A failed trial just handed APOE4 carriers something more useful than a miracle

Hi Phoenix friend,

The drug the internet crowned "Ozempic for Alzheimer's" ran its biggest test. 3,808 people. Two years. And the semaglutide group declined just as fast as the placebo group.

Here's why that's actually useful news for APOE4 carriers.

This is the full written synthesis of the EVOKE/EVOKE+ semaglutide Alzheimer's trial results. Prefer to watch? Here's the 28-minute deep dive on YouTube (studies on screen, full stories): https://youtu.be/eUOIIPXTUIM

Introduction

My stomach dropped when I saw the headlines. I carry two copies of APOE4, and like a lot of carriers, I'd been watching this trial. Not with full hope, but with genuine interest. An already-approved drug, already on tens of millions of prescriptions, that might also quietly protect the brain.

Then the Lancet published. Both trials discontinued for negative clinical outcome.

Here's what I want to walk you through: what the trial actually found, why a drug can fix your biology and still do nothing for your memory, and the one distinction almost no coverage is making. "The treatment trial failed" and "GLP-1s do nothing for your brain" are two completely different sentences. Once you can tell them apart, this result goes from depressing to genuinely useful.

I'm Kevin. I'm a Doctor of Pharmacy, and before Phoenix I spent years reading clinical trial data for a living. So I skipped the headlines and went to the actual numbers.

What the EVOKE Trials Actually Found

EVOKE and EVOKE+ were two phase 3 randomized controlled trials testing oral semaglutide 14 mg against placebo in people with early Alzheimer's disease [1]. The primary measure was CDR-SB, a 0-to-18 clinician scale covering memory, judgment, daily function, and personal care. Higher is worse [2].

The trials were built with 95% statistical power to catch even a modest win [2].

The result: a flat tie. Estimated difference at week 104 was minus 0.08, confidence interval straight through zero, p = 0.57 [1]. The Lancet's conclusion: "not efficacious in slowing clinical progression" [1]. Both trials discontinued for negative clinical outcome [1].

One number every carrier should pause on: roughly 60% of participants were APOE4 carriers. 61.7% in EVOKE, 57.8% in EVOKE+ [3]. This was the closest thing we've ever had to testing this drug in a room full of carriers. The room didn't get better.

Worth stating clearly: no genotype-stratified efficacy results were published. Nobody can tell you whether carriers responded differently from non-carriers. Anyone claiming otherwise is making it up.

💡 KEY INSIGHT: When a company discontinues its own well-powered trials for "negative clinical outcome," that's a verdict, not a setback.

Biology Moved. Cognition Didn't.

Here's the part that genuinely messes with your head. The drug did something. Just not for the people.

In a spinal-fluid substudy, several Alzheimer's biomarkers dropped on semaglutide: p-tau181, p-tau217, total tau, and the neuroinflammation marker YKL-40, all down roughly 10% or less [4]. Systemic inflammation (hs-CRP) dropped about 30% [4]. But those spinal-fluid changes weren't mirrored in blood, and they produced zero cognitive benefit [4].

The researchers put it plainly: "statistically significant reductions in p-tau or YKL-40 do not constitute evidence of meaningful clinical benefit" [4]. We saw this exact disconnect in the anti-amyloid era. It applies here too.

⚠️ CAVEAT: The modest drops in p-tau and YKL-40 are real findings from the CSF substudy. They do not constitute partial efficacy. Moving a number on a lab report is not the same as saving a memory.

Why It Failed: Blood-Brain Barrier and the Stage Problem

Two explanations carry most of the weight.

The drug barely gets in. Semaglutide was engineered to stay in the bloodstream for a long time. Convenient as a metabolic drug. But that same design means it does not cross the blood-brain barrier readily [2] [7].

Think of two tiers of effect. Top tier, proven in humans: lower glucose, lower weight, lower systemic inflammation. Solid, real, well-documented. Bottom tier, the brain-cell rescue effects (clearing tau, repairing neuronal energy, calming microglia): almost all of that evidence comes from rodent studies, at doses and timing that may not translate to humans [4] [5]. The drug barely crosses the front door of the brain, so the beautiful brain-rescue theory was never on solid human footing.

There's a clean idea underneath all this. Some researchers now describe Alzheimer's as partly brain insulin resistance, sometimes called "type 3 diabetes" [5]. If your brain cells have gone deaf to insulin, a drug that fixes insulin signaling should help. That's the dream. The problem is most of semaglutide's power stays on the wrong side of the door.

The timing may have been wrong, too. The people in EVOKE already had established amyloid and tau pathology. The argument is that a metabolic drug might only help during the preclinical window, before neurons are lost. Once the damage is done, there's less to preserve [5].

