Talking risk: how well do LLMs communicate radiation exposure?

Abstract

Artificial intelligence is becoming used more frequently in healthcare, particularly to generate patient education material to improve communication around patient concerns, general medical conditions and treatment overviews. This has the potential to enhance understanding, reduce health literacy barriers and support shared decision making, which can improve patient outcomes. ​In radiology, no current study investigates the ability of large language models (LLMs) to communicate radiation risk. Patient anxiety around the exposure to ionising radiation is common, hence effectively communicating these risks to patients to enable informed consent is essential, but is an inherently complex task.