Warfarin is the drug most often used to illustrate what a narrow therapeutic index means: too little and a clot forms, too much and a bleed starts, and the window between them is measured by a blood test taken every few weeks. A drug class that slows the stomach and suppresses appetite looks, on paper, like exactly the kind of thing that should disturb it. What follows is what has actually been measured, which is less than the mechanism suggests and less than a reader would expect.
What the mechanism predicts
Delayed gastric emptying changes how fast a tablet reaches the small intestine, not how much of it eventually gets there. Across the wider interaction literature described in the oral medications article, the consistent pattern is a lower and later peak concentration with total exposure unchanged. Warfarin is highly soluble and highly permeable, so the prediction for it is a slightly later peak and no change in the amount absorbed — which is a prediction about a drug whose effect is measured days after a dose anyway.
Four pharmacokinetic studies, and what two of them measured directly
Subcutaneous semaglutide at steady state was tested against a single warfarin dose in healthy subjects. Exposure ratios fell inside the prespecified 0.80 to 1.25 interval, peak concentration was not meaningfully changed, and the paper reports no clinically relevant change in international normalized ratio response.[1] The semaglutide tablet was tested the same way in two crossover trials, with no apparent effect on warfarin exposure or peak concentration from either the drug or its absorption enhancer.[2]
The two studies that measured the anticoagulant effect rather than the drug level found it going the wrong way for the popular theory. With exenatide in 16 healthy men, R- and S-warfarin pharmacokinetics did not change significantly, and the ratios of geometric means were 0.94 (0.93 to 0.96) for the area under the INR-time curve and 0.88 (0.84 to 0.92) for the maximum observed INR response.[3] With dulaglutide, maximum INR was not significantly affected and the area under the INR curve rose by 2%.[4] Four industry studies, none showing a clinically relevant change, and the only measured direction of effect was slightly less anticoagulation, not more.
What those four studies were not built to see
All four share a design, and the design has a blind spot worth naming. Each enrolled healthy volunteers, gave a single 25 mg warfarin dose in three of the four, and followed the response for days rather than months.[1][2][3] Healthy volunteers are not anticoagulated to a target, do not lose weight on the drug being tested, and cannot show a drift that accumulates across a dose-escalation schedule.
More to the point, none of these subjects had their appetite altered for long enough to change what they ate. A study measuring whether the stomach delays a tablet is answering a pharmacokinetic question, and the thing most likely to disturb warfarin in this population is not pharmacokinetic at all. That mismatch — the right studies for the wrong mechanism — is the shape of the evidence gap on this page.
One label carries a warfarin instruction, and it is not the one people buy
The exenatide label is the only product label in this class that treats warfarin as a named interaction with its own monitoring instruction. It states that the drug interaction study found no significant effect on INR, then records postmarketing reports of increased INR with concomitant warfarin use, sometimes associated with bleeding, and instructs that prothrombin time be monitored more frequently after initiation or alteration of therapy until stable.[5] That instruction rests on spontaneous reports rather than on the trial, which is an unusual and honest thing for a label to show side by side.
The molecules this market sells carry nothing equivalent. The tirzepatide labels name warfarin only as a parenthetical example of a narrow therapeutic index drug to monitor when oral medications are coadministered.[6] The semaglutide labels list warfarin among the drugs whose pharmacokinetics showed no clinically significant difference, and give their one concrete interaction figure for levothyroxine instead — the subject of a separate page. Nobody on semaglutide or tirzepatide is being told by a label to check an INR more often.
