Optimal Experimental Design to Estimate Insulin Response in Type 2 Diabetes

Abstract

In late-stage type 2 diabetes, automated titration algorithms provide a promising alternative to the current standard-of-care. Many published methods rely on personalized dose-response models to predict a safe and effective insulin dose. In this case study, we address the challenge of how to collect an informative data set to ensure practical identifiability of such models. We apply optimal experimental design to enhance the performance of a published titration algorithm. For a 24-hour experiment, we solve an optimization problem to select the size of three meals and the hourly fast-acting insulin infusion rate. In simulation, we demonstrate how the optimized protocol improves the safety of the algorithm's dose-predictions. The results indicate that optimal experimental design has the potential to improve model-based algorithms and may be used as a qualitative tool when planning clinical experiments.