2926-LB: A CGM-Based Machine Learning Model Predicts Fasting Glycemic Complications during Ramadan

Le résumé fourni par la source

Introduction and Objective: Over 116 million Muslims with diabetes fast during Ramadan each year, placing them at risk of dysglycemia and acute complications. Current pre-Ramadan risk assessment relies largely on complex expert consensus-based tools without quantitative predictive tools. We aimed to develop and internally validate the first CGM-based model to predict fasting glycemic complications Methods: We analyzed data from 550 individuals with diabetes who fasted during Ramadan and used CGM during Ramadan and the preceding month. Candidate predictors included demographic and clinical variables and pre-Ramadan CGM metrics: time in range (TIR), time below range (TBR), glycemia risk index (GRI), and coefficient of variation (CV). The primary outcome was fasting glycemic complications during Ramadan: defined as DKA, an ED visit, fasting interruption on >2 days due to dysglycemia, or Ramadan TIR <70%. A multivariable logistic regression model was developed. Model discrimination was assessed using the area under the receiver operating characteristic curve (AUC), and internal validation used 10-fold cross-validation Results: The median age was 27 years [IQR 19-42]; 60.0% were female; 77.1% had type 1 diabetes. Overall, 410 (74.5%) experienced fasting glycemic complications. Lower pre-Ramadan TIR, higher GRI, diabetes treatment modality, and longer diabetes duration were strong predictors of fasting glycemic complications. A prediction model based solely on pre-Ramadan CGM metrics demonstrated excellent discrimination (AUC of 0.91, 95% CI 0.89-0.94) and maintained performance after internal validation (cross-validated AUC of 0.89; 95% CI 0.87-0.92), comparable to models incorporating an extensive list of demographic and clinical variables Conclusion: This innovative CGM-based prediction model identifies PWD at risk of fasting glycemic complications and provides a simple, scalable, and practical tool for real-world risk stratification and personalized counseling. Disclosure M.E. Al-Sofiani: Speaker's Bureau; Ended; Medtronic, Dexcom, Inc. Research Support; Current; Dexcom, Inc. Research Support; Ended; Medtronic. Speaker's Bureau; Ended; Insulet Corporation, Abbott Diabetes, Sanofi. S.K. Alharthi: None. M. Abusamaan: None. S.A. Meo: None. D. Klonoff: Advisory Panel; Current; Afon Technology, Atropos Health, Embecta, Glooko, Inc., Glucotrack, Lifecare, Inc. Advisory Panel; Ended; Novo Nordisk. Advisory Panel; Current; Sanofi, Synchneuro, Thirdwayv Inc.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.