{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,13]],"date-time":"2026-10-13T21:37:01Z","timestamp":1791927421586,"version":"4.3.4"},"reference-count":19,"publisher":"Oxford University Press (OUP)","issue":"1","license":[{"start":{"date-parts":[[2025,1,10]],"date-time":"2025-01-10T00:00:00Z","timestamp":1736467200000},"content-version":"vor","delay-in-days":15,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/100000092","name":"National Library of Medicine","doi-asserted-by":"publisher","award":["R00LM014308-02"],"award-info":[{"award-number":["R00LM014308-02"]}],"id":[{"id":"10.13039\/100000092","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100000092","name":"National Library of Medicine","doi-asserted-by":"publisher","award":["R01LM012973-05"],"award-info":[{"award-number":["R01LM012973-05"]}],"id":[{"id":"10.13039\/100000092","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100000050","name":"National Heart, Lung, and Blood Institute","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100000050","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2024,12,26]]},"abstract":"<jats:title>Abstract<\/jats:title>\n               <jats:sec>\n                  <jats:title>Objective<\/jats:title>\n                  <jats:p>To evaluate large language models (LLMs) for pre-test diagnostic probability estimation and compare their uncertainty estimation performance with a traditional machine learning classifier.<\/jats:p>\n               <\/jats:sec>\n               <jats:sec>\n                  <jats:title>Materials and Methods<\/jats:title>\n                  <jats:p>We assessed 2 instruction-tuned LLMs, Mistral-7B-Instruct and Llama3-70B-chat-hf, on predicting binary outcomes for Sepsis, Arrhythmia, and Congestive Heart Failure (CHF) using electronic health record (EHR) data from 660 patients. Three uncertainty estimation methods\u2014Verbalized Confidence, Token Logits, and LLM Embedding+XGB\u2014were compared against an\u00a0eXtreme Gradient Boosting (XGB) classifier trained on raw EHR data. Performance metrics included AUROC and Pearson correlation between predicted probabilities.<\/jats:p>\n               <\/jats:sec>\n               <jats:sec>\n                  <jats:title>Results<\/jats:title>\n                  <jats:p>The XGB classifier outperformed the LLM-based methods across all tasks. LLM Embedding+XGB showed the closest performance to the XGB baseline, while Verbalized Confidence and Token Logits underperformed.<\/jats:p>\n               <\/jats:sec>\n               <jats:sec>\n                  <jats:title>Discussion<\/jats:title>\n                  <jats:p>These findings, consistent across multiple models and demographic groups, highlight the limitations of current LLMs in providing reliable pre-test probability estimations and underscore the need for improved calibration and bias mitigation strategies. Future work should explore hybrid approaches that integrate LLMs with numerical reasoning modules and calibrated embeddings to enhance diagnostic accuracy and ensure fairer predictions across diverse populations.<\/jats:p>\n               <\/jats:sec>\n               <jats:sec>\n                  <jats:title>Conclusions<\/jats:title>\n                  <jats:p>LLMs demonstrate potential but currently fall short in estimating diagnostic probabilities compared to traditional machine learning classifiers trained on structured EHR data. Further improvements are needed for reliable clinical use.<\/jats:p>\n               <\/jats:sec>","DOI":"10.1093\/jamiaopen\/ooae154","type":"journal-article","created":{"date-parts":[[2025,1,10]],"date-time":"2025-01-10T21:54:40Z","timestamp":1736546080000},"source":"Crossref","is-referenced-by-count":17,"title":["Uncertainty estimation in diagnosis generation from large language models: next-word probability is not pre-test probability"],"prefix":"10.1093","volume":"8","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-9341-7360","authenticated-orcid":false,"given":"Yanjun","family":"Gao","sequence":"first","affiliation":[{"name":"Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO 80045,","place":["United States"]},{"name":"Department of Medicine, University of Wisconsin-Madison , Madison, WI 53792,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9524-8436","authenticated-orcid":false,"given":"Skatje","family":"Myers","sequence":"additional","affiliation":[{"name":"Department of Medicine, University of Wisconsin-Madison , Madison, WI 53792,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-7999-7410","authenticated-orcid":false,"given":"Shan","family":"Chen","sequence":"additional","affiliation":[{"name":"Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School , Boston, MA 02114,","place":["United States"]},{"name":"Department of Radiation Oncology, Brigham and Women\u2019s Hospital\/Dana-Farber Cancer Institute , Boston, MA 02114,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-2585-2707","authenticated-orcid":false,"given":"Dmitriy","family":"Dligach","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Loyola University Chicago , Chicago, IL 60660,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4513-403X","authenticated-orcid":false,"given":"Timothy","family":"Miller","sequence":"additional","affiliation":[{"name":"Computational Health Informatics Program, Boston Children\u2019s Hospital , Boston, MA 02115,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0345-2232","authenticated-orcid":false,"given":"Danielle S","family":"Bitterman","sequence":"additional","affiliation":[{"name":"Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School , Boston, MA 02114,","place":["United States"]},{"name":"Department of Radiation Oncology, Brigham and Women\u2019s Hospital\/Dana-Farber Cancer Institute , Boston, MA 02114,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-9314-2037","authenticated-orcid":false,"given":"Guanhua","family":"Chen","sequence":"additional","affiliation":[{"name":"Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison , Madison, WI 53792,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3010-6960","authenticated-orcid":false,"given":"Anoop","family":"Mayampurath","sequence":"additional","affiliation":[{"name":"Department of Medicine, University of Wisconsin-Madison , Madison, WI 53792,","place":["United States"]},{"name":"Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison , Madison, WI 53792,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Matthew M","family":"Churpek","sequence":"additional","affiliation":[{"name":"Department of Medicine, University of Wisconsin-Madison , Madison, WI 53792,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6368-4652","authenticated-orcid":false,"given":"Majid","family":"Afshar","sequence":"additional","affiliation":[{"name":"Department of Medicine, University of Wisconsin-Madison , Madison, WI 53792,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"286","published-online":{"date-parts":[[2025,1,10]]},"reference":[{"key":"2025030418413283700_ooae154-B1","author":"Sox","year":"1989"},{"key":"2025030418413283700_ooae154-B2","doi-asserted-by":"crossref","first-page":"2217","DOI":"10.1056\/NEJMra054782","article-title":"Educational strategies to promote clinical diagnostic reasoning","volume":"355","author":"Bowen","year":"2006","journal-title":"N Engl J Med"},{"key":"2025030418413283700_ooae154-B3","doi-asserted-by":"publisher","first-page":"138","DOI":"10.1186\/s12911-016-0377-1","article-title":"Cognitive biases associated with medical decisions: a systematic review","volume":"16","author":"Saposnik","year":"2016","journal-title":"BMC Med Inform Decis Mak"},{"key":"2025030418413283700_ooae154-B4","doi-asserted-by":"crossref","first-page":"e2347075","DOI":"10.1001\/jamanetworkopen.2023.47075","article-title":"Artificial intelligence vs clinician performance in estimating probabilities of diagnoses before and after testing","volume":"6","author":"Rodman","year":"2023","journal-title":"JAMA Netw Open"},{"key":"2025030418413283700_ooae154-B5","doi-asserted-by":"crossref","first-page":"78","DOI":"10.1001\/jama.2023.8288","article-title":"Accuracy of a generative artificial intelligence model in a complex diagnostic challenge","volume":"330","author":"Kanjee","year":"2023","journal-title":"JAMA"},{"key":"2025030418413283700_ooae154-B6","doi-asserted-by":"crossref","first-page":"20","DOI":"10.1038\/s41746-024-01010-1","article-title":"Diagnostic reasoning prompts reveal the potential for large language model interpretability in medicine","volume":"7","author":"Savage","year":"2024","journal-title":"NPJ Digit Med"},{"key":"2025030418413283700_ooae154-B7","author":"Kapoor","year":"2024"},{"key":"2025030418413283700_ooae154-B8","first-page":"6577","author":"Geng"},{"key":"2025030418413283700_ooae154-B9","doi-asserted-by":"publisher","first-page":"ocae254","DOI":"10.1093\/jamia\/ocae254","article-title":"Large language model uncertainty proxies: discrimination and calibration for medical diagnosis and treatment","volume":"32","author":"Savage","year":"2025","journal-title":"J Am Med Inf Assoc"},{"key":"2025030418413283700_ooae154-B10","first-page":"1","article-title":"Evaluation and mitigation of the limitations of large language models in clinical decision-making","volume":"4","author":"Hager","year":"2024","journal-title":"Nat Med"},{"key":"2025030418413283700_ooae154-B11","first-page":"74952","author":"Turpin"},{"key":"2025030418413283700_ooae154-B12","author":"Jiang","year":"2023"},{"key":"2025030418413283700_ooae154-B13","author":"Meta AI"},{"key":"2025030418413283700_ooae154-B14","doi-asserted-by":"crossref","first-page":"e51391","DOI":"10.2196\/51391","article-title":"Learning to make rare and complex diagnoses with generative AI assistance: qualitative study of popular large language models","volume":"10","author":"Abdullahi","year":"2024","journal-title":"JMIR Med Educ"},{"key":"2025030418413283700_ooae154-B15","doi-asserted-by":"publisher","author":"Churpek","year":"2024","DOI":"10.1101\/2024.02.05.24301960"},{"key":"2025030418413283700_ooae154-B16","doi-asserted-by":"publisher","first-page":"e47736","DOI":"10.2196\/47736","article-title":"Comparing explainable machine learning approaches with traditional statistical methods for evaluating stroke risk models: retrospective cohort study","volume":"7","author":"Lolak","year":"2023","journal-title":"JMIR Cardio"},{"key":"2025030418413283700_ooae154-B17","doi-asserted-by":"crossref","first-page":"e1116","DOI":"10.1097\/CCE.0000000000001116","article-title":"Development and validation of a machine learning COVID-19 veteran (COVet) deterioration risk score","volume":"6","author":"Govindan","year":"2024","journal-title":"Crit Care Explor"},{"key":"2025030418413283700_ooae154-B18","first-page":"5414","author":"Gao","year":"2024"},{"key":"2025030418413283700_ooae154-B19","doi-asserted-by":"crossref","first-page":"e12","DOI":"10.1016\/S2589-7500(23)00225-X","article-title":"Assessing the potential of GPT-4 to perpetuate racial and gender biases in health care: a model evaluation study","volume":"6","author":"Zack","year":"2024","journal-title":"Lancet Digit Health"}],"container-title":["JAMIA Open"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/academic.oup.com\/jamiaopen\/article-pdf\/8\/1\/ooae154\/61411131\/ooae154.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/academic.oup.com\/jamiaopen\/article-pdf\/8\/1\/ooae154\/61411131\/ooae154.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,3,4]],"date-time":"2025-03-04T20:48:03Z","timestamp":1741121283000},"score":1,"resource":{"primary":{"URL":"https:\/\/academic.oup.com\/jamiaopen\/article\/doi\/10.1093\/jamiaopen\/ooae154\/7951510"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024,12,26]]},"references-count":19,"journal-issue":{"issue":"1","published-print":{"date-parts":[[2024,12,26]]}},"URL":"https:\/\/doi.org\/10.1093\/jamiaopen\/ooae154","relation":{},"ISSN":["2574-2531"],"issn-type":[{"value":"2574-2531","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2025,2]]},"published":{"date-parts":[[2024,12,26]]},"article-number":"ooae154"}}