USF Libraries
USF Digital Collections

Selection of clinical trials

MISSING IMAGE

Material Information

Title:
Selection of clinical trials knowledge representation and acquisition
Physical Description:
Book
Language:
English
Creator:
Nikiforou, Savvas
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
Clinical trials -- Data processing   ( lcsh )
Knowledge acquisition (Expert systems)   ( lcsh )
agents
medical
expert systems
knowledge representation
knowledge acquisition
Dissertations, Academic -- Computer Science -- Masters -- USF   ( lcsh )
Genre:
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: When medical researchers test a new treatment procedure, they recruit patients with appropriate health problems and medical histories. An experiment with a new procedure is called a clinical trial. The selection of patients for clinical trials has traditionally been a labor-intensive task, which involves matching of medical records with a list of eligibility criteria. A recent project at the University of South Florida has been aimed at the automation of this task. The project has involved the development of an expert system that selects matching clinical trials for each patient.If a patient's data are not sufficient for choosing a trial, the system suggests additional medical tests. We report the work on the representation and entry of the related selection criteria and medical tests. We first explain the structureof the system's knowledge base, which describes clinical trials and criteria for selecting patients. We then present an interface that enables a clinician to add new trials and selection criteria without the help of a programmer. Experiments show that the addition of a new clinical trial takes ten to twenty minutes, and that novice users learn the full functionality of the interface in about an hour.
Thesis:
Thesis (M.S.C.S.)--University of South Florida, 2002.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Savvas Nikiforou .
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 42 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001413333
oclc - 50138092
notis - AJJ0749
usfldc doi - E14-SFE0000029
usfldc handle - e14.29
System ID:
SFS0024720:00001


This item is only available as the following downloads:


Full Text

PAGE 1

OfceofGraduateStudies UniversityofSouthFlorida Tampa,Florida CERTIFICATEOFAPPROVAL Thisistocertifythatthethesisof SAVVASNIKIFOROU inthegraduatedegreeprogramof ComputerScience wasapprovedonJanuary30,2002 fortheMasterofScienceinComputerSciencedegree ExaminingCommittee: MajorProfessor:EugeneFink,Ph.D. Member:LawrenceO.Hall,Ph.D. Member:DmitryB.Goldgof,Ph.D.

PAGE 2

SELECTIONOFCLINICALTRIALS: KNOWLEDGEREPRESENTATIONANDACQUISITION by SAVVASNIKIFOROU Athesissubmittedinpartialfulllment oftherequirementsforthedegreeof MasterofScienceinComputerScience DepartmentofComputerScienceandEngineering CollegeofEngineering UniversityofSouthFlorida May2002 MajorProfessor:EugeneFink,Ph.D.

PAGE 3

c CopyrightbySavvasNikiforou2002 Allrightsreserved

PAGE 4

Dedication ToMarina,forhelpingmefulllmydreams,oneatatime.

PAGE 5

Acknowledgments IgratefullyacknowledgethehelpofEuge neFink,whohassupervisedmythesis workandguidedmethroughallstepsofresear chandwriting.IamgratefultoLawrence HallandDmitryGoldgoffortheirvaluablegui danceandfeedback.IalsothankPrinceton Kokku,TimIvanovskiy,andBhaveshGoswamifortheircommentsandsuggestions. Ihavereceivedinvaluablesupportfrommywife,MarinaNikiforou,whohasprovidedhelpandencouragementthroughmygr aduatestudies.Iamalsothankfultomy parents,ChristodoulosandYiannoulaNikifor ou,fortheirsupportandunderstanding.Finally,IwouldliketothankJoshJohnsonandM aryParrishfortheirfeedbackandhelpwith grammar.Theworkhasbeenpartiallyspons oredbytheBreastCancerResearchProgram oftheU.S.ArmyMedicalResearchandM aterielCommandundercontractDAMD17-001-0244.

PAGE 6

TableofContents ListofTables......................................ii ListofFigures......................................iii Abstract.........................................iv Chapter1Introduction1 Chapter2PreviousWork3 2.1ExpertSystems................................3 2.2KnowledgeRepresentation..........................5 2.3KnowledgeAcquisition..... .......................6 Chapter3SelectionofClinicalTrials8 3.1Example....................................9 3.2KnowledgeBase...............................11 3.2.1Questions...............................11 3.2.2MedicalTests.............................12 3.2.3EligibilityCriteria... .......................13 3.3OrderofTests.................................13 Chapter4EnteringEligibilityCriteria16 4.1Example....................................16 4.2TestsandQuestions..............................24 4.2.1AddingTests.............................24 4.2.2ModifyingaTest...........................24 4.2.3AddingaQuestion..........................24 4.2.4DeletingQuestions..........................25 4.2.5GeneralInformation.........................25 4.3EligibilityConditions...... .......................26 4.3.1AddingClinicalTrials........................26 4.3.2SelectingTests............................26 4.3.3SelectingQuestions..........................26 4.3.4DeninganExpression........................27 4.4LogicalExpressions..............................27 4.5EntryTime..................................32 Chapter5ConcludingRemarks35 References........................................36 i

PAGE 7

ListofTables 4.1Entrytimeformedicaltestsandrelatedquestions................33 4.2Entrytimeforeligibilitycriteria.........................33 ii

PAGE 8

ListofFigures 3.1EligibilitycriteriaforClinicalTrialA......................9 3.2Examplequestions...............................10 3.3EligibilitycriteriaforClinicalTrialB......................11 3.4Checkingtheeligibilityfortwoclinicaltrials..................12 3.5Logicalexpressions...............................13 4.1Enteringtestsandquestions...........................17 4.2Enteringeligibilitycriteria...........................17 4.3“Modifyingatest” screen............................17 4.4Eligibilitycriteriawithadisjunctivecondition..................19 4.5Addingnewtests................................19 4.6Selectingatest.................................20 4.7“Modifyingatest” screen............................20 4.8Addingnewquestions..............................21 4.9“Modifyingatest” screenwithalistofquestions................21 4.10Addinganewclinicaltrial...........................22 4.11Choosingtestsandquestiontypes........................22 4.12Selectingquestionsandanswers.........................23 4.13Combiningquestionsintoalogicalexpression..................23 4.14Viewingquestions...............................25 4.15Deletingtwoquestions.............................28 4.16“Addingclinicaltrials” screen.........................28 4.17Constructingarejectionexpression.......................29 4.18Convertinganexpressionintoadisjunctivenormalform.............30 4.19Acceptanceandrejectionexpressions......................31 4.20Entrytimeformedicaltestsets.........................34 4.21Entrytimeforeligibilitycriteria.........................34 4.22Dependencyofthetimeonthenumberofquestions...............34 iii

PAGE 9

SELECTIONOFCLINICALTRIALS: KNOWLEDGEREPRESENTATIONANDACQUISITION by SAVVASNIKIFOROU AnAbstract ofathesissubmittedinpartialfulllment oftherequirementsforthedegreeof MasterofScienceinComputerScience DepartmentofComputerScienceandEngineering CollegeofEngineering UniversityofSouthFlorida May2002 MajorProfessor:EugeneFink,Ph.D. iv

PAGE 10

Whenmedicalresearcherstestanewtreatmentprocedure,theyrecruitpatientswithappropriatehealthproblemsandmedicalhistor ies.Anexperimentwithanewprocedureis calleda clinicaltrial. Theselectionofpatientsforclinicaltrialshastraditionallybeena labor-intensivetask,whichinvolvesmatchingofmedicalrecordswithalistofeligibility criteria. ArecentprojectattheUniversityofSouthFloridahasbeenaimedattheautomation ofthistask.Theprojecthasinvolvedthede velopmentofanexpertsystemthatselects matchingclinicaltrialsforeachpatient.Ifa patient'sdataarenotsufcientforchoosinga trial,thesystemsuggestsa dditionalmedicaltests. Wereporttheworkontherepresentationandentryoftherelatedselectioncriteria andmedicaltests.Werstexplainthestru ctureofthesystem'sknowledgebase,which describesclinicaltrialsandcriteriaforselectingpatients.Wethenpresentaninterfacethat enablesacliniciantoaddnewtrialsandsel ectioncriteriawithoutth ehelpofaprogrammer. Experimentsshowthattheadditionofanewclinicaltrialtakestentotwentyminutes,and thatnoviceuserslearnthefullfunctionalityoftheinterfaceinaboutanhour. AbstractApproved: MajorProfessor:EugeneFink,Ph.D. AssistantProfessor,ComputerScienceandEngineering DateApproved: v

PAGE 11

Chapter1 Introduction Cancercauses550,000deathsintheUnitedStateseveryyear [ Keppel etal. ,2002 ] ,and thetreatmentofcancerisanactiveresearcharea.Medicalexpertsexplorenewtreatment methods,includingdrugs,surge rytechniques,radiationthera pies,andgenetherapies.An experimentwithanewtreat mentprocedureiscalleda clinicaltrial .Whenresearchers testanewprocedure,theychoosesubjectsw ithanappropriatecancertypeandmedical history.Theselectionofsubjectshastraditionallybeenamanualprocedure,whichinvolves signicanthumaneffort,andcliniciansoftenmisseligiblepatients [ Yusuf etal. ,1990; Kotwall etal. ,1992;Tu etal. ,1993;S eroussi etal. ,1999a;GennariandReddy,2000 ] ArecentprojectattheUniversityofSouthFloridahasbeenaimedatautomaticselectionofpatientsforclinicaltrials.Fletcherandhercolleagueshavedevelopedanexpert systemthatpromptsaclinicianforapatient'sdataandidentiesallmatchingtrials [ Bhanja etal. ,1998 ] .Experimentshaveshownthatthesystemimprovesthematchingaccuracyand reduceshumaneffort.Kokku etal. [2002a]haveaddedamechanis mfororderingrelated medicaltests;itspurposeistominimizethecostoftestsinvolvedintheselectionprocess. Thesystemincludesaknowledgebasewith informationaboutavailableclinicaltrials,criteriaforselectingpatients,andrelat edmedicaltests.Whenintroducingnewtrials, cliniciansneedtoaddthemtotheknowledge base.Fletcherdidnotprovideaninterface foraddingnewtrials,andsheencodedtheeligibilitycriteriainaspecialprogramming language.Thetimerequiredtoa ddanewtrialvariedfromtwentytothirtyhours.The 1

