Acute Inflammation Treatment via Particle Filter State Estimation and MPC

Models of acute inflammatory disease may have the potential to guide treatment decisions in critically ill patients. Model Predictive Control (MPC) leverages the predictive capacity of a model to determine a control strategy that guides a system to a target trajectory. As applied to acute inflammation, MPC might be used to guide a patient from disease to health by monitoring the patient state, computing and applying an optimal intervention strategy, and updating the strategy if the patient state diverges from predictions. A key challenge to the application of MPC is mapping the observable patient state into the complete state space of the model. We propose that a Particle Filter (PF) is a suitable algorithm for state estimation in nonlinear models of acute inflammation. As a proof of concept, we apply MPC and PF to the administration of hemoadsorption (HA) treatment in an 8-state model of endotoxemia in rats. In silico tests demonstrate that the PF generates accurate state estimates from limited observations in the presence of noise and parameter uncertainty. Furthermore, we explore the maximal predicted benefits of HA treatment with a standard single column configuration and hypothetical multi-column configurations, where each column has a specificity for a target cytokine. Simulations suggest that two column HA will improve treatment efficacy, but physiological restrictions on HA will limit benefits from higher order configurations.