Loading ...

A large real-world study is raising the possibility that GLP-1 drugs can improve long-term survival after breast cancer. But critics charge that the analysis is full of fundamental flaws.

The study, which combed electronic health records from over 800,000 US women with breast cancer, found that GLP-1 users had a substantially reduced risk of dying over 10 years — as much as 91% lower for women who used the drugs to manage diabetes than for those using other diabetes medications. The risk for breast cancer recurrence was also significantly curbed.

The researchers say their findings, published in JAMA Network Open, warrant further evaluation in randomized trials.

But outside experts pointed to a list of potential biases in the analysis that put the authors’ conclusions into doubt.

The steep drop in mortality with GLP-1 use, in particular, is a “screaming red flag,” according to Matt Spick, health-data scientist at the University of Surrey in Guildford, England. He characterized the survival benefit as implausible.

‘Big’ Data on GLP-1s and Cancer

As use of GLP-1 medications, particularly semaglutide (Ozempic/Wegovy), has soared in recent years, questions about their broader health effects have followed. Several studies — also using large real-world databases — have suggested that GLP-1 users have lower risks of developing certain cancers, particularly those related to obesity. Fewer have looked at whether GLP-1 use has any relationship with cancer outcomes.

For the new study, researchers led by Kristina L. Tatum, PsyD, of Virginia Commonwealth University (VCU) in Richmond, Virginia, analyzed data from the TriNetX research platform. TriNetX aggregates electronic health record data from over 100 million patients at dozens of US healthcare organizations and has built-in analytical tools.

The researchers identified over 800,000 women who were diagnosed with stages I-III breast cancer between 2006 and 2023, then used propensity-score matching to construct three cohorts: One compared GLP-1 use against nonuse among 1610 patients with obesity but not type 2 diabetes. Another compared GLP-1 use with insulin or metformin use among 2323 patients with diabetes. The third, involving 4052 patients with diabetes, compared GLP-1 use against use of SGLT2 inhibitors.

The index date for all groups was the date of breast cancer diagnosis, and GLP-1 use was defined as two or more prescriptions within 6 months before or any time after breast cancer diagnosis.

Among patients with obesity, GLP-1 use was tied to a significantly lower risk of all-cause mortality over 10 years (hazard ratio [HR], 0.35). The benefit was even greater among patients with diabetes when compared against metformin or insulin use (HR, 0.09). Adjusting for covariates did not change the estimates much.

The comparison of GLP-1s and SGLT2s did not yield significant results in unadjusted analysis — though after adjustment, GLP-1 drugs were again linked to lower all-cause mortality (HR, 0.77).

Recurrence-free survival was also significantly better among GLP-1 users — both those with obesity (HR, 0.44) and those with diabetes (HR, 0.33 vs insulin or metformin).

Study co-author Bernard F. Fuemmeler, PhD, also with VCU, called the findings “very promising.”

“But,” he said, “we need to follow this study up with carefully designed and conducted randomized controlled trials to determine clinical recommendations.”

Differing Interpretations

An editorial published with the study described the results as “striking” and said that while there could be other explanations for the improved outcomes tied to GLP-1 use, they are “not immediately apparent.”

Other experts begged to differ. Both Spick and Joshua Wang, researcher at Taipei Tzu Chi Hospital in New Taipei City, Taiwan, who trains his colleagues on TriNetX, highlighted key flaws in the analysis.

One is calendar-time bias: Because GLP-1 use has only taken off in recent years, many patients in that study group would have survived into the 2020s, when breast cancer care and outcomes had already improved overall. Meanwhile, patients in the control groups would have met their inclusion criteria throughout the study period, dating back to 2006.

“Any benefit seen in the GLP-1 cohort could simply be due to secular improvements in breast cancer mortality rates over the past 20 years,” Wang said.

He added that when he’s corrected for calendar-time bias in other TriNetX studies examining GLP-1 drugs, “the primary conclusions of the study can be changed.”

Healthy-user bias is another issue. Spick noted that, especially in the years before GLP-1s surged in popularity, patients who received prescriptions would have been well-insured and “engaged,” with doctors who were particularly motivated, as well.

Those characteristics would independently predict better cancer outcomes, Spick said.

Then there’s the matter of immortal-time bias, a classic problem in pharmaco-epidemiology that refers to the simple fact that patients cannot die until they get their prescriptions.

Take a patient diagnosed with breast cancer in 2006: Metformin and insulin were commonly used at the time, meaning that patients in the control group would likely qualify for the study around the time of diagnosis. In contrast, GLP-1 use was rare in 2006, so patients in the GLP-1 group would have likely survived many years after their cancer diagnosis.

The authors did try to get around this problem with a technique called landmark analysis and found that it attenuated the association between GLP-1 use and mortality. However, they used adjustments of only 6 and 12 months, which Wang said was likely insufficient.

He also brought up yet another issue: The authors did not detail the parameters they used to perform their analysis on the TriNetX platform, making it “impossible to replicate their study for a closer analysis.”

One reason that’s a problem, Wang said, is that the paper’s description of the timing of patients’ prescriptions cannot actually be operationalized on the platform and would require some kind of workaround — potentially introducing further bias.

Fuemmeler defended the work, noting that the study’s methodological issues are known limitations with any electronic record data. He also said all of the study’s queries and methods are “reproducible by someone who has access to TriNetX” and stressed that his team had worked with biostatisticians, epidemiologists, and representative statistical consultants from TriNetX.

Despite several requests, however, the authors did not share the query parameters they used in the study.

For his part, Spick worried that, despite the study’s issues, the results could have real-world effects — potentially driving “medically unnecessary demand” for GLP-1s among people with breast cancer.

The study was funded by VCU, with support from the National Institutes of Health. A coauthor is an employee of TriNetX, LLC. Spick and Wang reported having no relevant disclosures.