Trial evidence

Does tirzepatide cause cancer?

By David Chen·Editor·Reviewed September 2026

What the warning actually says¹

The first bullet of Zepbound’s boxed warning, verbatim:

In rats, tirzepatide causes dose-dependent and treatment-duration-dependent thyroid C-cell tumors at clinically relevant exposures. It is unknown whether ZEPBOUND causes thyroid C-cell tumors, including medullary thyroid carcinoma (MTC), in humans as human relevance of tirzepatide-induced rodent thyroid C-cell tumors has not been determined.

Two sentences, two different kinds of claim. The first is a finding in rats, stated flatly. The second is an admission of ignorance about people. A boxed warning built on an unknown is unusual, and it is why the contraindication is absolute rather than dose-dependent: tirzepatide is not for anyone with a personal or family history of medullary thyroid carcinoma, or with Multiple Endocrine Neoplasia syndrome type 2. Mounjaro, the same molecule for type 2 diabetes, carries the same warning word for word.³ The full warning and the contraindications.

The rat study, decoded²

Section 13.1 of the label is where the animal data lives, and it is more specific than the warning it supports. A two-year carcinogenicity study gave rats tirzepatide twice weekly at three doses, which the label puts at 0.1-, 0.4- and 1-fold the exposure of the maximum human dose of 15 mg once weekly, measured as AUC. What it found:

• a statistically significant increase in thyroid C-cell adenomas — benign tumors — in male and female rats, from the middle dose in males and the lowest dose in females;
• a statistically significant increase in C-cell adenomas and carcinomas combined in male and female rats at all doses examined, including the dose that gave the rats roughly a tenth of a human’s exposure.

Two other results in the same section point the other way, and they are part of the picture. In a six-month study in a strain of mouse bred for cancer testing, tirzepatide was not tumorigenic at any of three doses. And tirzepatide was not genotoxic in the bone-marrow test the label reports — it does not appear to damage DNA, which is one of the ways a drug can cause cancer and evidently not this one.

Why C-cells, and why this family of drugs? Thyroid C-cells carry receptors for both of the hormones tirzepatide imitates, and animal studies of the older single-receptor drugs pointed at medullary thyroid carcinoma before tirzepatide existed — that is the account given in the review of the tirzepatide trials below.⁴ Whether human C-cells respond the way a rat’s do is the open question the label refuses to close.

What the label does not report³

Both current tirzepatide labels were read in full for cancer, carcinoma, malignant, neoplasm, tumor, C-cell and calcitonin. Every appearance is in the boxed warning, the contraindications, the warnings section, the animal section or the patient-counseling section — the rodent finding and what follows from it.

Neither label reports a count of cancers from its human trials. The adverse-reaction table lists reactions that occurred in at least 2% of participants and more often than on placebo, and no tumor of any kind reaches it; the postmarketing section lists gastrointestinal, hypersensitivity, lung and kidney events, and no cancer. What the table does report. So the human numbers on this page do not come from the label. They come from the trials, pooled by people who went looking for them.

What the trials recorded⁴

One analysis has gathered the cancer counts from the tirzepatide trials specifically: 13 randomized trials, 13,761 participants, treatment lasting 26 to 72 weeks, with control arms that were placebo, insulin or a GLP-1 drug depending on the trial.

Read the counts before the ratios. A confidence interval running from 0.22 to 5.12 is what six events look like: it cannot rule out a fivefold increase and it cannot rule out a large decrease. The denominators differ by row because only the trials that reported a given cancer contribute to that row, and the tirzepatide arms were larger than the control arms in several trials — in SURMOUNT-1, three of the four arms received tirzepatide — so the raw counts are not rates. The ratios account for that; the counts tell you how little is behind them.

The same analysis found comparable risks in participants with and without diabetes, against each type of control separately, and at each dose. Its conclusion, as written: tirzepatide use in these trials “did not increase overall or specific cancer risk.” An earlier and smaller analysis of nine tirzepatide trials in type 2 diabetes reached the same place and called its own result “hypothesis-generating”, naming the small number of events and the short follow-up as the reasons.⁵

One bookkeeping note, because a reader who finds the abstract elsewhere will hit it. The abstract of that analysis says the included trials reported “no cases of papillary thyroid carcinoma”; its own results table prints 5 papillary cancers on tirzepatide against 1 in the controls, and its results text says the trials reported no cases of medullary thyroid carcinoma. The full text is what is used above.

