meta分析-選修課第二軍醫(yī)衛(wèi)生統(tǒng)計學(xué)教研室

第 臨床問 問：如何減少早產(chǎn) ：7個高質(zhì)量的RCT，其中5個試驗結(jié)果2生活中的問查 3 系統(tǒng)評價/系統(tǒng)綜Systematic4Thebuzzwordinmedicineinthe21stWhereisthe 56 ysisreferstothe ysisof yses...thestatistical ysisofalargecollectionof resultsfromindividualstudiesforthepurposeofintegratingthefindings…（G.Glass, ---after,morecomprehensive,英國教育心理7

中文定統(tǒng)、全面地收集全世界所有已或未的臨床研究，采用Meta8Systematicreview(日趨規(guī)范的用法 SystematicPooling

Numberof ysisPublicationsbyNumberofNumberof ysisPublicationsNumberof01993199419951996199719981999200020012002系統(tǒng)評價的主要特清楚地表明題目和目的：采用綜合檢索策略：例 詞檢索 檢如果可能使用Meta大規(guī)模應(yīng)對信息時代 、海量信息需要整系統(tǒng)綜

信息合

應(yīng)對信息時代 1986 推薦常規(guī)使用靜脈鏈激酶治療應(yīng)對信息時代 低早產(chǎn) 率和呼吸窘迫綜合征的發(fā)生率5

敘述性文獻(xiàn)綜述（Narrative 敘述性文獻(xiàn)綜述與系統(tǒng)評價的較通過對同一多個小樣本研究結(jié)果的綜Cochrane Cochrane系統(tǒng)評價是循證醫(yī)學(xué)實(shí)踐中最高質(zhì)量的證CochranedatabaseofsystematicreviewsSCIIF：5.715(2011)Meta分析步

Meta

關(guān)鍵環(huán)

Meta向該領(lǐng)域 請以“腎移植”，“爭議檢索，或以transplantationcontroversydispute

Meta是很久以前做的或者前面的研究有進(jìn)一步研究Meta4 應(yīng)該采取什么樣的治療方法？不Meta

Meta JAMA.2006;295:74- suggestthatstatinsreduce riskof cancer. Pharmacologically,statinsmayreducecancerviareductionsininflammation,neovasculaformation,andcellprolifeationbutc inhibitselenoproteinsynthesisanddecreaenaturalkillercellfunction,whi enhancecancerrisk.

ysesof randomizedrandomizedstatintrials制定計劃系統(tǒng)評價/Meta背景資料（國內(nèi)外現(xiàn)狀（二）原則：多途徑、 臨床試 登記系

這些文獻(xiàn)中可能包研究結(jié)注意：未正

Search MEDLINEfrom1966throughJuly EMBASEfrom1990throughJuly CINAHLfrom1982throughJuly WebofSciencefrom1994throughJuly CANCERLITfrom1975throughJuly TheCochraneSystematicReview

SearchMedicalSubjectHeadingand HMG-CoAredu einhibitor,HMGCoARI,statin,pravastatin,simvastatin,lovastatin,atorvastatin,cerivastatin,rosuvastatin,andSearchAmanualsearchof spresentedattheannualmeetingsAmanualreviewofreferencesfromprimaryorreviewarticlesAllpotentiallyrelevantarticleswerereviewedAllpotentiallyrelevantarticleswerereviewedindependentlyby3investigators 主要研究結(jié)果：35天內(nèi) （三）可分為三步實(shí)施根據(jù)題目 初篩，選出可能合格閱讀全文，納入肯定合格 文獻(xiàn)選擇的步 排 納Study納入標(biāo) 試驗設(shè)Randomizedtrialsof

Placebo-orroutine隨訪時Haveamean(or隨訪時Enrollaminimumof試驗結(jié)Reportdataontheincidenceof試驗結(jié)評價納入研究的偏倚風(fēng)險：評估單個臨少系統(tǒng)誤差（偏倚）和隨機(jī)誤差的程度。為什么要進(jìn)行納入研究真實(shí) 作為納入原始文獻(xiàn)的閾值解釋不同文獻(xiàn)結(jié)果差異的原因進(jìn)行系統(tǒng)評價敏感性分析偏倚的類型統(tǒng)差異，特別 判斷研究結(jié)果時常會出現(xiàn)報告偏倚Cochrane協(xié)作網(wǎng)的“偏倚風(fēng)險評估”工6ValidityThefollowingmethodologicalfeaturesmostrelevanttothecontrolofbiaswereassessed:randomization,randomallocationconcealment,maskingoftreatmentallocation,blinding,andwithdrawals.Allstudieswereevaluatedby3independentreviewers，withdisagreementresolvedby納入研究的資料提制定資料摘錄表格，按表格內(nèi)容提應(yīng)變量并填表提取的主要內(nèi)一般資料(試驗名稱、作者、出處等基線資料 質(zhì)量資料(隨機(jī)、盲法、隨訪等干預(yù)措施資料(治療方法、劑量、療程，對照措施結(jié)局資料 （五）收集數(shù)建立數(shù)據(jù)庫：Excel、Revman注意對計量資料必須注明單兩人同時獨(dú)立地對文獻(xiàn)提取信 Authoridentification,yearofpublication,geographiclocationofthestudyfundingsource,typeofstudydesign pectiveorrandomizedorobservational,presenceand

ndedopen-

n,samplesize,durationofpatientfollow-statinused,typeofcancerdiagnosesincluded

respiratory,gastrointestinal,ormelanoma),definitionsofcancertrialsforcancerend andcancerdeath(whenreported),andmethodtrialsforcancerend及真實(shí)性及真實(shí)性評系統(tǒng)評價不一定都要做Meta分析，更要要的是全面系統(tǒng)收Meta分析的步提出問提出問收集和評價文數(shù)據(jù)分結(jié)論與討擬定研究擬定研究計文獻(xiàn)檢異質(zhì)性檢報告和解釋結(jié)和和文獻(xiàn)的選合并效得出結(jié)提取、整理數(shù)亞組分局限性的討敏感性分析偏效應(yīng)量的統(tǒng)計描OddsratioRelativeriskRiskdifferenceMeandifferenceStandardmeandifference例如表1口 OR(95％CI)：2.20（1.26-

