Large Language Models in Action: Supporting Clinical Evaluation in an Infectious Disease Unit

Area of Science:

• Infectious Diseases
• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Healthcare-associated infections (HAIs), including sepsis, pose significant clinical and systemic challenges.
• Large language models (LLMs) present a novel approach for analyzing clinical data and generating guideline-based infection management recommendations.

Purpose of the Study:

• To evaluate the performance of LLMs in extracting and assessing clinical data for infection prevention and management.
• To determine the appropriateness of LLM-generated recommendations against international guidelines for sepsis patients.

Main Methods:

• Retrospective proof-of-concept study involving seven sepsis patients' clinical documentation.
• Analysis of five domains: antibiotic therapy, isolation, urinary catheter, infusion line, and pressure ulcer management.
• Utilized ChatGPT-4o to assess Italian clinical records against international guidelines.

Main Results:

• LLM demonstrated accuracy in antibiotic therapy and urinary catheter management, including indication identification and removal protocols.
• Identified errors in isolation measures (incorrect contact precautions) and pressure ulcer care (false lesion identification).

Conclusions:

• LLMs show potential as tools to support evidence-based practice in infection management.
• Further development is needed to improve LLM accuracy in specific clinical areas like isolation and wound care.