基于MedDRA系统的药物安全性多重比较贝叶斯层次模型构建及应用  

Construction and Application of Bayesian Hierarchical Model Based on Multiple Comparisons of Drug Safety Data in MedDRA System

作　　者：陈振明 李太顺[1] 杨嘉莹[1] 王诗远[1] 刘沛[1] Chen Zhenming;Li Taishun;Yang Jiaying(Department of Epidemiology and Health Statistics,School of Public Health,Southeast University(210009),Nanjing)

机构地区：[1]东南大学公共卫生学院流行病与卫生统计学系,210009

出　　处：《中国卫生统计》2020年第3期367-371,共5页Chinese Journal of Health Statistics

基　　金：中央高校基本科研业务费专项资金和江苏省研究生科研与实践创新计划项目(KYCX17_0186)。

摘　　要：目的在处理药物不良事件的安全性分析时会面临多重性问题,基于不良事件类型间的相似性,应用贝叶斯层次模型进行多重比较。方法利用MedDRA词典的层次结构构建贝叶斯层次模型,比较不同层次结构对模型的影响及其收缩作用。分析超过数概率进行统计推断的优势,以便标记出潜在的不良事件信号。结果贝叶斯层次模型使得不良事件间的数据可以借用同一层次内的信息,达到收缩数据的作用,收缩的程度与层次结构有关。使用后验超过数概率进行分析使得结果更具临床意义。结论本研究将贝叶斯层次模型引入不良事件的安全性分析中,并以实例说明其统计特性,为解决多重性问题提供了新思路。Objective The safety analysis of adverse events data will carry multiplicity problems.Based on the similarity of adverse event types,using Bayesian hierarchical model to adjust the results.Methods The hierarchical structure of MedDRA dictionary is used to construct the Bayesian hierarchical model.The effects of different hierarchical structures on the model and their shrinkage are compared.The advantages of statistical inference based on the posterior exceedance probability to detect the potential safety signals of adverse events are discussed.Results The Bayesian hierarchical model enables the adverse events data to borrow information from the same level to shrink the data.The degree of shrinkage is related to the hierarchical structure.The results are more clinically meaningful by using the exceedance probability.Conclusion In this study,Bayesian hierarchical model is introduced into the safety analysis of adverse events,and its statistical characteristics are illustrated with examples,which provides a new way to solve the multiplicity problems.

关 键 词：贝叶斯层次模型 多重性 药物安全性 

分 类 号：R195.1[医药卫生—卫生统计学]