The organ allocation process: a natural extension of the ...jvazquez/docs/Carrel-AIcom.pdf2 Vazquez-Salceda´ et al. / The organ allocation process: a natural extension of the Carrel

1

Theorganallocationprocess:anaturalextensionof theCarrelAgentMediatedElectronicInstitution

Javier Vazquez-Salceda�, UlisesCortes

�,

JulianPadget�, Antonio Lopez-Navidad

�,

FranciscoCaballero�,�

Technical Universityof Catalonia.c/Jordi Girona1 & 3. 08034Barcelona,Spain.E-mail:

�jvazquez,ia� @lsi.upc.es�

Departmentof ComputerScience. UniversityofBath.Bath,BA27AY. UnitedKingdom.E-mail: [email protected]�BancdeTeixits.Hospitaldela SantaCreui Sant

Pau.c/ St.AntoniM. Claret,167.E08025Barcelona,Spain.

In thispaperwe extendtheformalizationof Carrel, a virtualorganizationfor theprocurementof tissuesfor transplant,todealalsowith organs.We focuson theorganallocationpro-cessto show how it canbeformalizedwith the ISLANDERspecificationlanguageandwealsopresentthefirst versionofa mechanismto federateseveral geographically-distributedCarrelplatformswith theobjectiveof addressingsomeof thetechnicalissuesin the establishmentof a pan-Europeantis-sueandorgandistributionservice.

1. Intr oduction

Organtransplantsareamongthe mostcomplicatedmedicalproceduresperformedtoday. Organtransplan-tationfrom humandonorsarebecomingcommonlife-saving therapiesand the only option available whenthereis majordamageto or a malfunctionin anorgan.Today, mostdonatedorgansandtissuescomefrom pa-tientswho arepronouncedbraindeadasresultof dis-easeor injury. At the time of writing, more thanonemillion peoplein theworld havesuccessfullyreceivedan organ,and thereafter, in most cases,beenable toleadnormal lives.But while theseoperationsarebe-comingmorecommonplace,thereis a majorshortageof organs.At the time of writing, tenpeopledie dailydue to the shortageof transplantableorgans,while a

Organ Allowabletime from Donorto Recipient

Heart 4 to 5 (hours)Lung 5 to 6 (hours)

Heart-Lung 4 to 6 (hours)Pancreas 12 to 15 (hours)

Liver 12 to 18 (hours)Kidney 15 to 18 (hours)

Fig. 1. Cold ischaemiatime for solid organtransplantation

new nameis addedto thetransplantwaiting list every18minutes.

Over theyears,transplanttechniqueshave evolved,knowledge of donor-recipient compatibility has im-provedandsohaveimmuno-suppressantdrugregimes,leadingnotonly to anincreasein thedifferentkindsoforgansthatcanbetransplanted,but alsoin therangeoftransplants,moving beyondorgans(heart,liver, lungs,kidney, pancreas)to tissues(bones,skin, corneas,ten-dons).The allocationprocessfor tissuesis quite dif-ferentfrom thatfor organs,becauseof thetime organscanbepreservedoutsidethehumanbody(seetable1).

Tissuesareclustersof relatively homogeneouscells,so the optimal temperaturefor preservation of all thecells composingthe tissueis almostthe same.Thus,tissuescanbe preserved for several days(from sevendays in the caseof corneasto years in the caseofbones)in tissuebanks.For tissues,theallocationpro-cessis demand-driven,triggeredwhenthereis a recip-ientwith aneedfor a certaintissue,atwhich timesev-eraltissuebanksaresearchedfor a suitablepiece.

Organs,on the otherhand,arevery complex struc-tureswith severalkindsof cell typeseachwith differ-entoptimalpreservationtemperatures.That fact leadsto quite shortpreservation times(hours),no needforanorganbank,andanallocationprocessthatis supply-driven, triggered when a donor appears,taking theform of a searchfor a suitablerecipientin somenum-berof hospitals.

AI CommunicationsISSN0921-7126,IOS Press.All rightsreserved

2 Vazquez-Salcedaetal. / Theorganallocationprocess:anaturalextensionof theCarrelAgentMediatedElectronicInstitution

1.1. Theneedof software systemsfor theorganandtissuemanagement

As explainedin [22], the increasingrateof successof tissuetransplants,thatis literally asecondchanceatlife, is leadingto anincreasein thenumberof requestsandthisvolumeis startingto overwhelmthehumanco-ordinatorsathospitalswho areresponsiblefor manag-ing the transplantprocess,andfurthermoreis leadingto tissueloss,becauseavailable tissuesarenot beingassignedbeforethey exceedtheir shelf life asa resultof thelengthof time it is takingto processrequests.

In the caseof organs,successfultransplantshavealsoled to anincreasein demandfor organsfor trans-plantationpurposes.However, thereis not an increas-ing volumeof donationsto matchthedemandandthisis creatinglistsof transplantcandidatesor waitinglists.Typically, thereis onelist per kind of organ.Becauseof theorganshortage,theprocessof managinganddis-tributing the organsthat areavailableis complex andoften surroundedby controversy. Much researchhasbeendone to enablethe definition and implementa-tion of policiesfor donoridentification(to increasethenumberof availabledonors)[10], organallocation(tofind a suitablerecipientfor eachorgan)[20,7,16] andin extraction,preservationandimplantprocedures(toincreasethechancesof success).

The relative scarcityof donorshasled to the cre-ationof internationalcoalitionsof transplantorganiza-tions.Thisnew, moregeographicallydistributed,envi-ronmentmakesan evenstrongercasefor the applica-tion of distributedsoftwaresystemsto solve:

– thedataexchangeproblem:exchangeof informa-tion is a major issue,as eachof the actorscol-lectsdifferentinformationandstoresit in differ-ent formats.Theobvious,andeasilystated,solu-tion is thedefinitionof standarddatainterchangeformats.

– the communicationproblem: countriestypicallyusedifferent languagesand terminologiesto tagthesameitemsor facts.Eitherastandardnotationor astandardontologyor evenatranslationmech-anismis neededto avoid misunderstandings.

– the coordination issues:in order to managere-questsat an internationallevel, thereis the needto coordinategeographicallydistributed surgeryteams,andto coordinatepiecedelivery at an in-ternationallevel.

– the variety of regulations: an additionalissueisthenecessityof accommodatingacomplex setof,in somecasesconflicting, nationaland interna-

tional regulations,legislationandprotocolsgov-erningtheexchangeof organs[20,7]. Theseregu-lationsalsochangeover time,makingit essentialthatthesoftwareis adaptable.

