1296-OR: Continuous Glucose Monitoring Metrics Are Associated with Diabetic Ketoacidosis across Pre-Event Time Windows

Introduction and Objective: While diabetic ketoacidosis (DKA) prediction models have been successfully developed using electronic health record (EHR) data, incorporating continuous glucose monitoring (CGM) metrics may further improve risk stratification. To explore this possibility, we examined the association between CGM metrics and DKA events in adults with type 1 diabetes (T1D). Methods: We identified adults with T1D treated in our health system between 1/1/2017 - 3/20/2025 by adapting the SUPREME-DM and Klompas algorithms to our EHR. We identified DKA events using the ADA 2024 Consensus criteria. We included those whose Dexcom CGM data was retrievable from Dexcom Clarity prior to DKA-related hospitalization. For these individuals, we extracted pre-DKA data (i.e., all available CGM data from the 90-day window preceding each DKA event) and non-DKA data (i.e., all available CGM data, up to 450 days, separated ≥90 days from any DKA event). We used mixed-effects models to compare CGM metrics (e.g., time-in-range, variability, Glycemia Risk Index) between pre-DKA periods and matched non-DKA periods across four prespecified time windows (1, 7, 30, and 89 days before DKA). Results: Thirteen adults (mean age 56.5 ± 19.2 years, 46% female) experienced 28 DKA-related hospitalizations. On the day immediately preceding DKA, mean glucose was markedly elevated compared with 1-day non-DKA periods (263 vs. 199 mg/dL, p0.001), time 250 mg/dL doubled (50% vs. 24%, p0.001), and time-in-range was halved (24% vs. 49%, p0.001). Glycemic variability was increased in the 7-day pre-DKA window (mean amplitude of glycemic excursions MAGE 180 vs. 157, p=0.006). In the 30- and 89-day pre-DKA windows, the Hypoglycemia Component of the Glycemia Risk Index was significantly reduced (0.4 vs. 1.0 at 30 days, p=0.012; 0.3 vs. 0.8 at 89 days, p=0.008). Conclusion: CGM metrics were associated with DKA across time windows from 1 to 89 days. Future work will integrate CGM-derived metrics into our EHR-based DKA prediction model to enhance risk stratification. Disclosure J. Kohlenberg: Research Support; Current; Dexcom, Inc. Advisory Panel; Current; Glooko, Inc. Other - Speaker for Med Learning Group (funded by Sanofi); Ended; Sanofi. M. Xu: None. R. Coopergard: None. K.L. Gachet: None. N. Mathioudakis: None. M. Clements: Employee; Current; Glooko, Inc. Research Support; Ended; Dexcom, Inc., Abbott Diabetes. E. Helgeson: Research Support; Current; Fisher Current; Dexcom, Inc. S. Ma: None. Funding NIH NIDDK Diab-Docs K12DK133995