The question of GLP-1 and cancer has swung back and forth for two decades, from thyroid tumors in rodents to hopeful headlines about lower cancer rates in large health-record studies.
A research letter published November 3, 2026, in JAMA Network Open adds a careful new data point.
Using electronic health records from UT Southwestern Medical Center in Dallas, researchers found that people with type 2 diabetes who started a GLP-1 receptor agonist had a lower estimated risk of obesity-related cancer over three and five years than people who started insulin or a sulfonylurea.
Against three other diabetes drug classes, the difference was not statistically clear.
That distinction is the story. A drug comparison that looks favorable against insulin may look much less remarkable when the alternative is metformin, an SGLT2 inhibitor or a DPP-4 inhibitor.
The evidence is from humans, but it is observational. It can identify associations, not prove that GLP-1 drugs prevent cancer. The authors explicitly caution against choosing a diabetes medicine for that purpose.
How the GLP-1 and Cancer Study Was Built
The research team, led by Jesús Gibran Hernández-Pérez and including David S. Lopez of UT Southwestern’s School of Public Health and Simmons Comprehensive Cancer Center, examined electronic health records from a source population of 359,595 adults who received care between January 2010 and July 2025.
That figure describes the starting database, not the number of patients included in every analysis.
From those records, the investigators separately identified eligible adults with type 2 diabetes, no previous obesity-related cancer and a 365-day washout period without the treatment being compared.
They then constructed five separate target trial emulations.
A target trial emulation uses observational records to approximate the structure of a randomized clinical trial. Here, the researchers compared people beginning a GLP-1 receptor agonist with people beginning one of five alternative diabetes treatments.
The five alternatives were metformin, insulin, sulfonylureas, SGLT2 inhibitors and DPP-4 inhibitors.
Each comparison involved a different group of eligible patients. Depending on the comparator, between 5,835 and 18,007 people started a GLP-1 drug, with separate groups starting the alternative treatment.
The outcome was the first recorded diagnosis of an obesity-related cancer during follow-up.
Obesity-related cancers are cancers associated with excess body weight. Researchers have reason to investigate whether medicines that improve metabolic health and reduce weight might influence their incidence, but that biological possibility does not establish a cancer-prevention effect.
The team also classified patients by the severity of their diabetes before treatment began.
Low severity meant no recorded complications covered by the study’s severity index. Moderate severity included microvascular complications or earlier stages of chronic kidney disease without a high-severity condition. High severity included macrovascular disease or advanced chronic kidney disease.
That distinction matters because doctors do not prescribe diabetes medicines at random. Patients who start insulin often have different disease histories from those who start a newer treatment.
The researchers used statistical weighting and additional adjustments to make the treatment groups more comparable. Those techniques improve the analysis, but they cannot remove differences that were never measured or recorded.
What the Cancer Risk Comparisons Showed
The strongest results favored GLP-1 drugs over insulin and sulfonylureas.
At three years, people starting a GLP-1 receptor agonist had an estimated obesity-related cancer risk of 2.01%, compared with 3.26% among those starting insulin.
That was an absolute risk difference of 1.25 percentage points and a hazard ratio of 0.62, with a 95% confidence interval of 0.49 to 0.77.
A hazard ratio below 1 indicates a lower estimated cancer hazard in the GLP-1 group. It should not be confused with the absolute percentage-point difference in the probability of developing cancer.
At five years, the insulin comparison remained favorable to GLP-1 drugs, with an absolute risk difference of 1.67 percentage points and a hazard ratio of 0.61.
The sulfonylurea comparison followed a similar pattern.
At three years, GLP-1 initiation was associated with a 1.14-percentage-point lower absolute cancer risk and a hazard ratio of 0.67.
At five years, the sulfonylurea comparison produced a hazard ratio of 0.75, with a 95% confidence interval of 0.61 to 0.92.
However, the five-year absolute risk difference was 0.72 percentage points lower, with a confidence interval extending from 1.83 points lower to 0.31 points higher. That interval crossed zero, so the absolute difference was not statistically clear.
Against the remaining three drug classes, the apparent advantage weakened.
