Imperial College London

Academic Institute

Tim Orchard

Unknown

No

NIBDAPC_2024_0031

A discharge summary (also called a ‘discharge letter’) provides a summary of a patient’s hospital stay and includes key information such as why the patient was admitted to hospital, and tests and treatments received; it also includes instructions for GPs about ongoing care the patient needs (for example, stitches to be removed, medications to be started or stopped). Discharge summaries are usually written by junior doctors who work under extreme pressure. The discharge summary can be time-consuming to write and it is very difficult for doctors find time to do this administrative task whilst also seeing and treating patients who are ill in hospital. But patients can experience delays in their discharge or problems with their care after they leave the hospital if the discharge summary contains errors. This project will explore whether ChatGPT has the potential to help junior doctors to produce high-quality discharge summaries. Microsoft ChatGPT is a special type of artificial intelligence (AI) application that can generate and summarise human-like text based on information it is presented with. We want to understand whether ChatGPT can generate discharge summaries when presented with documents from the patient’s hospital record (including, for example, ward round notes, operation notes, test results). This project will not use the publicly available version of ChatGPT; a special version will be installed in the iCARE secure data environment. The iCARE secure data environment is a secure platform that holds patient data which has been de-identified (names and personal details removed) for research and audit purposes. Individual patients cannot be identified from the data and the iCARE environment can only be accessed by researchers who have received approval to do so. Patient data is not shared with Microsoft. Junior doctors on the project team will create text ‘prompts’ – instructions that tell ChatGPT what to do, for example: “You are doctor responsible for looking after hospital inpatients. Write a discharge summary including the following information […].” The prompts will also tell ChatGPT what NOT to do to ensure it doesn’t produce any false information, for example:“Generate the patient discharge summary based solely on the information documented in the patients’ notes.” We will do a study to assess whether Chat-GPT can generate discharge summaries that are as good as the ones written by junior doctors. If the results of this study are positive, we will do further research to understand how ChatGPT could improve on the discharge summaries written by junior doctors, for example, by tailoring them to different audiences (patients, carers, GPs, care homes, languages other than English) and by using recognised standards for discharge summaries (published by the Professional Records Standards Body – an organisation that makes recommendations about documentation in patients’ medical records).

This project is an initial steps towards patient benefitting patients. Assuming this project shows that ChatGPT can produce discharge summaries of the same quality as those written by junior doctors, implementation of the technology at Imperial College Healthcare NHS Trust (and beyond) could lead to to the following benefits: • reduced delays for patients waiting to be discharged from hospital • time usually spent writing discharge summaries could be released back to clinicians to spend directly assessing/treating patients or communicating with them face-to-face about their discharge • patients, carers and primary care practitioners having an improved understanding of discharge instructions from the acute setting thereby improving care continuity and minimising the risk of errors (e.g. patients taking their medications incorrectly or missing an important follow-up appointment).

Public Health Research

03/05/2024

06/02/2024

De-Personalised

Not applicable

Not applicable

One-off

18/09/2025

TRE