The application of Model Predictive Control to normalize glycemia of critically ill patients

In this paper we propose a Model based Predictive Controller (MPC) to be used for glycemia control in critically ill patients. A model, that is particularly developed for describing the glucose and the insulin dynamics of these patients, is estimated for each individual patient and re-estimated as new measurements are obtained. Both a quantitative and a qualitative analysis are performed with respect to a real-life dataset from 19 critically ill patients. In the first analysis the robustness of the MPC is tested assuming a once per hour or a once per four hours insulin adaptation frequency is imposed. The second analysis is characterized by a comparison between the MPC insulin infusion sequence and the insulin flows (determined by the nurse) that were effectively administered to the patient. The contribution of this paper is the development of an MPC for glycemia control in the Intensive Care Unit (ICU). The penalty index, which is a specific concept for quantitative analysis of glycemia control in the ICU, is also proposed. The results of the developed MPC are satisfactory both in terms of control behavior (reference tracking and the suppression of unknown disturbance factors) and clinical acceptability.