{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,1]],"date-time":"2026-10-01T16:58:36Z","timestamp":1790873916120,"version":"4.1.0"},"reference-count":35,"publisher":"Oxford University Press (OUP)","license":[{"start":{"date-parts":[[2026,10,1]],"date-time":"2026-10-01T00:00:00Z","timestamp":1790812800000},"content-version":"am","delay-in-days":0,"URL":"https:\/\/academic.oup.com\/pages\/standard-publication-reuse-rights"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:sec>\n                    <jats:title>Objectives<\/jats:title>\n                    <jats:p>To review the clinical evidence for small language models (SLMs), 4 billion parameters or fewer, for performance, safety, and deployment feasibility.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Materials and methods<\/jats:title>\n                    <jats:p>We searched 5 databases through February 10, 2026 for English-language reports of an SLM on a clinical task, following PRISMA 2020 under a PROSPERO-registered protocol. Two reviewers independently assessed eligibility (Cohen \u03ba\u2009=\u20090.93) and rated methodological quality and transparency. We calculated a relative task score (RTS): the primary metric of each study\u2019s best-performing qualifying SLM, divided by a study-specific reference comparator selected by a uniform hierarchy, calculated by the review authors.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>Eleven studies (7 peer-reviewed, 4 preprint or technical-report) were eligible. Across the 9 studies with a ratio-scale reference comparator, RTS ranged from 0.30 to 2.36; 1 was a raw difference on a non-ratio scale and 1 had no comparator. We did not pool estimates, given heterogeneity. Hallucination was evaluated in 6 of 11 studies, a scalar calibration-error metric or epistemic uncertainty in 0 of 11 (1 reported a calibration curve only), and on-hardware inference timing in 2 of 11. Our memory model estimated that a 4-billion-parameter model requires about 9.3 GB at a 2048-token context, indicating single-GPU memory feasibility rather than demonstrated deployment.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Discussion<\/jats:title>\n                    <jats:p>Domain-adapted SLMs are memory-feasible, but the evidence base is too small and heterogeneous for inferential comparison with larger models, and safety properties (including adversarial robustness) go unmeasured.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusion<\/jats:title>\n                    <jats:p>Routine reporting of calibration and uncertainty and measurement of end-to-end clinical performance are prerequisites for responsible SLM deployment.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>PROSPERO Registration<\/jats:title>\n                    <jats:p>CRD420261331444.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.1093\/jamia\/ocag164","type":"journal-article","created":{"date-parts":[[2026,9,8]],"date-time":"2026-09-08T11:32:31Z","timestamp":1788867151000},"source":"Crossref","is-referenced-by-count":0,"title":["Small language models in clinical medicine: a systematic review of performance, safety, and deployment feasibility"],"prefix":"10.1093","author":[{"ORCID":"https:\/\/orcid.org\/0009-0000-7542-8608","authenticated-orcid":false,"given":"Alon","family":"Gorenshtein","sequence":"first","affiliation":[{"name":"BRIDGE GenAI Lab, Beth Israel Deaconess Medical Center, Harvard Medical School , Boston, MA 02215,","place":["United States"]},{"name":"Department of Neurology, Harvard Medical School , Boston, MA 02215,","place":["United States"]},{"name":"Department of Neurology, Beth Israel Deaconess Medical Center , Boston, MA 02215,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0001-0438-0827","authenticated-orcid":false,"given":"Mahmud","family":"Omar","sequence":"additional","affiliation":[{"name":"BRIDGE GenAI Lab, Beth Israel Deaconess Medical Center, Harvard Medical School , Boston, MA 02215,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yiftach","family":"Barash","sequence":"additional","affiliation":[{"name":"BRIDGE GenAI Lab, Beth Israel Deaconess Medical Center, Harvard Medical School , Boston, MA 02215,","place":["United States"]},{"name":"Department of Radiology, Beth Israel Deaconess Medical Center, Harvard Medical School , Boston, MA 02215,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jonathan B","family":"Kruskal","sequence":"additional","affiliation":[{"name":"Department of Radiology, Beth Israel Deaconess Medical Center, Harvard Medical School , Boston, MA 02215,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Muneeb","family":"Ahmed","sequence":"additional","affiliation":[{"name":"Department of Radiology, Beth Israel Deaconess Medical Center, Harvard Medical School , Boston, MA 