Abstract
Hemophilia A is a rare inherited bleeding disorder typically managed with coagulation factor VIII (FVIII) replacement therapy. Designing personalized prophylactic regimens requires accurate pharmacokinetic (PK) characterization; current population PK (popPK) and Bayesian approaches provide a principled framework for individualized dosing, but their routine clinical implementation may still be constrained by model specification requirements and practical considerations in data collection and analysis. Here we present a machine learning (ML) framework, incorporating state-of-the-art language models, to predict individual FVIII PK parameters in pediatric patients. Using minimal sampling and routinely collected clinical data, our approach achieves superior performance over the widely adopted WAPPS-Hemo platform in predicting in vivo recovery (IVR) and FVIII half-life. These findings highlight the potential of AI-driven methods to reduce patient burden while improving accuracy in individualized treatment planning for children with severe hemophilia A.