{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,17]],"date-time":"2026-10-17T20:24:38Z","timestamp":1792268678885,"version":"4.4.0"},"reference-count":38,"publisher":"Public Library of Science (PLoS)","issue":"6","license":[{"start":{"date-parts":[[2026,6,18]],"date-time":"2026-06-18T00:00:00Z","timestamp":1781740800000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["www.plosdigitalhealth.org"],"crossmark-restriction":false},"short-container-title":["PLOS Digit Health"],"abstract":"<jats:p>Bias and inequity in palliative care disproportionately affect marginalised groups. Large language models (LLMs), such as GPT-4o, hold potential to enhance care but risk perpetuating biases present in their training data. This study aimed to systematically evaluate whether GPT-4o propagates biases in palliative care responses using adversarially designed datasets. In July 2024, GPT-4o was probed using the Palliative Care Adversarial Dataset (PCAD), and responses were evaluated by three palliative care experts in Canada and the United Kingdom using validated bias rubrics. The PCAD comprised PCAD-Direct (100 adversarial questions) and PCAD-Counterfactual (84 paired scenarios). These datasets targeted four care dimensions (access to care, pain management, advance care planning, and place of death preferences) and three identity axes (ethnicity, age, and diagnosis). Bias was detected in a substantial proportion of responses. For adversarial questions, the pooled bias rate was 0.33 (95% confidence interval [CI]: 0.26, 0.40); \u201callows biased premise\u201d was the most frequently identified dimension of bias (0.47; 95% CI: 0.40, 0.55), such as failing to challenge stereotypes. For counterfactual scenarios, the pooled bias rate was 0.26 (95% CI: 0.20, 0.32), with \u201cpotential for withholding\u201d as the most frequently identified dimension of bias (0.25; 95% CI: 0.17, 0.34), such as withholding interventions based on identity. Bias rates were consistent across care dimensions and identity axes. GPT-4o has the potential to perpetuate biases in palliative care, with implications for clinical decision-making and equity. The PCAD datasets provide novel tools to assess and address LLM bias in palliative care.<\/jats:p>","DOI":"10.1371\/journal.pdig.0001451","type":"journal-article","created":{"date-parts":[[2026,6,18]],"date-time":"2026-06-18T17:21:38Z","timestamp":1781803298000},"page":"e0001451","update-policy":"https:\/\/doi.org\/10.1371\/journal.pdig.corrections_policy","source":"Crossref","is-referenced-by-count":1,"title":["Large language models perpetuate bias in palliative care: Development and analysis of the Palliative Care Adversarial Dataset (PCAD)"],"prefix":"10.1371","volume":"5","author":[{"ORCID":"https:\/\/orcid.org\/0009-0006-3856-4356","authenticated-orcid":true,"given":"Naomi","family":"Akhras","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Fares","family":"Antaki","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Fannie","family":"Mottet","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Olivia","family":"Nguyen","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shyam","family":"Sawhney","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sabrina","family":"Bajwah","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Joanna M.","family":"Davies","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"340","published-online":{"date-parts":[[2026,6,18]]},"reference":[{"key":"pdig.0001451.ref001","unstructured":"Palliative care. https:\/\/www.who.int\/news-room\/fact-sheets\/detail\/palliative-care"},{"key":"pdig.0001451.ref002","unstructured":"Definition of inequity. https:\/\/www.merriam-webster.com\/dictionary\/inequity"},{"issue":"3","key":"pdig.0001451.ref003","doi-asserted-by":"crossref","first-page":"303","DOI":"10.1016\/j.soncn.2018.06.011","article-title":"Social Inequalities in Palliative Care for Cancer Patients in the United States: A Structured Review","volume":"34","author":"R Elk","year":"2018","journal-title":"Semin Oncol Nurs"},{"issue":"1","key":"pdig.0001451.ref004","doi-asserted-by":"crossref","first-page":"46","DOI":"10.1186\/s12904-022-00923-6","article-title":"Does ethnicity affect pain management for people with advanced disease? A mixed methods cross-national systematic review of \u201cvery high\u201d Human Development Index English-speaking countries","volume":"21","author":"G Clarke","year":"2022","journal-title":"BMC Palliat Care"},{"issue":"1","key":"pdig.0001451.ref005","doi-asserted-by":"crossref","first-page":"43","DOI":"10.1186\/s12904-023-01168-7","article-title":"How does ethnicity affect presence of advance care planning in care records for individuals with advanced disease? A mixed-methods systematic