Insulin Absorption Assessment Based on Bioimpedance Measurements: Model optimization for Improving Accuracy in Clinical Application

A model selection procedure to improve the accuracy in bioimpedance-based insulin absorption assessment is presented. An experimental campaign was implemented on a healthy subject to collect bioimpedance data during the controlled administration of insulin vehicle. Different regression models were statistically compared in the plane of percentage impedance variations and amount of vehicle injected. The residual standard error, the R-squared, the adjusted R-squared, and the ANOVA F-test were used to assess the goodness of fitting. The second-order polynomial model resulted the optimal model by considering both the goodness of fitting and the model complexity. With respect to the linear model, the second-order model guarantees an accuracy below the typical volume of an insulin unit of 10 $\mu 1$.