Research progress on bleeding risk assessment models in anticoagulant therapy

Balancing thromboembolic prevention against bleeding complications remains a critical challenge in anticoagulant therapy. While established bleeding risk assessment models (RAMs) such as HAS-BLED and HEMORR 2 HAGES were initially developed for warfarin-treated patients, their applicability to non-vitamin K antagonist oral anticoagulant (NOAC) users and venous thromboembolism (VTE) populations remained uncertain. This review synthesized recent advancements in bleeding risk stratification for atrial fibrillation (AF) and VTE patients, focusing on model performance, drug-specific adaptations, and emerging biomarker-driven approaches. For AF patients, traditional RAMs (HAS-BLED, HEMORR 2 HAGES, ATRIA) demonstrated moderate predictive accuracy (AUC: 0.55–0.74) in NOAC cohorts, with HEMORR 2 HAGES showing superior discrimination in certain studies. The biomarker-integrated ABC (incorporating GDF-15, troponin, hemoglobin) and the NOAC-specific DOAC score, have shown improved risk stratification, with the latter demonstrating a higher C-statistic than HAS-BLED. In VTE populations, the IMPROVE (AUC: 0.62–0.73) effectively identified high-risk medical inpatients, while the RIETE (major bleeding incidence: 0.1%–6.2%) and EINSTEIN (C-statistic: 0.68–0.74) addressed dynamic risks during anticoagulation. However, heterogeneity in validation cohorts, endpoint definitions (e.g., ISTH vs. TIMI criteria), and static risk factor selections limited cross-model generalizability. Current RAMs exhibited variable performance across anticoagulant regimens and clinical contexts highlighting the need for next-generation models that integrate dynamic risk modifiers (e.g., transient anemia, antiplatelet use) and biomarker-based approaches. While NOAC-specific tools such as the DOAC may be more appropriate for AF patients, context-adapted models like IMPROVE and RIETE are better suited for VTE populations. Future research should prioritize real-world validation, machine learning integration, and the standardization of bleeding definitions to advance precision anticoagulation strategies.