Physician Perspectives on Large Language Models in Health Care: A Cross-Sectional Survey Study

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Physician Practice Patterns

Background:

• Large language models (LLMs) are increasingly integrated into healthcare.
• Physician adoption and perception of LLMs are crucial for effective implementation.
• Understanding current practices and concerns is vital for developing supportive policies.

Purpose of the Study:

• To evaluate physicians' current practices and perspectives on large language models (LLMs) in healthcare.
• To compare physician attitudes towards LLMs at academic medical centers with and without institutional access.
• To identify key areas for physician education regarding LLM use in clinical settings.

Main Methods:

• A cross-sectional survey was conducted among 306 physicians (faculty and trainees) at two academic medical centers.
• One center had institutional LLM access, while the other did not.
• Data collected included LLM usage, perceived importance of evaluation metrics, liability concerns, and training needs.

Main Results:

• 64.4% of physicians reported using LLMs, with accuracy being the top priority.
• Physicians at the center with institutional LLM access reported significantly lower liability concerns (49.2% vs. 66.7%).
• 94% of physicians desired further training, prioritizing clinical applications and risk management; only 2.6% recommended LLMs to patients.

Conclusions:

• Physician adoption of LLMs is prevalent, with accuracy as a primary concern.
• Institutional LLM access may mitigate liability concerns among physicians.
• Medical centers should develop targeted policies and educational programs to support responsible LLM integration.