Model identification using stochastic differential equation grey-box models in diabetes.

Area of Science:

• Biomedical modeling
• Computational physiology
• Systems biology

Background:

• Virtual preclinical testing requires robust models, but ordinary differential equation (ODE) models lack statistical validation.
• Stochastic differential equations (SDEs) enable statistically validated models with uncertainty prediction.
• SDEs separate prediction errors into uncorrelated noise terms, identifying model deficiencies.

Purpose of the Study:

• Develop a stochastic-differential-equation-based grey-box (SDE-GB) model for type 1 diabetes mellitus (T1DM) patient glucoregulation.
• Enhance model validation using statistical tools.
• Track parameter variations, specifically the "time to peak of meal response".

Main Methods:

• Utilized an identifiable glucoregulatory model for T1DM patients as a basis for the SDE-GB model.
• Estimated model parameters using clinical data from four T1DM patients.
• Determined the optimal SDE-GB model via likelihood-ratio tests and employed parameter tracking.

Main Results:

• Transformed ODE models into SDE-GB models significantly improved prediction accuracy and uncorrelated errors.
• Parameter tracking revealed meal-type-dependent variations in the "peak time of meal absorption" parameter.
• Demonstrated the efficacy of SDE-GBs in capturing dynamic physiological responses.

Conclusions:

• SDE-GBs offer a robust framework for diabetes modeling and control algorithm development.
• Separation of prediction error in SDE-GBs leads to enhanced model predictability.
• The study highlights the utility of SDE-GBs for statistical validation and pinpointing model deficiencies in physiological systems.