Related Experiment Video
Updated: Aug 5, 2026
09:16
Multi-Modal Signals for Analyzing Pain Responses to Thermal and Electrical Stimuli
Published on: April 5, 2019
Assessing Pain Catastrophizing Through Free-Text Responses: A Validation of Large Language Models
Angela Lee1, Dokyoung Sophia You2, Troy C Dildine1
1Department of Anesthesiology, Perioperative and Pain Medicine, Stanford University School of Medicine, Stanford University, 1070 Arastradero Road, Suite 200, MC5596, Palo Alto, CA, 94304, USA.
Medrxiv : the Preprint Server for Health Sciences
|August 1, 2026
Summary
Large language models show potential for measuring fluctuating pain catastrophizing in chronic pain patients. Certain models implicitly captured state pain catastrophizing, offering new research avenues.
Area of Science:
- Psychology
- Artificial Intelligence
- Health Informatics
Background:
- Pain catastrophizing is often treated as a stable trait, but it can fluctuate contextually.
- Ecologically valid methods are needed to capture dynamic changes in pain catastrophizing.
- Large language models (LLMs) offer a novel approach for implicit measurement.
Purpose of the Study:
- To evaluate LLMs as implicit markers of state pain catastrophizing.
- To assess the validity of LLM-derived scores against established measures.
- To explore LLM performance across different pain-coping conditions.
Main Methods:
- Ninety-one adults with chronic pain completed trait measures and writing tasks under varied pain-coping conditions.
- Free-text responses were analyzed using four LLMs (Claude Opus 4, GPT Mini 4o, Llama 4 Maverick, Gemini 2.5 Pro).
- LLM-derived scores were compared with trait and state pain catastrophizing scale (PCS) scores, affect, and pain ratings.
Main Results:
- All four LLMs differentiated between negative and positive/neutral pain-coping conditions.
- Gemini-derived scores showed correlations with state catastrophizing and pain unpleasantness.
- LLM scores correlated with negative affect, indicating potential for implicit state measurement.
Conclusions:
- Certain LLMs may serve as implicit markers for state pain catastrophizing.
- LLM-based assessment offers a promising, albeit preliminary, tool for dynamic pain research.
- Further investigation is required to refine LLM applications in pain assessment.
