Streamlining evidence based clinical recommendations with large language models

Area of Science:

• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems
• Evidence-Based Medicine

Background:

• Integrating clinical evidence into practice is difficult due to workload and time constraints.
• Current methods for evidence synthesis are time-consuming and resource-intensive.

Purpose of the Study:

• To develop and evaluate Quicker, a Large Language Model (LLM)-powered system for automating clinical evidence synthesis and recommendation generation.
• To assess Quicker's ability to replicate standard guideline development workflows.

Main Methods:

• Developed Quicker, an end-to-end pipeline from clinical questions to recommendations.
• Created Q2CRBench-3, a benchmark dataset from guideline development records for three diseases.
• Conducted system-level testing with participants to evaluate efficiency and accuracy.

Main Results:

• Quicker demonstrated precise question decomposition, expert-aligned evidence retrieval, and comprehensive screening.
• Assisted data extraction improved accuracy; generated recommendations were more comprehensive and coherent than clinician-written ones.
• System-level testing showed Quicker reduced recommendation development time to 20-40 minutes per participant.

Conclusions:

• Quicker significantly enhances the speed and reliability of evidence-based clinical decision-making.
• The LLM-powered system shows potential to streamline guideline development processes.
• Quicker supports customized decision-making through integrated tools and interactive interfaces.