Models and Algorithms for Personalized Glycemic Control

Sammanfattning: Glycemic control is resource-demanding in the treatment of Intensive Care Units (ICUs) patients where blood glucose measurements are taken continuously at few-hour intervals to avoid the risk of having a patient in a coma, or in the worst case, causing death. The number of diabetes patients, mostly type 2 diabetes, is increasing around the world, and at the same time, we have an ongoing pandemic of Covid-19 with health workers stretched to their limits. Half of all ICU patients are treated with some form of glycemic control whereas many of the patients with diabetes do not have an adequate diagnosis which creates uncertainties in the treatment. An automated and personalized glycemic control will therefore contribute to patient safety and decrease the workload at ICUs. Each patient responds very differently to insulin due to stress and medication where researchers have identified an individual time-varying insulin sensitivity parameter to model the response of injected insulin. A recently conducted study in the area of glycemic control for patients with diabetes type 1 has developed an MPC (Model Predictive Control) algorithm based on virtual patient groups defined by the insulin sensitivity parameter. In this work, an MPC algorithm developed for five virtual patient groups based on data of insulin-sensitivity levels within ICUs included in a patient simulator is proposed. Three insulin glucose models suitable for an adaptive MPC were identified. The MPC algorithm is based on the ICING model which includes a time-varying insulin-sensitivity parameter. It was compared to an MPC algorithm based on the ICU Minimal Model which includes a constant insulin sensitivity parameter. Constraints of the input variables and a target range of the blood glucose concentration were identified. 50 patients in each virtual patient group were simulated with different values of the insulin sensitivity parameter. The simulations of the MPC algorithm based on the ICING model reached the target range for three of five patient groups where insulin sensitivity has a big impact on the patient’s blood glucose concentration. Improvements compared to a non-adaptive solution is a patient-specific controller that can handle time-varying insulin sensitivity and disturbances such as glucose given together with medication. This shows that the future of personalized glucose control together with continuous medication pumps can improve patient safety, and reduce stress and the workload at ICUs.