Bleeding classification in clinical trials: observer variability and clinical relevance.

To evaluate the bleeding classification in a recent trial on venous thrombosis treatment, a selection of reported bleeding episodes was adjudicated twice by an independent committee and graded by the treating physician and independent clinical experts on the clinical severity and impact on the patient's life. The kappa values for the dichotomy major bleeding versus minor or no bleeding were 0.79 (95% CI, 0.57-1.0) for the agreement between the two members of the adjudication committee and 0.77 (95% CI, 0.52-1.0) for the agreement between both adjudication sessions. The kappa values for the dichotomy major or minor bleeding versus no bleeding were 0.42 and 0.44. The weighted kappa values for the agreement between the treating physician and the independent experts were 0.76 for the clinical severity and 0.79 for the impact on the patient's life (95% CI, 0.63-0.88 and 0.70-0.89). The association between the adjudication result expressed as major bleeding or minor or no bleeding and the clinical grading by the treating physician resulted in an ROC curve with an area under the curve of 0.98 for the clinical severity and 0.99 for the impact on the patient's life. The dichotomy major or minor bleeding versus no bleeding resulted in areas under the curve of 0.70 and 0.66. In conclusion, the applied criteria for major bleeding are reproducible and clinically relevant. The criteria for minor bleeding are not reproducible and are less associated with the observed clinical relevance.