UnivariateAnalysisandTable1MarkDrazner,MD,MScUniversityofTexasSouthwesternMedicalCenterJune11,2004UnivariateAnalysisandTable1RiskFactor

OutcomeExposureExplanatoryvariableIndependentvariableDependentvariableResponseDiseaseTypesofVariablesinYourExperimentRiskFactor OutcomeExp2ObjectiveIstheriskfactorassociatedwiththeoutcome?Isthereapattern?ObjectiveIstheriskfactoras3ValidityandGeneralizabilityInternalValidity

Ifindapattern,isitreal(reproducible)?ChanceBiasConfoundingExternalValidityDoesitapplytootherpatients?GeneralizabilityValidityandGeneralizabilityI4ToolsAgainstInvalidityChance:StatisticsBias:Studydesign(Wasitafairstudy?)Confounding:MultivariableanalysisToolsAgainstInvalidityChance5StatisticalAnalysesUnivariate1RiskFactor 1OutcomeMultivariableMultipleriskfactors1OutcomeStatisticalAnalysesUnivariate6WhichStatisticalTestDoIUse?ItdependsontypeofdatabeinganalyzedTypesofdatainclinicalresearchCategorical:Nominalvs.OrdinalDichotomous>2categoriesContinuousSurvivalPaireddataLimitedtypesofdatameanslimitednumberofstatisticaltestsWhichStatisticalTestDoIUs7HypertensionConcentricLVHDilatedCardiomyopathyParadigmofHypertensiveHeartDisease“TransitiontoFailure”HypertensionConcentricLVHDila8Rameet.al,AJC,2004RetrospectiveStudy:Parkland

EchocardiographyDatabase(1992–1994)

LowEF

NormalEFn=28 n=131(18%)

(82%)159subjectsConcentricLVH(>1.3cm)NormalLVEFFollow-upecho.≥1yearlater4yearf/uRameet.al,AJC,2004Retrospe9Rame,AJC,2004Rame,AJC,200410SomePurposesofTable1ProvidebaselinecharacteristicsWhatpatientswerestudied?Aretheyapplicabletothereader’spatients(generalizability)?Isthestudybelievable–e.g.,authorssaytheystudiedadvancedheartfailure,dopatientshavecharacteristicsofadvancedheartfailure?Whataredifferencesatbaselinebetween2groups(e.g.,case-control)?Randomizedclinicaltrial:Didrandomizationwork?SomePurposesofTable1Provid11HowDidWeMakeTable1?HowDidWeMakeTable1?12MethodsSection“Dataareexpressedasproportions,means±SD,ormedians(25thand75thpercentiles).DifferencesincharacteristicsbetweenpatientswhodidordidnotprogresstoadepressedEFwerecomparedbyFisher’sexacttest,Student’sttestwithequalorunequalvariances,orWilcoxonrank-sumtest,whereappropriate.”Rame,AJC,2004MethodsSection“Dataareexp13WhatTypeofDataare

YouAnalyzing?WhatTypeofDataare

YouAna14DichotomousOutcomeEitherOrDichotomousOutcomeEither15DichotomousRiskFactorsDichotomousRiskFactors16EpidemiologistsLove2x2TablesDisease/OutcomeRiskFactor/Exposure+-+-EpidemiologistsLove2x2Tab17StatisticalTestsforDichotomousRiskFactor+DichotomousOutcomeChi-square(approximation)Fisher’sExactTestStatisticalTestsforDichotom18Chi-SquareObservedeffect(whatwesee)–Expected(undernull)Expected(undernull)Chi-SquareObservedeffect(wha19DiseaseExposure+-+-A BCDn1n2c1 c2X2=N(AD-BC)2n1n2c1c2AorD=patternDiseaseExposure+20DichotomousRiskFactors18/28whodeveloplowEFareAA104/131whodonotdeveloplowEFareAAP=0.14DichotomousRiskFactors18/2821高血壓英文課件Univariate22DichotomousRiskFactorsOnly129subjects(not159)=missingdata10(45%)whodeveloplowEFhadCAD19(18%)whodidnotdeveloplowEFhadCADP=0.01DichotomousRiskFactorsOnly123高血壓英文課件Univariate24LearnaStatisticalPackage!SASSPSSSTATALearnaStatisticalPackage!SA25ContinuousRiskFactorsContinuousRiskFactors26ContinuousRiskFactor

DichotomousOutcomeStudent’sTtestEqualvarianceUnequalvarianceWilcoxonRankSum(Mann-Whitney-U)Ttestvs.WilcoxonRankSum:Aredatanormallydistributed?ContinuousRiskFactor

Dichoto27HowDoYouTellifDataare

NormallyDistributed?LookforcharacteristicsofnormalityIsitbell-shaped(histogram)?Mean=median=modeSymmetryofdata(skewness)Lengthoftails/howcommonarevaluesnearandfarfrommean(kurtosis)StatisticaltestsofnormalityNormalprobabilityplotWilk-ShapirotestHowDoYouTellifDataare

N28NormalDistributionNormalDistribution29LeftskewRightskewHighkurtosisLotoftailsLowkurtosis“Boxy”LeftskewRightskewHighkurtos30高血壓英文課件Univariate31高血壓英文課件Univariate32Table1AseriesofstatisticaltestsDichotomousriskfactors+outcomeChisquareFisher’sexacttestContinuousriskfactorsanddichotomousoutcomeTtestWilcoxonranksumHavenotaddressedcategoricalriskfactorsoroutcomeswith>2levels,survivalanalysis,paireddataTable1Aseriesofstatistical33Confounder3rdfactorthatisassociatedwithexposure/riskfactor(height)Independentdeterminantofoutcome(wealth)Howtohandleinanalysis:StratificationMultivariableanalysisConfounder3rdfactorthatisa34UnivariateAnalysisIt’sastartbut…NeedtomovetoMultivariableAnalysistoknowifyourriskfact