Abstract
We aimed to identify factors associated with bleeding in patients with venous thromboembolism (VTE) when using direct oral anticoagulants (DOACs), to construct and externally validate a predictive model for bleeding, and to provide a validated tool for clinical assessment of bleeding. The prediction model in the development cohort was constructed using logistic regression and visualized through a Nomogram. We conducted external validation of the model. The accuracy and calibration were evaluated through the area under the curve (AUC), calibration curves, and the Hosmer-Lemeshow test. This multicenter retrospective study recruited 1121 patients treated with DOACs for VTE. The training set consisted of 806 patients from 11 centers and the external validation set consisted of 315 patients from 9 centers. The Alfalfa-VTE-Major model (AUC = 0.883) is composed of five variables: Age ≥ 65 years, history of bleeding, malignancy, and coadministration of antiplatelet drugs/nonsteroidal antiinflammatory drugs (NSAIDs) were independent risk factors for major bleeding, and coadministration of gastrointestinal protectants was a protective factor. The Alfalfa-VTE-Minor model (AUC = 0.875) was composed of five independent risk factors for minor bleeding variables: Age ≥ 65 years, anemia, history of bleeding, vascular disease, and coadministration of antiplatelet drugs/NSAIDs. We externally validate the currently widely used VTE bleeding model. The predictive power of the three bleeding scores was RIETE (AUC = 0.773), VTE-BLEED (AUC = 0.746), and Hokusai (AUC = 0.558) in descending order. We constructed and externally validated a predictive model for the occurrence of major and minor bleeding in VTE patients using DOACs. Both models have good predictive value and may be effective tools to help reduce the incidence of bleeding in patients with DOACs.