making sure we don’t forget the basics when using machine learning aaron n winn-匠人.pdf

该【making sure we don’t forget the basics when using machine learning aaron n winn-匠人 】是由【妙玉】上传分享，文档一共【2】页，该文档可以免费在线阅读，需要了解更多关于【making sure we don’t forget the basics when using machine learning aaron n winn-匠人 】的内容，可以使用淘豆网的站内搜索功能，选择自己适合的文档，以下文字是截取该文章内的部分文字，如需要获得完整电子版，请下载此文档到您的设备，方便您编辑和打印。JNCIJNatlCancerInst(2019)111(6):djy179doi:://academic./jnci/advance-article-abstract/doi/’,’af?:,MD,MPH,CenterforAdvancingPopulationScience,MedicalCollegeofWisconsin,8701WatertownPlankRd,Milwaukee,WI53226(e-mail:******@).Electronichealthrecords(EHRs):andareincreasinglyusedinphysicianpracticesthatcaptureEvenwhenwethinkourfacilityservesalargepopulation,thestructureddata(eg,labvaluesanddiagnosiscodes)andun-diseasecohortthatfulfillsourinclusioncriteriaoftenresultsinstructureddata(eg,physiciannotes).Moreover,-esResearchInstitute’s“”(1)ethislimita-andothersimilarprojects,ancersinsteadofjustonecan-,oftencontainedinex-cercohort;still,wewouldexpectthatdisease-specifictremelylargedatasets,havebeenmetwithnewmethodologicalpredictionmodels,-,wouldperformmuchbetterthansuchabroaddiseaseco-Usingthefirstdataset,,throughtheuseofinteractionsbe-variablesincludedandtheircoefficientsarecontinuallymodi-tweenthedisease(breastvslungcancerormetastaticvsfied(bythe“machine”),localized)andothercovariates,machinelearningcanappropri-“themachine”,,theinthisstudychosetonotincludeinteractionsbetweenparametersofthepredictionmodelaremodifiedor“tuned”un--Anotherkeyconsiderationforcreatingpredictionmodelsismonlythatthecovariatesneedtomeasurewhatwethinktheyarereferredtoasthe“test”dataset),allowingadescriptionofmodelmeasuring(5).-’(2)seminalbook,.(3)(includingHastie),weseealeading-edgequalityonthepredictionmodeland,inturn,physicians’-exampleofhowmachinelearningcanbeusedwithinthecur-,specificallyusingexistingimplementationofamachine-learningpredictiontoolinaclini-EMRdata,,ifapredictiontoolisimplementedinastudyandshowsthatusingmachinelearningcanbetterpredicthealthsystemandadministratorsuseittoimplementtime-(eg,requiringphysicians’mendationstoHowever,eventhoughthisisanewandexcitingtechnique,eithermirrorthepredictionmodelortoprovideadditionaljusti-theexistingrulesaboutcreatingpredictionmodelsshouldnotfication),mendthetreatmentthatofapredictionmodel(4)?,becausetheimpactAthirdconcernwhencreatingpredictionmodelsistheap-eshouldbesimilarwithinpropriatetreatmentofmissingvariables(4).GiventhelargeReceived:August24,2018;Accepted:September6,2018?TheAuthor(s),pleaseemail:journals.******@12|JNCIJNatlCancerInst,2019,,,andtheacknowledgmentthatmanyvari-Notesableshavelittleimpactontheoverallfitofthemodel,itseemsthatthetrade-pletedataandalargersampleAffiliationsofauthors:DepartmentofClinicalSciences,issomethingthatshouldbeexplicitlyexploredinfuturema-SchoolofPharmacy(ANW)andDepartmentofMedicineandchine-,isitworthlosing5%ofCenterforAdvancingPopulationScience(JMN),MedicalollegeofWisconsin,Milwaukee,,isnotstatisticallysignificantonitsown,-ease?AlthoughitishardtounderstandhowtoappropriatelyReferencesDownloadedfromhttps://academic./jnci/advance-article-abstract/doi/,,HudsonKL,BriggsJP,LauerMS.:turningadreamintoreal-;21(4):576–(ie,unstructuredfreetextdatafields).,TibshiraniR,?Thisispartic-NewYork,NY:Springer;,HenryAS,WoodDJ,-ancerpatientsusinghigh-;111(6):-,VickersAJ,CookNR,-dictionmodels:-learningtechniquesthatcangobeyondpredictionmod-2010;21(1):128–(6,7).Currently,machine-,PeatG,BelcherJ,CollinsGS,-learningapproachesarebestsuitedforpredictionproblems,:..(3)However,,ScarpaJ,BruzeliusE,TamlerR,BasuS,,epoorlymeasuredcova-:amachinelearning-basedpost-;5(10):808–-:;(6324):483–485.