Type 1 Diabetes Management using GLIMMER: Glucose Level Indicator Model with Modified Error Rate

• DOI: 10.48550/arxiv.2502.14183
• Publication Date: 2025-02-19
• Authors: Saman Khamesian; Asiful Arefeen; Adela Grando; Bithika Thompson; Hassan Ghasemzadeh
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Machine Learning (cs.LG)

Managing Type 1 Diabetes (T1D) demands constant vigilance as individuals strive to regulate their blood glucose levels avert the dangers of dysglycemia (hyperglycemia or hypoglycemia). Despite advent sophisticated technologies such automated insulin delivery (AID) systems, achieving optimal glycemic control remains a formidable task. AID systems integrate continuous subcutaneous infusion (CSII) and monitors (CGM) data, offering promise in reducing variability increasing time-in-range. However, these often fail prevent dysglycemia, partly due limitations prediction algorithms that lack precision abnormal events. This gap highlights need for proactive behavioral adjustments. We address this with GLIMMER, Glucose Level Indicator Model Modified Error Rate, machine learning approach forecasting levels. GLIMMER categorizes values into normal ranges devises novel custom loss function prioritize accuracy dysglycemic events where patient safety is critical. To evaluate potential T1D management, we both use publicly available dataset collect new data involving 25 patients T1D. In predicting next-hour values, achieved root mean square error (RMSE) 23.97 (+/-3.77) absolute (MAE) 15.83 (+/-2.09) mg/dL. These results reflect 23% improvement RMSE 31% MAE compared best-reported rates.