Overly restrictive eligibility criteria are a major cause of enrollment failure in clinical trials, yet existing recommendations for loosening them often lack scalability. RECITE (Revising Eligibility Criteria Incorporating Textual Evidence) introduces an automated pipeline for discovering, extracting, and implementing evidence‑based criterion revisions. The system leverages protocol amendment documents from ClinicalTrials.gov to build a curated dataset of 11,913 instances across 5,735 trials, each linking original criteria, documentary evidence, and target revisions. We benchmarked fifteen LLM configurations on a 3,116‑instance sample under both truncated‑context and retrieval‑augmented generation settings, using an LLM‑as‑judge protocol validated against ten domain experts, yielding a judge‑vs‑human quadratic‑weighted Cohen’s $\kappa_{qw}$ of $0.694 \pm 0.096$. The top open‑weight model, Gemma 2 9B, achieved a $\kappa_{qw}$‑aligned ordinal score of $3.0 \pm 0.9$ on a 0–4 scale (normalized $0.742$), while the best closed‑weight model, GPT‑4o, reached $3.0 \pm 0.8$ (normalized $0.745$). Using an agentic literature discovery system, we identified 44 paper‑trial matches across five publications that provide trial‑design and eligibility‑criterion modification evidence. These findings suggest that existing clinical knowledge can be operationalized into actionable trial design recommendations.
Review: RECITE demonstrates a practical approach to harness textual evidence for refining eligibility criteria, offering a scalable tool to improve trial recruitment efficiency.