Welcome back. In the previous lesson, a randomized trial gave a fairly direct answer about some short-term drinking outcomes, but not about every addiction or long-term recovery. Cognition presents a different evidence puzzle: people prescribed GLP-1 receptor agonists have sometimes had lower recorded rates of dementia, yet large trials of semaglutide in people who already had early Alzheimer’s disease did not show slower clinical decline.
This lesson separates risk of developing dementia from treatment of established Alzheimer’s disease. You will then examine why changes in blood or cerebrospinal fluid biomarkers cannot settle the clinical question on their own. Allow about 40–45 minutes, including the readings and figures.
What the observational association actually says
Dementia is an umbrella term for impairments in thinking and daily function; Alzheimer’s disease is one cause, while vascular disease and other processes can also contribute. That distinction matters before considering any drug result. An association with all-cause dementia in people with type 2 diabetes is not automatically evidence that a medicine prevents Alzheimer’s disease specifically.
Read the observational review with two questions in mind: who received the medicines, and what counted as an outcome?
Read Tang and colleagues’ review for both its estimate of the association and its account of why that estimate is difficult to interpret causally.
In the Abstract, read the Results through Conclusion, beginning with the pooled findings. In Results, find “GLP-1RAs and Risk of Dementia”; read from the study descriptions through the pooled estimate, including the heterogeneity figure that follows. Finally, in the later discussion of real-world treatment decisions, read from the limitations. Focus on differences between medicine users and nonusers that the researchers could not fully remove.
Across the studies pooled for the GLP-1 comparison, users had a lower recorded risk of all-cause dementia than nonusers: relative risk , with a 95% confidence interval from to . Put plainly, the estimated risk was about 28% lower relative to the comparison group. That is not a 28-percentage-point reduction, and the review does not establish an individual’s absolute benefit. The studies also varied substantially; the reported heterogeneity statistic was .
Why not call this proof of prevention? Prescribing is not random. Diabetes severity, other illnesses, age, access to care, and other treatments can influence both which medicine someone receives and whether dementia is later diagnosed. Records-based dementia diagnoses may also be imperfect. Comparing users with nonusers makes it particularly difficult to know which differences reflect the drug rather than the circumstances of receiving it. Statistical adjustment helps with measured differences, but cannot guarantee that the groups are otherwise comparable.
There are credible reasons to investigate the association. GLP-1 medicines can improve several metabolic or cardiovascular factors associated with later cognitive problems, and laboratory research suggests possible effects on brain-related processes. Neither route proves that the medicines prevent dementia in humans. In particular, the review found too little Alzheimer’s-specific evidence to pool a GLP-1 estimate for that diagnosis.
What EVOKE and EVOKE+ tested instead
EVOKE and EVOKE+ addressed a more focused, later-stage question: among people with biomarker-confirmed early Alzheimer’s disease, does oral semaglutide slow worsening compared with placebo? Together, the two randomized trials enrolled about 3,800 participants. This is a test of treatment after disease is already present, not a randomized test of whether taking a GLP-1 medicine years earlier prevents dementia.
The primary clinical measure was the Clinical Dementia Rating–Sum of Boxes, or CDR-SB. It combines assessments of cognition and everyday function. A rising score means worsening, so evidence of slower decline would require the semaglutide group’s score to rise less than the placebo group’s.
[PDF] evoke and evoke+: Two phase 3 randomised placebo-controlled ...
Use the presentation’s clinical-outcome slides to see what the randomized comparison found. These are trial-presentation materials; concentrate on the specified outcomes and time points rather than treating every plotted measure as a separate proof of benefit or harm.
Find the slide headed “Oral semaglutide did not slow cognitive and functional decline in participants with early AD versus placebo in either trial.” In its CDR-SB results, locate the EVOKE comparison and the EVOKE+ comparison at week 104. Then look at the following slide on the ADCS-ADL-MCI daily-function measure. Notice whether that separate clinical measure supports a benefit.
At week 104, the estimated CDR-SB difference between semaglutide and placebo was points in EVOKE (95% confidence interval to ) and points in EVOKE+ (95% confidence interval to ). In both trials, the confidence interval spans no difference. Daily-function results likewise did not establish a benefit. These findings do not prove that the true effect is exactly zero, but they do mean that these large trials did not demonstrate clinically meaningful slowing of early Alzheimer’s decline with this oral semaglutide regimen.
The figure below shows the trajectories through week 156. Both groups worsened over time, and their lines remained close. Keep its later follow-up estimates separate from the week-104 comparisons just reported; they refer to different time points.
