A Scenario-Driven Approach to the Construction and Application of High-Quality Datasets for Drug Regulation

Expand
  • Center for Information of NMPA Beijing 100163, China

Received date: 2026-04-09

  Revised date: 2026-06-23

  Accepted date: 2026-06-26

  Online published: 2026-06-29

Abstract

Objective: To address the insufficient supply of high-quality datasets in advancing AI + drug regulation and to explore dataset construction methods applicable to the field of drug regulation.Methods: Starting from the 15 scenarios outlined in the List of Typical Application Scenarios for Artificial Intelligence in Drug Regulation a scenario-driven dataset construction method was proposed forming a four-stage framework requirement definition data collection quality control and iterative optimization. Key techniques including scenario requirement decomposition multi-source heterogeneous data fusion expert-AI collaborative annotation and dynamic iterative optimization were developed. Empirical validation was conducted in the scenario of Class II medical device registration review.Results: A three-tier dataset system consisting of pre-training fine-tuning and evaluation datasets was established. In the Class II device review empirical study instruction-tuning data a RAG rule base and a test case set were successfully constructed. The model-assisted review achieved an accuracy of 91.4% and a recall rate of 93.6% which meet the expected effect.Conclusion: The scenario-driven approach effectively addresses the unique challenges of building high-quality datasets in drug regulation improves dataset construction efficiency and model application performance and provides methodological guidance for the development of AI data infrastructure in drug regulation.


Cite this article

YOU Yuwei, CHEN Feng .

A Scenario-Driven Approach to the Construction and Application of High-Quality Datasets for Drug Regulation

[J]. CHINESE JOURNAL OF MEDICINAL GUIDE, 2026 , 28(5) : 509 -509-515 . DOI: 10.1009-0959.2026.050028

References

        [1 国务院.国务院关于深入实施“人工智能+”行动的意见(国发〔202511号)[EB/OL.2025-08-26)[2026-03-13.https//www.gov.cn/zhengce/content/202508/content_7037861.htm.

         2  国务院.政府工作报告——202635日第十四届全国人民代表大会第四次会议上[EB/OL.2026-03-05)[2026-03-13.https//www.gov.cn/yaowen/liebiao/202603/content_7062625.htm.

         3  新华社.中华人民共和国国民经济和社会发展第十五个五年规划纲要[EB/OL.2026-03-13)[2026-03-13.https//www.gov.cn/yaowen/liebiao/202603/content_7062633.htm.

         4  国家药品监督管理局.国家药监局关于“人工智能+药品监管”的实施意见(国药监综〔20266号)[EB/OL.2026-03-11)[2026-03-13.https//www.gov.cn/zhengce/zhengceku/202604/content_7064591.htm.

         5  陈锋.实施“人工智能+药品监管”行动的思考[J.中国药事,2026401):3-10.

         6  中国信息通信研究院,国家数据发展研究院,中国电子技术标准化研究院,等.高质量数据集建设指引[EB/OL.2025-08-28)[2026-03-13.https//www.nda.gov.cn/sjj/ywpd/szkjyjcss/0830/20250830210000366789341_pc.html.

         7  国家数据局. 刘烈宏出席“2025科创大会”并致辞[EB/OL.2025-12-03)[2026-03-13.https//www.nda.gov.cn/sjj/jgsz/jld/llh/llhldhd/1203/20251203123534300772332_pc.html.

         8  国家发展和改革委员会.激活数据要素市场化配置改革动能,筑牢“人工智能+”行动发展基石 [EB/OL.2025-08-28)[2026-03-13.https//www.ndrc.gov.cn/xxgk/jd/jd/202508/t20250828_1400104.html.

         9  国家药品监督管理局.国家药监局综合司关于印发药品监管人工智能典型应用场景清单的通知[EB/OL.2024-06-18)[2025-09-26.https//www.nmpa.gov.cn/xxgk/fgwj/gzwj/gzwjzh/20240618144318144.html.

         10 陈锋,吴欣然.药品监管领域大语言模型(LLM)一体化建设策略研究[J.中国医药导刊,2025271):1-6.

         11 陈锋,吴欣然.大语言模型在药品监管中的应用实践与思考[J.中国食品药品监管,2025,(3):4-13.

         12 中央网络安全和信息化委员会办公室,国家发展和改革委员会.政务领域人工智能大模型部署应用指引[EB/OL.2025-10-11)[2026-03-13.https//www.gov.cn/lianbo/bumen/202510/content_7043861.htm.

         13 央视新闻.国家数据局:分三类建设高质量数据集赋能AI发展[EB/OL.2025-04-29)[2026-03-13.https//www.nda.gov.cn/sjj/ywpd/szkjyjcss/0429/20250429221703099105731_pc.html.

         14 光明网.刘烈宏:释放数据要素价值 推动数据要素与人工智能深度融合[EB/OL.2025-12-04)[2026-03-13.https//economy.gmw.cn/2025-12/04/content_38458370.htm.

         15 陈锋,由玉伟.药品监管人工智能伦理治理初探[J.中国医药导刊,20252711):1121-1126.

         16 国家药品监督管理局,国家卫生健康委员会.药物临床试验质量管理规范[EB/OL.2020-04-23)[2026-03-13.https//www.gov.cn/gongbao/content/2020/content_5525106.htm.

         17 世界医学会.赫尔辛基宣言[EB/OL.2024-10-19)[2026-03-13.https//www.wma.net/wp-content/uploads/2024/12/DoH-2024-in-Simplified-Chinese.pdf.

         18 吴欣然,刘洋.大语言模型在药品监管典型场景应用分析[J.中国食品药品监管,2025,(9):30-39.

         19 唐菀晨,孙浩,由玉伟,等.基于两级数据中心的药品监管数据治理实践与思考[J.中国医药工业杂志,2024557):1037-1043.

         20 中国信息通信研究院人工智能研究所,清华大学计算社会科学与国家治理实验室,中国人工智能产业发展联盟数据委员会.人工智能高质量数据集建设指南[EB/OL.2025-08-18)[2026-03-13.https//www.caict.ac.cn/kxyj/qwfb/ztbg/202508/P020250829553828220259.pdf.

         21 国务院.医疗器械监督管理条例[EB/OL.2024-12-06)[2026-03-13.https//www.nmpa.gov.cn/xxgk/fgwj/flxzhfg/20250416172904188.html.

         22 国家市场监督管理总局.医疗器械注册与备案管理办法(国家市场监督管理总局令第47号)[EB/OL.2021-08-26)[2026-03-13.https//www.samr.gov.cn/zw/zfxxgk/fdzdgknr/fgs/art/2023/art_568880e3ee344c45b38d073bba1c53ad.html.

         23 国家药品监督管理局医疗器械技术审评中心. 关于发布医疗器械注册电子申报目录文件夹结构的通告(2021年第15号)[EB/OL.2021-11-15)[2026-03-13.https//www.cmde.org.cn//xwdt/shpgzgg/gztg/20211115161000245.html.

         24 国家互联网信息办公室,国家发展和改革委员会,教育部,等.生成式人工智能服务管理暂行办法[EB/OL.2023-07-10)[2026-03-13.https//www.gov.cn/gongbao/2023/issue_10666/202308/content_6900864.html.

Outlines

/