口 心肌梗死有是用者的2.20性森林森林

原始文獻(xiàn)的原始文獻(xiàn)的編號

所有納入研的綜合結(jié)1.HeterogeneityStatisticalheterogeneitywasmeasuredusingtheQstatistic(P.10wasconsideredrepresentativeofsignificantstatistical 異質(zhì)性檢驗(Heterogeneity了解各獨(dú)立研究結(jié)果合并的合理性（可合并性）異質(zhì)性檢驗(Heterogeneity定義 、結(jié)局及其測量方法不觀察對象不同（協(xié)變量的存在1.Heterogeneity異質(zhì)性的識別：1.通過森林各研究的可信區(qū) 越多，同質(zhì)性越Heterogeneity異質(zhì)性的識別：1.通過森林各研究的可信區(qū) 越多，同質(zhì)性越0Heterogeneity0異質(zhì)性的識別 1.通過森林各研究的可信越少，異質(zhì)性明顯者異質(zhì)性有顯著性差Heterogeneity異質(zhì)性的識軟件完通過統(tǒng)計檢驗識別異質(zhì)性：軟件完Q服從于2分布，Q值（2值當(dāng)P值0.10 ysis--HeterogeneityNostatisticallysignificantheterogeneitywasobservedbetweentrialsforeither withtheQstatistic(P.37).HowmuchistoomuchI2 Meta回亞組分模Meta回亞組分模不合模Meta分析中異質(zhì)性資料處理的方的處異質(zhì)異質(zhì)異質(zhì)合并效應(yīng)量與統(tǒng)計推（Data

比值比（odds 相對 度（relativerisk，RR） 差（risk 用于度量衡單位相同的連續(xù)性變用于度量衡單位相同的連續(xù)性變量標(biāo)準(zhǔn)化均數(shù)差（standardizedmean度量衡單位不同的和均數(shù)相差較大的連續(xù)性變量合并效應(yīng)量與統(tǒng)計推The95%CIcross1.00foroverallincidenceofcancerdiagnosis ysesforpatientsreceivingstatintherapyvscontrolsOR1.0295CI0.97-1.07StatinshaveaStatinshaveaneutraleffectoncancerandcancerdeathriskinrandomizedcontrolledtrials.森林各個研究合后的效應(yīng)估亞組分析 亞組分析 Toestablishtheeffectofclinicalheterogeneitybetweenstudiesonmeta ysis’conclusions,subgroup wasconducted.Sincetheeffectofstatinsoncancermayvaryfromonesubtypeofcancertoanother,theeffectoftheseagentsontheincidenceofdifferenttypesofcancerdiagnosis(breast,prostate,colon,respiratory,gastrointestinal,ormelanoma)wasevaluated.Inaddition,becausenumerouspharmacologicdifferencesexistbetweenavailablestatins,comparisonswereconductedofeachindividualstatinstudied(atorvastatin,pravastatin,simvastatin,lovastatin,andfluvastatin),aswellasofhydrophilic(atorvastatin,pravastatin,andfluvastatin)orlipophilic(simvastatin,cerivastatin,andlovastatin)agentsandnaturally(pravastatin,simvastatin,andlovastatin)orsynthetically(atorvastatin,cerivastatin,andfluvastatin)derivedagents.亞組分析 The95%CIs1.00foreachof The95%CIs1.00foreachofthese 敏感性分析 檢查一定假設(shè)條件下結(jié)果穩(wěn)定性 敏感性分析 Studiesofpoorermethodologicalquality,suchasunblindedoropen-labeltrials,mayexhibitexaggeratedtreatmenteffects.Excludingthemmayresultinincreasedinternalvaliditybutcould Studiesofpoorermethodologicalquality,suchasunblindedoropen-labeltrials,mayexhibitexaggeratedtreatmenteffects.Excludingthemmayresultinincreasedinternalvaliditybutcould ysiswasconductedwhereby ysiswas yzedexcluding oropenlabelstudiesandusinga - fixed-effectsfixed-effects敏感性分析 Insensitivity ysis,themeta conclusionsre inedrobusttomethodologicalchanges. Theresultsoftheincidenceofcancerdiagnosisorcancerdeath yseswere notalteredwhenafixed-effectsmodelwasused Whenonlyrandomizeddoubleblind,placebo-controlledstatintrialswereevaluated,therewasnosignificantdifferenceincancerincidence。偏倚（Publicationbias 偏倚（PublicationbiasSeveralmethodswereusedtoassessthepotentialforpublicationbias.Visualinspectionoffunnelplotsforboththecancerdiagnosisandcancerdeathendpointswasconducted.偏倚（Publicationbias偏倚的漏斗圖分析（漏斗圖分析（funnel計算失效安全數(shù)（fail-safe 漏斗圖分析（funnelOur ysis’funnelplotsappear asymmetrical.Therefore,inevaluatingourplots,wecanasymmetrical.Therefore,inevaluatingourplots,wecanreportonlythattheremay補(bǔ)充：針灸治療中風(fēng)的49個試驗的漏斗圖分 失安全數(shù)（fail-safenumber，Nfs）： 用途：估 偏倚的程P為0.05和0.01時計算公式如Nfs0.05

(

1.64)2Nfs0.01

(

2.33)2 偏倚（Publicationbias