Nonlinear modeling of glucose metabolism: comparison of parametric vs. nonparametric methods

This paper presents the results of computational studies that compare simulated parametric and nonparametric models in terms of their ability to obtain reliable quantitative descriptions of the dynamic effects of variable infusions of insulin on blood glucose concentration in human subjects. In the nonparametric modeling approach, we employ the general class of Volterra-type models that are estimated from input-output data. The parametric models considered are the extensively studied "minimal model" and an augmented variant of the latter that incorporates the process of insulin secretion by the pancreas in response to elevated blood glucose. This model represents the actual closed-loop operating conditions of the system. The presented results demonstrate the feasibility of obtaining data-driven (i.e. inductive) nonparametric models in a realistic operating context, without resorting to the restrictive prior assumptions of model structure that are necessary for the numerous parametric (compartmental) models proposed previously. The rationale underpinning the nonparametric approach is that prior assumptions regarding the model structure may lead to results that are improperly constrained or biased by preconceived notions. Thus, it may be preferable to let the data guide the inductive selection of the appropriate model within the general class of Volterra-type models that imposes no such constraints.