Thefirst two pointscanlargely beresolvedby stan-dardsoftwaresolutions.For instance,theEU projectsRETRANSPLANT andTECN have largely focusedonthecreationof a) standardformatsfor thestorageandexchangeof informationaboutpieces,donorsandre-cipientsamongorganizations,b) telematicnetworks,or c) distributed databases. Another projectESCU-LAPE usesconventionalsoftwareto help in checkingtissuehistocompatibility.

In the USA the United Network for OrganSharing(UNOS) is supportingtools like ULAM [19] that areusedfor thesimulationandanalysisof nationalcadav-erickidney andkidney-pancreasallocationpoliciesfortransplantationto permitcomparisonof multiple liverallocationpolicy proposalsso that the policiescanbetestedprior to implementation.

Thethird point(coordination)is harderto solvewithconventionalsoftware.But a sound,if relatively new,approachis the useof software agents. An Agent is acomputerprogramcapableof takingits own decisionswith no externalcontrol (autonomy), basedon its per-ceptionsof theenvironmentandtheobjectivesit aimsto reach[23]. It not only reactsto theenvironment(re-activity) but alsopro-activelytakesinitiatives.Theso-cial ability of agentsallow themto grouptogether(inagencies) sharingcommonobjectivesanddividing thetasksin ordertoachievethoseobjectives.All theseuse-ful attributessuggestthatmulti-agentsystemsarewell-suitedfor solvingcoordinationissues.

It is the last point (variety of regulationschangingover time) which underpinsour casefor theuseof so-calledelectronic institutions, whosepurposeis to pro-videover-archingframeworksfor interactionof agentscapableof reasoningaboutthe normsgoverningindi-viduals’ actions,in the sameway asphysicalinstitu-tions andsocial normsdo in the real world (see � 2).Electronicinstitutionsandthenormsthatgovernthemarethekey to a systemthat is ableto adaptautomati-cally to changesin regulations.

In summary, our proposaladdressesall four issuesidentifiedabove,by theuseof multi-agenttechnology,not only for coordinationand regulationbut also forservingasalanguageinterfaceamongteamsusingdif-ferentterminology, andactively distributing the infor-mationto beshared.

Vazquez-Salcedaetal. / Theorganallocationprocess:anaturalextensionof theCarrelAgentMediatedElectronicInstitution 3

1.2. Organizationof thepaper

In 2 we introduceour point of view of electronicinstitutionsassocial structuresandthe useof norms.In 3 wegiveadescriptionof Carrel ourelectronicor-ganizationfor the procurementof humanorgansandtissuesfor transplantationpurposes.Here,wefocusonthe organ allocation processto show how it can beformalizedwith theISLANDER [4] specificationlan-guage.In 4 we discussthe main characteristicsofCarrel andfinally presentsomeconclusionsin 5.

2. Electronic Institutions

Building aMulti-Agent System(MAS) is acomplextask,asdevelopersneedto find anacceptablebalancebetweenthe autonomythat agentsshouldhave in or-derto actin unforeseencircumstances,andthecontrolthedesignerwantsto haveoveremergentbehaviour inorderto ensurethat the systemwill achieve its goals.Oneway to tamethe complexity of building a MASis to createa centralizedcontroller, that is, a specificagent that ensurescoordination.Coordinator agentsareagentswhich havesomekind of controloverotheragents’goalsor, at leaston partof the work assignedto anagent,accordingto theknowledgeabouttheca-pabilitiesof eachagentthat is underthe CoordinatorAgent’scommand.However thisapproachis notgood,as this specialagentbecomesa bottleneck in the in-formationflow. An alternative is to distributenot onlythework loadbut alsothecontrolamongall theagentsin the system(distributedcontrol). That meansto in-ternalizecontrol of eachagent,which hasnow to beprovidedwith reasoningandsocialabilitiesto make itableto reasonaboutintentionsandknowledgeof otheragentsplustheglobalgoalsof thesocietyin orderto beableto coordinatesuccessfullywith othersandalsotoresolve conflictswhenthey arise.However, asMosesandTennenholtzstatein [13], in thosedomainswherethecostof a conflict is dear, or if conflict resolutionisdifficult, completelyindependentbehaviour becomesunreasonable.Thereforesomekind of structureshouldbedefinedin orderto easecoordinationin adistributedcontrol scenario.A goodoptiontakenfrom animalin-teractionsis thedefinitionof socialstructures.

Social structures1, definea social level where themulti-agentsystemis seenas a societyof entitiesin

1They arealsocalledArtificial SocialSystemsby Shoham,MosesandTennenholtzin [13,18]

order to enhancethe coordinationof agentactivities(suchasmessagepassingmanagementandthealloca-tion of tasksandresources)by definingstructuredpat-ternsof behaviour. Socialstructuresreducethedangerof combinatorialexplosion in dealingwith the prob-lems of agentcognition, cooperationand control, asthey imposerestrictionsto the agents’actions.Theserestrictionshaveapositiveeffect,asthey:

– avoid potentialconflicts,or easetheir resolution– makeeasierfor agivenagentto foreseeandmodel

the otheragents’behaviour in a closedenviron-mentandfit its own behaviour accordingly.

An Agent-MediatedElectronic Institution(e-institu-tion for short)is kind of socialstructure wherethein-teractionsamonga groupof agentsaregovernedby asetof explicit normsexpressedin a languagerepresen-tationthatagentscaninterpret.

Therootsof this ideacomefrom thestudyof humanorganizations.Most humaninteractionsaregovernedby conventionsor rulesof somesort,having their ori-gin in society(emergent)or the laws (codificationofemergent rules) that societyhasdeveloped.Thus wefind that all humansocieties,even the mostprimitiveones,have somekind of social constraintsupontheirmembersin order to structureand regulate the rela-tionsamongtheir members.Someof theseconstraintsarequite informal (taboos,customs,traditions)whilesomeothersareformally defined(written laws,consti-tutions).In fact,modernhumansocietieshave definedcollectionsof expectedbehavioural patternsthathaveaneffect in specificscenarios(suchasa shop,bank,aconversation,a lectureor anexclusiveclub).

DouglasNorth2 refersto this corpora of constraintsasinstitutions. In his studies[15], North hasanalyzedthe effect of institutionson the behaviour of humanorganizations(includinghumansocieties)andhecon-cludesthat institutionalconstraintseasehumaninter-action(reducingthecostof thisinteractionby ensuringtrust),shapingchoicesandmakingoutcomesforesee-able.By thecreationof theseconstraints,theorganiza-tionsandtheinteractionsthey requirecaneachgrow incomplexity while interactioncostsremainstaticor areeven reduced.Having establishedtheseinstitutionalconstraints,every competentparticipantin the institu-tion mayact—andexpectothersto act—accordingtoa list of rights,duties,andprotocolsof interaction.