⚠️ CAVEAT: That prevention-window framing is explicitly described as speculative by the authors [4]. Competing explanations carry equal weight. Hold it as a hypothesis, not a settled answer.

"Treatment Failed" Is Not a "Prevention Verdict"

This is the distinction almost no coverage is making. And getting it wrong costs you either way.

EVOKE asked one specific question: can semaglutide treat people who already have early Alzheimer's? The answer came back no. Clear, well-powered, two-year verdict.

It did not ask: can semaglutide prevent cognitive decline in cognitively normal people at high risk? That trial has not been run. No randomized controlled trial has yet shown that semaglutide prevents cognitive decline or dementia in anyone [4].

File it this way:

• Treatment of existing disease: failed, full stop.

• Prevention in healthy APOE4 carriers: still an open question.

Where does the prevention optimism come from? Real-world data, and there's a lot of it. In people with type 2 diabetes (observational, so associative not causal): a large US target-trial emulation found a dementia hazard ratio of 0.67 [8]; a multinational cohort of over 200,000 people found Alzheimer's onset roughly 23% lower [9]; a 7-year retrospective cohort put dementia risk at HR 0.63 [11]. Pooling three cardiovascular outcome trials in T2D patients showed 53% fewer dementia diagnoses, but that was a post hoc look (dementia was never what those trials were built to measure) with a wide confidence interval, 0.25 to 0.86 [10] [2].

The signal is genuinely big. The catch the hype always skips: almost all of it is observational, all of it is in people with type 2 diabetes, not cognitively normal APOE4 carriers. The biggest limitation, named by the researchers themselves, is confounding by indication and healthy-user bias [10]. People who stay on these drugs long-term are different from people who don't. That gap in dementia rates might be the drug. Or it might be who ends up on the drug. Observational data cannot tell you which.

💡 KEY INSIGHT: The prevention question is genuinely open. The treatment question got a clear answer. Keeping those two in separate boxes is the whole skill.

Four Questions to Read Any Drug Headline in 30 Seconds

This is the one takeaway that makes every future trial result useful instead of terrifying.

1. Treatment or prevention? Who did they test: sick people or healthy people?

2. Primary or secondary endpoint? Was the win the pre-registered goal, or something they went fishing for after the fact?

3. Randomized or observational? A coin-flip assignment, or a confounded comparison?

4. Biomarker or outcome? Did a lab number move, or did a person actually get better?

Run EVOKE through all four. Treatment population. Primary endpoint. Randomized. Clinical outcome. That is the strongest kind of evidence on every question, and the drug still came up empty. That's why it counts as a verdict, not noise. The headline tells you what happened. The endpoint tells you what it means.

One more flag worth knowing: there's a single-author re-analysis arguing the trial "actually worked" at a later timepoint, week 130, on a secondary functional scale [7]. Before that gets your hopes up: it uses converted summary statistics, not patient-level data, on a secondary endpoint after the primary already failed [7]. And the author is affiliated with a company that develops next-generation BBB-penetrant GLP-1 drugs, meaning there's a direct commercial interest in the conclusion that semaglutide was onto something [7]. Watch this hypothesis. Do not call it a rebuttal of the failure.

The Metabolic Levers Are Already Yours

Here's the reframe that turns this from depressing to useful.

The proven part of semaglutide is metabolic: inflammation down, glucose down, weight down [4]. Safety in older adults held up in a pooled analysis of over 3,500 people aged 65 and above [6]. Main downsides: gastrointestinal side effects were the most common adverse events [6] and roughly 3.8% weight loss over a year [6], which in a frail or low-BMI older person can actually be a problem, not a benefit.

As a metabolic drug, with your doctor's agreement, it remains a reasonable option on several levels. As a treatment for Alzheimer's in someone who already has it: the data said no, and it's not formally indicated for that use [5]. Two different jobs. Be clear which one you're hiring it for.

Here's what I do, as someone who carries two copies and is not on a GLP-1. I go straight for the markers the drug was only borrowing. My ApoB is down 39%, on ezetimibe 10 mg stacked with diet and a lot of exercise. My VO2max is up 31%. My HbA1c is down 8%. My body fat is down 43%. None of that needed a GLP-1. (It's one person, not a trial.)

Every one of those is the upstream metabolic lever the prevention hypothesis is actually pointing at. The brain-insulin-resistance angle, the "type 3 diabetes" framing, all of it lives in the metabolic layer you can already reach [5].

You don't need the drug to win the metabolic game. You need the markers it moves. And every one of those is yours to move.

A failed trial didn't take anything from you. It handed you a cleaner map.

Key Takeaways

💡 Quick-Start Protocol (This Week):

1. Separate the two questions every time. Treatment of existing disease (EVOKE: failed). Prevention in healthy people (still untested by RCT). Never let anyone glue them together.

2. Run the four questions on any drug headline: treatment or prevention, primary or secondary, randomized or observational, biomarker or outcome.

3. Chase the markers, not the drug. hs-CRP, HbA1c, glucose, weight, VO2max. Those levers exist with or without a prescription.

4. If you're considering a GLP-1, frame it with your doctor as a metabolic decision, not an Alzheimer's decision. That's where the honest evidence sits.

5. Keep watching the prevention question. The review behind this result argues the next trials should test high-risk people who are still cognitively well, and it names APOE4 carriers [4]. When those trials open, they will show up in the Clinical Trial Engine below.

Read the Next Trial in Plain English

When I filmed the video, I said we were building a Clinical Trials section. It's live now, and it's free: the APOE4 Clinical Trial Engine. Every APOE4 trial from ClinicalTrials.gov, translated into plain English, with what it means for carriers. Refreshed daily, no signup.

So when the next big readout lands, and it will, you don't have to sort it out alone at midnight.

Open the Clinical Trial Engine

And the one thing to do first from this story: get two numbers drawn, hs-CRP and HbA1c. You don't need a prescription for that blood draw. You need a Tuesday morning.

Sources

Match each numbered citation in the article to the same number below. Select View source to open the original paper or trial record.

1. Cummings JL, Atri A, Sano M, et al. (2026). Efficacy and safety of oral semaglutide 14 mg (flexible dose) in early-stage symptomatic Alzheimer's disease (evoke and evoke+): two phase 3, randomised, placebo-controlled trials. Lancet. View source

2. Cummings JL, Atri A, Feldman HH, et al. (2025). evoke and evoke+: design of two large-scale, double-blind, placebo-controlled, phase 3 studies evaluating efficacy, safety, and tolerability of semaglutide in early-stage symptomatic Alzheimer's disease. Alzheimer's Research & Therapy. View source

3. Scheltens P, Atri A, Feldman HH, et al. (2026). Baseline characteristics from evoke and evoke+: Two phase 3 randomized placebo-controlled trials of semaglutide in participants with early-stage symptomatic Alzheimer's disease. Alzheimer's & Dementia (New York, N.Y.). View source

4. Alhowail AH, Al Mouslem AK, Almatrafi MA, Aldubayan MA (2026). Semaglutide in cognitive dysfunction: neuroprotective potential, clinical trial limitations, and a prevention-focused framework. Frontiers in Aging Neuroscience. View source

5. Gandhi A, Parhizgar A (2025). GLP-1 receptor agonists in Alzheimer's and Parkinson's disease: endocrine pathways, clinical evidence, and future directions. Frontiers in Endocrinology. View source

6. Sabbagh M, Boschini C, Cohen S, et al. (2025). Safety considerations of semaglutide in the potential treatment of Alzheimer's disease: A pooled analysis of semaglutide in adults aged >= 65 years. Alzheimer's & Dementia (New York, N.Y.). View source

7. Holscher C (2026). Semaglutide showed limited improvements in patients with Alzheimer's disease: Revisiting the evoke and evoke+ clinical trials. Journal of Alzheimer's Disease. View source Important context: This is a single-author re-analysis using converted summary statistics, not patient-level data. The author is affiliated with Kariya Pharmaceuticals, which develops BBB-penetrant GLP-1 drugs. A commercial conflict of interest applies to the "late signal" claim.

8. Tang H, Donahoo WT, DeKosky ST, et al. (2025). GLP-1RA and SGLT2i Medications for Type 2 Diabetes and Alzheimer Disease and Related Dementias. JAMA Neurology. View source

9. Schechter M, Fishkin A, Mosenzon O, et al. (2025). Neurodegeneration onset with glucagon-like peptide-1 receptor agonists in people with type 2 diabetes: a real-world multinational cohort study. Cardiovascular Diabetology. View source

10. Hui EK, Mukadam N, Kohl G, Livingston G (2025). Effect of diabetes medications on the risk of developing dementia, mild cognitive impairment, or cognitive decline: A systematic review and meta-analysis. Journal of Alzheimer's Disease. View source

11. Lin HT, Tsai YF, Liao PL, Wei JC (2025). Neurodegeneration and Stroke After Semaglutide and Tirzepatide in Patients With Diabetes and Obesity. JAMA Network Open. View source