Automated Generation of Hospital Discharge Summaries Using Clinical Guidelines and Large Language Models

Simon Ellershaw, Christopher Tomlinson, Oliver E Burton, Thomas Frost, John Gerrard Hanrahan, Danyal Zaman Khan, Hugo Layard Horsfall, Mollie Little, Evaleen Malgapo, Joachim Starup-Hansen, Jack Ross, Martinique Vella-Baldacchino, Kawsar Noor, Anoop D. Shah, Richard Dobson

Track: Non-traditional track

Keywords: Large Language Models, Discharge Summaries, Clinical Guidelines, Automation

TL;DR: We evaluate the ability of large language models to generate hospital discharge summaries when prompted by clinical guidelines.

Abstract: Discharge summaries are essential yet time-consuming documents doctors write at the end of a patient's hospital stay. They are the primary form of communication between hospital and community care teams. The automatic generation of summaries could reduce the administrative burden on doctors. We propose to use large language models, few-shot prompted by clinical guidance, to perform this task. Unlike previous supervised approaches, our method does not require a large training dataset, can accept full-length physician notes as inputs and is explicitly guided by clinical best practice. We implemented such a system using Royal College of Physicians London guidelines, GPT-4-turbo and MIMIC-III physician notes. 53 summaries were evaluated by 10 clinicians and found to have a micro accuracy of 0.81. Finally, we discuss methodical limitations and the required future improvements to the evaluation framework.

Presentation And Attendance Policy: I have read and agree with the symposium's policy on behalf of myself and my co-authors.

Ethics Board Approval: No, our research does not involve datasets that need IRB approval or its equivalent.

Data And Code Availability: Yes, we will make data and code available upon acceptance.

Primary Area: Software demonstrations and position papers

Student First Author: Yes, the primary author of the manuscript is a student.

Submission Number: 4