Real-World Implementation of Large Language Models for Writing Clinical Discharge Summaries Within a Secure Data...

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Documentation

Background:

• Discharge summaries are crucial clinical reports detailing patient hospital stays.
• Manual creation of discharge summaries is time-consuming, with 80% of electronic health record data being free text.
• Automation of discharge summary generation using large language models (LLMs) is a potential solution.

Purpose of the Study:

• To develop and evaluate a template-based prompting system for generating clinically acceptable discharge summaries.
• Focus on the "clinical summary" and "plan and requested actions" sections.
• Utilize routinely collected electronic patient records for automated summary generation.

Main Methods:

• Employed electronic health record data from Imperial College Healthcare NHS Trust.
• Used a dataset of 52 diverse inpatient encounters (43 development, 9 test).
• Applied OpenAI's generative pretrained transformer-4 with structured template prompts to synthesize clinical notes into discharge summaries.

Main Results:

• 89% of generated summaries in the test dataset received a positive global confidence rating.
• High completeness (89%) and accuracy (78%) for the "clinical summary" section.
• Positive results for "plan and requested actions" section completeness (78%) and accuracy (78%), readability, formatting, bias, and harm.

Conclusions:

• The developed pipeline demonstrates feasibility for automated discharge summary generation.
• Further rigorous statistical evaluation in larger patient cohorts is necessary.
• LLM-based automation shows promise for improving clinical documentation efficiency.