Enhanced blood glucose levels prediction with a smartwatch

Area of Science:

• Biomedical Engineering
• Health Informatics
• Endocrinology

Background:

• Stable blood glucose (BG) control is critical for preventing long-term complications in Type 1 diabetes (T1D).
• Accurate forecasting of future BG levels is essential for effective diabetes management.
• Current methods relying on manual activity logging for BG prediction are burdensome and often inaccurate.

Purpose of the Study:

• To develop and evaluate a framework for enhanced blood glucose forecasting.
• To reduce participant burden associated with manual data logging for diabetes management.
• To improve the accuracy and efficiency of BG level predictions.

Main Methods:

• Developed a predictive model integrating continuous glucose monitoring (CGM) data, smartwatch sensor data (heart rate, step count), and user-documented activities.
• Collected data from a cohort of participants with T1D.
• Analyzed the impact of smartwatch sensor data versus manual activity logs on prediction accuracy.

Main Results:

• Manual documentation of daily activities did not significantly improve BG level predictions.
• Incorporating smartwatch sensor data, specifically heart rate and step detection, substantially enhanced BG prediction accuracy.
• The model demonstrated improved prediction capabilities using physiological sensor data alongside recent BG measurements.

Conclusions:

• Smartwatch sensor data offers a more objective and less burdensome approach to improving BG forecasting compared to manual logging.
• Physiological data from wearables can significantly augment CGM data for more precise diabetes management.
• This framework presents a promising avenue for optimizing T1D self-management through technology.