Bleeding risk and performance of bleeding risk assessment models in patients with venous thromboembolism on anticoagulation: results from the prospective observational bleeding and assess long-term outcomes on health in patients with venous thromboembolism (BACH-VTE) study.

BACKGROUND

Patients with venous thromboembolism (VTE) are at risk of bleeding during anticoagulation.

OBJECTIVES

This study aimed to assess bleeding risk and the performance of risk assessment models (RAMs) VTE-PREDICT, HAS-BLED, RIETE, and VTE-BLEED in patients with acute VTE initiating anticoagulation.

METHODS

We used data from a prospective observational cohort study (BACH-VTE) including patients with acute VTE who initiated anticoagulation with a follow-up period of up to 2 years. Exclusion criteria were active cancer, pregnancy, and postpartum period. Major bleeding, clinically relevant nonmajor bleeding (CRNMB), and minor bleeding were recorded and their frequencies calculated. RAM performance was evaluated by discrimination and calibration. Predictors associated with clinically relevant bleeding (CRB; composite of major and CRNMB) were assessed.

RESULTS

In total, 308 patients (median age, 55 years; 42% women, 47% pulmonary embolism, 62% unprovoked VTE) were included. During a median follow-up time of 12.6 months, we observed 2 major bleedings, 41 CRNMBs, and 66 minor bleedings, corresponding to 2-year cumulative incidences (95% CI) of 0.9% (0%-2.1%), 16.2% (10.7%-21.3%), and 20.6% (15.1%-25.8%), respectively, and of 33.4% (26.7-39.5) for any bleeding. RAM discrimination was poor to moderate with C-statistics (95% CI) for CRB of 0.71 (0.61-0.80) for VTE-PREDICT, 0.59 (0.49-0.68) for HAS-BLED, 0.52 (0.41-0.62) for RIETE, and 0.56 (0.45-0.68) for VTE-BLEED. Calibration analysis revealed underestimation of bleeding risk. Female sex, lower hemoglobin, and bleeding history were associated with CRB in a univariable but not in a multivariable model.

CONCLUSION

In patients treated with anticoagulants for VTE, we found high rates of CRB. Only the VTE-PREDICT model showed acceptable discrimination, but poor calibration.