At three years, the hazard ratio was 0.78 against metformin, 0.81 against SGLT2 inhibitors and 0.84 against DPP-4 inhibitors. Their confidence intervals included 1, meaning the estimates did not demonstrate statistically clear differences in cancer hazard.
The five-year estimates were similarly inconclusive for those three comparators.
That does not establish that GLP-1 drugs and the other medicines have identical effects on cancer. It means the analysis did not provide sufficiently clear evidence of a difference in those comparisons.
The apparent cancer advantage was clear against insulin and sulfonylureas. Against three other diabetes drug classes, it was not statistically established.
That pattern is what makes the study more useful than a simple headline claiming GLP-1 drugs lower cancer risk.
The Diabetes Severity Twist
The researchers also asked whether the associations changed depending on how advanced a patient’s diabetes was before treatment.
Among patients with low diabetes severity, starting a GLP-1 drug was associated with lower three-year cancer risk than starting insulin or a sulfonylurea.
The absolute risk differences were 1.10 percentage points lower against insulin and 1.22 percentage points lower against sulfonylureas.
The results were less consistent among patients with more severe diabetes.
One comparison moved in the opposite direction.
In the high-severity group, GLP-1 initiators had a 1.61-percentage-point higher estimated three-year cancer risk than patients starting an SGLT2 inhibitor. The 95% confidence interval extended from 0.17 to 2.89 percentage points higher.
The five-year absolute risk difference in that subgroup also favored SGLT2 inhibitors, although the corresponding hazard-ratio estimates remained highly uncertain.
It would be a mistake to turn this subgroup result into a claim that GLP-1 drugs raise cancer risk in patients with advanced diabetes.
The investigators classified the severity-based comparisons as exploratory. Some subgroups included relatively few patients and cancer events, and the confidence intervals were not adjusted for the large number of comparisons being examined.
In particular, the high-severity SGLT2 comparison included only 237 patients in the SGLT2 treatment group, with fewer than 11 recorded cancer events at three years.
Small event counts can make estimates unstable, especially when patients receiving different treatments already differ in their underlying health.
One unexpected finding in a sea of exploratory comparisons should be treated as a question, not an answer.
Why the Comparison Drug Changes the Cancer Story
The contrast between insulin and the other diabetes medicines is not necessarily evidence that GLP-1 drugs have a direct anticancer effect.
Insulin is often introduced later in the course of type 2 diabetes, when patients may already have more advanced disease, longer exposure to high blood sugar or other metabolic complications.
Sulfonylureas can also be used in patients with different treatment histories, and both insulin and sulfonylureas may contribute to weight gain.
A comparison favoring GLP-1 drugs against those treatments could therefore have several explanations.
It may reflect a genuine biological benefit related to weight loss or metabolic improvements. It may reflect differences in who receives the drugs. Or it may involve both mechanisms.
The new-user, active-comparator design helps address some of these problems.
Rather than comparing people taking GLP-1 medicines with everyone not taking them, the researchers compared patients beginning alternative diabetes treatments and attempted to account for measured baseline differences.
That is a stronger approach than a simple treated-versus-untreated comparison, but it does not turn electronic health records into randomized evidence.
The question of GLP-1 and cancer also has a separate safety history.
Semaglutide products carry an FDA boxed warning about thyroid C-cell tumors observed in rodents. The human relevance of those findings remains uncertain, including whether semaglutide causes medullary thyroid carcinoma in people.
The warning includes a contraindication for patients with a personal or family history of medullary thyroid carcinoma or multiple endocrine neoplasia syndrome type 2.
Grey Pepper previously examined that distinction in its explainer on the rodent evidence behind semaglutide’s thyroid cancer warning.
The new JAMA Network Open letter examines obesity-related cancers as a combined outcome. It does not establish what GLP-1 medicines do to the risk of every individual cancer, and it cannot resolve uncertainty around rare medullary thyroid tumors.
For the wider scientific debate, see Grey Pepper’s coverage of the GLP-1 and cancer evidence discussed in a Science special issue.
This analysis adds a useful piece of evidence to that debate, but its central contribution is not a new molecular explanation.
