Challenging common bolus advisor for self-monitoring type-I diabetes patients using Reinforcement Learning

TL;DR

This paper applies Reinforcement Learning to optimize insulin dosing for self-monitoring type-I diabetes patients, demonstrating that learned policies differ from standard advice and can reduce hypoglycemia.

Contribution

It introduces a Reinforcement Learning approach to derive an optimal bolus rule, challenging the existing standard and showing potential improvements in patient safety.

Findings

Reinforcement Learning-derived bolus rules differ from standard advice.

Optimized policies can prevent hypoglycemia episodes.

Simulation results suggest improved insulin management.

Abstract

Patients with diabetes who are self-monitoring have to decide right before each meal how much insulin they should take. A standard bolus advisor exists, but has never actually been proven to be optimal in any sense. We challenged this rule applying Reinforcement Learning techniques on data simulated with T1DM, an FDA-approved simulator developed by Kovatchev et al. modeling the gluco-insulin interaction. Results show that the optimal bolus rule is fairly different from the standard bolus advisor, and if followed can actually avoid hypoglycemia episodes.