...SmithAF,JobeJB,MingayDL.Retrievalfrommemoryofdietaryinformation.ApplCognPsychol1991;5:269–96.KipnisV,CarrollRJ,FreedmanLS,LiL.Implicationsofanewdietarymeasurementerrormodelforestimationofrelativerisk:applicationtofourcalibrationstudies.AmJEpidemiol1999;150:642–51.PotterJD.Vegetables,fruit,andcancer.Lancet2005;366:527–10.VoorripsLE,GoldbohmRA,vanPoppelG,SturmansF,HermusRJJ,vandenBrandtPA.Vegetableandfruitconsumptionandrisksofcolonandrectalcancerinaprospectivecohortstudy.AmJEpidemiol2000;152:1081–92.SchatzkinA,LanzaE,CorleD,etal.Lackofeffectofalow-fat,high-fiberdietontherecurrenceofcolorectaladenomas.NEnglJMed2000;342:1149–55.AlbertsDS,MartinezME,RoeDJ,etal.Lackofeffectofahigh-fibercerealsupplementontherecurrenceofcolorectaladenomas.NEnglJMed2000;342:156–62.KolarAS,PattersonRE,NeuhouserML,etal.Apracticalmethodforcollecting3-dayfoodrecordsinalargecohort.Epidemiology2005;16:579–83.HunterCP,SpiegelmanD,AdamiH,etal.Non-dietaryfactorsasriskfactorsfrombreastcancer,andaseffectmodifiersoftheassociationoffatintakeandriskofbreastcancer.CancerCausesControl1997;8:49–56. IsItTimetoAbandontheFFQ? CancerEpidemiolBiomarkersPrev2005;14(12).December2005 gold-standardcriterionmeasures:urinarynitrogenexcretiontomeasureproteinintakeanddoublylabeledwatertomeasureenergyintake.Thecorrelationsforenergywere0.1forwomenand0.2formen;forprotein,thecorrelationswere0.3forbothmenandwomen.TheseresultsimplythatastudyusinganFFQwouldobserveatruerelativeriskof2.0as1.06forenergyand1.11forprotein.Whereassomenutrients,suchascarotenoidsorcalcium,areprobablybettermeasuredthanproteinorenergy,theseresultssuggestthateventhelargestcohortstudiesareunlikelytodetectmodestassociationswhenusinganFFQfordietaryassessment.ThethirddevelopmentwasthepublicationfromacohortstudyinwhichbothfoodrecordsandFFQswerecollected(4).Inthisstudy,therewasastatisticallysignificantassociationofdietaryfatwithbreastcancerriskbasedonthefoodrecords,butnotbasedontheFFQ.Asoon-to-bepublishedarticlefromasecondstudywillconfirmthisimportantfinding.Theevidenceismountingthatmuchoftheinconsistency,andsomeofthenullresults,instudiesofdietandcancerareduetopoordietaryassessment.Oneincontrovertibleconclusionisthatweneednewstrategiesfordietaryassessmentthatcanbepracticallyincorporatedintolargecohortstudies.FourProposalsforImprovingDietaryAssessmentinLargeCohortStudiesSomestrategiesfornewresearchornewepidemiologicpracticetakeadvantageofrecentlyavailabletechnologiesorarebasedonnewthinkingaboutnutritionalassessmentdogma.Thesestrategiesarebynomeansexhaustive.Wehopetheyillustratethattherearemanyopportunitiestoimprovedietaryassessmentandthattheseopportunitieswillgrowascomputerizedsystemsforcapturinginformationbecomesmaller,cheaper,morepowerful,andeasilyintegratedintowirelesscommunicationnetworks.ImproveFoodFrequency-TypeMeasures.Foodfrequenciesarelimitedintheirabilitytocollectcomplexinformationduetopracticalrestrictionsinherentinprintedquestionnaireformats.CurrentFFQsrequirecategorizedresponseswhichmaynotcaptureimportantvariabilityinuseofenergy-ornutrient-densefoods.Forexample,carbonatedbeverageportionsizesof‘‘small,’’‘‘medium,’’and‘‘large’’donotcapturethecurrentmarketplacerangeof8to36oz.servings.Complexskipalgorithmsarealsonotfeasibleforprintedquestionnairesbutarelogicalwhenmultipledetailsaboutfoodpurchasingandpreparationarerequiredtoproperlycharacterizeafood.Oneexampleistheuseandcompositionofmultiplemixeddishes,characteristicofAsiandietarypatterns,whichcannotbereadilycapturedusingcurrentFFQs.Finally,althoughitmakessensethatpicturesoffoodswouldmakeiteasiertoreportportionsizes,itisnotfeasibletoembedmultipleprintedportion-sizepicturesintotheresponseoptionsforeachfood.Computer-administeredquestionnaires,deliveredviainternetorontouch-screentabletcomputers,canaddresseachoftheseproblems.WedonotknowifusingnewtechnologytodesignanddeliverFFQswouldimprovethevalidityofdietaryassessment,butitwouldbeworthwhiletofindout.MeasureDietaryBehavior,NotJustNutrients.Inadditiontofocusingattentionontryingtomeasurenutrients,wecouldalsoformulatehypothesesintermsofdietarybehaviors.Questionsaboutusualdietarypractices(e.g.,‘‘Whenyouatebread,howoftenwasitwholewheatorotherwholegrainbread?’’