A standard procedure for creating a frailty index | BMC Geriatrics

The index is often expressed as a ratio of deficits present to the total number of deficits considered. For example, if 40 deficits were ... Skiptomaincontent Advertisement SearchallBMCarticles Search DownloadPDF Researcharticle OpenAccess Published:30September2008 Astandardprocedureforcreatingafrailtyindex SamuelDSearle1,ArnoldMitnitski1,2,3,EvelyneAGahbauer4,ThomasMGill4&KennethRockwood1,2,5  BMCGeriatrics volume 8,Article number: 24(2008) Citethisarticle 63kAccesses 1362Citations 143Altmetric Metricsdetails AbstractBackgroundFrailtycanbemeasuredinrelationtotheaccumulationofdeficitsusingafrailtyindex.Afrailtyindexcanbedevelopedfrommostageingdatabases.Ourobjectiveistosystematicallydescribeastandardprocedureforconstructingafrailtyindex.MethodsThisisasecondaryanalysisoftheYalePrecipitatingEventsProjectcohortstudy,basedinNewHavenCT.Non-disabledpeopleaged70yearsorolder(n=754)wereenrolledandre-contactedevery18months.Thedatabaseincludesvariablesonfunction,cognition,co-morbidity,healthattitudesandpracticesandphysicalperformancemeasures.Datacamefromthebaselinecohortandthoseavailableatthefirst18-monthfollow-upassessment.ResultsProceduresforselectinghealthvariablesascandidatedeficitswereappliedtoyield40deficits.Recodingprocedureswereappliedforcategorical,ordinalandintervalvariablessuchthattheycouldbemappedtotheinterval0–1,where0=absenceofadeficit,and1=fullexpressionofthedeficit.Theseindividualdeficitscoreswerecombinedinanindex,where0=nodeficitpresent,and1=all40deficitspresent.Thevaluesoftheindexwerewellfitbyagammadistribution.Betweenthebaselineandfollow-upcohorts,theage-relatedslopeofdeficitaccumulationincreasedfrom0.020(95%confidenceinterval,0.014–0.026)to0.026(0.020–0.032).The99%limittodeficitaccumulationwas0.6inthebaselinecohortand0.7inthefollow-upcohort.MultivariateCoxanalysisshowedthefrailtyindex,ageandsextobesignificantpredictorsofmortality.ConclusionAsystematicprocessforcreatingafrailtyindex,whichrelatesdeficitaccumulationtotheindividualriskofdeath,showedreproduciblepropertiesintheYalePrecipitatingEventsProjectcohortstudy.Thismethodofquantifyingfrailtycanaidourunderstandingoffrailty-relatedhealthcharacteristicsinolderadults. PeerReviewreports BackgroundFrailtyisastateofincreasedvulnerabilitytoadverseoutcomes.Howbesttooperationalizefrailtyiscontroversial[1–3],butonemethodusesafrailtyindex[4].Theprincipleistocountdeficitsinhealth(whichcanbesymptoms,signs,diseases,disabilitiesorlaboratory,radiographicorelectrocardiographicabnormalities)onthegroundsthatthemoredeficitsapersonhas,themorelikelythatpersonistobefrail.Theindexisoftenexpressedasaratioofdeficitspresenttothetotalnumberofdeficitsconsidered.Forexample,if40deficitswereconsidered,and10werepresentinagivenperson,thatperson'sfrailtyindexwouldbe10/40=0.25.Althoughtheideaandapproacharerelativelysimple,theresultsyieldedbythefrailtyindexhavebeenconsistentbetweensurveysevaluatedbyourgroup[4–7]andbyothers[8–11]eventhoughnoteveryfrailtyindexconsidersthesamedeficits,oreventhesamenumberofdeficits.Forexample,acrossseveralfrailtyindexmeasures,peopleaccumulatedeficits,onaverage,atabout0.03/year[4,5].Ineachstudy,thefrailerthepersonis(thehigherthedeficitcount)themorevulnerabletheyaretoadverseoutcomes[5,8,11].Thefrailtyindexisstronglyassociatedwiththeriskofdeath,institutionalizationandworseninghealthstatus,especiallywhenatleast30variablesareincluded[5].Thefrailtyindexalsoshowsaconsistent,sub-maximallimitatabout2/3ofthedeficitsthatareconsidered.Forexample,ifafrailtyindexiscomposedof60items,themostthatanyonewillhavewrongwiththemisnot60,but40[5].Thereproducibilityofthefindingsinrelationtothefrailtyindexisofsomeinterestbecausenoneofthesamplesinwhichthefrailtyindexhasbeenoperationalizedhasconsideredthesamedeficits.Tobeclear,itdoesnotmatterifstudyAconsidered40deficitsfromsetXofdeficitsandstudyBconsidered60deficitsfromsetYofdeficits;theestimatesfromeach(e.g.therateofdeficitaccumulation,therelationshipbetweendeficitaccumulationandmortality,orthelimittodeficitaccumulation)appeartobesimilar.Thisfindingsuggeststhatfrailtyisarealphenomenon,whichisapropertyofabiologicallycomplexsystem.Itindicatesthatfrailtycanbemeasuredinmanyways,andthereforecanbestudiedinmanyexistingdatasetsthatmightnothavesetouttomeasurefrailtyperse.Toencouragemorewidespreadevaluationoffrailty–agoalencouragedbymanygroups[12–14]–wepresentadetailed,step-by-stepproceduretodescribewhichpotentialvariablescanbeincludedinafrailtyindex,andhowtoestablishcut-pointsforcontinuousvariables.Here,frailtyindexeswerenewlycreatedusingbaselineandfollow-upsamplesfromanexistingcohortstudy,andtheirproperties(e.g.rateofincrease,limit,andrelationshipwithmortality)werecomparedwitheachother,andwithearlierwork.MethodsTheStudySampleTheYalePrecipitatingEventsProject(PEP)isacohortstudybasedinNewHavenCTthatenrolledindividualsaged70yearsorolder.Itsmethodshavebeenpublishedelsewhere[15,16].Briefly,754communitydwelling,Englishspeaking,non-disabledpersonswithlifeexpectancyandplanstostayintheareaformorethan12monthswereenrolledinthestudy.Comprehensivehome-basedassessmentswerecompletedatbaselineandevery18months.The18-monthassessmentincluded681participantsaged72to98years.Thisreportusesthebaselineand18monthfollowupdatatocontrastwitheachotherandtocomparethepropertiesofthenewindexescreatedinthisdatasetwithpreviouslyreportedfrailtyindexes.Atbaseline,mostparticipants(n=487,64.6%)werewomen,andmost(n=682,90.5%)werewhite,withameanMini-MentalStateExamination(MMSE)[17]scoreof26.8(SD=2.50).Mortalitywascheckedmonthlyfornineyearsfromthebaselineinterviewandwasconfirmedbyobituariesanddeathcertificates.SelectingcandidatedeficitsfortheFrailtyIndexAfrailtyindexcountsdeficitsinhealth.Thesedeficitsweredefinedassymptoms,signs,disabilitiesanddiseases[5].Allhealthdeficits,includingcontinuous,ordinalandbinaryvariables,weretakenfromthePEPsurveydatadictionary.Restrictedactivity,disabilityinActivitiesDailyLiving(ADL)andInstrumentalADL,impairmentsingeneralcognitionandphysicalperformance(e.g.impairedgripstrength,impairedwalking),co-morbidity,self-ratedhealth,anddepression/moodwereevaluated.Variablescanbeincludedinafrailtyindexiftheysatisfythefollowing5criteria:1)Thevariablesmustbedeficitsassociatedwithhealthstatus.Attributessuchasgrayinghair,whileage-related,areattributesandthereforenotincluded.2)Adeficit'sprevalencemustgenerallyincreasewithage,althoughsomeclearlyage-relatedadverseconditionscandecreaseinprevalenceatveryadvancedagesduetosurvivoreffects.3)Similarly,thechosendeficitsmustnotsaturatetooearly.Forinstance,age-relatedlenschangesresultinginproblemswithaccommodation(presbyopia)arenearlyuniversalbyage55;inotherwords,asavariable,presbyopiasaturatestooearlytobeconsideredasadeficithere.4)Whenconsideringthecandidatedeficitsasagroup,thedeficitsthatmakeupafrailtyindexmustcoverarangeofsystems–ifallvariableswererelatedtocognition,forexample,theresultingindexmightwelldescribechangesincognitionovertime,butwouldbeacognitiveimpairmentindex[18]notafrailtyindex.5)Ifasinglefrailtyindexistobeusedseriallyonthesamepeople,theitemsthatmakeupthefrailtyindexneedtobethesamefromoneiterationtothenext[19].Therequirementtousethesameitemsneednotapplytocomparisonsbetweensamples–i.e.samplesthatusedifferencefrailtyindexesappeartoyieldsimilarresults[5].Deficitsshouldbeaddeduntilthereareatleast30–40totaldeficits.Thereneedstobeaminimumnumberofdeficits.Ingeneral,themorevariablesthatareincludedinafrailtyindex,themorepreciseestimatesbecome.Similarly,estimatesareunstablewhenthenumberofdeficitsissmall–about10orless.Evenso,anindexwith30–40variableshasbeenshowntobesufficientlyaccurateforpredictingadverseoutcomes[6,14].Furthermore,afrailtyindexcanbeconstructedusinginformationthatisreadilyavailableinmosthealthsurveys,andisclinicallytractable–i.e.itusesanamountthatwouldbegatheredinmanyroutinehealthassessmentsofolderadults[5].CodingofindividualvariablesAllbinaryvariableswererecoded,usingtheconventionthat'0'indicatedtheabsenceofthedeficit,and'1'thepresenceofadeficit.Forvariablesthatincludedasingleintermediateresponse(e.g.'sometimes'or'maybe'),weusedanadditionalvalueof'0.5'.Frailtyindexvariablescanalsoaccommodateordinalandcontinuousvariablesasdeficits.Todosorequiresgradingthecontinuumorrankintoascorebetween0(wherenodeficitispresent)and1(wherethedeficitismaximallyexpressedbythegivenvariable).Forsomevariables,thisre-codingisself-evident.ConsiderthewidelyusedSelf-ratedHealthQuestion("Howwouldyourateyourhealth?Excellent,VeryGood,Good,Fair,Poor").Togradethisbetween'0'and'1',eachlowerself-ratingofhealthwascodedtorepresentalargerdeficit("Excellent=0","VeryGood=0.25","Good=0.5","Fair=0.75"and"Poor=1").Similarly,recognizedcut-pointscanbeusedforordinalandcontinuousvariables,suchastherapidwalktest[15].FortheMMSE,werecodeddeficitsaccordingtoseverityofimpairment[20].Weassigneda1forscoreslessthan10,denotingseveredementia,0.75forscores≥10and≤17,denotingmoderatedementia,0.5forscores≥18and≤20,denotingmilddementia,0.25forscores>20and<24,denotingmildcognitiveimpairment(MCI),and0forscores≥24,denotingnocognitiveimpairment[20].Somereadersmightobjectthatascoreof'1'seemssomethingofadiscount(notasufficientlyhighcount)forseveredementia,andthatlosingonly1pointforit,comparedwith0.25pointsforMCIisnotvalidonitsface.Consider,however,thatapersonwithseveredementiaislikelytohavemanymoredeficitsthanapersonwithMCI,e.g.moredisability,poorerphysicalperformance,higherdegreesofbehaviouralproblemsandsoforth.Becausenotallordinalorcontinuousvariableshavepublishedorself-evidentcut-points,additionalworkisrequiredtoestablishtheleastarbitrarycut-pointsforthesevariables.Methodstoaddressthiscanbebroadlycategorizedasthosebasedoncharacteristicsofthedistributionandthosebasedonjudgment(e.g.inrelationtosomeclinicallyrelevanthazard)[21].Here,weemploybothapproaches.Weusedallexistingpreviouslycodeddeficitstoestablishaninterimfrailtyindex,whosepurposewastohelpprovidecut-pointsfortheremainingvariables.Thisinterim/nearlycompletedindexwasthenplottedagainsttheremainingordinalandcontinuousvariablestounderstandwheretheircut-pointsmightbedetermined.Thevalueoftheindividualvariablethatcorrespondedto0.2ontheinterimfrailtyindex,i.e.thevalueofthevariableatwhich,onaveragepeoplehadafrailtyscoreof0.2orhigher,wasdenotedasthatdeficit'scut-point.Thevalue0.2onthefrailtyindexisrecognizedbymultiplefrailtymeasuresasapproachingafrailstate[7,8,22],sothatthismethodmettheconventionofdefiningdeficitcut-points.Inaddition,settingthevalueat,say,0.3seemsunreasonablyhigh,asthisisconsistentlywellintotherangeoffrailty,however,defined(includingbyanincreasedhazard)sowouldbeinsensitive.Greatersensitivityisobtainedatacut-pointof0.1,butwithlessspecificity.AnalysisofBaselineandFollowupCohortsTherateofaccumulationofdeficitswascalculatedbyevaluatingtheslopeofabestfitlogofthefrailtyindexinrelationtoage.Toevaluatetheimpactofagivenvariableonthefrailtyindex,weusedaniterative,re-samplingprocess,similarto"bootstrapping"asdetailedelsewhere[19,23].Weperformed1000iterationswhereeachiterationcalculatedthebaselineandfollowupfrailtyindexesusing80%oftheirvariables,plottedthelogofthesetwofrailtyindicesversusage,andrecordedtheslope.Byanalyzingtherangeoftheslopes,wewereabletocalculate95%confidenceintervals.Toobservetheupperlimitofthefrailtyindex,the99thpercentilesofeachcohort'sfrailtyindexwasplottedagainstage.Aflatteningofthiscurve(i.e.itsapproachtoa0slope)wouldsuggestacommonmaximumtothefrailtyindexateveryage,consistentwithearlierobservations[24].Statisticaldistributionsofthefrailtyindexwerecomparedwiththeoreticalmodels(Goodnessoffitbyleastsquares).Survivalanalysesweredoneusingbi-variateandmultivariateCoxRegressionanalyseswiththefrailtyindexastheindependentvariableandageandgenderascovariatesoneachofthetwosurveywaves.Thesurvivalcalculationswerebasedontheavailablenineyearmortalitydatafromthebaselinesurvey.The18monthfollowupsurvivalcalculationsbasedfromtheavailablesevenandahalfyear(from18monthinterview)mortalitydata.ComparisonoftheFrailtyIndexcalculatedforthePEPStudywithearlierestimatesThetwocalculatedindexes,onefrombaselineandoneatfollow-up,werecomparedwithpreviouslypublishedindexes,toseehowwelleachfitthefollowingcharacteristics:1)TheFrailtyIndexshouldhaveaskeweddensitydistribution(histogram)thatiswellapproximatedbyagammadistribution[4,8,10]2)Therateofdeficitaccumulation(priorestimateis0.03peryear)[4–6];3)Thepresenceofasub-maximal,age-invariantlimittotheFrailtyIndex(priorestimateis~0.67)[5,6,8];and4)AssociationofthemeanvalueoftheFrailtyIndexwithmortality[4–6,8,10,25].EthicsThestudyprotocolwasapprovedbytheYaleHumanInvestigationCommittee,NewHaven;allparticipantsprovidedinformedconsentatbaselineandatfollow-up.EthicalapprovalforsecondaryanalyseswasobtainedfromtheCapitalDistrictHealthAuthority,Halifax,NovaScotia.ResultsConstructionandcharacteristicsoftheFrailtyIndexatbaselineandatfollow-upOfthevariablesconsidered,40variablesthatmetallfrailtyindexcriteriaatbothbaselineandfollow-upwerechosen(Table1).Variableswereeliminatedbecausetheydidnotmeetatleastoneofthefivecriteria(unrelatedtoageandadverseoutcome,saturated,ortherewasalreadyamplerepresentationofthesystem)orbecausewehadidentified40variableswithwhichtopopulatetheindex.Somepotentialvariablesexcludedwere:Distancewalked(upto20ft.)(saturated),admittedtohospitalinthepastyear(non-ageassociated),useofawalkingdevice(sufficientvariables(i.e.n=40)werealreadyincluded),walkingaquartermile(alreadyaccountedforbytwovariables),measuredbloodpressure(sittingandstanding)(non-ageassociated),fractures(non-ageassociated),Parkinson'sDisease(lowprevalence),amputation(non-ageassociated),liverdisease(notpresentinbothsurveys),takingmedication(controversialinrelationtoadversehealthoutcome),light/medium/heavysports(Unreliableprevalenceandageassociation),measuredvision(saturation),andvarioustestsofphysicalperformance(alreadyaccountedforinothervariablesand40variablesalreadypopulatingtheindex),suchasfingertapandturninginacompletecircle.Table1HealthVariablesandCut-pointsfortheFrailtyIndexFullsizetable Ofthe40variablesincludedintheFrailtyIndex,threewerecontinuous,withnoclearcut-pointforinclusion.Thesewerepeakflow,shoulderstrengthandtimedusualpacewalkfor20ft.Thesevariables'deficitsweredeterminedbyplottingthemagainstthefrailtyindex(withoutthevariablesbeingadded)andidentifyingthevaluecorrespondingto0.2(Table2).Ofinterest,whenothercontinuousvariables(gripstrength,timedrapidwalkof20ft.)wereplottedagainstinterimfrailtyindexes,similarcut-pointstotheirpublishedcutoffswerefound(datanotshown).Table2ContinuousVariableCut-pointsFullsizetable ThebaselineandfollowupFrailtyIndexdistributionswerewellcorrelatedtoagammadistribution(Figure1,r2>0.90).Atbaseline,morepeoplehadFrailtyIndexvaluesbetween0–0.15,whereasatfollowup,morepeoplehadhigherFrailtyIndexvalues.Figure1 FrailtyIndexDistribution.Gammadistributionfit(lines)oftheobserveddistributionofthefrailtyindex(bar)inthebaseline(red)and18monthfollowup(blue)sample.Fullsizeimage Inrelationtoage,thebaselineaverageslopeofthedeficitaccumulationlinewas0.020(95%confidenceinterval0.014–0.026);i.e.onaverage,theestimatedmeanrateofdeficitaccumulationwas0.020peryear(Figure2).Forthecohortatfollow-up,theslopeofthelinerelatingdeficitstoagewas0.026(95%confidenceinterval0.020–0.032).Figure2 FrailtyIndexversusAgePlot.Frailtyindexversusageplotofbaseline(lightanddarkred)and18monthfollowup(lightanddarkblue).Shownherearetheaverage(darkblue/red)andtheobserved99thpercentile(lightblue/red)lines.Theslopeofthebestfitcurvesshowsnoaccumulationofdeficitsinthemostimpaired(99th)ofthesample.Bycontrastthefollowupaveragecurvehas2.6%deficitaccumulationperyear.Thebaselineaveragecurvehasa2.0%deficitaccumulationperyear;the99thpercentileslopealsoshowsnoaccumulationofdeficitswithage.Fullsizeimage Ininvestigatingtheupperlimits(99%sample)totheFrailtyIndex,wenotedthatboththebaselineandfollow-upcohortsnolongershowedarelationshipbetweenageanddeficitaccumulation(Figure2).Indeed,thebestfitlineofthe99%samplehasaslopestatisticallyindistinguishablefrom0.Theupperlimitusingthebaselinecohortwasaround0.6,whilethelimitusingthefollowupcohortwasabout0.7(therewerefourindividualswithslightlyhigherfrailtyvalues).Inbothcohorts,theconstructionoftheFrailtyIndexshowedlittlesensitivitytowhichvariableswereincluded(Figure3).Thedifferencesinslopeswerenegligiblewhen80%ofthevariableswerere-sampled;differencesintheinterceptsoftherelationshipbetweenageanddeficitaccumulationweremoreevident,butwithinnon-overlappingconfidenceintervals(Figure3).Figure3 VarianceintheSlopeoftheFrailtyIndex.TheBootstrappingofthefrailtyindex.Thefrailtyindexwascreatedandplotted1000times,eachtimerandomlypicking80%ofthevariablesoftheindex.Twentyiterationsareshownhere.Theexperimentalandbestfitregressionlinesoftheaverageindexvaluesareshowninthebaseline(red)andfollowup(blue).Fullsizeimage