Data-driven Solutions for Blood Glucose Management

This is a hybrid event. The Zoom link is here.

ABSTRACT: Type 1 diabetes (T1D) affects millions of people worldwide. People with T1D must regularly monitor their blood glucose level and administer insulin to manage it. This process is burdensome, necessitating frequent measurements, meal size estimation, and bolus calculations. Automated management solutions have been proposed, but still require patients to accurately estimate meal sizes and manually update user parameters. In this thesis, we identify and address challenges in the utilization of data-driven approaches for blood glucose management. First, accurately estimating meal size is challenging for individuals, resulting in noisy carbohydrate counts and in turn poor blood glucose management. Second, while accurate blood glucose forecast models could obviate the need for manual patient updates via their capacity for online learning, current forecasting approaches fall short of the level of accuracy required for safe automated blood glucose management. More specifically, learning the independent effects of carbohydrates and insulin boluses on future blood glucose values is challenging due to the relative sparsity of these variables and the strong correlation between them in the training data. In light of these challenges, we propose novel approaches for 1) correcting noisy patient-reported carbohydrate estimates, 2) forecasting in the presence of sparsity, and 3) control-ready forecasting when inputs are strongly correlated in training data. Combined our work represents a clear step towards data-driven blood glucose management. While inspired by challenges in blood glucose management, the solutions proposed are also applicable to other domains in healthcare and beyond.