Clinical Research

A resource to get the latest clinical evidence, studies, models and frameworks to advance knowledge of how best to manage chronic health conditions.

Welldoc is committed to our scientific research, advancing digital health, transforming chronic care, and driving value across healthcare. Areas of focus include digital health engagement, advancing artificial intelligence, cardiometabolic condition outcomes, cost and value, and real-world integration into the health ecosystem.

Behavioral Factors, Empowerment Bolsters Self-Management

The increasing use of web-based or technology-enabled solutions for health management presents opportunities to improve patient self-management…
A Novel Approach to Continuous Glucose Monitoring
Health Plan Opportunities to Control Costs Related to Chronic Disease
Moving the Dial in Lowering and Controlling A1C
The Power of Integrated Peer Support and Digital Health
A Novel Approach to Continuous Glucose Monitoring
Health Plan Opportunities to Control Costs Related to Chronic Disease
Health Plan Opportunities to Control Costs Related to Chronic Disease
Moving the Dial in Lowering and Controlling A1C
The Power of Integrated Peer Support and Digital Health
The Power of Integrated Peer Support and Digital Health
Topics
Clinical Bibliography Theme
Type
Clinical Bibliography Type
Artificial Intelligence (AI)CGM & Real-time Devices
welldocposter 150x150 1
Poster
Methods, Analysis, and Insights from a State-Of-The-Art Large Glucose Model
Proactive diabetes self-management requires accurate glucose value prediction and in-the-moment AI-driven coaching based on those predictions, all made possible by raw data from continuous glucose monitors (CGM).

Here, Welldoc builds upon our prior AI models that used CGM data only and expands to a new Large Glucose Model (LGM), which uses both CGM values and time series inputs to predict glucose trajectories at 30mins, 60mins and 2-hour time horizons. Results were analyzed across different Type 1 and Type 2 diabetes population subgroups (time of day, age group and total engagement levels) within a mobile diabetes management application.

This work will allow Welldoc to power new cardiometabolic focused capabilities and innovations in enhanced AI-driven personalization. Welldoc continues to drive this type of research to develop novel solutions leveraging data from real-world sensors, like CGM, and provide deep insights into subgroup level patterns and differences.

Junjie Luo, Abhimanyu Kumbara, Anand K. Iyer, Mansur E. Shomali, and Guodong “Gordon” Gao

Artificial Intelligence (AI)CGM & Real-time Devices
welldocposter 150x150 1
Poster
Evaluating Perplexity and Glucose Level Impact on State-Of-The-Art Generative Pre-trained Transformer (GPT) Model to Predict Glucose Values at Different Time Intervals
Welldoc continues to research how generative-AI models can be used in combination with data from real-time devices like continuous glucose monitors (CGM) to predict metrics like glucose risk indicator (GRI) and future health outcomes. Here, Welldoc developed a state-of-the-art GPT model to predict CGM trajectory at different time horizons and across two different prediction contexts. One particular context included model perplexity, which is an industry-standard metric of how well a model can predict a sample or next value in a sequence. The data show higher prediction uncertainty as the glucose profile complexity increases. This work supports Welldoc’s efforts to further optimize our diabetes solution, making the experience for individuals even more targeted and personalized.

Junjie Luo, Abhimanyu Kumbara, Anand K. Iyer, Mansur E. Shomali, and Guodong “Gordon” Gao

Artificial Intelligence (AI)Outcomes: AllOutcomes: Weight
welldocposter 150x150 1
Poster
Nutritional Analysis and Advanced Artificial Intelligence (AI) Predicts Weight Loss for People with Diabetes
Weight loss is a key factor in improving health outcomes for many cardiometabolic conditions, like diabetes. Building upon our previous AI-modeling research, Welldoc has now harnessed our AI to predict weight loss. Based on a large dataset of individuals with type 2 diabetes using our digital health solution’s food log, our AI models predicted at least 3% weight loss with high accuracy (93%). It also evaluated the effect of different nutrients on weight loss, as well as time-based correlations. Future research will dive deeper into nutrient- and time-of-day-based personalized digital nutrition coaching to better predict the likelihood and timing of achieving nutrition and weight loss goals for users. This work is integral in Welldoc’s ability to make our digital health solution for weight management even more targeted and accurate, as well as support weight loss-related outcomes across comorbid conditions in our platform.