PAGE 12

languagedidnotenforcestandardencoding, andtwoprogrammersc ouldproduceincompatibledescriptionsofeligibilitycriteria. Wehavedesignedaweb-basedinterfaceth atenablesacliniciantoaddnewtrials withoutthehelpofaprogrammer.Ithasre ducedtheentrytimefromtwentyhoursto abouttwentyminutes.Furthermore,itconvert stheeligibilitycriteriaintoastandardized formandensurescompatibilityofallknowledgeinthesystem.Wehaveusedtheinterface tobuildaknowledgebaseforclinicaltrialsattheMofttCancerCenter,locatedatthe UniversityofSouthFlorida. Webeginwithareviewofpreviousworkonmedicalexpertsystems(Chapter2).We thenexplaintheknowledgerepresentationint hedevelopedsystem(Chapter3),describe theinterfaceforaddingnewknowledge(Cha pter4),andconcludewithasummaryofthe resultsandfuturechallenges(Chapter5). 2

PAGE 13

Chapter2 PreviousWork Theautomationofmedicaldiagnosisandtreat mentselectionisani mportantproblem,and computerscientistshavedevelopedavarietyofmedicalexpertsystems.Theyhavecreatedrule-basedsystemsandBayesiannetworksthatcaptureexpertiseforseveralmedical domains,coveringbacterialdiseases,cancer,asthma,liverdiseases,andAIDS.Wereview someofthesesystems(Section2.1)andrela tedworkonknowledgerepresentationand acquisition(Sections2.2and2.3). 2.1ExpertSystems Researchersbegantoworkonmedicalapplicationsofarticialintelligenceintheearly seventies.ShortliffeandhiscolleaguesdevelopedthefamousMYCINsystem,whichdiagnosedbacterialdiseasesands uggestedappropriatetherapies [ Shortliffe,1974;Shortliffe etal. ,1975;BuchananandShortliffe,1984 ] .Itevolvedfromachemicalexpertsystem, calledDENDRAL[ Lederberg,1965;Buchanan etal. ,1969;Lederberg,1987 ] ,thatdeterminedmolecularstructuresbas edonspectrographyresults.MYCIN'sknowledgebaseconsistedof if-then rules,whichallowedtheanalysisof symptoms,selectionoftherapies,andevaluationoftheselectioncertainty.Forexample, thesystemcoulddeterminethatapatientwithuneededaspirinwith0.8certainty.ExperimentsconrmedthatMYCINcorrectlydiagnosedcommondiseases,whichledtothe developmentofothermedicalsystems [ BuchananandShortliffe,1984;Musen,1989 ] ,such 3

PAGE 14

asNEOMYCIN,PUFF,CENTAUR,andVM.Shortliffe etal. [1981]createdasystemforselectingchemotherapytreatments,calledONCOCIN,whichalsoevolvedfromMYCIN. Lucas etal. [1989]constructedarule -basedsystem,calledHEPAR,fordiagnosing liverandbiliary-tractdiseases,butitoftengavewrongdiagnoses [ KorverandJanssens, 1993;Onisko etal. ,1997 ] .KorverandLucas[1993]convertedtheinitialsystemintoa Bayesiannetwork,whichimproveditsperformance [ Lucas,1994 ] Musen etal. [1996]builtarule-ba sedsystem,calledEON,thatanalyzeddependencies amongtheavailabledataandassignedAIDSpatientstoclinicaltrials.Forexample,ifan onsetoflowbloodpressurecoincidedwiththebeginningofanewclinicaltrial,thesystem wouldnoticethatthetrialmay havecausedthelowpressure. Ohno-Machado etal. [1993]developedtheAIDS2system,whichalsomatchedAIDSpatientstoclinicaltrials.TheyintegratedlogicalruleswithBayesiannetworks,which helpedtomakedecisionsintheabsenceofsome dataandtoquantifythecertaintyofthese decisions. Bouaud etal. [1998;2000]createdacancerexpertsystem,calledONCODOC,that suggestedalternativeclinicaltrialsfor eachpatient,andallowedaphysiciantochoose amongthem.Itincludedagraphicalinterfacefo rinteractiveentryofapatient'sdataand considerationofalternativetrials.S eroussi etal. [1999a;1999b;2000;2001a;2001b]usedONCODOCtoselectparticipantsforclinicaltrialsattwohospitals,whichhelpedtoincrease thenumberofselectedpatientsbyafactorofthree. Theocharous[1996]developedaBayesiansy stemthatchoseclinicaltrialsforcancer patients.Itlearnedconditionalprobabilitiesofmedical-testoutcomesandusedthemto evaluatetheprobabilityofapa tient'seligibilityforeachtrial [ Papaconstantinou etal. 1998 ] .Onthenegativeside,theavailablemedi calrecordswereinsufcientforlearning accurateprobabilities.Furthermore,whenause raddednewclinicaltrials,hehadtochange thestructureoftheunderlyingBayesianne twork,whichwasoftenadifculttask. 4

PAGE 15

HammondandSergot[1996]builttheOaSiSar chitecture,whichcombinedthetechniquesfromseveralearliersystems,includingONCOCINandEON.Ithadagraphicalinterfaceforenteringpatients'dataan dextendingtheknowledgebase. Falloweld etal. [1997]studiedhowphysicianssel ectedcancerpatientsforclinical trials,andcomparedmanualandautomaticselection.Theyshowedthatexpertsystems couldimprovetheselectionaccuracy;howev er,theirstudyalsorevealedthatphysicians wereusuallyreluctanttousethesesystems.Carlson etal. [1995]conducteds imilarexperimentswithchoosingparticipantsforAIDSstudies,andalsoconcludedthatexpertsystems couldleadtoamoreaccurateselection. 2.2KnowledgeRepresentation Researchershavelongrealizedtheneedforageneral-purposerepresentationofmedical knowledge [ Clancey,1993;Clancey,1995 ] andinvestigatedavarietyofrelatedrepresentations. Inparticular,Ohno-Machado etal. [1998]proposedageneralformatformedical knowledge,calledtheGuideLineInterchangeFormat.Theirprojectinvolvedresearchers fromStanford,Harvard,andColumbia;theyus edthedevelopedrepresentationwithavarietyofalgorithms,andconcludedthatitwas sufcientformostmedicalknowledge.Onthe negativeside,itdidnotenforcecompatibilityamongknowledgebasesdevelopedbydifferentresearchers.Furthermore,itneededmajo rimprovementsforrepresentingconditional expressions,temporalreasoning,anduncertainty. Lindberg etal. [1993]proposedanalternativegen eral-purposeformat,calledthe UniedMedicalLanguageSystem,anddevel opedtoolsforconvertingvariousmedical databasesintothisf ormat.Later,LeDuff etal. [2000]studiedtechniquesfortranslating natural-languagedescri ptionofdiseasesintoLindberg'srepresentation. 5

PAGE 16

Rubin etal. [1999;2000]analyzedselectioncriteriaf orclinicaltrialsrelatedtothree cancertypesandproposedaformatforthesecriteria.Theybuiltamechanismforencodingnewcriteria,whichhelpedtheuserst oavoidsimplemistakes,suchasmissingor inconsistentselectionrules. Wang etal. [2001]comparedeightpreviouslydeve lopedformatsandidentiedmain elementsofmedicalknowledge,whichincludedpatientdata,treatmentdecisions,related actions,andaglobalstateofanexpertsyst em.Wangalsopointedouttheneedforabstractionandtemporalreasoning. 2.3KnowledgeAcquisition Earlyexpertsystemsdidnotincludeknowledge-acquisitiontools,andprogrammershandcodedtherelatedrules.Tosimplifyknowledgeentry,researchersimplementedspecialized toolsforsomesystems.Forexample,Musen etal. [1988;1989]developedtheOPALsystemforaddingnewknowledgetoONCOCIN,andMarcusandMcDermott[1989]builttheSALTsystem,whichhelpedengineerstos pecifyrulesforelevatordesign. Eriksson[1993]pointedouttheneedforgeneral-purposetoolsthatwouldallowefcientknowledgeacquisition,anddescribedasystemforbuildingsuchtools.Tallisand hiscolleaguesdevelopedalibraryofscriptsf ormodifyingknowledgebases,whichhelped toenforcetheconsistencyofthemodiedknowledge [ GilandTallis,1997;Tallis,1998; TallisandGil,1999;Tallis etal. ,1999 ] .KimandGil[2000a;2000b]consideredtheuseof scriptsforbuildingnewknowledge-acquisitiontools,andcreatedasystemforevaluating thesetools.Blythe etal. [2001]designedagenera lknowledge-acquisitioninterfacebased onprevioustechniques. MusenandhiscolleaguesdevelopedthePROTEGEenvironmentforcreatingknowledgebases [ Musen,1989 ] ;later,researchersuseditintheworkonAIDSexpertsystems [ Puerta etal. ,1992a;Puerta etal. ,1992b ] ,asthmatreatmentselection [ Johnsonand 6

