Active meal logging using an AI-driven food intelligence model was associated with a 30.04% increase in Time in Range over 14 days, compared to a 3.32% increase in non-loggers.
Observational (n=18,703)
Does AI-driven food image logging improve CGM-derived glycemic metrics in people with diabetes?
AI-enabled food image logging significantly improves continuous glucose monitoring metrics, including time in range, in people with diabetes over a 14-day period.
Absolute Event Rate: 30.04% vs 3.32%
p-value: p=<0.001
Introduction and Objective: Traditional food logging in diabetes care relies on manual self-reporting, limiting adherence and clinical utility. Image-based logging improves usability, but its impact depends on converting logs into personalized, actionable insights. This study evaluates an AI-driven food intelligence model that transforms food images into personalized dietary insights and assesses its effect on glycemic outcomes in people with diabetes. Methods: AI model was developed using over 100,000 real-world food logs from people with diabetes, integrating user persona, biomarkers, lifestyle, comorbidities, dietary preferences, and cultural eating patterns. The model evaluates composite meals per eating occasions to generate personalized food scores, healthier alternatives, and predicts post-prandial glucose impact. Glycemic outcomes were assessed over a 14-day CGM period, with Week 1 (W1) as baseline exposure to AI insights and Week 2 (W2) as continued engagement, comparing active meal loggers with non-loggers. Results: Over 14 days, active meal loggers (n=8,421) demonstrated greater percentage improvements in glycemic metrics compared with non-loggers (n=10,282). From W1 to W2, active loggers showed a 30.04% increase in TIR, a 14.45% reduction in TBR, and a 12.08% reduction in TAR (all p 0.001). In contrast, non-loggers exhibited a smaller 3.32% increase in TIR (p 0.001), a substantial upward shift in TBR (21.53%; p 0.001), and a modest upward shift in TAR (4.57%; p 0.001). Conclusion: This study shows that active food logging also improved food scores by 10.25% compared to the first week, contributing to improved glycemic control in the second week through enhanced meal-level awareness and timely dietary adjustments, and reinforcing its role in personalized dietary management, with potential added benefits from meal-specific glycemic index and glycemic load estimation. Disclosure S. Kumar: None. A.S. Shukla: None. A.M. Raymond: None. A. Sequeira: None. J. Joseline: None. C.G. Mehra: None.
KUMAR et al. (Fri,) conducted a observational in Diabetes (n=18,703). Active meal logging using an AI-driven food intelligence model vs. Non-loggers was evaluated on Percentage increase in Time in Range (TIR) from Week 1 to Week 2 (p=<0.001). Active meal logging using an AI-driven food intelligence model was associated with a 30.04% increase in Time in Range over 14 days, compared to a 3.32% increase in non-loggers.