GLP-1s and Cancer: What the Cohorts Actually Show

Two large cohort studies looked at GLP-1 drugs and cancer. One found a modest reduction in obesity-related cancers that is probably real. The other reported that breast cancer patients on these drugs had 91% lower mortality than those on insulin or metformin — and, in the same paper, no difference at all compared with a different diabetes drug. That contradiction is not a flaw in the study. It is the most instructive thing in it.

The believable finding

A nationwide cohort published in the Journal of the National Cancer Institute followed 919,609 overweight or obese adults with type 2 diabetes in the United States between 2016 and mid-2021. Across 2,086,526 person-years, 16,653 were newly diagnosed with one of thirteen obesity-related cancers.^[1]

GLP-1 users had an incidence of 7.5 per 1,000 person-years against 8.1 for users of other glucose-lowering drugs — an adjusted hazard ratio of 0.87. The association held across four separate comparators: against metformin, 0.90; against DPP-4 inhibitors, 0.88; against thiazolidinediones, 0.84.^[1]

The finding that teaches you something

A 2026 cohort in JAMA Network Open examined 841,831 patients with breast cancer, and after propensity score matching compared three pairings. The results were spectacular and internally inconsistent in a very specific way.^[2]

A hazard ratio of 0.09 means 91% lower mortality over ten years. No drug does that in breast cancer. Chemotherapy does not do that. What produces a number like that is not pharmacology — it is the comparison group.

Compared against insulin, the drug looks miraculous. Compared against another modern drug used in similar patients, it does nothing measurable.

People on insulin have, on average, longer-standing and worse-controlled diabetes, more complications and worse general health than people started on a GLP-1. Comparing the two measures how sick each group was before anyone prescribed anything. The SGLT2 comparison removes most of that difference — similar era, similar patients, similar prescribing logic — and the effect disappears with it.

How to read any study like this

1. Find the comparison group before the result. “Lower risk” is meaningless until you know lower than what. A drug compared against sicker patients will always look good.
2. Distrust enormous effects in observational data. Hazard ratios below about 0.5 in a non-randomized study are usually a description of who was prescribed the drug.
3. Look for consistency across comparators. The JNCI result held at 0.84 to 0.90 across four; the breast cancer result went from 0.09 to nothing across three.
4. Convert to absolute numbers. 7.5 against 8.1 per 1,000 person-years is the same finding as “13% lower risk”, and it lands very differently.
5. Note what the authors themselves say. The breast cancer authors call for randomized trials, which is the appropriate conclusion from their own data.

So what is actually known

That GLP-1 use is associated with a slightly lower rate of obesity-related cancers in people with type 2 diabetes, by roughly a tenth, consistently across comparators. That is a real finding and a modest one, and it is entirely plausible — obesity is a recognized driver of those thirteen cancer types, and these drugs reduce obesity.

The randomized evidence, which we cover in the trials’ answer on cancer, points at little or no effect in either direction — consistent with a small association that observational data can see and short trials cannot. What is not known is whether they improve survival in people who already have cancer. The evidence claiming so does not survive contact with a fair comparison group. If you are being treated for cancer, this is a question for your oncologist, and our thyroid cancer article covers the separate matter of the boxed warning that appears on every label in this class.