An Ensemble Learning Algorithm Based on Dynamic Voting for Targeting the Optimal Insulin Dosage in Type 1 Diabetes...

Area of Science:

• Endocrinology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Type 1 diabetes (T1D) management requires frequent exogenous insulin administration.
• Current mealtime insulin dosing relies on empirical formulas, often neglecting real-time continuous glucose monitoring (CGM) data trends.
• Suboptimal insulin dosing can lead to poor blood glucose (BG) control.

Purpose of the Study:

• To develop and evaluate an ensemble learning method for more accurate mealtime insulin bolus estimation in T1D.
• To integrate BG trend information from CGM sensors into insulin dosing algorithms.
• To improve BG management by leveraging machine learning.

Main Methods:

• An ensemble learning approach using dynamic voting to combine linear and nonlinear machine learning models was proposed.
• The ensemble model incorporates real-time BG trend information from CGM data.
• Model interpretability tools were used to assess feature contributions.
• The model was trained on a synthetic dataset of 100 virtual subjects and evaluated in a simulated environment.

Main Results:

• The ensemble method demonstrated superior performance compared to individual linear and nonlinear models.
• The model achieved a higher percentage of time within the target glycemic range.
• A favorable trade-off was observed in time spent below and above the target BG range.
• Model interpretation highlighted the critical role of BG dynamics in insulin dosage estimation.

Conclusions:

• Ensemble learning offers a promising approach to enhance mealtime insulin dosing accuracy in T1D.
• Integrating CGM-derived BG trend information significantly improves insulin dosage estimation.
• This data-driven approach has the potential to optimize BG control and reduce glycemic variability.