GLP-1s and Cancer Part 2 - Early Research

This is the Weight and Healthcare newsletter! If you like what you are reading, please consider subscribing and/or sharing!

In part 1 I discussed the basics of the deluge of studies coming out looking at GLP-1s and cancer. If you haven’t read part 1 yet, I would recommend doing so as I’ll be referring to it a lot as I move through the rest of this series.

Today we’re going to look at a study that considers relatively earlier data around GLP-1s and breast cancer. I want to point out that this will not be a full deep dive because a full deep dive is often not needed to understand the findings and their limitations. If you want a deeper dive into this (or any) study, let me know in the comments.

Finally, I also want to note that there are many types of cancer, including within the category of “breast cancer” and a lot of work around cancer treatment has focused on more precision treatments based on the type and subtypes of of cancers, so the idea of these broad studies of “cancer” without specifics about the cancer and/or the treatments are definitely a limitation that should be addressed.

Summary

A retrospective cohort analysis (a study that analyzed previous medical records, which creates significant limitations, including not knowing the income of participants which could represent a significant confounding variable) found a link (correlation, not causation,) between GLP-1 use and improved outcomes among cis women with breast cancer and BMIs over 30 or with Type 2 Diabetes. The study did not consider which GLP-1 drug people were on or at what dose. Because of the dates of the records the newest GLP-1 for weight loss wasn’t yet approved and through much of the study, earlier iterations of GLP-1s were the only options available. The inclusion criteria was just that participants had received at least two prescriptions for the drugs. The study authors themselves were clear that this study does not determine causation and that it only suggests that the subject is worthy of future study.

Deeper Dive

The study we are discussing today is Tatum KL, Dahman B, Stevenson A, et al. Survival and Recurrence With GLP-1 Receptor Agonists in Breast Cancer. JAMA Netw Open. 2026;9(5):e2612133. doi:10.1001/jamanetworkopen.2026.12133

Possible Conflicts

Funding: “Supported in part by the Virginia Commonwealth University Massey Comprehensive Cancer Center Biostatistics Shared Resource, which was supported in part with funding from National Institutes of Health National Cancer Institute Cancer Center Support grant P30CA016059, and by the C. Kenneth and Dianne Wright Center for Clinical and Translational Research, supported in part by Clinical and Translational Science Awards grant UM1TR004360.”

The Wright Center has an “industry collaboration” with TriNetX (which is the database used in this study and which I’ll discuss in a moment) but I didn’t find a connection between these funders and the weight loss industry.

Disclosures: The disclosures stated “Dr Tatum reported receiving nonfinancial support from TriNetX LLC during the conduct of the study. Dr Williford reported being an employee of TriNetX at the time this study was performed. Dr McGuire reported receiving personal fees from Hologic Inc, Kubtec Inc, and Axogen Inc outside the submitted work. No other disclosures were reported.”

Refreshingly, I didn’t find any connections to the GLP-1 drug manufacturers among the authors but I think these connections to TriNetX are important. As we move through this series we are going to see this database over and over again. In a moment I’ll talk about the limitations of the database itself but their zeal to be included in studies may be part of why we’re seeing so many of these retrospective cohort analyses.

Study Basics

The question they asked was “Is there a potential benefit associated with the use of glucagon-like peptide-1 receptor agonists (GLP-1 RAs) in female breast cancer survival and cancer recurrence among patients with ob*sity or with type 2 diabetes (T2D)?”

Immediately we see that they are clear that, at best, they can discover a “potential” benefit. That’s important because reporting (and social media) around these studies often overstate the conclusion, sometimes vastly.

This is a retrospective cohort analysis, the limitations and concerns of which I discussed extensively in Part 1 of this series.

Study database and possible issues

They used data from the TriNetX US Collaborative Network from “women (≥18 years) with breast cancer …from 68 health care organizations who received a diagnosis between April 1, 2006, and April 1, 2023”

Before we get further into the study, let’s discuss the TriNetX US Collaborative Network database because, again, it’s being used in a lot of this research and it’s worth understanding. This database was the subject of a 2025 paper called “TriNetX and Real-World Evidence: A Critical Review of Its Strengths, Limitations, and Bias Considerations in Clinical Research”. By Nassar et al.

