A proposal for providing accurate evidence in medicine by applying large language models

Abstract

[Background] Artificial intelligence (AI) technology has already permeated our daily lives. However, since LLMs are not specialized for medical knowledge, the risk of hallucination remains, leading to miscommunication in the field of medical treatment. Even health care professionals take time and effort to obtain appropriate information. In this situation, we have developed a method to obtain appropriate medical evidence from LLM by incorporating medical articles.[Methods] A web service was constructed using GPT, DeepL, and PubMed API and Flask. Question in Japanese was automatically translated into English with compulsory keywords to search relevant articles in PubMed. Based on these abstracts, GPT answered the question. Downloadable articles were also automatically downloaded and summarized in response to the question.[Results] LLM can now provide the accurate medical evidence in response to the Japanese questionnaires. Since the process is almost automatic, any special medical knowledge is not required to ask the question.[conclusion] LLM has made it possible to automate the collection of medical evidence that was previously been done manually. We now can access the accurate medical knowledge dramatically easily, not only for medical professionals but also for those who are not specialized in medicine.