GLP-1 and Thyroid Cancer Risk: The Warning and the Human Data

Every semaglutide box carries a warning about thyroid tumors, and it is the single most-asked question about this drug class. It is also the question most likely to be answered badly, because the answer has two halves that get mixed together constantly: what happened in rodents, and what has been observed in people. Those are separate bodies of evidence. They do not say the same thing, they are not measuring the same thing, and neither one cancels the other.

This article keeps them apart on purpose. First, exactly what the boxed warning says, quoted off the current Wegovy and Ozempic labels. Then what the rodent studies behind it actually found. Then, in a section that borrows none of those numbers, what the human cohort studies have reported.

The boxed warning, quoted

Both current labels open with a boxed warning headed WARNING: RISK OF THYROID C-CELL TUMORS. The Wegovy version reads:

"In rodents, semaglutide causes thyroid C-cell tumors at clinically relevant exposures. It is unknown whether WEGOVY causes thyroid C-cell tumors, including medullary thyroid carcinoma (MTC), in humans as the human relevance of semaglutide-induced rodent thyroid C-cell tumors has not been determined. WEGOVY is contraindicated in patients with a personal or family history of MTC or in patients with Multiple Endocrine Neoplasia syndrome type 2 (MEN 2). Counsel patients regarding the potential risk of MTC and symptoms of thyroid tumors¹."

Ozempic's box is the same warning with one clause missing — it states "In rodents, semaglutide causes thyroid C-cell tumors," without Wegovy's "at clinically relevant exposures," which on the Ozempic label appears in §5.1 rather than in the box. Everything else, including the MTC and MEN 2 contraindication, is identical².

Read that carefully, because the warning is more careful than most summaries of it. It does not say semaglutide causes thyroid cancer in people. It says the rodent finding exists, that the human relevance of that finding has not been determined, and that the drug is contraindicated in two specific populations regardless. Those are three different statements and only the first is a finding.

Section 5.1 on both labels adds one more piece of human data, and it is deliberately non-committal: "Cases of MTC in patients treated with liraglutide, another GLP-1 receptor agonist, have been reported in the postmarketing period; the data in these reports are insufficient to establish or exclude a causal relationship between MTC and GLP-1 receptor agonist use in humans¹²." Insufficient to establish or exclude — the label is explicitly declining to answer in either direction.

What the rodent studies actually found

This is the evidence the box is built on, and it is worth being precise about because it is genuinely stronger than the reassuring version and genuinely narrower than the alarming one.

Section 13.1 of both labels describes two lifetime carcinogenicity studies. In a two-year study in CD-1 mice, subcutaneous semaglutide produced "a statistically significant increase in thyroid C-cell adenomas and a numerical increase in C-cell carcinomas" in males and females. In a two-year study in Sprague Dawley rats, there was "a statistically significant increase in thyroid C-cell adenomas... in males and females at all dose levels, and a statistically significant increase in thyroid C-cell carcinomas... in males at greater than or equal to 0.01 mg/kg/day, at clinically relevant exposures¹²."

Two details are usually lost, and both cut against easy reassurance.

The rat effect was not confined to enormous doses. Ozempic's §13.1 states the rat doses as "below quantification, 0.2-, 0.5-, and 3-fold the exposure at the MRHD" of 2 mg/week. The C-cell carcinoma increase in males appeared at 0.01 mg/kg/day and above — which is the 0.2-fold rung. That is below human exposure at that label's maximum recommended dose, not a multiple of it. This is what "at clinically relevant exposures" means, and it is why the phrase is in the warning².

The "how many times the human dose" figure depends on which label you read, and that is not a contradiction. Both labels describe the same mouse study at the same doses — 0.3, 1 and 3 mg/kg/day in males. Ozempic reports those as 2-, 11- and 30-fold the maximum recommended human dose, against an MRHD of 2 mg per week. Wegovy reports the same doses as "up to 7-fold the MRHD," against an MRHD of 7.2 mg per week¹². Same rodents, same doses, different denominators — the human reference dose moved when the higher tier was approved. Anyone quoting "30 times the human dose" from one label against "7 times" from the other has found a ratio, not a discrepancy.

And then the sentence both labels put immediately after all of it: "Human relevance of thyroid C-cell tumors in rats is unknown and could not be determined by clinical studies or nonclinical studies¹²." That is the sponsor's own required disclosure that the rodent data cannot be translated.

What the human epidemiology has found

Nothing in this section derives from the rodent studies. These are separate observational studies in people, and they disagree with each other in a way that is largely explained by how each was designed.

The study that found an increase. Bezin et al. performed a nested case-control analysis in the French national health insurance database, identifying 2,562 patients with thyroid cancer among adults with type 2 diabetes on second-line antidiabetes drugs, matched with 45,184 controls. Their reported result: "Use of GLP-1 RA for 1-3 years was associated with increased risk of all thyroid cancer (adjusted hazard ratio [HR] 1.58, 95% CI 1.27-1.95) and medullary thyroid cancer (adjusted HR 1.78, 95% CI 1.04-3.05)³." This is the finding that drove most of the coverage on this topic, and it should not be waved away — it is a large national dataset with a specific MTC result. It also drew four separately published Letters in Diabetes Care in a single 2023 issue, plus a response from the authors, disputing the design and the interpretation⁴. Those are correspondence, not studies, and this article labels them as such.

