“Scientists discover why Ozempic may not work for some people,” ScienceDaily ran on June 5, eight weeks after Stanford had quietly announced the finding in April. The readership that scrolls past a headline like that is thinking about pant sizes, not blood-sugar logs. The paper underneath is thinking about the latter. Its primary outcome is whether Type 2 diabetes patients hit a recommended HbA1c target after six months on a GLP-1 drug, and almost none of its evidence touches the question the headline asks.

The study itself is a careful piece of pharmacogenomics. A team led by Anna Gloyn at Stanford and Markus Stoffel at ETH Zurich, working through Genome Medicine, identified two loss-of-function variants in the PAM gene that travel independently in people of European ancestry, with minor allele frequencies of roughly one percent for p.S539W and five percent for p.D563G; about one in ten people carries one of them. In a pooled analysis of three diabetes trials covering 1,119 participants, around a quarter of non-carriers reached the recommended HbA1c target after six months on a GLP-1. Among p.S539W carriers it was 11.5 percent. Among p.D563G carriers, 18.5 percent. Metformin, sulfonylureas, and DPP-4 inhibitors worked the same regardless of genotype. That class specificity is what makes the signal more than statistical noise.

Then comes the part the headline pretends does not exist. Only two of the trials in the main analysis collected weight data, and in those two the researchers saw no difference in weight loss between carriers and non-carriers. The Stanford press materials concede the obvious limitation in the next breath, calling the data “too limited to be conclusive.” A null result from a sample that small is not evidence the drugs work the same for weight, but it is even further from evidence they work less well, which is the inference a “why Ozempic may not work” headline manufactures. The piece that traveled imported a conclusion the paper explicitly declined to reach.

The biology cuts the same way. Carriers of the PAM variants do not have less GLP-1 in circulation; they have more. The enzyme PAM encodes, peptidyl-glycine alpha-amidating monooxygenase, normally caps peptide hormones with an amide group that lets them dock and signal, and the variants appear to blunt that step, leaving the body with elevated levels of hormone that does not fully land. The clean phrase “GLP-1 resistance” echoes “insulin resistance” and gives a familiar shape to a story that does not yet have a settled mechanism. Gloyn herself called it “the million-dollar question,” telling Stanford the team had “ticked off this enormous list of all the ways in which we thought GLP-1 resistance might come about” without nailing the answer. A vivid metaphor with the receipt still pending.

The buried complication is the one that should have set the headline writer’s hand off the keyboard. Two additional pharmaceutical-company-sponsored trials, excluded from the main meta-analysis on methodological grounds, used longer-acting GLP-1 receptor agonists and showed similar responses between carriers and non-carriers. Semaglutide, the molecule in Ozempic, is a long-acting GLP-1 receptor agonist. The set of trials in which the carrier effect appears at all is not the set of trials that resembles the drug the headline names. That is a fact the paper carries and the press cycle did not.

Read the disclosed competing interests and the press cycle starts to make a different kind of sense. The published paper, which is open access in Genome Medicine and trivial to look up, lists a Boehringer Ingelheim honorarium for one author, an AstraZeneca project grant for another, and a third author who is employed by and holds shares in Novo Nordisk A/S, the manufacturer of Ozempic. The funding line names, among others, the Novo Nordisk Foundation. That foundation is not a hands-off philanthropic relation: through its investment arm Novo Holdings, it controls roughly 77 percent of the voting rights in Novo Nordisk A/S and is required by its own articles to maintain a controlling interest. The single largest commercial beneficiary of the GLP-1 franchise sits at the top of an ownership chain that helped fund the study explaining why a slice of patients gets less out of the franchise. None of the consumer coverage flagged any of it.

The clinical pitch the lead author offers fits the moment. Gloyn told Stanford the goal was to help patients “get on the right drugs faster,” and floated the possibility of future medications that could “sensitize” people to GLP-1s. The first line is an argument for a pharmacogenomic test in front of a GLP-1 prescription, a billable layer of additional medicine bolted to a drug class that already commands premium prices. The second is an argument for a follow-on franchise. Both protect the brand from its real-world problem, which is the patient who tries the drug, sees a modest result, and stops refilling. A clean genetic story that explains those patients as carriers of a quiet biology is a better business outcome than a story in which the drug’s average effect is patchier than the trial sizzle.

The most honest sentence in the entire press cycle came from the senior author, on why the team kept digging. “We couldn’t understand this,” Gloyn said, “which is why we looked as many different ways as we could to see if this was a really robust observation.” A careful researcher describing a narrow investigation of a narrow question. The headline that traveled took that narrow question, swapped its outcome, dropped the trials that complicated it, and turned a diabetes pharmacogenomics paper into an Ozempic story. Somewhere between the journal and the feed, the work got bigger and the evidence got smaller.

Sources

  1. Genome Medicine – PAM gene variants and GLP-1 drug response, Gloyn & Stoffel et al. (2026)
  2. Stanford Medicine – One in 10 people may have resistance to GLP-1 diabetes drugs (2026)
  3. Stanford Medicine News – GLP-1 resistance research (2026)
  4. ScienceDaily – Scientists discover why Ozempic may not work for some people (2026)
  5. medRxiv – PAM variants and GLP-1 receptor agonist response, Umapathysivam et al. preprint (2023)
  6. Novo Nordisk Fonden – Ownership: Novo Holdings voting rights in Novo Nordisk A/S