You are on page 1of 9

FindaDoctor|Newsroom|Careers|DirectionsandLocations

Search

Ways to Give|AboutUs|Departments|ContactUs|Mobile ForPatientsandFamiliesYourChild'sHealthClinicalServices|ForHealthCareProfessionalsMedicalEducationMedicalResearch

Home>Stats

Stats Kaplan Meier (June 27, 2000) DearProfessorMean,WhenIreadmymedicaljournals,Ikeeponcomingacrosstermslike"Kaplan-MeierProduct Limitestimate"or"Kaplan-Meiersurvivalcurve."Whatdothesetermsmeanandwhenaretheyused? Oftenwewanttomeasurehowlongittakesforsomethingtooccur.Themostcommon(andthemost morbid)exampleishowlongittakesforsomeonetodie.Forthisoutcome,wewanttoestimatethe fractionofpatientswhosurviveforatleastonemonth,atleastthreemonths,etc.Thisestimateis knownasasurvivalcurve.

Thetermsurvivalissometimesmisleading,becausewecanuseitforotherlesssevereoutcomeslike howlonguntilacancerrelapse,orhowlonguntilaninfectionoccurs.Sometimesitcanevenbeused forapositiveoutcome,likehowlongittakesforacoupletoconceive.Butfortherestofthisexample, we'llkeepthingssimplebyassumingthattheoutcomeistimeuntildeath.

Estimatingasurvivalcurveisoftencomplicatedbytheuncooperativewayinwhichresearchsubjects sometimesbehave.Forexample,somesubjectsdecidetoleaveastudypartofthewaythrough. Othersrefusetodiebeforethestudyends.Welabeltheseuncooperativesubjectsascensored observations.Theysurvivedforatleastthreemonths,butthenwelosttouchwiththem.Orthey survivedatleastthreeyears,butthenwehadtoterminatethestudy. Short explanation TheKaplan-Meierestimateisasimplewayofcomputingthesurvivalcurveinspiteofallthese troublesomeresearchsubjects.Itinvolvescomputingthenumberofpeoplewhodiedatacertaintime point,dividedbythenumberofpeoplewhowerestillinthestudyatthattime.Wemultiplythese probabilitiesbyanyearliercomputedprobabilities,whichisonereasonthisiscalleda"productlimit estimate."

TheKaplan-Meiersurvivalcurveisoftenillustratedgraphically.Itlookslikeapoorlydesignedstaircase, withverticalstepsdownwardatthetimeofdeathofeachindividualsubject.

Oftenwewillcomparecurvesfortwodifferentgroupsofsubjects.Forexample,whatisthesurvival patternforsubjectsonastandardtherapycomparedtoanewertherapy.Wecanlookforgapsinthese curvesinahorizontalorverticaldirection.Averticalgapmeansthatataspecifictimepoint,onegroup hadagreaterfractionofsubjectssurviving.Ahorizontalgapmeansthatittooklongerforonegroupto experienceacertainfractionofdeaths. More details Tocomputeasurvivalcurve,youneedtonotethetimeofoccurrenceofevents(e.g.,failures,deaths)

Itispossiblefortwoormoreeventstooccuratthesametime,inwhichcasethenumberofdistinct timesislessthanthenumberofdeathsorfailures.Youneedtoplacethet'sinorderfromsmallestto largest.Thatis,

Youalsoneedtodefinethestartingpointofthestudy,

ThebasiccomputationsfortheKaplan-Meiersurvivalcurverelyonthecomputationofconditional survivalprobabilities.Inparticular,theprobability

whichcanbeinterpretedastheprobabilityofyoursurvivingtoaspecifictime,giventhatyousurvived totheprevioustime.Thisprobabilityiseasytocalculateifyouknowthenumberofdeathsorfailures ataspecifictimeandifyouknowthenumberofpatientsatriskatthatsametime. Amoredifficult(butmoreimportant)probabilityistheunconditionalprobabilityofsurvival,

whichrepresentsthesimpleprobabilityofsurvivaltoaspecifictime.Youcanusearelationship betweenthisunconditionalprobabilityandtheconditionalprobability:

