{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T05:03:14Z","timestamp":1789448594873,"version":"build-2803163510"},"reference-count":36,"publisher":"Springer Science and Business Media LLC","issue":"1","license":[{"start":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T00:00:00Z","timestamp":1789430400000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T00:00:00Z","timestamp":1789430400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"funder":[{"name":"Advancing Healthier Wisconsin Fund"}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["J Med Syst"],"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Diagnosis in real clinical settings is rarely finalized after a single encounter. Patients with complex conditions move through a referral network from primary care to specialists because physicians cannot maintain expertise across all diseases. Yet most diagnostic prediction tools apply a single generalist model across thousands of medical codes, which dilutes representational capacity for rare conditions and produces outputs poorly aligned with how clinicians actually work and reason. We present CPS-Net, a decentralized agent swarm that mirrors medical specialization hierarchies. The system connects specialized transformers trained on focused patient cohorts to language model agents organized into primary care, specialty, and subspecialty tiers. Agents collaborate without central supervision, communicating through shared case memory to make referral decisions transparent and auditable. We trained 41 transformers on Electronic Health Records (EHR) from 75,000 patients across 10 specialties and 30 subspecialties at Froedtert Hospital. Evaluated on 3,000 test cases, the framework achieved top-1, top-3, and top-5 accuracies of 48.8%, 71.00%, and 77.76% with 94.17% clinical relevance, substantially outperforming a monolithic transformer with 19.20%, 38.60% and 47.20% in top-1, top-3 and top-5 accuracy trained on all patients and a single language model agent baseline which achieved 12.70%, 26.40%, and 37.23% accuracy. This work demonstrates that aligning artificial intelligence architecture with clinical workflows improves both prediction accuracy and interpretability for diagnostic support.<\/jats:p>","DOI":"10.1007\/s10916-026-02460-8","type":"journal-article","created":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T04:37:19Z","timestamp":1789447039000},"update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["CPS-Net (Collaborating Physicians in Silico Network): A Decentralized Multi-Agent Transformer Framework for Specialty-Aware Disease Prediction"],"prefix":"10.1007","volume":"50","author":[{"given":"Mohammad Assadi","family":"Shalmani","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Masoud","family":"Khani","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Michael S.","family":"Harris","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Qiang","family":"Lu","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3900-643X","authenticated-orcid":false,"given":"Jake","family":"Luo","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"297","published-online":{"date-parts":[[2026,9,15]]},"reference":[{"issue":"1","key":"2460_CR1","doi-asserted-by":"publisher","first-page":"33","DOI":"10.1186\/s13012-025-01445-4","volume":"20","author":"C Derksen","year":"2025","unstructured":"C. Derksen et al., \u201cThe implementation challenge of computerised clinical decision support systems for the detection of disease in primary care: systematic review and recommendations,\u201d Implement. Sci., vol. 20, no. 1, p. 33, 2025, doi: https:\/\/doi.org\/10.1186\/s13012-025-01445-4.","journal-title":"Implement. Sci."