A pooled analysis of 78 studies found that people taking GLP-1 drugs developed fewer of ten obesity-linked cancers. Not one of the 78 studies was a randomized trial.
The review, published September 1 in Cancer Control ↗, started from 1,644 citations, narrowed to 139 full-text reviews, and pooled 78 observational studies of adults taking GLP-1 receptor agonists for any reason. These are the drugs behind Ozempic and Wegovy (semaglutide ↗) and Mounjaro and Zepbound (tirzepatide ↗), which act on the GLP-1 receptor ↗ to blunt appetite and drive large weight loss. Obesity is an established cause of more than a dozen cancers, so a drug that strips off weight is a plausible cancer-risk lever, and this is the largest attempt yet to add up whether the epidemiology bears that out.
It leans one way. GLP-1 use was tied to lower risk for ten of thirteen obesity-associated cancers: colorectal, endometrial, esophageal, gallbladder, liver, ovarian, pancreatic, stomach, plus meningioma and multiple myeloma. The sharpest numbers come from comparisons against insulin. Against insulin specifically, GLP-1 users had roughly half the colorectal cancer risk (relative risk 0.54, meaning about a 46 percent lower rate), about a third the liver cancer risk (0.35), and under half the pancreatic cancer risk (0.41). One cancer went the other way: thyroid risk was slightly higher among GLP-1 users, though the increase was not statistically significant and could be noise (relative risk 1.09, confidence interval 0.98 to 1.21).
That thyroid line is the one worth watching, because it is the signal regulators have flagged since the drugs launched, and it is the only direction here that would count as harm rather than benefit.
The problem is the comparator, and the authors say so. When the eye-catching numbers come from stacking GLP-1 against insulin, the contrast is partly a story about who gets insulin. People put on insulin tend to have longer, harder-to-control diabetes and more of the metabolic damage that itself raises cancer risk. Line a healthier drug up against a sicker-patient drug and the healthier group will look protected even if the pill did nothing. This is confounding by indication, and matched observational data cannot fully scrub it out.
The second catch is that weight loss and the drug are tangled together. If GLP-1 users get fewer obesity cancers because they are no longer obese, the causal actor is the weight, not a direct anticancer property of the peptide. That distinction matters for anyone trying to reposition these drugs as cancer prevention ↗ rather than metabolic treatment. The pooled data cannot separate the two.
None of this makes the pattern fake. Ten cancers moving the same direction across 78 studies is a real and consistent signal. It is just a signal that observational data is structurally unable to convert into cause. The authors close by calling for large randomized trials and balanced risk-benefit conversations, which is the honest reading: this is a strong reason to run the experiment, not evidence that the experiment already passed.
We have written the single-cohort version of this before. In April, a study of one large health system found GLP-1 users had 41 percent fewer obesity cancers, with the benefit missing in Black patients ↗, a reminder that a pooled average can hide who the effect does and does not reach. And the safety side has landed on reassurance from bias-resistant designs: earlier this year, GLP-1 drugs showed no melanoma signal across 390,000 matched pairs ↗. Fewer cancers on one side, no new cancers on the other, and on both sides the same limit: a database can tell you what happened to the people who took the drug, not what the drug did.