A bleeding risk index for estimating the probability of major bleeding in hospitalized patients starting anticoagulant therapy.
 PURPOSE: To construct and test prospectively a bleeding risk index for estimating the probability of major bleeding in hospitalized patients starting long-term anticoagulant therapy.
 PATIENTS AND METHODS: In an inception cohort of 617 patients starting long-term anticoagulant therapy in one hospital, data were gathered retrospectively and bleeding was classified using reliable explicit criteria.
 We constructed a bleeding risk index by identifying and weighting independent predictors of major bleeding using a multivariate proportional-hazards model.
 The bleeding risk index was tested in 394 other patients prospectively identified in a second hospital.
 The index was compared to physicians' predictions.
 RESULTS: Major bleeding developed before discharge in 61 of all 1,011 patients (6%).
 The bleeding risk index included four independent risk factors for major in-hospital bleeding: the number of specific comorbid conditions; heparin use in patients aged 60 years or older; maximal prothrombin or partial thromboplastin time 2.0 or more times control; liver dysfunction worsening during therapy.
 In the testing group, the index predicted major bleeding, which occurred in 3% of 235 low-risk patients, 16% of 96 middle-risk patients, and 19% of 63 high-risk patients (p less than 0.001).
 The bleeding risk index performed as well as physicians' predictions, and integration of the bleeding risk index with physicians' predictions led to a classification system that was more sensitive (p = 0.03) than physicians' predictions alone.
 In 86% of patients with a high risk of major bleeding, we identified specific ways of improving therapy, e.g., avoiding overanticoagulation and nonsteroidal anti-inflammatory agents.
 CONCLUSION: The bleeding risk index provides valid estimates of the probability of major bleeding in hospitalized patients starting long-term anticoagulant therapy and complements physicians' predictions.
 The possibility that bleeding can be prevented in high-risk patients warrants prospective evaluation.
