基于效应量服从偏正态分布的PM法

探讨Meta分析中异质性方差和偏倚的估计以及对合并效应量的影响。在Meta分析随机效应模型中,假设各研究的效应量服从偏正态分布的基础上,基于Q统计量及其一、二阶矩,得到PM估计方程组,采用搜索法求得方程组的解,即为异质性方差和偏倚系数的估计,由此进一步得到合并效应量的估计。在实例计算中,经典的DL法计算的合并OR值为1.370 6,而提出方法计算结果为1.365 3,计算结果十分相近。基于偏正态分布的PM法,不仅估计了异质性方差,而且估计了偏倚参数,即同时考虑了异质性和偏倚对合并效应量的影响,使Meta分析模型和方法更具一般性。

The PM Method in Meta-analysis Based on the Effects Obeying Skew Normal Distribution

Objective To discuss the estimations of heterogeneity variance and bias and the influence they have on the overall effect in meta-analysis. Methods In the random-effect model in meta-analysis, based on the assumption it is assumed that the effects in various researches obey skew normal distribution. The PM estimated equations can be obtained with Q-statistic and its first and second order moments. Then the solutions of the estimated equations, which are the estimations of heterogeneity variance and bias coefficient, are also can be obtained by using iteration method. After that we can get the estimation of overall effect. Results In practical calculations, the combined OR calculated by the classic DL method is 1.3706, while using the proposed method, the calculation result is 1.3653, which is very similar to the result obtained by the classic DL method. Conclusion Based on the skew normal distribution, the PM method estimates both the heterogeneity variance and the bias coefficient. Namely, due to taking into account the influence of heterogeneity and bias of the overall effect, the model and method of meta-analysis becomes more general.