shared-decision-making

This study compares zero-shot prompting of LLMs against supervised baselines for detecting shared decision-making in pediatric clinical encounters, finding that supervised models outperform zero-shot approaches and highlighting critical data leakage issues in evaluation pipelines.

This paper investigates the use of open-source smaller LLMs (OS-sLLMs) to automate the coding of shared decision-making in clinical consultations using the OPTION12 instrument, aiming for better privacy and sustainability. Preliminary results show that general-domain models like Mistral7b outperform medical-domain models, suggesting further development is needed for medical OS-sLLMs.