MortalityinrelationtotheFrailtyIndexThebaselineandfollowupcohort'sFrailtyIndexeswereeachassociatedwithmortality.InthebivariateCoxregressionanalysis,sex,ageandthefrailtyindexwereeachsignificantpredictorsofsurvivalinthebaselineandfollowupcohort(Table3).Inthemultivariableanalysis,allthesevariablesweresignificantlyrelatedtomortalityatbothbaselineandatfollowup.Table3CoxAnalysesFullsizetable DiscussionInasecondaryanalysisoftheYalePrecipitatingEventsProject,aFrailtyIndexwasconstructedforabaselineandafollow-upcohort,respectively.Eachstepintheprocesswasdescribed,toallowapreciseaccountofwhatconstitutesahealthdeficitforthispurpose,howtoselectwhichhealthdeficitstoincludeinafrailtyindex,howtooperationalizeanypossibledeficit(ordinal,continuousandbinary)toarangeof0–1andwhichcharacteristicsofthefrailtyindex(natureofthedistribution;slopeinrelationtoage;presenceofalimit)seemtobebroadlyreplicable.Severalreproduciblecharacteristics(e.g.thedistribution,theslopeandlimitofdeficitaccumulation)ofeachFrailtyIndexwereprovidedsothattheymaybeused,asinpreviouspapers[4–11],todescribetheoverallfrailtystateofthegroup.ThebaselineFrailtyIndexshowedarateofaccumulationtobe0.020peryear(per1yearincreaseinage)withanupperlimittothefrailtyindexofabout0.60whilethefollowupshowedarateof0.026deficitsaccumulatedperyearwithalimitaround0.70(Figure2).Weusedare-samplingbyvariableproceduretoconstructconfidenceintervalsfortheslopesofthelines(Figure3).Thisproceduregivesusinformationaboutthefrailtyconstruct,showingthatarangeofdeficitscaninfactbecombinedtogivearesultthatisinformativeintheaggregate.Inotherwords,theslopedependsontheoverallbehaviourofthedeficitaccumulation,andisnotdrivenbyasmallnumberofvariables.Inthisregard,earlierworkhasshownreasonableconsistencyoftherateofdeficitaccumulationacrosscommunity-dwellingrandomsamples[6].Here,wenotedthatthefollowupcohorthadfrailtyindexcharacteristics–frailtyindexvalues,rateandlimitsimilartothoseofpreviouslystudiedcommunitydwellingrandomsamples.Mostnotableisthe0.03accumulationofdeficitsandtheageindependentlimittofrailtyof0.67.Thebaselinesamplehadlowerestimates–aloweraverageFrailtyIndexandalowermaximallimit.Thissuggeststhatthebaselinecohortwasnotasfrailasthefollowupcohort.TherelationshipbetweentheFrailtyIndexandmortalityisofinterestonseveralgrounds,buthereispresentedchieflybecauseitrepresentsarelevantandnon-arbitrarytestofpredictivevalidity.Thisisimportantbecausepredictivevalidityisoneoftwotypesofso-calledcriterionvalidation,theotherbeingvalidationagainstaso-called"goldstandard"[21].Asthereisnogoldstandardforfrailtyassessment,predictivevalidationisanimportantmethodofvalidatinganyapproachtofrailtyoperationalization.Notethatourintentincheckingtheabilityofthefrailtyindextopredictmortalityisvalidationoftheindex,ratherthandevelopingamortalitypredictionindexthatincludedfrailty.Ifthefrailtyindexweremeanttobeamortalitypredictioninstrument,theremightbearationaleforweightingseveralitems,particularlyage[26].OnenotableresultfromtheCoxanalysesisthatincludingtheFrailtyIndexincreasedtheimpactofbeingmaleonmortality.Thislikelyreflectstheobservationfromearlierstudiesthatwhilemenaccumulatefewerdeficitsthandowomen,anygivenlevelofdeficitaccumulationismorelethalforthemandatanygivenage,femalesseemtobemorefrailthanmales[6,11].Therelationshipwithmortalityisalsoimportantinunderstandinghowdeficitaccumulationmightoperate.Classically,Gompertzdescribedtherateofmortalitybeingexponentiallyrelatedtoage[27].Equallyunsurprisingly,mortalityexponentiallyincreaseswiththeaccumulationofdeficits[5,8,19].Inaddition,accelerationofdeficitaccumulationischaracteristicofolderpeoplepriortodeath[8].Ourdatamustbeinterpretedwithcaution.Notallitemshadestablishedcut-points.Inaddition,cut-pointscanbedifficulttoapplyacrossasamplethatcoversmanyages,astheeffectsofcontinuoustraitscanbeage-specific.[28]Ourapproachderivedcut-pointsbasedonthe"interimfrailtyindex"proceduredescribedabove.Inaddition,thesampleissmall,sothatanyindividualestimatescanbeunstable;thisiswhereaggregationofitemsinafrailtyindexcanbehelpful,andwherethere-samplingstrategyisuseful.Ourpaperalsohassomestrengths.Inreplicatingmanyofthecharacteristicsofafrailtyindexinanewsample,wecangiveadditionalassuranceoftherobustnessoftheapproach.Byspellingoutindetailhoweachstepinconstructingafrailtyindexcanbeundertaken,andbysubmittingtoanopenaccessjournal,weareaimingtomakethemethodwidelyavailable.Wehavealsomademorepreciseamethodforestablishingcut-pointsforvariablesthatwerenotconstructedforinclusioninafrailtyindex,therebyfurtherallowingthemethodtobeused.Inthisregard,therelationshipofanygivenvariabletoameanfrailtyindexscoreof0.2mightseemarbitrary.InastudythatrelatedthefrailtyindexapproachtothephenotypicdefinitionoffrailtypopularizedfromtheCardiovascularHealthStudy[22],0.2correspondedtothemeanfrailtyindexvalueforpersonsdefinedas"pre-frail"[7,22].Amorerecentpaperfromanothergroupusedthe0.2cut-pointonaso-called"deficitindex"todistinguishpeoplewhowere"robust"formthosewhowere"pre-frail"[29].Finally,likemanyhealthsurveys,thePEPstudyhasmanymorevariablesthanareneededtoconstructa40-itemfrailtyindex.Severaleligiblevariableswerenotincludedonlybecausewehadreachedourtargetofa40-itemFrailtyIndex.Thereisnoscientificreasonnottoincludemore–wehaveconstructedanfrailtyindexof70items.Ontheotherhand,arecurringconcernaboutthefrailtyindexhasbeenthefeasibilityofcalculatingitifalotofvariablesareused[30].Here,asinsomeearlierstudies,[7,19,31]wehaveselectedvariablesatrandom(boot-strapping)fromalistofeligiblevariablestomakeuptheFrailtyIndexandhaveagainshownthattheresultsareinsensitivetotheprecisecompositionoftheindex.Changeinthehealthstatusofelderlypeopleisanobviousconcerntocliniciansandtopopulationplanners.Inthenextroundofanalyses,wewillbeinterestedtoknowwhetherthechangesinthefrailtystates(baselinefrailtystateversusthefollowupstate)canbedescribedusingasocalled"stochastic"transitionmodel[32]whichwehaveevaluatedinothercommunity-dwellingelderlysamples,althoughnotwithonesthatincludeasmanyperformancemeasuresasthePEPstudy[33].Thisintriguingpossibilityismotivatingfurtherinquiriesbyourgroup.ConclusionAsystematicprocessforcreatingaFrailtyIndexwaspresentedfortheYalePrecipitatingEventsProject,awellstudiedcohortinwhichdeficitaccumulationpreviouslyhadnotbeenevaluated.Theprocessallowsoperationalizationofthefrailtyindextobecarriedoutinotherdatasets.Thefrailtyindexrevealshowfrailty,understoodasavulnerabilitystatewithanincreasedriskofadverseoutcomes,canbequantified.Thismethodofquantifyingfrailtycanaidourunderstandingofhealthandfrailty-relatedhealthcharacteristicsandoutcomesinolderadults. ReferencesBergmanH,FerrucciL,GuralnikJ,HoganDB,HummelS,KarunananthanS,WolfsonC:Frailty:anemergingresearchandclinicalparadigm–issuesandcontroversies.JGerontolABiolSciMedSci.2007,62:731-737.Article  PubMed  PubMedCentral  GoogleScholar  FisherAL:Justwhatdefinesfrailty?.JAmGeriatrSoc.2005,53:2229-2230.10.1111/j.1532-5415.2005.00510.x.Article  PubMed  GoogleScholar  WhitsonHE,PurserJL,CohenHJ:Frailtythynameis...Phrailty?.JGerontolABiolSciMedSci.2007,62(7):728-730.Article  PubMed  PubMedCentral  GoogleScholar  MitnitskiAB,MogilnerAJ,RockwoodK:Accumulationofdeficitsasaproxymeasureofaging.TheScientificWorld.2001,1:323-336.CAS  Article  GoogleScholar  RockwoodK,MitnitskiA:Frailtyinrelationtotheaccumulationofdeficits.JGerontolBiolSciMedSci.2007,62A:722-727.Article  GoogleScholar  MitnitskiA,XiaoweiS,SkoogI,BroeGA,CoxJL,GrunfeldE,RockwoodK:Relativefitnessandfrailtyofelderlymenandwomenindevelopedcountriesandtheirrelationshipwithmortality.JAmGeriatrSoc.2005,53(12):2184-5189.10.1111/j.1532-5415.2005.00506.x.Article  PubMed  GoogleScholar  RockwoodK,AndrewM,MitnitskiA:AComparisonoftwoapproachestomeasuringfrailtyinelderlypeople.JGerontolABiolSciMedSci.2007,62A(No7):738-743.Article  GoogleScholar  KulminskiA,YashinA,ArbeevK,AkushevichI,UkraintsevaS,LandK:Cumulativeindexofhealthdisordersasanindicatorofaging-associatedprocessesintheelderly:ResultsfromanalysesoftheNationalLongTermCareSurvey.MechAgeingDev.2007,128:250-258.10.1016/j.mad.2006.12.004.CAS  Article  PubMed  GoogleScholar  KulminskiA,UkraintsevaSV,AkushevichIV,ArbeevKG,YashinAI:Cumulativeindexofhealthdeficienciesasacharacteristicoflonglife.JAmGeriatrSoc.2007,55(6):935-40.10.1111/j.1532-5415.2007.01155.x.Article  PubMed  PubMedCentral  GoogleScholar  GogginsWB,WooJ,ShamA,HoSC:FrailtyindexasameasureofpersonalbiologicalageinaChinesepopulation.JGerontolABiolSciMedSci.2005,60A:1046-1051.Article  GoogleScholar  WooJ,GogginsW,ShamA,HoSC:Socialdeterminantsoffrailty.Gerontology.2005,51:402-408.10.1159/000088705.CAS  Article  