The largest active-comparator cohort found nothing. Pasternak et al. published a Scandinavian cohort study in the BMJ covering Denmark, Norway and Sweden from 2007 to 2021, comparing patients starting a GLP-1 receptor agonist against patients starting a DPP-4 inhibitor — a design chosen, in their words, "to minimise risks of confounding and time related biases." Their results: "76 of 145,410 patients (incidence rate 1.33 events per 10,000 person years) treated with GLP1 receptor agonists and 184 of 291,667 patients (incidence rate 1.46 events per 10,000 person years) treated with DPP4 inhibitors developed thyroid cancer. GLP1 receptor agonist use was not associated with increased risk of thyroid cancer (hazard ratio 0.93, 95% confidence interval 0.66 to 1.31)... The hazard ratio for medullary thyroid cancer was 1.19 (0.37 to 3.86)⁵." Their own summary of the precision achieved: "the upper limit of the confidence interval was consistent with no more than a 31% increase in relative risk⁵."

The US study found a signal, in the first year only. Brito et al. ran a target trial emulation across US commercial and Medicare claims covering 351,913 patients. Overall, "GLP-1RA initiation was not significantly associated with increased overall risk for thyroid cancer compared to the other 3 diabetes drugs (HR, 1.24; 95% CI, 0.88-1.76). However, the risk for thyroid cancer was significantly higher within the first year after GLP-1RA initiation (HR, 1.85; 95% CI, 1.11-3.08)⁶." The authors' own reading of that pattern is in their conclusion: the finding "may have been due to enhanced early detection; therefore, further research is necessary to understand the underlying causes of this association⁶." A risk that appears immediately on starting a drug and then disappears is the classic signature of surveillance — people who start a new medication see clinicians more often, and thyroid cancer is unusually easy to find when someone looks.

A 2026 replication across three glycemic states. A target trial emulation using the TriNetX US network compared GLP-1 receptor agonist initiators against usual care and against SGLT2 inhibitor initiators in prediabetes, type 1 diabetes and type 2 diabetes, followed for up to five years. It reported modest increases in some non-malignant thyroid outcomes in type 2 diabetes, and on cancer specifically: "no increased risk of thyroid cancer was observed in any glycemic state⁷."

And a broader cancer cohort, reported precisely. Dai et al. compared 43,317 GLP-1 receptor agonist users against 43,315 matched non-users with obesity across 14 cancer types, finding lower overall cancer incidence (HR 0.83, 95% CI 0.76-0.91). Thyroid cancer was one of the 14 types studied. The abstract names reduced risks for endometrial and ovarian cancer and meningioma, and a marginally non-significant increase for kidney cancer; it does not report an elevated thyroid result⁸. That is worth stating exactly as it is — an absence from a list of highlighted findings is weaker evidence than a reported null, and this article is not going to upgrade it into one.

Why these studies disagree, and what that leaves

The disagreement is not random. It sorts almost entirely by study design.

The study that found an increase is a nested case-control analysis. The studies that found none are active-comparator new-user cohorts, a design specifically built to handle the confounding that troubles this question — because the comparison group is people starting a different diabetes drug at the same point in their disease, rather than people not being treated. And the one cohort that did find a signal found it only in the first year, and its own authors attributed it to detection.

Detection bias is not a hypothetical here. Thyroid cancer has a very high reservoir of subclinical disease, and its measured incidence in a population moves sharply with how hard anyone is looking. That is precisely why both labels tell prescribers not to screen routinely: "Routine monitoring of serum calcitonin or using thyroid ultrasound is of uncertain value for early detection of MTC in patients treated with WEGOVY. Such monitoring may increase the risk of unnecessary procedures, due to the low-test specificity for serum calcitonin and a high background incidence of thyroid disease¹." The label is warning about the same mechanism that most plausibly produced the first-year signal in the US claims data.

What remains genuinely unresolved is medullary thyroid carcinoma specifically, and the reason is arithmetic rather than controversy. MTC is rare. The Scandinavian cohort's MTC hazard ratio was 1.19 with a confidence interval running from 0.37 to 3.86⁵ — an interval that comfortably contains both a large protective effect and a near-quadrupling. No human study to date has enough MTC events to narrow that. The boxed warning's contraindication in personal or family history of MTC and in MEN 2 is not waiting on that evidence, and should not be read as if it were: it is a precaution applied to two populations where the baseline risk is already high, taken because the rodent mechanism is real and the human relevance is unknown.

What this means in practice

The contraindication is a hard one. Both labels list a personal or family history of MTC, or MEN 2, under §4 Contraindications — not as a caution to weigh, but as a condition under which the drug is not to be used¹². Both labels also instruct counseling patients on the symptoms of thyroid tumors: "a mass in the neck, dysphagia, dyspnea, persistent hoarseness¹."

For anyone buying a compounded GLP-1 rather than an approved one, there is an additional wrinkle worth naming: a compounded product does not come with the boxed warning, the contraindication list, or the counseling instruction attached, because it does not carry an FDA-reviewed label at all. The pharmacologic question is the same; the disclosure is not. Our explainer on what compounded semaglutide actually is covers that gap, our regulatory timeline covers how the legal basis for selling it narrowed, and our label-sourced dosing reference covers what the approved products specify. If what brought you here was a marketing claim about low-dose GLP-1 use, our piece on semaglutide microdosing is the companion to this one. And if you are comparing sellers, our tirzepatide provider rankings and our semaglutide provider comparison read each company's own published terms, including whether the product dispensed is the approved one.

The bottom line

The boxed warning is real, it is required, and it says what it says: in rodents, semaglutide causes thyroid C-cell tumors, at exposures the label itself calls clinically relevant, and whether that translates to humans has not been determined. In humans, the largest active-comparator cohort study to date found no increased risk of thyroid cancer over a mean 3.9 years of follow-up, with an upper confidence bound consistent with no more than a 31% relative increase; a French nested case-control study did find an increase; and a US target trial emulation found one confined to the first year after starting treatment, which its own authors attribute to enhanced early detection. Medullary thyroid carcinoma specifically is too rare for any of these studies to have resolved. Both facts belong in the answer, and neither one replaces the other.