页数:60页文件大小:590.0 KB来源: 流行病学ppt课件
本课件介绍系统综述与Meta分析的方法流程,从选题、文献检索与质量评价入手,讲解效应量合并、异质性检验与模型选择,并说明发表偏倚的识别与结果解读要点。
系统综述和Meta分析Systematical review and meta-analysis
系统综述(systematic review)被公认为客观地评价和综合针对某一特定问题的研究证据的最佳手段.A systematic review is a literature review focused on a singlequestion that tries to identify, appraise, select and synthesize allhigh quality research evidence relevant to that question.An understanding of systematic reviews and how to implementthem in practice is becoming mandatory for all professionalsinvolved in the delivery of health care.
Meta分析是指用统计学方法对收集的多个研究资料进行分析和概括,以提供量化的平均效果来回答研究的问题.The National Library of Medicinedefine meta-analysis as a“quantitative method of combining the results of independent studies(usually drawn from the published literature) and synthesizingsummaries and conclusions which may be used to evaluatetherapeutic effectiveness, plan new studies, etc., with applicationchiefly in the areas of research and medicine.”Meta分析(Meta-analysis)
第一节概述Outline第二节步骤和方法Steps and Methods第三节Meta分析常用统计方法Common statisticalmethods in meta-analysis第四节偏倚及其检查Bias andidentification目录outline
第一节概述
一、简史History系统综述的思想最早可以追溯到17世纪瑞利勋爵(Lord Rayleigh,1842~1919)最早描述这样一种方法1898年Gould进一步明确1904年Pearson首次提出数据合并的概念20世纪20年代,Fisher介绍了对若干独立试验结果的P值进行合并的方法
表1976 年,英国心理学家Glass首:本页为数据表,下列为源文件提取的原文数据,供对照核查。
1955年,首次对治疗的有效率进行Meta分析的文章发表1976年,英国心理学家Glass首次将合并统计量的文献综合研究称为Meta分析20世纪70年代,Chalmers医生经过10年的努力,对产科常规使用的226种诊疗方法进行了系统综述,由此开启了循证医学的序幕
本页含数据表,源文件由图片提取文本,数字与列边界粘连,无法安全自动切分为表格,需对照原课件人工还原。
Systematic reviews/meta-analyses indexed inPubMed–10 years
二、基本概念
系统综述的主要特征1.清楚地表明题目和目的2.采用综合检索策略3.明确的研究入选和排除标准4.列出所有入选的研究5.清楚地表达每个入选研究的特点并对它们的方法学质量进行分析Characteristics ofsystematical review1.Proclaim title and aim clearly2.Adopt systematical strategy forliterature search3.Define study inclusion andexclusion criteria4.List all included studies5.Describe the characteristics ofincluded studies and assess thequality of them
Clarify the causes
Meta分析简史History of meta-analysis1.Meta(after, more comprehensive, secondary)希腊词2.1976年以综合研究结果为目的而对不同的研究结果进行收集、合并及统计分析的一种方法(Glass )3.1986年对先前研究结果进行统计合并和评述的一种新方法(Sack )4.1987年用以汇总众多研究结果的各种定量分析(Hedge )5.1991年是一类统计方法,用来比较和综合针对同一科学问题所取得的研究结果(Fleiss和Gross )
定义以综合研究结果为目的,通过查阅文献收集与某一特定问题相关的多个研究并对这些研究的结果所进行的统计分析A statisticalmethod forsynthesizing theeffect sizes fromseveral studiesthat are relatedwith one specifiedquestionDefiniftion
几个、甚至上百个研究者在不同地区进行研究并发表研究结果进行整合后所得到的综合结果(证据)更有说服力优点The pooled resultssynthesized thestudiesthat were conducted by sveral orhundreds of researchers from differentcountries are more credible
与传统文献综述的区别Differences with traditional review传统文献综述的缺陷主观综合缺乏共同遵守的原则和步骤注重统计学是否“有意义”等价对待每篇文献,无权重定性而非定量Shortage of traditional reviewSubjective synthesisWithout consistent criteriaand stepsFocus on statisticalsignificanceDealing each study withoutweightProduce quantitativeresults, rather thanqualitative findings
系统综述和Meta分析的功能Function of systematical review and meta-analysis定量综合提供系统的、可重复的、客观的综合方法通过对同一主题多个小样本研究结果的综合,提高原结果的统计效能,解决研究结果的不一致性,改善效应估计值回答原各研究未提出的问题Quantitative synthesisSupply systmatical, duplicableobjective synthetic methodsIncrease statistical power andexplain why the results ofdifferent studies vary bysynthesizing the results fromseveral individual studies withsmall sample sizeAnswer the questions that werenot mentioned in originalstudies
第二节步骤和方法Steps and methods制定研究计划,收集资料make a study design根据入选标准选择合格的研究Search for related articles according to inclusion criteria异质性检验,敏感性分析identification of heterogeneity and sensitivity analysis计算各独立研究的效应大小Caculate effect size of individual study提取信息,填写过录表,建立数据库Data extraction, database set up复习每个研究并进行质量评估Quality assessment for reviewed studies总结报告have a conclusion and report
一、制定研究计划Make a study design明确研究目的提出检验假设特殊注意的亚组确定和选择研究的方法和标准确定提取和分析资料的方法和标准Clarify study aimPropose null hypothesisPay attention to subgroupsClarify and choose studymethod and criteriaClarify the method and criteriafor data extraction
二、收集资料
三、根据入选标准选择合格的研究Choose the qualified studies according to inclusion criteria检索大量的文献根据入选和排除标准进行仔细的筛选挑出合格的研究进行系统综述和Meta分析Extensive literature searchScreen the studies based oninclusion criteriaMake systematical reviewand meta-analyaia forqualified studies
四、复习每个研究并进行质量评估Quality assessment for reviewed studies321精确度(随机误差的程度)一般用可信限的宽度来表示Accuracy(Degrre of random error)Width of confidenceinterval方法学质量研究设计和实施过程中避免或减小偏倚的程度Methodology qualityAvoid and diminishbias in study designand the process ofenforcement外部真实性研究结果外推的程度External validityPossbility forextrapolating
随机对照试验(RCT)的方法学质量考察Quality assessment for random control trial受试者分组是否真正随机随机分组方案是否隐藏是否详细说明入选标准组间基线是否可比Whether the subjects areallocated randomlyWhether randomization isblindWhether explain the inclusioncriteria clearlyWhether the baseline ofgroups is comparable
研究过程中是否使用了盲法对失访、退出及不良反应病例是否进行了详细录,是否报告失访原因是否采用意向分析法分析结果患者的依从性如何Whether “blind method” isusedWhether record the details forloss to follow-up, exit or adverseeffect and the causes of loss tofollow-upWhether use (intention-to-treat,ITT) to analysis the resultsHow about the compliance ofpatients
Criteria
RCT质量评价的Jadad量表随机化方法Randomization恰当如计算机产生的随机数字或类似的方法(2分)Appropriate Random number table or computer generated number (2points)不清楚试验描述为随机试验,但没有告知随机分配产生的方法(1分)Unclearly Description of random trial, but without the method ofrandom allocation (1 point)不恰当如采用交替分配或类似方法的半随机化(0分)InappropriateAlternative allocation or other methods of semi-randomization (0 point )
失访与退出
五、提取信息,填写过录表,建立数据库Data extraction, database set up按事先制定的资料摘录表内容提取相应变量并填表用meta分析软件MetaView或SPSS、SAS、EXCEL等建立数据库Extract the correspondingvariables and fill in theformSet up database usingMetaView, SPSS, SAS orEXCEL
对计量资料必须注明单位;比较的两组有x和S;
计数资料要使用相同的率来表示双重录入注意Indicate the unit for quantitative materialThere are x and S in two comparative groups.Qualitative material should use same rateDouble computer enter
六、计算各独立研究的效应大小Caculate effect size of individual study通常两组间比较时连续变量(continuous variable)平均差值(mean difference)率差(rate difference)二分变量(binary variable)比值比(OR)相对危险度(RR)
七、异质性检验The identification of heterogeneity重要性Meta分析重要的环节目的检查各个独立研究的结果是否具有可合并性注意资料的“可合并性”ImportanceA important component ofmeta-analysisAimExamin the possibility ofsynthesispay attention to thefeasibility of synthezising
产生异质性的原因研究设计不同;
试验条件不同;
试验所定义的暴露、结局及其测量方法不同协变量的存在Causes of heterogeneitydifferent study design,different trial condition,different definition ofexposure, outcome,methods of measurementpresence of covariant
Meta分析中异质性资料处理的方法How to deal with heterogeneity异质性的处理忽略异质性固定效应模型检验异质性有异质性不合并合并随机效应模型解释亚组分析Meta回归
异质性的统计学检验
A better approach : quantifying heterogeneity-------I2I2描述了去除抽样误差(机遇)后的异质性。
异质性程度判定:Degree of inconsistency 0% no observed heterogeneity<25% low25%<I2<75% moderate>75%high
八、敏感性分析Sensitivity analysis检查一定假设条件下结果稳定性的方法目的发现影响meta分析研究结果的主要因素解决不同研究结果的矛盾性发现产生不同结论的原因A methodinvestigating thestability of a meta-analysiswith certain hypothesisAimdetect the main factorsinfluence the result ofmeta-analysisResolve the contradictionof different resultsDetect the causes ofdifferent conclusions
Stratification
九、总结报告conclusion and report研究结果(线宽表示其95%CI)Width of line indicate the95%CI of resultInvalid line各个研究合并后的效应估计Pooled effect size研究结果点估计值,其大小代表该研究在Meta分析中的权重Point estimate of effect size, size ofpoint indicate the weight in meta
