Physicians' Knowledge, Perceptions and Use of Large Language Models in Clinical Practice: A cross-sectional study at...

This study assessed the knowledge, perceptions and use of ChatGPT and other large language models (LLMs) among physicians at Sultan Qaboos University Hospital (SQUH), Oman. It also explored perceived benefits, barriers and ethical concerns regarding artificial intelligence (AI) integration into clinical practice.

This cross-sectional study was conducted between September and December 2024 using a structured online questionnaire distributed to physicians across different specialties. The survey covered demographics, familiarity with LLMs, applications in medical education, research, clinical care, administrative tasks and ethical considerations.

A total of 146 physicians were included (response rate = 48.7%); 65.1% were familiar with ChatGPT or other LLMs and 70.5% had used them, mainly for education (47.9%) and research (46.6%). Use in clinical practice (29.5%) and administrative tasks (18.5%) was less frequent. Most physicians perceived LLMs as enhancing research (82.9%), education (79.5%), administrative work (74.4%) and patient care (54.1%). While 89.8% believed LLMs could improve professional work, only 39.7% expressed confidence in integrating outputs while upholding academic standards. Ethical concerns were widespread (96.6%), focusing on reliability, accuracy and data privacy. Despite low awareness of ethical guidelines (15.1%), 87.0% expressed willingness to engage in AI-related training. Male physicians reported higher use for research and diagnostics (P <0.05 each), younger physicians (<30 years) more often used LLMs for education (P = 0.037) and senior physicians more frequently perceived bias in outputs (P = 0.045).

Physicians at SQUH demonstrated moderate familiarity and cautious optimism towards LLMs. Addressing gaps in training and ethical awareness is crucial for responsible AI adoption in clinical and academic practice.