Understanding the concerns and choices of public when using large language models for healthcare

Large language models (LLMs) have shown their potential in biomedical fields. However, how the public uses them for healthcare purposes such as medical Q&A, self-diagnosis, and daily healthcare information seeking is under-investigated. In this paper, we adopt a mixed-methods approach, including surveys (N=167) and interviews (N=17) to investigate how and why the public uses LLMs for healthcare. LLMs as a healthcare tool have gained popularity, and are often used in combination with other information channels such as search engines and online health communities to optimize information quality. LLMs provide more accurate information and a more convenient interaction/service model compared to traditional channels. LLMs also do a better job of reducing misinformation, especially in daily healthcare questions. Doctors using LLMs for diagnosis is less acceptable than for auxiliary work such as writing medical records. Based on the findings, we reflect on the ethical and effective use of LLMs for healthcare and propose future research directions.