},{"key":"2460_CR2","doi-asserted-by":"publisher","first-page":"b946","DOI":"10.1136\/bmj.b946","volume":"338","author":"C Heneghan","year":"2009","unstructured":"C. Heneghan et al., \u201cDiagnostic strategies used in primary care,\u201d BMJ, vol. 338, p. b946, 2009, doi: https:\/\/doi.org\/10.1136\/bmj.b946.","journal-title":"BMJ"},{"issue":"3","key":"2460_CR3","first-page":"215","volume":"51","author":"CB Forrest","year":"2002","unstructured":"C. B. Forrest, P. A. Nutting, B. Starfield, and S. von Schrader, \u201cFamily physicians\u2019 referral decisions: results from the ASPN referral study,\u201d J. Fam. Pract., vol. 51, no. 3, pp. 215\u2013222, 2002.","journal-title":"J. Fam. Pract."},{"key":"2460_CR4","doi-asserted-by":"publisher","unstructured":"L. Shi, \u201cThe impact of primary care: a focused review,\u201d Scientifica, vol. 2012, p. 432892, 2012, doi: https:\/\/doi.org\/10.6064\/2012\/432892.","DOI":"10.6064\/2012\/432892"},{"issue":"17","key":"2460_CR5","doi-asserted-by":"publisher","first-page":"1364","DOI":"10.1001\/jama.279.17.1364","volume":"279","author":"RA Rosenblatt","year":"1998","unstructured":"R. A. Rosenblatt, L. G. Hart, L. M. Baldwin, L. Chan, and R. Schneeweiss, \u201cThe generalist role of specialty physicians: is there a hidden system of primary care?,\u201d JAMA, vol. 279, no. 17, pp. 1364\u20131370, 1998, doi: https:\/\/doi.org\/10.1001\/jama.279.17.1364.","journal-title":"JAMA"},{"issue":"9","key":"2460_CR6","doi-asserted-by":"publisher","first-page":"e253192","DOI":"10.1001\/jamahealthforum.2025.3192","volume":"6","author":"R Karadakic","year":"2025","unstructured":"R. Karadakic et al., \u201cSubspecialization of surgical specialties in the US,\u201d JAMA Health Forum, vol. 6, no. 9, p. e253192, 2025, doi: https:\/\/doi.org\/10.1001\/jamahealthforum.2025.3192.","journal-title":"JAMA Health Forum"},{"issue":"3","key":"2460_CR7","doi-asserted-by":"publisher","first-page":"e190096","DOI":"10.1001\/jamanetworkopen.2019.0096","volume":"2","author":"ML Barnett","year":"2019","unstructured":"M. L. Barnett, D. Boddupalli, S. Nundy, and D. W. Bates, \u201cComparative accuracy of diagnosis by collective intelligence of multiple physicians vs individual physicians,\u201d JAMA Netw. Open, vol. 2, no. 3, p. e190096, 2019, doi: https:\/\/doi.org\/10.1001\/jamanetworkopen.2019.0096.","journal-title":"JAMA Netw. Open"},{"key":"2460_CR8","doi-asserted-by":"publisher","first-page":"8","DOI":"10.1038\/s43856-021-00003-5","volume":"1","author":"S Kundu","year":"2021","unstructured":"S. Kundu, \u201cHow will artificial intelligence change medical training?,\u201d Commun. Med., vol. 1, p. 8, 2021, doi: https:\/\/doi.org\/10.1038\/s43856-021-00003-5.","journal-title":"Commun. Med."},{"issue":"2","key":"2460_CR9","first-page":"139","volume":"42","author":"WW Rosser","year":"1996","unstructured":"W. W. Rosser, \u201cApproach to diagnosis by primary care clinicians and specialists: is there a difference?,\u201d J. Fam. Pract., vol. 42, no. 2, pp. 139\u2013144, 1996.","journal-title":"J. Fam. Pract."