GxP-Agent: Process-DAG Topology for Reliable Clinical Trial Programming with LLM Agents
GxP-Agent: Process-DAG Topology for Reliable Clinical Trial Programming with LLM Agents
GxP-Agent：用于可靠临床试验编程的流程有向无环图（DAG）拓扑与大模型智能体
Abstract: Clinical trial programming — transforming study protocols into analysis-ready datasets under CDISC standards — is a bottleneck in regulatory submissions, yet LLM-based code generation fails catastrophically on this task: across 11 single-shot attempts with five frontier models, none produces a valid subject-level analysis dataset.
摘要： 临床试验编程（即在 CDISC 标准下将研究方案转化为可分析数据集的过程）是监管申报中的一个瓶颈，然而基于大模型（LLM）的代码生成在此任务上表现极其糟糕：在对五个前沿模型进行的 11 次单次尝试中，没有一个模型能生成有效的主体级分析数据集。
We introduce GxP-Agent, a multi-agent system that encodes regulatory process ordering as a directed acyclic graph (DAG), decomposing monolithic dataset generation into 15 domain-specific nodes executed by worker agents with pharmaverse skill context, validation gates, and conditional retry.
我们引入了 GxP-Agent，这是一个将监管流程顺序编码为有向无环图（DAG）的多智能体系统。它将庞大的数据集生成任务分解为 15 个特定领域的节点，由具备 pharmaverse 技能背景、验证门控和条件重试机制的执行智能体来完成。
On CDISC-Bench, a new execution-based benchmark built from the FDA pilot submission CDISCPilot01 (254 subjects, 49 ground-truth ADSL variables), GxP-Agent with Claude Sonnet 4.6 achieves 100% structural match (49/49 variables, 254 correct records) across three independent runs, compared to 59.2% for the best retrieval-augmented baseline and 0% for all single-agent and flat multi-agent approaches.
在基于 FDA 试点申报 CDISCPilot01（包含 254 名受试者、49 个基准 ADSL 变量）构建的全新执行基准测试 CDISC-Bench 上，搭载 Claude Sonnet 4.6 的 GxP-Agent 在三次独立运行中均实现了 100% 的结构匹配（49/49 个变量，254 条记录完全正确）；相比之下，最佳的检索增强基线模型仅达到 59.2%，而所有单智能体和平铺式多智能体方法均为 0%。
The DAG topology also enables weaker models: GPT-4.1 achieves 59.2% mean structural match under the same DAG, where it scores 0% under every other architecture. The approach generalizes to ADAE (adverse events; 9-node branching DAG, 55 variables, 1,191 records), achieving 100% structural match on the first attempt.
DAG 拓扑结构也使得较弱的模型能够发挥作用：GPT-4.1 在相同的 DAG 下实现了 59.2% 的平均结构匹配率，而在其他任何架构下其得分均为 0%。该方法还推广到了 ADAE（不良事件；9 节点分支 DAG，55 个变量，1,191 条记录）任务中，并在首次尝试时就实现了 100% 的结构匹配。
These results demonstrate that encoding domain process knowledge as graph topology — rather than relying on LLM reasoning alone — is a key enabler for reliable, GxP-compliant clinical trial programming.
这些结果表明，将领域流程知识编码为图拓扑结构——而不是仅仅依赖大模型的推理能力——是实现可靠且符合 GxP 标准的临床试验编程的关键推动因素。