GLP-1 Receptor Agonists and Stroke: The Evidence Is Compelling — But Is It Actually About Stroke?

Here's what we tell ourselves: GLP-1 receptor agonists reduce stroke. The meta-analyses are clear. The guidelines give them a Class I, Level A recommendation. The mechanisms make sense — anti-inflammatory, anti-atherosclerotic, endothelial protection, weight loss, blood pressure reduction. The science is solid.

Here's what the trials actually show: in most individual studies, the stroke component didn't reach statistical significance. The largest trial in non-diabetic patients showed no stroke benefit whatsoever. And there has never been — not once — a trial designed with stroke as the primary endpoint.

The diagnosis drives the treatment — so let's get the diagnosis right. Are GLP-1 RAs stroke prevention drugs? Or are they cardiovascular drugs that happen to reduce MACE, with stroke carried along as a secondary benefit that may or may not hold up when you look closely?

Context determines management. The same drug, the same mechanism, two different patient populations — two completely different results. This isn't about cherry-picking data. It's about understanding what we're actually prescribing and why.

The Case For: This Is Real Stroke Prevention

Let's start with the strongest argument. The meta-analysis of 11 cardiovascular outcome trials involving 82,140 participants showed that GLP-1 receptor agonists reduced stroke by 16% (HR 0.84, 95% CI 0.76-0.93, p<0.001). Nonfatal stroke was reduced by 13%. This is not a trivial signal. This is the kind of consistent effect size that changes practice.

More important, look at the individual trials that drove that signal:

SUSTAIN-6 was the trial that made stroke neurologists pay attention. Published in the New England Journal of Medicine in 2016, it randomized 3,297 patients with type 2 diabetes to subcutaneous semaglutide versus placebo for 104 weeks. The primary endpoint was time to first MACE event — but the stroke component stood out. Nonfatal stroke occurred in 27 patients (1.6%) on semaglutide versus 44 (2.7%) on placebo. Hazard ratio 0.61 (95% CI 0.38-0.99, p=0.04). A 39% reduction. Statistically significant. Striking.

This wasn't a subset analysis. This was a prespecified component of the primary endpoint. And it reached significance in a relatively small trial with only 104 weeks of follow-up. That shouldn't happen unless the effect is real and substantial.

REWIND went further. Published in The Lancet in 2019, this trial enrolled 9,901 patients with type 2 diabetes — but here's what made it different: only 31% had established cardiovascular disease. This was a broader, more representative population than most CVOTs. Median follow-up was 5.4 years. Dulaglutide reduced total stroke events from 205 (4.1%) on placebo to 158 (3.2%) on treatment. Hazard ratio 0.76 (95% CI 0.62-0.94, p=0.010).

And the investigators went deeper. They looked at ischemic stroke specifically: HR 0.75 (95% CI 0.59-0.94, p=0.012). No effect on hemorrhagic stroke (HR 1.05), which makes mechanistic sense if the benefit is anti-atherosclerotic. Even more impressive: disabling stroke was reduced — HR 0.74 (95% CI 0.56-0.99, p=0.042). This wasn't just preventing minor events. This was preventing strokes that change lives.

The mechanisms align. GLP-1 receptors are expressed on endothelial cells, vascular smooth muscle, and in atherosclerotic plaques. Preclinical models show reduced infarct volume, decreased inflammation, and neuroprotection. In humans, GLP-1 RAs reduce systemic inflammation, improve endothelial function, stabilize atherosclerotic plaques, lower blood pressure, promote weight loss, and improve glucose control. Every one of those is a stroke risk factor. Why wouldn't reducing all of them together translate to stroke prevention?

The 2024 AHA/ASA Primary Stroke Prevention Guidelines gave GLP-1 RAs a Class I, Level of Evidence A recommendation for patients with type 2 diabetes who have high cardiovascular risk or established cardiovascular disease and HbA1c ≥7%. That's the highest level of evidence-based recommendation the guidelines can give. It's not speculative. It's not aspirational. According to the best available evidence synthesis, this is what we should be doing.

This is the case for. It's strong. It's not wrong.

But it's not complete.

The Case Against: Look at the Fine Print

Now let's look at what happened when the trials didn't cooperate.

LEADER, published in the New England Journal of Medicine in 2016, enrolled 9,340 patients with type 2 diabetes for a median of 3.8 years. Liraglutide reduced the primary composite MACE endpoint with a hazard ratio of 0.87 (p=0.01). That's significant. The drug works for cardiovascular protection.

But look at nonfatal stroke specifically: HR 0.89 (95% CI 0.72-1.11, p=0.30). Not significant. The confidence interval crosses 1.0. The benefit in LEADER was driven by cardiovascular death (HR 0.78), not by stroke.

Maybe liraglutide is just a weaker molecule. Fair point. Let's move to the biggest trial we have.

