GLP-1 Thyroid Cancer Risk: What the Boxed Warning Actually Says

The thyroid warning on every semaglutide and tirzepatide label comes from tumors that appeared in rodents, not from a rate measured in people. FDA can require a boxed warning on serious animal toxicity alone, and did. Whether the finding carries over to humans is still unknown, and the human studies published since do not agree with each other. What is not in doubt is the contraindication: if you or a blood relative has had medullary thyroid carcinoma, or you have MEN 2, these drugs are labeled as not for you.^[1]

What the warning actually says

Two sentences sit inside the box on the Ozempic label, and the second one is the honest half. The first attributes the tumors to rodents, says they grew both with the size of the dose and with how long dosing continued, and describes the exposures involved as clinically relevant, meaning within reach of what a person on the drug actually reaches. The second says the jump to people is unknown, and uses that word: whether the drug does this in humans, medullary thyroid carcinoma included, is written as something that “has not been determined”. That is a label marking the outer edge of its own evidence, on its own front page.^[1]

Wegovy carries an identical box with its own brand name swapped in, and so do the oral semaglutide tablets sold as Rybelsus and Ozempic tablets.^[2][3] Five US labels, one finding, and the same admission attached to each.

The tirzepatide labels say the same thing about a different animal. Mounjaro attributes the tumors to “both sexes of rats” and Zepbound to rats, where the semaglutide labels say rodents and mean mice as well.^[4][5] That difference is small on the page and real underneath it, and we come back to it below.

A boxed warning is a disclosure, not a measurement

This is the part almost every summary of the subject skips. A boxed warning is the most serious warning FDA can require on a drug label, and the rule that creates it says explicitly what it can be built on.^[6]

Certain contraindications or serious warnings, particularly those that may lead to death or serious injury, may be required by the FDA to be presented in a box. The boxed warning ordinarily must be based on clinical data, but serious animal toxicity may also be the basis of a boxed warning in the absence of clinical data.

— 21 CFR 201.57(c)(1), eCFR title 21 current as of August 13, 2026

The GLP-1 thyroid warning is the second case in that sentence. Serious animal toxicity, in the absence of clinical data. The regulation anticipated exactly this situation and told FDA it may box the finding anyway. So the box is doing what it was designed to do, which is force a disclosure onto the front page of the label while the human question stays open. It is not the agency reporting that the drug causes thyroid cancer in people, and reading it that way gets the meaning backwards.

Where the rodent finding came from

The label's Nonclinical Toxicology section carries the study detail, and it is more specific than the box. In a two-year study in CD-1 mice, semaglutide produced a statistically significant increase in thyroid C-cell adenomas at every dose tested, at exposures the label describes as clinically relevant. In a two-year Sprague Dawley rat study, adenomas rose at every dose in both sexes and carcinomas rose in males, at exposures up to roughly seven tenths of what a person on the maximum recommended dose reaches.^[2]

Tirzepatide's record is not identical. Rats showed the same pattern of C-cell adenomas across doses. A six-month study in rasH2 transgenic mice found tirzepatide was not tumorigenic.^[5] That is one species where the effect did not appear, which is worth knowing and is not the same as an all-clear.

There is also a mechanistic answer to why this might be a rodent problem specifically. A 2010 study localized the GLP-1 receptor to rodent C-cells and showed agonists triggered calcitonin release and C-cell growth in rats and, less so, in mice. In humans and cynomolgus monkeys, GLP-1 receptor expression on thyroid C-cells was low, and the agonists did not produce the same signaling or calcitonin response in primates. Twenty months of liraglutide in monkeys at more than sixty times human exposure produced no C-cell hyperplasia.^[7] The paper's own conclusion did not stop at reassurance, and neither should we: it said the long-term consequences of sustained GLP-1 receptor activation in the human thyroid remain unknown and merit further investigation.

Worth naming plainly, because it bears on how much weight the paper carries: that study was authored substantially by employees of the company that makes liraglutide and semaglutide. It is still the clearest published account of the species difference. Both things are true at once.

What the human studies found, and why they disagree

Three large observational studies have looked for thyroid cancer in people taking these drugs. They point in different directions, and we are not going to pick one for you.

A French analysis using the national health insurance database matched 2,562 people who developed thyroid cancer against 45,184 controls. Among those who had used a GLP-1 receptor agonist for one to three years, the adjusted hazard ratio was 1.58 for thyroid cancer of any type and 1.78 for medullary thyroid cancer specifically, with a confidence interval that only just cleared 1.^[8] That is the study most often cited when someone tells you the warning has been borne out.

A Scandinavian cohort study covering Denmark, Norway and Sweden from 2007 to 2021 compared people starting a GLP-1 receptor agonist against people starting a DPP-4 inhibitor. Thyroid cancer occurred in 76 of 145,410 GLP-1 users and 184 of 291,667 comparators, rates of 1.33 and 1.46 per 10,000 person-years. The hazard ratio was 0.93, and the authors noted the upper edge of the interval was consistent with no more than a 31% relative increase. For medullary thyroid cancer alone the estimate was 1.19 on very few events.^[9] That is the study most often cited when someone tells you the warning is a rodent artifact.

A 2026 US analysis of the TriNetX database put the five-year question to a type 2 diabetes population and came back with no increase against any of three comparators, every point estimate landing at or below 1.^[10] Its own conclusion flagged the limitation that matters most here: these drugs are recent, so a five-year window cannot speak to what a decade of exposure does.

