5 Key Takeaways

• 1

  Pathology reports are complex, free-text documents that vary widely in structure and content, complicating data extraction for clinical use.

• 2

  A three-stage reasoning-based framework using multiple large language models was developed to improve the extraction of structured data from pathology reports.

• 3

  Testing on over 6,000 reports showed the framework effectively generalizes across cancer types, achieving high accuracy in extracting key variables.

• 4

  Real-world reports from Moffitt Cancer Center demonstrated high extraction accuracy, highlighting the importance of evaluating AI systems on clinical data.

• 5

  Biomarker extraction is more challenging due to varied documentation styles and the need to synthesize information from multiple report sections.