IQVIA

TECHNOLOGIES

ExecutiveSummary

ApplyingAIinToday’s

RealityofQARAProcesses

AIinMedTechandpracticalrealitiesinQARA

ERDITGREMI,DirectorRegulatoryAffairs,Philips

DENISEMEADE,HealthcareandLifesciencesTechnologyLeader,Microsoft

RAJESHMIRSA,Principal,LifeSciencesQualityandRegulatoryServicesLeader,KPMGLLPCARLOSLUGO,VicePresidentofGlobalProductSafety&Surveillance,Philips

DONSOONG,SeniorDirectorandGeneralManager,QualityManagementSolutions,IQVIATechnologiesLORIELLIS,HeadofInsights,BioSpace(Moderator)

Tableofcontents

Keytakeaways1

Overview1

Context1

BeforetalkingaboutAI,wemustunderstandtheAIplayingfield1

ThelifesciencesandhealthcareindustriesintheU.S.arebehindothercountriesand

industriesinAIadoption2

Thetechnologyisonlyasgoodasyourdata2

Cleandatastartswithvalidation,buthandlingreal-worlddata(RWD)ismessy3

OrganizationsareeducatingQARAprofessionalstounderstandAIandpreparingfor

thefuture3

Conclusion4

Abouttheauthor5

Keytakeaways

?BeforetalkingaboutAI,wemustunderstandtheAI

playingfield.

?ThelifesciencesandhealthcareindustriesintheU.S.arebehindothercountriesandindustriesinAIadoption.

?Thetechnologyisonlyasgoodasyourdata.

?Cleandatastartswithvalidation,buthandlingreal-

worlddata(RWD)ismessy.

?Organizationsareeducatingqualityassuranceandregulatoryaffairs(QARA)specialiststounderstandAIandpreparingforthefuture.

Overview

Thegloballifesciencesindustryhasbeenslowto

adoptAI,particularlygenerativeAI(GenAI).AsGenAIbecomesmorewidelyadopted,QARAprofessionalsfacechallengesinthespaceandinhowitisappliedtoQualityandRegulatoryprocesses,whichrequiresanunderstandingofAItosuccessfullynavigate

datacleansing.

Context

QARAprofessionalsneedtocollaboratewithother

professionalstonavigatethechallengesthatAIbringsandreapthetechnology’sbenefitstoimprovepatientoutcomesandcommercialperformance.

BeforetalkingaboutAI,wemustunderstandtheAI

playingfield

ThepaneldiscussionbeganwithDeniseMeade,

healthcareandlifesciencestechnologyleaderat

Microsoft,settingilluminatingtheAIplayingfield

fortheaudience.SheexplainedthatAIisabroad

category.Machinelearning(ML)discussionstypicallyinvolvetheneedtotrain,testandreleasebasedonlargedatasetswhilelargelanguagemodels(LLMs),whicharealreadytrained,needtobegroundedin

data.ShehighlightedthatGenAIhashadagiantleapforwardinthelastfewyears.

“Toputitintoperspective,ittook

Netflixthreeandhalfyearstoreachonemillionusers.IttookgenerativeAIfivedays.”

—DeniseMeade,HealthcareandLifesciencesTechnologyLeader,Microsoft

TherearetworeasonshowquicklyGenAIwasadopted,Meadeexplained:accessibilityandvalue.“Essentiallyacoupleofcompaniestookabigleapforwardby

investinginitsotherestofusdonotneedtotraineverytimeyouuseLLMS,suchasChatGPT.Itcanbeappliedquicklyandeasilytogetinformation.”

Meadecautionedthatusersneedtohavesome

understandingofhowGenAIworksandhowtouseiteffectively.However,thereisadifferencebetweenLLMsandsmalllanguagemodels(SLMs),andwhatisbeingdonewithtraditionalAIcommonlyusedin

digitalmedicaldevices,roboticsandultrasoundtechnology.

“Withthesemodels,youaretakingwhathasalreadybeentrainedandgroundingitinyourowndata,”

Meadeexplained.“Abigimportantpartisthatdata

isaportionandsuperimportanttotraininmachine

learning.ButforGenAI,itismoreimportanttogroundthedataorgroundtheanswersinthedatathatyou

have.Youdon’tneedtotrainthem.”

|1

Thelifesciencesand

healthcareindustriesin

theU.S.arebehindother

countriesandindustriesinAIadoption

AspointedoutbothbyPhilips’ErditGremi,directorofregulatoryaffairs,andCarlosLugo,thecompany’svicepresidentofglobalproductsafety&surveillance,the

lifesciencesandhealthcareindustriesarebehindin

AIadoption.

