Abstract:
Characterization of glycemic patterns in individuals with type 1 diabetes is addressed in this work. Unlike other methodologies, no insulin or meal data will be considere...Show MoreMetadata
Abstract:
Characterization of glycemic patterns in individuals with type 1 diabetes is addressed in this work. Unlike other methodologies, no insulin or meal data will be considered available. This makes the method presented suitable for all CGM users, independently of the insulin delivery device used. As well, no manual input is required from the patient potentially increasing usability and adherence. Data of 44 type 1 diabetic patients, from 26-week study were used in this work. An unsupervised clustering algorithm based on Fuzzy C-Means was applied to classify event-to-event segments of CGM data. Events defining data partitioning were automatically generated based on: (1) an automatic meal detection algorithm (for day periods) and (2) time of day (for night periods). After pre-filtering the dataset, 985 meal segments were detected and 180 periods were labelled as night per patient on average. The optimal number of clusters was determined by optimising clusters' compactness and separation metrics, as given by the Fukuyama-Sugeno index. An average of 8 clusters for postprandial segments and 5 for nights was obtained, although, depending on patients' habits and the amount of training data, these numbers may vary between patients. For the sake of comparison, the method was also applied to CGM data partitioned based on the mealtime manually reported by the patient. The use of a meal detection algorithm instead proved superior performance in terms of reduced within-cluster variability and interpretation of the resulting prototypes, besides freeing from patient intervention. Clusters described CGM data segments with differentiated glycemic control metrics, such as time in range (TIR), with average values across all patients of \mathbf{45.2}\pm \mathbf{15}\ \% \ (\mathbf{mean}\pm\mathbf{SD}) for detected meals and \mathbf{49.9}\pm \mathbf{13.6}\ {\%} for announced meals data.
Date of Conference: 23-25 November 2022
Date Added to IEEE Xplore: 03 January 2023
ISBN Information: