Language models in nutrition and dietetics: a scoping review

Language models are a form of generative artificial intelligence that excel at natural language understanding and generation, with possible applications for providing evidence-based nutrition information in an interactive and personalized manner. However, they also pose risks to public health related to inaccuracy, bias, and data protection. It is unclear what quantity or type of research has been conducted on these models applied within nutrition and dietetics.

To identify the ways in which language models are being applied to nutrition and dietetics and the associated opportunities and challenges.

The review was conducted in accordance with a predefined protocol and included original studies from peer-reviewed journals. The databases Medline, EMBASE, PsycINFO, CINAHL, PubMed, Global Health, Web of Science, Scopus, DBLP, and IEEE Xplore were searched, and forward and backward citation chasing was conducted for included articles.

There were 2118 articles identified, and after screening 40 remained for data charting. All studies were published between 2022 and 2024. A total of 23 different language models were used with 9 studies creating domain-specific models using retrieval augmented generation or fine-tuning. A total of 17 nutrition-related diseases were considered. The tasks assigned to the language models included extraction or summarization of data from the scientific literature or electronic health records (n = 5), estimation of food composition (n = 9), creation of meal plans (n = 7), and provision of dietary advice (n = 21). Most studies (n = 38) focused on evaluating the models in hypothetical situations such as researcher-designed case studies or questions.

This is an emerging area with limited evidence for real-world implementations or domain-specific models in nutrition and dietetics. Future work is required in real-world settings to strengthen this evidence base and to broaden it to incorporate the views of nutritionists, dietitians, people with medical conditions, and the general public.