QUB-Cirdan at "Discharge Me!": Zero shot discharge letter generation by open-source LLM

TL;DR

This paper introduces a zero-shot, retrieval-augmented approach using open-source LLMs to automate discharge letter sections, aiming to reduce clinicians' administrative workload with high accuracy and efficiency.

Contribution

It presents a novel zero-shot, template-based method with RAG integration for discharge summary generation, improving reliability and contextual accuracy.

Findings

Achieved high evaluation scores across metrics

Demonstrated effective zero-shot generation with RAG

Provided insights from unsuccessful experiments

Abstract

The BioNLP ACL'24 Shared Task on Streamlining Discharge Documentation aims to reduce the administrative burden on clinicians by automating the creation of critical sections of patient discharge letters. This paper presents our approach using the Llama3 8B quantized model to generate the "Brief Hospital Course" and "Discharge Instructions" sections. We employ a zero-shot method combined with Retrieval-Augmented Generation (RAG) to produce concise, contextually accurate summaries. Our contributions include the development of a curated template-based approach to ensure reliability and consistency, as well as the integration of RAG for word count prediction. We also describe several unsuccessful experiments to provide insights into our pathway for the competition. Our results demonstrate the effectiveness and efficiency of our approach, achieving high scores across multiple evaluation metrics.