




版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
1Bias&Validity
inEpidemiologicalResearch
流行病学研究中的偏倚与真实性陈裕明教授医学统计与流行病学系Tel:873306052MaincontentsMeasurementError测量误差RandomerrorSystemicerror(Bias)Bias偏倚:sources,commonbias,andcontrolmethods来源、常见偏倚、控制方法Selectionbias选择偏倚Informationbias信息偏倚Confoundingbias混杂偏倚3ShouldIbelievetheresults?CoffeeDiabetesRR=0.7(95%CI,0.65-0.84)Chance?Confounding?Bias?Trueassociation
Causal Non-causal4MeasurementErrorRandomerror:leadstounreliablemeasurementSystematicerror:leadstobiasedmeasurement56ReliabilityandValidityReliableBiased
(NotValid)NotReliableBiased
(NotValid)NotReliableValidReliableValidRandomerror–measurementnotreliableSystematicerror–measurementbiased(notvalid)7Measurementerror:
ImplicationsforepidemiologicstudiesMeasurementerrorintheexposureortheeRANDOMERROR–Biastowardthenull(e.g.,RR=1)Systematicerror–biasineitherdirectionMeasurementerrorinaconfounderRandomerror–dilutesabilitytoadjustSystematicerror–confoundedestimateofeffect8Measurementerror:
ImplicationsforepidemiologicstudiesRANDOMERROR
Biastowardthenull(e.g.,RR=1)Systematicerror
biasineitherdirection9RandomerrorErrorduetochance“thatpartofourexperiencethatwecannotpredict”Usuallymosteasilyconceptualizedassamplingvariability10Draw1,000randomsamplesfrompopulationwith50%women11Attenuationoftherelations
随机误差减弱关联强度Physiologicalvariables********r=0.98r=0.6512Randomerror随机误差Errorduetochance“thatpartofourexperiencethatwecannotpredict”Usuallymosteasilyconceptualizedassamplingvariability13Randomerrorcanbeproblematic,but...Influencecanbecontrolledby…–increasingsamplesize
–changingdesign
–improvinginstrumentation14Howdowequantifyrandomerrorinatestorastudy?Foratestofexposures,es,confoundersCoefficientofvariation,CVRepeatmeasures,CV=SD/meanForatestofassociation,incidence,prevalenceHypothesistesting,p-value,95%CI15Bias16Definition
of
biasBias:anytrendinthedesign,conduct,analysis,interpretation,publicationorreviewofdatathatcanleadtoconclusionswhicharesystematicallydifferentfromthetruth.(DictionaryofEpidemiology,3rded.)
在设计、实施、分析、结果解释、发表或评阅过程中,能够导致结论系统地偏离真实值的任何趋势17Question?Canyoucontrolorreducebiasbyincreasingsamplesize?18Errorsinepidemiologicalstudies误差随机误差系统误差(偏倚)样本量19Classificationofbias分类Selectionbias
选择偏倚–differentialaccesstothestudypopulationInformationbias
信息偏倚–inaccuracyinmeasurementorclassificationConfoundingbias
混杂偏倚–unfaircomparison20SELECTIONBIAS-definitionSELECTIONBIAS选择偏倚:
Errorsintheprocessofidentifyingthestudypopulation,whichcausesthestudysampledoesnotrepresentthetargetpopulation.在确定研究人群时出现的系统误差,使得研究样本不能代表总体21SamplingandrepresentativenessSampleTargetPopulationSamplingPopulationTargetPopulationSamplingPopulationSample22SourcesofSelection
bias
选择性偏倚的来源1.Samplingbias抽样偏倚--thedegreetowhichthemeasuredsubjectsarenotrepresentativeofthepopulation,isdueto:
ConvenienceSample方便样本Violatedrandomsamplingscheme偏离随机抽样方案HaphazardSample偶然样本23SourcesofSelection
bias
选择性偏倚的来源2.Participationbias参加者偏倚–duetoparticipantsSelf-selection(volunteerism)
自我选择偏倚Healthy(ordiseased)peoplemayseekoutparticipationinthestudyNon-response,refusal无应答偏倚Response,orlackofit,dependsonexposureHealthyworkereffect
健康工人效应Differentiallosstofollow-up
选择性失访Biasduetodifferencesincompletenessoffollow-upbetweencomparisongroups24SourcesofSelection
bias
选择性偏倚的来源3.Ascertainmentbias确认偏倚--systematicerrorduetoadifferenceincharacteristicsbetweenwhochoosetoparticipateinastudyandwhosewhodonot.
