No GLP-1 drug is approved to treat or prevent Alzheimer’s disease. The evidence splits by study design. Randomized trials find nothing. Health records find a lot. This page sets out both, and why they disagree.
What the randomized trials found
ELAD tested liraglutide. It gave 204 people with mild to moderate Alzheimer’s daily liraglutide or placebo for 52 weeks [1]. None had diabetes. The primary outcome was cerebral glucose metabolism.
That outcome did not move. The difference was -0.17, with an interval from -0.39 to 0.06 and a P value of 0.14. The interval crosses zero.
One secondary outcome favored liraglutide. The ADAS executive domain differed by 0.15, with an interval from 0.03 to 0.28. Its P value of 0.01 is reported as unadjusted.
Pooling helps less than it sounds. A meta-analysis found four randomized trials in non-diabetic patients with Alzheimer’s or mild cognitive impairment [2]. Together they enrolled 112 people. The pooled effect was -0.10, with an interval from -0.53 to 0.34.
That is not evidence of absence. It is 112 patients. The correct statement is that a benefit has not been shown, not that it has been ruled out.
The phase 3 trials of oral semaglutide
The evoke and evoke+ trials tested oral semaglutide in early Alzheimer’s at phase 3 scale [4]. Participants were 55 to 85 years old. Each had mild cognitive impairment or mild dementia due to Alzheimer’s. Amyloid had to be confirmed by PET scan or spinal fluid. The dose climbed over eight weeks to 14 mg a day. The primary endpoint was change in CDR-SB at week 104.
The two trials randomized 3,808 people across 566 sites [5]. In evoke, CDR-SB rose 2.3 points on semaglutide and 2.3 on placebo. In evoke+, it rose 2.2 against 2.1. The differences were -0.08 (95% CI -0.35 to 0.20) and 0.10 (-0.17 to 0.38). Both trials were discontinued for a negative outcome. The full reading is in the trial that closed the question.
What the health records found
A target-trial emulation used a federated record system of over 29 million patients [3]. It matched 28,901 GLP-1 starters against 28,901 people starting a different diabetes drug. Matching used 30 baseline variables, plus index year and age band.
| Outcome | Hazard ratio | 95% CI |
|---|---|---|
| First Alzheimer's diagnosis | 0.46 | 0.29 to 0.73 |
| All-cause dementia | 0.66 | 0.55 to 0.79 |
| Mild cognitive impairment | 0.76 | 0.61 to 0.94 |
| All-cause mortality | 0.46 | 0.37 to 0.58 |
The semaglutide result reproduced at 0.56. The tirzepatide estimate was 0.60 and did not reach significance.
Why the two disagree
Two more limits apply. People who start a newer drug see doctors more often, which changes how fast a diagnosis is recorded. And that analysis appeared in Biology Methods & Protocols, which is not a dementia journal.
Matching on 30 variables does not fix this. It balances what was measured. It cannot balance the reason a prescriber chose one drug over another.
What is settled, and what is not
Settled: liraglutide was safe and tolerated over 52 weeks in this population. Daily injections in people with cognitive impairment had not been tested before.
Not settled: whether any GLP-1 drug changes the course of Alzheimer’s. Larger randomized trials of oral semaglutide have reported, and it did not slow early Alzheimer’s. A record-based hazard ratio remains a hypothesis.
One narrower question has its own page. The oral tablet was tested in cognitive impairment. It reached a similarly cautious place — see what the oral tablet did for cognitive impairment. The effect-size reading this site applies throughout is in sixteen percent lower, six hundredths of a point.