumbar spine
LVI: lymphovascular invasion
OS: overall survival
PINP: procollagen type I N‐terminal peptide 
PR: partial response
RANKL: receptor activator of nuclear factor kappa‐Β ligand
RCT: randomised controlled trial
SD: standard deviation
ULN: upper limit of normal

Characteristics of ongoing studies [ordered by study ID]

D‐BIOMARK.

Study name	An open label biomarker pilot study of the antitumoral activity of denosumab in the preoperative setting of early breast cancer Biomarker study of the antitumoral activity of denosumab in the pre operative setting of early breast cancer
Methods	Setting: randomised, 2:1 proportion, open‐label, early phase I Recruitment period: estimated: July 5, 2020 (final data collection date for primary outcome measure) Length of study: not reported Length of follow‐up: not reported
Participants	Eligibility criteria: women ≥ than 18 years, (the inclusion process will be modified to recruit at least 24 premenopausal patients); understand and sign Informed Consent for this study; capable, under investigator judgement, of understanding the non‐therapeutic nature of the study; diagnosed with invasive breast cancer in early, curable, stage (I or II); candidate for radical surgery as first therapeutic approach; Her2 negative receptor status; any oestrogen, progesterone status (the inclusion process will be modified to recruit at least 24 patients with TNBC tumours); adequate serum calcium or albumin‐adjusted serum calcium ≥ 2.0 mmol/L (8.0 mg/dL) and ≤ 2.9 mmol/L (11.5 mg/dL); general laboratory test within normality or with non‐relevant deviations of normality as per investigator judgement; normal organ and bone marrow function as defined by local standards: leukocytes, absolute neutrophil count, platelets, total bilirubin, AST/ALT/GOT/GPT, creatinine, creatinine clearance, magnesium, phosphorus; subject with reproductive potential must be willing to use, in combination with her partner, 2 acceptable methods of effective contraception or practice sexual abstinence throughout the study and continue for 6 months after study duration. Exclusion criteria: invasive breast cancer non‐amenable to surgical excision as first therapeutic approach; HER2‐positive breast cancer; metastatic breast cancer or other condition where the investigator recommends other treatment than surgery as the primary therapeutic approach; prior systemic treatment for any malignancy; treatment with denosumab contraindicated; bleeding diathesis or other concomitant condition that contraindicates inclusion in the study as per investigator judgement; high risk of ONJ or hypocalcemia: inadequate serum calcium or albumin‐adjusted serum calcium < 2.0 mmol/L (8.0 mg/dL) or > 2.9 mmol/L (11.5 mg/dL) prior history or current evidence of osteonecrosis of the jaw; active dental or jaw condition which requires oral surgery, including tooth extraction. Planned invasive dental procedures; known sensitivity to any of the products to be administered during the study; pregnant or breastfeeding or planning to become pregnant/breastfeed while on study through 6 months after the end of treatment; prior history or current evidence of osteonecrosis or osteomyelitis of the jaw; active dental or jaw condition which requires oral surgery, including tooth extraction; non‐healed dental or oral surgery, including tooth extraction; planned invasive dental procedures for the course of the study; ongoing treatment with denosumab or bisphosphonates. Mean age: not reported intervention: control: Menopausal status: not reported RANKL status: not reported Hormone receptor status: not reported Human epidermal growth factor receptor 2 status: not reported Participants randomised: not reported total: Country of participants: Spain
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention: denosumab 120 mg/1.7 mL subcutaneous solution [XGEVA®]; two injections on days 1 and 8 previous to surgery breast cancer excision; control: no intervention; supplemental; other. Previous bone‐modifying interventions: not reported Cancer treatment during study period: not reported
Outcomes	Primary: antiproliferative and/or pro‐apoptotic activity of denosumab Secondary: correlation between antiproliferative activity of denosumab and Rank/RankL expression; differential antiproliferative activity of denosumab amongst the different phenotypes of breast cancers; differential antiproliferative activity of denosumab amongst pre and postmenopausal patients; safety of denosumab and biopsy procedures in terms of frequency of adverse events (CTCAE V4).
Starting date	05/2018
Contact information	Contact: Eva González Suárez, PhD egsuarez@idibell.cat Contact: Silvia Casellas +34934515250 s.casellas@anagram‐esic.com
Notes	Funding sources: Conflicts of interest: not reported

ENDEAVOR.

