中国药物警戒 ›› 2023, Vol. 20 ›› Issue (6): 651-654.
DOI: 10.19803/j.1672-8629.20230109

• 上市后药品不良反应监测定量统计学方法研究专栏 • 上一篇    下一篇

零膨胀模型与狄利克雷过程结合在药品上市后不良反应信号检测中的应用

陈晨鑫1, 张朋朋2, 刘永梅2, 叶小飞1#, 贺佳1,*   

  1. 1海军军医大学卫勤系军队卫生统计学教研室,上海 200433;
    2联勤保障部队天津康复疗养中心,天津 300191
  • 收稿日期:2023-02-27 出版日期:2023-06-15 发布日期:2023-06-15
  • 通讯作者: * 贺佳,女,博士,教授,药物流行病学及新药评价。 E-mail:hejia63@yeah.net。#为并列通信作者。
  • 作者简介:陈晨鑫,男,在读博士,药物流行病学。
  • 基金资助:
    国家自然科学基金资助项目(82073671); 中国毒理学会临床毒理专项(CST2019CT201); 中国毒理学会临床毒理专项(CST2019CT101)

Combination of zero-inflated model and Dirichlet process in detection of adverse reaction signals during post-marketing surveillance

CHEN Chenxin1, ZHANG Pengpeng2, LIU Yongmei2, YE Xiaofei1#, HE Jia1,*   

  1. 1Department of Health Statistics, Faculty of Health Service, Naval Medical University, Shanghai 200433, China;
    2Tianjin Rehabilitation Center of Joint Logistic Support Force, Tianjin 300191, China
  • Received:2023-02-27 Online:2023-06-15 Published:2023-06-15

摘要: 目的 探索零膨胀模型与狄利克雷过程2种方法结合,在优化药品上市后安全性信号检测结果中的可能性。方法 通过文献检索,对零膨胀模型和狄利克雷过程的原理进行综述,分别总结2种方法的研究现状及优点,并提出二者结合以扩展信号检测方法的假设。结果 零膨胀模型能够校正超额“零计数”问题,减少信号检测结果的偏倚;狄利克雷过程通过丰富药品不良反应报告率先验分布的选择空间,能有效控制假阳性信号的产生。结论 将零膨胀模型与狄利克雷过程相结合,用于扩展现有贝叶斯信号检测方法存在一定可能,值得进一步研究。

关键词: 零膨胀模型, 狄利克雷过程, 上市后药品监测, 信号检测, 药品不良反应

Abstract: Objective To explore the possibility of the zero-inflated model being combined with Dirichlet process in optimizing the results of post-marketing safety signal detection of drugs. Methods By retrieving related literature, the principles of the zero-inflated model and Dirichlet process were reviewed. The current research on and strengths of the two methods were summarized. Thus, the combination of these two methods to extend technologies of signal detection was proposed. Results The zero-inflated model could minimize “excess zeros” and reduce the bias of signal detection results, while Dirichlet process could curb the false positive rate by enriching the selection of prior distribution of reporting rates of drug-adverse event pairs. Conclusion It is possible to combining the zero-inflated model with Dirichlet process to extend methods of Bayesian signal detection, and this approach deserves more study.

Key words: zero-inflated model, Dirichlet process, post-marketing surveillance, signal detection, adverse drug reaction

中图分类号: