M.S. AAI Capstone Chronicles 2024

Introduction

Type 1 diabetes mellitus (T1D) is a global healthcare challenge, affecting 9.5% of the population worldwide (Mobasseri, 2020) and 0.4% in the United States (Fang, 2024). Management requires precise insulin dosing to avoid complications such as hypoglycemia, which impairs cognitive function, and hyperglycemia, which can result in diabetic ketoacidosis (Lizzo, 2023). While CGM and insulin pumps have improved glycemic control, predicting and preventing dangerous glucose fluctuations remains a critical unmet need. Long-term control, assessed through HbA1c levels, reduces risks of complications such as retinopathy and neuropathy, underscoring the importance of accurate glucose and A1c predictions (WHO, 2011). To address these challenges, this project integrates predictive modeling with dietary insights to support personalized medicine, empowering patients to make data-driven decisions about their health. The proposed system benefits T1D patients using CGMs and insulin pumps, as well as individuals with brittle or refractory type 2 diabetes (Freckman, 2020). Additionally, healthcare providers and researchers can leverage its insights to refine treatment strategies. This project is built upon three complementary datasets: a.​ DiaTrend Dataset (Eyth, 2023): Provides CGM and insulin pump data from 54 participants, offering real-world insights into diabetes management. b.​ Tidepool Dataset (Tidepool, 2024): Contains three months of CGM and insulin data along with demographic metadata, enabling real-time glycemic event prediction.

2

236