GLP-1s and thyroid cancer: from rat C-cells to Nordic registries

Where the warning comes from

In rodents, GLP-1 agonists activate thyroid C-cells, raise calcitonin and drive proliferation. But human and monkey C-cells express little GLP-1 receptor, and 20 months of liraglutide at more than 60 times human exposure was not associated with hyperplasia in monkeys⁶. In humans, serial calcitonin measurements in more than 5,000 people on liraglutide or control for up to 2 years showed no drug-attributable rise¹¹. In mice, the effect does not appear without the GLP-1 receptor and was not associated with RET activation⁷. Even so, the authors cautioned that long-term consequences in the human thyroid "remain unknown"⁶.

That is why the tirzepatide label (revised August 2026) keeps the boxed warning: "In rats, tirzepatide causes dose-dependent and treatment-duration-dependent thyroid C-cell tumors at clinically relevant exposures", and the contraindication in a history of medullary thyroid carcinoma (MTC) or MEN2⁸. The semaglutide obesity label (revised June 2026) carries the same boxed warning on C-cell tumors¹⁰.

The French signal

In 2023 a case-control study in the French national database (2,562 cases, 45,184 controls) associated 1–3 years of GLP-1 use with all thyroid cancer (aHR 1.58; 95% CI 1.27–1.95) and medullary carcinoma (aHR 1.78; 1.04–3.05)¹. The main criticism was detection bias: people prescribed a drug with a thyroid warning tend to get more ultrasounds and calcitonin tests. After that study, the European Medicines Agency's pharmacovigilance committee (PRAC) concluded in October 2023 that the available evidence "does not support a causal association" and did not change the product information⁹. In the US, a target trial emulation with 351,913 patients found higher risk only in the first year (HR 1.85; 1.11–3.08), a pattern the authors say may reflect enhanced early detection¹⁴.

What the large studies found

• Scandinavia (BMJ 2024): 145,410 GLP-1 users vs 291,667 DPP-4 inhibitor users. HR 0.93 (0.66–1.31); incidence 1.33 vs 1.46 per 10,000 person-years. For MTC, HR 1.19 with a very wide interval (0.37–3.86)².
• Six countries (Thyroid 2025): 98,147 users, pooled HR 0.81 (0.59–1.12), with a median follow-up of only 1.8–3.0 years, "insufficient to rule out excess risk with long-term use"³.
• Cohort of 460,032 users with type 2 diabetes (Diabetes Care 2025), outcome thyroid tumors: HRs of 0.78 to 1.03 vs SGLT2 inhibitors, DPP-4 inhibitors and sulfonylureas; none elevated⁴.
• Meta-analysis of 48 trials and 94,245 participants (Ann Intern Med 2026): OR 1.37 (0.82–2.31), i.e. from 1 fewer to 9 more cases per 10,000 treated, moderate certainty, consistent for semaglutide and tirzepatide⁵.

What cannot be ruled out

The 2024–2026 data are reassuring for thyroid cancer overall: the absolute risk, if any, is a few cases per 10,000²⁵; in 89,646 long-term users (median follow-up 4.5 years) there was no excess either¹⁵. Not every meta-analysis agrees: one of 64 trials found an OR of 1.52 (1.01–2.29), with a fragility index of 1 and a 5-year number needed to harm of 1,349¹², and its update with 50 trials an OR of 1.55 (1.05–2.27)¹³. But medullary carcinoma is rare, its intervals are wide, and follow-up remains short for a slow-growing tumor²³. The French signal is plausibly surveillance bias, but it has not been disproven¹.