PubMed  GoogleScholar  BergmanH,FerrucciL,GuralnikJ,HoganDB,HummelS,KarunananthanS,WolfsonC:Frailty:anemergingresearchandclinicalparadigm–issuesandcontroversies.JGerontolABiolSciMedSci.2007,62(7):731-737.Article  PubMed  PubMedCentral  GoogleScholar  WalstonJ,HadleyEC,FerrucciL,GuralnikJM,NewmanAB,StudenskiSA,ErshlerWB,HarrisT,Fried:Researchagendaforfrailtyinolderadults:towardabetterunderstandingofphysiologyandetiology:summaryfromtheAmericanGeriatricsSociety/NationalInstituteonAgingResearchConferenceonFrailtyinOlderAdults.JAmGeriatrSoc.2006,54(6):991-1001.10.1111/j.1532-5415.2006.00745.x.Article  PubMed  GoogleScholar  FerrucciL,GuralnikJM,StudenskiS,FriedLP,CutlerGB,WalstonJD:Interventionsonfrailtyworkinggroup.Designingrandomized,controlledtrialsaimedatpreventingordelayingfunctionaldeclineanddisabilityinfrail,olderpersons:aconsensusreport.JAmGeriatrSoc.2004,52(4):625-634.10.1111/j.1532-5415.2004.52174.x.Article  PubMed  GoogleScholar  GillTM,GahbauerEA,AlloreHG,HamL:Transitionsbetweenfrailtystatesamongcommunity-livingolderpersons.ArchInternMed.2006,166:418-423.10.1001/.418.Article  PubMed  GoogleScholar  GillTM,DesaiMM,GahbauerEA,HolfordTR,WilliamsCS:Restrictedactivityamongcommunity-livingolderpersons:incidence,precipitants,andhealthcareutilization.AnnInternMed.2001,135:313-321.CAS  Article  PubMed  GoogleScholar  FolsteinMF,FolsteinSE,McHughPR:"Mini-mentalstate".Apracticalmethodforgradingthecognitivestateofpatientsfortheclinician.JPsychiatrRes.1975,13(3):189-198.10.1016/0022-3956(75)90026-6.Article  GoogleScholar  MitnitskiA,RockwoodK:Transitionsincognitivetestscoresover5and10yearsinelderlypeople:evidenceforamodelofage-relateddeficitaccumulation.BMCGeriatr.2008,8(1):3-10.1186/1471-2318-8-3.Article  PubMed  PubMedCentral  GoogleScholar  RockwoodK,MitnitskiA,SongX,SteenB,SkoogI:Long-termrisksofdeathandinstitutionalizationofelderlypeopleinrelationtodeficitaccumulationatage70.JAmGeriatrSoc.2006,54(6):975-979.10.1111/j.1532-5415.2006.00738.x.Article  PubMed  GoogleScholar  ReisbergB:FunctionalAssessmentStaging(FAST).PsychopharmacolBull.1988,24:653-659.CAS  PubMed  GoogleScholar  StreinerD,NormanG:HealthMeasurementscales:Aguidetotheirdevelopmentanduse.2003,Oxford,OxfordUniversityPress,172-193.3 GoogleScholar  FriedLP,TangenCM,WalstonJ,NewmanAB,HirshC,GottdienerJ,SeemanT,TracyR,KopWJ,BurkeG,McBurnieMA,CardiovascularHealthStudyCollaborativeResearchGroup:Frailtyinolderadults:evidenceforaphenotype.JGerontolABiolSciMedSci.2001,56A:M146-M156.Article  GoogleScholar  EfronB,TibshiraniR:AnIntroductiontothebootstrap.1993,NewYork,Chapman&HallBook  GoogleScholar  RockwoodK,MitnitskiA:Limitstodeficitaccumulationinelderlypeople.MechAgeingDev.2006,127(5):494-6.Article  PubMed  GoogleScholar  RockwoodK,MitnitskiA:frailty,fitnessandthemathematicsofdeficitaccumulation.RevClinGerontol.2007,17:1-12.10.1017/S0959259807002353.Article  GoogleScholar  LeeSJ,LindquistK,SegalMR,CovinskyKE:Developmentandvalidationofaprognosticindexfor4-yearmortalityinolderadults.JAMA.2006,295(7):801-8.10.1001/jama.295.7.801.Erratumin:JAMA2006,295(16):1900CAS  Article  PubMed  GoogleScholar  GompertzB:Onthenatureofthefunctionexpressiveofthelawofhumanmortalityandonanewmodeofdetermininglifecontingencies.PhilosophicalTransactionoftheRoyalSocietyofLondon.1825,SanDiego:AcademicPress GoogleScholar  YashinAI,ArbeevKG,AkushevichI,KulminskiA,AkushevichL,UkraintsevaSV:Stochasticmodelforanalysisoflongitudinaldataonagingandmortality.MathBiosci.2007,208:538-551.10.1016/j.mbs.2006.11.006.Article  PubMed  GoogleScholar  KulminskiAM,UkraintsevaSV,KulminskayaIV,ArbeevKG,LandK,YashinAI:Cumulativedeficitsbettercharacterizesusceptibilitytodeathinelderlypeoplethanphenotypicfrailty:lessonsfromtheCardiovascularHealthStudy.JAmGeriatrSoc.2008,56:898-903.10.1111/j.1532-5415.2008.01656.x.Article  PubMed  PubMedCentral  GoogleScholar  MartinFC,BrightonP:Frailty:differenttoolsfordifferentpurposes?.AgeAgeing.2008,37(2):129-31.10.1093/ageing/afn011.Article  PubMed  GoogleScholar  RockwoodK,AbeysunderaMJ,MitnitskiA:Howshouldwegradefrailtyinnursinghomepatients?.JAmMedDirAssoc.2007,8(9):595-603.10.1016/j.jamda.2007.07.012.Article  PubMed  GoogleScholar  