整合分析中的非参数检验:重复取样检验法的实例应用

Non-parametric test for meta-analysis: application of resampling tests

整合分析(meta-analysis)是对同一主题下多个独立实验结果进行综合的统计学方法。非参数检验整合分析——重复取样检验(resamplingtest)不考虑原文献数据的分布形式,故可在不知原文献数据分布形式时使用。其中的靴襻法(bootstrap)可用来给出总效应值的置信区间,但不能检验组内异质性是否显著。靴襻法与随机检验法(randomizationtest)可以有效弥补这一缺失,判断出组间差异性是否显著。实例应用表明,重复取样检验没有参数检验保守,又与参数检验的结果差异较小。

Meta-analysis is a statistical method to summarize independent studies under the same subject. Non-parametric test (resampling test)of meta-analysis do not consider the distribution patterns of data, so that it can be used when primary data distribution patterns are unknown. The confidence interval of overall effect size can be obtained by using bootstrap, but it can not confirm that if the heterogeneity among class is confident. The integration between bootstrap and randomization test can solve this problem. The sample application showed that resampling test is not more conservative than parametric test, and the difference between their outcomes was small. Key works: meta-analysis; nonparametric test; resampling test; bootstrap; randomization test