中国药物警戒 ›› 2024, Vol. 21 ›› Issue (2): 163-166.
DOI: 10.19803/j.1672-8629.20230632

• 法规与管理研究 • 上一篇    下一篇

基于知识图谱联合ERNIE-DPCNN模型的药品不良反应自动关联性评价方法研究

贾晋生1, 刘红亮1, 王青2, 侯永芳1, 李馨龄1,*   

  1. 1国家药品监督管理局药品评价中心,国家药品监督管理局药物警戒研究与评价重点实验室,北京 100076;
    2清华大学药物警戒信息技术与数据科学创新中心,北京 100084
  • 收稿日期:2023-10-13 出版日期:2024-02-15 发布日期:2024-02-06
  • 通讯作者: *李馨龄,男,主任药师,药品不良反应监测。E-mail：lixinling@cdr-adr.org.cn
  • 作者简介:贾晋生,男,本科,工程师,药品不良反应监测。
  • 基金资助:
    国家重点研发计划（2020YFC2005501）

Evaluation of the relevance of adverse drug reaction based on ERNIE-DPCNN

JIA Jinsheng1, LIU Hongliang1, WANG Qing2, HOU Yongfang1, LI Xinling1,*   

  1. 1Center for Drug Reevaluation, NMPA/NMPA, Key Laboratory for Research and Evaluation of Pharmacovigilance, Beijing 100076, China;
    2Research Center for Pharmacovigilance IT and Data Science, Tsinghua University, Tsinghua University, Beijing 100084, China
  • Received:2023-10-13 Online:2024-02-15 Published:2024-02-06

PDF (PC)

123

摘要： 目的 针对当前药品不良反应关联性评价存在的效率较低和主观性评估问题,通过建立药品不良反应关联性评价模型,探索药品不良反应自动关联性评价方法。方法 利用文献及互联网来源,对获取的不良反应报告标注数据（7 301条）进行知识抽取,构建药品不良反应知识图谱,建立知识驱动的ERNIE-DPCNN自动关联性评价模型。结果 提出的知识图谱联合ERNIE-DPCNN模型在测试集中的精确度、召回率和F1值分别达到82.18%、81.40%、81.21%,相对于其他基线模型各项评估指标均取得了最高值。结论 知识图谱联合ERNIE-DPCNN模型的方法能提高药品不良反应关联性评价效率,具备较强的准确性,并在一定程度上减少主观性评估误差,对基于人工智能的自动化评价有一定参考价值。

关键词: 药品不良反应, ERNIE-DPCNN模型, 知识图谱, 关联性评价, 文本分类, 深度学习, 人工智能

Abstract: Objective To establish a model for evaluation of ADR correlations in order to make the related evaluation more efficient and objective. Methods A total of 7 301 pieces of annotated data on adverse reaction reports obtained from literature and Internet sources were used for knowledge extraction to construct a knowledge graph about adverse drug reactions before a knowledge-driven automatic evaluation model of ERNIE-DPCNN relevance was established. Results The precision, recall and F1 value of the knowledge graph combined with the ERNIE-DPCNN model in the test set reached 82.18%, 81.40% and 81.21%, respectively, which yielded higher values than other models. Conclusion The method based on knowledge graphs combined with the ERNIE-DPCNN model can effectively improve the efficiency and accuracy of correlation evaluation of adverse drug reactions and reduce errors that arise from subjective evaluation, which is of referential value for automated evaluation based on artificial intelligence.

Key words: adverse drug reaction (ADR), ERNIE-DPCNN model, knowledge graph, correlations of evaluation, natural language processing, deep learning, artificial intelligence

中图分类号: