Noninvasive Hypoglycemia Detection in People With Diabetes Using Smartwatch Data

Abstract

OBJECTIVE To develop a noninvasive hypoglycemia detection approach using smartwatch data. RESEARCH DESIGN AND METHODS We prospectively collected data from two wrist-worn wearables (Garmin vivoactive 4S, Empatica E4) and continuous glucose monitoring values in adults with diabetes on insulin treatment. Using these data, we developed a machine learning (ML) approach to detect hypoglycemia (<3.9 mmol/L) noninvasively in unseen individuals and solely based on wearable data. RESULTS Twenty-two individuals were included in the final analysis (age 54.5 ± 15.2 years, HbA1c 6.9 ± 0.6%, 16 males). Hypoglycemia was detected with an area under the receiver operating characteristic curve of 0.76 ± 0.07 solely based on wearable data. Feature analysis revealed that the ML model associated increased heart rate, decreased heart rate variability, and increased tonic electrodermal activity with hypoglycemia. CONCLUSIONS Our approach may allow for noninvasive hypoglycemia detection using wearables in people with diabetes and thus complement existing methods for hypoglycemia detection and warning.

Smartwatches for non‐invasive hypoglycaemia detection during cognitive and psychomotor stress

“…Previously, we have shown the general concept of hypoglycaemia detection using smartwatch data 4 . The present analysis suggests that this concept extends to detecting pronounced hypoglycaemia even when people are involved in tasks related to cognitive and psychomotor stress, such as driving.…”

“…While electrodermal activity, heart rate variability and accelerometer features constitute relevant features for hypoglycaemia detection, 2,4 the accuracy of ML decision-making across different levels of hypoglycaemia is unknown. Electrodermal activity proved to be a decisive feature modality for detecting pronounced hypoglycaemia, but appeared to be less informative in detecting mild hypoglycaemia.…”

“…For evaluating the ML models, we used a leave‐one‐subject‐out approach, which allows for assessing generalizability of the models to unseen individuals. Unlike personalized models used in a previous study, 4 this approach offers scalability by eliminating the need to train individual models for each user. Interpretable ML allows assessing the decision‐making across different levels of hypoglycaemia.…”

Relationship between Short- and Mid-Term Glucose Variability and Blood Pressure Profile Parameters: A Scoping Review

“…The widespread use of such technology accompanied by automated and standardized reports including indices of interest could lead to a better understanding of several mechanisms with great clinical potential. Wearable devices allowing non-invasive glucose monitoring or cuffless devices performing frequent BP measurements could provide detailed assessments of glucose and BP profiles including GV and BPV [ 41 , 42 , 43 ]; however, such devices have not been validated with respect to their accuracy according to established protocols and, most importantly, in terms of their clinical utility and intended use [ 41 ]. The association between GV and BPV could highlight the importance of several common underlying mechanisms.…”

From Data to Insight: Predicting Interstitial Glucose in Healthy Cohort with Non-invasive Sensor Technology and Machine Learning

“…One approach to overcome this burden might be the use of data obtained from non-invasive measurements of physiological parameters by wearables which have been shown to respond to upon changes in the blood glucose level. Initial proof-of-concept studies show that hypoglycemic events can be detected by heart rate variability and electrodermal activity using wearables in prediabetes and diabetes [17][18][19] .…”