Methodological considerations for interpretation of rates of major haemorrhage in studies of anticoagulant therapy for atrial fibrillation

Abstract

Major haemorrhage is a significant concern for elderly patients with atrial fibrillation (AF) taking anticoagulant therapy. Two recently published studies have demonstrated conflicting results with regard to the incidence of major haemorrhage on warfarin in this age group. The Birmingham Atrial Fibrillation Treatment of the Aged Study1 (BAFTA) randomized 973 patients with AF aged 75 years or more to warfarin or aspirin. During the 2.7 year mean follow-up, the incidence of major haemorrhage was nearly identical in the two study arms (1.9% per year for warfarin vs. 2.2% for those taking aspirin). In contrast, a single centre, inception cohort study identified a significantly greater incidence of major haemorrhage in elderly patients initiating warfarin for stroke prevention in AF:2 7.2% per year in a population with mean age 77 years, increased to 13.1% per year among those aged 80 and older. How can such disparate rates be reconciled? The answers are found in the key elements of study design which guide the interpretation of event rates. These studies contrasting results demonstrate the importance of assessing study design, patient population (with particular regard to selection bias), length of observation, definition of major haemorrhage, specific anticoagulant, quality of anticoagulation control, and use of other risk-modifying therapy. Randomized controlled trials are the most valid assessment of an intervention. Although often criticized for the highly selected nature of the study population, a trial's first priority is to ensure internal validity to facilitate an objective assessment of efficacy. Anticipated drop-out rates should be low and medication adherence rates should be high. Translating trial results into real-world practice can be challenging because the patient populations are often considerably different (external validity). Observational studies provide important insight into clinical practice, but these designs are also subject to bias. A source of bias that affects both randomized …