GLP-1 drugs and thyroid cancer: what the evidence shows

The short answer

The warning is real and the evidence behind it is thin. It came from tumours in rats, in a receptor that rodents have plenty of and primates barely express. One large French study found a raised risk. Two larger and better designed studies since then found none.

If you have picked up a box of Ozempic, Wegovy, Saxenda or Victoza in the United States, it carried a boxed warning about thyroid tumours. A boxed warning is the strongest warning the US Food and Drug Administration puts on a label. These drugs are GLP-1 receptor agonists, which means they are copies of a natural gut hormone (GLP-1, or glucagon-like peptide 1) built to last much longer in the body and to switch on the same receptor the hormone does.

The warning did not come from anything that happened to a person. It came from rats and mice, which grew tumours in a particular thyroid cell type when given these drugs for most of their lives [4]. The whole question, ever since, has been whether that finding means anything for humans, and the honest summary is that the evidence has moved steadily toward no, without quite arriving.

Three large studies in people frame the argument. A French national database study published in 2023 reported an increased risk of thyroid cancer [1]. A Scandinavian study across three countries published in 2024 found no increase [2]. A multi-database study published in 2025 also found no increase [3]. The two that found nothing were larger, used a study design built to remove a specific kind of bias, and the second of them was explicitly designed to test the first result. That is the shape of a safety signal fading rather than firming up.

None of that makes the warning pointless. It does something narrower and more useful than "this drug causes cancer", and the last part of this article explains what.

Where the warning came from: rats, not people

Rodents given GLP-1 drugs long term grew thyroid C-cell tumours. The 2010 paper that established this also found the mechanism: rodent C-cells carry the GLP-1 receptor, and primate C-cells barely do. Twenty months of liraglutide in monkeys produced no C-cell hyperplasia at all.

Your thyroid contains two useful cell types. The famous ones make thyroid hormone. A much rarer population, called C-cells or parafollicular cells, makes a different hormone called calcitonin, which is involved in handling calcium. When C-cells multiply abnormally you get C-cell hyperplasia (too many of them, but not cancer), and when they turn malignant you get medullary thyroid carcinoma, a rare form of thyroid cancer that behaves quite differently from the common ones.

In 2010 a team reported the experiment that produced the warning, and it is worth reading past the headline. The GLP-1 receptor was localised to rodent C-cells. GLP-1 receptor agonists stimulated calcitonin release, turned up the calcitonin gene, and produced C-cell hyperplasia in rats and, to a lesser extent, in mice [4]. So far this is the sentence every article repeats.

The same paper then reports the part that usually gets dropped. Humans and cynomolgus monkeys had low GLP-1 receptor expression in thyroid C-cells, and in primates the drugs did not activate the signalling machinery or produce any calcitonin release at all. And 20 months of liraglutide at more than 60 times human exposure levels did not lead to C-cell hyperplasia in monkeys [4]. The species difference is not a caveat bolted on afterwards. It is the paper's actual finding.

A 2013 review of the rodent work in toxicologic pathology put it more bluntly still. Rat C-cells are more sensitive than mouse C-cells, the effects have not been documented in nonhuman primates or humans, and GLP-1 receptors are either not present or occur in low numbers on the C-cells of primates and humans. Its conclusion is that this represents an on-target species-specific effect that may not have relevance to humans [5]. That same review notes a fact worth knowing for its own sake: exenatide, the first GLP-1 agonist approved, is a synthetic version of exendin-4, a compound derived from the Gila monster [5].

The human thyroid is not entirely off the hook

A 2012 study looked for GLP-1 receptors in human thyroid tissue and found them. They appeared in medullary carcinoma and C-cell hyperplasia consistently, in 18 percent of papillary thyroid cancers, and in C-cells in five of 15 normal thyroids. Low expression is not no expression.

Presenting the species argument as settled would be doing the same selective quoting in the opposite direction, so here is the study that complicates it. In 2012 a group examined human thyroid tissue directly, staining samples of medullary thyroid carcinoma (12 samples), C-cell hyperplasia (9), papillary thyroid carcinoma (17) and normal thyroid (15) for calcitonin and for the GLP-1 receptor [6].

They found the receptor. Staining for calcitonin and for the GLP-1 receptor appeared together consistently in both medullary carcinoma and C-cell hyperplasia. The receptor also showed up in 18 percent of papillary thyroid carcinomas, which is the common kind, three cases out of 17 [6]. And in normal thyroid tissue it was found in five of the 15 cases examined, in roughly 35 percent of the C-cells assessed [6].

The authors' own conclusion is a model of not overclaiming: the consequence of long-term pharmacologically increased GLP-1 signalling on these receptor-expressing cells in the human thyroid remains unknown, and appropriately powered prospective studies to exclude an increase in medullary or papillary thyroid carcinoma are warranted [6]. That is the honest state of the mechanism. Rodents are clearly a bad model for humans here, and humans are not a clean zero either. Which is exactly why the argument had to be settled with population data instead.

