Data-Driven Blood Glucose Pattern Classification and Anomalies Detection: Machine-Learning Applications in Type 1...

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Diabetes Technology

Background:

• Diabetes mellitus is a chronic metabolic disorder requiring self-management, including blood glucose (BG) monitoring and adjustments to diet and insulin.
• BG anomalies, defined as undesirable readings, can stem from known or unknown causes.
• Machine learning (ML) is increasingly used in diabetes research for BG anomaly detection, yet current reviews on modeling strategies are lacking.

Purpose of the Study:

• To review and analyze state-of-the-art ML strategies for BG anomaly classification and detection.
• To focus on glycemic variability (GV), hyperglycemia, and hypoglycemia in type 1 diabetes.
• To assess ML's role in personalized decision support systems and BG alarm events for diabetes self-management.

Main Methods:

• A comprehensive literature search was conducted across multiple databases between September 2017 and November 2018.
• Peer-reviewed journals and articles were screened, with 47 selected for critical analysis.
• Information extraction followed predefined categories, with interrater agreement assessed using Cohen's kappa test.

Main Results:

• Various ML algorithms, including artificial neural networks, support vector machines, and deep belief networks, have been applied to BG anomaly detection.
• These ML models have demonstrated promising performance in classifying and detecting BG anomalies.
• The review identified common use of theoretical thresholds, which often fail to account for individual patient variations.

Conclusions:

• ML advancements, driven by diabetes technologies and self-collected data, are crucial for detecting hypoglycemia, hyperglycemia, and glycemic variability.
• Future studies must address inter- and intra-patient variations in BG dynamics and temporal changes.
• Emphasis on input data types and time lags is recommended for developing more robust and personalized BG anomaly detection systems.