Pharmacokinetic Modeling of Implantable Pumps

Abstract

Implantable drug infusion pumps deliver medications directly to target compartments -- intrathecal space for pain and spasticity management, intraperitoneal cavity for chemotherapy, and hepatic artery for liver-directed therapy -- at precisely controlled rates that produce pharmacokinetic profiles fundamentally different from systemic administration. The local delivery paradigm creates steep concentration gradients between the target compartment (high) and systemic circulation (low), achieving therapeutic concentrations at the target while minimising systemic exposure and associated adverse effects. This study presents the Implantable Pump Pharmacokinetic Modelling Framework (IPPKMF), evaluating five PK modelling approaches -- classical compartmental models, physiologically based pharmacokinetic (PBPK) models, population PK with nonlinear mixed-effects, computational fluid dynamics (CFD)-integrated distribution models, and machine learning-augmented PK prediction -- across four implantable pump applications: intrathecal baclofen for spasticity, intrathecal morphine for chronic pain, intraperitoneal chemotherapy, and hepatic arterial infusion. Our PK Modelling Performance Score (PKPS) integrates prediction accuracy, physiological plausibility, individual prediction capability, dose optimisation utility, and clinical decision support value. ML-augmented PK prediction achieved the highest PKPS (0.926) through neural ODE models that learned patient-specific PK parameters from sparse clinical sampling with 94% prediction accuracy for CSF drug concentrations, while CFD-integrated models achieved the highest spatial distribution accuracy (0.958) by simulating three-dimensional drug transport within anatomically realistic CSF and peritoneal geometries