A Prototype of AI Assistant for Summarizing Treatment Progress with Large Language Models

Abstract

In response to the regulatory limit on overtime work for physicians to be introduced in 2024, there is a growing demand for further efficiency improvements in medical practices. A time study in the clinical observation revealed the necessity for reducing the time spent creating medical documents. Consequently, we develop an AI assistant to support writing the treatment summary, which is time-consuming task. In a prototype of the AI assistant, we use the medical NER (named entity recognition) and LLM (large language model), which is fine-tuned or trained by our group. Our prototype application first extracts medical entities from a set of progress notes. Then, it generates the treatment summary from medical entities manually selected by physicians. We evaluated the effectiveness of our approach via a questionnaire survey of physicians after PoC trials. The results indicated our approach is expected to shorten the writing time for treatment summaries effectively.