)maybemoreeasilyandaccuratelyrecalledthanthefrequenciesandportionsizesofalonglistoffoods.Usingthisapproachalonetomeasuringdiet,studyhypothesesrequiringinformationonnutrientintakewouldbelimitedtothosethatcouldbeassessedusinganobjectivebiomarker.CollectReal-timeFood-UseInformationUsingComputer-AidedTechnologies.Therearemanyopportunitiesfortechnologicalsolutionstoassistbothincapturingandanalyzinginformationoncurrent,actualfooduse.Studyparticipantscoulduseadigitalphonewithanembeddedcameratotransmitpicturesanddescriptionsoffoodseatenonameal-by-mealbasis.Acomputer-administered24-hourrecallcouldbedeliveredovertheinternetoronapocketPC.Distributingsmall,inexpensivecomputerstostudyparticipantswouldbefarlesscostlyandcarrylessparticipantburdenthanadministeringrepeated24-hourCollectMultiple-DayFoodRecordsbutAnalyzeThemasforaCase-CohortorNestedCase-ControlStudy.dogmaisthatstudyparticipantsmustbetrainedtocompletefoodrecordsandthatrecordsmustbedocumentedorreviewedwiththeparticipantaftercompletiontoensurethatdescriptionsofeachfoodarecomplete.Itmaybe,however,thatanundocumentedfoodrecordisgoodenoughandthat,byfollowingasetofcodingrulesformissinginformation,evenanimperfectlymaintainedfoodrecordcouldbeanalyzedfornutrientintake.Ifthisweretrue,thenpartic-ipantscouldcompletemultiple-dayfoodrecords,whichcouldthenbestoredforlaterretrievalandanalysis.Weexaminedthishypothesisinapilotstudyandfoundthatcorrelationsbetweendocumentedandundocumented3-dayfoodrecordsrangedbetween0.87and1.0(14).Thehighcostoffoodrecordsisattributabletothedocumentationprocess,coding,anddataentryforanalysis.Bylimitingthenumberofrecordsanalyzedtothosethatareinformativeforacase-cohortornestedcase-controlstudy,foodrecordswouldcostlittlemorethanFFQs.Researchonwaystoenhancethequalityofundocumentedrecordswouldbewellmotivated.ConcludingRemarksWeshouldbeverycircumspectaboutanalysesofcurrentstudiesthathaveusedFFQsfordietaryassessment.AnalysesofthesestudieswillnodoubtcontinueasplannedandeffortssuchasthePoolingProjectofDietandCancer(15)mayyieldimportantfindings.However,wearenotlikelytolearnmuchmoreaboutdietandcancerriskbycontinuingtousestandardfoodfrequencyquestionnaires.Weneed,onceagain,toadoptacuriousandexploratoryattitudeaboutdietaryassessment.LargecohortstudiesarebeinginitiatedinAsiancountriesandtheyhaveauniqueopportunitytodevelopandevaluatedifferentapproachesfordietaryassessment.CohortstudiescurrentlyunderwayintheUnitedStatesandEuropecouldchangetheirmethodsfordietaryassessmentwhennextsurveyingtheircohortstoupdateexposureinformation.Whenitcomestodietaryassessment,weneedmorethoughtforfood.ReferencesByersT.Foodfrequencydietaryassessment:howbadisgoodenough?AmJEpidemiol2001;154:1087–8.SchatzkinA,KipnisV,CarrollRJ,etal.Acomparisonofafoodfrequencyquestionnairewitha24-hourrecallforuseinanepidemiologicalcohortstudy:resultsfromthebiomarker-basedObservingProteinandEnergyNutrition(OPEN)Study.IntJEpidemiol2003;32:1054–62.FowkeJH,ChungF-L,JinF,etal.Urinaryisothiocyanatelevels,brassica,andhumanbreastcancer.CancerRes2003;63:3980–6.BinghamS,LubenR,WelchA,WarehamN,KhawKT,DayN.Areimprecisemethodsobscuringarelationbetweenfatandbreastcancer?Lancet2003;362:212–4.DollR,PetoR.Thecausesofcancer:quantitativeestimatesofavoidablerisksofcancerintheUnitedStatestoday.JNatlCancerInst1981;66:1193–308.BurkeB.Thedietaryhistoryasatoolinresearch.JAmDietAssoc1947;23:1041–6.StefanikP,TrulsonM.Determiningthefrequencyoffoodsinlargegroupstudies.AmJClinNutr1962;11:335–43. CancerEpidemiology,Biomarkers&Prevention CancerEpidemiolBiomarkersPrev2005;14(12).December2005 IsItTimetoAbandontheFoodFrequencyQuestionnaire?AlanR.Kristal,UlrikePeters,andJohnD.PotterCancerPreventionProgram,FredHutchinsonCancerResearchCenter,Seattle,WashingtonAlthoughpainfultoadmit,itispossiblethatepidemiologistshavebeendeludedintheiracceptanceoffoodfrequencyquestionnaires(FFQ)asthestandardtoolfordietaryassess-mentinlargestudiesofdietandcancer.ThesubstantiallimitationsofFFQshavebeenknownforsometime(1)andpublishedstudiesbasedonFFQ-deriveddatahavelongincludedintheirdiscussionsectionsalitanyofweaknessesduetosuboptimaldietaryassessment.However,fewofusexpectedtheastonishinglypoormeasurementcharacteristics...