使用GRADEpro評價證據(jù)質(zhì)量

1、使用GRADEpro評價證據(jù)質(zhì)量武漢大學(xué)公共衛(wèi)生學(xué)院流行病教研室2011年11月GRADE系統(tǒng)簡介介GRADE（Grading of RecommendationsAssessment,DevelopmentandEvaluation）是由2000年建立的的GRADE工作組提提出的一一套評級級系統(tǒng)。GRADE系統(tǒng)使用用易于理理解的方方式評價價證據(jù)質(zhì)質(zhì)量和推推薦等級級，目前前已被世世界衛(wèi)生生組織（WHO）、Cochrane協(xié)作網(wǎng)等等一批著著名機(jī)構(gòu)構(gòu)所采用用。GRADE系統(tǒng)統(tǒng)較之其其他系統(tǒng)統(tǒng)的優(yōu)勢勢由一個具具有廣泛泛代表性性的國際際指南制制定小組組制定明確界定定了證據(jù)據(jù)質(zhì)量和和推薦強(qiáng)強(qiáng)度清楚評價價

2、了不同同治療方方案的重重要結(jié)局局對不同級級別證據(jù)據(jù)的升級級與降級級有明確確、綜合合的標(biāo)準(zhǔn)準(zhǔn)從證據(jù)到到推薦全全過程透透明明確承認(rèn)認(rèn)價值觀觀和意愿愿就推薦意意見的強(qiáng)強(qiáng)弱，分分別從臨臨床醫(yī)生生、患者者、政策策制定者者角度做做了明確確實用的的詮釋適用于制制作系統(tǒng)統(tǒng)評價、衛(wèi)生技技術(shù)評估估及指南南證據(jù)質(zhì)量量（GRADEpro）為達(dá)到透透明和簡簡化的目目標(biāo)，GRADE系統(tǒng)將證證據(jù)質(zhì)量量分為高高、中中、低低、極極低4級。一些些使用GRADE系統(tǒng)的組組織甚至至把低和和極低歸歸為一級級。推薦等級級使用GRADE系統(tǒng)時時，指南南小組用“強(qiáng)推推薦”表表示他們們確信相相關(guān)的干干預(yù)措施施利大于于弊。用“弱推推薦”表表示干

3、預(yù)預(yù)措施有有可能利利大于弊弊，但他他們把握握不大。GRADE中的證據(jù)據(jù)質(zhì)量和和推薦等等級符號字母/數(shù)字證據(jù)質(zhì)量 高質(zhì)量A 中等質(zhì)量B 低質(zhì)量C 極低質(zhì)量D推薦等級 支持使用某療法的強(qiáng)推薦1 支持使用某療法的弱推薦？2 反對使用某療法的強(qiáng)推薦1 反對使用某療法的弱推薦？2關(guān)鍵步驟驟一導(dǎo)導(dǎo)入RevMan文件關(guān)鍵步驟驟二判判斷結(jié)局局的重要要程度GRADE系統(tǒng)建議議采用9級分級判判斷結(jié)局局的重要要程度:13：不重要要結(jié)局（NOTIMPORTANT）46：重要結(jié)結(jié)局（IMPORTANT）79：關(guān)鍵結(jié)結(jié)局（CRITICAL）關(guān)鍵步驟驟三選選擇研究究設(shè)計類類型按研究設(shè)設(shè)計方法法劃分隨機(jī)試驗驗質(zhì)量等等級最高高

4、，為高高；觀察性研研究（包括隊隊列研究究、病例例-對照研究究、病例例-歷史對照照研究）質(zhì)量等等級低，為低；病例報告告、病例例分析等等其他研研究質(zhì)量量等級非非常低，為極低低。關(guān)鍵步驟驟四評評價證據(jù)據(jù)質(zhì)量基于研究究設(shè)計可可因5條降級因因素和4條升級因因素上升升或下降降級別。決定證據(jù)據(jù)質(zhì)量的的因素可能降低低證據(jù)質(zhì)質(zhì)量的因因素研究的局局限性結(jié)果不一一致間接證據(jù)據(jù)精確度不不夠發(fā)表偏倚倚可能增加加證據(jù)質(zhì)質(zhì)量的因因素效應(yīng)值很很大可能的混混雜因素素會降低低療效劑量-效應(yīng)關(guān)系系研究設(shè)計計與證據(jù)據(jù)質(zhì)量高：為不不降級的的隨機(jī)試試驗和升升2級的觀察察性研究究；中：為降降級的隨隨機(jī)試驗驗和升1級的觀察察性研究究；低：降

5、2級的隨機(jī)機(jī)試驗和和沒有升升降級的的觀察性性研究；非常低：降3級的隨機(jī)機(jī)試驗、 降1級的觀察察性研究究、病病例分析析/病例報告告。幫助教程程要點研究的局局限性（Risk of bias）這些局限限性包括括（RCTs）隱藏分組組缺失盲法缺失失（特別別是結(jié)局局指標(biāo)為為主觀性性指標(biāo)且且對其的的評估極極易受偏偏倚影響響時）失訪過多多未進(jìn)行意意向性分分析、觀觀察到療效效就過早早終止試試驗或未報道道結(jié)果（通常是是未觀察察到療效效的一些些研究）Torate study limitations:Ifyouthinkany limitations werenegligiblechoosenoIfyouthink

6、therewereserious limitationschooseseriousthiswilldowngradethequalityofevidenceforthisoutcome by1 levelIfyouthinktherewerevery seriouslimitationschoosevery seriousthiswilldowngrade thequality of evidence forthis outcomeby2 levels結(jié)果不一一致（Inconsistency）Inconsistency mayarisefrom differences in:populatio

