Abstract
Abstract
Here, an effort was made to develop and internally validate two predictive models for assessing thrombosis and bleeding risk in hospitalized patients with COVID-19. A retrospective cohort study was conducted on a cohort of 950 patients diagnosed with COVID-19. Least Absolute Shrinkage and Selection Operator (LASSO) regression combined with tenfold cross-validation was adopted for predictor screening, and bootstrap resampling was performed for internal validation. The overall incidence of thrombotic and bleeding complications during hospitalization was 4.9% (47/950) and 16.0% (152/950), respectively. Multivariable analysis confirmed that history of venous thromboembolism, cerebrovascular events, D-dimer, hormone therapy, and prophylactic/therapeutic low-molecular-weight heparin were independent predictors of thrombosis, while malignancy, platelet count, INR, D-dimer, fibrinogen, mechanical ventilation and COVID-19 severity were significant predictors of bleeding. The established models showed favorable discriminative ability, calibration and clinical utility, with Area Under the Receiver Operating Characteristic Curve (AUC) of 0.848 for thrombosis and 0.824 for bleeding. These validated models offer a practical tool for individualized risk stratification, helping to optimize anticoagulant decision-making and reduce the occurrence of thrombotic and bleeding events in hospitalized COVID-19 patients.