OBJECTIVES: The reported incidence of bleeding associated with oral anticoagulation therapy varies widely. The study objectives were to identify good quality evidence about risk of bleeding and to investigate the impact of study heterogeneity on outcomes. METHODS: A search was made of MEDLINE and EMBASE for randomised controlled trials and inception cohort studies between January 1990 and March 2002. Selection criteria were: anticoagulation monitored by INR, percentage time within range stated, and criteria for defining a major bleed stated. The sensitivity of diagnostic criteria for bleeding events was investigated by assembling a six-month cross sectional retrospective cohort of anticoagulated patients. Outcome events in this cohort were then assessed against the criteria for "major bleeds" proposed by each reviewed study. RESULTS: Twelve studies were identified that met the selection criteria. Significant variation was seen in the major bleed rates across the studies (P <0.0002, chi-square; range from 0.5 to 6.6 events per 100 patient years) confirming significant heterogeneity in the results. In the assembled cohort of 3998 patients there were 174 admissions in 6 months, of which 117 included some evidence of warfarin related bleeding. Applying the criteria from the 12 different studies resulted in a range of 24 to 117 of these events being classified as "major bleeds". The 3 studies with most restrictive definitions reported from 0.5 to 1.1 events per 100 patient years, compared with from 1.5 to 2.8 events per 100 patient years for the 4 studies with least restrictive definitions. CONCLUSIONS: It is essential for future outcomes research in anticoagulant therapy that consistent definitions of major adverse outcomes are applied. The inconsistent definitions used in the 12 published studies reviewed make it impossible to assess the impact of other differences between studies, for example in INR range or % of time within range.