Classification of Hypoglycemic Events in Type 1 Diabetes Using Machine Learning Algorithms.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Endocrinology

Background:

• Continuous and flash glucose monitoring (CGM/FGM) generate vast amounts of data.
• Improving the utilization of CGM/FGM data is crucial for diabetes management.
• Identifying root causes of hypoglycemic events can enhance patient safety and treatment efficacy.

Purpose of the Study:

• To test the hypothesis that a machine learning (ML) model can be trained to identify the root causes of hypoglycemic events using CGM/FGM data.
• To develop and validate an accurate ML model for hypoglycemia root cause analysis.

Main Methods:

• CGM/FGM data from 449 patients with type 1 diabetes were analyzed.
• 5041 hypoglycemic events were randomly selected and classified by clinicians into three main causes: overestimated bolus, overcorrection of hyperglycemia, and excessive basal insulin.
• A purpose-built convolutional neural network (HypoCNN) was developed and evaluated on training and validation datasets, including a "ground truth" dataset validated by insulin and carbohydrate recordings.

Main Results:

• The HypoCNN model achieved an average area under the curve (AUC) of 0.921 in the initial train/test split.
• Performance enhancements were achieved through time series masking, addition of time features, and class weighting.
• The HypoCNN model demonstrated strong performance on the "ground truth" dataset, achieving an AUC of 0.917.

Conclusions:

• Machine learning models can be effectively trained to interpret CGM/FGM data.
• The developed HypoCNN model offers a robust and accurate method for identifying the root causes of hypoglycemic events.
• This approach holds significant potential for improving diabetes management and patient outcomes.