Machine Learning for Warfarin Therapy: A Systematic Review.

Area of Science:

• Pharmacology
• Artificial Intelligence
• Clinical Informatics

Background:

• Warfarin remains crucial for specific patient groups despite newer anticoagulants.
• Traditional warfarin dosing has suboptimal therapeutic range achievement (55-65%).
• Suboptimal dosing increases risks of bleeding and thrombotic events.

Purpose of the Study:

• To systematically review machine learning (ML) approaches for warfarin dose prediction.
• To evaluate the performance and limitations of ML algorithms in warfarin dosing.

Main Methods:

• Systematic review of 14 studies (122,400 patients) from 2022-2025.
• Inclusion of studies using ML for warfarin dosing with performance metrics.
• Risk of bias assessment using PROBAST.

Main Results:

• Reinforcement learning showed superior performance in achieving therapeutic ranges.
• ML models consistently outperformed traditional methods in accuracy metrics (e.g., Mean Absolute Error).
• Significant limitations include a predominance of retrospective studies and lack of safety outcome data.

Conclusions:

• ML algorithms demonstrate potential for improved warfarin dosing accuracy.
• Absence of robust safety outcome data (bleeding, thromboembolism) is a major barrier to clinical implementation.
• Multicenter prospective trials with safety endpoints and diverse population validation are essential for adoption.