Adoption of Generative Large Language Models in Pathology: A National Survey of Chinese Pathologists (Preprint)

Abstract

<sec> <title>BACKGROUND</title> Pathologists are grappling with high workloads and uneven resource distribution, which can impede professional development and the delivery of quality patient care. The advent of generative large language models (LLMs) has the potential to revolutionize pathological field, where efficiency and resource accessibility are paramount. </sec> <sec> <title>OBJECTIVE</title> This study aimed to investigate the perceptions and willingness of Chinese pathologists to adopt generative LLMs. </sec> <sec> <title>METHODS</title> We conducted a questionnaire survey at the National Pathology Academic Annual Conference in April 2024, involving 339 certified Chinese pathologists. Participant responses were measured with a 5-point Likert scale for the performance of generative LLMs in clinical, research, and educational settings, with statistical analysis using mean and standard deviation (SD). Multivariable logistic regression models were employed to explore factors associated with the adoption of generative LLMs, reporting odds ratios (ORs) and 95% confidence intervals (CIs). </sec> <sec> <title>RESULTS</title> A total of 339 valid questionnaires were returned. The results revealed that pathologists generally supported the performance of generative LLMs in clinical (mean 3.87, SD 0.96), research (mean 3.88, SD 1.09), and educational (mean 4.04, SD 0.82) contexts. Positive attitudes towards the use of generative LLMs were prevalent. Notably, pathologists practicing in less developed urban areas (OR=1.99, 95% CI=1.07 to 3.69, p=0.030), those with higher caseloads (&gt;5000 cases/year; OR=2.12, 95% CI=1.01 to 4.44, p=0.047), and those engaged in research (OR=2.94, 95% CI=1.61 to 5.34, p&lt;0.001) and teaching (OR=2.37, 95% CI=1.42 to 3.96, p=0.001) activities, as well as those with prior experience with generative LLMs (OR=2.45, 95% CI=1.38 to 4.37, p=0.002), showed a greater inclination towards future adoption. </sec> <sec> <title>CONCLUSIONS</title> Chinese pathologists are receptive to generative LLMs, showing a positive inclination for their application. The study advocates for fostering the adoption of generative LLMs to improve the efficiency and accuracy of diagnosis, reduce the burden on pathologists, and improve the overall service level in the field of pathology. </sec>

Ähnliche Arbeiten