Improving Glucose Prediction Accuracy in Physically Active Adolescents With Type 1 Diabetes

Area of Science:

• Biomedical Engineering
• Endocrinology
• Diabetes Technology

Background:

• Physical activity poses challenges for glycemic control in type 1 diabetes.
• Accurate glucose prediction is crucial for automated decision-making systems like artificial pancreas (AP).
• Incorporating physiological variables may enhance glucose prediction during exercise.

Purpose of the Study:

• To evaluate the impact of including heart rate in glucose prediction models.
• To compare the performance of a new subspace identification model with and without heart rate data.
• To assess the proposed model against the existing metabolic state observer (MSO) in AP systems.

Main Methods:

• A predictor-based subspace identification technique was applied to a dynamic glucose prediction model.
• The model incorporated heart rate, carbohydrate intake, and insulin boluses.
• Performance was evaluated using experimental data from adolescents during ski camp, comparing models with (SID-HR) and without (SID-2) heart rate, and the MSO model.

Main Results:

• The SID-HR model demonstrated a statistically significant improvement in root-mean-square error compared to SID-2 (P < .001) and MSO (P < .001).
• The SID-HR model showed favorable performance over SID-2 and MSO, even when accounting for model complexity.
• Heart rate inclusion significantly enhances glucose prediction accuracy.

Conclusions:

• Including heart rate in glucose dynamics models improves prediction accuracy during physical activity.
• This methodology can enhance exercise-informed predictive control algorithms in artificial pancreas systems.
• The findings support the integration of heart rate monitoring for more robust diabetes management technology.