Meta分析中连续性变量二分化后的效应量估计

Estimation of effect-size after dichotomized continuous variables in Meta-analysis

背景:无论在医学研究领域以及行为科学或心理学研究领域,即使是设计类型为两组样本的比较研究,当对同一研究结局变量有的采用二分类变量,有的采用连续性变量,特别是研究结局实质上为连续性变量但在分析时采用二分类化处理这种情况时,在进行Meta分析时,也面临着巨大挑战。目的:基于2×2列联表的形式探讨采用7种不同的效应量指标对结局变量实质上为连续性变量却被二分类化的总体标准化均差效应量(δ)的估计效能。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(检索年限:1990/2009)及Pubmed数据库(检索年限:1979/2009)有关组间比较研究结局变量为连续性、二分类以及连续性与二分类混合型研究的Meta分析或系统评价方法学文献,中文检索词为"组间比较研究,连续性结局,二分类结局,合并的连续性与二分类结局,Meta分析,效应量";英文检索词为"group comparison study,continuous outcomes,dichotomous outcomes,combining continuous and dichotomous outcomes,meta-analysis,effect size"。排除非比较性研究以及相关的Meta分析与系统评价文献。结果与结论:符合标准的23篇文献主要探讨了两组比较研究中结局变量为连续性测量变量的效应量选择与连续性变量二分化处理7种效应量选择问题。在社会学与行为学研究领域,实质上为连续性结局变量为了解释与使用方便而在分析时采用了二分类化处理的情况比较普遍,但给系统评价与Meta分析带来了挑战,针对实质上为连续性变量二分类化处理的情况,文章比较分析了7种常用的效应量,结果表明两种效应量指标是一种不错的选择,一种是基于probit转换而另一种是基于logistic分布。

BACKGROUND：Both in the field of medical research and behavioral science or psychology in research,even if the type of design is a comparative study of the two samples;for the same outcome variable,there are dichotomized variables and continuous variables,especially for the continuous variables in the dichotomized treatment analysis.Therefore,it also faces enormous challenges in the meta-analysis.OBJECTIVE：To explore 7 different effect-size indices for estimating the population standardized mean difference （δ） based on a 2×2 contingency table in terms of bias and sampling variance.METHODS：The literatures related to the outcome variable of intragroup comparative study for the continuous,dichotomous,and continuous combined with dichotomous meta-analysis or systematic review methodology in CNKI database,VIP full text database,Wanfang Chinese Doctoral database （1990/2009）,Pubmed database （1979/2009） were retrieved,with key words of ＂group comparison study,continuous outcomes,dichotomous outcomes,combining continuous and dichotomous outcomes,meta-analysis,effect size＂ in Chinese,and the same key words in English.Non-comparative studies and relative meta-analysis and systematic evaluation literatures were excluded.RESULTS AND CONCLUSION：A total of 23 documents were included,and which focus on the problems of outcome variables are the continuous variable＇s effect-size selection and 7 effect-size selections of dichotomized variables in comparative studies of two groups.In the field of sociology and ethology,essentially as a continuous outcome variables in order to explain and easy to use in the analysis,the dichotomized treatment of continuous variables was widely applied,but it brought challenge for systemic evaluation and meta-analysis.For the dichotomized treatment of continuous variables,7 kinds of commonly used effect-size was performed comparative analysis.The results showed good performance for 2 indices,one based on the probit transformation and the other based on the logistic distribution.