99
Office of the Science Advisor Risk Assessment Forum EPA/100/R-14/004 July 2014 www.epa.gov/raf

Risk Assessment Forum White Paper: Probabilistic Risk Assessment

Embed Size (px)

Citation preview

Page 1: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

Office of the Science AdvisorRisk Assessment Forum

EPA/100/R-14/004 July 2014 www.epa.gov/raf

U.S. Environmental Protection Agency Office of Research and Development Washington, DC 20460

Official Business Penalty for Private Use $300

PRESORTED STANDARDPOSTAGE & FEES PAID

EPAPERMIT NO. G-35

lmarti03
Typewritten Text
MBRODER
Typewritten Text
MBRODER
Typewritten Text
MBRODER
Typewritten Text
MBRODER
Typewritten Text
MBRODER
Typewritten Text
MBRODER
Typewritten Text
MBRODER
Typewritten Text
jfitzpat
Typewritten Text
JFitzpat
Typewritten Text
JFitzpat
Typewritten Text
JFitzpat
Typewritten Text
JFitzpat
Typewritten Text
JFitzpat
Typewritten Text
JFitzpat
Typewritten Text
JFitzpat
Typewritten Text
MBRODER
Typewritten Text
Risk Assessment Forum White Paper: Probabilistic Risk Assessment Methods and Case Studies
JFitzpat
Typewritten Text
JFitzpat
Typewritten Text
JFitzpat
Typewritten Text
jfitzpat
Typewritten Text
jfitzpat
Typewritten Text
jfitzpat
Typewritten Text
Page 2: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

EPA/100/R‐14/004

RiskAssessmentForumWhitePaper:

ProbabilisticRiskAssessmentMethodsandCaseStudies

July25,2014

U.S.EnvironmentalProtectionAgencyOfficeoftheScienceAdvisorRiskAssessmentForum

ProbabilisticRiskAnalysisTechnicalPanelWashington,D.C.20460

Page 3: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

ii

Disclaimer ThisdocumenthasbeenreviewedinaccordancewithU.S.EnvironmentalProtectionAgency(EPA)policyandapprovedforpublication.Mentionoftradenamesorcommercialproductsdoesnotconstituteendorsementorrecommendationforuse.

ThisdocumentwasproducedbyaTechnicalPaneloftheEPARiskAssessmentForum(RAF).Theauthorsdrewontheirexperienceindoingprobabilisticassessmentsandinterpretingthemtoimproveriskmanagementofenvironmentalandhealthhazards.Interviews,presentationsanddialogueswithriskmanagersconductedbytheTechnicalPanelhavecontributedtotheinsightsandrecommendationsinthiswhitepaperandtheassociateddocumenttitledProbabilisticRiskAssessmenttoInformDecisionMaking:FrequentlyAskedQuestions.

U.S.EnvironmentalProtectionAgency(USEPA).2014.RiskAssessmentForumWhitePaper:ProbabilisticRiskAssessmentMethodsandCaseStudies.EPA/100/R‐09/001A.Washington,D.C.:RiskAssessmentForum,OfficeoftheScienceAdvisor,USEPA.

Page 4: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

iii

Foreword ThroughoutmanyoftheU.S.EnvironmentalProtectionAgency’s(EPA)programofficesandregions,variousformsofprobabilisticmethodshavebeenusedtoanswerquestionsaboutexposureandrisktohumans,otherorganismsandtheenvironment.Riskassessors,riskmanagersandothers,particularlywithinthescientificandresearchdivisions,haverecognizedthatmoresophisticatedstatisticalandmathematicalapproachescouldbeutilizedtoenhancethequalityandaccuracyofAgencyriskassessments.Variousstakeholders,insideandoutsideoftheAgency,havecalledforamorecomprehensivecharacterizationofrisks,includinguncertainties,toimprovetheprotectionofsensitiveorvulnerablepopulationsandlifestages.

TheEPAidentifiedtheneedtoexaminetheuseofprobabilisticapproachesinAgencyriskassessmentsanddecisions.TheRAFdevelopedthispaperandthecompaniondocument,ProbabilisticRiskAssessmenttoInformDecisionMaking:FrequentlyAskedQuestions,toprovideageneraloverviewofthevalueofprobabilisticanalysesandsimilarorrelatedmethods,aswellasprovideexamplesofcurrentapplicationsacrosstheAgency.Draftsofbothdocumentswerereleased,withslightlydifferenttitles,forpubliccommentandexternalpeerreviewinAugust2009.AnexternalpeerreviewwasheldinArlington,VirginiainMay2010.

Thegoalofthesepublicationsisnotonlytodescribepotentialandactualusesofthesetools,butalsotoencouragetheirfurtherimplementationinhuman,ecologicalandenvironmentalriskanalysisandrelateddecisionmaking.Theenhanceduseofprobabilisticanalysestocharacterizeuncertaintyinassessmentswillnotonlyberesponsivetoexternalscientificadvice(e.g.,recommendationsfromtheNationalResearchCouncil)onhowtofurtheradvanceriskassessmentscience,butalsowillhelptoaddressspecificchallengesfacedbymanagersandincreasetheconfidenceintheunderlyinganalysisusedtosupportAgencydecisions.

