Investigating expectations and needs regarding the use of large language models at Bavarian university clinics

TL;DR

This study explores how medical professionals in Bavarian clinics view the use of large language models, highlighting both potential benefits and concerns.

Contribution

The paper provides statistically grounded insights into the expectations and needs of medical professionals regarding LLM adoption in healthcare.

Findings

Participants use LLMs for research support, summarization, translation, and report drafting.

Most believe LLMs can improve personalized, evidence-based, and cost-effective patient treatment.

Concerns include data privacy, model opacity, and lack of institutional preparedness for LLM adoption.

Abstract

Recent advancements in Artificial Intelligence (AI) have been driven by Large Language Models (LLMs), powerful tools capable of generating coherent text and solving diverse analytical tasks. While LLMs hold great potential to enhance healthcare by assisting physicians and improving patient treatment, their clinical adoption is limited, and there is a lack of statistically grounded information on the opinions of medical professionals, personnel, and students regarding LLM usage. To address this gap, we conducted an online survey from April to October 2024, gathering insights from 120 participants across five Bavarian university clinics (in Germany), including physicians, medical students, and administrative staff. Findings show that many participants already use LLMs for research support, summarization, translation, and report drafting. Most believe LLMs will positively influence their field, acknowledge their potential to automate mundane tasks, and believe they will help to achieve a more personalized, evidence-based, and cost-effective patient treatment. However, concerns were shown regarding their opaque nature, data privacy, and the potential loss of patient trust. Participants overwhelmingly feel their institutions are not well-prepared for LLM adoption, with suggestions for improvement including increased education and specialized training, investments in digitalization and infrastructure, ensuring legal compliance, and encouraging technological openness. We hope these insights will inform the design of future medical AI solutions.

Figures