One Drop Develops a Model to Predict Blood Glucose Levels for People Using CGMs

One Drop has developed a model to accurately predict blood glucose levels for people using continuous glucose monitors (CGMs). The capability is the latest development in using AI to improve self-care for people with diabetes, and builds on One Drop’s AI-driven 8-hour blood glucose forecasts for users with type 2 diabetes, available in the One Drop mobile app as of this summer. 
 
In a recent study, the majority of the model-predicted blood glucose levels were within 20 percent of the actual CGM-measured values up to 120 minutes in advance, ensuring people with type 1 and type 2 diabetes would be able to make informed decisions regarding what to eat, how much insulin to bolus and the degree to which they should be physically active.
 
“No one with diabetes likes to be surprised by a sharp drop in blood sugar levels, or an unexpected rise, both of which could lead to unsafe situations and potentially dangerous consequences. Knowing where blood sugars will be up to 2 hours in advance can help people on CGM make the right choices regarding what they’re going to do next and help avoid problems before they even occur,” said One Drop founder and CEO Jeff Dachis. “One Drop’s predictive model offers the most accurate blood glucose forecasts based on real-world data that have been published to date. It’s a huge step forward in empowering people with the information they need to take better control of their health.”
 
Using more than 10 million hours of CGM data from One Drop users, the model accurately predicted blood glucose levels within 30 mg/dL of the CGM-measured values up to 120 minutes in advance. At 30 minutes, predictions were within 30 mg/dL more than 97% of time. At 60 minutes, predictions were within 30 mg/dL more than 75% of the time. At 120 minutes, predictions were within 30 mg/dL more than 55% of the time.
 
The same model accurately predicted blood glucose levels 30 minutes in advance that were within 50 mg/dL of the CGM-measured values more than 99% of the time. At 60 minutes, predictions were within 50 mg/dL more than 90% of the time. At 120 minutes, predictions were within 50 mg/dL more than 80% of the time.
 
One Drop’s  approach combines personalized one-on-one coaching with AI-powered insights to deliver care to the greatest number of people possible. Proprietary machine learning—powered by health data collected from over 1.5 million One Drop users—generates actionable advice that connects behavior with outcomes to simplify decision making and promote lasting behavior change. 
 
According to the company, One Drop’s efficacy has been established in more than 20 peer-reviewed studies, which showed that people with diabetes using One Drop decreased their A1c (average blood glucose levels) in a little as one month. One Drop is the only provider of blood glucose forecasts for people with type 2 diabetes.