Prediction of Postprandial Blood Glucose Variability Using Machine Learning in Frequent Insulin Injection Therapy...

Area of Science:

• Artificial Intelligence in Medicine
• Diabetes Technology
• Machine Learning for Health

Background:

• Postprandial glucose variability presents a significant management challenge for individuals with diabetes using multiple daily insulin injections (MDI).
• Accurate prediction of glucose peaks and nadirs is crucial for optimizing glycemic control and preventing complications.

Purpose of the Study:

• To evaluate the efficacy of transformer-based machine learning models in predicting postprandial glucose peaks and nadirs.
• To assess the impact of different nutritional input types (full nutrition vs. carbohydrate-only) on model predictive accuracy.

Main Methods:

• An observational study involving 58 adults with diabetes who provided continuous glucose monitoring, insulin logs, and dietary records.
• Development and evaluation of three transformer-based machine learning models using pre-meal glucose, insulin dose, and nutritional information.
• Performance assessment using Mean Absolute Error (MAE), R-squared (R²), and Clarke error grid analysis.

Main Results:

• The Full Nutrition Model demonstrated MAEs of 32.2 mg/dL for glucose peaks and 21.8 mg/dL for nadirs, with R² values of 0.58 for both.
• Carbohydrate-based input models achieved comparable accuracy to the full nutrition model.
• The majority of model predictions fell within the clinically acceptable Zones A and B of the Clarke error grid.

Conclusions:

• Transformer-based machine learning models show significant potential for accurately predicting postprandial glucose variability in MDI-treated diabetes patients.
• Simplified dietary inputs focusing on carbohydrates are effective, suggesting feasibility for practical clinical implementation.
• These findings support the integration of AI-driven predictive tools into diabetes management strategies.