Study name	A multicenter, randomised, comparative study regarding the efficacy of denosumab on normal bone mineral density in women receiving adjuvant aromatase inhibitors for early breast cancer (ENDEAVOR trial)
Methods	Setting: multicentre, randomised, comparative study, phase 3, open‐label Recruitment period: estimated: 09/2022 (primary completion date), 09/2027 (study completion date) Length of study: 5 years Length of follow‐up: twelve weeks after the completion of surgery or postoperative chemotherapy (the completion of chemotherapy refers to the completion of the final course, involving the recovery phase)
Participants	Eligibility criteria: infiltrative breast cancer; females ≥ 20 years; pathologically diagnosed with stage I, II, or IIIA breast cancer (Cancer Management Regulations, 11th version); underwent appropriate surgery, such as mastectomy and breast‐preserving surgery; Oestrogen receptor (ER)‐ or progesterone receptor (PgR)‐positive patients on immunohistochemical (IHC) staining; females meeting one of the following criteria for menopause: those aged ≥ 55 years, without menstruation those aged < 55 years, with amenorrhoea for ≥ 12 months, or those diagnosed with menopause by attending physicians based on the FSH and oestradiol levels those who underwent bilateral oophorectomy; BMD for the lumbar vertebrae (L1‐L4) on DXA before the start of this study is ≥ ‐1.0 SD of the mean value of young adult females (YAM), and the BMD for the femoral neck is ≥ ‐1.0 SD of YAM; no lumbar vertebral or femoral fracture; ECOG PS of 0‐2; adequate organ functions (laboratory data within 4 weeks before case registration) leukocyte count, ≥ 3000/mm3 or neutrophil count, ≥ 1500/mm3 AST, ALT, ≤ 1.5‐fold of the upper limit of the institutional reference range; serum creatinine, ≤ 1.5‐fold of the upper limit of the institutional reference range; case registration should be performed before the following point: twelve weeks after the completion of surgery or postoperative chemotherapy (the completion of chemotherapy refers to the completion of the final course, involving the recovery phase); interval of ≥ 4 weeks after the discontinuation of therapy with bisphosphonates (oral preparations), oestrogen preparations, raloxifene, calcitonin preparations, vitamin K preparations, active vitamin D preparations, or ipriflavone preparations, which influence bones; those from whom written informed consent regarding study participation was obtained. Exclusion criteria: patients in whom distant metastasis was confirmed clinically or using imaging procedures at the time of case registration; those with bilateral breast cancer; those for whom postoperative hormonal therapy was started before consenting to study participation; those who received endocrine therapy within 52 weeks before consenting to study participation; those to whom bisphosphonate preparations were intravenously administered within 52 weeks before consenting to study participation; those with the following diseases that may affect DXA severe scoliosis, immobility, hyperostosis or osteosclerosis of the lumbar vertebrae, calcification of the abdominal aorta, and vertebral disease; those with a history of malignant tumours other than breast cancer within 260 weeks before consenting to study participation; those with dental diseases, such as infectious diseases of the teeth or jaw and tooth trauma. Those for whom tooth or jaw surgery is scheduled within 6 weeks after consenting to study participation (tooth extraction, implantation); others who are considered to be ineligible by the chief investigator. Mean age: intervention: control: Menopausal status: those aged ≥ 55 years, without menstruation; those aged < 55 years, with amenorrhoea for ≥ 12 months, or those diagnosed with menopause by attending physicians based on the FSH and oestradiol levels; those who underwent bilateral oophorectomy. RANKL status: not reported Hormone receptor status: not reported Human epidermal growth factor receptor 2 status: not reported Participants randomised: not reported total: Country of participants: Japan
Interventions	Bone‐modifying treatment (dose, application, frequency, length) intervention: drug: denosumab injection AI intake + denosumab injection per 6 months drug: denosumab injection; control: only AI; supplemental; other. Previous bone‐modifying interventions: not reported Cancer treatment during study period: not reported
Outcomes	primary: percentage change in the bone mineral density (BMD) for the lumbar vertebrae (L1‐L4) on dual‐energy X‐ray absorptiometry (DXA) [time frame: 12 months after the start of this study]; the change is a value obtained by subtracting 1 from the BMD after 12 months/baseline BMD is expressed as a percentage. Secondary: percentage change in the BMD for the lumbar vertebrae (L1‐L4) on DXA [time frame: after 2, 3, 4, and 5 years]; percentage change in the BMD for the femoral neck: after 2/3/4/5 years; percentage change in the BMD for the femoral neck [time frame: after 12 months and 2/3/4/5 years]; percentage change in the BMD for the femoral neck: after 12 months and 2/3/4/5 years; percentage change in the BMD for the radius (an ultrasonic bone densimeter is used), (only institutions in which ultrasonic bone densimeters are used) [time frame: after 2 and 4 weeks, every 4 weeks thereafter (for 2 years after registration)(only]; percentage change in the BMD for the radius (an ultrasonic bone densimeter is used): after 2 and 4 weeks, every 4 weeks thereafter (for 2 years after registration)(only institutions in which ultrasonic bone densimeters are used); changes in Ca and bone metabolism markers [time frame: after 24 weeks]; changes in Ca (mg/dL corrected by albumin level) and bone metabolism markers such as TRAP5b, bone‐specific alkaline phosphatase (BSAP), blood pentosidine by blood sampling every 6 months; appearance rate of morbid fracture in all participants [time frame: up to 3 years]; appearance rate of morbid fracture up to 3 years in all participants. Morbid fractures include all types of fractures; disease‐free survival [time frame: at least 5 years]; disease‐free survival at the end of the study; overall survival [time frame: at least 5 years]; overall survival at the end of the study; appearance of adverse events [time frame: at least 5 years]; appearance rate of adverse events (such as hypocalcemia and necrosis of the jaw); quality of life (QOL) [time frame: after 24 weeks]; quality of life (QOL), Japanese version of Euro‐Qol (EQ‐5D‐5L) evaluated by questionnaire every 6 months.
Starting date	04/2017
Contact information	Hisako Ono, PhD, hisako‐o@koto.kpu‐m.ac.jp Tetsuya Taguchi, PhD, ttaguchi@koto.kpu‐m.ac.jp
Notes	Funding sources: Japanese Breast Cancer Society Conflicts of interest: not reported