Thyroid cancer in the trials⁶

A second analysis asked only about thyroid cancer, across 15 randomized trials of this whole family of drugs and 84,237 participants: 28 thyroid cancers in the treated groups against 15 in the controls, an odds ratio of 1.52 (0.86–2.68), which its authors rate as very low certainty evidence because the events are so rare. Four of those 15 trials were tirzepatide trials — SURPASS-2, SURPASS-3, SURPASS-4 and SURMOUNT-1 — and between them they recorded 7 thyroid cancers: 5 among participants on tirzepatide and 2 among comparators, giving an odds ratio of 1.22 with an interval from 0.22 to 6.80.

The detail that matters most for the boxed warning is what those cancers were. Most thyroid cancers reported in these trials were papillary or otherwise differentiated — the common kind, which has nothing to do with C-cells. Medullary events were too rare to analyze separately, which the authors name as a limit on how much any of this says about the animal finding. The rodent signal is about medullary cancer; the human data are mostly about a different disease of the same gland.

A third analysis, of 48 placebo-controlled trials and 94,245 participants in this drug class, put the thyroid-cancer odds ratio at 1.37 (0.82–2.31) — which it translates into somewhere between 1 fewer and 9 more cases per 10,000 people treated — and pancreatic cancer at 0.84 (0.53–1.35), both rated moderate certainty for “little or no effect.” It reports that the results held in a sensitivity analysis restricted to the semaglutide and tirzepatide trials, and that the trials were not designed to measure cancer and did not follow anyone for long.⁷

Calcitonin, and what it is for¹

Calcitonin is the hormone thyroid C-cells make, so it is the blood test that tracks the cell type the rat study implicated. The label addresses it directly, and not in the way a worried reader might expect:

Routine monitoring of serum calcitonin or using thyroid ultrasound is of uncertain value for early detection of MTC in patients treated with ZEPBOUND. 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 disease.

The same section gives the scale that makes sense of a result: patients with medullary thyroid carcinoma “usually have calcitonin values >50 ng/L.” Against that yardstick, the trials reported greater percentage rises in calcitonin on the 10 mg and 15 mg doses than on placebo ⁴ — and a small prospective study measured what those rises look like in absolute terms: in 58 adults with obesity and no known thyroid disease, calcitonin went from a mean of 2.7 to 3.2 pg/mL over 12 weeks, a statistically significant change in which no participant crossed 20 pg/mL and no thyroid nodules were seen on ultrasound.⁸

A picogram per milliliter is the same quantity as a nanogram per liter, so those values sit far below the level the label associates with medullary cancer. Fifty-eight people for twelve weeks is also not a cancer study, and the authors describe their finding as supporting short-term safety in people selected for having no thyroid risk factors. Whether anyone should have the test is a question for a prescriber, and the label’s answer above is the reason it is not an obvious yes.

The records studies⁹

Trials are short; medical records go on longer. The largest observational study with a tirzepatide cohort of its own compared 19,682 matched pairs of adults with type 2 diabetes and overweight or obesity — tirzepatide against a different class of diabetes drug — over a mean of 435 days, about fourteen months. It counted twelve obesity-associated cancers, thyroid among them.

The result: 206 cancers in the tirzepatide group against 243 in the comparison group, a hazard ratio of 0.84 (0.69–1.01) that does not reach statistical significance. One individual cancer differed significantly — ovarian cancer, lower on tirzepatide — and, in the authors’ words, tirzepatide “was not associated with significantly different rates of other individual” cancers, thyroid included. In the same study, semaglutide over a longer follow-up was associated with lower rates of several cancers.

Nothing in that is proof of protection. People prescribed one drug differ from people prescribed another in ways matching cannot remove, fourteen months is short, and a study of records can only see the cancers someone diagnosed and coded. What it does establish is the direction of the signal: where tirzepatide-specific human data exist, they do not show more cancer.