The main reasongroundingthe creationof institu-tions is to createtrust whenthepartiesknow very lit-

2DouglasNorth received the Nobel Prizein 1993for his studieson therole of institutionsin theperformanceof organizations.


tle aboutothers.No institutions are necessaryin anenvironmentwherepartieshave completeinformationaboutothers(e.g. a village market wherevendorsandbuyersknow eachotherandinteractonaperiodicalba-sis).However, in environmentswith incompleteinfor-mation(e.g. internationalcommerce),cooperative so-lutions(basedon trust)couldbreakdown unlessinsti-tutionsarecreatedto providesufficient informationforindividualsto createtrustandto policedeviations.

The samestatementsand ideascan be brought tothe field of agents.An e-Institution is the modellingof aninstitution throughthespecificationof its normsin somesuitableformalism(s)thatcanbefollowedbyagents.One of the hypothesesof the field is that theessenceof an institution, throughits normsand pro-tocolscanbe capturedin a precisemachineprocess-ableform (this key ideaforms thecoreof thenascenttopicof institutionalmodelling).Themainobjective isto createa safeenvironmentwhereagentscantrust inotheragents,asany violationof anormsmayleadto acompensatingsanction.

Theeffect of normson agentsshouldnot beseenasa constraintbut a guidewhich reducesthecomplexityof theenvironment,becausetheoptionspaceis pruned,andhenceallows the agentto make betteruseof its(limited) resources.

3. An Institution for the distrib ution of organsandtissues

The Carrel agentplatform modelsan organizationthat receives a tissuerequestfrom one hospital andthentries to allocatethe besttissueavailablefrom allthe known tissuebanks.Within this organizationdif-ferententitiesplay differentrolesthat aredeterminedby their goalsat any particularmoment.Figure2 de-pictsall thepartiesthatinteractwith theCarrelsystem.Therearea)thehospitalsthatcreatethetissuerequests,b) the tissuebanks,and c) the nationalorgan trans-plantationorganizations,that own the agentplatformand act as observers (in the figure the organizationsin Spainare depicted:the Organizacion NacionaldeTransplantes3 (ONT) [17] andtheOrganitzacio CATa-lanadeTransplantaments4 (OCATT)). In theproposedsystemall hospitals,eventhoserunninga tissuebank,mustmaketheir requeststhroughCarrelin orderto en-surea fair distribution of piecesandto easethetrack-ingof all piecesfrom extractionto thetransplant,astheONT andOCATT requirefor organs(thatis,no insidertrading).

3NationalTransplantOrganization4CatalanTransplantOrganization

Fig. 2. Carrel: An Agent MediatedInstitution for TissuesAssign-ment

Theroleof theCarrelinstitutioncanbesummarizedin thefollowing tasks:

– to make surethat all the agentswhich enterintothe institution behave properly(that is, that theyfollow thebehaviouralnorms).

– to be up to dateaboutall the availablepiecesinthetissuebanks,andall therecipientsthatarereg-isteredin thewaiting lists.

– to checkthat all hospitalsandtissuebanksfulfiltherequirementsfor interactingwith Carrel.

– to ensurethatthecommitmentsmadewithin Car-rel arefulfilled.

– to coordinatethepiecedelivery from onefacilityto another.

– to record all incidents relating to a particularpiece.

The participationof hospitalsin Carrel is basedonthenotionof membership.That is, hospitalsbelongtoCarrelandrespectthe negotiation(assignation)rules,andtheagentsthatrepresenttheminsideCarrelareun-able to breaktheseconventions.A Hospital interactswith Carrelthroughthe TransplantCoordinationUnitAgency (UCTx). This agency (depictedin figure 16)serves as interfacebetweenthe surgeonsand Carrel.Whena surgeonneedsa piece,he makeshis requestthroughthe UCTx system,which analyzesthe infor-mation entered,addsinformation aboutthe recipientandfinally createsa Finder Agent, which goesto theinstitutionlooking for a suitablepiece.

The information requiredby the Finder Agent tolook for a piecein Carrel is enteredin an electronicSealedEnvelope. Theenvelopecontainsthefollowinginformation:

– Urgencylevel, thatworksasanelectronicpostagestampandsetstheurgency level of therequest(inSpain:normal,urgency-1 or urgency-0)


– Hospital identification, a certificateissuedby theCertificationAuthority associatedwith Carrel[2],to allow the authenticationof the senderof eachrequestto ensurethatonly FinderAgentswith re-questsfrom authorizedhospitalscanenterandne-gotiate.

– Tissueinformation(type,parameters,etc.) andre-cipientdata(age,sex, laboratoryanalysis,etc.).

– The selectionfunction which is composedof aset of rules,eachone being a constraintthe se-lectedpiece(e.g. a cornea)mustsatisfy. Someoftheseruleswill typically comefrom thepolicy ofthe transplantunit of the hospital,while the restwill beintroducedby thesurgeon,whocansettheconstraintsneededfor agivenrecipient.A rule oftheselectionfunctioncancontain: predicatesabout the piece: predicatesthat

describethe constraintsthe selectedtissuehasto satisfy, suchasthe ageof the donoror thedimensionsof thepieceitself. predicatesaboutthe tissuebank:predicatesthatcansetconstraintsaboutthetissuebankpreferredby thesurgeonor thehospital. predicatesabout the cost of the piece: apredicatethat can set a maximumcost forthepiece.Thiscostonly coverspieceextrac-tion and preservation, and is paid througha clearinghouseby the hospitalwhich re-ceivesthepiece.Thereis nopriceassociatedwith thepieceitself.

3.1. ExtendingtheCarrel institution

In order to extend the Carrel systempresentedin[22] to handleorganaswell astissuedistribution, wemust augmentit for the organ allocationprocess.Inmostof theofficial organallocationorganizations,theprocessis composedof two phases:

1. Eachhospitalinforms theorganizationaboutpa-tients that have beenaddedto or removed fromthewaiting list of thathospital,or patientseitherto beaddedto or removedfrom thenational-wideMaximumUrgency Level5 Waiting List.

2. When a donor appears,the hospital informs theorganizationof all the organssuitablefor dona-tion in theform of offerssentto theorganalloca-tion organization,which thenassignstheorgans.

5In SpaintheMaximumUrgency Level is calledUrgency-0

This processcanbe modelledformally by specify-ing interactionsbetweenagents.To give a formal de-scriptionof the interactionamongagentsin the Car-rel systemwe will follow the sameformalism usedfor the caseof tissues[22]. The ISLANDER formal-ism [4] views an agent-basedelectronicinstitution asa type of dialogical systemwhereall the interactionsinside the institution are the compositionof multipledialogic activities (messageexchanges).Theseinter-actions(calledillocutions [14]) arestructuredthroughagentgroup meetingscalled scenesthat follow well-definedprotocols.

However, insteadof creatinga separatemodel fortheorganallocationprocess,we will extendthemodelfor the tissueallocationprocess.Someof the scenesthatweredefinedfor thecaseof tissueswill besharedfor organsby extendingtheir functionalities,andafewnew scenesarecreated.The resultingsetof scenesisthefollowing:

– ReceptionRoom: the scenewhereall the exter-nal agentsidentify themselvesin order to be as-signedthe roles they are authorizedto play. Iftheseagentsarecarryingeithera requestfor oneor moretissuesor anoffer of oneor moreorgans,thenthis informationis checkedto makesurethatit is well-formed.

– ConsultationRoom: the scenewherethe institu-tion is updatedaboutany eventor incidentrelatedto a piece.Agentscomingfrom tissuebanksup-datetheinstitutionabouttissueavailability, whileagentscomingfrom hospitalsupdatethe institu-tion aboutwaiting lists andinformationon piece(organor tissue)reception,transplantoperationsandthehealthof recipients.

– ExchangeRoom: thescenewheretheassignationprocessis made.In fact, there are specific ex-changeroomsfor managingtissuerequests(Tis-sueExchangeRoom) andfor organoffers(OrganExchangeRoom).

– ConfirmationRoom: the scenewherethe provi-sional assignmentsmadein either a TissueEx-change Roomor a Organ Exchange Roomareconfirmedor cancelledbecauseof the arrival ofanotherrequestwith higherpriority. In caseof theconfirmedones,adeliveryplanis built.

Anotherkey elementof the ISLANDER formalismis thedefinitionof agentroles. Eachagentcanbeasso-ciatedwith oneor moreroles,andtheserolesdefinethescenesthe agentcanenterandthe protocolsit shouldfollow (the sceneprotocolsare definedas multi-roleconversationalpatterns).Therearetwo kindsof roles:


ConsultationRoom

ENTERReception

RoomEXIT

OrganExchange

Room

TissueExchange

Room

ConfirmationRoom

[y:hf,z:hi,w:hc,v:tb]�[x:rrm] �

[x:cfrm] �

[x:crm] �

[x:trm] �[x:orm] �

[y:hf]

� � �[y:hf,z:hc]

� � � �

[y:h

i,z:tb

]

� � � �

[y:hf] �[y:hf] �

[y:hf,z:hc] �[y:hf,z:hc] �

[y:hf] �

[y:hf] �

[y:hi,z:tb]

�

[y:hf]�

[y:hf,z:hc]�

Fig. 3. TheCarrelInstitutionperformative structure

the external roles (rolesfor incomingagents)andtheinstitutional roles (roles for agentsthat carry out themanagementof the institution).Theexternalrolesarethefollowing:HospitalFinder Agent(hf): agentssent by hospitals

with tissuerequestsor organoffers that areseenfrom the point of view of the institution as re-questsfor finding an acceptabletissueor recipi-ent,respectively.

HospitalContactAgent(hc): agents from a certainhospitalthatarecontactedby theinstitutionwhenanorganhasappearedfor arecipientthatis onthewaiting list of thathospital.Theagentthenentersthe institution to acceptthe organandto receivethedeliveryplan.

Hospital Inf ormation Agent(hi): agentssentby hos-pitalsto keeptheCarrelsystemupdatedaboutanyeventrelatedto a pieceor thestateof thewaitinglists.They canalsoquerytheCarreldatabase.

Tissuebanknotifier (tb): agentssentby tissuebanksin orderto updateCarrelabouttissueavailability.

The institutionalrolesconsistof oneagentto man-age eachsceneand one agent to coordinateall thescenerelationships.Thus we have: Institution Man-ager (im) which coordinatesthe scenemanagersRe-ceptionRoomManager (rrm), TissueExchange RoomManager (trm),OrganExchangeRoomManager (orm),ConfirmationRoomManager (cfrm) andConsultationRoomManager (crm).

3.2. Theperformativestructure

The connectedgraphof scenesconstitutesthe per-formativestructure. It is a network of scenesthat de-

a� a� w �

w a a! w !

w " a" a# w #

a$ a% w %1

2

3a

3b

3c

3d 4a4b

5(ok)

5(faulty-void)

6

6

7a

7b

8 9(ok)

9(faulty-void)

Fig. 4. Theconversationgraphfor theReceptionroom

Msg& Illocution

1(request (?x hf|hc|tb) (?y rrm)(admission ?id agent ?role?hospital certificate))

2 (deny (!y rrm) (!x hf ' hc ' tb) (deny?deny reason))

3a (accept (!y rrm) (!x hi) (accept hi))

3b (accept (!y rrm) (!x tb) (accept tb))

3c (accept (!y rrm) (!x hf) (accept hf))

3d (accept (!y rrm) (!x hc) (accept hc))

4a(inform (?x hf) (?y rrm) (petition tissue?id hospital ?urgency level?time to deliver ?piece type(?piece parameters) (?info recipient)))

4b(inform (?x hf) (?y rrm) (petition organ?id hospital ?time for availability?piece type (?piece parameters)(?info donor)))

5 (inform (!y rrm) (!x hf) (petition state?id petition ok ' faulty))

6 (inform (?y rrm) (?x hf) (init exchange?piece type ?id exchange room))

7a(request (?x hf) (?y rrm)(tissue exchange entrance request!id exchange room))

7b(request (?x hf) (?y rrm)(organ exchange entrance request!id exchange room))

8(inform (?x hc) (?y rrm)(called for organ ?id hospital!id petition))

9 (inform (!y rrm) (!x hf) (called state!id petition ok ' faulty))

Fig. 5. Theillocutionsfor theReceptionroom

finesthe possiblepathsfor eachagentrole. In accor-dancewith its role,anagentmayor maynotbepermit-tedto follow aparticularpaththroughtheperformativestructure.

With all thescenesandrolesidentifiedin theprevi-


oussection,theperformativestructurecanbedrawn,asdepictedin figure3. Nodesarethesceneslistedaboveplus enter andexit nodesin order to definestartandendpointsin thediagram.Arcs arelabelledwith tagsvariable:role, wherevariableis an agent( and role isoneamongthe identifiedroles for the Carrelsystem.Thediagramin figure3 shows,for instance,thatscenemanagersgo directly from theenternodeto thescenethey shouldmanage(the* meansthatthey aretheonescreatingthescene),while all theexternalagentsmustproceedthroughtheReceptionRoomscenein ordertoberegisteredandthenbe directedto the propersceneaccordingto their roles.

