Adults with asymptomatic intracranial atherosclerosis who were on a GLP-1 drug at diagnosis died at about a third the rate of matched non-users over the next year. The signal held after matching for HbA1c, LDL cholesterol, body mass index, and systolic blood pressure. That last point is what makes this real-world cohort worth reading.
The cohort came from the TriNetX US Collaborative Network, which pools electronic health records from seventy-one healthcare organizations. The analysis, published May 21 in Neurology International ↗, pulled adults with an intracranial atherosclerotic stenosis (ICAS) diagnosis between January 2016 and December 2025, excluded anyone with a prior stroke, cerebral infarction, or other cerebrovascular ischemic syndrome, then asked who was on a GLP-1 receptor agonist in the six months before the diagnosis or on the diagnosis date itself.
Asymptomatic intracranial atherosclerotic stenosis is exactly what it sounds like. A patient has a narrowing of an artery inside the skull that shows up on imaging, often during a workup for something else, but has not yet had a stroke. The standard management is vascular risk-factor optimization. Statins, blood pressure control, diabetes control. No routine antiplatelet, no stent, no surgical intervention. The condition mostly sits and waits.
Before propensity matching the GLP-1 cohort had 1,746 patients. The non-user cohort had 71,792. After one-to-one matching on demographics, vascular risk factors, comorbidities, antithrombotic and lipid and diabetes therapies, plus a set of cardiometabolic lab and physiologic measures (HbA1c, LDL, BMI, and systolic blood pressure are the named ones in the abstract), 1,728 patients sat in each arm. The five GLP-1 receptor agonists captured in the exposure definition were lixisenatide ↗, semaglutide ↗, liraglutide ↗, tirzepatide ↗, and dulaglutide ↗.
The three one-year outcomes the authors prespecified are ischemic stroke, all-cause mortality, and a composite of stroke or mortality.
Ischemic stroke happened in 4.40 percent of GLP-1 users versus 6.10 percent of non-users (hazard ratio 0.70, 95% CI 0.52 to 0.95, p=0.044). That is consistent with the cerebrovascular benefit GLP-1 drugs have shown in the larger cardiometabolic outcome trials.
All-cause mortality is the bigger number. 3.40 percent in GLP-1 users versus 9.40 percent in non-users at one year, hazard ratio 0.35 (95% CI 0.26 to 0.47, p<0.001). That is a sixty-five percent relative reduction. The composite endpoint, stroke or death, ran 7.50 percent versus 15.00 percent (HR 0.48, 95% CI 0.39 to 0.59, p<0.001).
A 65 percent relative cut in one-year mortality is large enough that the natural first explanation is residual confounding. People who get GLP-1 drugs are different from people who don't, even in matched cohorts. The strength of this analysis is the matching list itself. HbA1c is in there. LDL is in there. BMI is in there. Systolic blood pressure is in there. The four levers that GLP-1 drugs are known to pull, plus the major non-cardiometabolic comorbidities and concurrent medications, were balanced between arms before the outcomes were counted. The signal still ran 65 percent on mortality and 30 percent on stroke.
The authors stop short of attributing the effect to any specific mechanism. They flag the result as hypothesis-generating and explicitly call for prospective studies. Two readings are worth keeping in mind. One, that there is a real cerebrovascular and survival benefit that does not show up in HbA1c, LDL, BMI, or systolic blood pressure, possibly mediated by GLP-1 signaling in vascular endothelium, microglia, or some downstream inflammatory pathway. Two, that the unmeasured confounders are bigger than the measured ones, propensity matching does only what it does, and people who get prescribed a GLP-1 are systematically healthier in ways the EHR does not capture (engagement with care, adherence patterns, frailty unrecorded in problem lists). The TriNetX cohort cannot distinguish them.
For the receptor-class picture this matters most for the GLP-1R ↗ page on peptidemodel. The five drugs in the exposure definition are the five GLP-1 receptor agonists currently in widest US use, and four of them are peptides that the platform hosts as cards in their own right (lixisenatide, semaglutide, liraglutide, dulaglutide, with tirzepatide ↗ as the GIP and GLP-1 dual). The piece of work the trial does not do is separate the five drugs from each other. Whether the cerebrovascular signal sits with the long-acting agents or runs across the class, and whether the dual-agonist sits cleanly above or below, is what a prospective head-to-head trial would have to address.