AI-Powered Metabolic Health Program from Accurately Predicts Individualized Glycemic Response in People with Type 2 Diabetes

MENLO PARK, Calif.–()–New data presented today at the 80th American Diabetes Association Scientific Sessions – A Virtual Experience unveiled a new AI algorithm from In an in-house study of 1,022 participants, the algorithm effectively predicted individualized glycemic response to specific meals in people with type 2 diabetes and pre-diabetes. This algorithm is critical to the company’s goal of developing and offering a program that provides people with highly personalized food and activity recommendations that drive positive behavior change and improved health.

In the Sugar Challenge study, applied its algorithm to data from each participant’s food and medication logs, heart rate monitor and continuous glucose monitor (CGM) to predict how their body would respond to each specific meal and activity. After a few days of data-gathering to develop an individualized model, the algorithm accurately predicted glucose response to future meals in the absence of any further CGM data, a finding that supports further exploration of the impact of AI models with intermittent CGM use.

While general guidance has existed for some time regarding the effect of dietary composition on blood glucose of people with diabetes, the reality is that food, exercise and medicine impact each of our bodies differently and a ‘one-glove-fits-all’ approach does not work,” said Michael Snyder, PhD, Director of Genomics and Personalized Medicine, and Co-Founder. “Using technology to provide individualized predictions about the glucose response to specific meals may provide benefits beyond general dietary guidelines, enabling people with diabetes to make better ‘personalized’ food choices.”

This study is the first in a series of research to establish a body of evidence for the program and its ability to motivate behavior change and help participants achieve better clinical outcomes. The program uses technology to unify the corpus of science with multiple data sources to produce actionable insights. plans to make the program commercially available later this summer.

Despite extensive investment and effort to address type 2 diabetes, it is still one of the fastest growing and most costly public health issues,” said Noosheen Hashemi, founder and CEO, “We believe AI-enabled technology can be used to deliver a very scalable program that helps people make positive behavior modifications through solid, science-based personalization. The results of this initial study are just the beginning of that journey for us.”

About the Study

In this observational human study, over 22,000 candidates were screened and 1,022 participants were selected. The participants included healthy volunteers, people with pre-diabetes and people with type 2 diabetes. The company developed a series of underlying technologies including derived nutritional values, glycemic index and glycemic load, which estimates how a person’s blood sugar will rise based on the food they eat, for 16 million foods. built its own mobile application to capture and unify various data points into one AI platform, collecting nearly 25 million data points for the study.

The participants wore a heart rate monitor and a CGM for 10 days. They provided body weight at baseline and agreed to comprehensive logging of their activity, food, medication, and water consumption.

After collecting four days of input, the algorithm learned an individualized model for each participant. Blinded to any further CGM data, the algorithm predicted participants’ glycemic responses to foods logged, ready for comparison to the actual CGM data.

The study measured the accuracy of the prediction at 60 minutes and 120 minutes post-meal, meeting industry standards of deviation.

  • For people with type 2 diabetes, the mean average error was 18.4 mg/dL at 60 minutes and 23 mg/dL at 120 minutes.
  • For those with pre-diabetes, the mean average error was 12.7 mg/dL at 60 minutes and 15 mg/dL at 120 minutes.
  • For healthy volunteers, the mean average error was 10.8 mg/dL at 60 minutes and 12.8 mg/dL at 120 minutes.

About is a precision health company founded in 2017. applies artificial intelligence (AI) to a mix of biological and behavioral data to deliver radically personalized insights and specific recommendations that drive positive behavioral change for people with metabolic syndrome conditions, including diabetes and pre-diabetes. The system is designed to encourage small lifestyle changes and habit-building that are readily achievable, but add up to a significant improvement in one’s health. The company closed $10.7M in seed round funding in July of 2019. More information can be found at


About the Author: Biotech Today

You might like