SELECT, published in the New England Journal of Medicine in 2023, was a landmark study. It enrolled 17,604 overweight or obese patients without diabetes — the largest cardiovascular outcomes trial for GLP-1 RAs to date. Semaglutide 2.4 mg reduced the primary MACE endpoint with a hazard ratio of 0.80 (p<0.001). This was the trial that earned semaglutide FDA approval for cardiovascular risk reduction regardless of diabetes status. A major regulatory win.

Nonfatal stroke: HR 0.93 (95% CI 0.74-1.15). Completely null. The benefit in SELECT was driven by nonfatal myocardial infarction (HR 0.72). If you're a stroke neurologist prescribing semaglutide for stroke prevention based on SELECT, the trial doesn't support you.

SOUL, published in the New England Journal of Medicine in 2025, tested oral semaglutide in 9,650 patients with type 2 diabetes. MACE was reduced with a hazard ratio of 0.86. Stroke reduction was approximately 12% — again, not individually significant. The benefit was driven by a 26% reduction in myocardial infarction.

Finding an abnormality isn't the same as finding the cause. We found a signal in meta-analysis. But when you look at individual trials — especially the largest and most recent ones — the stroke component keeps disappearing.

The SUSTAIN-6 Problem

SUSTAIN-6 is the trial everyone cites for stroke benefit. But let's be honest about what it was: a safety trial with 3,297 patients and 104 weeks of follow-up. There were 71 total stroke events — 27 on semaglutide, 44 on placebo. The hazard ratio was 0.61 with a 95% confidence interval of 0.38 to 0.99. It barely crossed the significance threshold.

Is a finding based on 71 total events, in a trial not powered for stroke, enough to change practice? SUSTAIN-6 was designed to assess cardiovascular safety — it needed to exclude a hazard ratio of 1.8 for MACE. It succeeded. But the stroke finding was exploratory. The confidence intervals are wide. And no subsequent trial with semaglutide has replicated that magnitude of stroke benefit.

The REWIND Exception

REWIND had the strongest and most consistent stroke signal of any GLP-1 RA trial. But even REWIND was powered for composite MACE, not stroke alone. The stroke analysis was prespecified, yes — but it was still a secondary endpoint. The trial was designed to answer a cardiovascular question, not a cerebrovascular one.

And here's the fundamental problem: no trial has ever been designed with stroke as the primary endpoint. Not one. We are building stroke prevention recommendations on secondary endpoints, post-hoc analyses, and meta-analyses pooling trials that weren't asking the stroke question.

The workup doesn't end when you find something. It ends when the pieces fit. Right now, the pieces don't fit.

The Generalizability Problem

Let's assume for a moment that the meta-analytic stroke benefit is real. Who does it apply to?

Most cardiovascular outcome trials enrolled patients with type 2 diabetes. SUSTAIN-6, LEADER, REWIND, SOUL — all diabetes trials. The question they asked was: in patients with type 2 diabetes and cardiovascular risk, does this drug reduce cardiovascular events?

Does the benefit extend to non-diabetic stroke patients? SELECT is the only major trial in non-diabetic patients — and it showed no stroke benefit.

REWIND had a broader population than most trials — only 31% had established cardiovascular disease at enrollment, which is more representative of primary prevention populations. But it's still a diabetes trial. The patients had HbA1c ≥7%, history of cardiovascular disease or cardiovascular risk factors, and a median age in the mid-60s. This is a metabolic syndrome population.

What about the 45-year-old with cryptogenic stroke and a PFO? What about the 78-year-old with small vessel disease? What about the young woman with moyamoya? The man with cervical artery dissection? Context determines management. The same finding, two different patients, two different answers.

The 2024 guidelines recommend GLP-1 RAs for primary stroke prevention in patients with diabetes. But most stroke neurologists see patients for secondary prevention — patients who've already had a stroke or TIA. The evidence gap here is significant. We have no trial data on whether starting a GLP-1 RA after an acute ischemic stroke reduces recurrent stroke risk.

Are we preventing stroke — or are we preventing the metabolic and inflammatory milieu that leads to atherosclerotic events generally? If it's the latter, then stroke reduction is a byproduct of better cardiometabolic health, not an independent cerebrovascular effect. And that changes who should be prescribing these drugs.

The Elephant in the Room: Should Neurologists Prescribe These?

GLP-1 receptor agonists are not simple drugs. They require dose titration to minimize gastrointestinal side effects. They carry risks — pancreatitis, gallbladder disease, potential thyroid concerns based on preclinical data. They need metabolic monitoring. Most neurologists don't manage diabetes medications day-to-day.

Are we expanding our scope appropriately — or are we creating fragmented care?

The counterargument is straightforward: neurologists already prescribe anticoagulants with bleeding risk that requires monitoring. We prescribe antihypertensives. We prescribe statins, often at high doses. If GLP-1 RAs have Class I evidence for stroke risk reduction in the right population, why shouldn't stroke neurologists own this?

The practical barriers are real. These drugs cost $800 to $1,000 per month without insurance. Supply shortages have been ongoing since 2022. Insurance denials and prior authorization requirements create substantial administrative burden. Even when approved, patient adherence is limited by cost and side effects.

