---
id: 20260820-T0-04
title: "GxP-Agent: 用过程DAG拓扑提升临床试验编程成功率"
title_en: "GxP-Agent: Process-DAG Topology Boosts LLM Reliability in Clinical Trials"
url: https://ai.daily.yangsir.net/daily/20260820-T0-04
issue_date: 2026-08-20
publish_date: 2026-08-19T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2608.16890
---

# GxP-Agent: 用过程DAG拓扑提升临床试验编程成功率

一篇来自arXiv的新论文指出，LLM在将临床试验方案转换为符合CDISC标准的分析数据集时表现不佳，在11次单次尝试中均以失败告终。针对此问题，论文提出GxP-Agent，一种基于过程DAG拓扑的解决方案，以提升LLM在临床试验编程任务中的可靠性。

## English Version

**GxP-Agent: Process-DAG Topology Boosts LLM Reliability in Clinical Trials**

A new arXiv paper highlights that LLMs fail catastrophically at converting clinical trial protocols into analysis-ready datasets under CDISC standards, failing all 11 single-shot attempts. To address this, the paper introduces GxP-Agent, a solution based on Process-DAG topology to improve LLM reliability in clinical trial programming.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2608.16890)

**详情页**：https://ai.daily.yangsir.net/daily/20260820-T0-04

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*