Why two honest studies land in different places

• What they compare against. The French study compared GLP-1 use to other second-line diabetes treatment; the Scandinavian one compared new starters against new starters of a specific rival drug class. Choosing a different comparator changes who ends up in the control group and what they were already at risk of.
• Detection. People who start a drug with a thyroid warning get their necks examined and imaged more often. Thyroid cancers are frequently found because someone looked, which can manufacture an association out of surveillance alone.
• Time. Every one of these studies covers a few years of exposure. A tumor process measured over a rat's lifetime is not something a median follow-up of three or four years can rule in or out.
• Rarity. Medullary thyroid carcinoma, the specific cancer the label names, is rare enough that even a study of hundreds of thousands of people is working from a handful of cases. Both the alarming and the reassuring estimates rest on small numbers.

Our reading is that the evidence is genuinely unsettled, and that any page telling you it is settled in either direction is telling you something the published record does not support. If you want one number to carry away, carry the absolute one: in the Scandinavian data, thyroid cancer turned up in roughly 1.3 people per 10,000 per year on a GLP-1, against 1.5 per 10,000 on the comparison drug.^[9]

The largest pooled answer, and what it still cannot settle

In 2026 the Annals of Internal Medicine pooled 48 randomized placebo-controlled trials covering 94,245 participants and graded the certainty of every result.^[11] For thyroid cancer it found an odds ratio of 1.37, with a confidence interval running from 0.82 to 2.31 — graded moderate certainty that these drugs probably have little or no effect. Stated as absolute risk, which is the form worth carrying around, that interval spans one case fewer to nine more per 10,000 people treated.

The same analysis found no signal for pancreatic, breast or kidney cancer at moderate certainty, and its results held when restricted to trials at low risk of bias and to semaglutide or tirzepatide specifically.^[11]

Its authors put the limitation in their own conclusion, and it is the same one that governs everything above: the pooled trials were not designed to measure cancer, and their follow-up was short. Solid tumors generally take years to become detectable. What this rules out is a large, fast effect. A small or slow one would produce exactly these numbers.

The manufacturer’s own analysis, and its lower bound

In 2026 Novo Nordisk published a three-source assessment of the same question: 93 of its liraglutide and semaglutide trials totalling 101,732 participants and 206,950 patient-years, its own post-marketing safety database, and a US commercial claims database.^[12] Its conclusion is that the totality of the data does not suggest an association. The individual numbers are worth reading before accepting that summary.

The claims-database result points the other way — a hazard ratio of 0.87 against SGLT2 inhibitors, comfortably including no difference. Comparing against another active drug rather than against nothing is a deliberate and good choice, because it partly controls for the fact that people prescribed these medicines differ from people who are not. It does not make SGLT2 users an ideal comparison group.

Two things push the trial estimate in opposite directions and neither can be resolved here. Surveillance bias inflates it: people on these drugs see clinicians more often and get imaged more, and thyroid cancers are frequently found incidentally. Short follow-up deflates it: 206,950 patient-years is a large number and thyroid tumors are slow.

The part that is not unsettled

The contraindication is categorical, and all five labels word it the same way. If medullary thyroid carcinoma has occurred in you or in a blood relative, or you carry the inherited syndrome MEN 2, the labeled answer is that these products are not to be used.^[1][5] Not watched more closely. Not started lower. Ruled out.

The labels are equally direct about what does not help. Checking serum calcitonin as a matter of routine, or booking a thyroid ultrasound the same way, is written up as of uncertain value for catching this cancer early in someone taking these drugs. The reasoning given is that the blood test is not specific enough and that thyroid abnormalities are common in the background population anyway, so the practice can push patients into procedures they never needed.^[2] Asking for a screening test as a way to feel safer is therefore not the reassurance it sounds like, and that verdict is the label's, not ours.

What the labels do ask for is that you know the symptoms they list: a mass in the neck, difficulty swallowing, shortness of breath, hoarseness that does not go away.^[1] Those go to a clinician, promptly, and they are not a reason to make any decision about your medication on your own. A separate question, and one that gets confused with this one constantly, is autoimmune thyroid disease — Hashimoto’s and drug-induced thyroiditis, which are immune conditions rather than cancers. The cohort evidence there, and the reason to be skeptical of it, is set out in GLP-1s and autoimmune thyroiditis.

If what you are taking is compounded

Everything above is drawn from FDA-approved labels. A compounded semaglutide or tirzepatide preparation has no approved label of its own, because the specific preparation was never reviewed. The molecule is the same one the rodent studies were run on, so the underlying question does not change, but three practical things do.

1. There is no document that has to tell you any of this. An approved product arrives with a boxed warning and a Medication Guide. What a compounded vial arrives with depends on the pharmacy and the telehealth service in front of it.
2. The contraindication screening depends on who asked. A family history of medullary thyroid carcinoma only excludes you if someone asks about it. If your intake was a web form, look back at whether it did.
3. Your adverse event may never reach FDA. A state-licensed 503A pharmacy is not required to report adverse events to the agency, which is one reason the compounded safety record is thinner than the approved one. We cover that gap in compounded tirzepatide side effects and in what is actually in a compounded vial.

If you are trying to work out who is dispensing what, our pharmacy register records the licensed entity behind each seller we track, and our compounded semaglutide board and live price tracker carry a verified date beside every figure. The related labeled risk that gets confused with this one is the eye question, which we handle separately in GLP-1 eye risks.