Performance evaluation of five large language models for assisting in the interpretation of urinalysis reports for kidney diseases: a real-world study

Key Points

• This study aims to evaluate the effectiveness of large language models in interpreting urinalysis reports for diagnosing kidney diseases.
• Evaluated five large language models in urinalysis interpretation using real-world data.
• Focused on instrument-specific flag interpretation and hallucination mitigation.
• Explored Retrieval-Augmented Generation integration and human oversight for improved accuracy.
• Proprietary models showed superior reasoning capabilities compared to others.
• Significant challenges in flag interpretation and hallucination resistance were identified, indicating areas for improvement.
• Human oversight is recommended to enhance model performance during interpretation.