GLP-1 agonists have accumulated a long list of possible off-label uses, and Alzheimer's disease is now on it. The mechanistic case is coherent enough to take seriously. Type 2 diabetes and insulin resistance travel with cognitive decline. GLP-1 receptors sit in the hippocampus and other regions involved in memory. Activate them in a mouse and the brain benefits.

From there the evidence separates into three tiers that point in different directions. The preclinical data are strong. The observational data are large and favorable. The randomized trials are close to null. A patient asking whether their semaglutide protects their memory is asking a question the field cannot yet answer.

News-Medical published a careful review of where things stand. What follows works through the three tiers and where each one breaks down.

In animal models the effect is real and mechanistically legible. Liraglutide reduces amyloid-β accumulation and tau-related changes. It cuts neuroinflammation and oxidative stress, preserves glucose metabolism, and props up mitochondrial function. In rats poisoned with okadaic acid, liraglutide lowers BACE1 and with it amyloid production. Knock the GLP-1 receptor out of a mouse and synaptic plasticity and memory formation degrade, which is the kind of loss-of-function result that makes a mechanism feel load-bearing rather than incidental.

The strongest preclinical result is in monkeys. Pretreatment with liraglutide before exposure to toxic amyloid reduced abnormal tau and synaptic loss. That is prevention in a primate, near the ceiling of what preclinical Alzheimer's evidence can show.

The observational estimates are large. A post hoc analysis of the LEADER, SUSTAIN 6, and PIONEER 6 cardiovascular trials suggested GLP-1 agonists might reduce dementia risk in diabetics by up to 53 percent. Real-world cohorts in the US and Denmark reported reductions of 42 and 64 percent. The REWIND analysis of dulaglutide found less cognitive decline. Taken at face value, these are striking numbers.

The same review supplies the counterweight. In a UK cohort of adults over 60 with type 2 diabetes, starting a GLP-1 agonist showed no reduction in all-cause dementia compared with a DPP-4 inhibitor in the intention-to-treat analysis (HR 0.95, 95% CI 0.87–1.04). A benefit appeared only when the analysis was restricted to continuous users (HR 0.79, 95% CI 0.64–0.97). Restricting to continuous users introduces the healthy user effect: people who stay on a medication for years differ systematically from those who stop, in ways correlated with better health outcomes independent of the drug.

The randomized evidence is weak. A 26-week liraglutide trial in Alzheimer's patients preserved cerebral glucose metabolism relative to placebo but did not alter amyloid or cognition. A 12-month trial showed better preservation of cortical volume and some cognitive measures. These are modest, exploratory signals.

The 2026 meta-analysis is the central number. Across 14 randomized trials and 1,260 non-diabetic participants with Alzheimer's, mild cognitive impairment, or Parkinson's, global cognition improved by a standardized mean difference of 0.14 (95% CI 0.01–0.27). The authors estimated roughly a 1 percent probability that this reached the threshold for a clinically important benefit. Verbal fluency was worse in the pooled analysis (SMD −0.43), and gastrointestinal adverse events were more common. Because several neurodegenerative populations were pooled, the estimates are not specific to Alzheimer's.

Across the tiers, the effect size shrinks as the evidence gets more rigorous, and the randomized trials are the ones that determine practice. EVOKE and EVOKE+ are testing oral semaglutide in early symptomatic Alzheimer's with the biomarkers and follow-up designed to settle the question. Until they report, the drugs remain approved for diabetes and obesity, and use for Alzheimer's neuroprotection is off-label and investigational.

The program is proceeding as it should. A coherent mechanism supported by primate prevention data is now being tested in a properly powered phase 3 trial. The open risk is interpretive: treating the preclinical and observational signals as settled before the trials report.