AI: aromatase inhibitor
ALT: alanine transaminase
AST: aspartate aminotransferase
BMD: bone mineral density
BSAP: bone‐specific alkaline phosphatase
CA: calcium
CTCAE: common terminology criteria of adverse events
DXA: dual‐energy x‐ray absorptiometry 
ECOG: Eastern Cooperative Oncology Group
EQ‐5D‐5L: European Quality of Life 5 Dimensions 5 Level version
ER: oestrogen receptor 
FSH: follicle‐stimulating hormone
GOT: glutamic oxaloacetic transaminase
GPT: glutamate pyruvate transaminase
IHC: immunohistochemical 
ONJ: osteonecrosis of the jaw
PgR: progesterone receptor
PS: performance status
QOL: quality of life
RANKL: receptor activator of nuclear factor kappa‐Β ligand
SD: standard deviation
TNBC: triple‐negative breast cancer
YAM: young adult females

Differences between protocol and review

Comparisons

As currently only one RANKL‐inhibitor (denosumab) is evaluated in randomised controlled studies, we could not assess the following comparison of two different RANKL‐inhibitors: one type of RANKL‐inhibitor versus another type of RANKL‐inhibitor.

Outcomes

1. We planned to analyse the outcome 'proportion of participants with pain response' and therefore extract data from pain scores and analgesic consumption. We also planned to analyse the adverse event 'bone pain right after administration'. When extracting data from studies, we realised that 'bone pain' was reported as an adverse event and therefore analysed this outcome. None of the studies reported on 'proportion of participants with pain response' which, in the case of non‐metastasised patients, would also be an outcome of interest, when analysing the effects of bisphosphonates and RANK‐L inhibitors. Therefore, we did not focus on pain response but rather bone pain as an adverse event in this analysis.

2. We considered adverse events independent of their grade. We planned to only focus on grade 3‐4 adverse events, but then realised that most studies did not report the grade of the occurring adverse event. Therefore, we decided to analyse adverse events independent of the grade.

3. Initially we planned to analyse the outcomes: bone mineral density, quality of life and survival outcomes at four time points of follow‐up. However, because of data availability, we did not analyse these outcomes at different time points but at longest follow‐up.

Grading

We initially planned to use the GRADEpro GDT online program to assess the certainty of the evidence. Since the program is not designed to evaluate the results of network meta‐analysis, we created Summary of Findings tables manually.

Subgroup analyses

We planned to conduct the following subgroup analyses on all efficacy and safety outcomes for network meta‐analysis, but unfortunately data were not available to perform them.

• Participants receiving endocrine therapy versus those not receiving endocrine therapy, or hormone receptor (HR)‐positive versus HR‐negative, also compared to human epidermal growth factor receptor 2 (HER2)‐positive. Because of lack of data availability, this subgroup analysis could not be performed.

• Type of endocrine therapy (e.g. tamoxifen alone versus aromatase inhibitor alone versus ovarian function suppression (OFS) in combination with tamoxifen versus OFS in combination with aromatase inhibitor). Because of lack of data availability, this subgroup analysis could not be performed.

• Type of bone‐modifying agent (bisphosphonate versus RANKL inhibitor). Since this corresponds to the main analysis using network meta‐analysis, no subgroup analysis was carried out.

• Bisphosphonates of the first (non‐amino bisphosphonates: etidronate, clodronate) and second generation (amino‐bisphosphonates: alendronate, risedronate, pamidronate, ibandronate, zoledronate) were to be assessed, independently. Since this would only exclude clodronate from the network, we did not conduct this analysis.

• Duration of bone‐modifying intervention: one year versus two‐to‐five years. Since in most studies, duration of bone‐modifying agents was longer than one year, we did not conduct this subgroup analysis.

• Participants with high risk of relapse (defined as receiving chemotherapy additionally to endocrine therapy) versus participants only receiving endocrine therapy. Because of lack of data availability, this subgroup analysis could not be performed.

• Participants with status N1, N2, N3 versus status N0. Because of lack of data availability, this subgroup analysis could not be performed.

Contributions of authors

Anne Adams: review development, screening, data extraction, risk of bias assessment, grading, statistical evaluation, interpretation of results, writing of the review

Tina Jakob: review development, screening, data extraction, risk of bias assessment, writing of the review

Alessandra Huth: screening, risk of bias assessment, writing of the review

Ina Monsef: search strategy development

Marco Kopp: data extraction, risk of bias assessment

Julia Caro‐Valenzuela: data extraction, risk of bias assessment

Moritz Ernst: screening, methodological expertise, interpretation of results

Achim Wöckel: clinical expertise, interpretation of results

Nicole Skoetz: review development, methodological expertise, screening, data extraction, grading, interpretation of results

Sources of support

Internal sources

• University Hospital Cologne, Department I of Internal Medicine, Center for Integrated Oncology Aachen Bonn Cologne Duesseldorf, Germany

  Provision of the offices, including technical equipment

• Institute of Medical Statistics and Computational Biology, Germany

  Support with statistical expertise

External sources

• Federal Ministry of Education and Research, Germany

  Grant number: 01KG1806

Declarations of interest

Action: edits to be checked by the author team

Anne Adams: awarded a grant from the Federal Ministry of Education and Research to perform this systematic review; this did not lead to a conflict of interest, as she is a statistical editor with Cochrane Haematology, but was not involved in the editorial process for this
review.

Tina Jakob: awarded a grant from the Federal Ministry of Education and Research to perform this systematic review; this did not lead to a conflict of interest.

Ina Monsef: none known; she is the Information Specialist of Cochrane Haematology, but was not involved in the editorial process for this review.

Alessandra Huth: none known.

Marco Kopp: none known.

Julia Caro‐Valenzuela: none known; a member of Cochrane Haematology, but was not involved in the editorial process for this review.

Moritz Ernst: none known; he is a member of Cochrane Haematology, but was not involved in the editorial process for this review.

Achim Wöckel: received consultancy fees to provide advice on metastatic breast cancer topics for Amgen (until end of 2020), Eli Lilly (ongoing), Hoffman‐La Roche (ongoing), Novartis (ongoing) and Pfizer (ongoing). No relevant relationships were declared in relation to early breast cancer topics.

Nicole Skoetz: none known; she is Co‐ordinating Editor of Cochrane Haematology, but was not involved in the editorial process for this review.

New