MitnitskiA,BaoL,RockwoodK:Goingfrombadtoworse:astochasticmodeloftransitionsindeficitaccumulation,inrelationtomortality.MechAgeingDev.2006,127:490-493.10.1016/j.mad.2005.09.007.Article  PubMed  GoogleScholar  MitnitskiA,SongX,RockwoodK:Improvementanddeclineinhealthstatusfromlatemiddleage:modelingage-relatedchangesindeficitaccumulation.ExpGerontol.2007,42(11):1109-15.Article  PubMed  GoogleScholar  FlegalKM,GraubardBI,WilliamsonDF,GailMH:Cause-specificexcessdeathsassociatedwithunderweight,overweightandobesity.JAMA.2007,298(17):2028-2037.10.1001/jama.298.17.2028.CAS  Article  PubMed  GoogleScholar  Pre-publicationhistoryThepre-publicationhistoryforthispapercanbeaccessedhere:http://www.biomedcentral.com/1471-2318/8/24/prepub DownloadreferencesAcknowledgementsFundsforsecondaryanalysiscamefromtheCanadianInstitutesofHealthResearch,throughanoperatinggranttoDr.Mitnitski(MOP62823)andbyagrantfromtheFountainInnovationFundoftheQueenElizabethIIHealthSciencesFoundation.Dr.GillistherecipientofaMidcareerInvestigatorAwardinPatient-OrientedResearch(K24AG021507)fromtheNationalInstituteonAging.KennethRockwoodissupportedbytheDalhousieMedicalResearchFoundation,astheKathrynAllenWeldonProfessorofAlzheimerResearch.DatacollectionforthePEPstudyhadbeenfundedbygrantsfromtheRobertWoodJohnsonFoundation,theAmericanFederationforAgingResearch,thePatrickandCatherineWeldonDonaghueMedicalResearchFoundation,andtheNationalInstituteonAging(R37AG17560,R01AG022993).AuthorinformationAffiliationsGeriatricMedicineResearchUnit,DalhousieUniversity&CapitalDistrictHealthAuthority,Halifax,CanadaSamuelDSearle, ArnoldMitnitski & KennethRockwoodDepartmentofMedicine,DalhousieUniversity,Halifax,CanadaArnoldMitnitski & KennethRockwoodDepartmentofMathematics&Statistics,DalhousieUniversity,Halifax,CanadaArnoldMitnitskiDepartmentofInternalMedicine,YaleUniversitySchoolofMedicine,NewHaven,CT,06504,USAEvelyneAGahbauer & ThomasMGillDivisionofGeriatricMedicine,DalhousieUniversity,Halifax,CanadaKennethRockwoodAuthorsSamuelDSearleViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarArnoldMitnitskiViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarEvelyneAGahbauerViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarThomasMGillViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarKennethRockwoodViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarCorrespondingauthorCorrespondenceto KennethRockwood.AdditionalinformationCompetinginterestsNosponsorhadaroleinthedecisiontoundertaketheseanalysesortosubmitthestudyforpublication.Eachauthorassertsnoproprietaryinterestintheresultandnofinancialconflictofinterest.Authors'contributionsSamSearlecarriedouttheanalysesandwrotethefirstdraftaspartofhisPhDprogram.ArnoldMitnitskisupervisedtheseanalyses.EvelyneGahbauerprovidedthedatasets.ThomasGillisthePIofthePEPstudyandcritiquedeachdraft.KennethRockwoodconceivedoftheideawithArnoldMitnitski,withwhomhearrangedfundingandco-wrotethefirstandsubsequentdrafts.Allauthorsreviewedandapprovedthefinaldraftofthepaper.Authors’originalsubmittedfilesforimagesBelowarethelinkstotheauthors’originalsubmittedfilesforimages. Authors’originalfileforfigure1Authors’originalfileforfigure2Authors’originalfileforfigure3Rightsandpermissions OpenAccess ThisarticleispublishedunderlicensetoBioMedCentralLtd.ThisisanOpenAccessarticleisdistributedunderthetermsoftheCreativeCommonsAttributionLicense( https://creativecommons.org/licenses/by/2.0 ),whichpermitsunrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited. ReprintsandPermissionsAboutthisarticleCitethisarticleSearle,S.D.,Mitnitski,A.,Gahbauer,E.A.etal.Astandardprocedureforcreatingafrailtyindex. BMCGeriatr8,24(2008).https://doi.org/10.1186/1471-2318-8-24DownloadcitationReceived:09May2008Accepted:30September2008Published:30September2008DOI:https://doi.org/10.1186/1471-2318-8-24SharethisarticleAnyoneyousharethefollowinglinkwithwillbeabletoreadthiscontent:GetshareablelinkSorry,ashareablelinkisnotcurrentlyavailableforthisarticle.Copytoclipboard ProvidedbytheSpringerNatureSharedItcontent-sharinginitiative KeywordsMildCognitiveImpairmentFrailtyIndexFrailtyStateBaselineCohortHealthDeficit DownloadPDF Advertisement BMCGeriatrics ISSN:1471-2318 Contactus Submissionenquiries:[email protected] Generalenquiries:[email protected]