A Performance-Based Adaptation Index for Automated Insulin Delivery Systems.

Area of Science:

• Biomedical Engineering
• Endocrinology
• Diabetes Technology

Background:

• Automated insulin delivery (AID) systems require personalized tuning for optimal performance.
• Individual insulin needs vary, necessitating adaptive algorithms.
• Continuous glucose monitoring (CGM) provides key data for algorithm adjustment.

Purpose of the Study:

• To introduce the Performance-Based Adaptation Index (PAI) for automatic adjustment of AID system aggressiveness.
• To enable AID systems to adapt to individual user needs and varying glycemic conditions.
• To improve glycemic outcomes in type 1 diabetes management.

Main Methods:

• Developed PAI integrating hypoglycemia and hyperglycemia CGM metrics into a single index.
• Proposed two computation methods for PAI: time in range (TIR) and glycemic risk indices.
• Assessed PAI feasibility in-silico using the UVA/Padova Type 1 Diabetes Simulator with a model predictive control (MPC) algorithm.

Main Results:

• PAI adjustment significantly improved Time in Range (TIR) from 35.1% to 71.8% in a conservative scenario.
• PAI reduced hypoglycemia exposure (TBR from 2.6% to 1.4%) in an aggressive scenario.
• PAI demonstrated safe and effective automatic tuning of the UVA-MPC controller in simulations.

Conclusions:

• PAI enables automatic tuning of AID controllers for improved glycemic control.
• Simulations show PAI can achieve TIR >70% in various physiological states.
• Clinical validation is recommended to confirm in-silico findings.