Prevalence-incidencebias
现患与新发病例偏差Prevalentcaseislikelytobemild,andsurvivelongerReferralbias
转诊偏倚Sickerpatientsarereferredtomajorhealthcenters25SourcesofSelection
bias
选择性偏倚的来源3.Ascertainmentbias确认偏倚Diagnosticbias诊断偏倚DiagnosisofeassociatedwithexposureAdmissionratebias(Berkson’sbias)入院偏差Hospitalizationratesdifferforbydiseaseandpresence/absenceoftheexposureofinterestLeadtimebias领先时间偏倚26Sampling
bias抽样偏倚Howrepresentativearehospitalisedtraumapatientsofthepopulationwhichgaverisetothecases?OR=627Sampling
bias抽样偏倚OR=6OR=3628Diagnostic
bias诊断偏倚OralContraceptiveuse
breakthroughbleeding
increasedchanceofdetectinguterinecancerDiagnosticapproachrelatedtoknowingexposurestatus
Overestimationof“a”
overestimationofOR29Non-responsebias无应答偏倚Controlschosenamongwomenattheirhomes:13000homescontacted
1060controlsUnderestimationof“b”
overerestimationofORControlsmainlyhousewiveswithlowerchanceofhavingtestthanwomengainfullyemployedPapersmearHadtestDidn’thavetest30Healthyworkereffect
健康工人效应31Healthyworkereffect32Admissionbias
HCV与肝癌的meta分析结果病例-对照来源研究数目OR(95%CI)医院—医院87.80(5.04—12.06)医院—社区76.70(3.28—13.68)社区—社区93.34(2.44—4.56)(吴江南
等.中华肝脏病杂志
2007)33Losstofollow-upbias
失访偏倚Biasduetodifferencesincompletenessoffollow-upbetweencomparisongroupsExampleStudyofdiseaseriskinmigrantsMigrantsmorelikelytoreturntoplaceoforiginwhenhavingdisease
losttofollow-up
lowerdiseaserateamongexposed(=migrant)34QuickQuizToassesstheassociationofexerciseandCHDinacohortstudySupposethatinfact,theannualincidenceofCHDdeathamongexercisersis5/105andamongnon-exercisersit’s10/105.Subjectswithouttelephonesaremorelikelythanphoneownerstobenon-exercisers.Doesnotincludingtheminthesamplerepresentabias?35QuizAnswerIftheno-phonesubjectsdifferonlyintheirexercisepatterns,thenno,thisisnotabias.Westillhaveexercisersandnon-exercisers,andpresumablywillstillfindthesameincidenceratesineachgroup.But,iftheydifferalsointheirotherrisksforCHDdeath,thenthiswouldbeabias.36Answercontinued...Forinstance,theno-phonenon-exercisersmayhaveothermajorriskfactors,likesmoking,obesity,anddiabetes.Losingthemwouldremovesomeoftheriskseenwithnon-exercising–maybenowwewouldfind5/105fortheexercisersand8/105forthenon-exercisers.Exclusioncausesabiasiftheexcludeddifferinboththeexposureandthee.37MethodstoreduceselectionbiasRigorous,butfeasiblerandomsamplingscheme严谨可行的随机抽样方案Cleardefinitionofstudypopulation
清楚定义研究人群Explicitcase/control,
expose/non-exposedefinitions 明确病例/对照,暴露/非暴露的定义Casesandcontrolsfromsamepopulation 病例和对照来自同一人群38MethodstoreduceselectionbiasSelectionofcasesandcontrolswithoutknowingexposurestatus(case-controlstudy)选择病例对照不要知道暴露状态Selectionofexposedandnon-exposedwithoutknowingdiseasestatus(retrospectivecohort)选择暴露与非暴露人群不要知道结局Striveforhighparticipationrate(incentive)尽量提高参与率39Information
bias
信息偏倚40Information