Atfirstglance,thisdoesnotseemtohelp,becausetherighthandsideoftheequationstillincludesan unconditionalprobability.Butwecanapplythisapproachagaintoget

andwecancontinuealongtheselinestoget

Thislastprobabilityrepresentstheprobabilityofsurvivingatthestartofthestudy.Unlesswe intentionallyrecruitdeadsubjects,thisprobabilityhastobe1.Therefore,theunconditionalprobability isequaltothecumulativeproductofconditionalprobabilities. Ateachtimepoint,youshouldcount

Youshouldalsocount

Armedwiththisinformation,youcannowcomputeaKaplan-Meiersurvivalcurve.Firstyouneedto calculatethenumberofpatientsatrisk,

Inotherwords,thenumberatriskatanyspecifictimepointisjustthenumberatriskattheprevious timepoint,minusthenumberofdeaths/failuresandthenumberofcensoredobservations.For convenience,wedefine

Nextyoucomputetheconditionalprobabilityofsurvival:

Finally,theunconditionalprobabilityofsurvivalissimplythecumulativeproductoftheconditional probabilities:

Example ThefollowingexampleisfromChadhaetal(2000).Theauthorsstudiedasampleof36pediatric patientsundergoingacuteperitonealdialysisthroughCookCatheters.Theywishedtoexaminehow longthesecathetersperformedproperly.Theynotedthedateofcomplication(eitherocclusion, leakage,exit-siteinfection,orperitonitis). Halfofthesubjectshadnocomplicationsbeforethecatheterwasremoved.Reasonsforremovalofthe catheterinthisgroupofpatientswerethatthepatientrecovered(n=4),thepatientdied(n=9),orthe catheterwaschangedtoadifferenttypeelectively(n=5).Ifthecatheterwasremovedpriorto complications,thatrepresentedacensoredobservation,becausetheyknewthatthecatheterstayed complicationfreeatleastuntilthetimeofremoval.

Figure3.1Failuresandcensoredobservationsforcatheterstudy.

Thetableaboveliststhedaysatwhichfailuresand/orcensoredobservationsoccurred.

Figure3.2Computationofnumberofpatientsatrisk

TocomputeaKaplan-Meiersurvivalcurve,youfirstneedtocomputethenumberofcathetersatrisk oneachday.Thisisjustthenumberofcathetersthatwerenotpreviouslycensoredorfailures.These calculationsappearinthetableshownabove.

Figure3.3Compuationofconditionalprobabilityofsurvival.

Nextyouneedtocomputetheconditionalprobabilityofsurvival.Thisistheprobabiltythatacatheter willsurviveataspecifictimepoint,giventhatitsurvived(andwasnotcensored)atanyprevioustime point.Thesecalculationsappearinthetableshownabove.

Figure3.4Computationofunconditionalsurvivalprobabilities.

Finally,youneedtocomputethecumulativeproduct:theproductofeachconditionalprobabilitywith allpreviousconditionalprobabilities.Thisprovidestheestimatesofsurvivalprobabilityusedinthe Kaplan-Meiercurve.Thesecalculationsappearinthetableshownabove.

Figure3.5Graphofunconditionalsurvivalprobabilities(Kaplan-Meiercurve).

ThegraphyouseeaboveistheKaplan-MeiercurveascomputedbySPSS.SelectANALYZE|SURVIVAL |KAPLAN-MEIERfromthemenutogetthisgraph.

Figure3.6SPSSdialogboxforKaplan-Meierprocedure.[Imageisalreadyfullsize]

ThefigureaboveshowstheSPSSdialogbox.Thedateoftheevent(eitherfailureorcensoring)goesin theTIMEfield.IntheSTATUSfield,youshouldplacethevariablewhichindicateswhethertheevent wasafailureoracensoredobservation.ClickontheDEFINEEVENTbuttontotellSPSSwhatcodesyou used.

