The LAMP trial's post-hoc analysis landed in Stroke ↗ today. Liraglutide cut stroke recurrence from 18.1 percent to 5.8 percent in patients with insulin resistance. In the half of the cohort without insulin resistance, the curves overlapped.
LAMP itself was an open-label randomized trial run across 27 hospitals in China between June 2019 and December 2023, enrolling adults with a minor ischemic stroke or high-risk transient ischemic attack plus type 2 diabetes. Six hundred thirty-six patients were randomized 1:1 to liraglutide ↗ plus standard secondary-prevention therapy or to standard therapy alone. The main trial result, published earlier, was a 90-day reduction in recurrent vascular events. The new analysis takes the 510 patients with complete metabolic data and asks a simpler question. Is the benefit evenly distributed, or does it sit in a subgroup?
It sits in a subgroup. Using the homeostasis model assessment of insulin resistance (HOMA-IR), a routine bedside calculation from fasting insulin and fasting glucose, the authors split the cohort at a value of 2.5, the established cutoff for clinically meaningful insulin resistance in Asian populations. The treatment-by-insulin-resistance interaction was significant for both stroke recurrence and composite vascular events (p equals 0.02 for both). In plain terms, the model is saying the average effect across the whole trial is hiding two different things happening to two different groups.
In the insulin-resistant half (HOMA-IR at or above 2.5), recurrent strokes hit 5.8 percent of patients on liraglutide versus 18.1 percent on standard care alone. The absolute risk reduction is 12.3 percentage points (95 percent CI, 5.6 to 19.0). The number needed to treat is 8 to prevent one recurrent stroke in three months. Composite vascular events tell the same story: 5.8 percent on liraglutide versus 19.2 percent on standard care, with an absolute risk reduction of 13.4 percentage points and the same NNT of 8. In the insulin-sensitive half (HOMA-IR below 2.5), neither outcome showed a significant difference. The drug was doing essentially nothing for them.
Why the subgroup story is the story
Eight is a remarkably small number to prevent a recurrent stroke. The current secondary-prevention algorithm after a minor ischemic stroke runs about NNT 20 to 40 for antiplatelet therapy added to risk-factor control, depending on the specific population. Recurrent stroke after a minor index event is common (roughly 10 to 15 percent of patients have another event in the first 90 days), expensive, and on balance worse for the patient than the first event because functional reserve is already compromised. An NNT of 8 from a peptide that is already approved, already on formularies, and already prescribed daily by endocrinologists across the same patient population is the kind of signal that gets a Phase 3 confirmation written into the next round of acute-stroke trial protocols.
The interaction signal also raises a real mechanistic question. Liraglutide's brain effects have been ascribed variously to direct GLP-1 receptor activation on cerebrovascular endothelium, to anti-inflammatory action on microglia, and to systemic improvements in glycemic and lipid control. The LAMP post-hoc result is most consistent with one of two readings. Either the drug works mainly through a pathway that requires insulin resistance to be operating (clearance of free fatty acids, suppression of hepatic gluconeogenesis, restoration of muscle insulin signaling), or the absolute risk in the insulin-resistant subgroup is just much higher to begin with and there is more room to move it. The trial data does not let those two readings be teased apart. A future trial that enrolls only insulin-resistant T2D stroke patients would resolve it.
What does not survive the post-hoc framing
The usual caveats apply harder than they normally do. This is a post-hoc analysis with a pre-specified threshold borrowed from the Asian-population literature; a Western trial would have used HOMA-IR 2.7 or 3.0 and the cohort split would look different. The follow-up is three months, so the analysis cannot speak to whether the benefit persists or whether the non-insulin-resistant subgroup catches up at one year. The trial is open-label, so the clinicians knew which patients were on liraglutide and which were not, and the outcome ascertainment for recurrent stroke (which often presents with subtle deficits) is not blinded. The sample is from a single Chinese network, and ethnic differences in insulin-secretion versus insulin-resistance dominance of type 2 diabetes mean the findings will need confirmation in a more mixed cohort before they enter guidelines. None of that is a reason to ignore the result. All of it is a reason to power a confirmatory trial the right way the first time.
The platform angle, briefly. Liraglutide is one of the longer-tenured molecules on peptidemodel, with a card that already lists cardiovascular among its indications. The LAMP secondary-stroke signal adds a specific subgroup-stratified data point to that picture. The cerebrovascular axis is where the rest of the GLP-1 class is heading: semaglutide's SELECT trial reduced stroke in non-diabetic patients with established cardiovascular disease, and the SUSTAIN-6 and LEADER trials both showed nominal stroke reductions that were never the headline result. LAMP is the first piece of GLP-1 stroke data with a clean subgroup interaction and a defined biomarker for who benefits, and that makes it the right primary read for clinicians who have been triangulating from class-effect arguments for the past three years.