},{"issue":"1","key":"2460_CR10","doi-asserted-by":"publisher","first-page":"54","DOI":"10.1038\/s41746-025-01460-1","volume":"8","author":"L D\u2019Adderio","year":"2025","unstructured":"L. D\u2019Adderio and D. W. Bates, \u201cTransforming diagnosis through artificial intelligence,\u201d NPJ Digit. Med., vol. 8, no. 1, p. 54, 2025, doi: https:\/\/doi.org\/10.1038\/s41746-025-01460-1.","journal-title":"NPJ Digit. Med."},{"issue":"5","key":"2460_CR11","doi-asserted-by":"publisher","first-page":"1589","DOI":"10.1109\/JBHI.2017.2767063","volume":"22","author":"B Shickel","year":"2018","unstructured":"B. Shickel, P. J. Tighe, A. Bihorac, and P. Rashidi, \u201cDeep EHR: a survey of recent advances in deep learning techniques for electronic health record (EHR) analysis,\u201d IEEE J. Biomed. Health Inform., vol. 22, no. 5, pp. 1589\u20131604, 2018, doi: https:\/\/doi.org\/10.1109\/JBHI.2017.2767063.","journal-title":"IEEE J. Biomed. Health Inform."},{"issue":"10","key":"2460_CR12","doi-asserted-by":"publisher","first-page":"1419","DOI":"10.1093\/jamia\/ocy068","volume":"25","author":"C Xiao","year":"2018","unstructured":"C. Xiao, E. Choi, and J. Sun, \u201cOpportunities and challenges in developing deep learning models using electronic health records data: a systematic review,\u201d J. Am. Med. Inform. Assoc., vol. 25, no. 10, pp. 1419\u20131428, 2018, doi: https:\/\/doi.org\/10.1093\/jamia\/ocy068.","journal-title":"J. Am. Med. Inform. Assoc."},{"key":"2460_CR13","doi-asserted-by":"publisher","first-page":"102021","DOI":"10.1016\/j.artmed.2021.102021","volume":"112","author":"JM Lee","year":"2021","unstructured":"J. M. Lee and M. Hauskrecht, \u201cModeling multivariate clinical event time-series with recurrent temporal mechanisms,\u201d Artif. Intell. Med., vol. 112, p. 102021, 2021, doi: https:\/\/doi.org\/10.1016\/j.artmed.2021.102021.","journal-title":"Artif. Intell. Med."},{"key":"2460_CR14","doi-asserted-by":"publisher","first-page":"103980","DOI":"10.1016\/j.jbi.2021.103980","volume":"126","author":"F Xie","year":"2022","unstructured":"F. Xie et al., \u201cDeep learning for temporal data representation in electronic health records: a systematic review of challenges and methodologies,\u201d J. Biomed. Inform., vol. 126, p. 103980, 2022, doi: https:\/\/doi.org\/10.1016\/j.jbi.2021.103980.","journal-title":"J. Biomed. Inform."},{"key":"2460_CR15","unstructured":"E. Choi, M. T. Bahadori, A. Schuetz, W. F. Stewart, and J. Sun, \u201cDoctor AI: predicting clinical events via recurrent neural networks,\u201d JMLR Workshop Conf. Proc., vol. 56, pp. 301\u2013318, 2016."},{"issue":"4","key":"2460_CR16","doi-asserted-by":"publisher","first-page":"e281","DOI":"10.1016\/S2589-7500(24)00025-6","volume":"6","author":"Z Kraljevic","year":"2024","unstructured":"Z. Kraljevic et al., \u201cForesight\u2014a generative pretrained transformer for modelling of patient timelines using electronic health records: a retrospective modelling study,\u201d Lancet Digit. Health, vol. 6, no. 4, pp. e281\u2013e290, 2024, doi: https:\/\/doi.org\/10.1016\/S2589-7500(24)00025-6.","journal-title":"Lancet Digit. Health"},{"key":"2460_CR17","doi-asserted-by":"publisher","first-page":"132","DOI":"10.1186\/s40854-025-00875-9","volume":"11","author":"J Hong","year":"2025","unstructured":"J. Hong and Y. M. Hwang, \u201cLong short-term memory networks in learning memory inconsistencies of stock markets,\u201d Financ. Innov., vol. 11, p. 132, 2025, doi: https:\/\/doi.org\/10.1186\/s40854-025-00875-9.","journal-title":"Financ. Innov."