review","volume":"22","author":"J Crooks","year":"2023","journal-title":"BMC Palliat Care"},{"issue":"11","key":"pdig.0001451.ref006","doi-asserted-by":"crossref","DOI":"10.1503\/cmaj.190655","article-title":"Association between Chinese or South Asian ethnicity and end-of-life care in Ontario, Canada","volume":"192","author":"CJ Yarnell","year":"2020","journal-title":"CMAJ"},{"issue":"2","key":"pdig.0001451.ref007","doi-asserted-by":"crossref","first-page":"142","DOI":"10.1136\/bmjspcare-2020-002719","article-title":"Hospice care access inequalities: a systematic review and narrative synthesis","volume":"12","author":"J Tobin","year":"2022","journal-title":"BMJ Support Palliat Care"},{"issue":"4","key":"pdig.0001451.ref008","doi-asserted-by":"crossref","first-page":"731","DOI":"10.1016\/j.jpainsymman.2018.12.338","article-title":"Examining Age Inequalities in Operationalized Components of Advance Care Planning: Truncation of the ACP Process With Age","volume":"57","author":"LC Prater","year":"2019","journal-title":"J Pain Symptom Manage"},{"key":"pdig.0001451.ref009","author":"J Dixon","year":"2015","journal-title":"Equity in the provision of palliative care in the UK: review of evidence"},{"issue":"3","key":"pdig.0001451.ref010","article-title":"Changing patterns in place of cancer death in England: a population-based study","volume":"10","author":"GUIDE_Care project","year":"2013","journal-title":"PLoS Med"},{"key":"pdig.0001451.ref011","doi-asserted-by":"crossref","first-page":"34","DOI":"10.1186\/s12904-015-0033-z","article-title":"Comparison of terminally ill cancer- vs. non-cancer patients in specialized palliative home care in Germany - a single service analysis","volume":"14","author":"S Stiel","year":"2015","journal-title":"BMC Palliat Care"},{"issue":"3","key":"pdig.0001451.ref012","doi-asserted-by":"crossref","first-page":"81","DOI":"10.15406\/hpmij.2024.07.00244","article-title":"Incorporating artificial intelligence in palliative care: opportunities and challenges","volume":"7","author":"U Munive Jesus","year":"2024","journal-title":"HPMIJ"},{"issue":"1","key":"pdig.0001451.ref013","doi-asserted-by":"crossref","first-page":"141","DOI":"10.1038\/s43856-023-00370-1","article-title":"The future landscape of large language models in medicine","volume":"3","author":"J Clusmann","year":"2023","journal-title":"Commun Med (Lond)"},{"issue":"44","key":"pdig.0001451.ref014","doi-asserted-by":"crossref","DOI":"10.1073\/pnas.2313790120","article-title":"Large language models show human-like content biases in transmission chain experiments","volume":"120","author":"A Acerbi","year":"2023","journal-title":"Proc Natl Acad Sci U S A"},{"issue":"1","key":"pdig.0001451.ref015","doi-asserted-by":"crossref","first-page":"195","DOI":"10.1038\/s41746-023-00939-z","article-title":"Large language models propagate race-based medicine","volume":"6","author":"JA Omiye","year":"2023","journal-title":"NPJ Digit Med"},{"issue":"12","key":"pdig.0001451.ref016","doi-asserted-by":"crossref","first-page":"3590","DOI":"10.1038\/s41591-024-03258-2","article-title":"A toolbox for surfacing health equity harms and biases in large language models","volume":"30","author":"SR Pfohl","year":"2024","journal-title":"Nat Med"},{"key":"pdig.0001451.ref017","article-title":"Evaluation of biases in large language models in palliative care","author":"N Akhras","year":"2024","journal-title":"King\u2019s College London"},{"key":"pdig.0001451.ref018","article-title":"A Toolbox for Surfacing Health Equity Harms and Biases in Large Language Models","author":"SR Pfohl","year":"2024","journal-title":"arXiv"},{"issue":"1","key":"pdig.0001451.ref019","doi-asserted-by":"crossref","first-page":"60","DOI":"10.1038\/s41591-024-03425-5","article-title":"The TRIPOD-LLM reporting guideline for studies using large language models","volume":"31","author":"J Gallifant","year":"2025","journal-title":"Nat Med"},{"key":"pdig.0001451.ref020","article-title":"Counterfactual Fairness","author":"MJ Kusner","year":"2017","journal-title":"arXiv [stat.ML]"},{"key":"pdig.0001451.ref021","article-title":"Towards Expert-Level Medical Question Answering with Large Language Models","author":"K Singhal","year":"2023","journal-title":"arXiv"},{"key":"pdig.0001451.ref022","article-title":"A statistical framework for evaluating the repeatability and reproducibility of large language models","author":"C Shyr","year":"2025","journal-title":"medRxiv"},{"key":"pdig.0001451.ref023","unstructured":"BLEU and ROUGE score for NLP evaluation. GeeksforGeeks. 