A Novel Risk Prediction Score for Clinically Significant Bleeding in Patients Anticoagulated for Venous Thromboembolism with Active Cancer

Abstract

Background Cancer-associated venous thromboembolism (Ca-VTE) treatment with anticoagulation is associated with bleeding complications and there are limited data on risk factors. Current models do not provide accurate bleeding risk prediction. Methods UK Clinical Practice Research Datalink data (2008–2020) were used to generate a cohort of patients with anticoagulant initiation for first Ca-VTE. Patients were observed up to 180 days for significant bleeding including major bleeding and clinically relevant nonmajor bleeding requiring hospitalization (CRNMB-H). A scoring scheme was developed from sub-distribution hazard ratios, and its discrimination (expressed by the C-statistic) estimated from cross-validation. Results A total of 15,749 patients with Ca-VTE and anticoagulant treatment were included. In total, 537 significant bleeding events, 161 major bleeds, and 376 CRNMB-H were identified after adjudicated review in 4,914 person-years of observation. Incidence rates of 3.3 and 7.7 per 100 person-years were noted for major bleeding and CRNMB-H. Independent predictors of significant bleeding included cancer of the bladder, central nervous system, cervix, kidney, melanoma, prostate and upper gastrointestinal tract, metastases, minor surgery, minor trauma, and history of major bleeding or CRNMB (before or after the Ca-VTE diagnosis). Patients recognized as low, medium, and high risk (30.4, 56.8, and 1.7% of the population, respectively) had a 6-month significant bleeding incidence rate of 5.1, 19.0, and 56.5 per 100 person-years, respectively. Overall C-statistic for significant bleeding was 0.70 (95% confidence interval: 0.65–0.75), and 0.76 (0.68–0.84) and 0.67 (0.61–0.73) for major bleeding and for CRNMB-H, respectively. Conclusion This risk score may identify patients at risk of significant bleeding, while also helping to determine treatment duration.