Finding Optimal Alphabet for Encoding Daily Continuous Glucose Monitoring Time Series Into Compressed Text

Area of Science:

• Biomedical Engineering
• Data Science
• Endocrinology

Background:

• Continuous glucose monitoring (CGM) has transformed diabetes care.
• Novel data encoding methods are needed to leverage CGM insights.
• This study introduces a text-based encoding for CGM daily profiles.

Purpose of the Study:

• To develop a method for compressing CGM daily profiles into a text string.
• To preserve essential clinical metrics within the encoded data.
• To optimize the encoding process for accuracy and efficiency.

Main Methods:

• Defined eight alphabets to represent glucose ranges.
• Utilized Akaike information criterion (AIC) to assess error.
• Estimated compression ratios for optimal alphabet selection.
• Analyzed data from six diverse studies including individuals with type 1 diabetes (T1D), type 2 diabetes (T2D), and healthy controls.

Main Results:

• A 9-letter alphabet was found optimal for encoding CGM profiles in T1D and T2D, minimizing information loss (lowest AIC).
• Fewer letters were required for encoding data from healthy individuals due to lower glucose variability.
• The 9-letter alphabet demonstrated strong correlation (0.945-0.965) with the coefficient of variation.

Conclusions:

• Text-based encoding of CGM data facilitates profile classification and search engine utilization.
• Potential applications include predictive modeling, anomaly detection, and AI-driven diabetes research.
• This method enhances the utility of CGM data for clinical practice and research.