The RADAR Study - Wearable-Based Dysglycemia Detection and Warning in Diabetes

Completed

• Conditions: Diabetes Mellitus, Insulin-Dependent; Diabetes

• Registration Number: NCT04689685
• Lead Sponsor: Insel Gruppe AG, University Hospital Bern

Brief Summary

The study RADAR aims at developing a wearable based dysglycemia detection and warning system for patients with diabetes mellitus using artificial intelligence.

Detailed Description

Prior research has investigated the general potential of data analytics and artificial intelligence to infer blood glucose levels from a variety of data sources. In this study patients with insulin-dependent diabetes mellitus will be wearing a continuous glucose meter (CGM) and a smartwatch for a maximum duration of 3 months in an outpatient setting. The gathered data will be used to develop a non-invasive and wearable based dysglycemia detection and warning system using artificial intelligence.

Study Timeline

• Study First Submitted: 2020-12-24
• Study First Submitted QC: 2020-12-24
• Study First Posted: 2020-12-30
• Study Start: 2021-02-19
• Primary Completion: 2022-03-28
• Study Completion: 2022-03-28
• Status Verified: 2022-09
• Last Update Submitted: 2022-09-13
• Last Update Posted: 2022-09-14

Recruitment & Eligibility

• Status: COMPLETED
• Sex: All
• Target Recruitment: 40

Inclusion Criteria

• Informed consent as documented by signature
• Age ≥ 18 years
• Diabetes mellitus treated with multiple daily insulin injections (MDI) or continuous subcutaneous insulin infusion (CSII)

Exclusion Criteria

• Smartwatch cannot be attached around the wrist of the patient
• Known allergies to components of the Garmin smartwatch or the Dexcom G6 system
• Pregnancy, intention to become pregnant or breast feeding
• Cardiac arrhythmia (e.g. atrial flutter or fibrillation, AV-reentry tachycardia, AV-block > grade 1)
• Pacemaker or ICD (implantable cardioverter defibrillator)
• Treatment with antiarrhythmic drugs or beta-blockers
• Drug or alcohol abuse
• Inability to follow the procedures of the study, e.g. due to language problems, psychological disorders, dementia, etc. of the participant
• Physical or psychological disease likely to interfere with the normal conduct of the study and interpretation of the study results as judged by the investigator

Study & Design

• Study Type: OBSERVATIONAL
• Study Design: Not specified

Primary Outcome Measures

Name	Time	Method
Accuracy of the RADAR model: Diagnostic accuracy of wearable based physiological data in detecting dysglycemia (glucose > 13.9mmol/L and glucose < 3.9 mmol/L) quantified as the area under the receiver operator characteristics curve (AUC-ROC)	4-12 weeks	Accuracy of the RADAR-model will be assessed using machine learning technology and physiological data recorded by the smartwatch compared to continuous glucose measurements (ground truth)

