我国“循证”冠名杂志刊载干预类系统评价/Meta分析的统计学现状调查

The Statistical Status of Systematic Reviews or Meta Analyses on Interventions Published in Chinese Journals of Titled “Evidence-Based”

目的调查我国"循证"冠名杂志发表的干预类系统评价/Meta分析统计学软件和统计学分析方法现状。方法以《中国循证医学杂志》、《循证医学》、《中国循证儿科杂志》、《中国循证心血管医学杂志》4本杂志官方网站为数据来源,根据纳入排除标准,纳入干预类的系统评价/Meta分析/荟萃分析/系统综述,检索时间截止2011年12月31日。2名研究者独立提取资料,并输入Excel软件分析。结果共纳入干预类系统评价/Meta分析487篇,RevMan为最常用的统计学软件(445篇,91.38%),其次为Stata(13篇,2.67%)。92.81%(452篇)系统评价/Meta分析使用1种统计软件,1.64%(8篇)使用2种统计软件,5.54%(27篇)未报告使用的统计软件。统计学分析方法的描述和执行情况存在结果部分和方法学部分的报告率不等,尤其体现在敏感性分析、亚组分析和发表偏倚的评价,且其本身的报告比例并不高。结论调查显示RevMan是目前我国干预类系统评价/Meta分析的首选统计软件;正确地执行敏感性分析、亚组分析及评估发表偏倚可有效地避免各种偏倚。

Objective The aim of this study is to explore the statistical problems current situation of systematic reviews or meta analyses on intervention published in Chinese journals of titled “evidence-based”. Methods The official websites of the Journal of Evidence-Based Medicine , the Chinese Journal of Evidence-Based Medicine , Chinese Journal of Evidence Based Pediatrics and Chinese Journal of Evidence-Based Cardiovascular Medicine being the sources of data , according to the selection criteria and exclusion criteria, the studies that met the condition were selected and then analyzed with the Excellsoftware. The deadline was December 31, 2011. Results A total of 487 studies were fitted into at last. RevMan is the most commonly used statistical software （445 studies, 91.38%）, followed by Stata （13studies, 2.67%）. Software combinations： A kind of software used （452 studies, 92.81%）, the use of two kinds of softwares （8 studies, 1.64%）, and 5.54% （27 studies） did not report any software. The reporting rate of description and implementation for statistical methods was different in result section and methodological section , especially reflected in the sensitivity analysis , subgroup analysis , publication bias evaluation , and the ratio itself was poor. Conclusion The survey showed that the RevMan was the first choice on systematic reviews or meta analyses on intervention. The uses of subgroup analysis, sensitivity analysis, and publication bias were poor. More attention should be paid to these files in future studies.