It is showing how the apparent association changes when researchers change the treatment used for comparison.
What the Study Cannot Tell Us
The authors list several limitations, and they are substantial.
First, the analysis used retrospective electronic health records from one medical system. Diagnoses, treatments and follow-up encounters occurring elsewhere may not have been fully captured.
Second, statistical weighting cannot remove confounding from information that is missing or poorly measured.
Differences in disease duration, glycemic exposure, treatment access, cancer surveillance and other patient characteristics may continue to influence the results even after adjustment.
Third, the treatment groups did not always have fully overlapping characteristics. Comparing patients who were unlikely to receive the same drugs makes causal interpretation more difficult.
Fourth, some severity subgroups had few cancer events. That makes apparently striking differences less dependable than they might look in a headline.
Fifth, many GLP-1 users began treatment relatively recently. As a result, the five-year analysis depended disproportionately on people who started treatment earlier, potentially limiting how well those results apply to current prescribing patterns.
The timing also creates a broader problem.
Cancer can take many years to develop and become detectable. A three- or five-year observation window may capture some meaningful associations, but it cannot provide a complete picture of long-term carcinogenic or protective effects.
Five years is a long time in a drug’s commercial life and a short time in a tumor’s.
The study also evaluated a broad outcome covering obesity-related cancers rather than providing sufficiently detailed evidence for each cancer type.
One drug might plausibly affect different tumor types in different ways. A combined outcome could obscure those distinctions, particularly when uncommon cancers contribute few events.
Finally, this was not a randomized clinical trial designed to test whether GLP-1 drugs prevent cancer.
The target trial approach makes better use of existing records, but it cannot eliminate all the biases that distinguish observational research from random assignment.
The researchers call for longer-term studies across multiple health systems.
The Takeaway: Reassurance Is Not Cancer Prevention
For people with type 2 diabetes, the findings are broadly reassuring about the overall obesity-related cancer outcome examined.
Across the five main treatment comparisons, GLP-1 initiation was not associated with a statistically clear increase in overall obesity-related cancer risk.
Against insulin and sulfonylureas, the analysis found lower estimated cancer risk, although the strength and precision of the evidence differed depending on the outcome measure and follow-up period.
For anyone hoping these medicines can double as cancer-prevention treatments, the result is more sobering.
The apparent advantage was not consistently demonstrated against metformin, SGLT2 inhibitors or DPP-4 inhibitors.
That pattern raises the possibility that treatment selection and underlying diabetes severity explain part of the apparent benefit.
The investigators’ conclusion is appropriately narrow: the findings do not support choosing glucose-lowering treatment specifically to prevent cancer.
Established benefits involving blood sugar control, body weight, cardiovascular disease and kidney health should remain central to treatment decisions, alongside a patient’s clinical history and safety considerations.
The next stage of this research will require longer follow-up, more diverse patient populations and more precise analyses of individual cancer types.
Randomized studies that track cancer outcomes over sufficient periods could eventually provide stronger evidence, although such questions are difficult and expensive to answer.
Until then, this research letter demonstrates something worth remembering whenever a health-record study produces an impressive percentage.
The question is not only whether a treatment looks better.
It is better than what, in which patients, and under what assumptions?
For GLP-1 drugs and cancer, changing the comparison drug changed how convincing the apparent benefit looked. That is a finding worth reporting without turning it into a promise of cancer prevention.
Sources and Further Reading
- Hernández-Pérez JG, Abdelgadir O, Torres-Sánchez L, et al. GLP-1 Receptor Agonists and Obesity-Related Cancer by Diabetes Severity. JAMA Network Open. November 3, 2026;9(11):e2641869. DOI: 10.1001/jamanetworkopen.2026.41869.
- PubMed. GLP-1 Receptor Agonists and Obesity-Related Cancer by Diabetes Severity. PMID: 42975902. November 2026.
- PubMed Central. Full-text archive of the JAMA Network Open research letter and associated methodological information. PMCID: PMC13732504.
- DailyMed. Ozempic (semaglutide) Full Prescribing Information, including boxed warning about thyroid C-cell tumors and relevant contraindications.
