We conducted the CDDIAB study at the Montpelier University Hospital with Pr. Renard, and in partnership with INSERM and CNRS, to evaluate our machine learning approach to predicting blood glucose levels of each individual from a collection of personal blood glucose measurements and contextual data.
14 type 1 diabetic patients were included, treated via multiple daily injections or pump therapy and using a continuous glucose monitoring device. One month of data was collected for each patient, with no external intervention. The datasets gathered allowed us to build personalized predictive models and validate accuracy by comparing the predicted values with actual blood glucose levels from the sensor.
The results of the study are extremely satisfying up to 90 minutes prediction horizon, and an abstract based on the study is being published at ATTD 2019 (details below).
Integrated inside an app to support decision-making, this open-loop technology will help patients anticipate hypo- and hyper-glycaemia in particular during night time. It could also be used on top of an Artificial Pancreas particularly during unstable phases with rapid glucose changes.