Model predictive control with sub-frequency actuation for long acting insulin treatment in type 2 diabetes

Type 2 diabetes (T2D) accounts for 90% of the diabetes patients, and more than 60% of patients on insulin fail to reach treatment targets. The main reasons include low adherence to treatment, caused by, among other reasons, complexity and fear of overdosing. In insulin titration, pre-breakfast SMBG measurements are used for dose calculations. However, glucose values may be lower between the daily SMBG measurements. Not considering this may lead to over-titration. We propose a dose guidance algorithm to identify individualized optimal dosing of long acting insulin for T2D patients. We use a compartment model of fasting glucose and long acting insulin dynamics in T2D to develop a model predictive control (MPC) based decision support system. To promote safety, the controller optimizes the daily dose based on predicted fasting glucose values with faster sampling than available data allows. We test the performance of the controller with respect to safety and efficacy in a simplified in silico environment and compare the results with standard of care dose guidance. The aim of this paper is to illustrate the importance of fast sampling in prediction between the slow input samples. This importance is observable in the results, which are indicative of the potential of such an approach.