Risk of Hemorrhage on Warfarin

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: Fang MC, et al. A new risk scheme to predict warfarin-associated hemorrhage: The ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation) study. J Am Coll Cardiol 2011;58:395-401.

In this study, the authors attempt to develop a risk stratification score to predict bleeding in patients treated with warfarin oral anticoagulation. The Anticoagulation and Risk Factors in Atrial Fibrillation (ATRIA) study follows 13,559 adults with nonvalvular permanent atrial fibrillation enrolled in the Kaiser Permanente Health System of Northern California. Clinical characteristics for these patients are maintained in a large system-wide database. This database was searched for discharge diagnoses for extracranial hemorrhages and for primary and secondary diagnoses of intracranial bleeding events including intracerebral, subarachnoid, or subdural hemorrhages. The records of patients with an identified hemorrhagic event were reviewed by a clinical outcomes committee and classified using a formal protocol. Only events that occurred within 5 days of warfarin exposure were included. Hemorrhages that were not present on admission but occurred during a hospitalization or were procedurally related were excluded. Major hemorrhages were defined as those that were either fatal, required a transfusion of greater than or equal to two units of packed cells, or involved a critical anatomic site.

The data in this report were obtained in 9186 subjects in the ATRIA study cohort who contributed 32,888 person-years of warfarin exposure. The group was split into "derivation" and "validation" cohorts for developing and testing the risk scheme. Among these patients, there were 461 warfarin-associated major hemorrhages for an annualized rate of 1.4% per year. Five variables were identified as having the highest predictive value: baseline anemia, severe renal disease, age greater than 75 years, any prior hemorrhage diagnosis, and a diagnosis of hypertension. Stratification of these variables resulted in anemia and severe renal disease being assigned 3 points, age greater than 75 being assigned 2 points, and prior hemorrhage diagnosis and hypertension being assigned 1 point each. Therefore, the final risk scheme had a possible range of 0 to 10 points. The score was essentially equally effective in both the derivation and validation cohorts. In the combined derivation and validation group, events per 100 patient years ranged from 0.4 for those with a risk score of 0, 2.6 for those with a risk score of 4, and up to 12.4 and 17.25 for patients with risk scores of 9 or 10. When grouped into low (0 to 3 points), intermediate (4 points), and high-risk (5-10 points) category groups, the event rates were 0.76, 2.62, and 5.76 events per 100 patient years. The calculated c-index statistics for the three category score system was 9 and for continuous scores 0.74. These values were superior to the c-index scores for six other published risk schemes.

The authors conclude that a relatively simple scoring system based on these five parameters will help clinicians estimate the magnitude of hemorrhagic risk when prescribing or continuing anticoagulant therapy.


Anticoagulation therapy in patients with non-valvular atrial fibrillation requires physicians and patients to carefully weigh the potential risks and benefits of long-term therapy. The two most prevalent stroke risk scoring systems are the CHADS2 and the CHA2DS2VASc, and practice guidelines guideline rely heavily on them. Bleeding risk has received less attention in the atrial fibrillation population. Therefore, the new system proposed here should be quite valuable to clinicians planning to start therapy.

Several other factors also must be considered. Favoring a decision for anticoagulation is the observation that only intracranial bleeding is frequently fatal and/or disabling, while many strokes are not. Against anticoagulation is the frequent "minor" bleeding that adversely affects the lifestyle of many patients. We should also note that the recent release of dabigatran, an oral direct thrombin inhibitor, and the expected approval of one or more Factor Xa inhibitors for stroke prevention in atrial fibrillation have changed the playing field in oral anticoagulation. Although it is likely that the bleeding risk score here will also work for patients treated with these new agents, that will have to be documented clinically. It is also somewhat unfortunate that the authors chose not to compare their risk scheme to the recently published HAS-BLED scheme used in the recent European Society of Cardiology atrial fibrillation guidelines. The HAS-BLED score includes points for a history of a labile INR, hepatic disease, use of drugs or alcohol, and age 65-75. Both of these two new bleeding risk scoring systems are relatively simple and should prove helpful to clinicians as they deal with atrial fibrillation patients.