Updates and analysis on technology, impact investing, and our venture companies.

Next Page Navigation Link
Previous Page Navigation Link

Now that we’re well into the 2nd week of beta testing our Soft CGM application, we’re finally getting to see some new data from external users on the app’s performance; and for our developers and data scientists in the venture lab, it’s a little bit like Christmas morning. Following a month-long alpha test with internal users, 11 type 1 diabetes patients outside of Aspire have been using the diabetes management tool on their phones every day, inputting finger stick, carbohydrate, and insulin data to get blood glucose predictions. And for just one week of use, those predictions are looking pretty accurate.

Our Senior Software Engineer Jason Hertzog created a great data visualization tool that lets us compare each user’s Soft CGM blood glucose predictions with their actual blood glucose level, along with all of their finger stick, carbohydrate, and insulin data inputs throughout the day. In this video, a small sample of data shows just how accurately Soft CGM can predict glucose levels, even when we’re looking ahead four hours into the future. Without invasive sensors or expensive hardware, our adaptive algorithm can achieve predictions that are nearly as accurate as a traditional CGM. But there is still plenty of room for improvement, as a few predictions miss the mark. 

Overall, out of 468 predictions made for our beta users so far, 85 percent of the predictions fall within zones A and B of the Clarke Error Grid. That means that 85 percent of Soft CGM’s predictions would lead to appropriate treatment. We will be addressing areas of improvement over the coming weeks with new operational features and a new model to include in our adaptive algorithm, which should greatly improve our results in the future. 

Now that Soft CGM is finally in the hands of a few T1D’s, we’re getting extremely valuable data and feedback that will help us fine tune our models, improve accuracy, and enhance the personalized user experience. As data pours in over the next several weeks, we’re very excited to learn more ways we can refine this truly innovative approach to diabetes management. Stay tuned for more updates.