PAGE 17

Musen,1996 ] ,andelevator-designrules [ Rothenuh etal. ,1996 ] .Musen etal. [2000] extendedPROTEGEandbuiltanewversion,calledPROTEGE-2000. Severalresearchershavestudiedtechniquesforextractingmedicalknowledgefrom natural-languagedocumen ts.Inparticular,HahnandSchnattinger[1997a;1997b;1998] builtaparserforprocessingGermanmedicaltextsongastro-intestinaldiseases.Romacker andHahn[2001]improvedtheparserandsho wedthattheresultingaccuracyofsemanticrepresentationswasbetween80%and93%;however,itseffectivenessinconstructing knowledgebaseswasverylow. Researchershavealsoconsidereddata-min ingtechniquesforlearningmedicalknowledgefromclinicaldatabases [ Cimino etal. ,1988;Shusaku,1998;Mendonc aandCimino, 2000 ] .Althoughthesetechniquesgeneratedbasicd iagnosisrules,theyallowedthecorrect diagnosisonlyin8%ofthetestcases. Thereadercanndamoredetailedreviewoftheworkonknowledgeentryinthe bookbyBooseandGains[1990],whodescribed knowledge-acquisitiontoolsnotonlyfor medicalsystemsbutalsoforotherapplicati ons.RinglandandDuce[1988]presentedstandardtechniquesforknowledge representation,includingf unctionalapproachesandtemporalreasoning.Price[1990]alsoreviewedgene raltoolsforknowledgerepresentationand acquisition. 7

PAGE 18

Chapter3 SelectionofClinicalTrials PhysiciansattheMofttCancerCenterhave about150clinicaltrialsavailableforcancer patients.Theyhaveidentiedcriteriathatdet ermineapatient'seligibilityforeachclinical trial,andtheyusethesecriteriatoselectappr opriatetrialsforeligiblepatients.Traditionally,physicianshaveselectedtrialsbyama nualanalysisofapatient 'sdata.Thereviewof resultingselectionshasshownthattheyusuallydonotcheckallclinicaltrialsandoccasionallymissanappropriatetrial. Toaddressthisproblem,Fle tcherandhercolleaguesbu iltasystemforautomatic selectionofclinicaltrials,andaknowledgebaseforbreastcancer [ Bhanja etal. ,1998 ] Theirsystempromptedacliniciantoentertheresultsofmedicaltestsforapatient,and identiedappropriatetrials.Itselectedtheorderofquestionsthatminimizedtheexpected amountofdataentry.Forinstance,ifsome questioncouldrevealthatapatientwasnot eligibleforanytrial,thesystemaskeditbeforetheotherquestions.Experimentsshowed thatthesystemreducedthehumaneffortinvolvedintrialselectionandhelpedtoavoid inaccuracies.Kokkuhascontinuedthiswor kandaddedinformationaboutthecostsof medicaltestsandthepainlevelsofrelatedprocedures [ Kokku etal. ,2002b ] .Hehas developedatechniqueforndingtheorderof testproceduresthatreducestheexpected costandpain. WereviewKokku'ssystemforselectionofclinicaltrials.Webeginwithanexample oftheselectionprocess(Section3.1),desc ribethemainelementsoftheknowledgebase (Section3.2),andexplainheuristic sfortestordering(Section3.3). 8

PAGE 19

Generalinformation 1.Thepatientisfemale. 2.Sheisatmostforty-veyearsold. Mammogram: Costis$150,painlevelis1 3.CancerstageiseitherIIorIII. 4.Cancerisnotinvasive. Electrocardiogram: Costis$160,painlevelis1 5.Thepatienthasnocongenitalheartdisease. 6.Thepatienthasnocardiacarrhythmias. Biopsy: Costis$200,painlevelis3 7.Atmostthreelymphnodeshavetumorcells. 8.Alltumorsaresmallerthanthreecentimeters. Figure3.1:EligibilitycriteriaforClinicalTrialA.3.1Example InFigure3.1,wegiveasimpliedexampleofeligibilitycriteriaforacertainclinicaltrial, calledTrialA.Wecanusethistrialforyoung andmiddle-agewomenwithanoninvasive breastcanceratstageIIorIII.Apatientiseligibleifshehasatmostthreeaffectedlymph nodes,allhertumorsaresmallerthanthree centimeters,andshehasnoheartproblems. Whenacliniciantestsapatient'seligibilityforthistrial,hehastoorderthreemedical tests.TocheckConditions3and4,aclinicia nsendsapatientforamammogram,whichis almostpainlessandcosts$150.Ifthepatientm eetstheseconditions,sheneedsanelectrocardiogram,whichisthenextcheapesttest.Finally,ifshesatisesConditions5and6,the cliniciansendsherforabiopsy,whic hisanexpensiveandpainfulprocedure. Thesystemrstpromptsacliniciantoenterthepatient'ssexandage(Figure3.2a). Ifthepatientsatisesthecorrespondingc onditions,thesystemasksforthemammogram results(Figure3.2b),andtheclinicianord ersamammogram.Then, thesystemrequests theelectrocardiogram(Figure3 .2c)andbiopsy(Figure3.2d). Iftheclinicianhasinformationaboutsomeofthepatient'soldtests,hemayanswer thecorrespondingquestionsalongwithenteri ngpersonaldata,beforethesystemselects 9

PAGE 20

Whatisthepatient'ssex? FemaleMale Whatisthepatient'sage? (a)Generalquestions. Whatisthecancerstage? IIIIIIIV Doesthepatienthaveinvasivecancer?Yes No Unknown (b)Mammogramresults. Doesthepatienthavecongenitalheartdisease?Yes No Unknown Doesthepatienthavecardiacarrhythmias?Yes No Unknown (c)Electrocardiogramresults. Howmanylymphnodeshavetumorcells? Whatisthegreatesttumordiameter? (d)Biopsyresults. Figure3.2:Examplequestions.Thesystemguidesaclinicianthroughaseriesofquestions, groupedbytestprocedures,andusestheanswerstoselectappropriateclinicaltrials.testprocedures.Forexample,ifheknowsthatthepatienthasinvasivecancer,hemayenter italongwithsexandage,andthenthesystemimmediatelyrejectsTrialA. InFigure3.3,wegiveanotherexampleofeligibilitycriteria,andwerefertothe correspondingclinicaltrialasTrialB.Ifbotht rialsareintheknowledgebase,thesystem cancheckwhetherapatientiseligibleforeitherofthem.First,itpromptstheclinician toenterthegeneralinforma tion(Figure3.4a),andthenasksforthemammogramresults, whicharerelevanttobothtrials(Figure3.4b).Iftheresultssatisfytheeligibilitycriteriafor TrialB,thesystemrequeststheliver-testda ta(Figure3.4c),andthenoutputsthedecision forTrialB.TodeterminetheeligibilityforTrialA,itrequeststheelectrocardiogramdata (Figure3.4d).Iftheresultssatisfytheeligibilitycriteria,thesystemasksforthebiopsy data(Figure3.4e)andthenoutputsthedecisionforTrialA. 10

PAGE 21

Generalinformation 1.Thepatientisfemale. 2.Sheisatleasttwenty-sevenyearsold. Mammogram: Costis$150,painlevelis1 3.CancerstageisIII. 4.Cancerisnotrecurrent. Livertest: Costis$150,painlevelis1 5.ThepatienthasnohepatitisB. 6.Thepatienthasnoliverinfections. Figure3.3:EligibilitycriteriaforClinicalTrialB.3.2KnowledgeBase Thesystem'sknowledgebaseincludesquesti ons,medicalprocedures,andlogicalexpressionsthatrepresenteligibilityconditions. 3.2.1Questions Thesystemsupportsthreetypesofquestions.T hersttypetakesayes/noresponse,the secondisamultiplechoice,andthethirdrequiresanumericanswer.Whenthesystemasks ayes/noquestion,itacceptsoneofthreeanswers:YES,NO,orUNKNOWN.Theusercan disablethe unknown optionforsomequestions;forexample,wedonotaccept unknown for theelectrocardiogramresultsinFigure3.4(d).Whenthecliniciangetsamultiple-choice question,suchasacancerstage,hehastoselect oneoftheavailableanswers(Figure3.4b). Ananswertoanumericquestionisarealvalue,whichmustbewithinthelegalrangefor thisquestion;forexample,apatient'sagei sbetween0and150(Figure3.4a),andatumor diameterisbetween0and25centimeters(Figure3.4e). 11

PAGE 22

Whatisthepatient'ssex? FemaleMale Whatisthepatient'sage? (a)Generalquestions. Whatisthecancerstage? IIIIIIIV Doesthepatienthaveinvasivecancer?Yes No Unknown Doesthepatienthaverecurrentcancer?Yes No Unknown (b)Mammogramresults,relevanttobothtrials. DoesthepatienthavehepatitisB?Yes No Unknown Doesthepatienthaveliverinfections?Yes No Unknown (c)Liver-testresults,relevanttoTrialB. Doesthepatienthavecongenitalheartdisease?Yes No Doesthepatienthavecardiacarrhythmias?Yes No (d)Electrocardiogramresults,relevanttoTrialA. Howmanylymphnodeshavetumorcells? Whatisthegreatesttumordiameter? (e)Biopsyresults,relevanttoTrialA. Figure3.4:Checkingtheeligibilityfortwoclinicaltrials.Thesystembeginswiththequestions relatedtobothtrials.3.2.2MedicalTests Thedescriptionofamedicaltestincludesthetestname,dollarcost,estimatedpainlevel, andlistofquestionsthatcanbeansweredbasedonthetestresults.Forexample,the mammograminFigure3.1hasacostof$150andpainlevelof1,anditprovidesdatafor Criteria3and4.Twodifferenttestsmayans werthesamequestion;forinstance,boththe mammogramandthebiopsyshowthecancerstage. 12

