Abstract
Background and Aims
Accurate predictions of blood glucose (BG) concentration for large prediction horizons (PH) might improve type-1 diabetes therapy by allowing patients to adjust the therapy based on BG future values. Identification of seasonal local models after clustering BG data enhances other BG prediction approaches when used offline on available data sets. However, a methodology for the use under free-living conditions is needed.
Methods
Long-term full days BG historical data including post-prandial and nocturnal periods is partitioned into a set of event-to-event time subseries, driven by meals and night periods, and seasonality is enforced. Preprocessed data are then clustered into similar glycemic profiles and a seasonal model is identified for each cluster. The online BG prediction is obtained by local model integration through real-time membership-to-cluster estimation. As a proof of concept, the framework is tested over 6 months data of UVA/Padova simulator extended with several variability sources. Additionally to the BG prediction, an online monitoring system informs about prediction confidence and abnormal behavior detection.
Results
The framework exhibits high prediction accuracy for large PHs: a MAPE of 4.10%, 5.95%, 8.43%, 11.32%, 13.65%, and 13.97% has been achieved for 15-, 30-, 60-, 120-, 180-, and 240-min PHs, respectively.
Conclusions
The proposed system allows the use of seasonal models for BG prediction under free-living conditions, and therefore allows diabetic patients to anticipate therapeutic decisions and detect abnormal states.