There is also a single case report — a 68-year-old woman taking tirzepatide for type 2 diabetes, diagnosed with a mixed medullary and papillary thyroid carcinoma after presenting with a neck mass. Its own authors note that human studies have not confirmed an association between these drugs and differentiated thyroid cancer, and that no consensus exists on screening for it. One case is a case: it cannot tell you whether the drug was involved, and it is here because leaving it out would be selective.¹⁰

The studies about other drugs¹¹

Search this question and you will meet two famous cohorts, usually presented as being about the drug you are taking. Neither one contains a single tirzepatide user.

• The French study that found a risk. In national insurance data, use of a GLP-1 receptor agonist for one to three years was associated with an increased rate of all thyroid cancer (hazard ratio 1.58, 1.27–1.95) and of medullary thyroid cancer specifically (1.78, 1.04–3.05), from 2,562 cancer cases matched to 45,184 controls. Its cohort ran from 2006 to 2018 and its cancers were identified between 2014 and 2018 — years before tirzepatide was approved in 2022.
• The Scandinavian study that did not. Across Denmark, Norway and Sweden, 76 thyroid cancers among 145,410 people starting a GLP-1 receptor agonist and 184 among 291,667 starting a different diabetes drug gave a hazard ratio of 0.93 (0.66–1.31), and 1.19 (0.37–3.86) for medullary cancer alone. Its data end in 2021 — again before tirzepatide existed as an approved medicine.¹²

This is not a technicality. Tirzepatide is not the same kind of drug as the ones those cohorts measured: its own label describes it as a GIP receptor and GLP-1 receptor agonist, acting on two receptors where the older drugs act on one.¹ A paper whose title says “GLP-1 receptor agonist” may or may not have enrolled anyone taking tirzepatide, and the only way to know is to read which drugs it names. Every study on this page has been read for exactly that, and the source list below says what each one enrolled.¹⁵

Reports are not rates¹³

A third kind of evidence circulates as though it were a risk estimate: counts of adverse-event reports sent to regulators. In the US database, a disproportionality analysis covering 2004 to early 2024 reported thyroid cancer more often than expected for four drugs in this family, tirzepatide among them — a reporting odds ratio of 2.09 (1.51–2.89) for tirzepatide, against 7.61 for semaglutide, 3.59 for dulaglutide and 15.59 for liraglutide. In the European database, across 34,956 reports, semaglutide was less likely than tirzepatide to carry a thyroid-cancer report (0.54, 0.37–0.81), while the analysis by indication found no significant signal at all.¹⁴

Notice that the two databases disagree about which drug looks worse, which is the clearest possible demonstration of what these numbers are. A reporting odds ratio compares how often an event appears in reports about one drug versus the rest of the database; it has no denominator of people treated, and a drug with a boxed warning about thyroid tumors attracts thyroid reports. The label states the principle for its own postmarketing section: because such reactions are reported voluntarily from a population of uncertain size, “it is not always possible to reliably estimate their frequency or to establish a causal relationship to drug exposure.”³ Both sets of authors say the same in their own words.

What a compounded vial changes¹

What none of this settles

Cancer is a slow question asked of fast evidence. The randomized trials ran 26 to 72 weeks; the records study followed people for about fourteen months; the two long cohorts that did run for years were measuring other drugs. Solid tumors can take longer than any of those windows to appear, so “no increase detected” is a statement about what has been looked at, not a clean bill of health — and the reverse is also true: the handful of cancers counted so far are far too few to establish a risk.

What is settled is narrower and more useful. The rat finding is real and it is in the label. The contraindications are absolute, and they are the part of this page a reader can act on with a prescriber: a personal or family history of medullary thyroid carcinoma, or MEN 2, rules the drug out. Everything else is a question that needs longer follow-up than anyone has yet published.

This page is not medical advice and recommends no test, no screening interval and no decision. It reports what the label says, what the trials counted and which molecules each study enrolled. Symptoms the label asks patients to report — a lump in the neck, trouble swallowing, shortness of breath or a hoarseness that does not go away — belong to a clinician the day they appear, not to a web page. See the medical disclaimer.