Stata软件在Meta-分析发表性偏倚识别中的探讨  

STUDY ON STATA SOFTWARE IN INVESTIGATING PUBLICATION BIAS IN META-ANALYSIS

作　　者：王丹[1] 牟振云[1] 翟俊霞[2] 宗红侠[3] 赵晓东[2] 

机构地区：[1]河北医科大学公共卫生学院流行病与卫生统计教研室,石家庄050017 [2]河北省医学情报研究所 [3]河北医科大学图书馆

出　　处：《现代预防医学》2008年第15期2819-2822,共4页Modern Preventive Medicine

基　　金：河北省卫生厅指令性计划课题(07025)

摘　　要：[目的]探讨Stata软件在Meta-分析发表性偏倚的识别中的应用。[方法]针对研究实例,采用Stata9软件中漏斗图法、Begg秩相关法、Egger回归法、剪补法的相应命令语句。[结果]4种方法分别绘制出精美的漏斗图、Begg漏斗图、Egger回归图和剪补法漏斗图。Begg法中z=0.81,P=0.417,差异无统计学意义;Egger回归法中t=1.19,P=0.254,差异无统计学意义;剪补法补充4个缺失研究,剪补前后logRR及其95%的可信区间分别为0.022[-0.015,0.058]和0.012[-0.021,0.046]。[结论]采用4种方法对本研究的实例分析均识别出无发表性偏倚。因此,Stata软件具有高效、快速、方便地完成Meta-分析中发表性偏倚识别的优点。[ Objective ] To explore the application of Stata software in investigating publication bias in meta-analysis. [ Methods ] The corresponding commands of four methods in Stata9 software were applied to the analysis of the example in the study. The four methods were funnel plot, Begg＇s adjusted rank correlation test, Egger＇s regression test, trim and fill method. [Results] Fine plots, which were funnel plot, Begg＇s funnel plot, Egger＇s regression plot, trim and fill funnel plot, were produced by the four methods respectively. The results of Begg＇s test （z = 0.81, P = 0.417） and Egger＇s test （t = 1.19, P = 0.254） were not statistically significant. Four missing studies were filled by the trim and fill method. LogRR and its 95%CI were 0.022 [-0.015, 0.058] and 0.012 [-0.021, 0.046] respectively before and alter the application of trim and fill method. [Conclusion ] There is no publication bias in the meta-analysis of the example by the four methods. The merit of Stata software is that it can complete to investigate publication bias in meta-analysis efficiently, quickly and conveniendy.

关 键 词：META-分析 STATA软件 发表性偏倚 

分 类 号：R195.1[医药卫生—卫生统计学]