Classification of glucose records from patients at diabetes risk using a combined permutation entropy algorithm

Area of Science:

• Biomedical Engineering
• Data Science
• Endocrinology

Background:

• Electronic glucose monitors generate vast datasets of glucose readings.
• Conventional visual inspection of glucose data has limitations.
• Mathematical methods can extract deeper clinical insights from glucose time series.

Purpose of the Study:

• To evaluate Permutation Entropy (PE) for distinguishing glucose patterns in healthy versus potentially diabetic subjects.
• To assess the predictive value of PE in the context of diabetes risk.

Main Methods:

• Utilized Permutation Entropy (PE), a time series analysis technique based on ordinal patterns.
• Applied PE to glucose records from subjects at risk of diabetes.
• Employed a Cox regression model for risk stratification analysis.

Main Results:

• PE identified significant differences in glucose records between diabetic and non-diabetic patients up to three years pre-diagnosis.
• Quantitative PE values differed significantly between the two groups (non-diabetic: 3.5878 ± 0.3916; diabetic: 3.1564 ± 0.4166).
• Achieved classification accuracy exceeding 70%, indicating PE's potential as a risk stratification tool.

Conclusions:

• Permutation Entropy (PE) shows potential as a tool for early diagnosis of dysregulation in the glucoregulatory system.
• PE can aid in the early identification of individuals at risk for diabetes using continuous glucose monitoring data.