Patients on warfarin who started a GLP-1 drug spent 2.1 percentage points less time in their target blood-thinner range. The mechanism appears to run through the dinner plate, not the liver.
That is the headline finding of a retrospective cohort study ↗ published online May 11 in the Journal of Thrombosis and Thrombolysis. The investigators pulled records from the TriNetX research network, a federated database of de-identified electronic health records covering tens of millions of US patients, and screened 53,943 warfarin patients before narrowing to 1,021 who had documented warfarin use and INR measurements both six months before and six months after starting a GLP-1 receptor agonist. It is the largest dataset assembled to date for this specific question.
Time in therapeutic range, the standard quality metric for warfarin anticoagulation, dropped from 64.2 percent before GLP-1 initiation to 62.1 percent after (95% CI for the change −3.7% to −0.6%, p=0.01). Mean INR did not move. INR variability did. Time above the therapeutic window increased by 1.3 percentage points (p=0.04), the INR standard deviation drifted up by 0.03 (p=0.051), and clinicians shortened the interval between INR rechecks by about 0.7 days on average (p=0.049). The picture is not of a single dose-shift but of a coagulation system whose pharmacokinetic envelope has started to wobble.
Why a peptide changes a coumarin drug
Warfarin's dose-response ↗ is famously sensitive to dietary vitamin K, the cofactor it antagonizes. Leafy greens, broccoli, certain plant oils, even some fermented soy products all push INR down by feeding the clotting-factor machinery warfarin tries to starve. Dietitians counsel warfarin patients to keep vitamin K intake consistent rather than zero, because variability moves INR more than absolute intake.
GLP-1 receptor agonists work primarily by slowing gastric emptying and suppressing appetite at the central and gastrointestinal levels. Total food intake falls by roughly a third in the early weeks for many patients. The paper's authors propose that the appetite reduction does not reduce vitamin K intake uniformly, and that the day-to-day variability of intake is what pushes INR around. They did not measure vitamin K directly. The signal is consistent with that mechanism and inconsistent with a pharmacokinetic interaction, because pharmacokinetic interactions would move mean INR, not just its variance.
This puts GLP-1 receptor agonists in a category most drug-interaction databases were not built to flag. Standard interaction screens look for shared metabolic enzymes (CYP2C9 for warfarin), shared transporters, or shared protein-binding sites. None of those apply here. The interaction is mediated by behavior, the way a drug that changes what someone eats changes the bioavailability of something else they eat. Pharmacology has a name for this category, the food-drug interaction, but most of the textbook examples run in the other direction: grapefruit juice inhibits CYP3A4, St. John's wort induces it. A drug class that systematically nudges patient diet at population scale is a newer thing to think about.
What clinicians do with it
The clinical implication the authors land on is closer INR monitoring after GLP-1 initiation, not contraindication. The mean is fine. The variance is the problem, and variance responds to more frequent measurement.
The piece of the population for whom this matters most is warfarin patients with atrial fibrillation or mechanical heart valves who do not have a clean DOAC alternative. For everyone else on warfarin, the wobble argues quietly in favor of direct oral anticoagulants ↗ (apixaban, rivaroxaban, dabigatran, edoxaban), which do not depend on dietary vitamin K and have no equivalent variability signal. The decision was already trending that direction in most guideline updates. A reproducible diet-mediated wobble from a drug class that 12 percent of US adults will plausibly be on within five years is one more push.
Platform note
The peptide drugs at the center of this signal are semaglutide ↗, tirzepatide ↗, liraglutide ↗, dulaglutide ↗, and exenatide ↗. The study did not break results out by agent, which is the right call given the sample size but a real limitation: semaglutide and tirzepatide cause more pronounced appetite suppression than liraglutide or exenatide, and the diet-mediated wobble should track that gradient. A follow-up analysis stratified by agent and by weight loss achieved would test the mechanism harder. So would direct vitamin K intake measurement in a prospective cohort. Neither is in the field's queue yet, as best the GLP-1R pipeline ↗ can tell.
What this run leaves on the table is that the same logic should apply to any drug whose target therapeutic window depends on a nutrient the patient is now eating less of. Antiepileptics with narrow indices, thyroid replacement, certain immunosuppressants. None of those have a 1,021-patient TriNetX cohort yet. They will.