A viral diabetes-and-dementia explainer ran on The Conversation this week and got republished by ScienceDaily the next morning, and it is missing the single most important fact in its own GLP-1 section. Less than three weeks before it published, The Lancet ran the full Phase 3 results of the EVOKE and EVOKE+ trials of oral semaglutide in early Alzheimer’s disease. The trials enrolled 3,808 patients across two studies and ran for 104 weeks against placebo. The change in the Clinical Dementia Rating Sum of Boxes was 2.3 versus 2.3 in EVOKE (p=0.57) and 2.2 versus 2.1 in EVOKE+ (p=0.46). The drug did nothing. The explainer presents both trials in the present tense, as if we are waiting on the answer.
| Arm | EVOKE | EVOKE+ |
|---|---|---|
| Semaglutide | 2.3 | 2.2 |
| Placebo | 2.3 | 2.1 |
Phase 3 is where you find out. Phase 3 said no.
The biology the explainer gets right is what makes the omission sting. Diabetics carry roughly 60 percent higher dementia risk. Frequent hypoglycemic episodes lift that further. High glucose damages the small vessels that feed cortex, and the blood-brain barrier weakens behind it. None of that is hype. It is documented, replicated, and most readers have never heard it spelled out.
The part I came in ready to love is the insulin-receptor section. According to the explainer, the single biggest genetic risk factor for Alzheimer’s appears to do part of its damage by trapping the insulin receptor inside the neuron, so it cannot reach the cell surface where insulin can dock. From the outside the cell looks insulin-resistant, the same way a muscle in a type 2 diabetic does. From the inside it is starving in a sea of sugar. That is the kind of mechanism that explains why some researchers now call Alzheimer’s “type 3 diabetes,” a label coined back in 2008 by Suzanne de la Monte’s group at Brown. The brain runs on about 20 percent of your body’s energy for 2 percent of its mass, and dementia patients steadily lose the ability to burn glucose properly. As a way of looking at the disease, at least, the metabolic problem and the cognitive problem are one process showing up in two organs at once.
Then I hit the GLP-1 paragraphs, and the piece quietly fell apart.
The authors, Craig Beall and Natasha MacDonald, write that semaglutide is being tested in EVOKE and EVOKE Plus and that “records show” GLP-1 users have lower dementia risk than people on other diabetes drugs. The records line is the load-bearing one, and it does not say what they need it to say. The dementia signal for GLP-1s comes from registries and electronic health records, not trials. People who get prescribed Ozempic are not the people who get prescribed sulfonylureas. They are younger, wealthier, more compliant, less sick at baseline, more medicated everywhere else. Researchers call it confounding by indication, and even the 2024 metformin paper Beall and MacDonald cite for their metformin claim warns that the apparent protective effect “may be biased” and causality cannot be inferred from this design. The same caveat applies to every observational GLP-1 dementia paper they leaned on.
Then EVOKE ran. With 3,808 patients randomized, with placebo control, with a primary endpoint chosen years in advance and locked in, the answer came back null. Novo Nordisk did not yank the whole molecule, but it did discontinue the planned one-year open-label extension, which is the window where any late benefit on top of the 104-week curve would have shown. When a pharma company designs a trial to prove brain protection, sees a 0.0 versus 0.1 point gap on an 18-point scale at two years, and then cancels its own follow-up, the read is not “still promising.”
The SGLT2 paragraph has a related problem. The piece calls the gliflozins (Jardiance, Farxiga) “superior” to GLP-1s at preventing dementia, citing one large South Korean retrospective cohort from early 2025. A separate US electronic-health-record analysis in JAMA Neurology found no difference between the two classes. “Superior” is doing a lot of work in a sentence supported by one of two studies and zero randomized trials.
Now the funding. Beall and MacDonald disclose grants from Diabetes UK, Breakthrough T1D, the Steve Morgan Foundation, the Medical Research Council, NC3Rs, the Society for Endocrinology, and the British Society for Neuroendocrinology. None of those are pharma checks, and I am not accusing anyone of writing to order. What I am pointing at is the ecosystem the disclosure sits inside. Diabetes UK’s longest-running corporate partnership is with Novo Nordisk, the maker of semaglutide. The partnership has run since 2014, funds the charity’s Clinical Champions programme, and includes silver sponsorship of the Diabetes UK Professional Conference. UK diabetes research, conference culture, and patient-facing messaging all run through this circuitry. The framing it produces, intentional or not, is the framing the explainer landed on: optimistic about approved drug classes, silent about the trial that failed, ready to call one Korean cohort “superior” evidence.
Strip the brand stories out, though, and the rest still matters. Diabetics get dementia more often. Hypoglycemia hurts the brain. Insulin signaling in neurons is broken in Alzheimer’s, and APOE4 partly explains the mechanism. Glucose-damaged microvessels in the cortex are the kind of finding that, if it holds up at scale, would justify treating cognitive symptoms in a diabetic with the same vascular urgency you treat their feet. None of that needed a drug-class endorsement to land.
If I were watching this beat as a reader, I would not bet on semaglutide for cognition. I would bet on the unsexy work: glycemic control without hypoglycemia, blood pressure control, vascular health, and the next round of randomized trials in the SGLT2 class, where the cognition question is still open. The viral version of the story tells you a pill is coming. The Lancet readout, the one with 3,808 names and two years of follow-up, says it is not. The viral version did not mention the Lancet readout.
Sources
- Beall C, MacDonald N. “Ten ways diabetes and dementia are linked.” The Conversation, June 2026
- ScienceDaily republication: “10 surprising ways diabetes and dementia are connected,” June 16, 2026
- Cummings JL, Atri A, Sano M, et al. “Efficacy and safety of oral semaglutide 14 mg in early-stage symptomatic Alzheimer’s disease (EVOKE and EVOKE+).” The Lancet, May 30, 2026
- Alzheimer Europe: Novo Nordisk announces topline EVOKE/EVOKE+ results, November 2025
- Diabetes UK: “Our Partnership with Novo Nordisk” (corporate partner page)
- Comparative effectiveness of SGLT2 inhibitors and GLP-1 receptor agonists in preventing Alzheimer’s, vascular, and other dementia in type 2 diabetes (Korean retrospective cohort, 2025)
- AJMC: “SGLT2 Inhibitors, GLP-1 Agonists Similarly Reduce Dementia Risk for Patients With Diabetes”
- Underwood B-R, et al. “Incident dementia risk among patients with type 2 diabetes receiving metformin versus alternative oral glucose-lowering therapy.” UK CPRD observational cohort, 2024
- NeurologyLive: “GLP-1 Semaglutide Fails to Outperform Placebo in Phase 3 EVOKE Trial of Alzheimer Disease”