ADA 79th Scientific Session Poster Presentation
In 2019 Diabits presented at the American Diabetes Association conference.
Accuracy was also measured using more than 1.6 million predictions made by the Diabits algorithm. Using the Parkes Error Grid Analysis (link to epsilon page), the 60 minute predictions had a 96.83% accuracy in the A and B regions. This means the predictions are clinically acceptable and will not negatively impact a user’s treatment decisions. Diabits does not require manual inputs, and therefore this value is affected by the fact that many users likely had unknown food or insulin events between the prediction moment and the moment for which the prediction was made. If only points with no such events were chosen, this accuracy would be even higher.
This research shows that Diabits predictive technology can help people with diabetes better manage their diet and insulin by providing an estimate of likely future blood glucose levels. The algorithm has the potential of augmenting the currently available artificial pancreas technologies to better regulate insulin in the presence of metabolic factors such as activity and stress. Both the in-range percentages and estimated Hba1c demonstrate significant improvement when the patients use Diabits more each day.
This research is just the beginning. Diabits is currently researching the best ways of using our predictive models as part of a closed loop artificial pancreas system that would enable automated, more efficient blood glucose control. Keep checking back for updates on our latest research!