In November, Novo Nordisk said its two big trials of semaglutide in early Alzheimer's disease had failed. The drug, the same molecule sold as Ozempic and Wegovy, did not slow cognitive decline at the point the trials were designed to measure, and the company shut the studies down. A new paper argues the company stopped looking too soon.
The trials, EVOKE and EVOKE+, randomized 3,808 adults aged 55 to 85 with mild cognitive impairment or mild dementia from Alzheimer's to oral semaglutide or placebo. The prespecified readout was a clinical dementia score at week 104. It came back negative, the full results were published in The Lancet ↗00459-9/fulltext), and Novo discontinued the extension phases. Brain biomarkers had moved in the right direction, but the thinking and daily-function scores had not.
The trials kept collecting data past week 104, out to weeks 130 and 156, that was never formally analyzed once the primary endpoint missed. In a paper in the Journal of Alzheimer's Disease ↗, neuroscientist Christian Holscher took the published group averages from those later visits, reconstructed the spread, and ran his own statistical tests. On one daily-function scale in EVOKE at week 130, he reports a difference favoring semaglutide that clears the usual significance bar (p = 0.0039). A common cognition test trended the same way at week 156, and some spinal-fluid markers separated too.
Read carefully, this is a contrarian reanalysis, not new trial data, and the caveats are heavy. Holscher worked from summary statistics, the averages and error bars in the published reports, not the underlying patient records. Tests run after the fact on data points that were not the trial's prespecified endpoint, without correction for the many comparisons involved, are exactly the kind of analysis that turns up false positives. He is also a long-standing advocate of using GLP-1 drugs against neurodegeneration and is affiliated with a company developing them, a conflict the paper discloses. One author re-reading someone else's failed trial is not the same as the trial succeeding.
The mechanistic argument underneath is the more interesting part. Semaglutide was engineered to linger in the bloodstream for a week at a time, which is great for weight loss and bad for the brain, because a molecule built to stay in the blood crosses into brain tissue poorly. Holscher's reading is that the drug may have nudged Alzheimer's at the edges precisely because so little of it reached the target organ, and that a GLP-1 drug designed to cross the blood-brain barrier could do more. That is a hypothesis, not a finding, but it is testable, and several brain-penetrant GLP-1 candidates are in development.
Either way the semaglutide card ↗ now carries a failed Alzheimer's indication, and the GLP-1R ↗ story in the brain did not end cleanly. The official result stands: semaglutide did not slow Alzheimer's on the measure that counted. Whether the GLP-1 idea in dementia is dead or just aimed at the wrong molecule is the question this paper is really asking, and it is a fair one to keep open.