Do GLP-1 Drugs Raise Cancer Risk? What Trials Show

Current randomized-trial evidence does not show that GLP-1 drugs raise cancer risk over roughly 2 to 5 years. That matters because medicines such as Ozempic, Wegovy, and Mounjaro carry a thyroid-tumor warning rooted in rodent studies, while more than 40 million people now use this drug class. The reassuring part is real. So is the boundary around it: trials cannot answer whether a cancer risk appears after a decade of treatment.

The largest analysis so far, Ko et al. (2026), combined 48 randomized controlled trials with 94,245 participants. Across 13 obesity-related cancer types, no pooled result reached statistical significance. Every confidence interval crossed 1.0, which means the trials did not establish either increased or reduced risk.

What the best trial evidence actually found

Randomized trials offer the cleanest comparison because participants are assigned to a GLP-1 receptor agonist or comparator rather than selecting treatment based on their health, access, or clinician. But these trials were built mainly to measure metabolic and cardiovascular outcomes. Cancer was a secondary or exploratory outcome, and cancer events were rare.

For four cancers, Ko and colleagues rated the evidence moderate certainty. None showed a statistically significant difference:

• Thyroid cancer: odds ratio (OR) 1.37, 95% confidence interval (CI) 0.82–2.31; an estimated absolute difference ranging from 1 fewer to 9 more cases per 10,000 treated.
• Pancreatic cancer: OR 0.84, 95% CI 0.53–1.35; 9 fewer to 6 more cases per 10,000 treated.
• Breast cancer: OR 0.95, 95% CI 0.60–1.49; 10 fewer to 12 more cases per 10,000 treated.
• Kidney cancer: OR 1.12, 95% CI 0.78–1.60; 5 fewer to 13 more cases per 10,000 treated.

Those ranges matter more than a bare statement that the findings were "neutral." The pancreatic estimate points modestly below 1.0, for example, but its confidence interval includes both lower and higher risk. It does not prove protection. The thyroid estimate sits above 1.0, but it also does not prove harm.

For colorectal, esophageal, liver, gallbladder, ovarian, endometrial, gastric, multiple myeloma, and meningioma outcomes, the evidence was low to very uncertain because there were too few events. Gastric cancer had especially little usable trial evidence. A null result from sparse data is not a clean bill of health; it is an unanswered question with a confidence interval attached.

Why the thyroid warning has remained so unsettling

The boxed warning traces back to rodents. In those studies, chronic GLP-1 receptor stimulation produced thyroid C-cell changes that could progress through hyperplasia and adenoma to carcinoma. Regulators had a sensible reason to take that finding seriously rather than wave it away because mice are, in many respects, terrible tiny humans.

Human thyroid biology appears materially different. Vilsbøll et al. (2026), a thyroid-cancer-focused review, found no clinical evidence that GLP-1 receptor agonists cause thyroid cancer in people. Its mechanistic argument is specific: rodent thyroid C-cells express GLP-1 receptors densely, whereas human thyroid C-cells have negligible to undetectable receptor expression. If the receptor is largely absent, the rodent C-cell pathway has no obvious pharmacologic route in human tissue.

That biology makes a human thyroid-cancer effect less plausible, but it does not erase statistical uncertainty. Ko et al.'s thyroid estimate had an upper confidence-limit of 2.31. This is not evidence of a 130% increase; it is the largest increase the available trial data could not rule out. Rare cancers make even large trial programs imprecise.

Pancreatic cancer: a concern that has not held up in trials

Early concern about pancreatic toxicity came from mechanistic hypotheses and adverse-event reporting, neither of which can reliably establish that a drug causes cancer. Controlled evidence has been more consistent. A pancreatic-focused randomized-trial meta-analysis by Nreu et al. (2023) found no significant association between GLP-1 receptor agonist use and pancreatic cancer in people with type 2 diabetes.

Ko et al.'s pancreatic OR of 0.84 (95% CI 0.53–1.35) fits that earlier null result. A separate broad review, Jaradat et al. (2025), also concluded that clinical data do not consistently support the thyroid and pancreatic signals raised by animal studies and pharmacovigilance reports. Independent reviews can overlap in their included trials, so they are not three entirely separate experiments. Still, their conclusions converge.

The long-term question remains open

The central limitation is follow-up. The randomized trials in the 2026 meta-analysis had median follow-up periods of 2 to 5 years. Many solid tumors take 5 to 20 years to develop and become clinically detectable. These data support safety during the period observed; they cannot exclude a risk that requires longer exposure or latency to emerge.

A nationwide retrospective cohort study by Levy et al. (2025) adds longer post-market observation in 1.1 million patients. Unlike the class-pooled trial meta-analysis, it reported different cancer-risk profiles among individual GLP-1 drugs. That finding deserves attention, but it cannot establish that one drug caused a difference. Patients prescribed different agents can differ in obesity severity, diabetes, other illnesses, cancer screening, treatment access, and baseline risk. Those problems—confounding by indication, channeling bias, and differential surveillance—can create apparent drug differences in observational data.

The discrepancy with the trial meta-analysis is therefore unresolved rather than mysterious. Trials provide stronger protection against confounding but had too few cancer events and too little follow-up to detect small, agent-specific effects. The large cohort has statistical scale and longer real-world exposure but weaker causal control. Agent-specific registries with long follow-up will matter more than another broad class average.

Could effects differ by tumor type?

They could, at least biologically. Ungvari et al. (2025) analyzed GLP-1 receptor expression across multiple tumor types and found that expression varied substantially and correlated with survival in a tumor-type-specific way. This establishes an association, not proof that GLP-1 treatment changes a tumor's behavior. It does explain why a single pooled estimate across cancer types could miss effects limited to receptor-expressing tumor subtypes.

Weight loss complicates interpretation further. Lower adiposity can reduce obesity-related cancer risk through changes in insulin and IGF-1 signaling, sex-hormone bioavailability, and inflammatory adipokine activity. If a study eventually finds lower cancer rates among GLP-1 users, that result may reflect weight loss rather than a direct anticancer effect of the medication. Current evidence cannot separate those explanations cleanly.

What this means if you take a GLP-1 drug

The evidence does not support stopping a GLP-1 receptor agonist because of a general cancer concern within the studied 2-to-5-year window. Across 48 trials and 94,245 participants, no statistically significant cancer increase emerged, including for thyroid and pancreatic cancer. Modern-agent sensitivity analyses for semaglutide and tirzepatide were consistent with the overall result.

Current evidence is reassuring for short- to medium-term use, not a lifetime oncology guarantee.

People with active cancer, a high cancer risk, or questions about a specific agent should discuss their individual situation with their treating clinician. The trial evidence is strongest for the four cancer outcomes with moderate-certainty ratings and weaker for the other nine. For everyone else, the honest answer is pleasantly unglamorous: current human data do not show a cancer signal, while decade-scale risk remains unmeasured.