A Data-Driven Approach to Classifying Daily Continuous Glucose Monitoring (CGM) Time Series

Area of Science:

• Biomedical Engineering
• Data Science
• Endocrinology

Background:

• Diabetes affects over 422 million globally, necessitating precise blood glucose monitoring.
• Continuous Glucose Monitoring (CGM) is increasingly vital for diabetes management in Type 1 (T1D) and Type 2 (T2D) patients.
• Current CGM use is rising, highlighting the need for efficient data analysis methods.

Purpose of the Study:

• To develop a data-driven method for identifying a finite set of representative daily glucose profiles (motifs).
• To enable accurate matching of individual patient CGM data to these representative motifs.
• To support advanced applications like predictive modeling and automated diabetes control systems.

Main Methods:

• Utilized a large dataset of daily CGM profiles from T1D and T2D patients across various treatment modes.
• Employed a data-driven approach to identify candidate sets of representative daily glucose profiles (motifs).
• Validated and selected a final set of motifs (Ω) using training and validation datasets, then tested on a large independent dataset.

Main Results:

• Identified 8 candidate sets of motifs from 9,741 training profiles.
• Selected the final set Ω using 14,175 validation profiles.
• Achieved 99.0% classification accuracy on 42,595 daily CGM profiles in the testing dataset.

Conclusions:

• The developed set of representative daily glucose profiles (Ω) robustly classifies CGM data.
• This classification enables enhanced predictive modeling, decision support, and automated control systems for diabetes.
• The findings support the advancement of personalized and automated diabetes management strategies.