},{"key":"2460_CR18","doi-asserted-by":"publisher","unstructured":"A. Saelmans et al., \u201cImplementation and updating of clinical prediction models: a systematic review,\u201d Mayo Clin. Proc. Digit. Health, vol. 3, no. 3, p. 100228, 2025, doi: https:\/\/doi.org\/10.1016\/j.mcpdig.2025.100228.","DOI":"10.1016\/j.mcpdig.2025.100228"},{"key":"2460_CR19","unstructured":"A. Vaswani et al., \u201cAttention is all you need,\u201d in Advances in Neural Information Processing Systems, vol. 30, 2017, pp. 5998\u20136008."},{"issue":"1","key":"2460_CR20","doi-asserted-by":"publisher","first-page":"7155","DOI":"10.1038\/s41598-020-62922-y","volume":"10","author":"Y Li","year":"2020","unstructured":"Y. Li et al., \u201cBEHRT: transformer for electronic health records,\u201d Sci. Rep., vol. 10, no. 1, p. 7155, 2020, doi: https:\/\/doi.org\/10.1038\/s41598-020-62922-y.","journal-title":"Sci. Rep."},{"issue":"12","key":"2460_CR21","doi-asserted-by":"publisher","first-page":"1833","DOI":"10.1093\/jamia\/ocaf160","volume":"32","author":"E Redekop","year":"2025","unstructured":"E. Redekop et al., \u201cZero-shot medical event prediction using a generative pretrained transformer on electronic health records,\u201d J. Am. Med. Inform. Assoc., vol. 32, no. 12, pp. 1833\u20131842, 2025, doi: https:\/\/doi.org\/10.1093\/jamia\/ocaf160.","journal-title":"J. Am. Med. Inform. Assoc."},{"issue":"1","key":"2460_CR22","doi-asserted-by":"publisher","first-page":"7857","DOI":"10.1038\/s41467-023-43715-z","volume":"14","author":"Z Yang","year":"2023","unstructured":"Z. Yang, A. Mitra, W. Liu, D. Berlowitz, and H. Yu, \u201cTransformEHR: transformer-based encoder-decoder generative model to enhance prediction of disease outcomes using electronic health records,\u201d Nat. Commun., vol. 14, no. 1, p. 7857, 2023, doi: https:\/\/doi.org\/10.1038\/s41467-023-43715-z.","journal-title":"Nat. Commun."},{"key":"2460_CR23","doi-asserted-by":"publisher","unstructured":"J. Wang et al., \u201cRecent advances in predictive modeling with electronic health records,\u201d in Proc. 33rd Int. Joint Conf. Artificial Intelligence (IJCAI), 2024, pp. 8272\u20138280, doi: https:\/\/doi.org\/10.24963\/ijcai.2024\/914.","DOI":"10.24963\/ijcai.2024\/914"},{"key":"2460_CR24","doi-asserted-by":"publisher","unstructured":"Z. Yang, S. Kwon, Z. Yao, and H. Yu, \u201cMulti-label few-shot ICD coding as autoregressive generation with prompt,\u201d in Proc. 37th AAAI Conf. Artificial Intelligence, vol. 37, no. 4, 2023, pp. 5366\u20135374, doi: https:\/\/doi.org\/10.1609\/aaai.v37i4.25668.","DOI":"10.1609\/aaai.v37i4.25668"},{"issue":"1","key":"2460_CR25","doi-asserted-by":"publisher","first-page":"11730","DOI":"10.1038\/s41598-021-90000-4","volume":"11","author":"V Ardulov","year":"2021","unstructured":"V. Ardulov et al., \u201cRobust diagnostic classification via Q-learning,\u201d Sci. Rep., vol. 11, no. 1, p. 11730, 2021, doi: https:\/\/doi.org\/10.1038\/s41598-021-90000-4.","journal-title":"Sci. Rep."},{"issue":"6","key":"2460_CR26","doi-asserted-by":"publisher","first-page":"631","DOI":"10.3390\/bioengineering12060631","volume":"12","author":"S Maity","year":"2025","unstructured":"S. Maity and M. J. Saikia, \u201cLarge language models in healthcare and medical applications: a review,\u201d Bioengineering, vol. 12, no. 6, p. 631, 2025, doi: https:\/\/doi.org\/10.3390\/bioengineering12060631.","journal-title":"Bioengineering"},{"key":"2460_CR27","doi-asserted-by":"publisher","first-page":"e55138","DOI":"10.2196\/55138","volume":"26","author":"O Ayo-Ajibola","year":"2024","unstructured":"O. Ayo-Ajibola, R. J. Davis, M. E. Lin, J. Riddell, and R. L. Kravitz, \u201cCharacterizing the adoption and experiences of users of artificial intelligence-generated health information in the United States: cross-sectional questionnaire study,\u201d J. Med. Internet Res., vol. 26, p. e55138, 2024, doi: https:\/\/doi.org\/10.2196\/55138.","journal-title":"J. Med. Internet Res."},{"issue":"1","key":"2460_CR28","doi-asserted-by":"publisher","first-page":"305","DOI":"10.1146\/annurev-biodatasci-103123-094851","volume":"8","author":"Y Yang","year":"2025","unstructured":"Y. Yang et al., \u201cBeyond multiple-choice accuracy: real-world challenges of implementing large language models in healthcare,\u201d Annu. Rev. Biomed. Data Sci., vol. 8, no. 1, pp. 305\u2013316, 2025, doi: https:\/\/doi.org\/10.1146\/annurev-biodatasci-103123-094851.","journal-title":"Annu. Rev. Biomed. Data Sci."},{"issue":"1","key":"2460_CR29","doi-asserted-by":"publisher","first-page":"135","DOI":"10.1038\/s41746-023-00879-8","volume":"6","author":"M Wornow","year":"2023","unstructured":"M. Wornow et al., \u201cThe shaky foundations of large language models and foundation models for electronic health records,\u201d NPJ Digit. Med., vol. 6, no. 1, p. 135, 2023, doi: https:\/\/doi.org\/10.1038\/s41746-023-00879-8.","journal-title":"NPJ Digit. Med."},{"key":"2460_CR30","unstructured":"Y. Zhu et al., \u201cPrompting large language models for zero-shot clinical prediction with structured longitudinal electronic health record data,\u201d arXiv:2402.01713, 2024. [Online]. Available: https:\/\/arxiv.org\/abs\/2402.01713"},{"key":"2460_CR31","doi-asserted-by":"publisher","first-page":"107200","DOI":"10.1016\/j.neunet.2025.107200","volume":"185","author":"Y Wang","year":"2025","unstructured":"Y. Wang, Z. Wu, J. Yao, and J. Su, \u201cTDAG: a multi-agent framework based on dynamic task decomposition and agent generation,\u201d Neural Netw., vol. 185, p. 107200, 2025, doi: https:\/\/doi.org\/10.1016\/j.neunet.2025.107200.","journal-title":"Neural Netw."},{"issue":"1","key":"2460_CR32","doi-asserted-by":"publisher","first-page":"541","DOI":"10.1038\/s41746-025-01940-4","volume":"8","author":"R Li","year":"2025","unstructured":"R. Li, X. Wang, D. Berlowitz, J. Mez, H. Lin, and H. Yu, \u201cCARE-AD: a multi-agent large language model framework for Alzheimer\u2019s disease prediction using longitudinal clinical notes,\u201d NPJ Digit. Med., vol. 8, no. 1, p. 541, 2025, doi: https:\/\/doi.org\/10.1038\/s41746-025-01940-4.","journal-title":"NPJ Digit. Med."},{"issue":"1","key":"2460_CR33","doi-asserted-by":"publisher","first-page":"159","DOI":"10.1038\/s41746-025-01550-0","volume":"8","author":"X Chen","year":"2025","unstructured":"X. Chen et al., \u201cEnhancing diagnostic capability with multi-agents conversational large language models,\u201d NPJ Digit. Med., vol. 8, no. 1, p. 159, 2025, doi: https:\/\/doi.org\/10.1038\/s41746-025-01550-0.","journal-title":"NPJ Digit. Med."},{"issue":"3","key":"2460_CR34","doi-asserted-by":"publisher","first-page":"191","DOI":"10.1007\/s10489-007-0085-8","volume":"30","author":"H Gonz\u00e1lez-V\u00e9lez","year":"2009","unstructured":"H. Gonz\u00e1lez-V\u00e9lez et al., \u201cHealthAgents: distributed multi-agent brain tumor diagnosis and prognosis,\u201d Appl. Intell., vol. 30, no. 3, pp. 191\u2013202, 2009, doi: https:\/\/doi.org\/10.1007\/s10489-007-0085-8.","journal-title":"Appl. Intell."