Multimodal Glucose Prediction in Type 2 Diabetes

CGM- and Behavior-based Large Health Model for Just-in-time Diabetes Management

The primary objective of this research, funded by Samsung Strategic Alliance for Research and Technology, is to develop multi-modal foundation models that integrate Continuious Glucose Monitoring (CGM) data with patient behavior data (food intake, medication, and physical activity) to improve real-time glucose prediction and personalized diabetes management for patients with Type 2 diabetes (T2D), delivered via mobile apps and digital health tools.

Study Overview

• Status: Not yet recruiting
• Conditions: Type 2 Diabetes
• Intervention / Treatment: Device: Digital Health Data Collection System

• Study Type: Observational
• Enrollment (Estimated): 36

Contacts and Locations

• Study Contact: Name: Nestoras Mathioudakis, MD, MHS; Phone Number: 410-955-3663; Email: nmathio1@jhmi.edu
• Study Contact Backup: Name: Gordon Gao, PhD; Phone Number: 410-234-9450.; Email: ggao8@jh.edu

Study Locations

• United States
  • Maryland
    • Baltimore, Maryland, United States, 21287
      • Johns Hopkins Medicine
      • Contact: Nestoras Mathioudakis, MD, MHS; Phone Number: 410-955-3663; Email: nmathio1@jhmi.edu
      • Contact: Gordon Gao, PhD; Phone Number: 410-234-9450.; Email: ggao8@jh.edu
      • Principal Investigator: Nestoras Mathioudakis, MD
      • Principal Investigator: Gordon Gao, PhD

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Non-Probability Sample

Study Population

Participants will be selected from the Johns Hopkins Medicine adult clinical population. The study population will include adults with type 2 diabetes who receive diabetes care through Johns Hopkins Medicine, including primary care and endocrinology clinics. Participants will be recruited from patients whose routine diabetes care includes use of continuous glucose monitoring and who may be eligible to contribute glucose, wearable, app-based behavioral, and electronic medical record data for development and validation of glucose prediction models.

Description

Inclusion Criteria:

• 18-75 years old
• Registered patient under Johns Hopkins Medicine (JHM)
• Type 2 Diabetes diagnosis
• Diabetes managed by a primary care physician or endocrinologist at JHM
• Android Smartphone user
• Must have a Dexcom G7 or FreeStyle Libre 3 CGM and using a mobile app to access their CGM data (G7 or Libre 3 apps)
• 2 weeks of usage (with at least 50% wear time) prior to study participation required
• CGM Time in Range of <70% in 14 days prior to enrollment
• Must be able to read, understand, and communicate in English
• Must not have hearing or vision impairments
• Willingness to Download the Welldoc app
• Agree to wear a SAMSUNG Galaxy Watch at least 12 hours per day
• Download SAMSUNG Health (Non-SAMSUNG Phone user)
• Download Google Health Connect
• Use CGM at least 80% of the time
• Take a photo of all meals

Exclusion Criteria:

• Pregnant
• Non-English speaker
• Has hearing or vision impairment
• Use of an insulin pump (i.e. automated insulin delivery system)
• Diagnosed with other forms of diabetes (e.g. Type 1 Diabetes, Latent Autoimmune Diabetes in Adults (LADA), Maturity-Onset Diabetes of the Young (MODY), or Gestational diabetes)
• Non-Android smartphone user (i.e., Apple iOS)
• CGM time-below-range > 4% (i.e. hypoglycemia) in the 14 days prior to enrollment.
• Hospitalization for Diabetic Ketoacidosis (DKA) or severe hypoglycemic episode within the previous 6 months.

Study Plan

How is the study designed?

Number of groups / cohorts

1

Cohorts and Interventions

Group / Cohort

Intervention / Treatment

Adults With Type 2 Diabetes Using CGM; Adults with type 2 diabetes receiving care through Johns Hopkins Medicine will participate in a single site observational cohort study. Participants will continue usual diabetes care and will not receive a treatment intervention from the study team. Participants will contribute CGM, smartwatch, app-based behavioral, and electronic medical record data for development and validation of glucose prediction models. Study-generated messages and summary reports will be reviewed by the study team and will not be delivered to participants.

