Attitudes Toward Large Language Models in Health Care and Preferences for Their Adoption and Oversight Among Health Care Professionals: Cross-Sectional Survey.

Abstract

Large language models (LLMs) are rapidly entering health care, but limited empirical data exist on health care professionals' perceptions. Understanding health care professionals' attitudes is essential for responsible implementation as LLMs transition from experimental to routine tools. To characterize health care professionals' perspectives on LLM use in health care, including exposure, knowledge, perceived clinical utility, safety and bias concerns, and oversight preferences. This cross-sectional survey was distributed online through a health care news platform mailing list. A total of 335 health care professionals responded, including attending physicians (n=230, 68.7%), residents or fellows, nurse practitioners, physician assistants, and researchers. Most were aged 30 to 59 years (n=243, 72.5%) and practiced in the Northeast United States (n=261, 77.9%). Outcomes included LLM use patterns, knowledge levels, perceived applications, safety and bias concerns, and preferences for regulatory oversight. Analyses included descriptive statistics, Wilcoxon rank-sum tests, <i>χ</i>² tests, and Spearman correlations. Of 335 participants, 62.7% (n=210) reported current or contemplated LLM use. Users reported significantly higher self-reported knowledge than nonusers (<i>P</i><.001). Age was not associated with knowledge (ρ=-0.072; <i>P</i>=.19). Participants identified literature review (n=246, 73.4%), decision support (n=191, 57%), and patient communication (n=184, 54.9%) as the most valuable applications. Concerns included decision errors (n=253, 75.5%) and algorithmic bias (n=245, 73.1%); nearly all respondents (n=323, 96.4%) expressed concern about bias, and those who had observed bias reported higher concern levels (<i>P</i><.001). Participants favored regulation by professional associations (n=219, 65.4%) over technology companies (n=97, 29%), with 87.8% (n=294) supporting professional guidelines. Confidence in existing oversight was low, with 66.6% (n=223) reporting none. In this exploratory convenience sample, health care professionals reported early adoption of LLMs for lower-risk tasks while expressing concerns about safety, bias, and governance. Given the low response rate and recruitment through a health care innovation-focused mailing list, these findings may not reflect the views of the broader health care professional population. Respondents preferred professional organizations over industry for oversight and suggested that successful integration of LLMs into health care will require careful planning, human supervision, transparent disclosure, and auditing. Future studies using more representative sampling methods are needed to better characterize health care professionals' attitudes toward LLMs.