Clinical Research

Nutrition is an essential component of managing chronic health conditions like diabetes and obesity. However, logging meals within health apps is time-consuming and cumbersome for many people. Welldoc continues to research opportunities to streamline this process through advanced AI.

In this research, we successfully developed a novel approach for automatedly detecting recent food intake solely based on continuous glucose monitoring (CGM) data and AI with an accuracy of 77.8%.

This research is a blueprint for the next generation of digital health. By enabling tools to gather consistent, complete dietary information without any input from the user, we are moving toward zero-effort self-management.

At Welldoc, our goal is to leverage these deep insights to build more personalized capabilities. By analyzing engagement across different clinical segments, we are developing AI models that can better predict individual needs and tailor digital coaching to truly support every unique path.

In this research, we studied the probabilities of weight loss, stratified by starting body mass index (BMI) band. We developed a probability state transition matrix to estimate the weight loss effect of a digital health tool on groups of individuals with and without an anti-obesity medication (AOM) in a weight management program.

In general, for users with starting BMI <36, use of the digital health tool proved to be an effective mechanism to lose weight without an AOM. When combined with engagement patterns, this research is foundational to developing AI-based weight loss prediction capabilities, based on both starting BMI level and the level of digital health engagement.

We investigated user engagement patterns within our digital health application, comparing individuals utilizing a weight management program both with and without anti-diabetic/anti-obesity medications (AOMs).

Studying engagement patterns is a key focus area for Welldoc as we continue to focus our research on further personalizing the weight management journey.

This research indicates that engagement can vary based on medication usage. We observed significantly higher engagement with the non-AOM group. This supports the idea that the need for tracking, guidance, and support needs may be more pronounced for individuals navigating their weight without these medications.

In this research, Welldoc and Johns Hopkins Carey Business School leveraged dense, real-time glucose data from Continuous Glucose Monitoring (CGM), similar to how fitness trackers collect biometric signals. Our work shows that by combining this data with food intake, we can significantly improve the accuracy of future glucose predictions. This takes Welldoc one step closer to the real world application of reliable prediction and prevention of overnight hyperglycemia.

Why It Matters:

• Smarter Predictions: We developed a "large health model" (LHM) that uses AI to analyze dense data from CGMs and food intake data. This model is more accurate in predicting future glucose levels over longer periods, with a greater improvement for the 2-hour interval when food's impact is highest. This improved accuracy is critical to establishing trust in the next generation of AI driven digital health coaching capabilities.
• Better Health Outcomes: This work establishes a clear pathway for integrating various types of health data into AI models to enhance their predictive power. The ultimate goal is to apply this improved accuracy to real-world clinical challenges, such as reliably predicting and preventing overnight hyperglycemia.
• A New Approach: Welldoc’s research goes beyond existing models by not only collecting data but also using it to predict future biometric values and offer automated coaching based on those predictions. This is a novel approach that leverages the power of AI to provide actionable insights for.

Health systems are increasingly integrating digital health solutions to provide personalized support to patients and timely insights to clinicians. The Allina Health system,, has partnered with Welldoc to integrate Welldoc’s FDA-cleared cardiometabolic digital app into their diabetes program. The Welldoc App syncs with continuous glucose monitoring (CGM) devices. This integration enables the system to capture a new metric, Glucose Management Indicator (GMI) to help support patients with diabetes. GMI is a calculated value based on CGM data that provides an estimate of a person's average blood sugar (A1C) over a shorter period. This is valuable because GMI can show changes in glucose levels faster than a traditional A1C test, which is helpful for both patients and clinicians. The poster outlines the GMI as a new metric for success in diabetes and key aspects of health system-health tech collaboration in diabetes.

Key Takeaways:

• New Metric for Success: Welldoc is the first to utilize GMI as a quality metric within a digital health solution. The GMI is a calculated value used to estimate A1C based on CGM data. It provides a unique value in that it can be reported in a shorter time period (10-14 days) and allows for faster observation of glucose changes. The GMI's inclusion as a quality metric in the 2025 HEDIS measures recognizes the value of CGM data in assessing diabetes management.
• The power of health system-health tech collaboration: Allina and Welldoc partnered to effectively integrate the Welldoc solution into Allina’s diabetes program, which included workflow, eHR integration and incorporating data into clinical interventions and treatment plans.
• Eye on Quality of Care: The incorporation of GMI as a quality metric is essential for maintaining high Health Plan Ratings and Star ratings for value-based care. This integration helps health systems meet their quality goals and ultimately improves care for people with diabetes.

Obesity is a costly condition. Welldoc is able to shift the BMI curve through digital health engagement alone – translating to significant ROI and cost savings. Here, we presented a modeling tool to show the ROI and economic impact of shifting individuals with obesity into a lower BMI band and our ability to do so through AI-driven digital health.

Key Takeaways:

• ~25% of individuals with obesity successfully shifted to a BMI below 30
• $1,527 reduction in estimated average cost per person at 6 months

This was achieved without GLP-1 medications, highlighting how AI-driven digital health can be used in multiple ways to support weight management – as a standalone solution, or a precursor or adjunct to more costly treatment pathways.