An Artificial Intelligence-Assisted Umbrella Review of Eating Disorder Meta-Analyses (2020–2025): Mapping Treatment Efficacy, Comorbidity, and Persistent Gaps in Sample Diversity

Background/Objectives: Eating disorders (EDs) are serious psychiatric illnesses that result in physical and mental-health impairments. A unifying summary of the broad collection of ED research findings is lacking. This umbrella review leveraged artificial intelligence (AI) to summarize ED-focused meta-analyses, to identify directions for future research and treatment. Methods: PubMed, PsycINFO, Web of Science, Embase, and ProQuest were searched for meta-analyses of human ED research published between 1 June 2020 and 31 May 2025 (inclusive). A large language model (LLM) conducted risk of bias assessment and data extraction for studies meeting all inclusion criteria. Neural topic modeling (BERTopic) identified the themes present across included meta-analyses, and an LLM summarized findings of articles in each category. Results: 220 meta-analyses were included. AI tools demonstrated good accuracy in quality appraisal, data extraction, and data synthesis (97%, 93%, and 96%, respectively). Most meta-analyses had at least one critical flaw (per AMSTAR 2 quality appraisal). Neural topic modeling identified 16 themes within the set of included articles: Neurocognition, Treatment Effectiveness, Intervention Efficacy, Biomarkers, Premorbid Conditions, Clinical Features, Epidemiology, Psychiatric Comorbidity, Behavioral Addictions, Long-Term Outcomes, Prevention, Psychological Factors, Treatment Gaps, COVID-19, Pharmacotherapy, and Avoidant Restrictive Food Intake Disorder. Article summaries collectively identified ED prevalence of 1.69% and associations of EDs with mortality and disability. Studies covered a range of illness mechanisms and treatments, though disproportionately included Western females. Conclusions: This review highlights the broad scope but limited generalizability of ED research, as well as the potential of AI to accelerate systematic review workflows.