“AlthoughwesaythatUnitedStateslifesciencesandhealthcareindustrysayisadvancedininnovationandtechnology,weareextremelybehindtherestoftheworldandotherindustries,”Lugoexplained.“AsmuchasIunderstandwewanttocontinuetobeopento

usingartificialintelligence,there’sstillthatregulatorystop.Ican’teventellyouhowoftenIheardFDAsay,‘Weloveit.Wewanttolearnmoreaboutit.’Westill

needadecidingfactor.Westillneedthathumaninteractiontosayyesorno.”

WhiletheFDAishesitanttoadoptAI,regulatorsin

othercountriesarenot.Australia’sTherapeuticGoodsAdministration(TGA)hasbeensteadilyincreasingitsadoptionofAIandBigPharmaareapproachingPhilipstopartnerinthespace.

AspointedoutbyGremi,LLMsandAIingeneralrequireafundamentallydifferentproductdesignapproach,onenotbasedontraditionalrolesorhierarchicalif-thenstatements.

“Howdoyoumakesurethatthe

datathatyouhaveinputintothisAIorintothismodelaretruly

representativeofallofthetypesofpatientsorcasesthatyouwillseethroughouttheentirelifetimeofthisproduct?”

—ErditGremi,DirectorRegulatoryAffairs,Philips

Instead,regulatorsandproductdesignersneedtoconsiderotherchallenges.

“Areyoustatisticallysoundinthatjudgment,andhaveyouacquireditsufficientlysothatsomethingthat

youmissedtodayinyourvaluationmodel,oryourvalidationsetdoesn’tbecometheadverseeventsayearfromnow?”Gremimused.

Thetechnologyisonlyasgoodasyourdata

Aspreviouslymentioned,GenAIandLLMsarealreadytrainedbutneedtobegroundedindata.ThisiswhereQARAprofessionalsneedtobesavvyenoughto

understandthedataanddatasources.DonSoong,

seniordirectorandgeneralmanagerofquality

managementsolutionsatIQVIA,suggestedthatQARAprofessionalsanddatascientistscollaborate.“Thedatascientistisgoingtounderstandallthetechniquesof

cleansingdata,buttheQARAisgoingtounderstandthenuancesinthedata,sotheymustpartner.”

PhilipshasQARAanddatascientistsinthesame

departmenttopromotecollaborationandreduce

downtime.Withthesetwotypesofexpertiseworkingtogether,researcherscangainatrueunderstandingofthedata,thedemographics,geographyandotherelementsthatbiasthedata.Tomitigatethatbias

throughcleansing,thetwodepartmentsbalancethedatasotherearethesamenumberofparameters

percategory,whichwillgiveafairresponsewhenthealgorithmsrun.

RajeshMirsa,principaloflifesciencesqualityand

regulatoryservicesleaderatKPMGLLP,wasnot

surprisedthatthediscussionturnedtowardsdata

quality.“I’vebeendoingthisforcloseto30yearsandwehavebeenhearingthesamethingforlast30years,thedataqualityisaproblem.Nothinghaschangedthelast30years.”Mirsabelievesthattheindustryneedstorethinkitsstrategy,puttinginplaceapproachesthatwillgeneratedataofsufficientquality.“Dataisnota

staticthing.Itchanges.”

2|ApplyingAIinToday’sRealityofQARAProcesses

Cleandatastartswith

validation,buthandlingReal-WorldData(RWD)ismessy

ToLugo,thekeyisdatavalidation.“Weknowthatdatamaynotbe100%pure,butcanwevalidatewhatwe

haveandmoveforward?”Beingabletoaskandanswer

thisquestionensurestherightqualitydecisions

aremade.Gremiaddedthatdataacquisitionexerciseistrulyidealbutnotalwaysfeasible.Thebest

availabletypeofdataisreal-worlddata(RWD),asitisrepresentativeofwhatthealgorithmormodelbeingdevelopedisgoingtobeencounteringintheworld.“Relyingonreal-worlddataandunderstandingwhatyoucansiftthroughandalreadyhaveavailablein

somewaysisactuallymorerepresentativethanatrueclinicalvalidationofaprospectivestudybecauseitishappeninginclinics,”Gremiexplained.

Mirsaemphasizedthatcorrectdataarecriticalwhendealingwithcomplaintsorotherspecifictasks.In

addition,hesaidthatthereisacertainamountofacceptableriskwhendealingwithdatasinceitwillneverbe100%pure.Heexplainedthequestionsheproposestohisteamsandclients.