Secondary Outcome Measures

Name	Time	Method
Accuracy of the RADAR-forecast model: Diagnostic accuracy of CGM data in combination with wearable based data (physiological, time, fasting glucose, and motion) in forecasting severe hyperglycemia (glucose > 13.9mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR forecasts will be assessed using machine learning technology, historical continuous glucose measurements data, and historical wearable based data (physiological, time, fasting glucose, and motion) compared to future continuous glucose measurements (ground truth).
Accuracy of the RADAR-forecast model: Diagnostic accuracy of CGM data in combination with wearable based data (physiological, time, fasting glucose, and motion) in forecasting severe hyperglycemia (glucose < 3.0mmol/L) quantified as AUC-ROC.	4-12 weeks	Accuracy of the RADAR forecasts will be assessed using machine learning technology, historical continuous glucose measurements data, and historical wearable based data (physiological, time, fasting glucose, and motion) compared to future continuous glucose measurements (ground truth).
Change of sleep pattern in dysglycemia (< 3.9 mmol/l and > 13.9 mmol/l) compared to eugylcemia.	4-12 weeks	Sleep pattern will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
Influence on activity (number of steps and stairs climbed per day) on daily time in glycemic target range (3.9 - 10 mmol/L)	4-12 weeks	Number of steps and stairs climbed per day will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
Accuracy of the RADAR+model: Diagnostic accuracy of wearable based data (physiological, time, fasting glucose, and motion) in detecting dysglycemia (glucose > 13.9mmol/L and glucose < 3.9 mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR+-model will be assessed using machine learning technology and wearable based data (physiological, time, fasting glucose, and motion) compared to continuous glucose measurements (ground truth)
Accuracy of the RADAR-forecast model: Diagnostic accuracy of CGM data in combination with wearable based data (physiological, time, fasting glucose, and motion) in forecasting mild hypoglycemia (glucose < 3.9mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR forecasts will be assessed using machine learning technology, historical continuous glucose measurements data, and historical wearable based data (physiological, time, fasting glucose, and motion) compared to future continuous glucose measurements (ground truth).
Change of electrodermal activity (< 3.9 mmol/l and > 13.9 mmol/l) compared to eugylcemia.	4-12 weeks	Electrodermal activity will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
Accuracy of the RADAR+model: Diagnostic accuracy of wearable based data (physiological, time, fasting glucose, and motion) in detecting hypoglycemia (glucose < 3.9 mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR+-model will be assessed using machine learning technology and wearable based data (physiological, time, fasting glucose, and motion) compared to continuous glucose measurements (ground truth)
Accuracy of RADAR-forecast model: Diagnostic accuracy of CGM data in combination with wearable based data (physiological, time, fasting glucose, and motion) in forecasting glucose levels quantified as the mean absolute error.	4-12 weeks	Accuracy of the RADAR forecasts will be assessed using machine learning technology, historical continuous glucose measurements data, and historical wearable based data (physiological, time, fasting glucose, and motion) compared to future continuous glucose measurements (ground truth).
Change of heart rate variability (< 3.9 mmol/l and > 13.9 mmol/l) compared to eugylcemia.	4-12 weeks	Heart rate variability will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
Change of stress level (< 3.9 mmol/l and > 13.9 mmol/l) compared to eugylcemia.	4-12 weeks	Stress level will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
Influence of sleep duration on daily time in glycemic target range (3.9 - 10 mmol/L)	4-12 weeks	Sleep duration will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
Accuracy of the RADAR model: Diagnostic accuracy of wearable based physiological data in detecting hypoglycemia (glucose < 3.9 mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR-model will be assessed using machine learning technology and physiological data recorded by the smartwatch compared to continuous glucose measurements (ground truth)
Accuracy of the RADAR model: Diagnostic accuracy of wearable based physiological data in detecting severe hypoglycemia (glucose < 3.0 mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR-model will be assessed using machine learning technology and physiological data recorded by the smartwatch compared to continuous glucose measurements (ground truth)
Accuracy of the RADAR model: Diagnostic accuracy of wearable based physiological data in detecting severe hyperglycemia (glucose > 13.9mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR-model will be assessed using machine learning technology and physiological data recorded by the smartwatch compared to continuous glucose measurements (ground truth)
Change of skin temperature (< 3.9 mmol/l and > 13.9 mmol/l) compared to eugylcemia.	4-12 weeks	Skin temperature will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
Influence of movement on daily time in glycemic target range (3.9 - 10.0 mmol/l)	4-12 weeks	Movement will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
Accuracy of the RADAR+model: Diagnostic accuracy of wearable based data (physiological, time, fasting glucose, and motion) in detecting severe hypoglycemia (glucose < 3.0 mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR+-model will be assessed using machine learning technology and wearable based data (physiological, time, fasting glucose, and motion) compared to continuous glucose measurements (ground truth)
Accuracy of the RADAR+ model: Diagnostic accuracy of wearable based data (physiological, time, fasting glucose, and motion) in detecting severe hyperglycemia (glucose > 13.9mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR+-model will be assessed using machine learning technology and wearable based data (physiological, time, fasting glucose, and motion) compared to continuous glucose measurements (ground truth)
Accuracy of the RADAR-forecast model: Diagnostic accuracy of CGM data in combination with wearable based data (physiological, time, fasting glucose, and motion) in forecasting dysglycemia (glucose>13.9mmol/L and glucose<3.9 mmol/L) quantified as AUC-ROC	4-12 weeks	Accuracy of the RADAR forecasts will be assessed using machine learning technology, historical continuous glucose measurements data, and historical wearable based data (physiological, time, fasting glucose, and motion) compared to future continuous glucose measurements (ground truth).
Change of heart rate in dysglycemia (< 3.9 mmol/l and > 13.9 mmol/l) compared to eugylcemia.	4-12 weeks	Heart rate will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
Influence on stress-level on daily time in glycemic target range (3.9 - 10 mmol/L)	4-12 weeks	Stress level will be recorded by the smartwatch and glucose values are measured with the continuous glucose meter (CGM).
24. Analysis of user requirements for smartwatch based dysglycemia warning systems	4-12 weeks	User requirements for the smartwatch based dysglycemia warning system will be assessed in a semi-quantitative interview.

Trial Locations

Locations (1)

Department of Diabetes, Endocrinology, Nutritional Medicine and Metabolism; 🇨🇭 Bern, Switzerland