Meta-分析固定效应模型及其扩展应用

Fixed effects models and extensive application in Meta-analysis

Meta-分析是对多个研究效应量的汇总分析,异质性检验的结果如果拒绝无效假设,该采用何种方法处理是进行Meta-分析之前要解决的问题。解决此问题,通常有3种途径:①假定超出对象水平的变异是随机的,将采用随机效应模型。②继续采用固定效应模型,但附加一个假设条件,假定超出对象水平的变异或为零或来源于研究水平间可识别的变异来源,分解效应量变异的固定效应模型。③假定超出对象水平的变异部分来源于研究水平间的系统性因素,部分来源于随机性因素,这种情况需采用混合效应模型。文章针对第2种途径介绍两种处理方法:一种是模拟方差分析模型,另一种是校正的加权多元回归模型,希望引起读者或科研人员对该方法的注意。

Meta-analysis is one way to summarize and integrate many research findings focused on some theme. Heterogeneity test is an essential work before carrying out meta-analysis, if test result is statistically significant. There are three analytic approaches, by which the meta-analyst can take for this situation. First, the excess variability can be assumed to stem from random differences between the studies can be modeled, random effects model can be used. Second, the excess variability can be assumed to be either zero or completely systematic, fixed effects model can continue to be used. Third, it can be assumed that a portion of the excess variability is systematic and can be statistically modeled and that another portion is random and cannot be modeled, thus, mixed effects models can be used. In the article, we will focus on the second approach in order to induce correct application. The main methods of modeling the systematic variance in effect sizes are the analog to the analysis of variance （ANOVA） for categorical variables and Weighted least squares regression for continuous variables.