基于不同结构反应变量比较研究Meta分析的效应量选择

Effect size selection of Meta-analysis based on different structural response variable:A comparative study

背景:成组设计资料的比较研究,比较的是两组或两组以上反应人群某反应变量测量结果的差异,这种设计类型资料的Meta分析理论上比较成熟和完善,但研究者或系统评价者在进行成组设计资料的Meta分析时仍面临不少困境。组间差异比较研究需根据反应变量的结构慎重选择不同的效应量。目的:探讨成组设计资料的Meta分析中针对不同结构反应变量效应量的选择情况及其注意事项。方法:应用计算机检索CNKI全文数据库、维普全文数据库、万方中国博硕数据库1990/2009及Pubmed数据库1979/2009有关组间比较研究结局变量为连续性、二分类以及两者混合型研究的Meta分析或系统评价方法学文献,排除内容陈旧、重复的文献。结果与结论:纳入30篇符合标准的文献进行综述。结果显示结局变量以连续性或二分化两种形式呈现在研究文献中是比较普遍的现象,可能更适用,但转换为百分数分析后存在的问题是切点的确定可能过于武断,而且连续性变量二分化处理后失去了部分信息。针对这种情况的资料进行Meta分析仍面临着巨大的方法学挑战。文章在介绍几种常用的处理方法的基础上提出了一种处理方法,即基于等级贝叶斯模型的贝叶斯重构法,将克服切点值确定的武断与信息缺失的缺陷。

BACKGROUND：The comparative study of groups design data is to compare the difference of response variable measurements on two or more groups of respondents.The Meta-analysis on the studies of this kind of design information is theoretically comparative maturity and consummate,but the researchers and system evaluators still face many difficulties during the Meta-analysis of groups design data.The Meta-analysis on group comparison study requires selecting carefully different effect sizes based on the structure of response variable.OBJECTIVE：To explore the effect size selection and announcements of response variable of different structure in Meta-analysis of groups design data.METHODS：Articles related to Meta-analysis or systematic review methodology literature about group comparison studies for the continuous,dichotomous,and combined outcomes in CNKI database,VIP database,Wanfang Chinese Doctoral database （1990/2009）,and Pubmed database （1979/2009） were retrieved by computer.Outdated and repetitive researches were excluded.RESULTS AND CONCLUSION：Totally 30 literatures were involved for summarization according to inclusion criteria.It is relatively common that outcome variables were presented in continuous or dichotomous forms in the research literature,in some situations,it may be more applicable,but when continuous outcome variables were converted to percentage,the determination of cut points may be too arbitrary after converted to percentage analysis,and some information were lost after the treatment of continuous variable dichotomy.It is still facing a huge methodological challenge for meta-analysis.In this review,we introduced a approach based on several treatments commonly used,namely,bayesian reconstruction method,which is based on bayesian model class （Hierarchical Bayesian Models） will overcome the shortcomings of arbitrary determination of cut points and information default.