},{"issue":"1","key":"2460_CR35","doi-asserted-by":"publisher","first-page":"51","DOI":"10.32604\/cmc.2021.012632","volume":"67","author":"M Elhoseny","year":"2021","unstructured":"M. Elhoseny et al., \u201cA new multi-agent feature wrapper machine learning approach for heart disease diagnosis,\u201d Comput. Mater. Contin., vol. 67, no. 1, pp. 51\u201371, 2021, doi: https:\/\/doi.org\/10.32604\/cmc.2021.012632.","journal-title":"Comput. Mater. Contin."},{"key":"2460_CR36","unstructured":"H. Wang, S. Zhao, Z. Qiang, N. Xi, B. Qin, and T. Liu, \u201cBeyond direct diagnosis: LLM-based multi-specialist agent consultation for automatic diagnosis,\u201d arXiv:2401.16107, 2024. [Online]. Available: https:\/\/arxiv.org\/abs\/2401.16107"}],"container-title":["Journal of Medical Systems"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/link.springer.com\/content\/pdf\/10.1007\/s10916-026-02460-8.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/link.springer.com\/article\/10.1007\/s10916-026-02460-8","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/link.springer.com\/content\/pdf\/10.1007\/s10916-026-02460-8.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T04:37:33Z","timestamp":1789447053000},"score":1,"resource":{"primary":{"URL":"https:\/\/link.springer.com\/10.1007\/s10916-026-02460-8"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,15]]},"references-count":36,"journal-issue":{"issue":"1","published-online":{"date-parts":[[2026,12]]}},"alternative-id":["2460"],"URL":"https:\/\/doi.org\/10.1007\/s10916-026-02460-8","relation":{},"ISSN":["1573-689X"],"issn-type":[{"value":"1573-689X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,15]]},"assertion":[{"value":"12 February 2026","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"8 September 2026","order":2,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"15 September 2026","order":3,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"This study was approved by the Institutional Review Board under protocol number PRO00041276 (\n                      Building Clinical Decision Aids Using Patient Similarity and Other Machine Learning Algorithms\n                      ). The research involved retrospective analysis of electronic health record data and was conducted in accordance with institutional policies and applicable federal regulations. The IRB granted a waiver of informed consent as the study posed minimal risk and used de-identified data. All data were securely stored and handled in compliance with HIPAA and institutional data protection standards.","order":1,"name":"Ethics","label":"Ethics Approval","group":{"name":"EthicsHeading","label":"Declarations"}},{"value":"All authors provided consent for publication.","order":2,"name":"Ethics","label":"Consent for Publication","group":{"name":"EthicsHeading","label":"Declarations"}},{"value":"Not Applicable.","order":3,"name":"Ethics","label":"Consent to Participate","group":{"name":"EthicsHeading","label":"Declarations"}},{"value":"Not Applicable.","order":4,"name":"Ethics","label":"Clinical Trial Number","group":{"name":"EthicsHeading","label":"Declarations"}},{"value":"The authors declare no competing interests.","order":5,"name":"Ethics","label":"Competing interests","group":{"name":"EthicsHeading","label":"Declarations"}}],"article-number":"131"}}