Control-relevant models for glucose control using a priori patient characteristics

Area of Science:

• Biomedical Engineering
• Control Systems
• Endocrinology

Background:

• Accurate glucose control in type 1 diabetes mellitus (T1DM) is hindered by patient-model mismatch in artificial pancreas systems.
• Existing model-based control algorithms struggle to prevent both postprandial hypoglycemia and hyperglycemia.

Purpose of the Study:

• To develop control-relevant mathematical models for T1DM that prioritize safety by minimizing hypoglycemia risk.
• To personalize these models using patient characteristics to account for inter-subject variability.
• To implement and evaluate these models within a zone model predictive control (ZMPC) algorithm.

Main Methods:

• Development of T1DM models focused on control relevance rather than prediction error minimization.
• Conservative parameter selection to reduce hypoglycemia likelihood.
• Personalization of models using a priori patient data.
• Implementation in a ZMPC algorithm and in silico robustness evaluation.

Main Results:

• Hypoglycemia was completely avoided in silico, even with significant meal disturbances.
• The developed control-relevant models demonstrated robustness in simulated environments.
• The proposed control approach proved effective in preventing dangerous glucose level drops.

Conclusions:

• Personalized, control-relevant models can enhance the safety and reliability of artificial pancreas systems for T1DM.
• The developed ZMPC algorithm offers a robust solution for glucose control, minimizing hypoglycemia risks.
• This approach simplifies controller setup for physicians, requiring no specialized control expertise.