One study found a risk. Two larger ones did not

A French study of 2,562 thyroid cancer cases reported a 58 percent higher risk after one to three years of use. A Scandinavian cohort of more than 145,000 users found a hazard ratio of 0.93, and a multi-database study across millions of patients found nothing on any comparison.

The 2023 French study used the national health insurance database to compare people with type 2 diabetes on second-line diabetes drugs between 2006 and 2018. It identified 2,562 thyroid cancer cases and matched them against 45,184 control subjects. Use of a GLP-1 agonist for one to three years was associated with an increased risk of all thyroid cancer, at a hazard ratio of 1.58 (with a range of statistical uncertainty running from 1.27 to 1.95), and of medullary thyroid cancer at a hazard ratio of 1.78 (1.04 to 3.05) [1]. A hazard ratio of 1.58 means a 58 percent higher rate in the exposed group; the range in brackets is the confidence interval, the band of values the data cannot rule out.

The 2024 Scandinavian study set out to test that. It followed people starting a GLP-1 agonist in Denmark, Norway and Sweden from 2007 to 2021 and compared them against people starting a different diabetes drug class. Thyroid cancer occurred in 76 of 145,410 GLP-1 users and in 184 of 291,667 comparison patients, giving rates of 1.33 and 1.46 cases per 10,000 person-years. The hazard ratio was 0.93 (0.66 to 1.31), and for medullary thyroid cancer specifically it was 1.19 (0.37 to 3.86) [2].

That result is worth stating precisely, because "found no risk" is vaguer than what the study can support. The authors put it as a ceiling: given the top of the confidence interval, the findings are incompatible with an increased relative risk of more than 31 percent, which in absolute terms is no more than 0.36 extra cases per 10,000 person-years against a background rate of 1.46 per 10,000 in the comparison group [2]. A ceiling is more useful than a reassurance, because you can check your own tolerance against it.

The 2025 study went wider again, pooling administrative claims and health record databases and comparing new GLP-1 users against new users of three other drug classes. Between roughly 316,000 and 460,000 patients starting a GLP-1 agonist entered the analyses, against comparison groups running into the millions, and there was no statistically significant increase in thyroid tumours on any comparison or any analytical approach [3]. That paper also describes a meta-analysis of 45 clinical trials covering 52,600 patients on GLP-1 agonists which likewise found no significant increase, while noting its own limits of sample size and imprecision [3].

Why a risk that shows up immediately is suspicious

Cancer takes years to develop, so a real drug-caused cancer risk should grow with duration of use. In the French data the risk appeared early and stayed roughly flat. The Scandinavian authors read that missing dose-response as a sign of detection bias rather than of cause.

This is the part the consumer coverage leaves out, and it is the most useful thing on this page. When two good studies disagree, the interesting question is not which one to believe but what would produce that exact pattern of disagreement.

Start with what a genuine cancer risk looks like in data. Tumours take years to form and more years to be found, so exposure that actually causes cancer usually shows a dose-response relationship: longer use, higher risk. The French study did not show that. Its point estimates were of similar magnitude for use under one year, one to three years, and over three years. The Scandinavian authors flag this directly: since an effect of these drugs on thyroid cancer is unlikely to emerge after short-term use, a risk that appears immediately and then stops growing might indicate confounding, and an alternative explanation is detection bias [2].

Detection bias is simple and easy to miss. Someone who starts a new injectable medicine sees their doctor more often, gets examined more often, and gets scanned more often. Thyroid cancer is unusually vulnerable to this, and the scale is not small. A study covering 63 countries estimated that 1,736,133 of the 2,297,057 thyroid cancer cases diagnosed between 2013 and 2017, or 75.6 percent, were attributable to overdiagnosis, the term for finding a tumour that was never going to cause harm [9]. In four countries the figure among women was above 85 percent [9]. Look harder at a group of people and you will find more of these, without a single extra tumour having formed. The Scandinavian study's own supporting analysis found a nominally raised risk confined to the first year after starting treatment, which is the fingerprint of looking rather than of causing [2].

There is a second, more technical difference. The French study compared GLP-1 users against non-users. The Scandinavian and 2025 studies compared them against people starting a different drug for the same condition at the same stage, a design called an active-comparator new-user study. The reason that matters is that people who are prescribed a newer, more expensive drug differ from people who are not, in ways no database fully records. Comparing them to non-users measures those differences along with the drug. The BMJ authors note that in the French study the exposure groups might have been misaligned on important baseline characteristics [2], and the 2025 authors make the same point and add that their own design was chosen specifically to address it [3].

None of this proves the French study wrong. It explains why two competent teams reading the same drug class got different answers, and why the later ones carry more weight.

The warning is American, not universal

The United States contraindicates GLP-1 agonists for people with a personal or family history of medullary thyroid carcinoma or MEN2, and boxes the warning. European product information carries no equivalent contraindication or warning, and monitors the question through routine pharmacovigilance instead.