2024. https:\/\/www.geeksforgeeks.org\/nlp\/understanding-bleu-and-rouge-score-for-nlp-evaluation\/"},{"key":"pdig.0001451.ref024","doi-asserted-by":"crossref","first-page":"102545","DOI":"10.1016\/j.mex.2023.102545","article-title":"K-Alpha Calculator-Krippendorff\u2019s Alpha Calculator: A user-friendly tool for computing Krippendorff\u2019s Alpha inter-rater reliability coefficient","volume":"12","author":"G Marzi","year":"2024","journal-title":"MethodsX"},{"issue":"1","key":"pdig.0001451.ref025","doi-asserted-by":"crossref","first-page":"159","DOI":"10.2307\/2529310","article-title":"The measurement of observer agreement for categorical data","volume":"33","author":"JR Landis","year":"1977","journal-title":"Biometrics"},{"key":"pdig.0001451.ref026","article-title":"GPTBIAS: A comprehensive framework for evaluating bias in large language models","author":"J Zhao","year":"2023","journal-title":"arXiv"},{"issue":"23","key":"pdig.0001451.ref027","doi-asserted-by":"crossref","first-page":"2255","DOI":"10.1001\/jama.2023.22557","article-title":"Automation Bias and Assistive AI: Risk of Harm From AI-Driven Clinical Decision Support","volume":"330","author":"R Khera","year":"2023","journal-title":"JAMA"},{"issue":"1","key":"pdig.0001451.ref028","doi-asserted-by":"crossref","first-page":"19","DOI":"10.1186\/s12910-017-0179-8","article-title":"Implicit bias in healthcare professionals: a systematic review","volume":"18","author":"C FitzGerald","year":"2017","journal-title":"BMC Med Ethics"},{"issue":"11","key":"pdig.0001451.ref029","doi-asserted-by":"crossref","first-page":"1504","DOI":"10.1007\/s11606-013-2441-1","article-title":"Physicians and implicit bias: how doctors may unwittingly perpetuate health care disparities","volume":"28","author":"EN Chapman","year":"2013","journal-title":"J Gen Intern Med"},{"issue":"10","key":"pdig.0001451.ref030","doi-asserted-by":"crossref","DOI":"10.1371\/journal.pone.0257794","article-title":"Is physician implicit bias associated with differences in care by patient race for metastatic cancer-related pain?","volume":"16","author":"K Fiscella","year":"2021","journal-title":"PLoS One"},{"key":"pdig.0001451.ref031","article-title":"Should ChatGPT be Biased? Challenges and Risks of Bias in Large Language Models","author":"E Ferrara","year":"2023","journal-title":"arXiv"},{"key":"pdig.0001451.ref032","article-title":"The Equitable AI Research Roundtable (EARR): Towards Community-Based Decision Making in Responsible AI Development","author":"J Smith-Loud","year":"2023","journal-title":"arXiv"},{"key":"pdig.0001451.ref033","article-title":"PALLM: Evaluating and Enhancing PALL iative Care Conversations with L arge L anguage M odels","author":"Z Wang","year":"2025","journal-title":"ACM Trans Comput Healthcare"},{"key":"pdig.0001451.ref034","article-title":"Review of Large Language Models for Patient and Caregiver Support in Cancer Care Delivery","volume":"9","author":"R Kouzy","year":"2025","journal-title":"JCO Clin Cancer Inform"},{"issue":"11","key":"pdig.0001451.ref035","doi-asserted-by":"crossref","first-page":"1329","DOI":"10.1089\/jpm.2013.9468","article-title":"Racial and ethnic disparities in palliative care","volume":"16","author":"KS Johnson","year":"2013","journal-title":"J Palliat Med"},{"issue":"10","key":"pdig.0001451.ref036","doi-asserted-by":"crossref","first-page":"1371","DOI":"10.1136\/bjo-2023-324438","article-title":"Capabilities of GPT-4 in ophthalmology: an analysis of model entropy and progress towards human-level medical question answering","volume":"108","author":"F Antaki","year":"2024","journal-title":"Br J Ophthalmol"},{"issue":"4","key":"pdig.0001451.ref037","doi-asserted-by":"crossref","first-page":"100324","DOI":"10.1016\/j.xops.2023.100324","article-title":"Evaluating the Performance of ChatGPT in Ophthalmology: An Analysis of Its Successes and Shortcomings","volume":"3","author":"F Antaki","year":"2023","journal-title":"Ophthalmol Sci"},{"key":"pdig.0001451.ref038","article-title":"Personas shift clinical action thresholds in large language models","author":"E Klang","year":"2026","journal-title":"medRxiv"}],"container-title":["PLOS Digital Health"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dx.plos.org\/10.1371\/journal.pdig.0001451","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,18]],"date-time":"2026-06-18T17:21:44Z","timestamp":1781803304000},"score":1,"resource":{"primary":{"URL":"https:\/\/dx.plos.org\/10.1371\/journal.pdig.0001451"}},"subtitle":[],"editor":[{"given":"Po-Chih","family":"Kuo","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"editor"}]}],"short-title":[],"issued":{"date-parts":[[2026,6,18]]},"references-count":38,"journal-issue":{"issue":"6","published-online":{"date-parts":[[2026,6,18]]}},"URL":"https:\/\/doi.org\/10.1371\/journal.pdig.0001451","relation":{},"ISSN":["2767-3170"],"issn-type":[{"value":"2767-3170","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,6,18]]}}}