2697-LB: AI-Based Clustering of Hypoglycemic Clamp Responses Identifies Awareness Status in Subjects with Type 1 Diabetes

Introduction and Objective: We applied an AI-driven clustering approach to clamp responses to determine whether integrated symptom and physiological responses can more objectively distinguish individuals with Impaired awareness of hypoglycemia (IAH) . Methods: 19 individuals with T1D underwent a stepped clamp with glucose sequentially lowered to 90, 65, 55, and 45 mg/dL. K-means clustering was applied using symptoms alone, physiological responses alone, and combined responses. Clustering performance was evaluated using both absolute responses and response changes Results: Clustering using response changes separated participants into two groups consistent with preserved hypoglycemia awareness (n=10) and impaired awareness (n=9). Analysis using responses at 45 mg/dL produced a nearly identical partition (11 vs 8), with 18 of 19 individuals assigned consistently across models. Symptom responses were the primary drivers of cluster separation. The Δ90→45 analysis captures the fold change and produces the clearest overall separation of awareness phenotypes when using all responses (Figure 1). Conclusion: AI-based clustering of hypoglycemic clamp responses identifies awareness status in T1D using integrated physiological and symptom signals. Compared with questionnaire-based approaches such as the GOLD score, AI-driven phenotyping provides an objective framework that may more accurately distinguish individuals with IAH.