Precision Medicine in Diabetes Using Continuous Glucose Monitoring

Abstract

Diabetes affects millions of individuals around the world and is a leading cause of death. The risk of serious long-term complications in diabetes can be mitigated through early interventions in the form of medication and behavioral changes. However, the pathophysiology of diabetes and the course of the disease are markedly heterogeneous, making it essential that disease management is tailored to the individual. Continuous glucose monitoring (CGM) helps patients manage their disease through the collection of real-time measurements of interstitial glucose, providing insight into glycemic dynamics that laboratory measurements cannot capture. In this thesis, we investigate how CGM can be used to enable personalized disease management in diabetes using modern methods from machine learning and signal processing. We first investigate a model-based approach to estimate metabolic parameters from CGM data. We show that parameters estimated from daily CGM data correlate with parameters derived from in-clinic laboratory measurements. Then, we explore how the rapidly emerging field of generative artificial intelligence can be integrated into diabetes care through analysis of CGM data. We show how large language model agents hold promising potential to assist patients and clinicians in managing diabetes through the synthesis and narrative summarization of large amounts of CGM data. Finally, we leverage observational CGM data to understand heterogeneity in type 2 diabetes. The work in this thesis shows how modern computational methods in machine learning can enable precision medicine in diabetes by leveraging wearable CGM data for improved disease management.