Improving minimal model identifiability in insulin resistant patients utilising insight from the graphical structural model identifiability method

Abstract

The Minimal Model is often used to characterize participant responses to glucose by fitting model simulations to measured data. Although the model has met significant success defining test responses of normo-glucose tolerant (NGT) participants, the model has experienced practical identifiability issues with insulin resistant (IR) individuals. A previous investigation of practical model identifiability hypothesized that Minimal Model identifiability of IR individuals would be improved if a bolus impulse response was followed by a sustained period of mild hypoglycemia. In this investigation, an in-silico Monte Carlo analysis is undertaken to observe the effect of incorporating a clamp-like period to the end of an intravenous glucose tolerance test (IVGTT-EIC) protocol. N=100 virtual patient responses to the IVGTT-EIC and frequently sampled IVGTT (FS-IVGTT) protocols are defined. Minimal Model parameter values are identified for each test includeing measurement error M=500 times. The robustness of model parameters from the two protocols is assessed via paired coefficient of variation (CV). The CV values for the Minimal Model parameters derived from the proposed IVGTT-EIC protocol were significantly lower than the CV values from the FS-IVGTT. In particular, median CV(%) for Minimal Model parameters SG, p2, p3 and VG derived from IVGTT-EIC data was: 3.8%, 16.8%, 8.7% and 2.2%, respectively, compared to 9.3%, 43.0%, 36.0% and 2.9% respectively for the FS-IVGTT. Minimal Model simulation of FS-IVGTT data is a well established method of measuring insulin sensitivity. However, the proposed IVGTT-EIC protocol significantly improved the identifiability of the Minimal Model over the FS-IVGTT. The success of the proposed protocol validates the utility of the graphical structural model identifiability analysis for defining robust clinical protocols.