7、ns(e.g.drugsmayhavelargerrelativeeffectsinsickerpopulations)interventions (e.g. largereffects withhigher drugdoses)outcomes(e.g.diminishingtreatmenteffect withtime)差異可能能源于人人群（如如藥物對對重癥人人群的療療效可能能相對顯顯著）、干預(yù)措措施（如如較高藥藥物劑量量會使療療效更顯顯著）或或結(jié)局指指標(biāo)（如如隨時間間推移療療效降低低）。I2Oneset of criteria would saythat an I2ofless tha

8、n40% is low3060%may be moderate5090%may be substantial75100% is considerableTorate inconsistency:Ifyouthinkany inconsistencywasnegligiblechoosenoIfyouthinktherewasserious inconsistencychooseseriousthiswilldowngradethequalityofevidenceforthisoutcome by1 levelIfyouthinktherewasvery seriousinconsistenc

9、ychoosevery seriousthiswilldowngradethequalityofevidenceforthisoutcome by2 levels間接證據(jù)據(jù)（Indirectness)Therearetwo types of indirectness:1.Indirectcomparison2.IndirectpopulationinterventioncomparatoroutcomeTorate indirectness:Ifyouthinkthe evidence isdirectchoosenoIfyouhaveserious doubtsaboutdirectness

10、chooseseriousthiswilldowngradetheevidencefor thisoutcomeby1 levelIfyouhavevery seriousdoubts about directnesschoosevery seriousthiswilldowngradetheevidencefor thisoutcomeby2 levels精確度不不夠（Imprecision）當(dāng)研究納納入的患患者和觀觀察事件件相對較較少而致致可信區(qū)區(qū)間較寬寬時，指指南小組組將降低低該研究究的證據(jù)據(jù)質(zhì)量。A thresholdrule-of-thumb valueTotalnumberofev

11、ents is lessthan300（dichotomous）Totalpopulation sizeislessthan 400（continuous）Torate imprecision:Ifyouthinkthe resultswereprecisechoosenoIftherewasserious imprecisionchooseseriousthiswilldowngradethequalityofevidenceforthisoutcome by1 levelIftherewasvery seriousimprecisionchoosevery seriousthiswilld

12、owngradethequalityofevidenceforthisoutcome by2 levels發(fā)表偏倚倚（Publicationbias）Publicationbias ariseswhen investigatorsfailtoreportstudiesthey haveundertaken(typically those thatshownoeffect).Methods to detectthepossibilityofpublicationbiasinsystematic reviewsexist, although authorsofthe reviewsand guid

13、elinepanelsmustoftenguessaboutthelikelihoodofpublicationbias.A prototypicalsituationthatshouldelicit suspicionofpublicationbias occurswhen publishedevidenceislimited to asmallnumber of trials,all of which areshowing benefits of thestudied intervention.若研究者者未能發(fā)發(fā)表研究究（通常常是陰性性結(jié)果的的研究）時，證證據(jù)質(zhì)量量亦會減減弱。典典型情況況

14、是當(dāng)公公開的證證據(jù)僅局局限于少少數(shù)試驗驗而這些些試驗全全部由企企業(yè)贊助助，此時時不得不不質(zhì)疑存存在發(fā)表表偏倚。FunnelPlotTorate probability of thepublicationbias:IfyouthinkthereisnoevidenceofpublicationbiaschooseunlikelyIfthereishigh probability of publication biaschooselikelythiswilldowngradethequalityofevidenceforthisoutcome by1 levelIfthereisvery high

15、probabilityofpublicationbiaschoosevery likelythiswilldowngradethequalityofevidenceforthisoutcome by2 levels效應(yīng)值很很大（Largeeffect）療效越顯顯著，證證據(jù)越有有力。Torate magnitudeoftheeffect:Iftheeffect wasnotlarge(RRbetween0.5and 2.0)choosenoIftheeffect waslarge(RReither 2.0or2or5.0or5or0.2thiswillupgradethequalityofevi

16、denceforthisoutcome by2 levels可能的混混雜因素素會降低低療效（Plausibleconfoundingwouldchange theeffect）Onoccasion,all plausibleconfoundingfrom observationalstudiesorrandomised trialsmaybeworkingtoreducethe demonstratedeffectorincreasethe effectifnoeffectwas observed.Forexample,ifonlysickerpatientsreceiveanexperime

17、ntalintervention or exposure,yettheystillfare better,itislikely thatthe actualinterventionorexposureeffect is largerthan thedata suggest.Torate theeffectofall plausibleresidualconfounding:Ifthereisnoevidencethat theinfluenceofall plausibleconfoundingwouldreduceademonstratedeffect or suggestaspurious

18、effectwhenresults shownoeffect choosenoIfthereisevidencethat theinfluenceofall plausibleconfoundingwouldreduceademonstratedeffect or suggestaspuriouseffectwhenresults shownoeffect chooseyesthiswillupgradethequalityofevidenceforthisoutcome by1 level劑量-效應(yīng)關(guān)系系（Dose-response gradi）Thepresenceofadose-resp

19、onse gradient mayincreaseourconfidenceinthefindingsofobservationalstudies andthereby increase thequality of evidence.Torate thepresenceofdose-response gradient:Ifthereisnoevidenceofdose-response gradient choosenoIfthereisevidenceofdose-response gradient chooseyesthiswill upgradethe qualityofevidencefor thisoutcomeby1 level評級流程程關(guān)鍵步驟驟五完完善數(shù)據(jù)據(jù)關(guān)鍵步驟驟六生生成表格格EvidenceProfile vs.Summary of Findings TableGRADEEvidenceprofileParticularlyuseful forguidelinedevelopersPresentsinformationaboutthebodyofevide