The largest real-world dataset finds a small effect and a different cause
A 2026 retrospective cohort screened 53,943 patients in a research network and found 1,021 with documented warfarin use both before and after starting a GLP-1 receptor agonist. Comparing the six months on either side of that start, mean time in therapeutic range fell by 2.1 percentage points (95% CI, −3.7 to −0.6; P = 0.01), from 64.2% to 62.1%. Mean INR did not change at all, at 0.00 (95% CI, −0.02 to 0.03; P = 0.79). Time above the therapeutic range rose by 1.3% (P = 0.04), time below by 0.8% (P = 0.25), and the interval between INR rechecks shortened by 0.7 days.[7]
The authors frame the mechanism as indirect, and the framing matters more than the effect size: GLP-1 receptor agonists do not affect warfarin pharmacokinetics, but their appetite-reducing effect may decrease vitamin K intake and increase INR variability.[7] That is not an absorption interaction. It is a diet interaction, and it belongs to the same family of second-order effects as the reduced food intake that makes insulin dosing harder to get right. A person eating half as much salad is taking half as much vitamin K, and warfarin has been titrated against the old amount.
The case literature is two reports, and they contradict each other
Searching PubMed for tirzepatide and warfarin together returns two records. Both are single case reports, and they describe opposite events.
In the first, a 66-year-old woman with grade III obesity and a mechanical mitral valve prosthesis started tirzepatide in hospital. After ten days, routine testing showed an unexpectedly significant reduction in INR despite a strict and stable diet, and the therapeutic target was regained only after roughly three weeks of progressive increases in the daily warfarin dose.[8] In the second, a 64-year-old man on warfarin for chronic deep vein thrombosis developed pulmonary hemorrhage and acute kidney injury shortly after a tirzepatide dose increase, with an INR of 8.7. That case was confounded by markedly elevated c-ANCA and a working diagnosis of ANCA-associated vasculitis, the patient died before biopsy confirmation, and the authors state that definitive causality remains unclear.[9]
One under-anticoagulation, one over-anticoagulation, in a literature of two. Neither establishes a direction, and quoting whichever one fits a narrative would misrepresent the size and the disagreement of the published record.
For the direct oral anticoagulants the count is worse
Apixaban, rivaroxaban and their relatives have displaced warfarin for most indications, and they carry no routine blood test at all — which removes both the monitoring burden and the instrument that would detect a problem. A PubMed search pairing semaglutide with the four marketed direct oral anticoagulants returns 15 records, none of which is an interaction study of that pair; a search for the drug class against the category returns 7, likewise none.
The closest published dataset used a molecule nobody can buy. A 2026 phase 1 program assessed lotiglipron, an oral small-molecule GLP-1 receptor agonist, against dabigatran and found peak concentration reduced by up to 55% with no significant change in total exposure — the same shape seen across this class. Both studies were terminated early along with the drug's development after liver transaminase elevations.[10] A small molecule is not a peptide, and a discontinued compound is not semaglutide, so that finding does not transfer. It is simply the nearest thing that exists.
What an absence this size means for a cash intake
Anticoagulant bleeding is not a rare or theoretical harm. National surveillance estimated 1,270,259 emergency department visits for oral anticoagulant-related bleeding across 2016 to 2020, of which 47.8% resulted in hospitalization, at a rate of 13.0 visits per 100 patients dispensed warfarin and 5.9 per 100 dispensed a direct oral anticoagulant.[11] Against that baseline, a 2.1 point loss of time in therapeutic range is small, and the honest summary is that no study has been designed or sized to detect whether it translates into events.
The practical consequence is narrow and worth stating plainly. There is no evidence that an anticoagulant should stop, no labeled reduction for either drug, and no reason from the pharmacology to expect a large effect. What the evidence does support is that the months of appetite change after starting are a period of greater INR variability, and an INR is the one instrument that would show it. That requires somebody to know both prescriptions exist — the recurring gap described in the telehealth intake article.
Finally, the product distinction this market turns on. Compounded semaglutide and tirzepatide are not approved by the FDA, and are not reviewed by the agency for safety, efficacy or quality before a pharmacy dispenses them. Nothing above changes on that basis, because there is almost nothing there to change; the published record on this pair is thin for the branded products and empty for the compounded ones. How claims on this site are established is set out in the methodology.