欢迎访问《中国循证儿科杂志》官方网站，今天是
English Version
中国循证儿科杂志

中国循证儿科杂志 ›› 2019, Vol. 14 ›› Issue (3): 212-216.DOI: 10.3969/j.issn.1673-5501.2019.03.010

• 循证医学方法学 • 上一篇    下一篇

单臂试验连续型数据的贝叶斯Meta分析方法及实现

张天嵩   

  1. 复旦大学附属静安区中心医院 上海,200040
  • 收稿日期:2019-04-10 出版日期:2019-06-25
  • 通讯作者: 张天嵩,E-mail:zhangtiansong@fudan.edu.cn

Bayesian methods for meta-analysis of continuous data in single-arm trials and its application

ZHANG Tian-song   

  1. Jing'an District Central Hospital,Fudan University, Shanghai 20004
  • Received:2019-04-10 Online:2019-06-25
  • Contact: ZHANG Tian-song,E-mail: zhangtiansong@fudan.edu.cn

PDF

1041

摘要： 目的 介绍单臂试验连续型数据的Meta分析模型、贝叶斯方法及实现。方法 阐述正态-正态层次模型,基于该模型框架,以贝叶斯方法拟合随机效应模型,对效应参数μ和异质性参数τ分别选择不同的先验,使用R软件的bayesmeta包对两个文献数据重新分析。结果 在正态-正态层次模型框架下,基于不同的先验信息,贝叶斯Meta分析结果为:数据1参数μ的点估计及95%CI分别为-4.26(-6.97, -1.92)和-4.50(-9.27, -0.53),参数τ点估计及95%CI分别为1.51(0.41, 2.75)和2.28(0.00, 6.57);数据2参数μ的点估计及95%CI分别为-4.07(-5.54, -2.71)和-4.12(-5.96,-2.46),参数τ点估计及95%CI分别为1.54(0.78, 2.48)和1.81(0.74, 3.51)。结论 不同的先验可能影响参数估计值。基于NNHM框架下的贝叶斯方法适用于单臂试验连续型数据的Meta分析。Bayesmeta包以其简单、快速、准确、可重量性算法等可以用于实现贝叶斯随机效应模型Meta分析。

关键词: bayesmeta包, Meta分析, 贝叶斯方法, 单臂试验, 连续型数据, 正态-正态层次模型

Abstract: Objective To introduce a model of meta-analysis, and to explain how to fit the model with bayesian Methods.Methods The normal-normal hierarchical model(NNHM)was explained. Random effects model was used to reanalyze two worked examples from literature in the framework of the NNHM. Bayesian Methods were used to fit the NNHM with the bayesmeta R package. The priors for the effect parameter μ and heterogeneity parameter τ were given different choice respectively.Results In the framework of NNHM, based on the different priors, the Results of data 1 were shown that point estimates and 95% credit interval(CI) of parameter μ using the bayesian method were -4.26(-6.97, -1.92) and -4.50(-9.27, -0.53) respectively and point estimates and 95%CI of parameter τ were 1.51(0.41, 2.75) and 2.28(0.00, 6.57) respectively. And the Results of data 2 were shown that point estimates and 95%CI of parameter μ were -4.07(-5.54, -2.71) and -4.12(-5.96, -2.46) respectively and point estimates and 95% CI of parameter τ were 1.54(0.78, 2.48) and 1.81(0.74, 3.51) respectively.Conclusion Different priors might have an effect on the estimation of parameters. Bayesian Methods were preferable for the meta-analyses of continuous data in single-arm trials within the framework of NNHM, and the bayesmeta package was allowed to perform bayesian random-effects meta-analyses because of its easy, quick, accurate and reproducible computation.

Key words: Bayesian methods, Bayesmeta package, Continuous data, Meta-analysis, Normal-normal hierarchical model, Single-arm trials