Digital twin–supported behavioral intention in mothers of young children to prevent childhood obesity: a large language model–based intervention study

Abstract

Introduction Childhood obesity is a major determinant of lifelong non-communicable disease risk, yet early intervention may modify this trajectory. Digital twins and large language models may provide a scalable means of translating individualized risk prediction into understandable and motivating lifestyle guidance. Methods We developed a precision-preventive intervention integrating a childhood-overweight digital twin with an empathic, supervised-fine-tuned open-source nutrition-guidance large language model. The model was based on Llama 3-70B and fine-tuned using the Standard Tables of Food Composition in Japan 2020 and its 2023 amendment. Response appropriateness was evaluated using 30 nutrition-related questions independently assessed by three registered dietitians. The digital twin adopted a published Tohoku Medical Megabank Birth and Three-Generation cohort model predicting overweight risk at 36–47 months, 6 years, 11 years, and 14 years. A total of 121 women with a child under 14 years of age participated in the digital-twin visualization and a single chat-based test. Results The off-the-shelf Llama 3-70B produced a mean of 18 appropriate responses out of 30, whereas the fine-tuned model produced 27 out of 30. The digital-twin visualization and chat consultation were rated highly for clarity, perceived value, behavioral intention, and willingness to continue, with core-item means above 4.9 on a five-point scale. Thematic analysis showed that 112 of 121 participants perceived the recommended lifestyle changes as feasible and achievable gradually. Discussion The integrated intervention demonstrated preliminary operational feasibility and acceptability. Individualized future-risk visualization combined with supportive conversational guidance was associated with high self-reported behavioral intention. Controlled and longitudinal studies are required to determine whether these preliminary motivational responses translate into sustained behavior change and improved health outcomes.