Scientific Evidence for Clinical Text Summarization Using Large Language Models: Scoping Review.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Natural Language Processing

Background:

• Clinicians face information overload from electronic health records.
• Automatic summarization using large language models (LLMs) is a growing area of research.
• A structured overview of LLM-based clinical text summarization is needed.

Purpose of the Study:

• To review the state of the art in clinical text summarization using LLMs.
• To evaluate the evidence level of existing research.
• To assess the clinical applicability of current summarization findings.

Main Methods:

• Scoping review following PRISMA-ScR guidelines.
• Searched 5 databases for literature from January 2019 to June 2024.
• Included studies on transformer-based models for clinical text summarization using free-text data.

Main Results:

• 30 studies analyzed, predominantly retrospective observational designs with real patient data.
• Research focus is narrow, often on radiology reports from intensive care units, primarily in the US.
• Summarization methods are mainly abstractive, with inconsistent reporting and heterogeneous evaluation frameworks; external validation and safety analyses are rare.

Conclusions:

• Significant barriers exist for translating current research into clinical practice.
• The field is exploratory, with limited scope and insufficient evaluation of performance and clinical impact.
• Advancement requires broader scope, robust evaluation, and focus on real-world applicability, safety, and fairness.