流行病学数据分析简介theodore epi analy 2014.pptx

AnalysisofEpidemiologicDataIntroductionTheodoreR.HolfordYaleSchoolofPublicHealthDepartmentofBiostatistics

ReviewStatisticalinferenceEstimationPointestimatesIntervalestimates—confidenceintervals(CI)Interpretation—Evidenceforrealeffect?DoestheCIincludevaluethatindicatesnoeffect?HypothesistestingIndicationofstrengthofevidenceforaneffectStudydesignAdjustmentforconfounding

IncidenceandPrevalenceIncidencerates(Numbernewcases)/(Totalfollow-uptime)RateoftransitionfromhealthytodiseasedPrevalencerate(Numbercasesatt)/(Populationatt)Casesoccurredbeforetandstilldiseased(notcuredanddidnotdie)Affectedbyincidencerate,survivalandcureIncidencepreferredforetiologicstudiesbecauseonlydependsonhealthtodiseasetransition

AnalysisofEpidemiologicDataAlternativeStudyDesigns

ProportionsDenominatorisnumberofsubjects(nottotalfollow-upormid-yearpopulation)Numerator(numberofcasesisasubsetofthedenominator)StudydesignsthatyieldproportionsasesCross-sectionalCohortorfollow-upstudywitheveryonefollowforsamelengthoftimeCase-control

Typicallayoutofa2x2tableDiseasestatus(j)Factor(i)HealthyIllTotalLevel1n11n12n1+Level2n21n22n2+Totaln+1n+2n++

Cross-sectionalstudydesign

Cross-sectionalstudydesignDiseasestatus(j)Factor(i)HealthyIllTotalLevel1π11π12π1+Level2π21π22π2+Totalπ+1π+21

CohortstudydesignExample:Effectofexposureduringpregnancy Durationisfixed Twolevele,e.g.,lowbirthweightornormalbirthweight

CohortstudydesignDiseasestatus(j)Factor(i)HealthyIllTotalLevel1π1|1π2|11Level2π1|2π2|21

MeasureofassociationOddsRatioofprobabilityofdiseasetoprobabilityhealthyLevel1:π11/π12orπ1|1/π2|1SimilartobettingoddsRatioofoddsusedasameasureofassociationEqualityofproportiondiseasedequivalenttoequalityofoddsOddsratio:OR=1

Cohortandcross-sectionaloddsratios

Case-controlstudydesign

Case-controlstudydesignDiseas