Showmethedata,notlackofdata

SDTMassumesthatifthereisnorecordthennothinghappened.Thisworksbutonlyifitwascheckedindatacapture,whichrequiresaquestionandrecord(e.g.,Were

thereanyAEs?)

Absenceofevidenceisnotevidenceofabsence:mustcheckthatmissingdatais

missing

SomethinkCDISC’sCDASHdatacapturestandardisunnecessary.Theysayit’sverysimilartoSDTM,andthefewdifferencescreateconfusionandextrawork.CDASHis similartoSDTM,buttheysolvedifferentproblems.Usedtogethertheypositivelyimpactdatacapture,quality,usability,

repurposing,andtraceability.

WeexploredifferencesbetweenCDASHandSDTMandwhyboth standardsarecritical.

CDASHandSDTMareinfactverysimilar.• 67%ofCDASHv2.0mapsdirectlytoSDTMIGvariables,andCDASHv2.0includesmapping• 86%ofCDASHmapsdirectlywithstandardmappingsincluded(e.g.,dates)• 14%aredifferentforareason

SDTMisoptimizedfortabulation,analysisdatasetcreation,&datasubmission.CDASHisoptimizedfordatacapture,investigatorsiteactivities,&dataquality.

Differentrequirements,differentapproaches,butwiththesameendinmind.

Machine-readabledata:- ISO8601Dates/Times:1

variable,YYYY-MM-DDThh:mm:ss

-Duration:P1M3D

SDTMmachine-readableformatsforvariablessuchasdatesaregoodfordatareusabilitybutarenotuser-friendlyfordatacapture.Sitesrecordingdatain

unfamiliarformatsincreasesriskoferrors

Human-readable:-Dates/Times:2ormorevariables,DD-MMM-YYYY,

HH:MM:SS-Duration:1month,3days

Variablesmustbeinorderbydomain;non-standardvariablesarestoredin

differentdatasets(e.g.,FA,SUPP--)

Domain-drivenorganizationiscriticalforstandardtools,butdatamustmakesensetothesite.ThiscanmeantosplitdomainsacrossCRFsandCRFsacrossdomains,and

notsplitcustomandstandardvariables

DatastructureharmonizedwithSDTMbutvariablescanbearrangedtomakedata

captureeasier.

Collectedrelationshipsbetweendataare

representedinRELREC,aseparatedataset

RELRECisbasedoncollecteddata,butdataisnotcapturedlikethat.Enteringlinenumbersintherelateddatasetsissimpler,requiringnoderivations(e.g.,addingAEline#to

relatedconmed)

Linksamongrecordsareexplicit(e.g.,thisAErelatedtothatCM),orimplicit(e.g.,AEseveritychangesgoingintoFA)

indatacollection

Findingsdatamustbeinanormalizedorverticalstructure;answersarealreadyinvariables

Normalizedstructurescanstorenewtestswithoutchangingdatasetstructures,butmostEDCsystemscan’t

dothis;also,differenttestsinadomainmayneeddifferentcontrolledterms(e.g.,differentanswersfordifferent

questionsinasurvey)

Findingsdatamaybehorizontal,lettingeachtesthaveadifferentcodelist;

SDTMCTisusedforvariablenames&CRFprompts

Metadatacentersontabulations,e.g.,variable

labelsandroles

SDTMlabelsidentifytabulationdata.CDASHhasquestiontextsandpromptsdesignedtoelicitclearresponseson

CRFs.CRFinstructionsconveySDTMandCDASHassumptionsinadatacapturecontext

Metadataincludescaptureneeds,e.g.,question

text/prompt,CRFcompletioninstructions

KitHoward,DirectorEducation,CDISC,[email protected],VPEducation,CDISC,[email protected]

WiththankstotheCDASHandSDSteams

SDTM WhytheDifference? CDASH

Authors&Acknowledgements

WhetherRegulatoryAffairsassemblingasubmission,FDAreviewersseekingsafetysignals,orBigDataminerssearchingforas-yetunknownreasons,futureusersmustbeconfidentthatthedatarepresentsthe“truth.”

UsingCDASHfacilitatesconsistent,well-defineddataacrossstudies.Withoutthatconfidence,atbestthedatawillproducevagueassociations;atworst,itmaykillus.

TouseSDTMinsteadofCDASHfordatacapture,takeoutderivedvariables,recordsanddatasets;addindataqualityindicatorvariables;putallcustomandFAvariablesintoparentdatasets;reformatvariablesthatarenotuser-friendly;rewordvariablelabelstoquestions;andrestructureverticaldatatohorizontal.

ThiseffectivelyproducesCDASH.Excepteachorganizationwilldoitdifferently,resultinginreduceddataqualityandtraceability

Conclusions