No clinical prediction model has been specifically developed or validated to identify patients with unprovoked venous thromboembolism (VTE) who are at high risk of major bleeding during extended anticoagulation. In a prospective multinational cohort study of patients with unprovoked VTE receiving extended anticoagulation after completing ≥3 months of initial treatment, we derived a new clinical prediction model using a multivariable Cox regression model based on 22 pre-specified candidate predictors for the primary outcome of major bleeding. This model was then compared with modified versions of five existing clinical scores. A total of 118 major bleeding events occurred in 2516 patients (annual risk, 1.7%; 95% confidence interval, 1.4-2.1). Incidence of major bleeding events per 100 person-years in high- and non-high-risk patients, respectively, were 3.9 (95% confidence interval, 3.0-5.1) and 1.1 (0.8-1.4) using the newly derived CHAP model (creatinine, hemoglobin, age, and use of antiplatelet agent), 3.3 (2.6-4.1) and 1.0 (0.7-1.3) using modified ACCP, 5.3 (0.6-19.2) and 1.7 (1.4-2.0) using modified RIETE, 3.1 (2.3-3.9) and 1.1 (0.9-1.5) using modified VTE-BLEED, 5.2 (3.3-7.8) and 1.5 (1.2-1.8) using modified HAS-BLED, and 4.8 (1.3-12.4) and 1.7 (1.4-2.0) using modified OBRI scores. Modified versions of the ACCP, VTE-BLEED, and HAS-BLED scores help identify patients with unprovoked VTE who are at high risk of major bleeding and should be considered for discontinuation of anticoagulation after 3-6 months of initial treatment. The CHAP model may further improve estimation of bleeding risk by using continuous predictor variables, but external validation is required before its implementation in clinical practice.-