AI-Based Personalized Warfarin Dose Optimization
- Prelegent(ci)
- Ali Hashemi Baghi
- Afiliacja
- Ph.D. Applicant at the MIMUW
- Język referatu
- angielski
- Termin
- 18 listopada 2026 14:15
- Odnośnik
- https://uw-edu-pl.zoom.us/j/95979582337?pwd=XvPqZ9IWQfyFvd1h1pWVU4w9wOmj0r.1
- Seminarium
- Seminarium Zakładu Biomatematyki i Teorii Gier
Warfarin is a widely used oral anticoagulant for the prevention and treatment of thromboembolic disorders. However, dose optimization is challenging due to its narrow therapeutic range and substantial interindividual variability. Underdosing may increase thrombotic risk, whereas excessive anticoagulation can lead to serious bleeding. In this study, the problem of personalized warfarin dose optimization was first formulated as a constrained multi-objective optimization model, aiming to balance bleeding and thrombotic risks while satisfying clinical constraints. Subsequently, artificial neural networks (ANNs) were developed to predict bleeding risk, thrombotic risk, and international normalized ratio (INR) response using patient-specific clinical and laboratory data. Model interpretability was further investigated using SHapley Additive exPlanations (SHAP), partial dependence plots (PDPs), and individual conditional expectation (ICE) plots to examine feature effects at both population and individual levels. Finally, the ANN predictions were integrated into the optimization framework to identify personalized dose candidates. This approach aims to support safer and more individualized warfarin therapy.
Nie jesteś zalogowany |