02215,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Olga R","family":"Brook","sequence":"additional","affiliation":[{"name":"Department of Radiology, Beth Israel Deaconess Medical Center, Harvard Medical School , Boston, MA 02215,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ben","family":"Illigens","sequence":"additional","affiliation":[{"name":"Hasso Plattner Institute for Digital Health, Icahn School of Medicine at Mount Sinai , New York, NY 10029,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Girish N","family":"Nadkarni","sequence":"additional","affiliation":[{"name":"Hasso Plattner Institute for Digital Health, Icahn School of Medicine at Mount Sinai , New York, NY 10029,","place":["United States"]},{"name":"The Windreich Department of Artificial Intelligence and Human Health, Icahn School of Medicine at Mount Sinai , New York, NY 10029,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Eyal","family":"Klang","sequence":"additional","affiliation":[{"name":"BRIDGE GenAI Lab, Beth Israel Deaconess Medical Center, Harvard Medical School , Boston, MA 02215,","place":["United States"]},{"name":"Department of Radiology, Beth Israel Deaconess Medical Center, Harvard Medical School , Boston, MA 02215,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"286","published-online":{"date-parts":[[2026,10,1]]},"reference":[{"key":"2026100112051279800_ocag164-B1","doi-asserted-by":"publisher","first-page":"e0000662","DOI":"10.1371\/journal.pdig.0000662","article-title":"Large language models in medicine: a review of current clinical trials across healthcare applications","volume":"3","author":"Omar","year":"2024","journal-title":"PLOS Digit Health"},{"key":"2026100112051279800_ocag164-B2","doi-asserted-by":"publisher","author":"Gorenshtein","year":"2025","DOI":"10.1101\/2025.08.22.25334232"},{"key":"2026100112051279800_ocag164-B3","doi-asserted-by":"publisher","first-page":"631","DOI":"10.3390\/bioengineering12060631","article-title":"Large language models in healthcare and medical applications: a review","volume":"12","author":"Maity","year":"2025","journal-title":"Bioengineering (Basel)"},{"key":"2026100112051279800_ocag164-B4","doi-asserted-by":"publisher","first-page":"109713","DOI":"10.1016\/j.isci.2024.109713","article-title":"The application of large language models in medicine: a scoping review","volume":"27","author":"Meng","year":"2024","journal-title":"iScience"},{"key":"2026100112051279800_ocag164-B5","doi-asserted-by":"publisher","author":"Tahera","year":"2026","DOI":"10.48550\/arXiv.2601.10004"},{"key":"2026100112051279800_ocag164-B6","doi-asserted-by":"publisher","author":"Auger","year":"2026","DOI":"10.64898\/2026.02.24.26346972"},{"key":"2026100112051279800_ocag164-B7","doi-asserted-by":"publisher","first-page":"1821","DOI":"10.1093\/jamia\/ocae122","article-title":"BioInstruct: instruction tuning of large language models for biomedical natural language processing","volume":"31","author":"Tran","year":"2024","journal-title":"J Am Med Inform Assoc"},{"key":"2026100112051279800_ocag164-B8","doi-asserted-by":"publisher","author":"Subramanian","year":"2025","DOI":"10.48550\/arXiv.2501.05465"},{"key":"2026100112051279800_ocag164-B9","doi-asserted-by":"publisher","author":"Malaquias Junior","year":"2025","DOI":"10.48550\/arXiv.2501.02068"},{"key":"2026100112051279800_ocag164-B10","doi-asserted-by":"publisher","author":"Garg","year":"2026","DOI":"10.1016\/j.cosrev.2026.100999"},{"key":"2026100112051279800_ocag164-B11","doi-asserted-by":"publisher","author":"Dettmers","DOI":"10.52202\/068431-2198"},{"key":"2026100112051279800_ocag164-B12","author":"Frantar"},{"key":"2026100112051279800_ocag164-B13","first-page":"87","author":"Lin","year":"2024"},{"key":"2026100112051279800_ocag164-B14","doi-asserted-by":"publisher","first-page":"159","DOI":"10.2307\/2529310","article-title":"The measurement of observer agreement for categorical data","volume":"33","author":"Landis","year":"1977","journal-title":"Biometrics"},{"key":"2026100112051279800_ocag164-B15","doi-asserted-by":"publisher","first-page":"e082505","DOI":"10.1136\/bmj-2024-082505","article-title":"PROBAST+AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods","volume":"388","author":"Moons","year":"2025","journal-title":"BMJ"},{"key":"2026100112051279800_ocag164-B16","doi-asserted-by":"publisher","first-page":"611","DOI":"10.1145\/3600006.3613165","author":"Kwon","year":"2023"},{"key":"2026100112051279800_ocag164-B17","doi-asserted-by":"publisher","first-page":"340","DOI":"10.1093\/jamia\/ocac225","article-title":"A comparative study of pretrained language models for long clinical text","volume":"30","author":"Li","year":"2023","journal-title":"J Am Med Inform Assoc"},{"key":"2026100112051279800_ocag164-B18","doi-asserted-by":"publisher","first-page":"357","DOI":"10.1038\/s41586-023-06160-y","article-title":"Health