PAGE 23

sex =FEMALEand sex =MALEorage [0 45]and age (45 150]orcancer-stage {II,III} and cancer-stage{I,IV} orinvasive-cancer =NOand invasive-cancer {YES,UNKNOWN}orlymph-nodes [0 3]and lymph-nodes (3 100]ortumor-diameter [0 3]and tumor-diameter (3 25]orheart-disease =NOand heart-disease {YES,UNKNOWN}orcardiac-arrhythmias =NO cardiac-arrhythmias {YES,UNKNOWN}(a)Acceptanceexpression.(b)Rejectionexpression. Figure3.5:LogicalexpressionsforthecriteriainFigure3.1.Theacceptanceexpressionrepresents theeligibilityconditions(a),whereastherejectionexpressiondescribesineligiblepatients(b).3.2.3EligibilityCriteria Weencodeeligibilityforaclinicaltrialbyalogicalexpressionthatdoesnothavenegations, calledthe acceptanceexpression .Itincludesvariablesthatrepresenttheavailabledata,as wellasequalities,inequalities,"set-element"relations,conjunctions,anddisjunctions.For example,weencodethecriteriainFigure3 .1bytheexpressiongiveninFigure3.5(a). Inaddition,thesystemusesthelogicalcomplementoftheeligibilitycriteria,calledthe rejectionexpression, whichalsodoesnotincludenegations(Figure3.5b).Itdescribesthe conditionsthatmakeapatientineligiblefortheclinicaltrial. ThesystemcollectsdatauntilitcandeterminewhichofthetwoexpressionsisTRUE. Forexample,ifthepatient'ssexisMALE,thentherejectionexpressioninFigure3.5(b)isTRUE,andthesystemimmediatelydeterminestha tthistrialisinappropriate.Ontheother hand,ifthesexisFEMALE,andtheothervaluesareunknown,thenneitherexpressionisTRUE,andthesystemhastoaskmorequestions. 3.3OrderofTests Whenaclinicianentersmedicaldataforapatient,thesystemidentiesallappropriate trials.Thetotalcostandpainlevelofthetestsinvolvedinthetrialselectionmaydependon theirordering.Forinstance,ifwebeginw iththemammogram,anditshowsthatthecancer 13

PAGE 24

stageisI,thenwecanimmediatelyrejectthetrialinFigure3.1andavoidmoreexpensive tests. Kokku etal. [2002a;2002b]havestudiedheuristicsfor orderingthetests;theirheuristicsaccountforthecostandpainleveloftest s,thestructureofacceptanceandrejection expressions,andthenumberofexpressionsth atrequireeachtest.Theheuristicsuseadisjunctivenormalformofacceptanceandrejec tionexpressions;thatis,eachexpressionmust beadisjunctionofconjunctions. Kokkuhasdenedtheoverall"payment"for medicaltestsasalinearcombinationof theircostsandpainlevels;thatis,ifapatientneeds n tests,thepaymentis a ni =1costi+ b ni =1paini. Ausersetsthevaluesof a and b ,andthesystemchoosestheorderofquestionsthatreduces theexpectedpayment.Aftergettingtheresultsofthersttest,itre-evaluatestheneedfor othertestsandrevisestheirordering.Thechoiceofthersttestisbasedonthreecriteria. 1. Costandpainlevelofthetest. Thesystemgivespreferencetotestswithsmaller payments.Forexample,itmaystartwiththemammogram,whichischeaperand lesspainfulthantheothe rtwotestsinFigure3.1. 2. Numberofclinicaltrialsthatrequirethetest. Whenthesystemchecksapatient's eligibilityforseveraltrials,itprefersteststhatprovidedataforlargernumberof trials.Forexample,iftheelectrocardiogramgivesdataforthreetrials,thesystem maypreferittothemammogr amdespiteitshighercost. 3. Immediatedecisionsforsometrials. Ifatestcanleadtoanimmediateacceptanceor rejectionofsometrials,thesystemprefersittoothertests.Forinstance,iftheliver testshowsthatthepatienthashepatitisB,thesystemcanimmediatelyrejectTrialB. 14

PAGE 25

Toselectthersttest,thesystemscoresallrequiredtestsaccordingtothethree criteria.Itcomputesalinearcombinationoft hesethreescoresforeverytest,andchooses thetestwiththehighestscore.Kokku etal. [2002b]haveevaluatedthisstrategyusing retrospectivedatafor187patientsattheMof ttCancerCenter,anddemonstratedthatit signicantlyreducesthecost. 15

PAGE 26

Chapter4 EnteringEligibilityCriteria WhenFletcherdevelopedtheinitialsystem,shedidnotincludeaninterfaceforadding eligibilitycriteria,andaprogrammerhadtoen codethecriteriainaspecializedlanguage. Wehavedesignedaweb-basedinterfacefora ddingnewcriteria,whichconsistsoftwo mainparts;therstpartisforenteringinfo rmationaboutmedicaltests(Figure4.1),and thesecondisforspecifyingeligibilitycriteria(Figure4.2). Theinterfaceincludesfteenscreens;threeofthemare"startscreens,"whichcan bereacheddirectlyfromanyotherscreen.Forexample,considerthe “Modifyingatest” screeninFigure4.3,whichallowschangingth etestname,cost,andpainlevel.Ithasfour buttonsatthebottomformovingtorelatedscr eens,andthreebuttonsontheleftformoving tothestartscreens. Wegiveanexampleofenteringeligibilitycriteria(Section4.1),describethetwomain partsoftheinterface(Sections4.2and4.3),givealgorithmsforgeneratingacceptanceand rejectionexpressions(Section4.4),andprese ntexperimentsontheeffectivenessofthe interface(Section4.5). 4.1Example SupposethattheuserneedstoentertheclinicaltrialsinFigures3.1and4.4,andthesystem initiallyhasnoinformationabouttherelatedte sts.Theuserhastodescribethetestsand questions,andthenspecifytheeligibilityconditions.Weassumethatherstentersthe trialinFigure3.1,andlateraddsthetrialinFigure4.4. 16

PAGE 27

Adding tests Modifying a test question Adding a yes/no question Adding a numeric Deleting questions Adding a multiplechoice question Change the cost and pain View all questions Add a new question View all questions Delete questions View all questions Add a new test View an old test Add a new question View all questionsView all questions Add a new question Change the test name(a)Testsandquestions. question Adding a yes/no choice question Adding a multiplequestion Adding a numeric General information Deleting questions Add a new question View all questions Delete questions View all questions Add a new question View all questionsView all questions Add a new question View all questions(b)Generalquestions.Figure4.1:Enteringtestsandquestions.Weshowthescreensbyrectanglesandtransitionsbetweenthembyarrows.Theboldrectanglesarethestartscreens. Adding clinical trials Initialize new criteria Finalize criteria View criteria Selecting tests Choose relevant tests Choose questions make a patient eligible Specify answers that an eligibility expression Arrange questions into Selecting questions Defining an expressionFigure4.2:Enteringeligibilitycriteria. Figure4.3:“Modifyingatest” screen.Thethreebuttonsontheleftareformovingtothestart screens;everyscreeninthesystemhasthesebuttons.17

PAGE 28

First,heusesthe “Addingtests” screentoenterthenewtests;weillustratetheentry oftwotestsinFigure4.5.Then,heenterstherelatedquestions;toenterquestionsfora specictest,heselectsthetestandclicks “Modify” (Figure4.6),whichtakeshimtothe “Modifyingatest” screen(Figure4.7). Toaddaquestion,theuserclickstheappr opriatebuttonatthebottom(Figure4.7) andthentypesthequestion(Figure4.8).Foramultiple-choicequestion,hehastoinclude theansweroptions(Figure4.8b);foranumericquestion,heneedstospecifytherangeof allowedvalues(Figure4.8c).Ifothertestsp rovidedataforthesamequestion,theuserhas toselectallrelatedtestsinthelowerbox(F igure4.8b).Thenewlyaddedquestionsappear onthe “Modifyingatest” screen(Figure4.9). Afteraddingthequestionsfora lltests,theusergoestothe “Addingclinicaltrials” screen,initializesanewtrial(Figure4.10) ,andselectsitforaddingeligibilityconditions.Hegetsthe “Selectingtests” screenandchoosesthetestsrelatedtothetrial(Figure4.11).Then,heselectsrelevantquesti onsandtheanswersthatmakeapatienteligible (Figure4.12). NowsupposethattheuserneedstoaddtheclinicaltrialinFigure4.4.Thenew eligibilityconditionsrequirealivertest,whi chisnotintheknowledgebase,andtheuser hastoaddtheinformationrelatedtothistest.Furthermore,hehastoaddthequestion aboutrecurrentcancertothemammographytest .Aftermakingtheseadditions,heisready toentertheeligibilitycriteria. Condition5includesadisjunction,whichrequiresthe “Combinedquestion” option atthebottomofthequestionsscreen(Figure4 .12).Theusercheckstheelementsofthedisjunctivequestion,marksthea ppropriateanswers,andclicks “Combinedquestion,” which takeshimtothescreenforcomposinglogicale xpressions(Figure4.13).Afterentering Condition5,headdstheothercriteriausingthe “Simplequestions” option(Figure4.12). 18

PAGE 29

Generalinformation 1.Thepatientisfemale. 2.Sheisatleasttwenty-sevenyearsold. Mammogram: Costis$150,painlevelis1 3.CancerstageisIII. 4.Cancerisnotrecurrent. 5.Either € thetumorisatleasttwocentimeters,or € thecancerisnotinvasiveand atleasttwolymphnodeshavetumorcells. Livertest: Costis$150,painlevelis1 6.ThepatienthasnohepatitisB. 7.Thepatienthasnoliverinfections. Figure4.4:Eligibilitycriteriawithadisjunctivecondition. (a)Mammographytest. (b)Biopsytest.Figure4.5:Addingnewtests.19

PAGE 30

Figure4.6:Selectingatestforenteringtherelatedquestions. Figure4.7:“Modifyingatest” screen.Thesystemhasnoinformationaboutrelatedquestions,and theuserclicksoneofthebottombuttonsformovingtoaquestion-entryscreen.20

PAGE 31

(a)Yes/noquestion.(b)Multiple-choicequestion. (c)Numericquestion.Figure4.8:Addingnewquestions.Theusertypesaquestionandansweroptions.Ifthequestion isrelatedtoseveraltests,theusershouldcheckallthesetests. Figure4.9:“Modifyingatest” screenwithalistofquestions.21

