Physicians at three major university-affiliated hospitals in Chongqing, China, hold a cautiously optimistic view of large language models in healthcare, according to a stratified random survey conducted between October and December 2023. Of the 419 physicians who responded, 211, or 50.3 percent, expressed a positive view of the technology, while the remainder stopped short of endorsing it.
The survey used a stratified random sampling approach across the three hospitals, a design intended to capture a representative mix of physicians rather than a self-selected convenience sample. Participants answered a 20-item questionnaire covering their demographic characteristics, their perception of large language models, the demand for specific applications in medical practice, and the factors shaping their acceptance. Researchers then analyzed the responses with a Generalized Linear Mixed Model to identify which characteristics were statistically associated with holding positive attitudes toward the technology.
One factor stood out clearly: hands-on experience. Physicians who had previously used large language models were significantly more likely to hold positive attitudes, with an odds ratio of 1.58 and a 95 percent confidence interval of 1.05 to 2.37.
Demand for specific applications was substantial. Respondents most frequently wanted large language models for diagnostic assistance, cited by 73.7 percent, followed by literature searches at 70.6 percent and data analysis at 69.2 percent.
Yet enthusiasm was tempered by persistent reservations. Concerns about the maturity of the technology were the most common, voiced by 69.0 percent of respondents. Data privacy and security followed at 58.5 percent, and 58 percent worried about the risk of large language models generating erroneous content.
The near-even split in overall attitudes, with just over half of respondents positive, captures the tension embedded in those numbers. Physicians see genuine utility in applications that accelerate literature review, streamline data analysis, and assist diagnosis, but they also recognize that the technology is still maturing.
The authors conclude that the study reveals a cautiously optimistic stance among physicians from the three university-affiliated hospitals in Chongqing, highlighting both growing demand for and reservations about these technologies.
Several limitations shape how the findings should be read. The survey covered physicians at three hospitals in a single Chinese city, all university-affiliated, so the results describe that population and may not generalize to physicians elsewhere in China or in other countries. The data were collected between October and December 2023, a period when large language models in medicine were evolving rapidly, and attitudes may have shifted since. As a cross-sectional survey, the study captures associations at one point in time; the link between prior use and positive attitudes is statistical and does not establish that experience causes favorable views.
Even with those caveats, the survey provides a quantitative baseline for a debate that has often been driven by anecdote. It quantifies both the appetite for diagnostic assistance, literature searching, and data analysis among working physicians and the specific worries, about technological maturity, privacy, and erroneous output, that temper that appetite.
Subject of Research: Physician attitudes toward large language models in healthcare surveyed in Chinese hospitals
Article Title: Attitudes of physicians towards large language models in healthcare: insights from a stratified random survey in China
Article References: Wei, Q., Zhang, D., Qu, L., & Xu, X. (2026). Attitudes of physicians towards large language models in healthcare: insights from a stratified random survey in China. BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03941-z
Image Credits: AI Generated
DOI: 10.1186/s12911-026-03941-z
Keywords: large language models, physicians, survey, China, healthcare AI, diagnostic assistance, data privacy, clinical workflow, stratified sampling, technology acceptance, Attitudes, towards
News Source: Ophelia Keating. (November 4, 2026). Half of surveyed Chinese physicians view AI language models positively, hospital poll finds. Scienmag.
