Retrieval-augmented clinical decision support for structured hip-joint disease assessment

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Orthopedic Diagnostics

Background:

• Hip-joint disease assessment is complex, requiring integration of clinical symptoms, imaging, staging, and management.
• Clinician-facing decision-support systems can aid structured reasoning but need rigorous evaluation.

Purpose of the Study:

• To validate a novel hip-joint decision-support system (Hip-Agent) using a retrospective case-based approach.
• To assess the system's diagnostic accuracy, physician feedback, and performance across varying case complexities.

Main Methods:

• Retrospective validation of Hip-Agent involving 74 hip-joint cases across five disease categories.
• Cases were stratified as easy, moderate, or complex and evaluated by 15 physicians.
• Endpoints included diagnostic accuracy, physician ratings of clinical domains, confidence, decision time, and acceptability.

Main Results:

• Hip-Agent achieved 85.1% overall diagnostic accuracy.
• Accuracy was high for easy (94.4%) and moderate (92.0%) cases but significantly lower for complex cases (46.2%).
• Physicians rated the output format highly, with good to excellent inter-observer reliability for clinical domain ratings.

Conclusions:

• The Hip-Agent system shows feasible retrospective performance for hip-joint case evaluation, though accuracy varies with complexity.
• Low accuracy in complex cases necessitates caution and further investigation.
• Prospective studies are required to evaluate the system's impact on clinical practice and human-AI interaction.