University of Twente Student Theses


Identification of Type 1 Diabetes Phenotypes through Glycemic Features based on Continuous Glucose Monitoring Data

Have, Laura ten (2023) Identification of Type 1 Diabetes Phenotypes through Glycemic Features based on Continuous Glucose Monitoring Data.

[img] PDF
Abstract:Objective: No personalized therapy options are yet available for T1D patients. It is believed that CGM data could possibly be the source for distinguishing different phenotypes which could eventually aid developing health care. However no studies were yet able to identify and validate distinct clinically relevant CGM-based phenotypes. Therefore, the aim of the current study is to distinguish different T1D phenotypes based on cluster analyses to ultimately contribute to better T1D health care. Methods: The study sample included patients from the DIABASE cohort who were clustered based on 20 selected CGM features measuring hypoglycemia, hyperglycemia and glycemic variability during different times of the day. Agglomerative hierarchical clustering was used as the clustering method in combination with the ’agnes’ function in R, with Euclidean distance as the distance metric and Ward’s minimum variance method as the linkage method. The optimal number of clusters was determined based on analysis of the elbow plot and the dendrogram. Statistical analysis of the numeric features for overall-tests of difference was carried out using the Kruskal-wallis test. Inter-cluster testing was performed with the Dunn’s test. An extra method of validation called the silhouette value was used to analyse the placement of patients in the clusters. Results: The study sample included 78 adult patients with T1D (53,56% male, mean age 53 years). Five clusters were identified with overall significant differences (p<0,05), inter-cluster differences were not all significant, especially between cluster 1 and 5. Cluster 1 had overall moderate feature values which stayed closed to the average values of all patients. The silhouette plot showed that not every patients was correctly placed in this cluster. Cluster 2 was characterized by the lowest hypoglycemia metrics and severe hyperglycemic incidence. Cluster 3 showed severe hyperglycemic exposure compared to other clusters and had the highest mean glucose value. Cluster 4 was characterized with the lowest hyperglycemia metrics, lowest glycemic variability and the lowest mean glucose value. Cluster 5 showed severe hypoglycemia and glycemic variability and the silhouette width indicated that patients were placed incorrect. Conclusion: The current study showed that CGM data analysis can be used to discover different subgroups of T1D, which could eventually result in more personalized health care. Five clusters were identified by using agglomerative hierarchical clustering analysis based on glycemic features. However, to validate if distinct phenotypes of T1D were found, additional research is needed.
Item Type:Essay (Bachelor)
Faculty:TNW: Science and Technology
Subject:44 medicine
Programme:Biomedical Technology BSc (56226)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page