What actually happened with GLP-1 drugs and Alzheimer's?
In November 2025 the largest Alzheimer's trial ever to include a GLP-1 reported its primary endpoint. EVOKE, run by Novo Nordisk, tested oral semaglutide in roughly 4,000 people across 40 countries over two years. The result on cognitive decline: no difference versus placebo. The company discontinued the open-label extension because there was nothing worth following.
Four weeks later, in December 2025, a much smaller trial published in Nature Medicine reported the opposite. The ELAD trial tested liraglutide in 204 people with mild-to-moderate Alzheimer's over one year. It showed about 50% less brain volume loss and 18% slower cognitive decline compared to placebo.
Same drug class. One month apart. Headlines called the results a contradiction. They are not. The two drugs hit the brain differently, the patients were at different disease stages, and the trials measured success with different tools. Walk through the mechanism and the apparent conflict dissolves.
Why does blood sugar matter so much for Alzheimer's?
Alzheimer's-flavored dementia carries a strong metabolic signature. Some researchers call it type 3 diabetes because the neurons themselves become insulin resistant. High blood sugar drives oxidative damage. You get advanced glycation end products, the rusting of tissue in the presence of sugars. The same glycation and oxidative-stress machinery shows up across dementia subtypes. In Lewy body dementia, the edges of the Lewy bodies get glycated and tear through tissue. In frontotemporal dementia, the picture leans more vascular, a wear-and-tear and blood-pressure phenomenon in the frontal and temporal regions where productive language lives. This is the type Bruce Willis went public with as his verbal resources declined.
A quick vocabulary note. Dementia is a symptom, not a disease. It describes the failure to form and retrieve memory, and it usually arrives alongside another dysregulation: agitation, anger, or dysregulated fear. Underneath the different flavors sits a common metabolic driver, the glycation and oxidative stress that degrades tissue over roughly 30 years. You do not get Alzheimer's all at once. You get it across three decades.
Early-onset Alzheimer's in your 40s or 50s is rare and almost always genetic. The presenilin gene runs in specific families, and you know it because everyone in the family gets it in their 50s. Without that marker, the relevant question is whether you are on a 30-year metabolic decline, not whether you have it now. If you are concerned, a metabolic screen makes sense. Programs like Apollo Health's PreCode and ReCode, built on Dr. Dale Bredesen's work, map the 30-plus factors that act as fuel for the fire across the dementia pathways. Functional medicine clinicians run similar panels, including APOE status. For the broader picture of when these changes begin, see the critical aging window.
Why did the two GLP-1 trials get opposite results?
The drugs are not interchangeable, and three differences explain most of the gap.
Route of administration. EVOKE used oral semaglutide. ELAD used daily subcutaneous liraglutide. An oral drug has to survive first-pass metabolism in the liver, which knocks the concentration down hard before it reaches systemic circulation. A subcutaneous injection bypasses that gatekeeper, gets deposited, and absorbs gradually. My hunch is that the route of administration matters more here than the difference between the two molecules.
Pulsatile versus constant dosing. The oral pill peaks and drops off, peaks and drops off. The subcutaneous depot delivers a more constant, steady signal. Those are different exposure profiles for tissue that responds to sustained signaling.
Blood-brain barrier penetrance. Semaglutide has higher relative CNS penetrance than liraglutide. The interesting wrinkle: that advantage did not rescue EVOKE. The likely reason is that the lower bioavailability of the oral route, combined with pulsatile dosing, swamped the penetrance advantage. Better at crossing the barrier does not help much if far less of the drug ever makes it into the body.
Disease stage and endpoint. EVOKE enrolled mild cognitive impairment and mild Alzheimer's and measured the CDR-SB, the clinical dementia rating sum of boxes, the standard regulatory endpoint that captures cognitive and functional decline. ELAD enrolled mild-to-moderate, later-stage disease with potentially more inflammation and more rampant blood-sugar dysregulation, and its primary endpoint was an FDG-PET scan measuring glucose utilization in the temporal lobes, plus the ADAS-Cog 11. The metabolic endpoint may be more sensitive in more advanced disease, where there is more metabolic dysfunction to spare.
EVOKE was a well-run, large, rigorous trial. A clean negative result in a good trial is real information. ELAD was smaller, 5% of the size, but the volume preservation and the metabolic sparing in the temporal lobes were dramatic effects. Read together, the two trials look less like a contradiction and more like hypothesis generation: one route and dosing scheme produced a strong individual-level signal, another failed to reproduce it at 20 times the sample for twice as long.
Do GLP-1 drugs protect the brain beyond blood sugar?
They appear to. There are GLP-1 receptors across many brain regions, so the effect on cognition is not purely a downstream consequence of blood sugar. The mechanistic bridge runs through insulin signaling. GLP-1 drugs improve insulin sensitivity, which in the brain means better glucose uptake, better clearance, and less oxidative stress. Even when a particular GLP-1 cannot cross the blood-brain barrier well enough to directly activate hippocampal neurons, it may still improve brain function by reducing peripheral insulin resistance.
The real-world data backs a signal. A 2025 meta-analysis in Diabetes by Chang pooled 18 observational studies covering more than three million people and found GLP-1 receptor agonist use associated with reduced dementia risk versus other anti-diabetic medications, with hazard ratios well below 1 for all-cause dementia and for Alzheimer's specifically. One large analysis reported roughly a 58% reduction in Alzheimer's risk. The comparison is the telling part: both groups had reasonably controlled blood sugar, often better controlled in the non-GLP-1 medication arms, yet the GLP-1 cohorts still did better. The benefit shows up where blood sugar alone does not account for it.
Read these numbers with appropriate caution. People who get GLP-1 prescriptions tend to be healthier and more engaged in their care. There is time-window bias in the period before dementia develops. And dementia diagnostics vary widely across regions, especially for mild cases. The signal is strong, consistent, and large, but it is observational.
What about Parkinson's and GLP-1 drugs?
The Parkinson's story rhymes with the Alzheimer's one. A 2025 Lancet trial of exenatide was negative for motor progression, despite an earlier promising phase 2 and good blood-brain barrier penetrance. A 2024 NEJM trial of lixisenatide, another high-penetrance compound, was positive, with slowed motor symptom progression. Same drug class, different molecules, drug-by-drug heterogeneity in disease impact.
Parkinson's itself is a category more than a single disease clinically. Multiple system atrophy, Parkinsonian non-specific diagnoses, Parkinson's proper, and Lewy body dementia all involve the basal ganglia and the dopaminergic system but differ in mechanism. And like Alzheimer's, it builds slowly. The pars compacta of the substantia nigra can lose up to 75 to 80% of its dopamine generation before you see tremor, stiffness, or rigidity.
That fact also undermines the chemical-imbalance model of mental illness. There is no absolute correct level of a neurotransmitter. The system retunes itself across a wide range by changing receptor density, receptor sensitivity, and the number of cells in the synapse. Do not over-read neurotransmitter level tests. Even if you could measure them past metabolic degradation downstream, we do not know what a given number means.
Is your memory slip early Alzheimer's, or just bad sleep?
I run direct brain map reviews with about 25 people a week, and at least half raise a memory concern. Many are 30, 40, or 50 and quietly terrified they are watching early Alzheimer's. Most of the time they are watching the EEG signature of chronic underrest.
Here is what shows up on a quantitative EEG when you push a tired brain. The alpha rhythm, the idling speed of the system, slows and spreads, and the left hemisphere stops synchronizing as tightly. Meanwhile delta, the background rest-and-repair band, climbs in both amplitude and speed as the cleaning crew rushes around full-time. For most rested people, delta sits near average. When you are sleep-deprived, ill, injured, or your metabolism is tanking from something like apnea or heavy drinking, delta pushes into the foreground. That feels foggy and tired.
The common experience that pairs with slipped alpha and climbing delta is delayed recall for words and names. The tip-of-the-tongue moment, the "wait, I know this." That is usually a timing and synchronization problem, not a memory storage problem. The handoff between receptive language areas and productive language areas is sensitive to delays. When you are tired, you may stutter or lose a word and then recover it minutes later or the next day. The information was stored fine. The loading and handoff failed.
The fix is sleep, specifically deep, dreamless, slow-wave sleep. That is the number that matters on a sleep tracker. Aim to grow deep sleep toward two to two and a half hours, or around 25% of total sleep. String a few of those nights together and the alpha waves snap back into relative position, often speeding up, while delta backs off in power and speed. For the broader protocol, see biohacking sleep and biohacking brain fog.
What can neurofeedback and HEG do for the aging brain?
Two readers asked whether neurofeedback could slow cognitive decline, and one asked specifically about HEG.
I use passive infrared HEG to train cerebral blood flow, what I think of as vasomotion. The vasculature in the brain has pacemaker neurons that send waves of energy washing across the cortex, supplying energy where it is needed and responding somewhat to demand. Healthy aging includes maintaining that vasomotion. Accelerated aging, the low-and-slow EEG phenotype with low amplitude and excess slow energy, tends to come with reduced vasomotion.
Passive infrared HEG trains the middle of the forehead, a broad rather than focused target. I find it a blunt but effective tuner for migraines, post-concussion fatigue, post-COVID fog, dysautonomia, and aspects of autism. In migraine, many common flavors involve spreading cortical depression, a mismatch between vasomotion and metabolic demand that cramps a region metabolically and produces the sparkly lights and odd neurological symptoms before the pain. Training more stable, reliable blood flow reduces severity first and then incidence. I have also watched HEG and frontal training shift social function in higher-functioning autistic clients, including timing, prosody, eye contact, and sense of humor. This is clinical observation, not RCT-grade evidence, so weigh it accordingly.
On the EEG side, vascular dementia and the temporal-lobe metabolic loss in Alzheimer's both leave a pathology signature you can see. Once enough tissue is lost, you see excess delta or theta over the affected regions, no radioactivity required. For training, SMR and beta protocols have a long track record with sleep stabilization, attention, and self-regulation. See SMR neurofeedback and the neurofeedback research overview for the evidence base. One clinical truism worth holding onto: outside of major tissue damage, the worse the dysregulation, the faster it tends to change. That holds for impulsivity, anxiety, and disrupted sleep.
A note on GLP-1 tolerance and motivation
A few readers report that GLP-1 appetite suppression hits hardest the first time and weakens on a second course, even months later, without resetting. The pattern points toward the serotonin system rather than simple receptor downregulation. Some readers also describe blunted motivation, a flatter hedonic tone, on these drugs, and a lingering flatness after stopping. That profile resembles what some people report after SSRIs or finasteride, agents that act on the serotonin 1A receptor. If you feel blunted in that way after withdrawing, I would expect a lift over time and would pursue interventions that support that receptor system.
Should you take a GLP-1 for brain health right now?
For me, at 55, with Alzheimer's, Parkinson's, and diabetes in my family, the answer is not yet. The reasons I would consider these drugs, metabolic health, appetite control, behavioral control, brain protection, are things I can largely reach through diet, sleep, neurofeedback, and meditation. I would rather make the lightest touch that drives the system, and an injectable is a heavy touch. I also have APOE2/3 status, which is neuroprotective, and a handful of other factors that make me likely to age well, so I am not pushing hard in that direction.
If you are concerned about your trajectory, start here. Get a metabolic screen and address the out-of-range factors. Protect deep sleep and stop eating in the couple of hours before bed, because melatonin shuts off insulin at the end of the day and a late snack leaves blood sugar elevated, suppressing the single nightly growth hormone pulse you get after 30 or 40. For the metabolic-signaling angle on diet, see strategic fasting. If you want to understand and train your own brain, a QEEG brain map gives you a real picture of your alpha speed, delta load, and regulatory features, and remote neurofeedback can act on them.
If you are weighing a GLP-1 specifically, look up the individual drug, not the class. Check its route of administration, its blood-brain barrier penetrance, its dosing profile, and which dementia subtype the trials actually measured. More studies are landing every few months, and the drugs are not one category. Pick the lightest effective intervention you can sustain, measure what it does, and build from there.