{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,27]],"date-time":"2026-08-27T15:30:45Z","timestamp":1787844645052,"version":"build-2784847793"},"reference-count":28,"publisher":"SAGE Publications","license":[{"start":{"date-parts":[[2026,2,1]],"date-time":"2026-02-01T00:00:00Z","timestamp":1769904000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by-nc\/4.0\/"},{"start":{"date-parts":[[2026,2,1]],"date-time":"2026-02-01T00:00:00Z","timestamp":1769904000000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/journals.sagepub.com\/page\/policies\/text-and-data-mining-license"}],"content-domain":{"domain":["journals.sagepub.com"],"crossmark-restriction":true},"short-container-title":["DIGITAL HEALTH"],"published-print":{"date-parts":[[2026,2]]},"abstract":"<jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Large language models (LLMs) are entering clinical workflows, yet their alignment with medical-ethical principles is unclear.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Objective<\/jats:title>\n                    <jats:p>To evaluate how LLMs reconcile deontological and utilitarian reasoning in medically adapted \u201cTrolley dilemma\u201d scenarios.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>\n                      We tested 14 LLMs, including GPT-o1-preview and DeepSeek-R1-Distill-Llama, using five medically adapted Trolley dilemmas sourced from philosophical and real-world scenarios. Each model generated 10 independent answers per case (overall 700 responses). Responses were forced to \u201cyes\u201d (utilitarian) or \u201cno\u201d (deontological). The primary outcome was the proportion of utilitarian choices. \u03c7\n                      <jats:sup>2<\/jats:sup>\n                      tests compared models.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      LLM responses varied. Some models, such as GPT-4 and Gemma, selected the utilitarian option in 20% of cases, while others, like Qwen-2-7B, reached 80% (\n                      <jats:italic toggle=\"yes\">p<\/jats:italic>\n                      \u2009&lt;\u2009.001). GPT-o1-preview and DeepSeek-R1-Distill-Llama selected utilitarian responses in 44% and 38% of cases, respectively. Some models selected the consequence-maximizing option in vignettes where doing so required overriding consent or endorsing intentional harm within the scenario framing. Seven of the fourteen models tested recommended impermissible medical actions, including nonconsensual limb amputation, killing a healthy person for organ harvesting, and transferring a patient to hospice against their wishes, in 44 of 700 outputs (6.3\u202f%).\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusions<\/jats:title>\n                    <jats:p>In this constrained forced-choice stress test, models occasionally endorsed boundary-violating actions within the vignette framing. These results do not generalize to routine clinical decision-making, but motivate further benchmarked evaluation of refusal behavior and boundary constraints in ethically sensitive medical contexts. We must carefully define how these models influence clinical decisions, establish robust ethical guidelines for their use, and ensure that fundamental ethical norms are not compromised.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.1177\/20552076261428395","type":"journal-article","created":{"date-parts":[[2026,2,26]],"date-time":"2026-02-26T12:42:21Z","timestamp":1772109741000},"update-policy":"https:\/\/doi.org\/10.1177\/sage-journals-update-policy","source":"Crossref","is-referenced-by-count":1,"title":["Alignment of large language models in solving medical ethical dilemmas"],"prefix":"10.1177","volume":"12","author":[{"ORCID":"https:\/\/orcid.org\/0000-0003-0509-4686","authenticated-orcid":false,"given":"Vera","family":"Sorin","sequence":"first","affiliation":[{"name":"Mayo Clinic"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Benjamin S.","family":"Glicksberg","sequence":"additional","affiliation":[{"name":"Icahn School of Medicine at Mount Sinai"},{"name":"Icahn School of Medicine at Mount Sinai"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-2516-1751","authenticated-orcid":false,"given":"Panagiotis","family":"Korfiatis","sequence":"additional","affiliation":[{"name":"Mayo Clinic"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jeremy D.","family":"Collins","sequence":"additional","affiliation":[{"name":"Mayo Clinic"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mei-Ean","family":"Yeow","sequence":"additional","affiliation":[{"name":"Mayo Clinic"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Megan","family":"Brandeland","sequence":"additional","affiliation":[{"name":"Mayo Clinic"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Girish N.","family":"Nadkarni","sequence":"additional","affiliation":[{"name":"Icahn School of Medicine at Mount Sinai"},{"name":"Icahn School of Medicine at Mount Sinai"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Eyal","family":"Klang","sequence":"additional","affiliation":[{"name":"Icahn School of Medicine at Mount Sinai"},{"name":"Icahn School of Medicine at Mount Sinai"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"179","published-online":{"date-parts":[[2026,2,26]]},"reference":[{"key":"e_1_3_2_2_2","first-page":"3","article-title":"The \u2018four principles\u2019 approach to health care ethics","volume":"29","author":"Beauchamp TL","year":"2007","unstructured":"Beauchamp TL. The \u2018four principles\u2019 approach to health care ethics. Principles Health Care Ethics 2007; 29: 3\u201310.","journal-title":"Principles Health Care Ethics"},{"key":"e_1_3_2_3_2","doi-asserted-by":"publisher","DOI":"10.1136\/bmj.309.6948.184"},{"key":"e_1_3_2_4_2","doi-asserted-by":"publisher","DOI":"10.1136\/jme.29.5.269"},{"key":"e_1_3_2_5_2","doi-asserted-by":"publisher","DOI":"10.1037\/a0031021"},{"key":"e_1_3_2_6_2","doi-asserted-by":"publisher","DOI":"10.1007\/s13164-012-0112-5"},{"key":"e_1_3_2_7_2","doi-asserted-by":"publisher","DOI":"10.4103\/2229-5070.175024"},{"key":"e_1_3_2_8_2","doi-asserted-by":"publisher","DOI":"10.1056\/AIp2400559"},{"key":"e_1_3_2_9_2","doi-asserted-by":"publisher","DOI":"10.1038\/s43856-023-00370-1"},{"key":"e_1_3_2_10_2","doi-asserted-by":"publisher","DOI":"10.1001\/jamanetworkopen.2024.40969"},{"key":"e_1_3_2_11_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41591-024-03097-1"},{"key":"e_1_3_2_12_2","first-page":"95","article-title":"Evaluation and mitigation of cognitive biases in medical language models","volume":"7","author":"Schmidgall S","year":"2024","unstructured":"Schmidgall S, Harris C, Essien I, et al. Evaluation and mitigation of cognitive biases in medical language models. NPJ Dig Med 2024; 7: 95.","journal-title":"NPJ Dig Med"},{"key":"e_1_3_2_13_2","doi-asserted-by":"crossref","unstructured":"Emelin D Le Bras R Hwang JD et al. Moral stories: situated reasoning about norms intents actions and their consequences. In: Online and Punta Cana Dominican Republic November 2021 pp.698-718. Association for Computational Linguistics.","DOI":"10.18653\/v1\/2021.emnlp-main.54"},{"key":"e_1_3_2_14_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-025-18489-7"},{"key":"e_1_3_2_15_2","unstructured":"Hendrycks D Collin B Basart S et al. Aligning AI with shared human values. 2020. DOI: 10.48550\/arXiv.2008.02275."},{"key":"e_1_3_2_16_2","doi-asserted-by":"crossref","unstructured":"Omar M Soffer S Agbareia R et al. Socio-demographic biases in medical decision-making by large language models: a large-scale multi-model analysis. medRxiv 2024: 2024.2010.2029.24316368. DOI: 10.1101\/2024.10.29.24316368.","DOI":"10.1101\/2024.10.29.24316368"},{"key":"e_1_3_2_17_2","doi-asserted-by":"publisher","DOI":"10.1007\/s41666-025-00211-x"},{"key":"e_1_3_2_18_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41746-025-01792-y"},{"key":"e_1_3_2_19_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-025-16716-9"},{"key":"e_1_3_2_20_2","unstructured":"Wei J Meng Z Xiao Z et al. MedEthicsQA: a comprehensive question answering benchmark for medical ethics evaluation of LLMs. arXiv preprint arXiv:250622808 2025."},{"key":"e_1_3_2_21_2","first-page":"404","volume-title":"Albuquerque","author":"Jin H","year":"2025","unstructured":"Jin H, Shi J, Xu H, et al. Medethiceval: evaluating large language models based on Chinese medical ethics. In: Albuquerque. New Mexico: Association for Computational Linguistics, 2025, pp.404\u2013421."},{"key":"e_1_3_2_22_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-023-43436-9"},{"key":"e_1_3_2_23_2","doi-asserted-by":"publisher","DOI":"10.1136\/jme-2023-109366"},{"key":"e_1_3_2_24_2","doi-asserted-by":"publisher","DOI":"10.1055\/a-2405-0138"},{"key":"e_1_3_2_25_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-023-45223-y"},{"key":"e_1_3_2_26_2","doi-asserted-by":"publisher","DOI":"10.5840\/monist197659224"},{"key":"e_1_3_2_27_2","doi-asserted-by":"publisher","DOI":"10.1093\/0195108590.001.0001"},{"key":"e_1_3_2_28_2","doi-asserted-by":"publisher","DOI":"10.1056\/NEJMp1804094"},{"key":"e_1_3_2_29_2","doi-asserted-by":"publisher","DOI":"10.1186\/s12939-025-02419-0"}],"container-title":["DIGITAL HEALTH"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/journals.sagepub.com\/doi\/pdf\/10.1177\/20552076261428395","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/journals.sagepub.com\/doi\/full-xml\/10.1177\/20552076261428395","content-type":"application\/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/journals.sagepub.com\/doi\/pdf\/10.1177\/20552076261428395","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,4,28]],"date-time":"2026-04-28T17:56:15Z","timestamp":1777398975000},"score":1,"resource":{"primary":{"URL":"https:\/\/journals.sagepub.com\/doi\/10.1177\/20552076261428395"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,2]]},"references-count":28,"alternative-id":["10.1177\/20552076261428395"],"URL":"https:\/\/doi.org\/10.1177\/20552076261428395","relation":{},"ISSN":["2055-2076","2055-2076"],"issn-type":[{"value":"2055-2076","type":"print"},{"value":"2055-2076","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,2]]},"article-number":"20552076261428395"}}