External validation of bleeding risk models for the prediction of long-term bleeding risk in patients with established cardiovascular disease

The long-term predictive performance of existing bleeding risk models in patients with various manifestations of cardiovascular disease (CVD) is not well known. This study aims to assess and compare the performance of relevant existing bleeding risk models in estimating the long-term risk of major bleeding in a cohort of patients with established CVD. Seven existing bleeding risk models (PRECISE-DAPT, DAPT, Ducrocq et al, de Vries et al, S2TOP-BLEED, Intracranial B2LEED3S and HAS-BLED) were identified and externally validated in 7,249 patients with established CVD included in the Utrecht Cardiovascular Cohort–second manifestations of arterial disease study. Predictive performance was assessed in terms of discrimination and calibration, both at 10 years and the original prediction horizon of the models. Major bleeding was defined as Bleeding Academic Research Consortium type 3 or 5. After a median follow-up of 8.4 years (interquartile range 4.5-12.5), a total of 233 (3.2%) major bleeding events occurred. C-statistics for discrimination at 10 years ranged from 0.53 (95%CI 0.49-0.57) to 0.64 (95%CI 0.60-0.68). Calibration plots after recalibration to 10 years showed best agreement between predicted and observed bleeding risk for De Vries et al, S2TOP-BLEED, DAPT and PRECISE-DAPT. The performance of existing bleeding risk models to predict long-term bleeding in patients with CVD varied. Discrimination and calibration were best for the models of de Vries et al, S2TOP-BLEED, DAPT and PRECISE-DAPT. Of these, recalibrated models requiring the least predictors may be preferred for use to personalize prevention with antithrombotic therapy.