PAGE 32

Figure4.10:Addinganewclinicaltrial. Figure4.11:Choosingtestsandquestiontypes.22

PAGE 33

Figure4.12:Selectingquestionsandanswers.Theuserchecksthequestionsforthecurrentclinical trialandmarkstheanswersthatsatisfytheeligibilitycriteria. Figure4.13:Combiningquestionsintoalogicalexpression.23

PAGE 34

4.2TestsandQuestions Wenowdescribethesix-screeninterfacefora ddingtestsandquestions(Figure4.1a).The startscreenallowsviewingtheavailabletestsanddeningnewones,whereastheother screensareformodifyingtestsandaddingrelatedquestions. 4.2.1AddingTests WeshowthestartscreeninFigure4.5;itsleft-handsideallowsviewingquestionsand goingtoamodicationscreen.Iftheuserselectsatestandclicks “View,” thesystem showstherelatedquestionsatthebottomo fthesamescreen(Figure4.14).Ifheclicks “Modify,” itdisplaysthe “Modifyingatest” screen(Figure4.7).Theright-handsideof thestartscreenallowsaddinganewtestbyspecifyingitsname,cost,andpainlevel. 4.2.2ModifyingaTest Thetest-modicationscreenshowstheinform ationaboutaspecictest,whichincludesthe testname,cost,painlevel,andrelatedquesti ons(Figure4.9).Theusercanchangethetest name,cost,andpainlevelbyenteringnewvaluesandclicking “Change.” Thefourbottom buttonsallowmovingtothescreensforadding newquestionsanddeletingoldquestions. 4.2.3AddingaQuestion Weshowthescreenforaddingyes/noquestionsinFigure4.8(a),multiple-choicequestions inFigure4.8(b),andnumericquestionsinFigur e4.8(c).Theusercanenteranewquestion forthecurrenttest,alongwithasetofallowe danswers.Ifthequestionisalsorelatedto othertests,theuserhastomarktheminthelowerbox(Figure4.8b). 24

PAGE 35

4.2.4DeletingQuestions Thisscreen(Figure4.15)isforremovingold andincorrectlyenteredquestions.Theuser hastomarkunwantedquestionsandclick “Delete.” 4.2.5GeneralInformation Thegeneralquestionsincludesex,age,ando therpersonaldata,collectedwithoutmedical tests.Themechanismforaddingsuchquesti onsconsistsofvescreens(Figure4.1b),and theuseraddsgeneralquestionsinthes amewayastest-relatedquestions. Figure4.14:Viewingthequestionsforaspecictest.25

PAGE 36

4.3EligibilityConditions Wenextdescribethemechanismforenteringeligibilitycriteria,whichconsistsoffour screens(Figure4.2). 4.3.1AddingClinicalTrials Thestartscreen(Figure4.16)allowstheusertoinitializeanewclinicaltrial,viewthe criteriaforoldtrials,andnalizecompletedtrials.Thelowerpartofthescreenisforinitializinganewtrial,whichrequiresenterin gthetrial'snameanduniquenumber.Theupper partisalistoftrialswithunnishedeligibilitycriteria.Theusercanviewthequestionsfor anunnishedtrialbyclicking “View,” andhecangotoamodicationscreenbyclicking “Modify.” Aftercompletingtheeligibilitycriteria,theusernalizesthetrialbyclicking “Activate.” Thelistofnalizedtrialsisinthemiddleofthescreen;theusercanviewthese trials,buthecannotmodifythem. 4.3.2SelectingTests Iftheuserclicks “Modify” onthestartscreen,thesystemdisplaysthetest-selectionscreen (Figure4.11).Theuserthenchoosesrela tedtestsandquestiontypes,andclicks “Continue” togetthequestionlist.Forinstance, ifhechoosesmammogramandbiopsyon theleft,andthetoptwoquestiontypesontheright,thenhegetsalistofallyes/noand multiple-choicequestionsrelatedtothemammogramandbiopsy. 4.3.3SelectingQuestions Thenextscreen(Figure4.12)allowstheusertoselectspecicquestionsandmarkanswers thatmakeapatienteligible.Foramultiple-choicequestion,theusermayspecifyseveral eligibilityoptions;forexample,apatientmaybeeligibleifhercancerstageisIIorIII.For 26

PAGE 37

anumericquestion,theuserhastospecifyar angeofvalues;forexample,apatientmay beeligibleifherageisbetween0and45years.Iftheuserclicks “Simplequestions,” the systemgeneratesaconjunctionoftheselected criteria.Iftheeligibilityconditionsinvolve amorecomplexexpressi on,theuserhastoclick “Combinedquestion,” whichtakeshim tothescreenforcomposinglogicalexpressions. 4.3.4DeninganExpression Thisscreen(Figure4.13)allowstheusertoa rrangetheselectedquestionsintoanexpressionthatincludesnestedconjunctionsanddisj unctions;however,thesystemdoesnotallow negations. 4.4LogicalExpressions Whentheusernalizesaclinicaltrial,thesystemcombinestheeligibilitycriteriainto anacceptanceexpression,andthengenerates thecorrespondingrejectionexpression.In Figure4.17,wegiveanalgorithmthatconstruc tstherejectionexpressionbyrecursive applicationofDeMorgan'slaws;theresultingexpressiondoesnotincludenegations. IfthesystemusestheorderingheuristicsdescribedinSection3.3,ithastoconvertthe acceptanceandrejectionexpressionsintoadisj unctivenormalform,thatis,adisjunctionof conjunctions;weuseastandardconversionalgorithm [ Kenneth,1988;CramaandHammer, 2001 ] ,summarizedinFigure4.18.Forinstance,iftheeligibilitycriteriaareasshownin Figure4.19(a),thesystemgeneratestheaccep tanceexpressioninFigure4.19(b)andthe rejectionexpressioninFigure4.19(c). 27

PAGE 38

Figure4.15:Deletingtwoquestions. Figure4.16:“Addingclinicaltrials” screen.Itallowstheusertoaddnewtrials(bottompart), modifyandnalizeeligibilitycriteria(top),andviewthenalizedcriteria(middle).28

PAGE 39

Negate-Expression ( bool-exp ) Theinputisalogicalexpression, bool-exp ,whichrepresentseligibilitycriteria. Determinewhether bool-exp isaconjunction,disjunction, yes/noquestion,multiple-choicequestion,ornumericquestion. Calltheappropriatesubroutinebelowandreturntheresultingexpression. Negate-Conjunction ( bool-exp ) Theinputisaconjunctiveexpression;thatis, bool-exp is“ sub-exp and sub-exp and ... ” New-Exps := Foreveryterm sub-exp oftheconjunction bool-exp : New-Exps := New-Exps { Negate-Expression ( sub-exp ) } Returnthedisjunctionofalltermsin New-Exps. Negate-Disjunction ( bool-exp ) Theinputisadisjunctiveexpression;thatis, bool-exp is“ sub-exp or sub-exp or ... ” New-Exps := Foreveryterm sub-exp ofthedisjunction bool-exp : New-Exps := New-Exps { Negate-Expression ( sub-exp ) } Returntheconjunctionofalltermsin New-Exps. Negate-Yes-No ( bool-exp ) Theinputisayes/noquestion. If bool-exp is" Variable =YES,"thenreturn" Variable {NO,UNKNOWN} ." If bool-exp is" Variable =NO,"thenreturn" Variable {YES,UNKNOWN} ." If bool-exp is" Variable {YES,UNKNOWN} ,"thenreturn" Variable =NO." If bool-exp is" Variable {NO,UNKNOWN} ,"thenreturn" Variable =YES." Negate-Multiple-Choice ( bool-exp ) Theinputisamultiple-choicequestion;thatis, bool-exp is“ Variable Option-Set. ” Let All-Options bethesetofallansweroptionsfor Variable New-Options := All-Options Š Option-Set. (Thissetdifferenceincludesallanswersthatarenotin Option-Set .) Return" Variable New-Options. Negate-Numeric ( bool-exp ) Theinputisanumericquestion;thatis, bool-exp is“ Variable [ Min Max ] ” Let"[ Lower Upper ]"betherangeofallowedvaluesfor Variable ; thatis,thevalueof Variable isalwaysbetween Lower and Upper If Min = Lower ,thenreturn" Variable ( Max Upper ]." If Max = Upper ,thenreturn" Variable [ Lower Min )." Return" Variable [ Lower Min ) ( Max Upper ]." Figure4.17:Constructingarejectionexpression.The Negate-Expression procedureinputsan acceptanceexpressionandrecursivelyprocessesitssubexpressions.29

PAGE 40

Normalize ( bool-exp ) Theinputisalogicalexpression, bool-exp ;theoutput isanequivalentexpressionindisjunctivenormalform. If bool-exp isanequality,inequality,or"set-element"test, thenreturn bool-exp If bool-exp isadisjunction sub-exp1 sub-exp2, then norm-exp1:= Normalize ( sub-exp1); norm-exp2:= Normalize ( sub-exp2); return norm-exp1 norm-exp2. If bool-exp isaconjunction sub-exp1 sub-exp2, then norm-exp1:= Normalize ( sub-exp1); norm-exp2:= Normalize ( sub-exp2); return Merge ( norm-exp1, norm-exp2). Merge ( norm-exp1, norm-exp2) Theinputistwologicalexpressions, norm-exp1and norm-exp2,indisjunctivenormalform; theoutputisadisjunctivenormalformoftheirconjunction, norm-exp1 norm-exp2. New-Exps := 0. Foreveryterm sub-exp1of norm-exp1: Foreveryterm sub-exp2of norm-exp2: New-Exps := New-Exps { sub-exp1 sub-exp2} Returnthedisjunctionofalltermsin New-Exps Figure4.18:Convertinganexpressionintoadisjunctivenormalform.The Normalize procedure inputsanexpressionwithoutnegations,whichrepresentsacceptanceorrejectionconditions,and generatesanequivalentexpressionindisjunctivenormalform.30

