316P Interpreting small generative LLMs for oncology trial matching using gradient-based attribution under low-resource settings

Abstract

Recent research has employed various large language models (LLMs) to match patients with clinical trials. While these approaches have demonstrated promising results, some important challenges remain. In particular, understanding why the language model makes a correct or incorrect decision, and identifying which parts of the clinical narrative influence its predictions, remain unclear. This interpretability is crucial for real-world clinical applications, where AI-assisted clinicians must evaluate patient eligibility for clinical trials.

Ähnliche Arbeiten