Design, use and evaluation of prediction methods for blood glucose concentration
Sprache des Titels:
The research on algorithms for prediction of future blood glucose (BG) concentration based on historic data has started as continuous glucose measurement systems have became available more than a decade ago. Currently, many different approaches can be found in the literature and can be categorized according to the way how the prediction is computed mathematically and how the prediction algorithm is designed. The categorization can also be done conceptually: whether the prediction depends only on historical BG data or if other available information such as insulin doses or meal carbohydrates is utilized as well. Regardless of the categorization, there is the need for an independent evaluation of the quality of the predictions. Several performance metrics were proposed but currently there is no consensus on which ones are the most significant. Additionally, performance metrics themselves can vary significantly depending on the specific measurement conditions.