biasSystematicerrorinthemeasurementofinformationonexposureore
在测量暴露或结局时发生的系统误差Differencesinaccuracyofexposuredatabetweencasesandcontrolsofedatabetweendifferentexposuregroups41SourcesofInformationbias
偏倚的来源Respondent应答者,:inabilitytounderstand,recall,articulate;unwillingnesstodiscloseorsocialdesirabilityDatacollector资料收集者:unclearorambiguousquestions,lackofaneutraldemeanor,insufficientlyconscientious,inaccuratetranscription,fraudInstrument测量工具:inaccurateinstrument42SourcesofinformationbiasDatamanagers资料管理者:inaccuratetranscription,mis-reading,miscoding,programmingerrorsDataanalysts资料分析者:variablecodingandprogrammingerrorsDatainterpreters资料解释者:inadequateappreciationofthecharacteristicsofthemeasureoroftherelationsbeingstudied43MaintypesofinformationbiasReportingbias
报告偏倚Recallbias回忆偏倚Biasedreporting偏倚的报告Observerbias
观察者偏倚Datacollectionbias资料收集偏倚Interviewerbias访问者偏倚Biasedfollow-up
偏性随访detectionbias检测偏倚Measurementbias测量偏倚44DefinitionofcommoninformationBias-1Datacollectionbias资料收集偏倚Biasthatresultsfromabstractingcharts,interviewsorsurrogateinterviewsRecallbias回忆偏倚DiseaseoccurrenceenhancesrecallaboutpotentialexposuresDetectionbias检测偏倚Theexposurepromotedmorecarefulevaluationfortheeofinterest45DefinitionofcommoninformationBias-1Interviewerbias访问者偏倚Systematicdifferenceinsoliciting,recording,interpretinginformation.Reportingbias报告偏倚Exposuremaybeunder-reportedbecauseofattitudes,perceptions,orbeliefs46MothersofchildrenwithmalformationswillrememberpastexposuresbetterthanmotherswithhealthychildrenRecall
bias回忆偏倚Casesrememberexposuredifferentlythancontrols
Overestimationof“a”
overestimationofOR47Recall
bias回忆偏倚子女患风湿性关节炎与父母患病关联的病例对照研究风湿性关节炎风湿性关节炎父母风湿性关节炎子女病例普通对照OR子女病例同胞对照OR无31111.011201.0一方10745.023172.5双方61613.9633.6(SchullandCobb,1969)48InvestigatormayprobelisteriosiscasesaboutconsumptionofsoftcheeseInterviewer
bias访问者偏倚Investigatoraskscasesandcontrolsdifferentlyaboutexposure
Overestimationof“a”
overestimationofOR49Biased
follow-up偏性随访UnexposedarelessdiagnosedfordiseasethanexposedExampleCohortstudytoinvestigateriskfactorsformesothelioma
间皮瘤DifficulthistologicaldiagnosisHistologistmorelikelytodiagnosespecimenasmesotheliomaifasbestos(石棉)exposurekown50Impactofinformationbias:
MisclassificationMeasurementerrorleadstoassigningwrongexposureorecategory51Impactofinformationbias:
MisclassificationNon-differential无差异错误分类(错分)ErrorinassessingexposureordiseaseissimilarbetweenstudygroupsMeasureofeffecttendstoward1Differential差异错分Errorinassessingexposure(ordisease)differsindifferentstudygroupsMayincreaseordecreasemeasureofeffect52HypotheticalCase-Controlstudy60404060DEˉEDˉOR=60*60/40*40=2.2510010048325268DEˉEDˉ100100PercentExposureMisclassification:20%20%OR=48*64/36*52=1.96OR
Differentialmisclassification53DifferentialmisclassificationHypotheticalCase-Controlstudy60404060DEˉEDˉOR=60*60/40*40=2.2510010057324368DEˉEDˉ100100PercentExposureMisclassification:5%20%OR=48*64/36*52=2.7454Techniquestoreduceinformationbias(1)StandardizemeasurementinstrumentsAdministerinstrumentsequallytocases/controlsorexposed/unexposedUseobjectiveparametersifpossibleUsemultiplesourcesofinformationQuestionnairesDirectmeasurementsRegistriesCaserecords55Techniquestoreduceinformationbias(2)UsemultiplecontrolsUsere-classifiedsubjectstocheckforbiasUsememoryaidsandvalidateexposuresProvidestandardizedtrainingsessionsandprotocols56Techniquestoreduceinformationbias(3)Blinding:Blindparticipantsastostudygoalsandparticipants’classificationstatusBlindresearchersastosubjects’studystatusTrytoensurethatquestionareclearlyunderstoodthroughcarefulwordingandpretesting57Confounding混杂58ConfoundingConfoundingisadistortioninanassociationbetweenastudyexposureanddiseasebroughtaboutbytheinfluenceofextraneousfactors
混杂是由于外来因素的影响导致的暴露与结局联系的扭曲。Latin:confundere–“tomixtogether”
59ConfoundingConfounders混杂因素:extraneousfactorthatbiasestheassociationweareinterestedinarecalledconfounders.混杂因素:扭曲暴露与结局之间联系的外来因素60Confounding暴露因素混杂因素结局?混杂因素特征图(SzkloM&NietoFJ,2007)61混杂因素特征混杂因子必须同疾病有关联(作为病因或病因标志物,但不是疾病的结果),且该关联独立于暴露因素;混杂因子必须同暴露有关联;混杂因子既不能是暴露的效应,也不能是疾病结果,尤其不能是暴露与疾病之间的中间变量。6262Example1:Case-control:OC–breastCa;Covariate:SmokingSocialstatusOCUsers
NonusersORDNon-DDNon-DLow2550751501Middle505050501High1204030101Total1951401552101.89ConfoundingoccurredbecausesmokingispositivelyassociatedwithOCandbreastCa63常见混杂因素表现形式年龄分布地区死亡a?c孕妇年龄产次新生儿畸形?社会经济地位其他饮食、生活方式因素维生素C摄入量结肠癌?d总体健康状况性功能死亡?b64Avariablecannotbeaconfounderifitisastepinthecausalchainorpathway.ExampleModeratealcoholconsumption
高脂高热能饮食肥胖TC
TG
LDLHDL
FG
Insulin
冠心病糖尿病高血压低体力活动65Question?
Whichvariablesarepotentialconfounders?Howdoyouknowifavariableisaconfounderinyourdataornot?
Usually,riskfactorsfordisease.
Comparecrudeandadjustedmeasuresofassociation.Iftheydifferappreciably,theansweris“yes.”66ControlofconfoundingDesignstrategiesRandomizationRestrictionMatchingAnalysisstrategiesStandardizationStratificationMultivariateanalyses67
PartIII:
Validity68ValidityDefinition:thedegreetowhichameasurementorstudyreachesacorrectconclusion.Types:internalvsexternalvalidity69StudyValidityInternalvalidityThedegreetowhichtheobservedresultsofthestudyaretrue.Inferencesarecorrectregardingthepar
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 电视直销协议书
- 董事内部协议书
- 电杆质保协议书
- 继父建房协议书
- 粪肥还田协议书
- 线路租用协议书
- 签订恋爱协议书
- 松江区婚姻纠纷协议书
- 暑假生勤工俭学协议书
- 父亲和儿女签字协议书
- DLT 593-2016 高压开关设备和控制设备
- 6.2《青纱帐-甘蔗林》-【中职专用】高一语文课件(高教版2023·基础模块下册)
- 2022年港澳台联考语文真题
- 高压开关柜实习周记
- 市政工程投资估算编制办法(建标XXXX164号)
- 水性漆涂装线项目环境影响报告表
- 七年级(下)第一章 活动1 网络与社会生活(第一课时)
- (研究生)商业伦理与会计职业道德ppt教学课件(完整版)
- 机床刀具行业报告:以山特维克为鉴
- 高速铁路路基声屏障桩基试桩方案
- 手术质量与安全分析报告模板
评论
0/150
提交评论