Cohort 2 student Harry Emerson publishes in the Journal of Biomedical Informatics entitled ‘Offline reinforcement learning for safer blood glucose control in people with type 1 diabetes’.
Harry had the following to say about his publication: ‘The publication explores how machine learning algorithms can be used to improve insulin dosing decisions for people with type 1 diabetes. Artificial pancreas devices have shown great success in reducing the burden of diabetes management, but rely on simplistic and reactive control algorithms. This work applies offline reinforcement learning as a method for learning sophisticated and safe strategies from pre-collected patient data. The method is verified in a simulation of 30 people and explores practical challenges, such as human error, device malfunction and data quality. The presented method significantly improved blood glucose control compared to current state-of-the-art control algorithms and was shown to be more robust than previous reinforcement learning approaches to the constraints of real-world data. The algorithm demonstrated the greatest benefit in children, which represent a particularly important group as they are often unable to manage their diabetes without assistance.
I implemented a selection of reinforcement learning algorithms in the type 1 diabetes simulator and created a full pipeline to perform data collection, training and evaluation. I modified the established UVA/Padova simulator to incorporate common blood glucose scenarios in which to evaluate the approach.
This represents the first demonstration of the benefits of offline reinforcement learning in blood glucose control. This work provides a basis for continued reinforcement learning research, demonstrating the potential of the approach to improve the health outcomes of people with type 1 diabetes, while highlighting the method’s shortcomings and areas of necessary future development. The publication was picked up by Wired Science, who write an article about myself and my research https://www.wired.com/story/managing-type-1-diabetes-is-tricky-can-ai-help/.
Link to Paper: https://www.sciencedirect.com/science/article/pii/S1532046423000977