Privacy-Preserving Glycemic Management in Type 1 Diabetes: Development and Validation of a Multiobjective Federated Reinforcement Learning Framework | JoVE Visualize

Area of Science:

Artificial Intelligence in Healthcare
Machine Learning for Diabetes Management
Reinforcement Learning for Glycemic Control

Background:

Effective diabetes management necessitates precise glycemic control to prevent hypoglycemia and hyperglycemia.
Existing machine learning (ML) and reinforcement learning (RL) methods often struggle to balance competing objectives like minimizing insulin use and reducing glycemic variability.
Centralized data processing in current systems raises privacy concerns for sensitive health data, highlighting the need for decentralized, privacy-preserving solutions.

Purpose of the Study:

To develop and validate PRIMO-FRL (Privacy-Preserving Reinforcement Learning for Individualized Multi-Objective Glycemic Management Using Federated Reinforcement Learning).
To optimize clinical objectives including maximizing time in range (TIR), reducing hypoglycemia and hyperglycemia, and minimizing glycemic risk.
To ensure patient privacy through a novel federated reinforcement learning framework.

Main Methods:

Developed PRIMO-FRL, incorporating multiobjective reward shaping for dynamic balancing of glucose stability, insulin efficiency, and risk reduction.
Trained and tested the model using simulated data from 30 patients (children, adolescents, adults) via the FDA-approved UVA/Padova simulator.
Conducted comparative analysis against state-of-the-art RL and ML models using metrics like TIR, hypoglycemia, hyperglycemia, and glycemic risk scores.

Main Results:

PRIMO-FRL achieved a 76.54% overall TIR, with adults showing the highest at 81.48%.
The framework successfully eliminated hypoglycemia (0.0% below 70 mg/dL) across all age groups, significantly outperforming existing methods.
Adults exhibited the best control over mild hyperglycemia (18.52%), and the approach consistently reduced glycemic risk scores, indicating improved safety.

Conclusions:

PRIMO-FRL demonstrates transformative potential as a privacy-preserving approach to personalized glycemic management using federated reinforcement learning.
The framework eliminates hypoglycemia, improves TIR, and preserves data privacy by decentralizing model training, addressing limitations of centralized systems.
PRIMO-FRL offers a scalable, secure, and clinically viable solution for real-world diabetes care, paving the way for privacy-preserving AI in sensitive healthcare environments.