3.2.1. Authenticationof externalagentsAs explainedabove,in theReceptionRoomexternal

agentsenterandareregisteredinsidethe platform. Inthis roomanauthenticationmechanismbasedon elec-tronic certificatesensuresthat external agentscomeonly from authorizedorganizations(which previouslyreceivedtheelectroniccertificateto beused).Oncethesenderhasbeenidentifiedandauthorized,theexternalagentsarethendirectedto theproperroomaccordingto their roles.

Theprotocolof thisscenecanbeseenin figure5: Anagent( makesa requestfor admission(1) that canbeaccepted(messages3a,3b,3c,3d)or refused(message2, exit state)+* ). Dependingon therole of agent( :

– it is directedto theConsultationRoom(exit )-, ),– if it brings a requestfrom a hospital,a check-

ing of suchrequestis done(messages4 and5).Thenagent( waitsuntil a properExchangeRoomis availableto do theassignation(messages6 and7afor tissues,6 and7b for organs).

– if it wascalledby theinstitutionto receiveanor-ganoffer, the informationit bringsaboutthe re-cipient is checkedand,if all is correct,thenis di-rectedto theOrganExchangeRoomthatsentthecall.

3.2.2. Registeringtherecipientsandtheavailablepieces

In orderto managetheassignationof organsandtis-sues,the Carrel institution shouldbe kept up to dateabout:a) all the available tissuesfor transplantation,b) thestateof hospitalswaiting list for eachkind of or-gan,andc) thewhereaboutsaboutall piecesthathavebeenassignedby Carrel.

TheConsultationRoomallows agentscomingfromhospitalsor tissuebanksto keepCarrelupdatedaboutall the facts mentionedabove. The protocol of thissceneis shown in figure 7. The incomingagentscan

a. a/

a0 1 w 0

1413121110

15

1617 18

Fig. 6. Conversationgraphfor theConsultationroom

Msg2 Illocution

10(inform (?x hi) (?y crm) (piece arrival?id hospital ?id tissue bank ?id piece(?state)))

11(inform (?x hi) (?y crm)(transplantation eval ?id piece?id recipient ?date(?info transplantation)))

12(inform (?x tb) (?y crm)(tissue bank update ?id tissue bank?id piece (?specifications)))

13(inform (?x hi) (?y crm)(waiting list update ?id hospital?id piece ?id recipient ?time in(?info recipient) ))

14(inform (?x hi) (?y crm)(maximum urgency level update ?id hospital?id piece ?id recipient ?urgency level?time in (?info recipient) ))

15 (inform (!y crm) (!x hi 3 tb)(notification ack !id piece ok 3 error))

16 (query-if (?x hi) (?y crm) (?query))

17 (inform (!y crm) (!x hi) (query results(?results)))

18 (request (?x hi 3 tb) (?y im) (end))

Fig. 7. Illocutionsfor theConsultationroom

performnotifications(messages10 to 14) andarein-formedif the notification is successful(message15).The agentscoming from hospitals—whichrepresenttheHospitalTransplantCoordinator[3]—canalsoper-form queries(message16) abouthistorical facts(e.g.statisticson, say, successfulcorneatransplantsover acertainperiod).Thequeriesareanswered(message17)with the level of detail that is permittedfor a certainrole,asall accessto thedatabaseis controlledthrougha Role-BasedAccessModel [9]. When the incomingagentshaveperformedall thequeriesandnotifications,they exit theCarrelsystem(message18).

3.2.3. AllocatingorgansFor organassignment,a new scene,the Organ Ex-

change Roomhasbeenadded.Numerousfactorsin-fluencetheprobabilityof successin solid organtrans-


a454 a4 6 a487 w 4

w 9 w 6

a4 9 a4 : a4 ; a4 <

19

20

21(yes)

21(no)

22

22

23

2425

26

20

27

27 26

Fig. 8. Theconversationgraphfor theOrganExchangeroom

Msg= Illocution

19 (query-if (?x hf) (?y orm)(recipient for organ ?id petition))

20

(query-if (?x orm) (?y hc)(call for recipient ?id recipient!id petition ?time for availability?piece type (?piece parameters)(?info donor)))

21 (inform (!y hc) (!x orm) (call answer!id petition ?id hospital))

22 (inform (?x orm) (?y hf) (recipient found!id petition !id recipient !id hospital))

23(query-if (?x hc) (?y orm)(change recipient (!id previous recipient?id new recipient ?change reason)))

24 (inform (!y orm) (!x hc) (accept change))

25 (inform (!y orm) (!x hc) (reject changereason))

26 (request (?x hf > hc) (y im) (exit?exit reason))

27 (inform (?x orm) (?y hf)(recipient not found reason))

Fig. 9. Theillocutionsfor theOrganExchangeRoom

plantation.Bothpatientandgraftsurvival canbeinflu-encedby factorssuchasrace,histocompatibility, im-muneresponsiveness,age,sex, bodyhabitus,primarydisease,causeof donordeath,andothercharacteristicsof the candidateanddonors.It shouldbe possibleforeachhospitalto createa setof rules,seefor exampletherulesin figure10,for cadaversolidorgantransplan-tationthatmaximizesthenumberof additionalyearsofpatientlife systemwide or thatmaximizesthenumberof yearsof transplantfunction.

Theprotocolof this scene,depictedin figure9, canbedividedin two parts:

– the arrival of an Agent? (hospitalFinder Agent)with anoffer of anavailableorgan(states@BACA and@ AED ), waitingfor anotificationthataproperrecip-ient hasbeenfound (message22, exit stateFHG )

or not (message27 leadingto a requestfor exitthroughstateF A ).

– the loop of the scenemanagerlooking for recip-ients. Basedon the information of the waitinglists storedin Carrel’s database,the sceneman-agersendsacall to ahospital(message20)wherethereis asuitablerecipient.ThenanAgentI (hos-pital ContactAgent) entersthe sceneto answerthe call, sayingwhetherit acceptsthe organ ornot (message20). SometimesAgentI , represent-ing thehospitalTransplantCoordinator,expressestheintentionto usetheorganin adifferentrecipi-ent(message23),achangethat,dependingonthereasonsgiven,caneitherbe acceptedor rejected(messages24 and25). If the scenemanagerandAgentI agree,then Agent? is notified of the re-cipient,otherwiseAgentI exits thesceneandtheloopstartsagainwith acall to anotherhospitalforanotherrecipient.

