The following abstract is drawn from a recently published paper in the Journal of Pathology Informatics. We invite you to read the full paper and join the conversation, become a member of the Pathology News community to share your thoughts, ask questions, and engage with others around this work.

Authors: Chiara M.L. Loeffler^a,b,c, Nic G. Reitsam^a,d,e, Fabian Wolf^a, Esther H. Stueker^a, Hannah S. Muti^a,c,f, Isabella C. Wiest^a,i,j, Jakob Nikolas Kather^a,b,g,h,j

Abstract

Background
Accurate risk stratification in oncology is essential for guiding treatment decisions, yet current algorithms rely on a narrow set of structured variables, and hence potentially ignore the rich signal in narrative pathology reports. These reports contain nuanced morphological descriptions and expert clinical judgment, yet this narrative information remains largely unused in clinical decision-making as it gets lost in “prose” text-based reports. We hypothesized that large language models (LLMs) could extract prognostic information from complete free-text pathology reports and convert it into a binary survival biomarker.

Methods
We used the open-weight LLaMA 3.3 70B model to generate risk scores directly from publicly available pathology reports across three gastrointestinal cancer types. The model was prompted to synthesize the complete narrative reports into a binary prognostic score. We evaluated associations between the LLM-generated scores and survival outcomes, including overall survival, progression-free survival, and disease-specific survival.

Results
In colorectal cancer, LLM-generated risk scores demonstrated significant prognostic value for overall survival (Hazard ratio (HR) = 2.77, 95% confidence interval (CI) = 1.92–3.97, p < 0.001), progression-free survival (HR = 2.93, 95% CI = 2.11–4.08, p < 0.001), and disease-specific survival (HR = 5.85, 95% CI = 3.66–9.36, p < 0.001). Multivariate analysis confirmed the LLM-generated risk score as an independent prognostic factor for progression-free survival.

Conclusion
LLMs can turn narrative pathology reports into a single, independent survival biomarker. This approach leverages routinely available free-text documentation without requiring additional tissue analysis or pathologist workload, providing a deployable method to enhance risk stratification for treatment decision-making.

Read the full article:SCRIPT: Stratified clinical risk prediction from pathology reports using large language models – ScienceDirect

^aElse Kroener Fresenius Center for Digital Health, Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, 01307 Dresden, Germany

^bDepartment of Medicine I, Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, 01307 Dresden, Germany

^cNational Center for Tumor Diseases Dresden (NCT/UCC), a partnership between DKFZ, Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, and Helmholtz-Zentrum Dresden – Rossendorf (HZDR), Dresden, Germany

^dPathology, Faculty of Medicine, University of Augsburg, Augsburg, Germany

^eBavarian Center for Cancer Research, BZKF, Augsburg, Germany

^fDepartment for Visceral, Thoracic and Vascular Surgery, University Hospital and Faculty of Medicine Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany

^gMedical Oncology, National Center for Tumor Diseases (NCT), University Hospital Heidelberg, Heidelberg, Germany

^hPathology & Data Analytics, Leeds Institute of Medical Research at St James’s, University of Leeds, Leeds, United Kingdom

ⁱDepartment of Medicine II, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany