Project title

The feasibility of large language models in extracting discharge summaries (FLEX-DS)

Country

UK

Background

Comprehensive, accessible, and timely discharge summaries are critical to the safe transfer of care from hospital to home. The summaries mark a handover of caring responsibilities to primary care services for patients and families. However, writing discharge summaries can often be delegated to overburdened junior medical staff, with limited training and/or limited knowledge of the patient.

Given the importance of these summaries on patient care and safety, we need to identify ways in which discharge summaries can be improved. This will reduce completion time and physician burnout, speeding up patient discharge, improving patient flow, and increasing both the accuracy and accessibility of discharge summaries.

Summary

This feasibility study will assess if a large language model (LLM), informed by patients and healthcare staff, can generate accurate, accessible, comprehensive, and acceptable discharge summaries. 

To achieve this, the project will a dopt a 4-step sprint cycle:

  • Step 1: Create discharge summaries using an LLM
  • Step 2: Review of summaries by multidisciplinary healthcare staff
  • Step 3: Review of summaries by patient and public group
  • Step 4: Feedback any comments from steps 2 and 3 to improve the AI output

Each sprint will look to enhance the quality and accessibility of the discharge summaries across different healthcare specialties.

Outcome

The research will provide two main results:

1. Components that make up a good discharge summary

As part of the sprints, we will determine the most important components of a discharge summary according to both clinicians and patients. These components may align or they may differ, meaning one summary may be required specifically for patients and families and one for clinicians.

2. An LLM to write patient summaries

After the sprints have been completed, we will have an LLM model that is able to draft discharge summaries that meet the requirements of clinicians, patients, and their families. If the approach proves feasible, we will look into integrating the LLM into the workflows of a secondary care setting, assessing its real-world impact on both staff and patients.