-
摘要: 人群研究, 尤其是临床研究需要重复, 但重复研究往往会产生不一致的研究结果, 采用系统综述及时综合分析同一问题相关研究便成为循证决策的基础, 系统综述中可以对原始研究数据实现统计合并的Meta-分析已在越来越多领域得到广泛应用。但Meta-分析本身存在局限性, 是否适用于所有类型研究的统计合并, 尚存争议。有鉴于此, 本文在介绍系统综述步骤和Meta-分析基本原理及统计合并计算步骤的基础上, 进一步分析Meta-分析的误差来源, 解释随机效应模型与固定效应模型的区别, 旨在引导人们客观理解Meta-分析的方法学局限性, 审慎使用Meta-分析, 在进行临床决策时审慎解读Meta-分析结果, 尤其是基于随机效应模型的统计合并。Abstract: Researches, especially population studies and clinical studies, need to be repeated. However, repeated studies on the same question may produce conflict results. Therefore, a systematic review, a timely comprehensive analysis of individual studies on the same question, is the basis of evidence-based decision making. Meta-analysis of systematic review, which is regarded as the method for producing the evidence of top level, has been widely used in more and more fields. However, there are some methodological limitations for meta-analysis. It is controversial whether meta-analysis could be applicable to all type of researches. In view of this fact, we introduced the procedure of systematic review, the basic principle and the statistical formula of meta-analysis, and further explored the sources of error in meta-analysis and explained the difference between the random effect model and the fixed effect model in order to understand the methodological limitations of meta-analysis and perform it deliberately. At the same time, we recommended that clinicians should carefully interpret the pooled results of a meta-analysis, especially of the one using the random effect model, when they are going to make a clinical decision.
-
Key words:
- systematic reviews /
- Meta-analysis /
- error /
- random effect model /
- fixed effect model
-
表 1 原始研究的结果整理
原始研究i 结局事件 无结局事件 合计 试验组 ai bi n1i 对照组 ci di n2i 合计 Ni 注:此表以结局变量为二分类变量为例 表 2 Meta-分析中需要处理的误差
误差 发生原因 原始研究水平 随机误差 1.设计阶段,如单纯随机抽样误差要大于分层随机抽样误差 2.样本量不足 3.测量精确度不足 系统误差 1.设计阶段,如采用非随机抽样获得研究对象 2.测量准确度不好 3.存在混杂偏倚 Meta-分析水平 系统误差 1.发表偏倚 2.实施过程中无充分的质量控制 -
[1] Smith ML, Glass GV. Meta-analysis of psychotherapy outcome studies[J]. Am Psychol, 1977, 32:752-760. http://cn.bing.com/academic/profile?id=f086d9b4c6f6c2f03b0f6847e87668f7&encoded=0&v=paper_preview&mkt=zh-cn [2] Spitzer WO. The challenge of Meta-analysis[J]. J Clin Epidemiol, 1995, 48:1-4. doi: 10.1007/1-4020-2648-X_12 [3] Onitilo AA. Is it time for the Cochrane Collaboration to reconsider its Meta-analysis methodology?[J]. Clin Med Res, 2014, 12:2-3. http://cn.bing.com/academic/profile?id=f8d35124e513988ed64dbc8506122cb9&encoded=0&v=paper_preview&mkt=zh-cn [4] Thompson S, Ekelund U, Jebb S, et al. A proposed method of bias adjustment for Meta-analyses of published obser-vational studies[J]. Int J Epidemiol, 2011, 40:765-777. https://www.ncbi.nlm.nih.gov/pubmed/21186183 [5] Shuster JJ. Empirical vs. natural weighting in random effects Meta-analysis[J]. Stat Med, 2010, 29:1259-1265. http://cn.bing.com/academic/profile?id=caa9bb3983cfbd644435160ac2385e07&encoded=0&v=paper_preview&mkt=zh-cn [6] Deeks J, Higgins J. On behalf of the statistical methods group. Statistical Algorithms in Review Manager 5. 2010 [7] Borenstein M, Hedges L, Rothstein H. Meta-analysis: fixed effect vs. random effects (2007)[EB/OL]. http://www.Meta-analysis.com at 16/06/2008. [8] DoiSA, Thalib L. A quality-effects model for Meta-analysis[J]. Epidemiology, 2008, 19:94-100. http://cn.bing.com/academic/profile?id=7e1213e61e1233cb26120a324e776c9e&encoded=0&v=paper_preview&mkt=zh-cn [9] Wang YX, Peng X, Nie XL, et al. Burden of hypertension in China over the past decades:Systematic analysis of prevalence, treatment and control of hypertension[J]. Eur J Prev Cardiol, 2016, 23:792-800. http://cn.bing.com/academic/profile?id=51a5a53f07c1218b7e5fb787f06d759c&encoded=0&v=paper_preview&mkt=zh-cn [10] Liu TY, Nie XL, Wu ZH, et al. Can statistic adjustment of OR minimize the potential confounding bias for Meta-analysis of case-control study: a secondary data analysis[J]. BMC Med Res Methodol, 2017, In press.