Comparison of model Predictive control (MPC) algorithms to optimise blood glucose in fully closed loop (FCL) systems

Area of Science:

• Endocrinology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Fully closed-loop (FCL) systems represent an advancement in automating glucose regulation for Type 1 Diabetes.
• Model Predictive Control (MPC) is an emerging technology within FCL systems.

Purpose of the Study:

• To assess the clinical effectiveness of FCL systems.
• To explore future optimizations by comparing recent FCL system developments.

Main Methods:

• Comparison of MPC-based FCL systems with Proportional-Integral-Derivative (PID) controlled hybrid closed-loop (HCL) models.
• Analysis of three emerging FCL advancements: nonlinear MPC (NMPC), λ-Policy Iteration (λ-PI), and pulse-modulated artificial pancreas (PMCL) systems.
• Proposal of a novel hybrid model integrating benefits from emerging algorithms.

Main Results:

• MPC-based FCL systems demonstrated superior time-in-range (TIR) compared to PID-HCL (74.4% vs. 63.7%, P = 0.020).
• Key challenges remain, including postprandial hyperglycaemia and insulin absorption delays, with no system consistently exceeding the 70% TIR target.
• Emerging advancements include NMPC for dual-hormone systems, λ-PI for adaptive learning, and PMCL systems mimicking natural insulin secretion.

Conclusions:

• Current innovations show promise in silico but lack clinical validation.
• Barriers to clinical adoption include glucagon instability, CGM inaccuracies, cost, and patient adherence.
• Future research should focus on long-term trials addressing real-world factors and integrating predictive control, adaptive learning, and dual-hormone regulation for improved diabetes management.