Figure3.7SPSSdialogboxfordefiningevents.[Imageisalreadyfullsize]

ThefigureshownaboveistheSPSSdialogboxwhereyoudistinguishbetweenfailuresandcensoring. Inthisdataset,avalueof1indicatesafailureand0representscensoring.

Figure3.8SPSSdialogboxforKaplan-Meieroptions.[Imageisalreadyfullsize]

AlsobesuretoclickontheOPTIONSbuttoninthemaindialogbox.Thefigureaboveshowsyouthe dialogboxyouseewhenyouclickontheOPTIONSbutton.BesurethattheSURVIVALPLOTSoptionis checked. Reference Tenckhoff Catheters Prove Superior to Cook Catheters in Pediatric Acute Peritoneal Dialysis. ChadaV,WaradyBA,BloweyDL,SimckesAM,AlonUS. AmericanJournalofKidneyDiseases(2000),35(6):1111-1116. Further reading TherearemanybeginninglevelbooksonbiostatisticsthatdiscusstheKaplan-Meiercurve,suchas Woolson'sbook.YoucanfindamoreadvancedanddetailedapproachinCollett'sbook.
1. Modelling Survival Data in Medical Research. CollettD. LondonEngland:ChapmanandHall(1994). ISBN:0-412-44890-4. 2. Statistical Methods for the Analysis of Biomedical Data WoolsonRF. NewYorkNY:JohnWiley&Sons,Inc.(1987). ISBN:0-471-80615-3.

HereissomeextramaterialthatIneedtointegrateintotheabovedescription.

Survivalprobabilitiesinvolvetheestimationofthetimetosomeevent.Usually,theeventinvolves deathorfailureofsomesort.Someofthepatientsmaynotexperiencetheevent,becausethestudy endsbeforetheydie,orwelosetouchwiththempartwaythroughthestudy.Forthesepatientswe havepartialinformation,weknowthattheeventoccurred(orwilloccur)sometimeafterthedateof lastfollow-up.Werefertothesepatientsascensoredobservations.Wedon'twanttoignorethese patients,becausetheyprovidesomeinformationaboutsurvival,butweneedtohandlethem differently. Thefirststepinasurvivaldataanalysisistoestimatesurvivalprobabilitiesforeachgroup.Whenwe knowtheexactdateofdeath(orfailure)foreachpatient,thiscomputationistrivial.Inmost situations,however,wewillhavepartialinformationonsomeofthepatients.Wewillknowthatthey survivedbeyondacertainpoint,butbecausethestudyendedbeforeallthepatientsdied,orbecause welosttouchwithsomeofthepatients,orbecausetheywithdrewfromthestudy,wedonotknowthe exactdateofdeath.Thesepatientsrepresentcensoredobservations,observationsthatyouhaveto accountfordifferentlythanothers. Asimpleexampleofcensoreddatainvolvesfailureofadevice,andnotthedeathofaperson.Ina studyofcathetersforperitonealdialysis,thesecatheterscanfailduetoocclusion,leakage,or infection.Somecathetersareremovedpriortofailure,usuallyeitherbecausethepatientcompleted dialysisorthepatientdied.Ifthecatheterisremovedpriortofailure,thatisconsideredacensored observation. Catheters Day removed priorto failure 1 2 3 4 5 6 7 10 12 13 8 2 1 1 5 2 2 1 1 3 2 1 2 2 1 Catheters failed

Ifyouwantedtoestimatetheprobabilitythatacatheterwillsurviveitsfirstday,that'seasy.There were34catheters,2didnotsurvivethefirstday,15failedondays2-13.For17ofthecatheters,we didnotknowwhentheywouldhavefailed,butwedoknowthattheyallsurvivedatleastoneday. Sotheprobabilityofsurvivingthefirstdayis32/34=94%. Buthowwouldweestimatetheprobabilityofsurvivingtwodays?fourdays?tendays? Thisistricky,becausethecensoredobservationsprovideinformationuptothedayofcensoring,but cannottellusanythingmoreaboutsurvivingbeyondthatday.Whatweneedtodoiscomputethe numberofcathetersatriskoneachday.Thisisthenumberofcathetersthatwouldbeatriskfor failureonthatday.Itwouldexcludeanycathetersthatfailedonpreviousdaysanditwouldexclude anycathetersthatwerecensoredonpreviousdays. Catheters Day removed priorto failure 1 2 3 4 5 6 7 10 12 13 8 2 1 1 5 2 2 1 1 3 2 1 2 2 1 34 34-8-2=24 24-2-2=20 20-1-1=18 18-1-1=16 16-5-3=8 8-2=6 6-1=5 5-2=3 3-2=1 Catheters failed Catheters atrisk

Wethenneedtocomputetheconditionalprobabilityofsurvivingateachtimepointgiventhatthe cathetersurvivedtheprevioustimepoint.Thisconditionalprobabilitywouldbe (numberatrisk-numberoffailures)/(numberatrisk) Catheters Day removed priorto failure 1 2 3 4 5 6 7 10 12 13 8 2 1 1 5 2 2 1 1 3 2 1 2 2 1 34 34-8-2=24 24-2-2=20 20-1-1=18 18-1-1=16 16-5-3=8 8-2=6 6-1=5 5-2=3 3-2=1 32/34 =0.94 22/24 =0.92 19/20 =0.95 17/18 =0.94 13/16 =0.81 6/8 =0.75 5/6 =0.83 3/5 =0.60 1/3 =0.33 0/1 =0.00 Catheters failed Catheters atrisk Conditional probability

Thenwecomputethecumulativeproductoftheseprobabilities.ThisrepresentstheKaplan-Meier estimateofthesurvivalprobability. Catheters Day removed priorto failure 1 2 3 4 5 6 7 10 12 13 8 2 1 1 5 2 2 1 1 3 2 1 2 2 1 34 34-8-2=24 24-2-2=20 20-1-1=18 18-1-1=16 16-5-3=8 8-2=6 6-1=5 5-2=3 3-2=1 32/34 =0.94 22/24 =0.92 19/20 =0.95 17/18 =0.94 13/16 =0.81 6/8 =0.75 5/6 =0.83 3/5 =0.60 1/3 =0.33 0/1 =0.00 0.94 0.94*0.92 =0.86 0.86*0.95 =0.82 0.82*0.94 =0.77 0.77*0.81 =0.62 0.62*0.75 =0.46 0.46*0.83 =0.38 0.38*0.60 =0.23 0.23*0.33 =0.08 0.08*0.00 =0.00 Catheters failed Catheters atrisk Conditional probability Cumulative product

Hereisagraphofthesesurvivalprobabilities.

Theplothasa"stairstep"pattern,becausewedon'tknowthesurvivalprobabilityatfractionaldays (suchas2.5days)andatsomeintegerdays(suchas9days).Byconvention,weestimatethesurvival probabilityforthesevaluesasequalingthesurvivalprobabilityoftheclosestvaluethatisstillsmaller (the2daysurvivalprobabilityfor2.5days,andthe7daysurvivalprobabilityat9days). Noticethattheestimatedmediansurvivaltime(thetimeatwhich50%ofthecatheterssurvived)issix days. Tenckhoff Catheters Prove Superior to Cook Catheters in Pediatric Acute Peritoneal Dialysis. ChadhaV.AmericanJournalofKidneyDiseases2000:35(6);1111-1116. Category:AskProfessorMean,Category:Survivalanalysis

Copyright1995-2012TheChildren'sMercyHospital Legal|PrivacyandTermsofUse|SiteMap

CMH Employees

You might also like