Catherine Brown, MS, RD, Anand Iyer, PhD, MBA, Abhimanyu Kumbara, MS, MBA, Maxwell Ebert, MPH

Artificial Intelligence (AI)CGM & Real-time DevicesEngagementIntegrated Care ModelsOutcomes: AllOutcomes: Blood PressureOutcomes: GlycemicOutcomes: WeightPopulation HealthReal World Evidence
welldoc article
Article
The Critical Elements of Digital Health in Diabetes and Cardiometabolic Care
In the latest issue of “Frontiers in Endocrinology,” Welldoc highlights key elements for designing and implementing successful diabetes digital health tools in clinical practice. We explore topics like the importance of regulatory oversight, looking beyond A1C, addressing technology literacy, and clinical integration to increase efficiency. Additionally, we outline the practical steps needed to become a digital health-ready practice. These insights are integral to Welldoc’s commitment to developing best-in-class digital health solutions and transforming the care continuum.

Mansur Shomali, Pablo Mora, Grazia Aleppo, Malinda Peeples, Abhimanyu Kumbara, Janice MacLeod, Anand Iyer

Artificial Intelligence (AI)CGM & Real-time Devices
welldocposter 150x150 1
Poster
Evaluating a State-of-the-art Generative Pre-trained Transformer Model to Predict Continuous Glucose Monitoring Values at Different Time Intervals
Welldoc is actively developing generative-AI models specific to continuous glucose monitoring (CGM) value prediction. Our GPT-like AI modeling approach, based on a vast CGM dataset for individuals with type 1 and type 2 diabetes, has achieved state-of-the-art performance in predicting short-term CGM trajectories, at five-minute intervals, up to two hours in advance. Specifically, when predicting one of the 5 glucose categories (VeryLow, Low, InRange, High, VeryHigh), the model demonstrated an overall accuracy of 94% in predicting CGM values within 30 minutes. The absolute prediction accuracy to predict glucose values within 5mg/dL and 10 mg/dL error range in 30mins was 62% and 80% respectively. This exceptional accuracy highlights the model's potential to provide valuable insights into the dynamic interplay between blood sugar levels, lifestyle factors, and interventions. By leveraging these insights, personalized, automated coaching can be developed to optimize diabetes management and improve individual outcomes. This work is integral to Welldoc’s Advanced AI and our ability to transform the care continuum.

Mansur Shomali, MD, CM, Junjie Luo, MS, Abhimanyu Kumbara, MS, Anand Iyer, PhD, Gordon Gao, PhD

Artificial Intelligence (AI)CGM & Real-time DevicesOutcomes: AllOutcomes: Glycemic
welldocposter 150x150 1
Poster
CGM-GPT: A Transformer Based Glucose Prediction Model to Predict Glucose Trajectories at Different Time Horizons
Welldoc is pioneering the use of AI to revolutionize diabetes management. Our groundbreaking general purpose transformer based CGM research represents a significant leap forward in predicting glucose levels for individuals living with both Type 1 and Type 2 diabetes.

This research is the first in a series outlining Welldoc's novel methodology towards predicting future continuous glucose monitoring (CGM) glucose levels with high accuracy. The poster presents Welldoc's state-of-the-art AI models, which can predict glucose trajectories at 30-, 60- and 120-minute intervals for both type 1 and type 2 diabetes populations with ~ 50% less root mean square error, when compared to existing benchmark studies. Welldoc's GPT model, CGM-GPT, was trained only on CGM data sets from individuals living with Type 1 and Type 2 diabetes, and reflects the advanced opportunity to develop sophisticated large sensor models (LSM) by leveraging the vast data available via real-time sensors.

We are committed to further refining our models by incorporating additional data sources and exploring expanded applications. This ongoing research will drive the development of innovative diabetes management solutions. Stay tuned for additional information on our exciting transformer based CGM research.

Junjie Luo, MS, Abhimanyu Kumbara, MS, Mansur Shomali, MD, CM, Anand Iyer, PhD, Gordon Gao, PhD

small_c_popup.png

Taking Diabetes Self-Management to the Next Level