The searchandassignmentprocessesby the scenemanagerare driven by knowledgeof donor-recipientcompatibility that is codedin the form of rules suchas thosefor kidneys shown in Figure 10. Rules1 to8 are relatedto sizecompatibility, eitherconsideringageranges(rules1 to 7) or weight differences—herethe criterion permitsa 20% variationabove or below.Rules5 to 7 considerquality of the kidney 6 andas-sessnot only the limit that is acceptablebut also thetransplanttechniqueto be used(to transplantone orbothkidneys).Rules9 to 10 areexamplesof diseasesin the donor that do not lead to discardingthe organfor transplantation,if aproperrecipientis found(in theexample,a recipientthathashadalsothesamekind ofhepatitisB or C in thepast).Finally, rules11 to 13 areexamplesof rejectionrules7, asdeterminedby currentmedicalknowledge.

It is importantthat suchpoliciesnot behard-codedin the system,assuchrulesevolve with practice(forinstance,someyearsagodonorswith any kind of hep-atitiswerediscarded).Expressingtheknowledgein theform of rulesis a techniquethat allows the systemtobeadaptableto futurechangesin medicalpractice.

Anotherrestrictionwhich mustbeencodedinto thesystemis theallowabletimebetweenretrieving theor-ganfrom thedonorandtransplantingtheorganinto arecipient(seetable1).

6Glomerulosclerosis is a negative decayfactorin kidney filteringbehaviour, while CreatinineClearance is a measureof the amountof serumcreatininethekidney canfilter

7Therulesin figure 10 areorderedandgroupedonly for displaypurposes.The orderhasno precedencemeaning(for instance,rule11shouldprecedeany other).


1- (age_donor <= 1)-> (age_recipient < 2)

2- (age_donor > 1) AND (age_donor < 4)-> (age_recipient < 4)

3- (age_donor >= 4) AND (age_donor < 12)-> (age_recipient > 4) AND

(age_recipient < 60)

4- (age_donor >= 12) AND (age_donor < 60)-> (age_recipient >= 12) AND

(age_recipient < 60)

5- (age_donor >= 60) AND (age_donor < 74) AND(creatinine_clearance > 55 ml/min)

-> (age_recipient >= 60) AND(transplant_type SINGLE-KIDNEY)

6- (age_donor >= 60) AND (age_donor < 74) AND(glomerulosclerosis <= 15%)

-> (age_recipient >= 60) AND(transplant_type SINGLE-KIDNEY)

7- (age_donor >= 60) AND(glomerulosclerosis > 15%) AND(glomerulosclerosis <= 30%)

-> (age_recipient >= 60) AND(transplant_type DUAL-KIDNEY)

8- (weight_donor = X)-> (weight_recipient > X*0.8) AND

(weight_recipient < X*1.2)

9- (disease_donor Hepatitis_B)-> (disease_recipient Hepatitis_B)

10-(disease_donor Hepatitis_C)-> (disease_recipient Hepatitis_C)

11-(disease_donor VIH)-> (DISCARD-DONOR)

12-(glomerulosclerosis > 30%)-> (DISCARD-KIDNEY)

13- (HLA_compatibility_factors < 3)-> (DONOR-RECIPIENT-INCOMPATIBILITY)

Fig. 10.Exampleof someselectionrulesfor kidneys

3.2.4. AllocatingtissuesTheTissueExchangeRoomis theplacewherenego-

tiation over tissuesis performed.The protocolof thissceneis shown in figure 12: AgentJ (hospitalFinderAgent) asksthescenemanagerfor tissueoffers(tissuesmatchingthe requirementsincludedin their petition).Thenthescenemanagergivesa list of availabletissues(message29) that is evaluatedby the externalagentJ(message30).With thisinformationthescenemanagercanmakeaprovisionalassignmentandsolvecollisions(two agentsinterestedin the sametissue).Whenthisprovisionalassignmentis delivered(message31) thenagentJ exits thesceneto go to theConfirmationRoomrepresentedby state KHL . There is an alternative pathfor thecasewhenthereareno availablepiecesmatch-

ing therequirementsdescribedin thepetition(message9 with null list). In this caseagentJ requestsan exitpermissionfrom theinstitution(message32,exit stateK+M ), includingthereasonfor leaving. Thereasonpro-videdis recordedin theinstitutionlogsto form anaudittrail for the relevantauthoritiesto inspect.For furtherinformationaboutthisnegotiationprocesssee[2].

aN O aN P aQ R aQ5N

aQ5Q w N w Q

28 29(offers) 30

3129(nil-void)

32

Fig. 11.Theconversationgraphfor theTissueExchangeroom

MsgS Illocution

28 (query-if (?x hf) (?y trm)(offer list?id petition))

29(inform (!y trm) (!x hf) (offer list!id petition (list (?id piece1?info piece1) ... (?id piecen?info piecen))))

30(inform (?x hf) (?y trm) (weighted list!id petition (list (!id piece1 ?weight)... (!id piece1 ?weight))))

31(query-if (?y trm) (?x hf) (piece offer(?id petition ?id piece?cost estimation) T void))

32 (request (?x hf) (y im) (exit?exit reason))

Fig. 12.Theillocutionsfor theTissueExchangeRoom

3.2.5. ConfirmingtheassignationIn theConfirmationRoomscene,theprovisionalas-

signmentsmade in a Tissue Exchange Room or anOrgan Exchange Roomareeitherconfirmedor with-drawn. Figure14 shows theprotocolof this scene:theagentJ cananalyzetheassignedpiecedataandthenac-ceptor refuseit (message33). If theagentJ acceptsthepieceandno higher-priority requestsappearduring acertaintime window then the provisional assignmentis confirmedanda delivery plan is givento theagentJ(message34), andthenit exits theCarrelsystem(exitstateKHL ). Whenthereis a requestwith higherprioritythatneedsthepieceprovisionally assignedto agentJ aconflict arises.To resolve the conflict the sceneman-agernotifies the agentJ that the assignmenthasbeenwithdrawn (message35) and that he is then entitledto a freshrequestfor anotherpiece,if available,(mes-sage36) to benegotiatedagainin theExchangeRoomwhenceit came.


aU5V aU5W w U

aUYX w V

aUYU w Z

33(accept) 34

35 35

33(refused)

32

36

Fig. 13.Conversationgraphfor theConfirmationroom

Msg[ Illocution

32 (request (?x hf) (y im) (exit?exit reason))

33 (inform (?x hf) (?y cfrm) (piece eval?id petition ?id piece accepted \ refused))

34(inform (?y cfrm) (?x hf) (piece delivery?id petition ?id hospital ?id tissue bank?delivery plan))

35(inform (?y cfrm) (?x hf)(piece reassigned exception ?id petition?id piece ?reassignment reason))

36 (query-if (?x hf) (?y cfrm)(another offer list ?id petition))

Fig. 14. Illocutionsfor theConfirmationroom

3.2.6. TheMulti-agentarchitectureTheagentarchitecturethatsupportsthe institutional

roles(seefigure15)showsoneagentmanagingeachofthescenes:theRR Agent for theReceptionRoom, theCR Agent for the ConsultationRoom, an ER Agentfor eachExchangeRoom(eithertheonesfor organsortheonesfor tissues),anda CfR Agent for theConfir-mationRoom. Also thereis an agent(the IM Agent)playingthe institutionmanager role.

