JMIR Publications

ABSTRACT

Background:

Effective diabetes management requires precise glycemic control to prevent both hypoglycemia and hyperglycemia, yet existing machine learning (ML) and reinforcement learning (RL) approaches often fail to balance multiple competing objectives. Traditional RL-based glucose regulation systems primarily focus on single-objective optimization, overlooking critical factors such as minimizing insulin overuse, reducing glycemic variability, and ensuring patient safety. Furthermore, these approaches typically rely on centralized data processing, raising significant privacy concerns due to the sensitive nature of healthcare data. There is a critical need for a decentralized, privacy-preserving framework that can personalize blood glucose regulation while addressing the multi-objective nature of diabetes management.

Objective:

This study proposes PRIMO-FRL (Privacy-preserving, Reinforcement learning for Individualized Multi-Objective diabetes management using Federated Reinforcement Learning), a novel framework that optimizes multiple clinical objectives—maximizing Time in Range (TIR), reducing hypoglycemia and hyperglycemia, and minimizing glycemic risk—while preserving patient privacy.

Methods:

We developed PRIMO-FRL, integrating multi-objective reward shaping to dynamically balance glucose stability, insulin efficiency, and risk reduction. The model was trained using simulated patient data generated from the FDA-approved UVA/Padova simulator, covering diverse patient cohorts (children, adolescents, and adults). A comparative analysis was conducted against state-of-the-art RL and ML models, evaluating performance using key metrics such as TIR, hypoglycemia (<70 mg/dL), hyperglycemia (>180 mg/dL), and glycemic risk scores.

Results:

The PRIMO-FRL model achieved a robust overall TIR of 76.54%, with adults demonstrating the highest TIR (81.48%), followed by children (77.78%) and adolescents (70.37%). Importantly, the approach entirely eliminated hypoglycemia (0.0% time spent below 70 mg/dL) across all cohorts, significantly outperforming existing methods. Mild hyperglycemia (180–250 mg/dL) was observed in adolescents (29.63%) and children (22.22%), while adults exhibited the best control (18.52%). Furthermore, the PRIMO-FRL approach consistently reduced glycemic risk scores, demonstrating improved safety and long-term stability in glucose regulation.

Conclusions:

Our findings highlight the potential of PRIMO-FRL as a transformative, privacy-preserving approach to personalized glycemic management. By integrating FRL, this framework not only eliminates hypoglycemia and improves Time in Range (TIR) but also ensures patient data privacy by decentralizing model training. Unlike traditional centralized approaches that require sharing sensitive health data, PRIMO-FRL leverages federated learning to keep patient data local, significantly reducing privacy risks while enabling adaptive and personalized glucose control. This multi-objective optimization strategy offers a scalable, secure, and clinically viable solution for real-world diabetes care. The ability to train personalized models across diverse patient populations without exposing raw medical data makes PRIMO-FRL well-suited for deployment in privacy-sensitive healthcare environments. These results pave the way for future clinical adoption, demonstrating the potential of privacy-preserving AI in optimizing glycemic regulation while maintaining security, adaptability, and personalization.