PAGE 41

sex =FEMALEandage [27 150]andcancer-stage =IIIandrecurrent =NOand( tumor-size [2 25]or( invasive =NOandlymph-nodes [2 100] ))andhepatitis =NOandliver-infections =NO(a)Eligibilitycriteria. sex =FEMALEandage [27 150]andcancer-stage =IIIandrecurrent =NOandtumor-size [2 25]andhepatitis =NOandliver-infections =NO or sex =FEMALEandage [27 150]andcancer-stage =IIIandrecurrent =NOandinvasive =NOandlymph-nodes [2 100]andhepatitis =NOandliver-infections =NO (b)Acceptanceexpression.sex =MALEorage [0 27)orcancer-stage {I,II,IV}orrecurrent {YES,UNKNOWN}or( tumor-size [0 2)andinvasive {YES,UNKNOWN} )or( tumor-size [0 2)andlymph-nodes [0 2) )orhepatitis {YES,UNKNOWN}orliver-infections {YES,UNKNOWN}(c)Rejectionexpression. Figure4.19:AcceptanceandrejectionexpressionsfortheeligibilitycriteriainFigure4.4.We representbothexpressionsasdisjunctivenormalformswithoutnegations.31

PAGE 42

4.5EntryTime Toevaluatetheinterface,wehaverunexperimentswithsevennoviceusers.Allparticipants havebeenundergraduatestudents,whohadnopr iorexperiencewiththeinterface.First, everyuserhasenteredfoursetsofmedicaltests;eachsethasincludedthreetestsandten questions.Then,eachuserhasaddedeligibility expressionsfortenbreast-cancertrials usedattheMofttCancerCenter;thenumberofquestionsinaneligibilityexpressionhas variedfromtentothirty-ve. Wehavemeasuredtheentrytimeforeachtestsetandeachclinicaltrial(Tables4.1and4.2). WeshowthemeantimeforeverytestsetinFigure4.20(left),andthetimeperquestionfor thesamesetsinFigure4.20(right).Allusershaveenteredthetestsetsinthesameorder, from1to4;sincetheyhadnopriorexperience, theirperformancehasimprovedduringthe experiment.InFigure4.21,wegivesimila rgraphsfortheentryofclinicaltrials. InFigure4.22,weplotthedependencyoftheentrytimeonthesizeofaneligibility expression,fortheeighttrialsenteredafter theinitiallearningperiod.Theresultssuggest thatthetimelinearlydependsonthenumbero fquestions,whichmeansthatthetimeper questiondoesnotdependonth ecomplexityofanexpression. Theexperimentshaveshownthatnovicescanefcientlyusetheinterface;theyquickly learnitsfullfunctionality,andtheirlearning curveattensafteraboutanhour.Theaverage timeperquestionis31secondsfortheentryofmedicaltestsand37secondsforeligibility criteria,whichmeansthatausercanenterall breast-cancertrialsusedatMofttinabout ninehours.Ontheotherhand,codingthesamet rialswithouttheinterfaceisprojectedto takesevenweeksoffull-timework. 32

PAGE 43

Table4.1:Entrytimeformedicaltestsandrelatedquestions.Wegivethetimesforsevenusers, whohaveenteredfoursetsoftests.Everysetincludesthreetestsandtenquestions. Num.of Entrytime(seconds) atestset UserA UserB UserC UserD UserE UserF UserG Mean 1 575 726 575 874 412 420 468 579 2 348 505 375 430 383 300 345 390 3 339 430 345 338 323 275 321 339 4 303 355 382 302 336 205 314 316 Table4.2:Entrytimeforeligibilitycriteria.Weshowtheresultsforsevenusers;eachuserhas constructedeligibilityexpressionsfortenclinicaltrials.Thenumberofquestionsinanexpression variesfromtentothirty-ve. Num.of Num.of Entrytime(seconds) atrial questions UserA UserB UserC UserD UserE UserF UserG Mean 1 10 1380 590 406 566 970 420 563 586 2 12 225 322 580 700 640 437 475 526 3 15 443 466 570 340 775 300 507 493 4 18 622 443 812 712 1080 497 570 686 5 21 630 602 746 722 1230 828 760 815 6 27 683 597 700 612 972 882 579 724 7 28 753 742 1032 880 995 950 889 915 8 29 763 634 860 722 1020 763 865 811 9 30 431 561 623 460 765 443 605 576 10 35 1168 900 1265 1085 1555 1007 1160 1162 33

PAGE 44

1 2 3 4 0 200 400 600 800 1000 1200 1400 1600 number of a test setentry time (sec) 1 2 3 4 0 20 40 60 80 100 number of a test settime per question (sec) Figure4.20:Entrytimeformedicaltestsets(left)andthemeantimeperquestionforeachset (right).Weplottheaverageperformance(dashedlines)andthetimeofthefastestandslowestusers (verticalbars). 1 2 3 4 5 6 7 8 9 10 0 200 400 600 800 1000 1200 1400 1600 number of a clinical trialentry time (sec) 1 2 3 4 5 6 7 8 9 10 0 20 40 60 80 100 number of a clinical trialtime per question (sec) Figure4.21:Entrytimeforeligibilitycriteria.Weshowtheaveragetimeforeachclinicaltrialand thecorrespondingtimeperquestion(dashedlines),alongwiththeperformanceofthefastestand slowestusers(verticalbars). 0 5 10 15 20 25 30 35 0 200 400 600 800 1000 1200 1400 1600 number of questionsentry time (sec) 0 5 10 15 20 25 30 35 0 20 40 60 80 100 number of questionstime per question (sec) Figure4.22:Dependencyofthetimeonthenumberofquestionsinaneligibilityexpression.We plotthetimeofenteringeligibilityexpressions(left)andthecorrespondingtimeperquestion(right). Theresultsshowthatthetimeperquestiondoesnotdependonthecomplexityofanexpression.34

PAGE 45

Chapter5 ConcludingRemarks Wehavedevelopedknowledge-acquisitiontoolsforanexpertsystemthatselectsclinical trialsforcancerpatients.Wehavedescribedtherepresentationofselectioncriteriaanda web-basedinterfaceforaddingnewtrials. AlthoughcancerresearchatMoftthasprovidedthemotivationforthiswork,thedevelopedtoolsarenotlimitedtocancer,andwe canusethemtoenterselectioncriteriaforclinicaltrialsrelatedtootherdiseases. Theexperimentshaveshownthatausercanenteraclinicaltrialintentotwentyminutes,whereasprogrammingthesameknowle dgewithouttheinterfacetakesabouttwenty hours(personalcommunicationwithKokku).Nov icescanreadilyusetheinterfacewithout priorinstructions,andtheyreachtheirfullspeedafteraboutanhour. Theexperimentshavealsorevealedseverallimitationsofthedevelopedtools,and addressingthemmaybeasubjectoffuturework .First,theexpertsystemdoesnotestimateprobabilitiesofmedical-testresults.Weconjecturethatintegrationofprobabilistic methodswiththecurrentheuristicsmayfurthe rreducethecostofselectedtests.Second, thesystemdoesnotparsethetextofquestions ,anditcannotrecognizeidenticalorrelated questions.Ifauseraccidentlyentersthesamequestiontwice,thesystemwilltreatitastwo differentquestions.Third,theinterfacedoe snotallowausertoencodelogicalrelationshipsamongquestions.Forexample,weca nnotspecifythatapost-menopausalwomanis neverpregnant.Iftheknowledgebaseinclude smenopauseandpregnancyquestions,the systemmayaskaboutpregnancyevenafterl earningthatapatientispost-menopausal. 35

PAGE 46

References [ Bhanja etal. ,1998 ] SanjuctaBhanja,LynnM.Fletcher,LawrenceO.Hall,DimtryB. Goldgof,andJeffreyP.Krischer.Aqualitati veexpertsystemforclinicaltrialassignment.In ProceedingsoftheEleventhInternationalFloridaArticialIntelligenceResearchSocietyConference ,pages8488,1998. [ Blythe etal. ,2001 ] JimBlythe,JihieKim,SuryaRamachandran,andYolandaGil.An integratedenvironmentforknowledgeacquisition.In ProceedingsoftheInternational ConferenceonIntelligentUserInterfaces ,pages1320,2001. [ BooseandGains,1990 ] JohnH.BooseandBrianRGains. TheFoundationsofKnowledgeAcquisition .AcademicPress,SanDiego,CA,1990. [ Bouaud etal. ,1998 ] JacquesBouaud,BriggiteS eroussi, Eric-CharlesAntoine,Mary Gozy,DavidKhayat,andJean-Frans oisBoisvieux.Hypertextualnavigationoperationalizinggenericclinicalpracticeguidelinesforpatient-specictherapeuticdecisions. JournaloftheAmericanMedicalInformaticsAssociation ,5(suppl.):488492,1998. [ Bouaud etal. ,2000 ] JacquesBouaud,BriggiteS eroussi, Eric-CharlesAntoine,Laurent Zelek,andMarcSpielmann.ReusingONCODOC,aguideline-baseddecisionsupport system,acrossinstitutions:Asuccessfulexpe rimentinsharingmedicalknowledge.In ProceedingsoftheAmericanMedicalInformaticsAssociationAnnualSymposium ,volume7,2000. [ BuchananandShortliffe,1984 ] BruceG.BuchananandEdwardH.Shortliffe. RuleBasedExpertSystems:TheMYCINExperimentsoftheStanfordHeuristicProgramming Project .Addison-Wesley,Reading,MS,1984. [ Buchanan etal. ,1969 ] BruceG.Buchanan,GeorgiaL .Sutherland,andEdwardA. Feigenbaum.HeuristicDENDRAL:Aprogramforgeneratingexplanatoryhypotheses inorganicchemistry.InBernardMe ltzerandDonaldMichie,editors, MachineIntelligence ,volume4,pages209254.EdinburghUniversityPress,Edinburgh,United Knigdom,1969. [ Carlson etal. ,1995 ] RobertW.Carlson,SamsonW.Tu,NancyM.Lane,TzeL.Lai, CarolA.Kemper,MarkA.Musen,andEdwardH.Shortliffe.Computer-basedscreening ofpatientswithHIV/AIDSforclinicaltrialeligibility. OnlineJournalofCurrentClinical Trials ,4(179),1995. [ Cimino etal. ,1988 ] JamesCimino,L.J.Mallon,andOctoG.Barnett.AutomatedextractionofmedicalknowledgefromMEDLINE.In ProceedingsoftheTwelfthAnnual SymposiumonComputerApplicationsinMedicalCare ,pages180184,1988. 36

