{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,4]],"date-time":"2026-09-04T06:12:54Z","timestamp":1788502374097,"version":"build-2803163510"},"reference-count":45,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,4]],"date-time":"2026-09-04T00:00:00Z","timestamp":1788480000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/100007366","name":"Fondazione Italiana Sclerosi Multipla","doi-asserted-by":"publisher","award":["2024\/BR\/005"],"award-info":[{"award-number":["2024\/BR\/005"]}],"id":[{"id":"10.13039\/100007366","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Accessing large-scale clinical and biomedical databases remains a significant barrier for clinicians and researchers, requiring substantial computational expertise. Agentic artificial intelligence frameworks, in which large language models (LLMs) orchestrate multi-step reasoning and query execution under interactive human supervision, offer the potential to democratize data access and accelerate evidence generation.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>We applied the Model Context Protocol (MCP) to integrate, within a single agentic workflow, an Observational Medical Outcomes Partnership (OMOP)-standardized electronic health record (EHR) database from an academic health system (&amp;gt;7 million subjects), with the Scalable Precision Medicine Open Knowledge Engine (SPOKE), a curated knowledge graph integrating relationships biomedical concepts from expert-maintained resources. Using this implementation (MedCP), we evaluated the approach across 100 benchmarking clinical research tasks, 617 biomedical factual accuracy questions (BiomixQA), an integrative case study linking clinical co-occurrence with molecular similarity, and a standardized protocol for generating and replicating real-world studies.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>Across the 100-task benchmark, knowledge-graph access improved mean scores overall for GPT-5.5 and Claude Opus 4.8, though the pattern differed between models; on BiomixQA, SPOKE grounding raised multiple-choice accuracy for both models, without changing true\/false accuracy. In the case study, disease co-occurrence patterns extracted from the EHR correlated with molecular network similarity, surfacing mechanistic hypotheses from real-world data. Applied to the replication of published observational studies, the research protocol compressed timelines from months to hours, lowering the technical barrier to query generation and execution while study design and interpretation remained under expert supervision.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusions<\/jats:title>\n                    <jats:p>An agentic AI infrastructure that combines institutional EHR data with curated biomedical knowledge via MCP can serve as a transparent, domain-grounded, supervised research assistant for real-world evidence, supporting both hypothesis generation and testing.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/frai.2026.1883853","type":"journal-article","created":{"date-parts":[[2026,9,4]],"date-time":"2026-09-04T05:30:13Z","timestamp":1788499813000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["An agentic AI framework connecting language models to electronic health records and a biomedical knowledge graph for real-world evidence"],"prefix":"10.3389","volume":"9","author":[{"given":"Gianmarco","family":"Bellucci","sequence":"first","affiliation":[{"name":"Department of Neurology, Weill Institute for Neurosciences, University of California San Francisco, San Francisco","place":["CA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Wanjun","family":"Gu","sequence":"additional","affiliation":[{"name":"Department of Neurology, Weill Institute for Neurosciences, University of California San Francisco, San Francisco","place":["CA, United States"]},{"name":"Graduate Program in Biomedical Informatics, University of California San Francisco, San Francisco","place":["CA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Peter W.","family":"Rose","sequence":"additional","affiliation":[{"name":"San Diego Supercomputer Center, University of California San Diego, La Jolla","place":["CA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sergio E.","family":"Baranzini","sequence":"additional","affiliation":[{"name":"Department of Neurology, Weill Institute for Neurosciences, University of California San Francisco, San Francisco","place":["CA, United States"]},{"name":"Graduate Program in Biomedical Informatics, University of California San Francisco, San Francisco","place":["CA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,4]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"1773","DOI":"10.1038\/s41591-022-01981-2","article-title":"Multimodal biomedical AI","volume":"28","author":"Acosta","year":"2022","journal-title":"Nat. Med."