In orderto assisttheseagents,two agentsareaddedfor specifictasks:thePlanner Agent, to build thede-liveryplansthatareneededin theConfirmationRoom,andtheDB Agent, whichhandlesaccesscontrolto theinternaldatabase.For solid organsthePlanner Agenthasto considertheconstraintsimposedby theconser-vationof thosepiecesasdescribedin table1.

3.3. ExtendingtheUCTx

Adapting the UCTx agency in order to assistnotonly in thetissueallocationprocessbut alsoin theor-ganallocationprocessis notdifficult. In thecaseof tis-sues,it is thesurgeonswhoareresponsiblefor creatingthetissuerequeststhroughtheirSurgeonAgent [3]. Inthe caseof organs,however, it is the Hospital Trans-plantCoordinatorwhois responsiblefor issuingorgan

Fig. 15.Themulti-agentarchitectureof aCarrelplatform

offers to the institutionor answeringa call for recipi-ents.So the architecturepresentedin [3] (depictedinfigure16) is not modifiedbut ratherthe CoordinatorAgent functionalityis extended.

3.4. A networkof Carrel institutions

In theprecedingsectionstheCarrelsystemhasbeendescribedas an institution that works alone,manag-ing all therequestsandofferscomingfrom thehospi-tals.However a distributedsystemis neededin orderto managethe allocationproblemat an internationallevel, which is oneof theaimsof our scheme.

To do so,we proposethecreationof a federationofgeographically-distributedCarrelplatforms.Hospitalsandtissuebanksregister themselveswith the nearestplatformandinteractasdescribedin previoussections.

As a result, the searchprocessmay becomedis-tributedthroughtheplatformsexchanginginformationamongthemselvesvia theirDB Agents. Theprocessisthefollowing:

– The DB Agent of a certainplatform] receivesaquery, either from an Organ Exchange Room, aTissueExchangeRoomor theConsultationRoom

– It accessesthelocaldatabase.– If the informationis not availablelocally, thenit

sendspart of the query to other DB Agents inotherCarrelplatforms.

– All the differencesin terminologyare solved atthis point by theuseof domainontologiesshared


Fig. 16.Themulti-agentarchitecturefor theUCTx system

by all the platforms that define a commonex-changeformatfor theinformation.

All Carrel platformsareawareof the existenceoftheotherplatforms.Thecommunicationamongagentsof differentplatformsis donethroughthe mechanismthe FIPA specificationestablishesfor communicationamongAgentplatforms[5].

4. DISCUSSION

Thecurrentsystemsof organprocurementandallo-cationworksreasonablywell, but significantimprove-mentsin bothits fairnessandits effectivenesscouldbemade.We presenthereanAgent-MediatedElectronicInstitutionfor thedistributionof organsandtissuesfortransplantationpurposes.Thepolicy thatweareimple-mentingherefollows theSpanishmodelfor organandtissueprocurementandSpanishregulationsfor alloca-tion but webelieve thatthismodelis generalenough–andverysuccessful–to usedasanexample.

Our aim with this work is not only to apply multi-agenttechnologiesto modeltheorganandtissueallo-cationproblembut wealsohavedevotedsignificantef-fort to formalization,following the recommendationsin [6] abouttheneedfor formal designmethodswhenapplyingagentsto themedicaldomainin orderto en-surethe safetyandsoundnessof the resultingsystem.In our casewe have chosena formalism called IS-LANDER [4], basedonthedialogicalframework idea,to getanaccuratedescriptionof theinteractionsamongtheagents.By meansof suchformalismwe havebeen

ableto designa systemthatcombinesthestrengthsofagentswith theadvantagesof formal specification.

As far aswe know, therearevery few referencesinthe literatureto theuseof agentsin the transplantdo-main.[21] and[12] describesingleagentsto solvespe-cific tasksneededfor this domain(respectively, a re-ceiver selectionalgorithm basedin multi-criteria de-cision techniquesand a planner of transportroutesamonghospitalsfor organdelivery). [11] proposesamulti-agentsystemarchitectureto coordinateall theteammembersinvolved in a transplant.[1] alsopro-posesastatichierarchicalagentarchitecturefor theor-ganallocationproblem,but noformalismis usedin thedevelopmentof thearchitecture,andno mechanismispresentedto makethearchitectureadaptiveto changesin policiesor regulations.

5. CONCLUSIONS

Public interestin theutility andfairnessof solid or-ganandtissuetransplantationis amongthemotivationsof this work. Transplantsprotocolsasaboveexplainedhavemany characteristicsthatsuggesttheappropriate-nessof a multi-agentsolution:

– they aresafetycritical;– they involve largeamountsof data;– thedataarediversein sourceandformat;– complex inferencesmustbemadefrom combina-

tionsof data;– co-ordinatedactivity acrossnumerousagencies

maybeindicated;andfinally,– stronglegal andethicalobligationsunderpinthe

interactionsbetweenagenciesasONT or OCATT.Carrel is anagent-basedsolutionmeantto helppolicymakers in designing,implementingandverifying or-ganallocationpolicies (like for example[8]) so that,say, graft survival ratescanbemaximized.Finally, wesummarizethe propertiesof our proposal,which webelievemighthelpto addressconcernsaboutwhatfairdistributionmightmean:

– Experimentationwith allocation policies beforedeployment,for example:^ Old-for-the-old^ Childrenfirst^ Allocation basedon waiting timeor combinationsof theaboveandotherschemes.