PAGE 47

[ ClanceyandSoloway,1990 ] WilliamJ.ClanceyandElliotSoloway. Articialintelligenceandlearningenvironments .Elsevier,Amsterda m,Netherlands,1990. [ Clancey,1993 ] WilliamJ.Clancey.Notesonepistemologyofarule-basedexpertsystem. ArticialIntelligence ,59(1/2):191204,1993. [ Clancey,1995 ] WilliamJ.Clancey.Thelearningprocessintheepistemologyofmedical information. MethodsofInformationinMedicine ,34(1/2):122130,1995. [ CramaandHammer,2001 ] YvesCramaandPeterL.Hammer. BooleanFunctions Sringer-Verlag,Berlin,Germany,2001. [ Duff etal. ,2000 ] FranckLeDuff,AnitaBurgun,MireilleCleret,BrunoPouliquen,VictoireBarac'h,andPierreLeBeux.Knowledgeacquisitiontoqualifyuniedmedical languagesysteminterconceptualrelationships.In ProceedingsoftheAmericanMedical InformaticsAssociationAnnualFallSymposium ,2000. [ Eriksson,1993 ] HenrikEriksson.Specicationandgenerationofcustom-tailored knowledge-acquisitiontools.In ProceedingsoftheThirteenthInternationalJointConferenceonArticialIntelligence ,volume1,pages510518,1993. [ Falloweld etal. ,1997 ] LesleyFalloweld,D.Ratcliffe,andRobertSouhami.Clinicians'attitudestoclinicaltrialsofcancertherapy. EuropeanJournalofCancer 33(13):22212229,1997. [ GennariandReddy,2000 ] JohnH.GennariandMadhuReddy.Participatorydesignand aneligibilityscreeningtool.In ProceedingsoftheAmericanMedicalInformaticsAssociationAnnualFallSymposium ,pages290294,2000. [ GilandTallis,1997 ] YolandaGilandMarceloTallis.Ascript-basedapproachtomodifyingknowledgebases.In ProceedingsoftheFourteenthNationalConferenceonArticial Intelligence ,pages377383,1997. [ HahnandSchnattinger,1997a ] UdoHahnandKlemensSchnattinger.Anempiricalevaluationofasystemfortextknowledgeacquisition.In ProceedingsoftheTenthEuropean KnowledgeAcquisitionWorkshop ,pages129144,1997. [ HahnandSchnattinger,1997b ] UdoHahnandKlemensSchnattinger.Knowledgemining fromtextualsources.In ProceedingsoftheSixthInternationalConferenceonInformationandKnowledgeManagement ,pages8390,1997. [ HahnandSchnattinger,1998 ] UdoHahnandKlemensSchnattinger.Atextunderstander thatlearns.In ProceedingsoftheThirty-SixthAnnualMeetingoftheAssociationfor ComputationalLinguisticsandSeventeenthInternationalConferenceonComputational Linguistics ,volume1,pages476482,1998. [ HammondandSergot,1996 ] PeterHammondandMarekJ.Sergot.Computersupport forprotocol-basedtreatmentofcancer. JournalofLogicProgramming ,26(2):93111, 1996. 37

PAGE 48

[ JohnsonandMusen,1996 ] PeterD.JohnsonandMarkA.Musen.Developmentofa guidelineauthoringtoolwithPROTEGE-IIbasedontheDILEMMAgenericprotocoland guidelinemodel.TechnicalReportSMI-960620,StanfordMedicalInformaticsGroup, StanfordSchoolofMedicine,1996. [ Kenneth,1988 ] RosenH.Kenneth. DiscreteMathematicsandItsApplications .McGrawHill,NewYork,NY,thirdedition,1988. [ Keppel etal. ,2002 ] KennethG.Keppel,JeffreyN.Pearcy,andDianeK.Wagener.Trends inracialandethnic-specicratesfortheh ealthstatusindicators:UnitedStates,1990 98. HealthyPeople2000,StatisticalNotes ,23,2002. [ KimandGil,2000a ] JihieKimandYolandaGil.Acquiringproblem-solvingknowledge fromendusers:Puttinginterdependencymodelstothetest.In ProceedingsoftheSeventeenthNationalConferenceonArticialIntelligence ,pages223229,2000. [ KimandGil,2000b ] JihieKimandYolandaGil.Userstudiesofaninterdependencybasedinterfaceforacquiring problem-solvingknowledge.In ProceedingsoftheInternationalConferenceonIntelligentUserInterfaces ,pages165168,2000. [ Kokku etal. ,2002a ] PrincetonK.Kokku,LawrenceO.Hall,DimitryB.Goldgof,EugeneFink,andJeffreyP.Krischer.Acost-ef fectiveagentforclinicaltrialassignment. UnpublishedManuscript,2002. [ Kokku etal. ,2002b ] PrincetonK.Kokku,SavvasNikiforou,LawrenceO.Hall,DimitryB.Goldgof,EugeneFink,JeffreyP.Krische r,andTimIvanovkiy.Acost-effective agentforclinicaltrialassignm ent.UnpublishedManuscript,2002. [ KorverandJanssens,1993 ] M.KorverandA.R.Janssens.Developmentandvalidation ofHEPAR,anexpertsystemforthediagnosisofdisordersoftheliverandbiliarytract. MedicalInformatics ,16(3):259270,1993. [ KorverandLucas,1993 ] M.KorverandPeterJ.F.Lucas.Convertingarule-basedexpert systemintoabeliefnetwork. MedicalInformatics ,18(3):219241,1993. [ Kotwall etal. ,1992 ] CyrusKotwall,LeoJ.Mahoney,RobertE.Myers,andLindaDecoste.Reasonsfornon-entryinrandomizedc linicaltrialsforbreastcancer:Asingle institutionalstudy. JournalofSurgicalOncology ,50:125129,1992. [ Lederberg,1965 ] JoshuaLederberg.DENDRAL-64:Asystemforcomputerconstruction, enumerationandnotationoforganicmoleculesastreestructuresandcyclicgraphs.Part II.TechnicalReportN66-14074,NASAScienticandTechnicalAerospaceReports, 1965. [ Lederberg,1987 ] JoshuaLederberg.HowDENDRALwasconceivedandborn.In ProceedingoftheACMSymposiumontheHistoryofMedicalInformatics .NationalLibrary ofMedicine,1987. 38

PAGE 49

[ Lindberg etal. ,1993 ] DonaldA.B.Lindberg,BetsyL.Humphreys,andAlexaT.McCray.Theuniedmedicallanguagesystem. JournalofMethodsofInformationin Medicine ,4(32):281291,1993. [ Lucas etal. ,1989 ] PeterJ.F.Lucas,R.W.Segaar,andA.R.Janssens.HEPAR:Anexpert systemforthediagnosisofdisordersoftheliverandthebiliarytract. Liver ,9:266275, 1989. [ Lucas,1994 ] PeterJ.F.Lucas.RenementoftheHEPARexpertsystem:Toolsandtechniques. JournalofArticialIntelligenceinMedicine ,6(2):175188,1994. [ MarcusandMcDermott,1989 ] SandraMarcusandJohnP.McDermott.SALT:Aknowledgeacquisitionlanguageforproposeandrevisesystems. ArticialIntelligence 39(1):137,1989. [ Mendonc aandCimino,2000 ] EneidaA.Mendonc aandJamesJ.Cimino.Automatic knowledgeextractionfromMEDLINEcitations.In ProceedingsoftheAmericanMedical InformaticsAssocationAnnualFallSymposium ,pages575579,2000. [ Musen etal. ,1988 ] MarkA.Musen,DavidM.Combs,EdwardH.Shortliffe,and LawrenceM.Fagan.OPAL:Towardthecomputer-aideddesignofoncologyadvice systems.InPerryL.Miller,editor, SelectedTopicsinMedicalArticialIntelligence pages166180.Springer-Verlag,NewYork,NY,1988. [ Musen etal. ,1996 ] MarkA.Musen,SamsonW.Tu,AmarK.Das,andYuvalShahar.EON:Acomponent-basedapproachtoautoma tionofprotocol-Directedtherapy. Journal oftheAmericanMedicalInformaticsAssociation ,3(6):367388,1996. [ Musen etal. ,2000 ] MarkA.Musen,RayW.Fergerson,WilliamE.Grosso,NatalyaF. Noy,MonicaCrubzy,andJohnH.Gennari .Component-basedsupportforbuilding knowledge-acquisitionsystems.In ProceedingsoftheConferenceonIntelligentInformationProcessingoftheInternationalFederationforInformationProcessingWorld ComputerCongress ,2000. [ Musen,1989 ] MarkA.Musen. AutomatedGenerationofModel-BasedKnowledgeAcquisitionTools .MorganKaufmann,SanMateo,CA,1989. [ Ohno-Machado etal. ,1993 ] LucilaOhno-Machado,Eduard oParra,SuzanneB.Henry, SamsonW.Tu,andMarkA.Musen.AIDS2:Adecision-supporttoolfordecreasing physicians'uncertaintyrega rdingpatienteligibilityforHIVtreatmentprotocols.In ProceedingsoftheSeventeenthAnnualSymposiumonComputerApplicationsinMedical Care ,pages429433,1993. [ Ohno-Machado etal. ,1998 ] LucilaOhno-Machado,John H.Gennari,ShawnMurphy, NileshL.Jain,SamsonW.Tu,DianeE.O liver,andEdwardPattison-Gordon.The guidelineinterchangeformat:Am odelforrepresentingguidelines. JournalofAmerican MedicalInformaticsAssociation ,5(4):357372,1998. 39

