Artificial intelligence in eating disorder research: a scoping review with bibliometric analysis of trends, methodologies, and translational readiness (2016–2025)

Abstract

Abstract Purpose Artificial intelligence (AI) applications in eating disorder research have expanded rapidly, yet the structure, scope, and translational readiness of this literature have not been systematically mapped. This study aimed to characterise publication trends, AI methodologies, clinical application domains, data sources, and implementation readiness across the empirical literature on AI in eating disorders. Methods A scoping review with bibliometric analysis was conducted across three databases (PubMed [MEDLINE], the Web of Science Core Collection, and Scopus) on 20 November 2025. Studies were included if they reported original empirical research employing AI or machine learning as the primary analytical method in an eating disorder population. A total of 66 articles published between 2016 and 2025 were included. Each study was classified according to predefined taxonomies covering AI methodology, clinical application domain, data source type, implementation readiness, eating disorder type, and target population. Bibliometric indicators, including total citations, citations per year, and normalised citation counts, were retrieved from the Web of Science Core Collection. Co-authorship, keyword co-occurrence, bibliographic coupling, and co-citation networks were constructed using Python (version 3.11) with the NetworkX and matplotlib libraries, and reporting quality was appraised using the TRIPOD+AI and CONSORT-AI statements. Results Publication output grew from one article in 2016 to 16 in 2025. Machine learning was the dominant AI methodology (78.8%), followed by natural language processing (19.7%) and a single conversational AI application (1.5%). The most common clinical applications were risk prediction (33.3%), diagnosis (25.8%), and treatment outcome prediction (21.2%). Clinical data (42.4%) and neuroimaging (18.2%) were the most frequently used data sources. The majority of studies remained at the laboratory stage (77.3%), with only two describing clinically deployed systems. The United States contributed 33.3% of articles; no contributions were identified from Africa, South America, or Southeast Asia. Conclusions AI research in eating disorders is growing rapidly but remains geographically concentrated, methodologically dominated by traditional machine learning, and largely confined to laboratory settings. Adolescent populations (despite representing the peak age of eating disorder onset), bulimia nervosa, and low- and middle-income country contexts are systematically underrepresented. Closing the gap between research output and clinical implementation will require independent model validation, prospective evaluation, and sustained multidisciplinary collaboration. Level of Evidence Not applicable. This is a scoping review with bibliometric analysis that maps the scope and structure of a research field rather than evaluating clinical effectiveness; it therefore does not correspond to a traditional level-of-evidence rating.