Metformin for Cancer: How the Biggest RCT Killed the Hypothesis

The Story in Brief

For over a decade, metformin was the darling of cancer repurposing. Hundreds of observational studies showed that diabetics taking metformin had 20-40% lower cancer rates. The mechanism made sense: metformin activates AMPK, suppresses mTOR, lowers insulin and IGF-1. Researchers were excited. Trials were launched.

Then came MA.32. The largest, most rigorous randomized controlled trial ever conducted on a repurposed cancer drug. 3,649 non-diabetic breast cancer patients. Metformin 850mg twice daily vs. placebo. Five years of treatment.

Result: Hazard ratio 1.01 (95% CI 0.84-1.21, P=0.93) for invasive disease-free survival. Absolutely nothing.

What the Observational Data Showed

The excitement was understandable. Study after study found:

Diabetics on metformin had 20-40% lower cancer incidence than diabetics on other medications
Signals appeared across multiple cancer types: breast, colorectal, pancreatic, lung, prostate
The mechanism was plausible (AMPK/mTOR/insulin axis)
Metformin was cheap, safe, and available everywhere

This generated enormous research investment. ClinicalTrials.gov lists over 100 metformin-cancer trials.

Why the Observational Data Was Wrong

A 2023 paper in Diabetes Care dissected the methodological problems:

Immortal Time Bias

This is the key flaw. To be classified as a "metformin user" in observational studies, a patient had to survive long enough to receive a prescription, fill it, and take it for some minimum period. The time between diabetes diagnosis and metformin initiation was counted as "exposure time" even though the patient wasn't taking the drug yet. During this immortal time, the patient could not have the outcome (cancer death) attributed to the metformin group. This systematically inflated the apparent benefit.

Healthy User Bias

Patients who take metformin consistently tend to be more health-conscious overall. They're more likely to attend screenings, follow medical advice, exercise, and manage their diet. The "metformin effect" may have been a "compliant patient effect."

Confounding by Indication

Patients given metformin (first-line for mild diabetes) tend to be healthier than patients given insulin or sulfonylureas (used for more advanced diabetes). Comparing metformin users to non-metformin diabetics is comparing healthier to sicker patients.

The MA.32 Trial

Design: Multicenter, double-blind, placebo-controlled RCT
Patients: 3,649 non-diabetic women with high-risk operable breast cancer
Intervention: Metformin 850mg twice daily vs. matching placebo
Duration: 5 years of treatment
Primary endpoint: Invasive disease-free survival
Result: HR 1.01 (95% CI 0.84-1.21, P=0.93)
Subgroup analyses: HER2-negative, ER/PR-positive subgroups showed no benefit either
Secondary analysis: Metformin did not reduce risk of new primary cancers

Published in JAMA 2022. This is as close to a definitive negative result as cancer research gets.

The Meta-Analysis Context

A meta-analysis of 22 RCTs published in BMC Medicine (2022) found mixed results across cancer types. Some subgroup signals existed but the overall evidence was weak. The metformin-cancer narrative has essentially collapsed under the weight of rigorous evidence.

Is Metformin Completely Useless for Cancer?

Probably not completely. There may be specific subgroups where it helps:

Obese patients with hyperinsulinemia (where the insulin-lowering mechanism matters most)
Specific molecular subtypes of cancer (PI3K-mutated tumors, potentially)
As a metabolic optimization tool alongside other interventions

But the blanket "metformin prevents cancer" narrative is dead. If you're taking metformin for diabetes, there's no reason to stop. But starting it specifically for cancer prevention is not supported by the best available evidence.

The Lesson

Metformin's story is a cautionary tale for all repurposed drug research:

"Kills cancer in a dish" means almost nothing. Cell culture results rarely translate to humans.
Observational data is treacherous. Immortal time bias, healthy user bias, and confounding can create signals that don't exist.
Only RCTs settle the question. Until a drug has been tested in a properly randomized trial, all other evidence is hypothesis-generating, not hypothesis-confirming.

This is why we grade evidence strictly on this site. Mechanism + petri dish + epidemiology is not enough. The bar is human trial data.

Sources

Goodwin PJ et al. JAMA 2022;327:1963-1973: MA.32 trial. jamanetwork.com
Diabetes Care 2023;46(5):904: Immortal time bias analysis
BMC Medicine 2022: Meta-analysis of 22 RCTs. bmcmedicine.biomedcentral.com
JCO 2023: MA.32 secondary analysis (new primary cancers)