Deviation Analysis and Interval Modeling as Complementary Tools to Evaluate Insulin Dosing Algorithms in Diabetes

To test newly developed insulin dosing algorithms for type 1 diabetes (T1D) patients, researchers and engineers often use simulation studies with complex physiological models of the human glucose metabolism. However, since those models typically do not include any time-variant behavior, simulation studies can easily lead to an overestimation of control performance. In order to have an estimate of the effect of intrapatient variability on glucose levels, so called “Deviation Analyses” have recently been proposed. These combine real recorded data as a baseline with models of insulin action in order to predict the effect of dosing algorithms. Therefore, they include the intrapatient variability of the recorded glucose traces. As an alternative, the information about intrapatient variability can be stored in a parameter interval of so-called interval models and can be used to compute the envelope of possible glucose trajectories. The current paper proposes to use Deviation Analysis and interval models as complementary tools to enhance the performance assessment of insulin dosing algorithms under intrapatient variability: Interval models can be used to detect unrealistic Deviation Analysis trajectories or, alternatively, unconservative and overly conservative interval models can be searched for using Deviation Analysis results.