BACKGROUND: Warfarin is effective in the prevention of stroke in atrial fibrillation but is under used in clinical care. Concerns exist that published rates of hemorrhage may not reflect real-world practice. Few patients > or = 80 years of age were enrolled in trials, and studies of prevalent use largely reflect a warfarin-tolerant subset. We sought to define the tolerability of warfarin among an elderly inception cohort with atrial fibrillation. METHODS AND RESULTS: Consecutive patients who started warfarin were identified from January 2001 to June 2003 and followed for 1 year. Patients had to be > or = 65 years of age, have established care at the study institution, and have their warfarin managed on-site. Outcomes included major hemorrhage, time to termination of warfarin, and reason for discontinuation. Of 472 patients, 32% were > or = 80 years of age, and 91% had > or = 1 stroke risk factor. The cumulative incidence of major hemorrhage for patients > or = 80 years of age was 13.1 per 100 person-years and 4.7 for those < 80 years of age (P=0.009). The first 90 days of warfarin, age > or = 80 years, and international normalized ratio (INR) > or = 4.0 were associated with increased risk despite trial-level anticoagulation control. Within the first year, 26% of patients > or = 80 years of age stopped taking warfarin. Perceived safety issues accounted for 81% of them. Rates of major hemorrhage and warfarin termination were highest among patients with CHADS2 scores (an acronym for congestive heart failure, hypertension, age > or = 75, diabetes mellitus, and prior stroke or transient ischemic attack) of > or = 3. CONCLUSIONS: Rates of hemorrhage derived from younger noninception cohorts underestimate the bleeding that occurs in practice. This finding coupled with the short-term tolerability of warfarin likely contributes to its underutilization. Stroke prevention among elderly patients with atrial fibrillation remains a challenging and pressing health concern.