Related Experiment Video
Updated: Jun 13, 2026
03:14
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
SCRIPT: Stratified clinical risk prediction from pathology reports using large language models.
Chiara M L Loeffler1,2,3, Nic G Reitsam1,4,5, Fabian Wolf1
1Else Kroener Fresenius Center for Digital Health, Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, 01307 Dresden, Germany.
Journal of Pathology Informatics
|June 12, 2026
Summary
Large language models (LLMs) can extract prognostic information from free-text pathology reports, creating a novel survival biomarker. This approach enhances cancer risk stratification without increasing pathologist workload.
Area of Science:
- Oncology
- Computational Pathology
- Artificial Intelligence in Medicine
Background:
- Accurate cancer risk stratification is crucial for treatment decisions.
- Current methods often overlook valuable prognostic information within narrative pathology reports.
- Free-text pathology reports contain nuanced descriptions and expert judgment that are underutilized.
Purpose of the Study:
- To investigate the potential of large language models (LLMs) to extract prognostic information from free-text pathology reports.
- To develop a method for converting narrative pathology report data into a binary survival biomarker.
- To assess the utility of LLM-generated risk scores in predicting patient survival outcomes.
Main Methods:
- Utilized the LLaMA 3.3 70B large language model to process complete free-text pathology reports.
- Prompted the LLM to synthesize narrative reports into a binary prognostic risk score.
- Evaluated the association between LLM-generated risk scores and survival outcomes (overall survival, progression-free survival, disease-specific survival) in gastrointestinal cancers.
Main Results:
- LLM-generated risk scores showed significant prognostic value for overall survival, progression-free survival, and disease-specific survival in colorectal cancer.
- Hazard ratios indicated strong associations between LLM scores and adverse survival outcomes (e.g., HR=2.77 for OS, HR=2.93 for PFS, HR=5.85 for DSS).
- Multivariate analysis confirmed the LLM-generated risk score as an independent prognostic factor for progression-free survival.
Conclusions:
- Large language models can effectively transform narrative pathology reports into a singular, independent survival biomarker.
- This method leverages existing free-text data, requiring no additional tissue analysis or pathologist effort.
- The LLM-based approach offers a deployable solution to enhance cancer risk stratification and inform treatment decisions.