“WhatisthepurposeofthedatathatI’mtryingtodoifI’musingforsomesortofalgorithmicmodeling?

WhatsortofhypothesisamItryingtocreate?”In

somecases,hesaid,“Idon’tneed100%correctdata;Icanlivewith70%or80%.ThenItakeoutthe20%or

30%andoutliersIbelievearenotcorrect.Iwillgettothesamehypothesisofwhatismypatternislookingfor.”Whendesigningapattern,hesaidheaddressesthedatainconsistenciesbytakingthemoutofthe

calculationswhilebuildingthemodel.

RWDhasthepotentialtobecollectedinamore

pristinemanner.Meadespokefromexperiencewith

companiesthatcometoMicrosofttofixthecollectionofRWDoranydata.“Oftentimeswhatweendupdoingattheendoftheprojectisactuallystartingmoving

folksfrompaperprocessesjusttodigitalprocesses,”Meadenoted.“Itisamazinghowmanytimeswhenyougointoafactoryandpeopleareusingapenandpapertocollectdata,whichisthenlatertranscribedinto

asystem.”

OrganizationsareeducatingQARAprofessionalsto

understandAIandpreparingforthefuture

ThebiggestchallengeishowtokeepinfrontofAI.

Lugonotedthatconferencesandprivateeventsare

keytohelpingtheindustryadoptAI.Ascompanies

enterthespacemoreaggressive,Lugosaidhefinds

thatitisdifficulttoopendoorsandlowerwalls

becauselifesciencesareguardedaswholeinthe

UnitedStates,unliketherestoftheworld,whenit

comestoAIadoption.Theprocessisslow.However,

hedidnoteincreasingcybersecurityconcernsas

aconsequenceoftechnologicaladvancesincethe

discussiontookplaceduringtheCrowdStrikeincident,whichcreatedflightissuesforbothpanelistsand

audiencemembers.Atthetimeofthediscussion,therewerestill600flightscanceledthedaypriorbyDelta.

Mirsasuggestedthatthemostpressingconcernis

theworkforce.Inthecurrentenvironment,QARA

professionals’workloadconsistsof30%to40%

paperwork.Hesuggestedthatthisis15to20years

behindthetechnologicalcurvecomparedtoother

industries.ThisisindirectoppositiontoFDA’s

approvalof150AI-basedproductswithinthelasteightmonths,whichbringsittoatotalofover700productsbeingapprovedtodate.Whilestillbehindother

industries,QARAprocessesthataredependenton

paperworkslowdowntheprocessandwillnotbeabletoeffectivelyhandletheinfluxofinformationastheindustrycontinuestoblendAIintoscience.

Additionally,thefutureworkforcehasbeenraisedonAIsopaperprocessesmaybeforeigntothem.Mirsaquestioned,“Howdowetraintheworkforce?And

that’saveryimmediateproblemtodayforcompaniesontheworkforceperspective.”Fortheindustryto

moveforward,theworkplacemustmoveawayfrompaper.

LugofurtheremphasizedMirsa’spoint.Becausetheupcomingworkforcehasbeenraisedwithtechnology,trainingbecomesdifficultwhenworkingwith

newhires.Onekeyexamplehegavewasthrough

|3

communication.Lugoexplained,“IfI’mtryingtogetoneofmyengineerswhoIjustrecentlyhired,I’m

calling,callingandcalling.Heorsheneverpicksupthephone,butthemomentIsendatextoranemail,theresponseisimmediate.”ThequestionforLugoishowdoyoutrainanewhirewiththatcommunicationstyle.ItisagapheisactivelyworkingonfiguringoutforPhilips.

Soongfocusedonthecostefficiencyconcernsforleadership.

“Theindustryisdrivingustobemorecostefficient.Domorewithless,soleadershipwantsAIto

beused.”

Conclusion

QARAprocessesandproceduresneedtoevolvetoadopttechnology.Thelifesciencesandhealthcare

industryinUnitedStatesisbehindbothother

industriesandcountriesinadoption.However,there

isclearlyaneedforAI.Theupcomingworkforceis

comfortablewithAIbutwillneedtraining.ThistrainingcanonlybecompletedbythoseQARAprofessionals

whoareabletoclosetheknowledgegapbetweenthecurrentpaperprocesswiththetechnologicalprocessesofthefuture.Ultimately,theadoptionofAIintoQARAprocessesha