Experimental evaluation of a recursive model identification technique for type 1 diabetes

Area of Science:

• Biomedical Engineering
• Control Systems
• Endocrinology

Background:

• Artificial beta cell controllers require accurate glucose-insulin models for type 1 diabetes.
• Adaptive models are crucial for controller robustness against changing physiological conditions.
• Recursive parameter estimation enables model adaptation to new conditions for improved control accuracy.

Purpose of the Study:

• To compare the predictive accuracy of recursive and nonrecursive (batch) autoregressive exogenous input (ARX) models for glucose-insulin dynamics.
• To evaluate model performance under normal and reduced insulin sensitivity conditions.
• To assess the utility of these models for model-based artificial beta cell controllers.

Main Methods:

• Retrospective analysis of glucose-insulin data from nine type 1 diabetes subjects.
• Identification of empirical dynamic autoregressive exogenous input (ARX) models using batch and recursive techniques.
• Validation of model predictions against test data, including data from induced reduced insulin sensitivity (prednisone administration).
• Comparison of ARX model predictions with zero-order hold (ZOH) model-free predictions.

Main Results:

• Both batch and recursive ARX models demonstrated comparable prediction accuracy for normal and reduced insulin sensitivity data.
• ARX models provided marginally more accurate predictions than ZOH predictions across various prediction horizons (30-90 minutes).
• For reduced insulin sensitivity, batch ARX models showed 5-9% better accuracy than ZOH, while recursive models showed 2-10% improvement.

Conclusions:

• Recursively identified ARX models did not outperform batch models in predicting glucose-insulin dynamics, even during altered insulin sensitivity.
• Both ARX model types offered only marginal improvements over the model-free ZOH predictions.
• The simplicity and computational efficiency of ARX models may justify their use in model-based artificial beta cell controllers despite modest accuracy gains.