Hypoglycemia in T1D is considered the main factor limiting tight glycemic control, with well-documented consequences in morbidity, mortality, and economic cost. Hypoglycemia-related simultaneous glucose-insulin time series clustering has not been examined before. We analyzed data from 392 individuals (CSII and AID) from the T1DEXI dataset. A multivariate time series clustering algorithm on normalized CGM and IOB data was developed, using dynamic time warping and a threshold-based algorithm to discern postprandial (PP) glucose-insulin profiles related to hypoglycemic episodes. ANOVAs and independent t-tests were used to assess within- and between-profile differences of the AUC of IOB (total, basal rate, and bolus), respectively. 7500 (19.4%) out of 38744 3-hour PP periods had at least one hypoglycemic episode. Within this subset, we identified sixteen distinct glucose-insulin clusters, categorized into four behavioral profiles, P1-P4 (Fig. 1). Late PP hypoglycemia (P1), accounting for 34.8% of all episodes, is associated with a higher total basal rate and bolus than the desirable decision-making (P3) and late PP hyperglycemia (P4) profiles; all comparisons with p<0.001.
In conclusion, new glucose-insulin behavioral profiles may provide actionable insight at the population and individual levels to minimize hypoglycemia risk.
L.E. Cypris: None. J.F. Garcia-Tirado: Research Support; Dexcom, Inc. Other Relationship; Dexcom, Inc.
