Integrating Neural Differential Forecasting with Safe Reinforcement Learning for Blood Glucose Regulation

TL;DR

This paper introduces TSODE, a novel safety-aware reinforcement learning controller that combines NeuralODE forecasting with uncertainty calibration to improve safe and personalized blood glucose regulation in Type 1 Diabetes patients.

Contribution

The paper presents TSODE, integrating NeuralODE-based glucose prediction with Thompson Sampling RL and conformal calibration for safety, a novel approach in diabetes management.

Findings

Achieved 87.9% time-in-range in simulations.

Reduced time below 70 mg/dL to less than 10%.

Outperformed existing baselines in safety and effectiveness.

Abstract

Automated insulin delivery for Type 1 Diabetes must balance glucose control and safety under uncertain meals and physiological variability. While reinforcement learning (RL) enables adaptive personalization, existing approaches struggle to simultaneously guarantee safety, leaving a gap in achieving both personalized and risk-aware glucose control, such as overdosing before meals or stacking corrections. To bridge this gap, we propose TSODE, a safety-aware controller that integrates Thompson Sampling RL with a Neural Ordinary Differential Equation (NeuralODE) forecaster to address this challenge. Specifically, the NeuralODE predicts short-term glucose trajectories conditioned on proposed insulin doses, while a conformal calibration layer quantifies predictive uncertainty to reject or scale risky actions. In the FDA-approved UVa/Padova simulator (adult cohort), TSODE achieved 87.9% time-in-range with less than 10% time below 70 mg/dL, outperforming relevant baselines. These results demonstrate that integrating adaptive RL with calibrated NeuralODE forecasting enables interpretable, safe, and robust glucose regulation.