},{"key":"B2","doi-asserted-by":"publisher","first-page":"1443","DOI":"10.1111\/dom.70336","article-title":"Comparative effectiveness of sodium-glucose cotransporter-2 inhibitors and glucagon-like peptide-1 receptor agonists for incident dementia: a retrospective multicohort study","volume":"28","author":"Anagnostakis","year":"2026","journal-title":"Diabetes Obes. Metab."},{"key":"B3","doi-asserted-by":"publisher","first-page":"56","DOI":"10.1038\/nrg2918","article-title":"Network medicine: a network-based approach to human disease","volume":"12","author":"Barab\u00e1si","year":"2011","journal-title":"Nat. Rev. Genet."},{"key":"B4","doi-asserted-by":"publisher","first-page":"e1912869","DOI":"10.1001\/jamanetworkopen.2019.12869","article-title":"Feasibility of using real-world data to replicate clinical trial evidence","volume":"2","author":"Bartlett","year":"2019","journal-title":"JAMA Netw. Open"},{"key":"B5","doi-asserted-by":"publisher","first-page":"248","DOI":"10.1002\/ana.78033","article-title":"Comparative safety profiles of ocrelizumab and rituximab in multiple sclerosis treatment using real-world evidence","volume":"99","author":"Cerono","year":"2026","journal-title":"Ann. Neurol."},{"key":"B6","doi-asserted-by":"publisher","first-page":"6","DOI":"10.1007\/s44250-026-00343-w","article-title":"Transforming clinical medicine with multimodal artificial intelligence, agentic systems, and the model-context protocol: a perspective on future directions","volume":"5","author":"Choi","year":"2026","journal-title":"Discov Health Syst."},{"key":"B7","doi-asserted-by":"publisher","first-page":"94","DOI":"10.7861\/futurehosp.6-2-94","article-title":"The potential for artificial intelligence in healthcare","volume":"6","author":"Davenport","year":"2019","journal-title":"Future Healthc. J."},{"key":"B8","unstructured":"DeFalco\n              F.\n            \n            \n              Schuemie\n              M.\n            \n            \n              Sena\n              A.\n            \n            \n              Adulyanukosol\n              N.\n            \n            \n              Liu\n              S.\n            \n            \n              Black\n              A.\n            \n          \n          Eunomia: Standard Dataset Manager for Observational Medical Outcomes Partnership Common Data Model Sample Datasets\n          \n          2025"},{"key":"B9","doi-asserted-by":"crossref","DOI":"10.1109\/AIIoT65859.2025.11105280","article-title":"\u201cEnhancing clinical decision support and EHR insights through LLMs and the model context protocol: an open-source MCP-FHIR framework,\u201d","volume-title":"Proceedings of the 2025 IEEE World AI IoT Congress (AIIoT), Seattle, WA, USA","author":"Ehtesham","year":"2025"},{"key":"B10","doi-asserted-by":"publisher","first-page":"giae082","DOI":"10.1093\/gigascience\/giae082","article-title":"Knowledge graph\u2013based thought: a knowledge graph\u2013enhanced LLM framework for pan-cancer question answering","volume":"14","author":"Feng","year":"2025","journal-title":"GigaScience"},{"key":"B11","doi-asserted-by":"publisher","first-page":"7526","DOI":"10.1038\/s41467-025-62781-z","article-title":"Large language model powered knowledge graph construction for mental health exploration","volume":"16","author":"Gao","year":"2025","journal-title":"Nat. Commun."},{"key":"B12","doi-asserted-by":"publisher","first-page":"e58670","DOI":"10.2196\/58670","article-title":"Leveraging medical knowledge graphs into large language models for diagnosis prediction: design and application study","volume":"4","author":"Gao","year":"2025","journal-title":"JMIR AI"},{"key":"B13","doi-asserted-by":"publisher","first-page":"61","DOI":"10.1016\/j.neuron.2006.09.011","article-title":"The neurobiology of multiple sclerosis: genes, inflammation, and neurodegeneration","volume":"52","author":"Hauser","year":"2006","journal-title":"Neuron"},{"key":"B14","doi-asserted-by":"publisher","first-page":"1345","DOI":"10.1056\/NEJMp2113319","article-title":"Methods of public health research\u2014strengthening causal inference from observational data","volume":"385","author":"Hern\u00e1n","year":"2021","journal-title":"N. Engl. J. Med."},{"key":"B15","doi-asserted-by":"publisher","first-page":"2446","DOI":"10.1001\/jama.2022.21383","article-title":"Target trial emulation","volume":"328","author":"Hern\u00e1n","year":"2022","journal-title":"JAMA"},{"key":"B16","doi-asserted-by":"publisher","first-page":"574","DOI":"10.3233\/978-1-61499-564-7-574","article-title":"Observational health data sciences and informatics (OHDSI): opportunities for observational researchers","volume":"216","author":"Hripcsak","year":"2015","journal-title":"Stud. Health Technol. Inform."},{"key":"B17","doi-asserted-by":"publisher","first-page":"AIoa2400555","DOI":"10.1056\/AIoa2400555","article-title":"Autonomous LLM-driven research\u2014from data to human-verifiable research papers","volume":"2","author":"Ifargan","year":"2025","journal-title":"NEJM AI"},{"key":"B18","doi-asserted-by":"publisher","DOI":"10.1056\/AIdbp2500144","article-title":"MedAgentBench: a virtual EHR environment to benchmark medical LLM agents","author":"Jiang","year":"2025","journal-title":"NEJM AI"},{"key":"B19","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1038\/s41597-022-01899-x","article-title":"MIMIC-IV, a freely accessible electronic health record dataset","volume":"10","author":"Johnson","year":"2023","journal-title":"Sci. Data"},{"key":"B20","author":"Kallfelz","year":"2021","journal-title":"MIMIC-IV Demo Data in the OMOP Common Data Model"},{"key":"B21","doi-asserted-by":"publisher","DOI":"10.1101\/2025.02.28.25323115","article-title":"Medical Hallucinations in Foundation Models and Their Impact on Healthcare","author":"Kim","year":"2025","journal-title":"medRxiv [Preprint]"},{"key":"B22","doi-asserted-by":"publisher","first-page":"026120","DOI":"10.1103\/PhysRevE.73.026120","article-title":"Vertex similarity in networks","volume":"73","author":"Leicht","year":"2006","journal-title":"Phys. Rev. E"},{"key":"B23","author":"Masayoshi","year":"2025","journal-title":"EHR-MCP: Real-World Evaluation of Clinical Information Retrieval by Large Language Models via Model Context Protocol."},{"key":"B24","doi-asserted-by":"publisher","first-page":"btae353","DOI":"10.1093\/bioinformatics\/btae353","article-title":"KRAGEN: a knowledge graph-enhanced RAG framework for biomedical problem solving using large language models","volume":"40","author":"Matsumoto","year":"2024","journal-title":"Bioinformatics"},{"key":"B25","doi-asserted-by":"publisher","first-page":"1257601","DOI":"10.1126\/science.1257601","article-title":"Uncovering disease-disease relationships through the incomplete interactome","volume":"347","author":"Menche","year":"2015","journal-title":"Science"},{"key":"B26","doi-asserted-by":"publisher","first-page":"128","DOI":"10.1038\/s44172-025-00453-y","article-title":"Explainability in the age of large language models for healthcare","volume":"4","author":"Mesinovic","year":"2025","journal-title":"Commun. Eng."},{"key":"B27","unstructured":"Introduction. Model Context Protocol\n          \n          2025"},{"key":"B28","doi-asserted-by":"publisher","first-page":"btad080","DOI":"10.1093\/bioinformatics\/btad080","article-title":"The scalable precision medicine open knowledge engine (SPOKE): a massive knowledge graph of biomedical information","volume":"39","author":"Morris","year":"2023","journal-title":"Bioinformatics"},{"key":"B29","doi-asserted-by":"publisher","first-page":"AIra2400380","DOI":"10.1056\/AIra2400380","article-title":"RAG in health care: a novel framework for improving communication and decision-making by addressing LLM limitations","volume":"2","author":"Ng","year":"2025","journal-title":"NEJM AI"},{"key":"B30","unstructured":"OMOP Common Data Model.\n          \n          2025"},{"key":"B31","doi-asserted-by":"publisher","first-page":"4464","DOI":"10.1038\/s41380-022-01722-4","article-title":"Behavioural and functional evidence revealing the role of RBFOX1 variation in multiple psychiatric disorders and traits","volume":"27","author":"O'Leary","year":"2022","journal-title":"Mol. Psychiatry"},{"key":"B32","doi-asserted-by":"publisher","first-page":"330","DOI":"10.1038\/s43856-025-01021-3","article-title":"Multi-model assurance analysis showing large language models are highly vulnerable to adversarial hallucination attacks during clinical decision support","volume":"5","author":"Omar","year":"2025","journal-title":"Commun. Med."},{"key":"B33","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/btag216","article-title":"eGoT: enhanced graph-of-thoughts for multi-hop knowledge retrieval and hypothesis generation in biomedicine","volume":"42","author":"Sanda","year":"2026","journal-title":"Bioinformatics"},{"key":"B34","doi-asserted-by":"publisher","first-page":"172","DOI":"10.1038\/s41586-023-06291-2","article-title":"Large language models encode clinical knowledge","volume":"620","author":"Singhal","year":"2023","journal-title":"Nature"},{"key":"B35","doi-asserted-by":"publisher","first-page":"btae560","DOI":"10.1093\/bioinformatics\/btae560","article-title":"Biomedical knowledge graph-optimized prompt generation for large language models","volume":"40","author":"Soman","year":"2024","journal-title":"Bioinformatics"},{"key":"B36","doi-asserted-by":"publisher","first-page":"4114","DOI":"10.1038\/ncomms5114","article-title":"A fast and unbiased procedure to randomize ecological binary matrices with fixed row and column totals","volume":"5","author":"Strona","year":"2014","journal-title":"Nat. Commun."},{"key":"B37","author":"Su","year":"2025","journal-title":"KGARevion: An AI Agent for Knowledge-Intensive Biomedical QA"},{"key":"B38","unstructured":"Observational Health Data Sciences and Informatics\n          \n          2021"},{"key":"B39","doi-asserted-by":"publisher","first-page":"AIoa2300138","DOI":"10.1056\/AIoa2300138","article-title":"Towards generalist biomedical AI","volume":"1","author":"Tu","year":"2024","journal-title":"NEJM AI"},{"key":"B40","doi-asserted-by":"publisher","first-page":"4840","DOI":"10.1109\/ACCESS.2026.3651218","article-title":"Agentic AI in healthcare and medicine: a seven-dimensional taxonomy for empirical evaluation of LLM-based agents","volume":"14","author":"Vatsal","year":"2026","journal-title":"IEEE Access"},{"key":"B41","doi-asserted-by":"publisher","first-page":"1142","DOI":"10.1038\/s42256-026-01266-0","article-title":"Empowering biomedical evidence exploration and synthesis with deep knowledge graph research","volume":"8","author":"Wang","year":"2026","journal-title":"Nat. Mach. Intell."},{"key":"B42","doi-asserted-by":"publisher","first-page":"199","DOI":"10.1142\/9789819807024_0015","article-title":"Improving retrieval-augmented generation in medicine with iterative follow-up questions","volume":"2025","author":"Xiong","year":"2025","journal-title":"Biocomputing"},{"key":"B43","doi-asserted-by":"publisher","first-page":"894","DOI":"10.3390\/info16100894","article-title":"Large language model agents for biomedicine: a comprehensive review of methods, evaluations, challenges, and future directions","volume":"16","author":"Xu","year":"2025","journal-title":"Information"},{"key":"B44","article-title":"\u201cReAct: synergizing reasoning and acting in language models,\u201d","author":"Yao","year":"2023","journal-title":"International Conference on Learning Representations"},{"key":"B45","doi-asserted-by":"publisher","first-page":"20552076261427503","DOI":"10.1177\/20552076261427503","article-title":"Performance of latest AI models, RAG, and MCP on lung cancer-related questions","volume":"12","author":"Zhao","year":"2026","journal-title":"Digit Health"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1883853\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,4]],"date-time":"2026-09-04T05:30:17Z","timestamp":1788499817000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1883853\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,4]]},"references-count":45,"alternative-id":["10.3389\/frai.2026.1883853"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1883853","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,4]]},"article-number":"1883853"}}