Risk of Bleeding with Warfarin Therapy for Atrial Fibrillation

Abstract & Commentary

By John P. DiMarco, MD, PhD, Professor of Medicine, Division of Cardiology, University of Virginia, Charlottesville. Dr. DiMarco does research for Medtronic, is a consultant for Medtronic, Novartis, and St. Jude, and is a speaker for Boston Scientific.

Source: Gomes T, et al. Rates of hemorrhage during warfarin therapy for atrial fibrillation. CMAJ 2012; Nov. 26. [Epub ahead of print.]

This paper details the result of a population-based cohort study of all Ontario, Canada, residents older than 66 years of age who began warfarin therapy for atrial fibrillation over an 11-year period. Patients were identified by the authors from data in the Ontario Health Insurance Plan. Patients were included if they had a hospital or office visit diagnosis of atrial fibrillation and began warfarin during this time period. Medical records were then surveyed for the occurrence of major hemorrhages. This was defined as a visit to an emergency department or admission to a hospital for hemorrhage during warfarin therapy. Patients were followed for up to 5 years after starting warfarin. Hemorrhage was classified by anatomic site using standard definitions. If patients had more than one admission for hemorrhage, only the first event was included.

During the 13-year study period, there were 125,195 patients who began therapy with warfarin in the setting of a diagnosis of atrial fibrillation. This was 47% of all new users of warfarin in this age group during this time period. Of these patients, 69% had an estimated CHADS2 score ≥ 2. In this inception cohort, the cumulative incidence of hemorrhage was 1.0% at 30 days, 4.1% at 1 year, and 8.7% at 5 years. During the study, the overall risk of hemorrhage was 3.8% per patient year. The annualized risk was highest during the first 30 days of therapy (11.8%) and 3.4% during the follow-up period. Hemorrhage was more common as the CHADS2 score increased. Patients with a CHADS2 score ≥ 4 had a 16.7% hemorrhage rate per person year in the first 30 days and 6.0% per year afterwards. By contrast, those with CHADS2 scores < 2 had hemorrhage rates of 1.8% per person year with a score of 0 and 2.5% per person year with a score of 1. Hemorrhage rates were higher among patients older than 75 years (4.6% in older patients vs 2.9% in younger patients). Upper and lower gastrointestinal hemorrhage accounted for 63% of the hemorrhage-related hospitalizations, intracranial hemorrhage for 5%, and other sites, mostly genitourinary, for 39%. There were 1963 deaths due to hemorrhage in the hospital or within 7 days after discharge. Intracranial hemorrhage had the highest mortality (41.7%) compared to gastrointestinal hemorrhage (14.7%) and other sites of hemorrhage (12.6%).

The authors conclude that in a large inception cohort of patients with atrial fibrillation, hemorrhage is common during both the first 30 days and subsequent months of warfarin therapy and is related to risk factors in the CHADS2 score and also to age. The risk of hemorrhage is higher than has been seen in recent published randomized trials of anticoagulation therapy. The mortality rate associated with hemorrhage, particularly intracranial hemorrhage, is extremely high.


Since the new oral anticoagulants, dabigatran and rivaroxaban, were approved for stroke prevention in patients with atrial fibrillation, concerns have been raised about bleeding problems with these agents that had not been prominent in the randomized clinical trials. Reports in both the medical literature and the lay press have complained that the level of anticoagulation cannot be monitored with standard tests and that there is no available rapid antidote. Although these statements about the new oral anticoagulants are true, this paper points out that bleeding with warfarin is also a major concern. Particularly striking are the very high rates of bleeding with warfarin during the first month of therapy and the shockingly high mortality rates seen during hospitalizations for hemorrhage. We have scoring systems for both stroke risk and bleeding risk in patients on anticoagulant therapy. The data in this paper indicate that applying these risk scores in clinical practice is still problematic.