The same class of diabetes drugs pointed in opposite directions on the thyroid depending on which kind of diabetes the patient had. In type 1, they tracked with less thyroid disease. In type 2, with slightly more.
That split is the finding of a large real-world study published June 30 in Therapeutic Advances in Endocrinology and Metabolism ↗. Researchers led by Hao-I Hsieh, drawing on the TriNetX US records network, matched adults starting GLP-1 receptor agonists against similar patients on usual care and followed both groups for up to five years. GLP-1 drugs are the semaglutide ↗ family, sold as Ozempic and Wegovy and now taken by millions, and the thyroid has been a nagging question since rodents on the drugs grew thyroid C-cell tumors.
Split the patients by diabetes type and the picture stops being one thing. Among 7,428 matched pairs with type 1 diabetes, the GLP-1 group had less autoimmune thyroiditis (hazard ratio 0.41, roughly 60 percent fewer cases), less underactive thyroid (0.77), and less overactive thyroid (0.61). Among 43,167 pairs with prediabetes the pattern held, milder: fewer cases of an underactive thyroid (0.82) and of thyroid disorders generally (0.77).
Then type 2 diabetes, the 137,952-pair group, went the other way. GLP-1 users had more autoimmune thyroiditis (1.23, about a quarter more) and more nontoxic goiter, a benign enlargement of the gland (1.14). The one condition measured in both diabetes types, autoimmune thyroiditis, landed on opposite sides of the line: protective in type 1, elevated in type 2.
The number that should not be read straight is the mortality one. Every group on GLP-1 drugs died less often, by a lot (hazard ratios 0.47 to 0.68). A drug does not cut death in half. That gap is the healthy-adherer effect, the well-documented tendency of people who fill and keep taking prescriptions to be healthier in ways the records never capture. It is the standard reason to distrust real-world drug-benefit numbers, and it applies here in full.
Which is what makes the thyroid split interesting rather than dismissible. Healthy-adherer bias pushes every outcome the same way, toward looking good on the drug. It does not obviously explain why one thyroid condition would flip sign between two patient groups. And the type 2 harm signal held against an active comparator: versus patients started on SGLT2 inhibitors, another modern diabetes drug, GLP-1 users still had more autoimmune thyroiditis (1.16) and goiter (1.06). Comparing drug to drug rather than drug to usual care strips out much of the healthy-adherer problem, and the signal survived.
On the fear that started all this, cancer, the study is reassuring and consistent with what we found before. Thyroid cancer showed no significant change in any group. That matches the larger TriNetX analysis we covered in May ↗, where an early 41 percent bump in thyroid cancer melted to nothing by the third year, the signature of surveillance bias rather than a real tumor risk. Two independent looks at the same records network now put the cancer worry down and move the question to the smaller, non-cancer thyroid effects, which is where the type-specific split lives.
What a patient does with this is modest. The absolute risks are small, the study is records-based and cannot prove cause, and nobody should stop a working diabetes drug over a goiter hazard ratio of 1.14. But if the direction really does depend on diabetes type, thyroid monitoring on these drugs may not be one setting for everyone, which is the authors' own careful conclusion.