单臂试验连续型数据的贝叶斯Meta分析方法及实现

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

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。

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.