Breast Cancer Prevention Drugs: What an NMA Ranking Means

The ranking is provocative. The evidence behind it is uneven.

A new network meta-analysis put two diabetes drug classes above tamoxifen for preventing breast cancer. That sounds like a practice-changing result. It is not.

The analysis found relative risks of 0.18 for sulfonylureas and 0.25 for thiazolidinediones versus placebo, both statistically significant. But those estimates came from sparse breast cancer events recorded as secondary outcomes in diabetes trials, then compared indirectly with established prevention drugs through placebo. Tamoxifen, raloxifene, exemestane, and anastrozole remain the medications with guideline-backed evidence for women at elevated breast cancer risk.

The more credible signal in the analysis comes from newer selective estrogen receptor modulators, or SERMs: arzoxifene, bazedoxifene, and lasofoxifene. They remain investigational for breast cancer prevention, but their biology fits the question better than the apparent sulfonylurea result.

What the 43-trial analysis found

Pourali et al.'s 2025 network meta-analysis pooled 43 randomized controlled trials and ranked medications by their probability of being among the most effective options, using SUCRA scores. A SUCRA score is a ranking metric, not a measure of how much a drug reduces a person's absolute risk.

Sulfonylureas ranked first, with a SUCRA of 0.92 and a relative risk of 0.18 versus placebo (95% confidence interval, 0.04 to 0.91). Thiazolidinediones ranked next at 0.83, with a relative risk of 0.25 (95% CI, 0.08 to 0.78). Both confidence intervals exclude 1.0, so the model estimated statistically significant reductions compared with placebo.

Tamoxifen had a SUCRA of 0.23 and a reported number needed to treat of 150. That low rank does not show that tamoxifen performs poorly in real-world prevention. It partly reflects a feature of rankings: highly uncertain estimates can rise to the top when their point estimates look dramatic.

Within the SERM subgroup, arzoxifene, bazedoxifene, and lasofoxifene ranked above tamoxifen and raloxifene. These drugs are not FDA-approved for breast cancer prevention, and none has completed a dedicated prevention trial with breast cancer incidence as a pre-specified primary outcome.

Why a network ranking can mislead

Network meta-analysis can estimate comparisons that no trial tested directly. Here, drug classes were connected through placebo-controlled trials. The model can therefore compare sulfonylureas with tamoxifen without a trial that assigned women to one or the other.

That approach works best when the linked trials ask sufficiently similar questions. This network spans SERMs, aromatase inhibitors, diabetes drugs, aspirin, statins, and other interventions tested in different populations and trial settings. The analysis used a common between-study heterogeneity parameter across those comparisons. That assumption may not hold when the drugs, participants, outcome ascertainment, and trial purposes differ so sharply.

The diabetes-drug estimates also look fragile on their face. Sulfonylureas had a confidence interval from 0.04 to 0.91: a 22-fold span between its bounds. Thiazolidinediones ranged from 0.08 to 0.78. Such wide intervals point to very few breast cancer events. The point estimates imply large reductions, but the true effect could be far smaller than the headline numbers suggest.

More importantly, the breast cancer outcomes came from diabetes management trials as secondary endpoints, not from prevention trials designed and powered to measure breast cancer incidence. Secondary outcomes can be informative, but they may be inconsistently captured or adjudicated. Combining them in a network does not turn them into dedicated cancer-prevention evidence.

Diabetes drugs are the least actionable result

The populations in the diabetes trials were predominantly women with type 2 diabetes or prediabetes. They differ from the elevated-risk, otherwise healthy women who typically consider breast cancer chemoprevention. Diabetes, obesity, insulin resistance, inflammation, hormonal status, and comorbidity can all affect both cancer risk and the meaning of a drug's risks and benefits.

There is a biologically plausible case for thiazolidinediones, at least as a hypothesis. A 2025 review by Albini et al. describes how hyperinsulinemia, chronic inflammation, and dysregulated metabolic signaling can create cancer-promoting environments in obesity and type 2 diabetes. Because thiazolidinediones act through PPAR-gamma, they could plausibly affect some of those pathways. Plausibility is not proof of prevention efficacy.

The sulfonylurea result faces the opposite problem. Sulfonylureas stimulate insulin secretion, while chronic hyperinsulinemia is part of the metabolic environment implicated in tumor-promoting signaling. That does not prove sulfonylureas increase breast cancer risk, but it makes a relative risk of 0.18 biologically difficult to accept without direct replication.

A 2023 Mendelian randomisation analysis by Yarmolinsky et al. used genetic proxies for glucose-lowering drug targets to examine cancer risk and described the epidemiology as conflicting. It did not establish a definitive protective causal signal for glucose-lowering drug targets. The designs answer different questions, but the genetic evidence provides no independent reason to treat the network ranking as clinical guidance.

Safety makes the practical answer clearer. Thiazolidinediones have class-level concerns including heart failure exacerbation and increased fracture risk in women. Sulfonylureas lower glucose by increasing insulin secretion, an unsuitable mechanism for a person without diabetes seeking preventive treatment. Neither class should be used for breast cancer prevention outside a clinical trial.

Why newer SERMs deserve attention

Newer SERMs are a more coherent lead for future research. SERMs can block estrogen signaling in breast tissue while producing different effects in the uterus and bone. A 2021 review by Pinkerton et al. describes meaningful differences in tissue selectivity among these agents, including newer options with more favorable uterine and bone profiles than older SERMs.

A 2025 pharmacology review by Stanczyk et al. likewise notes that SERMs vary substantially in estrogen-receptor affinity and pharmacokinetics. That gives the higher rankings for arzoxifene, bazedoxifene, and lasofoxifene a reasonable mechanistic basis. It does not establish that any one of them prevents more breast cancers than tamoxifen.

Dedicated trials would need to test that claim in women selected for elevated breast cancer risk, with breast cancer incidence as a pre-specified primary endpoint and enough events to estimate both benefit and harm precisely.

What remains established

Jahan et al.'s review of endocrine prevention summarizes the established landscape: tamoxifen and raloxifene have large dedicated prevention-trial evidence and guideline support for women at elevated risk. Exemestane and anastrozole are additional endorsed options for postmenopausal women. These drugs target estrogen-related biology, the best-validated pharmacologic route to reducing breast cancer risk.

Hormonal exposure also matters in the opposite direction. A 2026 meta-analysis of menopausal hormone therapy found an overall association with higher breast cancer incidence across 34 studies (odds ratio 1.15, 95% CI 1.09 to 1.22), with substantial heterogeneity. The result does not answer which prevention drug to take, but it reinforces why estrogen signaling remains central to prevention research.

For a woman considering chemoprevention, a SUCRA leaderboard should not outrank trial design, absolute risk, adverse effects, and a clinician's assessment of her situation.

The network meta-analysis is useful because it spots hypotheses worth testing. Its diabetes-drug rankings are not treatment recommendations. Its most plausible contribution is a clearer research target: newer SERMs may deserve properly designed breast cancer prevention trials.