Background: COVID-19 patients are at increased risk of thrombosis and bleeding, but no standardized bleeding risk assessment tool has been recommended.
Objective: This study evaluates the predictive value of the IMPROVE Bleeding Risk Score (BRS) in hospitalized COVID-19 patients.
Design: A multicenter, prospective cohort of 3,886 hospitalized COVID-19 patients across six tertiary hospitals in China between December 1, 2022, and January 31, 2023.
Participants: Patients were objectively diagnosed with COVID-19 by pathogen or antibody detection and followed for 90 days.
Main measures: The primary outcomes were major bleeding (MB) and clinically relevant non-major bleeding (CRNMB). We evaluated the IMPROVE BRS predictive performance using hazard ratios (HRs), positive and negative predictive values, the area under the receiver operating characteristic curve (AUC), and calibration.
Key results: Among 3,886 hospitalized COVID-19 patients (median age 74, IQR 62-84), 42 MB (1.1%) and 47 CRNMB (1.2%) events occurred within 90 days. The IMPROVE BRS performed well in predicting MB events, with an AUC of 0.84 (95% CI, 0.77-0.91) at 90 days. Calibration plots indicated good calibration. High-risk patients had a significantly higher bleeding risk than low-risk patients, even after adjusting for low molecular weight heparin (LMWH) thromboprophylaxis (MB: adjusted HR 6.63, 95% CI 3.62-12.15; CRNMB: adjusted HR 3.69, 95% CI 2.04-6.71). Subgroup analysis indicated that LMWH thromboprophylaxis significantly increased MB risk in elderly patients with high bleeding risk (14 days: adjusted HR 5.45, 95% CI 1.15-25.94; 30 days: adjusted HR 4.16, 95% CI 1.11-15.53).
Conclusions: The IMPROVE BRS effectively predicted MB risk in COVID-19 patients and provided valuable guidance for LMWH thromboprophylaxis in elderly patients. Further research is needed to validate its applicability in different populations and refine threshold values for improved predictive accuracy.
Keywords: COVID-19; bleeding Risk; clinical Prediction Model; external Validation.
© 2025. The Author(s), under exclusive licence to Society of General Internal Medicine.