Nonlinear dynamic modeling and model predictive control of thrombin generation to treat trauma‐induced coagulopathy

Abstract

This article is motivated by the pressing need to robustly automate clinical interventions for trauma‐induced coagulopathy (TIC). TIC occurs after severe trauma and shock, and has poor outcomes and about 30% mortality. Although modulating the blood proteins that drive TIC can improve patient outcomes, no practical control‐oriented methodology exists to accurately capture biochemical process dynamics and satisfactorily regulate clotting. Hence, we propose a nonlinear dynamic coagulation model that distills the complex biochemical reactions of clotting and also simultaneously generalizes an existing linear phenomenological model. Using our new nonlinear model, we demonstrate the feasibility of model predictive control (MPC) to automate clinical treatments, first in a single‐input case that is similar to current open‐loop clinical practice, and second in a multi‐input case that administers three blood proteins as system inputs to attain satisfactory TIC treatment. The output in both cases is the key clotting protein thrombin. To test robustness, we confirm that both single‐input and multi‐input MPC are suitable for TIC treatment in the presence of an experimentally observed nonlinearity, an unknown state‐dependent power law input delay. Thus, this article provides a strong foundation to transition current open‐loop clinical approaches to closed‐loop process control.