Here is a fact that reframes the whole question and appears almost nowhere in consumer coverage. The thyroid warning is not a global scientific consensus. It is a regulatory decision made in one country.

In the United States, the product labels of GLP-1 receptor agonists carry boxed warnings about thyroid cancer, and the drugs are contraindicated in patients with a personal or family history of medullary thyroid cancer or multiple endocrine neoplasia type 2 [2]. Multiple endocrine neoplasia type 2, usually shortened to MEN2, is an inherited condition that makes medullary thyroid carcinoma very likely, so this is a rule aimed at people whose C-cells are already predisposed to go wrong. The 2025 paper lists the drugs it applies to by name: liraglutide, dulaglutide, exenatide and semaglutide [3].

The same paper then states the contrast plainly: no similar contraindications or warnings exist in product information in Europe, where the potential risk is monitored as part of routine pharmacovigilance instead [3]. The Scandinavian authors add that their findings support the conclusion of a European Medicines Agency investigation that the available evidence does not support a causal association between GLP-1 receptor agonist use and thyroid cancer [2].

Two regulators looked at the same literature and landed differently. That is not a scandal; it is what regulators do when a signal comes from an animal model and the human data are reassuring but not infinite. The practical reading is that the US label is precautionary, aimed hardest at a small, identifiable group, and not a statement that the drug has been shown to cause cancer in people. If you want the wider regulatory picture for peptide drugs, our guide to FDA-approved peptides covers how these decisions get made.

What calcitonin testing can and cannot tell you

Calcitonin is the blood marker for C-cell trouble, so it was the obvious thing to measure. Across more than 5,000 people on liraglutide and 9,340 in a five-year cardiovascular trial, levels stayed at the low end of normal and no medullary thyroid cancers occurred in the treated group.

Because calcitonin is what C-cells make, tracking it in the blood is the natural way to look for the rodent effect in humans. Two large efforts did exactly that, and their results are the quietest but perhaps strongest evidence in the file.

The first measured unstimulated serum calcitonin every three months, for up to two years, in a series of trials covering more than 5,000 people on liraglutide or a control therapy. Mean levels started at the low end of the normal range and stayed there. At two years the estimated geometric mean values were no greater than 1.0 ng per litre, well below the upper limits of normal, and the proportion of people crossing the clinically relevant cutoff of 20 ng per litre was very low in every group, with no consistent relationship to dose or time and no consistent difference between treatments [7].

The second followed the 9,340 participants of the LEADER cardiovascular outcomes trial, in which people with type 2 diabetes at high cardiovascular risk were randomly assigned to liraglutide or placebo, over a period of three and a half to five years. At 36 months there was no evidence of any increase in calcitonin in either the male or the female subgroup, and there were no episodes of C-cell hyperplasia or medullary thyroid carcinoma in the liraglutide-treated patients [8]. The 2010 rodent paper had already reported the same direction of travel in humans, with mean calcitonin in patients exposed for two years remaining at the lower end of the normal range and no difference in the proportion crossing a 20 pg/ml cutoff [4].

What this does not mean is that you should go and get a calcitonin test. The reason the trials had to enrol thousands of people is that the marker is noisy and the disease is rare, and a single measurement in one person carries far less information than these numbers suggest. This is a question for the doctor who prescribed the drug, not for a lab panel bought online.

What this means if you are taking one

For most people the honest answer is that the ceiling on any excess risk is now very low and the warning is aimed at a specific inherited group. If medullary thyroid cancer or MEN2 runs in your family, that is the conversation to have before starting rather than after.

The single most actionable thing on this page is the family history question. The contraindication exists for people with a personal or family history of medullary thyroid carcinoma or MEN2 [2], and that is knowledge most people either have or can get by asking a relative. If it applies to you, say so before a prescription is written rather than looking it up afterwards.

For everyone else the numbers are the reassurance, and they are worth carrying as numbers rather than as a vibe. The largest well-designed study puts the ceiling at no more than 0.36 extra thyroid cancers per 10,000 person-years [2], which is the kind of quantity that has to be weighed against what the drug is being taken for. It also has real limits: the mean follow-up was 3.9 years [2], so nobody can yet speak to twenty years of use, and the 2025 study could not include tirzepatide at all because it was too new for a retrospective cohort [3]. Newer agents in this family inherit the class warning without yet having the class evidence.

Symptoms are worth knowing simply because they are cheap to know: a new lump in the neck, a hoarse voice that does not resolve, or trouble swallowing are reasons to get looked at, whatever you are taking. That is ordinary advice, not GLP-1 advice.

If you want to understand these drugs properly rather than one risk at a time, our semaglutide mastery course and tirzepatide mastery course cover how each molecule works, what its trials measured and where the evidence stops, and the free foundations course starts with what a peptide is at all.