Development of a Contemporary Bleeding Risk Model for Elderly Warfarin Recipients

Background and purpose Develop and validate a contemporary bleeding risk model to guide the clinical use of warfarin in the elderly atrial fibrillation (AF) population. Methods Chart-abstracted data from the National Registry of Atrial Fibrillation was combined with Medicare part A claims to identify major bleeding events requiring hospitalization. Using a split-sample technique, candidate variables that provided statistically stable relationships with major bleeding events were selected for model development. Three risk categories were created and validated. The new model was compared to existing bleeding risk models using c-statistics and Kaplan-Meier curves. Results Model development and validation was conducted on 26,345 AF patients who were > 65 years of age and had been discharged from the hospital while receiving warfarin therapy. The following eight variables were included in the final risk score model: age ≥ 70 years; gender; remote bleeding; recent (ie, during index hospitalization) bleeding; alcohol/drug abuse; diabetes; anemia; and antiplatelet use. Bleeding rates were 0.9%, 2.0%, and 5.4%, respectively, for the groups with low, moderate, and high risk, compared to the bleeding rates for groups with moderate risk (1.5% and 1.0%) and high risk (1.8% and 2.5%) from other models. Conclusions Using a nationally derived data set, we developed a model based on contemporary practice standards for determining major bleeding risk among AF patients receiving warfarin therapy. The larger sample size afforded the opportunity to incorporate additional risk factors. In addition, since the majority of our population was > 65 years of age, we had greater ability to stratify risk among the elderly. Develop and validate a contemporary bleeding risk model to guide the clinical use of warfarin in the elderly atrial fibrillation (AF) population. Chart-abstracted data from the National Registry of Atrial Fibrillation was combined with Medicare part A claims to identify major bleeding events requiring hospitalization. Using a split-sample technique, candidate variables that provided statistically stable relationships with major bleeding events were selected for model development. Three risk categories were created and validated. The new model was compared to existing bleeding risk models using c-statistics and Kaplan-Meier curves. Model development and validation was conducted on 26,345 AF patients who were > 65 years of age and had been discharged from the hospital while receiving warfarin therapy. The following eight variables were included in the final risk score model: age ≥ 70 years; gender; remote bleeding; recent (ie, during index hospitalization) bleeding; alcohol/drug abuse; diabetes; anemia; and antiplatelet use. Bleeding rates were 0.9%, 2.0%, and 5.4%, respectively, for the groups with low, moderate, and high risk, compared to the bleeding rates for groups with moderate risk (1.5% and 1.0%) and high risk (1.8% and 2.5%) from other models. Using a nationally derived data set, we developed a model based on contemporary practice standards for determining major bleeding risk among AF patients receiving warfarin therapy. The larger sample size afforded the opportunity to incorporate additional risk factors. In addition, since the majority of our population was > 65 years of age, we had greater ability to stratify risk among the elderly.