LEON Farhy, United States of America
University of Virginia School of Medicine MedicinePresenter of 1 Presentation
CLUSTERING CGM DAILY PROFILES IN THE INTERNATIONAL DIABETES CLOSED-LOOP (IDCL) TRIAL
Abstract
Background and Aims
Each type 1 diabetes patient undergoes a recurrent process of glycemic variation described by a sequence of CGM daily profiles. The goals of this study is to develop a methodology to approximate this process based on clustering of CGM daily profiles. The new technology is designed to evaluate the dynamics of glycemic control and is applicable to treatment optimization and decision support.
Methods
We use data from iDCL Protocol 3 (NCT03563313) which generated over 30,000 CGM daily profiles and compared Closed Loop Control (CLC) vs. Sensor Augmented Pump (SAP). Using unsupervised algorithm daily CGM profiles were classified in 3 clusters with the following cluster centers:
| Cluster 1: Tight Control/Intensive Treatment | Cluster 2: Hyperglycemic exposure | Cluster 3: Intermediate/Average Control | |
| Mean Blood Glucose (BG) | 234.76 | 138.24 | 175.90 |
| SD of BG | 72.14 | 38.70 | 57.74 |
| Low BG index | 0.17 | 0.73 | 0.41 |
| High BG index | 20.74 | 3.46 | 9.38 |
Results
The table below presents the classification of CGM daily profiles in the 3 clusters for SAP and CLC groups. All study participants went through all clusters but CLC resulted in 19.1% more time in tight control, and less time in both average control and hyperglycemia.
| Comparing % time in each cluster, SAP vs. CLC | SAP | CLC | p-value |
| Cluster 1: Tight Control/Intensive Treatment | 41.7% | 60.8% | <0.001 |
| Cluster 2: Hyperglycemic exposure | 15.2% | 5.8% | <0.001 |
| Cluster 3: Intermediate/Average Control | 43.1% | 33.4% | <0.001 |
Conclusions
A novel technique was developed to classify daily CGM profiles into separable clusters. While each person goes through all/most clusters during their routin, clusters sequences differentiate treatment modalities, CLC vs SAP.