Machine Learning Models for Predicting Bleeding Risk in Anticoagulated Patients with Atrial Fibrillation and Venous...

Area of Science:

• Cardiology
• Medical Informatics
• Data Science in Medicine

Background:

• Predicting bleeding events in patients on oral anticoagulants for atrial fibrillation (AF) and venous thromboembolism (VTE) is crucial.
• Machine learning (ML) offers potential for more accurate risk prediction due to its ability to model complex interactions.

Purpose of the Study:

• To systematically review and synthesize evidence comparing ML-based bleeding risk models against traditional clinical scores.
• Focus on anticoagulated AF and VTE patient populations.

Main Methods:

• Systematic review and narrative synthesis of studies published between 2015 and 2025.
• Included studies applied ML algorithms to predict bleeding events in anticoagulated AF or VTE patients.
• Analyzed data from 13 studies encompassing 464,523 participants.

Main Results:

• ML algorithms (e.g., random forest, XGBoost, neural networks) generally outperformed traditional scores like HAS-BLED and RIETE/VTE-BLEED.
• ML models showed higher AUCs in both AF (0.64-0.76 vs. 0.52-0.61) and VTE (0.59-0.91 vs. 0.61-0.65) populations.
• Deep learning ensembles achieved the highest predictive performance (AUCs >0.8).

Conclusions:

• ML-based bleeding risk models demonstrate statistically superior discrimination compared to established scores.
• Observed improvements in discrimination were modest (ΔAUC 0.05-0.15), and clinical utility is not yet established.
• Further validation, calibration assessment, and impact studies are needed for routine clinical adoption.