Scholarlink

<p>Large language models (LLMs) have emerged as potential tools for early disease characterization and pandemic preparedness due to their ability to interpret complex textual data. This study evaluated the sensitivity of three LLMs: GPT-5, Claude Sonnet 4, and Gemini 2.5 Pro on early case reports of COVID-19, Mpox, and Zika. Each case report was modified to remove explicit diagnostic terms, and models were prompted to identify whether the presentation represented a disease of pandemic potential. Claude Sonnet 4 achieved the highest sensitivity overall across all three diseases. GPT-5 demonstrated inconsistent results, performing poorly on Mpox. Findings highlight significant variability in diagnostic reliability across LLMs, emphasizing the need for multimodal integration, dataset refinement, and ethical oversight. Limitations include the small sample size, retrospective English-language case reports, text-only inputs, and evaluation of known diseases.</p><p> Clinical Trial Number. Not applicable.</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

Preparing for Pandemics with Large Language Models: An Evaluation of Sensitivity Across COVID-19, Zika, and Monkeypox Case Reports

  • 期刊论文

  • 发表日期：

    2026年3月28日

查看全文
  • Dan Nguyen,
  • Arya S. Rao,
  • Aneesh Mazumder,
  • Bianca Arraiza,
  • Alex Aldrich,
  • William Marks,
  • Marc D. Succi

摘要

Large language models (LLMs) have emerged as potential tools for early disease characterization and pandemic preparedness due to their ability to interpret complex textual data. This study evaluated the sensitivity of three LLMs: GPT-5, Claude Sonnet 4, and Gemini 2.5 Pro on early case reports of COVID-19, Mpox, and Zika. Each case report was modified to remove explicit diagnostic terms, and models were prompted to identify whether the presentation represented a disease of pandemic potential. Claude Sonnet 4 achieved the highest sensitivity overall across all three diseases. GPT-5 demonstrated inconsistent results, performing poorly on Mpox. Findings highlight significant variability in diagnostic reliability across LLMs, emphasizing the need for multimodal integration, dataset refinement, and ethical oversight. Limitations include the small sample size, retrospective English-language case reports, text-only inputs, and evaluation of known diseases.

Clinical Trial Number. Not applicable.