The SELECT trial earned semaglutide 2.4 mg FDA approval in 2024 for cardiovascular risk reduction in overweight or obese patients with established cardiovascular disease — regardless of diabetes status. This expands the eligible population enormously. But again, the stroke signal in SELECT was null. If we're prescribing for cardiovascular protection broadly, that's one conversation. If we're prescribing specifically for stroke prevention, the evidence is much weaker.

What We Still Don't Know

The most important question: Will anyone ever do a stroke-primary endpoint trial?

It would require tens of thousands of patients followed for years. It would be expensive. And if the meta-analytic signal is real but modest — a 15-16% reduction — the trial would need to be enormous to detect that difference with adequate power. Pharmaceutical companies have little incentive to run such a trial when their drugs are already guideline-recommended for cardiovascular protection.

What about acute stroke? Preclinical data shows that GLP-1 RAs reduce infarct volume and promote neuroprotection when given in animal models of acute ischemia. Anti-inflammatory effects. Reduced oxidative stress. Improved cerebral blood flow. But no clinical trial has tested GLP-1 RAs in acute stroke or stroke recovery. Could early initiation after stroke improve outcomes? We have no idea.

Duration of benefit: most trials followed patients for 2 to 5 years. What happens at 10 years? Does the benefit persist? Does it increase with longer treatment duration? Or does it plateau?

Which GLP-1 RA is best for stroke? SUSTAIN-6 and REWIND showed stroke benefit with semaglutide and dulaglutide, respectively. LEADER and SELECT did not show stroke benefit with liraglutide or higher-dose semaglutide. Is this a difference in molecule? Dose? Patient selection? Or statistical noise? We're prescribing these drugs without knowing.

And then there's tirzepatide — a dual GIP/GLP-1 receptor agonist. It produces greater weight loss and greater A1c reduction than semaglutide. The SURPASS-CVOT trial is ongoing. Will it show stroke benefit? If it does, does that argue for a GLP-1-specific mechanism? Or does it just reinforce that cardiometabolic risk reduction translates to fewer events across the board?

The Diagnosis We're Actually Making

Here's what I think is happening. GLP-1 receptor agonists reduce cardiovascular events — convincingly, consistently, across multiple trials. They reduce myocardial infarction. They reduce cardiovascular death. And because stroke shares many of the same upstream risk factors — atherosclerosis, inflammation, endothelial dysfunction, hypertension — they probably reduce stroke too.

But calling them "stroke prevention drugs" overstates the directness of the evidence. The meta-analytic signal is real, but it's driven by a small number of trials, with wide confidence intervals, in highly selected populations. When you test the same drugs in broader populations — like SELECT — the stroke benefit disappears.

What we're looking at is not a cerebrovascular drug that happens to prevent MI. It's a cardiometabolic drug that improves the substrate from which both MI and stroke arise. That's still valuable. But it's different.

The pieces fit when you frame it this way: GLP-1 RAs are profoundly effective at reducing atherosclerotic cardiovascular disease burden. In patients with diabetes and high cardiovascular risk — the population most trials enrolled — that translates to fewer MIs, fewer cardiovascular deaths, and probably fewer strokes. In patients without diabetes, without established atherosclerotic disease, or with stroke mechanisms unrelated to large-artery atherosclerosis? The evidence is much thinner.

What This Means for Practice

Should stroke neurologists prescribe GLP-1 receptor agonists? In the right patient, yes. A 62-year-old with diabetes, hypertension, obesity, coronary artery disease, and a recent embolic stroke of undetermined source with aortic arch atheroma — absolutely. That patient fits the trial populations. The guidelines support it. The mechanisms align.

But we should be honest about what we're doing. We're not prescribing a stroke drug. We're prescribing a cardiometabolic drug that reduces atherosclerotic events broadly, with stroke as one component of that benefit.

And we should recognize the limits. The 45-year-old with cryptogenic stroke and no metabolic risk factors? The 80-year-old with lacunar disease? The patient who's already had a stroke despite optimal medical therapy including a statin and antihypertensive? We don't have trial data. We're extrapolating.

The diagnosis drives the treatment — so let's get the diagnosis right. If the diagnosis is "high atherosclerotic cardiovascular risk with metabolic syndrome," then GLP-1 RAs make sense. If the diagnosis is "stroke with unclear mechanism and I want to cover every possible pathway," we're guessing.

The question isn't whether GLP-1 receptor agonists reduce MACE. They do. The question is whether they're stroke drugs or cardiovascular drugs that happen to include stroke in the composite — and whether that distinction changes who we prescribe them to and why.

The workup doesn't end when you find something. It ends when the pieces fit. Right now, the pieces fit for cardiometabolic risk reduction in patients with diabetes and high cardiovascular risk. They fit less well for stroke-specific prevention in unselected populations.

That doesn't mean the drugs don't work. It means we need to know what question we're answering — and be honest when the trial didn't ask that question at all.