A Knowledge-Enhanced Platform (MetaSepsisKnowHub) for Retrieval Augmented Generation-Based Sepsis Heterogeneity and...

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Medicine
• Critical Care Medicine

Background:

• Sepsis presents a significant clinical challenge due to its heterogeneity and high mortality.
• Precision medicine approaches are crucial for improving sepsis diagnosis and treatment.
• Current clinical decision-making for sepsis requires enhanced support systems.

Purpose of the Study:

• To extract and consolidate sepsis biomarkers for comprehensive biomedical information.
• To enhance the accuracy, stability, and interpretability of large language model (LLM)-driven clinical decisions.
• To integrate retrieval augmented generation (RAG) and prompt engineering for improved sepsis management.

Main Methods:

• Developed MetaSepsisKnowHub, a knowledge-enhanced platform with 427 sepsis biomarkers from 423 studies.
• Curated a tailored LLM framework incorporating RAG and prompt engineering.
• Evaluated performance using the System Usability Scale and Net Promoter Score.

Main Results:

• RAG-based recommendations showed statistically significant improvements over baseline LLMs for textual questions.
• RAG demonstrated superior factual correctness, accuracy, and knowledge recall compared to baseline LLMs.
• High user satisfaction reported with an average System Usability Scale score of 82.20 and Net Promoter Score of 72.

Conclusions:

• The MetaSepsisKnowHub platform, combined with RAG, enhances LLM capabilities for precision in critical care.
• This approach bridges the gap between research and clinical practice, accelerating bench-to-bedside translation.
• Presents a knowledge-enhanced paradigm for AI applications in critical care medicine.