PAGE 50

[ Onisko etal. ,1997 ] AgnieszkaOnisko,MarkJ.Druzdzel,andHannaWasyluk.ApplicationofBayesianbeliefnetworkst odiagnosisofliverdisorders.In Proceedingsofthe ThirdConferenceonNeuralNetworksandTheirApplications ,pages730736,1997. [ Papaconstantinou etal. ,1998 ] ConstantinosPapaconstantinou,GeorgiosTheocharous, andSridharMahadevan.Anexpertsystemfo rassigningpatientsintoclinicaltrials basedonBayesiannetworks. JournalofMedicalSystems ,22(3):189202,1998. [ Price,1990 ] ChrisJ.Price. KnowledgeEngineeringToolkits .EllisHorwood,Chichester, WestSussex,UnitedKnigdom,1990. [ Puerta etal. ,1992a ] AngelR.Puerta,JohnW.Egar,SamsonW.Tu,andMarkA. Musen.Amultiple-methodknowledge-acquisitionshellfortheautomaticgeneration ofknowledge-acquisitiontools. KnowledgeAcquisition ,4(2):171196,1992. [ Puerta etal. ,1992b ] AngelR.Puerta,HenrikEriksson,JohnW.Egar,andMarkA. Musen.Generationofknowledge-acquisitiontoolsfromreusabledomainontologies.TechnicalReportKSL-92-81,ComputerSc ienceDepartment,StanfordUniversity, 1992. [ RinglandandDuce,1988 ] GordonA.RinglandandDavidA.Duce. Approachesto KnowledgeRepresentation.AnIntroduction .ResearchStudiesPress,Letchworth, UnitedKnigdom,1988. [ RomackerandHahn,2001 ] MartinRomackerandUdoHahn.Semanticinterpretationof medicallanguagequantitativeanalysisand qualitativeyield.InSilvanaQuaglini,PedroBarahona,andSteenAndreassen,editors, ArticialIntelligenceinMedicine ,pages 258267.Springer-Verlag,Berlin,Germany,2001. [ Rothenuh etal. ,1996 ] ThomasE.Rothenuh,JohnH.Gennari,HenrikEriksson,AngelR.Puerta,SamsonW.Tu,andMarkA.Mu sen.Reusableontologies,knowledgeacquisitiontools,andperformancesystems:PROTEGE-IIsolutionstoSisyphus-2. InternationalJournalofHuman-ComputerStudies ,44(34):303332,1996. [ Rubin etal. ,1999 ] DanielL.Rubin,JohnH.Gennari,S andraSrinivas,AllenYuen,HerbertKaizer,MarkA.Musen,andJohnS.Silva.Toolsupportforauthoringeligibility criteriaforcancertrials.In ProceedingsoftheAmericanMedicalInformaticsAssociationAnnualSymposium ,pages369373,1999. [ Rubin etal. ,2000 ] DanielL.Rubin,JohnH.Gennari,andMarkA.Musen.Knowledge representationandtoolsupportforcriti quingclinicaltrialprotocols.In Proceedingsof theAmericanMedicalInformaticsAssociationFallSymposium ,pages724728,2000. [ Sager etal. ,1994 ] NaomiSager,MargaretLymanan dChristineBucknall,NgoNha,and LeoJ.Tick.Naturallanguage processingandtherepresantationofclinicaldata. AmericanMedicalInformaticsAssociation ,1(2):142160,1994. 40

PAGE 51

[ S eroussi etal. ,1999a ] BriggiteS eroussi,JacquesBouaud,and Eric-CharlesAntoine.Enhancingclinicalpracticegui delinecompliancebyinvolvingphysiciansinthedecision process.InWernerHorn,YuvalShahar,GregerLindberg,SteenAndreassen,and JeremyC.Wyatt,editors, ArticialIntelligenceinMedicine ,pages7685.SpringerVerlag,Berlin,Germany,1999. [ S eroussi etal. ,1999b ] BriggiteS eroussi,JacquesBouaud,and Eric-CharlesAntoine. Users'evaluationofONCODOC,abreastcancertherapeuticguidelinedeliveredatthe pointofcare. JournaloftheAmericanMedicalInformaticsAssociation ,6(5):384389, 1999. [ S eroussi etal. ,2001a ] BriggiteS eroussi,JacquesBouaud,and Eric-CharlesAntoine.ON-CODOC:Asuccessfulexperimentofcomputer -supportedguidelinedevelopmentand implementationinthetreatmentofbreastcancer. ArticialIntelligenceinMedicine 22(1):4364,2001. [ S eroussi etal. ,2001b ] BriggiteS eroussi,JacquesBouaud, Eric-CharlesAntoine,Laurent Zelek,andMarcSpielmann.UsingONCODOCasacomputer-basedeligibilityscreening systemtoimproveaccrualontobreastcancerclinicaltrials.InWernerHorn,Yuval Shahar,G.Lindberg,SteenAndreassen,andJ.Wyatt,editors, ArticialIntelligencein Medicine ,pages413430.Springer-Verlag,Berlin,Germany,2001. [ Shortliffe etal. ,1975 ] EdwardH.Shortliffe,RandallDavis,StantonG.Axline,BruceG. Buchanan,CordellC.Green,andStanleyCohe n.Computer-basedconsultationsinclinicaltherapeutics:ExplanationandruleacquisitioncapabilitiesoftheMYCINsystem. ComputersandBiomedicalResearch ,8:303320,1975. [ Shortliffe etal. ,1981 ] EdwardH.Shortliffe,A.CarlisleScott,MiriamB.Bischoff, WilliamvanMelle,andCharlotteD.Jacobs.ONCOCIN:Anexpertsystemforoncology protocolmanagement.In ProceedingsoftheSeventhInternationalJointConferencein ArticialIntelligence ,pages876881,1981. [ Shortliffe,1974 ] EdwardH.Shortliffe.MYCIN:ARule-BasedComputerProgramfor AdvisingPhysiciansRegardingAntimicrobialTherapySelection .PhDthesis,Computer ScienceDepartment,StanfordUniversity,1974. [ Shusaku,1998 ] TsumotoShusaku.Automatedextrac tionofmedicalexpertsystemrules fromclinicaldatabasesbasedonroughsettheory. InternationalJournalinInformation Sciences ,112:6784,1998. [ TallisandGil,1999 ] MarceloTallisandYolandaGil.Designingscriptstoguideusersin modifyingknowledge-basedsystems.In ProceedingsoftheSixteenthNationalConferenceonArticialIntelligence ,pages242249,1999. [ Tallis etal. ,1999 ] MarceloTallis,JihieKim,andYolandaGil.Userstudiesofknowledgeacquisitiontools:Methodol ogyandlessonslearned.In ProceedingsoftheTwelfth WorkshoponKnowledgeAcquis ition,ModelingandManagement ,1999. 41

PAGE 52

[ Tallis,1998 ] MarceloTallis.Ascript-basedapproachtomodifyingknowledge-basedsystems.In ProceedingsoftheTenthInnovativeApplicationsofArticialIntelligenceConference ,pages11831195,1998. [ Theocharous,1996 ] GeorgiosTheocharous.Anexpertsystemforassigningpatientsinto clinicaltrialsbasedonBayesiannetworks.Master'sthesis,ComputerScienceandEngineeringDepartment,UniversityofSouthFlorida,1996. [ Tu etal. ,1993 ] SamsonW.Tu,CarolA.Kemper,NancyM.Lane,RobertW.Carlson, andMarkA.Musen.Amethodologyfordeter miningpatients'eligibilityforclinical trials. JournalofMethodsofInformationinMedicine ,32(4):317325,1993. [ Wang etal. ,2001 ] DongwenWang,MorPeleg,SamsonW.Tu,EdwardH.Shortliffe,and RobertA.Greenes.Representationofclinicalpracticeguidelinesforcomputer-based implementations. MedicalInformatics ,10:285289,2001. [ Yusuf etal. ,1990 ] SalimYusuf,PeterHeld,K.K.Teo,andElizabethR.Toretsky.Selectionofpatientsforrandomizedcontrolledtri als:Implicationsofwideornarroweligibilitycriteria. StatisticsinMedicine ,9:7386,1990. 42


xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001413333
003 fts
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 030703s2002 flua sbm s000|0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0000029
035
(OCoLC)50138092
9
AJJ0749
b SE
SFE0000029
040
FHM
c FHM
049
FHME
090
QA76
1 100
Nikiforou, Savvas.
0 245
Selection of clinical trials
h [electronic resource] :
knowledge representation and acquisition /
by Savvas Nikiforou .
260
[Tampa, Fla.] :
University of South Florida,
2002.
502
Thesis (M.S.C.S.)--University of South Florida, 2002.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 42 pages.
520
ABSTRACT: When medical researchers test a new treatment procedure, they recruit patients with appropriate health problems and medical histories. An experiment with a new procedure is called a clinical trial. The selection of patients for clinical trials has traditionally been a labor-intensive task, which involves matching of medical records with a list of eligibility criteria. A recent project at the University of South Florida has been aimed at the automation of this task. The project has involved the development of an expert system that selects matching clinical trials for each patient.If a patient's data are not sufficient for choosing a trial, the system suggests additional medical tests. We report the work on the representation and entry of the related selection criteria and medical tests. We first explain the structureof the system's knowledge base, which describes clinical trials and criteria for selecting patients. We then present an interface that enables a clinician to add new trials and selection criteria without the help of a programmer. Experiments show that the addition of a new clinical trial takes ten to twenty minutes, and that novice users learn the full functionality of the interface in about an hour.
590
Adviser: Fink, Eugene
650
Clinical trials
x Data processing.
690
Dissertations, Academic
z USF
Computer Science
Masters.
Knowledge acquisition (Expert systems)
653
agents.
medical.
expert systems.
knowledge representation.
knowledge acquisition.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.29