文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 工程科技 > 电子/电路 > How to Use SAS for Logistic Regression with Correlated Data
侵权投诉

How to Use SAS for Logistic Regression with Correlated Data

sas af

Paper261-27

HowtoUseSAS forLogisticRegressionwithCorrelatedDataOliver Kuss, Institute of Medical Epidemiology, Biostatistics, and Informatics,

Halle/Saale,Germany

ABSTRACT

Manystudydesignsinappliedsciencesgiverisetocorrelateddata.Forexample,subjectsarefollowedovertime,arerepeatedlytreatedunderdifferentexperimentalconditions,orareobservedinlogicalunits(e.g.clinics,families,litters).

StatisticalmethodsforregressionanalysisforthiskindofdatawithcontinuousresponsesarequiteestablishedandtheSAS systemoffersavarietyofprocedures(GLMprocedure,MIXEDprocedure)foranalysis.Fordiscreteresponses,however,wehavetofaceagreatermathematicalcomplexityandstatisticalanalysisisnotthatstraightforwardanylonger.

WeshowthedifferentoptionsthattheSAS systemoffersfortheanalysisofbinaryresponseswithcorrelateddata(GENMODprocedure,%GLIMMIX,%NLINMIX,NLMIXEDprocedure,PHREG/LOGISTICprocedureandMeta-Analyticmethods),investigatetheirstatisticalproperties,andillustratethembyanexampleofamulticenterstudy.

Weconcludethatitisdifficulttogivegeneralrecommendationswhichofthemethodstousebecausethisdependsonthedataathandandonthedesiredinterpretationofparameters,butinourdatasetwefeelmostcomfortablewiththeresultsfromtheNLMIXEDandthePHREG/LOGISTICprocedure.

INTRODUCTION

Thelogisticregressionmodelhasbecomethestandardanalyzingtoolforbinaryresponsesinavarietyofdisciplines.Reasonsforthisare:easeofinterpretationofparametersasadjustedoddsratios,possibilityofcalculatingprognosesfortheeventofinterest,andavailabilityofstandardsoftware.TheLOGISTICprocedureisthestandardtoolinSAS softwareforfittinglogisticregressionmodels,butsolutionswiththeGENMOD,thePROBIT,ortheCATMODprocedurearealsopossible.

Acrucialpointinstandardlogisticregressionanalysisisthatobservationsareindependentofoneanotheranditisknownthatviolationsofthisassumptionresultininvalidstatisticalinference.However,manystudydesignsinappliedsciencesgiverisetocorrelateddata.Forexample,subjectsarefollowedovertimeandresponsesareassessedatdifferenttimepoints,arerepeatedlytreatedunderdifferentexperimentalconditions,orareobservedinlogicalunits(e.g.clinics,families,litters).

InthefollowingweillustratethedifferentoptionsthatSAS softwareoffersfortheanalysisofbinaryresponseswithcorrelateddata(GENMODprocedure,%GLIMMIX,%NLINMIX,NLMIXED

procedure).Weshowthateachofthemestimatesparametersfromoneoftwodifferentstatisticalmodelsandcommentontheinterpretationofparametersandthestatisticalpropertiesofthemethodsinvolved.Themodelsandtheestimationproceduresareillustratedbyanexampleofamulticenterrandomizedcontrolledclinicaltrial.

THEDATA

ThedatasetthatweusetoexplainthevariousprocedureswasoriginallyreportedbyBeitler/Landis,1985,andwasalsousedbyWolfinger,1999,tointroducethenewNLMIXEDprocedure.Thedatahavebeencollectedinamulticenterrandomizedcontrolledclinicaltrialconductedineightdifferentclinics.Thepurposeofthestudywastoassesstheeffectofatopicalcreamtreatment

comparedtonotreatmentoncuringnonspecificinfections.Ineachoftheeightclinics(clinic),thenumberoftreated(n)andthe

numberofsuccessfullycuredpersons(x)wererecordedfor

treatment(treatment=1)andcontrol(treatment=0).TheSAS datasetisthefollowing:

datainfection;

inputclinictreatmentxn;datalines;111136101037211620202232311419307194121640117516175001261111600107115701981468067

run;

ACRUDEANALYSIS

Afirstpreliminaryanalysistoassessthetreatmenteffectwillmaybeignorethefactthatthedatawereobservedindifferentclinicsandcollapsethedatainasingle2x2-table.AsuitablemeasurefortreatmenteffectinthiscaseistheoddsratioandtheFREQ

procedurecanbeusedtoachieveanestimator.Somesmalldatamanipulation(actuallywritingeachpersonwithitsresponse(cure)inasingleline)hastoprecedetheanalysis(thevariablestatuswillbeneededinalateranalysis):

datainfection2(drop=i);

setinfection;doi=1ton;

ifi<=xthencure=1;ifi>xthencure=0;status=2-cure;output;end;

run;

procfreqdata=infection2;

tablestreatment*cure/relrisk;

run;

Thisanalysisyieldsanestimatedoddsratio[with95%-CI]of1.498[0.915;2.452],thatmeanstheoddsforcuringtheinfectionis

roughly50%higherinthetreatmentcomparedtothecontrolgroup.However,

How to Use SAS for Logistic Regression with Correlated Data

proceedingthiswaycompletelyignoresthefactthatthestudywasundertakeninseveralclinicsandwemightsuspectthatdifferentfeaturesofthem(personnel,environment,typicalpopulationofpatients)mightinfluencethetreatmenteffectintheindividual

clinic.Notethatthisalsoimpliesthatthereisacorrelationofpatientswithinindividualclinics.Inaclinicwithhightreatmenteffectmanypatientswillbecuredandacureinasinglepatientwillbe

accompaniedwithahigherprobabilitybyacureofanotherpatientfromthesameclinic.

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email