(P03-11-26) When Food Data Gaps Limit AI-Driven Nutrition Advice: Insights From a Process Evaluation of AI-Based Nutrition Consultant System in Viet Nam

Objectives: Artificial intelligence (AI) has increasing potential to support nutrition service delivery. However, evidence on the feasibility, acceptability, and perceived usefulness of AI-driven healthcare provision tools in low- and middle-income countries (LMICs) remains limited. This study aims to address this gap and provide a process evaluation of a newly developed AI-applied nutrition consultation system in Viet Nam.

Methods: An AI-powered nutrition consultation system was developed using a Vietnamese food composition database comprising 1,522 foods and anthropometric reference data from over 125,000 individuals stratified by age and sex. The system was pilot tested in approximately 6,000 community users and 4,290 consultations across four hospitals. Process evaluation data were collected as part of the pilot test through system usage logs, structured user feedback, and surveys of participating health professionals to assess usability, response time, perceived accuracy, and usefulness in nutrition assessment and meal planning.

Results: The system was perceived as feasible and acceptable for both community and clinical use. Most users reported satisfactory response time and a user-friendly interface. The system successfully supported dietary intake assessment, identification of nutrition-related conditions, and meal planning during all recorded consultations, with performance rated as good or fair in 33.3% and 60% of cases, respectively. 86.7% health professionals agreed that the system was useful and supportive of their routine work. Meal planning was identified as the most challenging component, primarily due to gaps and variability within food database.

Conclusions: This process evaluation shed light on AI application in nutrition consultation in Viet Nam, suggesting its overall feasibility and acceptability across both healthcare providers and community users in LMICs. Further refinement of food data and validation studies are needed to strengthen accuracy and scalability.

Funding Sources: This work was funded by Vietnam Ministry of Science and Technology.