00459-9/asset/8ec1c90a-c705-400f-abe4-6e4235385280/main.assets/gr2_lrg.jpg)
Randomization makes these trials much stronger evidence about the tested treatment effect than a comparison of people who happened to receive different prescriptions. Their scope still has boundaries: they tested oral semaglutide in people with existing early Alzheimer’s disease, not every GLP-1 medicine, every dementia subtype, or prevention before symptoms emerge.
Why biomarker changes do not overturn the clinical result
A biomarker is a measurable biological signal, such as a protein concentration in blood or cerebrospinal fluid (CSF). Biomarkers can help identify eligible participants and reveal whether a treatment changes some aspect of biology. But a person can have a changed biomarker without a detectable improvement in thinking or daily function.
The trials illustrate that gap. A CSF substudy examined Alzheimer’s-related and other markers; exploratory blood analyses examined additional markers. Review their sample sizes and timing before interpreting the forest plot.
[PDF] evoke and evoke+: Two phase 3 randomised placebo-controlled ...
Return to the trial presentation to connect its biomarker findings with its clinical conclusion.
On the slide “Trial design of the CSF sub-study,” read the substudy counts and note that CSF was assessed at week 78. Continue to the slides headed “CSF biomarkers – ratio to baseline at week 78” and “Exploratory blood-based biomarkers - ratio to baseline at week 104.” End at the “Conclusions” slide: read from the trial summary. Attend especially to its note that the biomarker comparisons were not adjusted for multiplicity.
00459-9/asset/7d84723d-f741-42b3-a880-9906d2699211/main.assets/gr4_lrg.jpg)
In the CSF substudy, some tau-related and other measures were lower with semaglutide than with placebo. Those findings are biologically interesting, but the substudy was far smaller than the full trials, and many biomarkers were examined without adjustment for multiple comparisons. Blood results were not a uniform pattern of improvement either: for example, GFAP was higher in both trials, while the plotted plasma p-tau comparisons did not show clear differences.
More fundamentally, a biomarker is not interchangeable with an outcome a person experiences. Lower measured CSF tau does not by itself show fewer brain tangles, better memory, or preserved independence. Even a genuine effect on the measured biology might be too small, occur too late, or fail to change the processes driving clinical decline.
Alzheimer’s Clinical Trials 2026: An Insider’s Look with Dr. Jeffrey Cummings
In BrightFocus Foundation’s “Alzheimer’s Clinical Trials 2026: An Insider’s Look with Dr. Jeffrey Cummings,” listen to the brief explanation of why a biological change and a clinical benefit require different evidence.
Watch biomarkers versus benefit. The speaker is discussing another experimental treatment, not semaglutide; use the passage for its general distinction between a marker shifting and cognitive decline demonstrably slowing.
Putting apparently conflicting findings together
The observational and randomized results need not be describing the same effect. They differ in population, timing, and outcome:
| Evidence | Population and comparison | What it can currently support |
|---|---|---|
| Observational dementia studies | Mainly people with type 2 diabetes who did or did not receive GLP-1 medicines | An association with lower recorded all-cause dementia risk, subject to confounding and diagnostic uncertainty |
| EVOKE and EVOKE+ | People with existing early Alzheimer’s disease randomized to oral semaglutide or placebo | No demonstrated slowing of clinical decline with the tested treatment over the primary trial period |
| Trial biomarker analyses | Trial participants, including a smaller CSF substudy | Evidence that some measured biological signals changed, not proof of cognitive benefit |
One possibility is that improving metabolic or vascular health before dementia develops could affect later risk without reversing or slowing established Alzheimer’s disease. Another is that some or all of the observational association reflects differences between people who received different medicines. The present evidence does not tell us which explanation accounts for the association. Nor do the negative treatment trials establish that prevention is impossible; they simply did not test prevention.
The evidence verdict is therefore precise: a preventive effect remains unproven, while oral semaglutide did not demonstrate benefit for slowing early Alzheimer’s disease in EVOKE or EVOKE+. Biomarker changes justify investigating mechanisms, not replacing the clinical result. Establishing prevention would require appropriately designed, sufficiently long studies that compare meaningful cognitive or dementia outcomes—not merely a favorable laboratory measure.
Conclusion
Three distinctions carry this lesson: association versus causation, prevention versus treatment, and biomarker change versus clinical benefit. Observational studies raise a worthwhile dementia-risk hypothesis; EVOKE and EVOKE+ provide stronger evidence against a slowing effect of the tested oral semaglutide regimen in people with early Alzheimer’s disease. Neither finding licenses a blanket claim about all GLP-1 medicines and all forms or stages of dementia.
The next lesson applies the same care to cancer claims, where animal findings, human safety observations, and long-term prevention questions also answer different things.