Device: Digital Health Data Collection System; Participants will use a digital health data collection system that includes the Welldoc app, a Samsung smartwatch, and the participant's existing continuous glucose monitor. The system will collect CGM data, smartwatch-derived activity, sleep, and vital sign data, and app-based behavioral information such as meals, physical activity, and medication use. Participants will continue usual diabetes care and will not receive treatment recommendations from the study team. Data will be used to develop and validate glucose prediction models and AI-generated research outputs that will be reviewed by the study team and not delivered to participants.; Other Names: Continuous glucose monitor; Dexcom G7; Welldoc; Samsung Galaxy Watch; FreeStyle Libre 3

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Root Mean Square Error of CGM Glucose Prediction Model	Model performance will be evaluated using root mean square error to compare predicted continuous glucose monitor glucose values with observed continuous glucose monitor glucose values. Model performance using continuous glucose monitor data alone will be compared with model performance using continuous glucose monitor data plus behavioral measures, including physical activity and diet logs.	Up to 3 Month follow-up

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Number of Meal Logs Submitted Per Participant	The total number of meal logs submitted by each participant in the study app will be summarized. A higher number indicates more frequent meal logging.	Up to 3 Month follow-up
Number of Physical Activity Logs Submitted Per Participant	The total number of physical activity logs submitted by each participant in the study app will be summarized. A higher number indicates more frequent physical activity logging.	Up to 3 Month follow-up
Number of Medication Logs Submitted Per Participant	The total number of medication logs submitted by each participant in the study app will be summarized. A higher number indicates more frequent medication logging.	3 month follow-up
Number of Mood Logs Submitted Per Participant	The total number of mood logs submitted by each participant in the study app will be summarized. A higher number indicates more frequent mood logging.	Up to 3 Month follow-up
Percent of Expected Continuous Glucose Monitor Data Captured Per Participant	The percentage of expected continuous glucose monitor data captured during the study period will be summarized for each participant. A higher percentage indicates greater continuous glucose monitor use.	Up to 3 Month follow-up
Mean Daily Samsung Smartwatch Wear Time Per Participant	Mean daily Samsung smartwatch wear time will be summarized as the average number of hours per day that each participant wears the Samsung smartwatch. A higher number indicates greater smartwatch wear.	Up to 3 Month follow-up
Percent of Study Days With Study App Use Per Participant	The percentage of study days with any recorded study app use will be summarized for each participant. A higher percentage indicates greater study app use.	Up to 3 Month follow-up
Clinician-Rated Accuracy of Artificial Intelligence-Generated Content as Assessed by a Study-Specific 5-Point Likert Scale	Artificial intelligence-generated research content will be reviewed by the study team for accuracy using a study-specific 5-point Likert scale. Scores range from 1 to 5, with higher scores indicating greater accuracy. These outputs will not be delivered to participants.	3 month follow-up
Clinician-Rated Safety of Artificial Intelligence-Generated Content as Assessed by a Study-Specific 5-Point Likert Scale	Artificial intelligence-generated research content will be reviewed by the study team for safety using a study-specific 5-point Likert scale. Scores range from 1 to 5, with higher scores indicating greater safety. These outputs will not be delivered to participants.	3 month follow-up
Clinician-Rated Communication Quality of Artificial Intelligence-Generated Content as Assessed by a Study-Specific 5-Point Likert Scale	Artificial intelligence-generated research content will be reviewed by the study team for communication quality using a study-specific 5-point Likert scale. Scores range from 1 to 5, with higher scores indicating better communication quality. These outputs will not be delivered to participants.	3 month follow-up

Collaborators and Investigators

• Sponsor: Johns Hopkins University
• Collaborators: Welldoc; Samsung Research America
• Investigators: Principal Investigator: Nestoras Mathioudakis, MD, MHS, Johns Hopkins University

Publications and helpful links

General Publications

• Healey E, Tan ALM, Flint KL, Ruiz JL, Kohane I. A case study on using a large language model to analyze continuous glucose monitoring data. Sci Rep. 2025 Jan 7;15(1):1143. doi: 10.1038/s41598-024-84003-0.

Study record dates

Study Major Dates

• Study Start (Estimated): June 15, 2026
• Primary Completion (Estimated): January 15, 2027
• Study Completion (Estimated): February 26, 2027

Study Registration Dates

• First Submitted: June 2, 2026
• First Submitted That Met QC Criteria: June 2, 2026
• First Posted (Actual): June 8, 2026

Study Record Updates

• Last Update Posted (Actual): June 8, 2026
• Last Update Submitted That Met QC Criteria: June 2, 2026
• Last Verified: June 1, 2026

More Information

Terms related to this study

• Additional Relevant MeSH Terms: Endocrine System Diseases; Metabolic Diseases; Glucose Metabolism Disorders; Diabetes Mellitus; Nutritional and Metabolic Diseases; Diabetes Mellitus, Type 2
• Other Study ID Numbers: IRB00523137

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No