The authors point out that the database, made up primarily of data “from insured populations in academic and acute care settings” has inherent limitations. After conducting a “comprehensive review examining TriNetX’s data architecture, quality metrics, and research applications, focusing on data integrity, platform architecture, and the external validity of research finding” they found “significant methodological considerations.” They point out that “TriNetX’s reliance on retrospective data introduces biases such as selection bias and confounding variables. The coding accuracy of electronic health records, which have not been independently validated, is a critical determinant of data reliability. The demographic representation is limited, affecting the generalizability of results.“ Ultimately, they recommend that “TriNetX is a valuable resource for healthcare research. However, its limitations must be acknowledged/”

To their credit, this Tatum et al, study about GLP-1s and breast cancer does acknowledge most of this in their limitations section, which is great. Unfortunately, the media reporting on this rarely, if ever, does.We’ll also come back to the limitations section in a bit.

Population

The study authors divided the group into three cohorts:

Patients with a BMI of 30 or above, divided into GLP-1 users and GLP-1 non-users (n=1610)

Patients with Type 2 Diabetes (T2D) divided into GLP-1 use versus insulin or metformin (n=2323)

Patients with T2D divided into GLP-1 use versus SGLT2 Inhibitors (n=4052)

The used propensity score matching which is a technique to try to control for possible confounding variables. Sometimes it is suggested that the use of propensity matching can make it possible for causal inference (meaning that rather than just saying that the study found that GLP-1 use and positive outcomes happened at the same time, they could claim that GLP-1s caused the positive outcomes.) The issue with this is that the causal inference is only as good as the items that were used in propensity matching. In this case they used “demographic characteristics, diagnosis, procedures, and medications.” Those demographic characteristics did not include income/socioeconomic status which could be what “GLP-1 use” is standing in for and thus could be a major confounding factor (as I discussed in Part 1.)

These authors specifically state that “causality cannot be inferred.”

Findings

Topline findings were:

Patent with T2D who took GLP-1s had lower hazard of all cause mortality and recurrence free survival (RFS) than those who took insulin or metformin.There were no significant differences between those who took GLP-1s and those who took and sodium-glucose cotransporter 2 inhibitor groups.

For patients with BMI of 30+, GLP-1 use correlated with lower hazard of all cause mortality and RFS over 10 year follow up.

Their conclusion states “In this cohort study of patients with BC, findings suggested a potential association between GLP-1 RA use and improved outcomes among patients with BC who have ob*sity and related metabolic conditions.”

This is a study from which it would be reasonable to consider additional studies to get more clarity or, as the authors put it “These findings support further evaluation of GLP-1 RA therapy in randomized clinical trials.” But, as the study authors themselves point out, it is not appropriate for clinical decision making and it has some serious weaknesses, some of which they were clear about, but some I think they missed.

Limitations

In the limitations section they mention the issues we talked earlier with the database, the fact that causality can’t be inferred, some limitations with their statistical analysis, the fact that the use of EHR diagnostic codes means that “the accuracy of cancer characteristics and causes of death is unclear,” the lack of standard definitions for disease progression (including recurrence and metastasis). They include the fact that propensity score matching “reduces the number of patients available for outcomes and certain comparison groups.” Further, the precision of the 10-year Kaplan-Meier survival estimates and hazard ratios was limited by “substantial administrative censoring around 5 years.”

One big issue (and they mention it) is that they grouped all GLP-1s together, regardless of what drug was being taken so there is no information about which GLP-1, or what dose, subjects were on. These cases were drawn from April 1, 2006 to April 1, 2023. Liraglutide wasn’t even launched until 2010 so the first four years it appears to me that every patient would have been on exenatide which only has a T2D dose. To me, it is a massive weakness that the study does not draw any distinction between exenatide and, for example, semaglutide at the mega weight loss dose. Mounjaro (tirzeptaide for T2D) was only approved 5/2021 and Zepbound (tirzepatide for weight loss) wasn’t approved until after this date range.

The authors point out that “The only requirement was that the participant in the GLP-1 group had 2 or more prescriptions for GLP-1s.” They note that “it remains unclear whether the observed long-term outcomes reflect initial metabolic changes or sustained therapy.”

In fact, it remains unclear if GLP-1s had anything to do with the outcomes at all. Beyond their, what I would call possible over-confidence, in GLP-1s here, this is a significant limitation. I think the fact that SGLT2s (a class of Type 2 Diabetes drug that includes Jardiance, Invokana, and Farxiga) and GLP-1s showed no meaningful difference is a point that should generate considerable interest here. The manufacturers of GLP-1s have been more aggressive, and I would say more successful, at generating research and marketing connecting their drug with benefits outside of T2D, but it’s possible that if GLP-1s offer a benefit in breast cancer, SGLT2 inhibitors might offer similar benefits with a favorable risk profile for at least some patients. Further, if patients were only required to have had two prescriptions for GLP-1s, for some of the drugs that means they may have barely gone beyond the subtherapeutic dose, assuming they took them (as getting a prescription and taking the drugs are two different things.)

