Closed-loop glycemic control for critically ill subjects based on data-driven model predictive control

Hyperglycemia is a frequent and serious issue in the intensive care units (ICU), which can result in negative outcomes or even death. Closed-loop glycemic control is a promising direction to deal with this issue. Through reducing the blood glucose level, negative outcomes and even mortality can be minimized. As a closed-loop control method, model predictive control (MPC) performs well in glycemic control due to its super ability of dealing with constraints and time delays. However, conventional MPC encounters difficulties when it is used in the ICU, because the individualized model of an ICU patient is usually unknown. Therefore, an online subspace identification method (SIM) was used to identify one subject´s individualized model; based on this model, MPC was implemented to design the insulin delivery rate automatically. This combination is termed as a SIM-based model predictive control (SIM-MPC) method, categorized as a data-driven control method. The effectiveness and robustness of the SIM-MPC method have been validated by using some simulation tests.