Scholarlink

<p>Published clinical case reports are a valuable yet underutilized source of evidence for drug repurposing. However, systematically identifying relevant reports remains a challenge due to the volume of literature and the diversity of candidate compounds. We present TheraMind, an AI system that leverages large language models (LLMs) to automate the identification and analysis of case reports supporting potential drug repurposing for non-small cell lung cancer (NSCLC). Our system screened 10,023 PubMed-indexed case reports across 18 candidate drugs using coordinated data extraction and standardized four-question prompts assessing diagnosis, drug administration, discontinuation, and clinical outcomes. We employed three evaluation strategies, rule-based classifiers, single-model validators, and a majority-vote ensemble integrating GPT-40-mini, Gemini-2.0-Flash, and LLaMA-3-8B. The ensemble approach achieved 92% recall and 99.7% specificity in detecting clinically relevant reports. Structured outputs included patient demographics, therapeutic responses, and case summaries. This LLM-driven framework offers a scalable approach to accelerate drug repurposing by mining real-world evidence from unstructured clinical literature.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

TheraMind: a multi-LLM ensemble for accelerating drug repurposing in lung cancer via case report mining

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月29日

查看全文
  • Vrushket More,
  • Lyra Lu,
  • Zeyu Ding,
  • Zhaohan Xi,
  • Seth Mizia,
  • Nancy L. Guo

摘要

Published clinical case reports are a valuable yet underutilized source of evidence for drug repurposing. However, systematically identifying relevant reports remains a challenge due to the volume of literature and the diversity of candidate compounds. We present TheraMind, an AI system that leverages large language models (LLMs) to automate the identification and analysis of case reports supporting potential drug repurposing for non-small cell lung cancer (NSCLC). Our system screened 10,023 PubMed-indexed case reports across 18 candidate drugs using coordinated data extraction and standardized four-question prompts assessing diagnosis, drug administration, discontinuation, and clinical outcomes. We employed three evaluation strategies, rule-based classifiers, single-model validators, and a majority-vote ensemble integrating GPT-40-mini, Gemini-2.0-Flash, and LLaMA-3-8B. The ensemble approach achieved 92% recall and 99.7% specificity in detecting clinically relevant reports. Structured outputs included patient demographics, therapeutic responses, and case summaries. This LLM-driven framework offers a scalable approach to accelerate drug repurposing by mining real-world evidence from unstructured clinical literature.