Noninvasive Hypoglycemia Detection in People With Diabetes Using Smartwatch Data

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Diabetes Management

Background:

• Hypoglycemia poses significant risks for individuals with diabetes on insulin therapy.
• Current hypoglycemia detection methods often require invasive monitoring or are reactive.

Purpose of the Study:

• To develop and validate a noninvasive approach for detecting hypoglycemia using data from wrist-worn wearables.
• To leverage machine learning to identify hypoglycemia solely from physiological signals captured by smartwatches.

Main Methods:

• Prospective data collection from two distinct wearables (Garmin vivoactive 4S, Empatica E4) and continuous glucose monitoring.
• Development of a machine learning model trained on wearable data to detect hypoglycemia (<3.9 mmol/L).
• Validation of the model in unseen individuals using wearable data exclusively.

Main Results:

• The machine learning model achieved an area under the receiver operating characteristic curve of 0.76 ± 0.07 for hypoglycemia detection.
• Key physiological features associated with hypoglycemia included increased heart rate, decreased heart rate variability, and elevated tonic electrodermal activity.
• The model demonstrated successful noninvasive detection of hypoglycemia using only wearable sensor data.

Conclusions:

• The developed approach offers a potential noninvasive method for real-time hypoglycemia detection in people with diabetes.
• This wearable-based system could serve as a valuable adjunct to existing continuous glucose monitoring and warning systems.
• Further research may lead to improved diabetes management and patient safety through wearable technology.