system-scale language models are all-purpose prediction engines","volume":"619","author":"Jiang","year":"2023","journal-title":"Nature"},{"key":"2026100112051279800_ocag164-B19","doi-asserted-by":"publisher","first-page":"e000068","DOI":"10.1136\/bmjdhai-2025-000068","article-title":"MedMobile: a mobile-sized language model with clinical capabilities","volume":"1","author":"Vishwanath","year":"2025","journal-title":"BMJ Digital Health AI"},{"key":"2026100112051279800_ocag164-B20","doi-asserted-by":"publisher","author":"Pradeepkumar","year":"2026","DOI":"10.48550\/arXiv.2601.22197"},{"key":"2026100112051279800_ocag164-B21","doi-asserted-by":"publisher","first-page":"100197","DOI":"10.1016\/j.ibmed.2024.100197","article-title":"Open-source small language models for personal medical assistant chatbots","volume":"11","author":"Magnini","year":"2025","journal-title":"Intell-Based Med"},{"key":"2026100112051279800_ocag164-B22","doi-asserted-by":"publisher","first-page":"527","DOI":"10.1007\/s44230-024-00085-z","article-title":"PharmaLLM: a medicine prescriber chatbot exploiting open-source large language models","volume":"4","author":"Azam","year":"2024","journal-title":"Hum-Cent Intell Syst"},{"key":"2026100112051279800_ocag164-B23","doi-asserted-by":"publisher","author":"Thomas","year":"2025","DOI":"10.1101\/2025.07.01.25330679"},{"key":"2026100112051279800_ocag164-B24","doi-asserted-by":"publisher","first-page":"159","DOI":"10.1007\/s10916-025-02297-7","article-title":"RAG-enhanced open SLMs for hypertension management chatbots","volume":"49","author":"Aguzzi","year":"2025","journal-title":"J Med Syst"},{"key":"2026100112051279800_ocag164-B25","doi-asserted-by":"publisher","author":"Ghaffarzadeh-Esfahani","year":"2026","DOI":"10.48550\/arXiv.2602.21374"},{"key":"2026100112051279800_ocag164-B26","doi-asserted-by":"publisher","first-page":"2440","DOI":"10.63317\/2c6cyoc8xohu","author":"Ferrazzi","year":"2026"},{"key":"2026100112051279800_ocag164-B27","doi-asserted-by":"publisher","author":"Watanabe","year":"2024","DOI":"10.48550\/arXiv.2412.16423,"},{"key":"2026100112051279800_ocag164-B28","doi-asserted-by":"publisher","author":"Abdin","year":"2024","DOI":"10.48550\/arXiv.2404.14219,"},{"key":"2026100112051279800_ocag164-B29","doi-asserted-by":"publisher","author":"Kamath","year":"2025","DOI":"10.48550\/arXiv.2503.19786,"},{"key":"2026100112051279800_ocag164-B30","doi-asserted-by":"publisher","author":"Zhang","year":"2024","DOI":"10.48550\/arXiv.2401.02385,"},{"key":"2026100112051279800_ocag164-B31","doi-asserted-by":"publisher","first-page":"81","DOI":"10.3390\/ijerph23010081","article-title":"LLM-assisted scoping review of artificial intelligence in Brazilian public health: lessons from transfer and federated learning for resource-constrained settings","volume":"23","author":"Borges","year":"2026","journal-title":"Int J Environ Res Public Health"},{"key":"2026100112051279800_ocag164-B32","doi-asserted-by":"publisher","first-page":"101707","DOI":"10.1016\/j.lanwpc.2025.101707","article-title":"Large language models and global health equity: a roadmap for equitable adoption in LMICs","volume":"63","author":"Chen","year":"2025","journal-title":"Lancet Reg Health West Pac"},{"key":"2026100112051279800_ocag164-B33","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 (Lond)"},{"key":"2026100112051279800_ocag164-B34","doi-asserted-by":"publisher","author":"Belcak","year":"2025","DOI":"10.48550\/arXiv.2506.02153,"},{"key":"2026100112051279800_ocag164-B35","doi-asserted-by":"publisher","DOI":"10.21203\/rs.3.rs-8872967\/v1","article-title":"When agentic LLMs trust poisoned tools: vulnerability of clinical LLMs to adversarial guidelines","author":"Omar","year":"2026","journal-title":"Res Square"}],"container-title":["Journal of the American Medical Informatics Association"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/academic.oup.com\/jamia\/advance-article-pdf\/doi\/10.1093\/jamia\/ocag164\/71560997\/ocag164.pdf","content-type":"application\/pdf","content-version":"am","intended-application":"syndication"},{"URL":"https:\/\/academic.oup.com\/jamia\/advance-article-pdf\/doi\/10.1093\/jamia\/ocag164\/71560997\/ocag164.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,1]],"date-time":"2026-10-01T16:05:50Z","timestamp":1790870750000},"score":1,"resource":{"primary":{"URL":"https:\/\/academic.oup.com\/jamia\/advance-article\/doi\/10.1093\/jamia\/ocag164\/8859226"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,10,1]]},"references-count":35,"URL":"https:\/\/doi.org\/10.1093\/jamia\/ocag164","relation":{},"ISSN":["1067-5027","1527-974X"],"issn-type":[{"value":"1067-5027","type":"print"},{"value":"1527-974X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,10,1]]},"article-number":"ocag164"}}