GLP-1s and Cancer: The Randomized Answer

We have already looked at what the observational cohorts say about cancer, and at how badly one of them behaved. Here is the randomized answer, from a systematic review in the Annals of Internal Medicine: GLP-1 drugs may have little or no effect on obesity-related cancers.^[1] The reason that is less reassuring than it sounds is in the review’s own limitations.

What the trials showed

For colorectal, esophageal, liver, gallbladder, ovarian and endometrial cancer, and for multiple myeloma and meningioma, the review found the drugs may have little or no effect — at low certainty. For gastric cancer, the effect is very uncertain.^[1]

Why 'no effect' is not the same as 'safe'

The review states its central limitation directly: the included trials were not designed to evaluate cancer outcomes and had short follow-up.^[1] That single sentence governs everything above.

Cancer takes years to appear. These trials ran for one or two. Absence of a signal in that window is largely a statement about the window.

Cancer outcomes in these studies were recorded as adverse events, not hunted for as endpoints, in trials built around weight or cardiovascular disease. The same structural weakness produced the inflated dementia signal we covered in the EVOKE article, only here it cuts toward a null rather than a false positive.

What is genuinely reassuring, and what is not

• Reassuring: results held consistent in sensitivity analyses restricted to low-risk-of-bias trials, and across semaglutide and tirzepatide specifically, and across subgroups by follow-up, population, dose and duration of action.^[1] Consistency is the property that makes a null believable.
• Reassuring: nothing points toward increased risk. The intervals sit around no effect rather than leaning toward harm.
• Not reassuring: the follow-up is far too short for the question. Longer-term studies are needed, as the review says.
• Not reassuring: the confidence intervals are wide enough to accommodate meaningful effects in either direction for several cancers.

How this sits with the cohort evidence

The cohorts suggested a small protective association — a hazard ratio of 0.87 for obesity-related cancers across 919,609 people. The randomized trials suggest little or no effect. Those are not in conflict.

The cohorts have the years the trials lack and the confounding the trials do not. The trials have the randomization the cohorts do not and none of the time. Each is strong exactly where the other is weak, which is why the honest answer combines them into: probably nothing dramatic in either direction, and nobody yet has the decade of data that would settle it.

The thyroid C-cell warning that appears on every label in this class is a separate question with its own evidence, covered in the thyroid cancer article.