All of which is to say that even if GLP-1s do have benefits around breast cancer, this could mean that taking a low dose for a short time is what creates the benefits (rather than, say, the weight loss megadose over a long period of time) so any future trials should consider this.

The authors explain “The lower-than-expected proportion of estrogen receptor–positive or estrogen receptor–negative tumors likely reflects incomplete capture of oncologic characteristics in structured EHR fields within TriNetX. This limitation is inherent to real-world EHR data and may affect other tumor and treatment characteristics.”

To me the most glaring issue that is not discussed in the limitations section is the lack of income in the propensity matching (which may be because it was not available in the database.) Still, as I wrote about extensively in part 1, GLP-1 use may be acting here as a proxy for money/access/other factors. The authors did include “socioeconomic risk” which is an ICD-10 code:

ICD-10-CM Codes › Z00-Z99 › Persons with potential health hazards related to socioeconomic and psychosocial circumstances Z55-Z65

Codes

Z55 Problems related to education and literacy

Z56 Problems related to employment and unemployment

Z57 Occupational exposure to risk factors

Z58 Problems related to physical environment

Z59 Problems related to housing and economic circumstances

Z60 Problems related to social environment

Z62 Problems related to upbringing

Z63 Other problems related to primary support group, including family circumstances

This is certainly better than nothing but it’s not the same thing as capturing income. People’s income could be too low to properly access healthcare or any of the other factors that those with more money might be able to access and that might impact the studied outcomes, but still not necessarily be captured in this code. Further, it’s unclear how accurately these codes were captured in the medical records that were used for this study.

So, as the authors themselves point out, the study does not show that GLP-1s cause improved cancer survival and/or reduced cancer recurrence among cis female breast cancer patients with BMIs of 30+ or with T2D. In fairness, the study wasn’t designed to make that determination, but that hasn’t stopped reporters and social media influencers from overstating the study conclusions.

Again, if there is a medication that prevents cancer and/or reduces recurrence and/or improves survival that would be great. As I’ve said, GLP-1s have actual benefits for Type 2 Diabetes and may have other actual health benefits as well, but these must be properly tested and not just launched into practice via a marketing campaign around studies with significant limitations.

I think it’s important to remember that the manufacturers of these drugs, their astroturf “patient advocacy” groups, and the researchers and doctors on their payroll who are quick to give media interviews and slow to disclose conflicts of interest and have a history of vastly overstating the benefits/evidence for these drugs and I would argue that it is worth it to be cautious lest people expose themselves to the side effects of these drugs (some of which are serious and can be fatal) for what may be no actual benefit.

Also, if income/access is the true driver of better outcomes, healthcare risks going in the entire wrong direction by focusing on getting as many people on these drugs as possible when, in truth, increasing income and access is what would help.

As I discussed in part 1, as you are reading these articles (or other articles about GLP-1s or other medications, ask yourself if the studies they are citing have answered the following questions:

What drug(s)? (Are they just saying “GLP-1s” or have they identified specific drugs?)

At what dose(s)? (Were the impacts different for different doses?)

For what duration? (How long did people have to be on the drugs for there to have been an effect)

At what time? (Did the person need to be taking the drugs prior to diagnosis, was there a benefit to starting after diagnosis? etc.)

For what benefit (specific numbers around prevention/slowed progression/lowered mortality etc. on what types of cancer with what other treatments)?

At what effect size?

For which people? (Were the results the same for all groups? Who wasn’t included/was underrepresented - typically People of Color and trans and nonbinary people, but could be others as well.)

With what side effects?

Are GLP-1s superior to other interventions (SGLT2 inhibitors/weight-inclusive interventions etc.)

In part 3 we’ll look at the Guardian Article that so many of you have asked me about.

Liked the piece? Share the piece!

More research
The Research Post

More resources
The Resource Post

*Note on language: I use “fat” as a neutral descriptor as used by the fat activist community, I use “ob*se” and “overw*ight” to acknowledge that these are terms that were created to medicalize and pathologize fat bodies, with roots in racism and specifically anti-Blackness. Please read Sabrina Strings’ Fearing the Black Body – the Racial Origins of Fat Phobia and Da’Shaun Harrison’s Belly of the Beast: The Politics of Anti-Fatness as Anti-Blackness for more on this.