Is Neural-ODE an Effective Method for Modeling Pharmacokinetics (PK) Data in Real-world Clinical Settings?
This study analyzes the effectiveness of Neural-ODEs in pharmacokinetics (PK) modeling by benchmarking ODEs against other state-of-the-art approaches for time-series modeling. The study also evaluates ODEs in noisy and missing data simulations to mimic real-world settings.
Is Neural-ODE an Effective Method for Modeling Pharmacokinetics (PK) Data in Real-world Clinical Settings?
Recent publications have demonstrated that Neural ordinary differential equations (Neural-ODEs) are the most accurate models in predicting the PK of untested treatment regimens in patients. Our study aims to first validate these claims by benchmarking ODEs against other Deep Learning approaches. Our study then evalutes the performance of Neural-ODEs in simulations that mimic real-world clinical settings, characterized by noisy and missing data. Our findings indicate a critical threshold beyond which the model's performance significantly declines.
Creative Commons Attribution 4.0 International License (CC-BY 4.0)
