基于偏正态分布的随机效应meta回归

The random-effects meta-regression based on the skewed normal distribution

将随机效应meta回归模型进行推广,构造基于偏正态分布的随机效应meta回归模型.在模型中同时考虑偏倚和异质性,以及二者对合并效应估计和回归系数估计的影响.在该模型下,对合并效应量和回归系数采用加权最小二乘估计,并进一步修正为无偏估计.与以往基于正态分布的随机效应meta回归模型相比,该模型将偏倚进行量化,并从合并效应估计和回归系数估计中予以剔除,消除了偏倚对合并效应量和回归系数的影响.提出的模型较基于正态分布的随机效应meta回归模型可以有效提高合并效应量和回归系数的估计精度.

The random-effects meta-regression is extended to the model based on the skewed normal distribution. Bias and heterogeneity are taken into account in the model. At the same time, their influences to the estimators of the overall effect and the regression coefficient also are considered. Under the model based on the skewed normal distribution, the overall effect and the regression coefficient use the weigh- ted least squares estimator which further are corrected to be the unbiased estimators. Compared with the random-effects meta-regression model based on the skewed normal distribution, the bias is quantified and removed from the overall effect estimator and the regression coefficient estimator, so that the bias influences on the combined effect and the regression coefficient are eliminated. The estimation accuracies of the combined effect and the regression coefficient are effectively improved under the new model.