进行总结报告观察性研究的Meta分析(Meta-analysis of Observational Studies inEpidemiology,MOOSE)随机对照试验的Meta分析报告质量(The quality of reporting of Meta-analyses ofrandomized controlled trials, QUOROM)
第三节分析常用统计方法
假定第i项研究的总体效应为(suppose the total effect of i study):其他随机效应为(other random effect),可写成随机效应模型(random effect model ):i = 1,2,…,k其加权均数为(weighting means):其估计方差分别为(estimated variance):其中为权数,可以是合计例数,也可以是合并方差的倒数
检验的零假设为(null hypothesis of test)当成立时,统计量为,;
计算值后,可以查值表得到相应的P值,若,则拒绝,支持随机效应模型的假定;
若,则支持固定效应模型的假定。
一致性检验Consistency check
加权合并设k
公式同两均数之差的合并和一致性检验公式一致性检验Consistency check
加权合并暴露组病例人数exposed cases暴露对照人数exposed controls非暴露组病例人数unexposed cases非暴露组对照人数unexposed controls令写成随机效应模型有,其加权均数:均数的估计方差分别为:其中:三、病例-对照研究OR值的合并Statistical mehod for combining data of OR incase-control study
合并OR及其95%CI分别为:pooled OR and 95%CI:
各研究间差异的估计方差为Estimated variance for the difference among studies其中
一致性检验检验的零假设为(null hypothesis of test)成立,Q值服从自由度的分布;
若,则拒绝,支持随机效应模型的假定,必须用计算及其95%CI;
若,则支持固定效应模型的假定,即使,也可令,计算平均效应大小及其95%CI。
三、病例-对照研究一致性检验Consistency checkin case-control study
第四节偏倚及其检查Bias and examination发表偏倚(publication bias)定位偏倚(location biases)引用偏倚(citation bias)多次发表偏(multiplepublication bias)有偏倚的入选标准(biased inclusion criteria)一、偏倚的种类
ConceptionThose studies withstatistically significant results aremore likely to be submitted andpublished than studies withoutsignificant results.CausesSerious problem of publication biasfor medical papersIf meta-analysis is only based onthe results of published studies, theconclusion of meta-analysis mayexaggerate the effect.定义具有统计学显著性意义的研究结果较无显著性意义和无效的结果被报告和发表的可能性更大。
产生原因医学文献中发表偏倚的问题相当严重如果meta分析只是基于已经发表的研究结果,可能会夸大疗效,甚至得到一个虚假的疗效(一)发表偏倚Publication bias
(二)定位偏倚Location bias定义在已发表的研究中,阳性结果的文章更容易以英文发表在国际性杂志,被引用的次数可能更多,重复发表的可能性更大,从而带来文献定位中的偏倚。
Those studies with statistically significant results aremore likely to be published in international journals.产生原因英语偏倚(English language bias)文献库偏倚(Database bias)
(三)引用偏倚citation bias产生原因支持阳性结果的研究比不支持的研究可能更多地被作为参考文献加以引用The studies supporting the significant result are more likely tobe cited as reference than studies without significant results.杂志的知名度对文章的引用也会产生影响Thedegree of journals’ celebrity could influence the articles
(四)多次发表偏倚Multiple publication bias产生原因阳性结果的研究更容易多次发表或作为会议报告,这就使得这些文章更容易被查到并纳入Meta分析中Meta分析中如果包括重复数据会高估疗效The studies with significant result are more likely to be published oraccepted as conference reports, which causes these studies have morechance to be included in meta-analysisIf there are repeated data in meta-analysis, the conclusion mayoverestimate the effect.
二、偏倚的检查Examination of publication biasMeta分析检查偏倚的最佳途径先根据一个基本的入选标准收集全部的研究再考虑不同的入选标准进行彻底的敏感性分析检查偏倚的程度漏斗图分析计算失效安全数Best way for examining bias inmeta-analysisCollect all studies based ona standard inclusion criteriaMake a sensitivity analysiswith the consideration ofdifferent inclusion criteriasDegree of biasFunnel plotsFail-safe number,Nfs
(一)漏斗图分析Funnel plot漏斗图相对于样本量的效应值,是以研究的效应估计值作为横坐标,样本量作为纵坐标画出的散点图漏斗图分析根据图形的不对称程度判断Meta分析中偏倚有无。
They are scatter plots of the effect estimates from individualstudies (horizontal axis) against sample size or some otherindicators of the estimates (vertical axis)
针灸治疗中风的49个试验的漏斗图分析(Tang TL, 1999)如果Meta分析中没有偏倚,图形构成一个对称的倒置“漏斗”;
如果图形呈现明显的不对称,表明偏倚可能存在(见右图)样本量小的研究结果通常分散在图形底部很宽的范围内样本量大的研究结果集中在图形上部一个较窄的范围内The most precise estimates (those from larger studies)are at the top of the funnel and those from less preciseor smaller studies are more likely spread at the base ofthe funnel because of their larger standard errors.
The end
下载地址:{{downurl}}
{{{a}}}
{{#b}}{{{b}}}
{{/b}} {{/title}} {{#nolog}}登陆注册
{{/nolog}} {{#djsx}}点我刷新
{{/djsx}} {{#gocz}} {{/gocz}} {{#ffqr}} {{/ffqr}}