– TheCarrel systemmayhelpto speedup theallo-cationprocessof solid organsfor transplantationandin thiswayto improvegraftsurvival rates.Re-call thetemporalconstraintsimposedby thecon-servationof thosepieces(seetable1)


– The federationof several platformsof this kind,asdescribedin _ 3.4,wouldmakeit possibleto ex-tendthebenefitsof sucha systemto a largergeo-graphicalarea(e.g. from countryto country)eas-ing the communicationamongthe different Or-ganExchangeOrganizations.This will allow theunificationof policiesin theEuropeanSpace.

Futurework aims to explore other alternative for-malismsto describemulti-agentsystemsin complexdomains.As partof thatexplorationweaim to expandthe PROforma [6] formalism,which is well-suitedtomodeldecisionsupportsystemsfor domainswith un-certainty, to includeagentcommunicationandthedef-inition of multiple roles(in thesensethateachrolehasa differentview of theproblemto solve).

Acknowledgements

U. Cortes,andJ.Vazquez-Salcedawould like to ac-knowledgethe IST-2000-28385Agentcities.NETandthe IST-1999-10176A-TEAM projects.The views inthispaperarenotnecessarilythoseof Agentcities.NETandA-TEAM consortia.

References

[1] A. Aldea, B. Lopez, A. Moreno, D. Riano, and A. Valls.A multi-agentsystemfor organ transplantco-ordination. InS.Quaglini,P. Barahona,andS.Andreassen,editors,Proceed-ingsof the8th.EuropeanConferenceon Artificial Intelligencein Medicine, Portugal, 2001, LectureNotesin Artificial Intel-ligence2101:Artificial Intelligencein Medicine,pages413–416.Springer-Verlag,2001.

[2] U. Cortes, A. Lopez-Navidad, J. Vazquez-Salceda,A. Vazquez,D. Busquets,M. Nicolas,S. Lopes,F. Vazquez,andF. Caballero.Carrel:An agentmediatedinstitutionfor theexchangeof humantissuesamonghospitalsfor transplanta-tion. In 3ba CongresCatala d’Intel.ligencia Artificial, pages15–22.ACIA, 2000.

[3] U. Cortes,J. Vazquez-Salceda,A. Lopez-Navidad,andF. Ca-ballero. UCTx: A multi-agentsystemto assista transplantco-ordinatorunit. AppliedIntelligence. Accepted, 2003.

[4] M. Esteva, J. Padget,andC. Sierra. Formalizinga languagefor institutionsandnorms. In John-JulesMeyer andMilindeTambe,editors,Intelligent AgentsVIII , volume2333 of Lec-ture Notesin Artificial Intelligence, pages348–366.SpringerVerlag,2001. ISBN 3-540-43858-0.

[5] The Foundation for Intelligent Phisical Agents, http://www.fipa.org/repository/fipa2000.html. FIPASpecifications, 2000.

[6] J. Fox andS. Das. Safeand Sound. MIT Press,1 cEd edition,1999.

[7] U.S. Government. 42 cfr part 121, organ procurementandtransplantationnetwork; final rule. 63 FederalRegister16295,at 16332,April 1998.

[8] M. Hild. Fair kidney allocationbasedon waiting time. SocialScienceWorking Paper1117,California Instituteof Technol-ogy, 2001.

[9] A. Lin. Integratingpolicy-drivenrole-basedaccesscontrolwithcommondatasecutiryarchitecture. TechnicalReportHPL-1999-59,HPLaboratoriesBristol, 1999.

[10] Antonio Lopez-Navidad. Professionalcharacteristicsofthe transplant coordinator. Transplantation Proceedings,(23):1607–1613,1997.

[11] A. Moreno, A. Valls, and J. Bocio. Managementof hospi-tal teamsfor organtransplantsusingmulti-agentsystems.InS.Quaglini,P. Barahona,andS.Andreassen,editors,Proceed-ingsof the8th.EuropeanConferenceon Artificial Intelligencein Medicine, Portugal, 2001, LectureNotesin Artificial Intel-ligence2101,pages374–383.Springer-Verlag,2001.

[12] A. Moreno,A. Valls, and A. Ribes. Finding efficient organtransportroutesusingmulti-agentsystems.In IEEE 3rd Inter-national Workshopon EnterpriseNetworkingandComputingin HealthCare Industry(Healthcom), L’Aquilla, Italy, 2001.

[13] Y. MosesandM. Tennenholtz.Artificial socialsystems.Com-puters andAI, 14(6):533–562,1995.

[14] P. Noriega. Agent-Mediated Auctions: The FishmarketMetaphor. Number 8 in IIIA MonographSeries.Institutd’Investigacio en Intel.ligencia Artificial (IIIA), Bellaterra,Spain,1997.PhDThesis.

[15] D.C. North. Institutions,InstutionalChange andEcononomicPerformance. CambridgeUniv. Press,Cambridge,1 cbd edition,1990.

[16] Commiteeon OrganProcurementandTransplantationPolicy,editors. Organ Procurementand Transplantation:AssessingCurrent Policies and the Potential Impactof the DHHS FinalRule. NationalAcademyPress,Washington,USA, 1999.

[17] Organizacion Nacional de Transplantes. http://www.msc.es/ont.

[18] Yoav ShohamandMosheTennenholtz.On sociallaws for ar-tificial agentsocieties:Off-line design. Artificial Intelligence,73(1-2):231–252,1995.

[19] S.E.Taranto,A.M. Harper, E.B. Edwards,andJ.D.Rosendale.Developinga nationalallocationmodelfor cadaveric kidneys.In J. A. Joines,R.R.Barton,K. Kang,andP. A. Fishwick,ed-itors, Proc. of the 2000Winter SimulationConference, pages1955–1962,2000.

[20] UNOS. Unosstatementof principlesandobjectives of equi-table organallocation. Seminars in Anesthesia, 14:142–166,March2000.

[21] A. Valls,A. Moreno,andD. Sanchez.A multi-criteriadecisionaidagentappliedto theselectionof thebestreceiver in a trans-plant. In J.Braz,M. Piattini,andJ.Filipe, editors,4dfe Interna-tional Conferenceon EnterpriseInformationSystems(ICEIS-2002), CiudadReal,Spain,2002.ICEISPress.

[22] J. Vazquez-Salceda,U. Cortes,andJ. Padget.Formalizinganelectronicinstitution for the distribution of humantissues.InArtificial Intelligencein Medicine, volume27,pages233–258.Elsevier, March2003.

[23] M. WooldrigeandN.R. Jennings. Intelligent agents:Theoryandpractice.TheKnowledge EngineeringReview, 10(2):115–152,1995.

Documents

The organ allocation process: a natural extension of the ...jvazquez/docs/Carrel-AIcom.pdf2 Vazquez-Salceda´ et al. / The organ allocation process: a natural extension of the Carrel