____________________________________________RobertKavlockInterimScienceAdvisorU.S.EnvironmentalProtectionAgency

Page 5: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

iv

ProbabilisticRiskAnalysisTechnicalPanel

HalûkÖzkaynak,EPAOfficeofResearchandDevelopment(Co‐Chair)RobertHetes,EPAOfficeofResearchandDevelopment(Co‐Chair)KathrynGallagher,EPAOfficeofWater(Co‐Chair)ChrisFrey,NorthCarolinaStateUniversity(whileontemporaryassignmentatEPA)(WhitePaperCo‐Lead)JohnPaul,EPAOfficeofResearchandDevelopment(WhitePaperCo‐Lead)PaskyPascual,EPAOfficeofResearchandDevelopment(WhitePaperCo‐Lead)MarianOlsen,EPARegion2(CaseStudyLead)MikeClipper,EPAOfficeofSolidWasteandEmergencyResponseMichaelMessner,EPAOfficeofWaterKeeveNachman,currentlyatTheJohnsHopkinsBloombergSchoolofPublicHealthZacharyPekar,EPAOfficeofAirandRadiationRitaSchoeny,EPAOfficeofResearchandDevelopmentCynthiaStahl,EPARegion3DavidHrdy,EPAOfficeofPesticideProgramsJohnLangstaff,EPAOfficeofAirandRadiationElizabethMargosches,EPAOfficeofChemicalSafetyandPollutionPrevention(retired)AudreyGalizia,EPAOfficeofResearchandDevelopmentNancyRios‐Jafolla,EPARegion3DonnaRandall,EPAOfficeofChemicalSafetyandPollutionPreventionKhoanDinh,EPAOfficeofChemicalSafetyandPollutionPrevention(retired)HarveyRichmond,EPAOfficeofAirandRadiation(retired)

OtherContributingAuthors

JonathanChen,EPAOfficeofChemicalSafetyandPollutionPreventionLisaConner,EPAOfficeofAirandRadiationJanetBurke,EPAOfficeofResearchandDevelopmentAllisonHess,EPARegion2KellySherman,EPAOfficeofChemicalSafetyandPollutionPreventionWoodrowSetzer,EPAOfficeofResearchandDevelopmentValerieZartarian,EPAOfficeofResearchandDevelopment

EPARiskAssessmentForumScienceCoordinators

JulieFitzpatrick,EPAOfficeoftheScienceAdvisorGaryBangs,EPAOfficeoftheScienceAdvisor(retired)

ExternalPeerReviewersScottFerson(Chair),AppliedBiomathematicsAnnetteGuiseppi‐Elie,DuPontEngineeringDaleHattis,ClarkUniversityIgorLinkov,U.S.ArmyEngineerResearchandDevelopmentCenterJohnToll,WindwardEnvironmentalLLC

Page 6: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

v

Table of Contents Foreword.................................................................................................................................................................iii 

ListofFiguresandTables...............................................................................................................................viii 

ListofAcronymsandAbbreviations..............................................................................................................ix 

EXECUTIVESUMMARY.........................................................................................................................................1 

1.  INTRODUCTION:RELEVANCEOFUNCERTAINTYTODECISIONMAKING:HOWPROBABILISTICAPPROACHESCANHELP..............................................................................................4 

1.1.  EPADecisionMaking......................................................................................................................................4 1.2.  TheRoleofProbabilisticRiskAnalysisinCharacterizingUncertaintyand

Variability............................................................................................................................................................4 1.3.  GoalsandIntendedAudience......................................................................................................................5 1.4.  OverviewofThisDocument.........................................................................................................................5 1.5.  WhatAreCommonChallengesFacingEPARiskDecisionMakers?............................................5 1.6.  WhatAreKeyUncertaintyandVariabilityQuestionsOftenAskedbyDecision

Makers?.................................................................................................................................................................6 1.7.  WhyIstheImplementationofProbabilisticRiskAnalysisImportant?.....................................9 1.8.  HowDoesEPATypicallyAddressScientificUncertaintyandVariability?............................10 1.9.  WhatAretheLimitationsofRelyingonDefault‐BasedDeterministicApproaches?........11 1.10. WhatIsEPA’sExperiencewiththeUseofProbabilisticRiskAnalysis?.................................12 

2.  PROBABILISTICRISKANALYSIS..............................................................................................................15 

2.1.  WhatAreVariabilityandUncertainty,andHowAreTheyRelevanttoDecisionMaking?..............................................................................................................................................................15 

2.1.1.  Variability..................................................................................................................................15 2.1.2.  Uncertainty...............................................................................................................................15 

2.2.  WhenIsProbabilisticRiskAnalysisApplicableorUseful?..........................................................17 2.3.  HowCanProbabilisticRiskAnalysisBeIncorporatedIntoAssessments?...........................17 2.4.  WhatAretheScientificCommunity’sViewsonProbabilisticRiskAnalysis,and

WhatIstheInstitutionalSupportforItsUseinPerformingAssessments?..........................18 2.5.  AdditionalAdvantagesofUsingProbabilisticRiskAnalysisandHowItCan

ProvideMoreComprehensive,RigorousScientificInformationinSupportofRegulatoryDecisions...................................................................................................................................19 

2.6.  WhatAretheChallengestoImplementationofProbabilisticAnalyses?...............................19 2.7.  HowCanProbabilisticRiskAnalysisSupportSpecificRegulatoryDecision

Making?..............................................................................................................................................................20 2.8.  DoesProbabilisticRiskAnalysisRequireMoreResourcesThanDefault‐Based

DeterministicApproaches?.......................................................................................................................21 2.9.  DoesProbabilisticRiskAnalysisRequireMoreDataThanConventional

Approaches?....................................................................................................................................................21 2.10.  CanProbabilisticRiskAnalysisBeUsedtoScreenRisksorOnlyinComplexor

RefinedAssessments?..................................................................................................................................22 2.11. DoesProbabilisticRiskAnalysisPresentUniqueChallengestoModelEvaluation?.........23 2.12. HowDoYouCommunicatetheResultsofProbabilisticRiskAnalysis?.................................23 2.13.  AretheResultsofProbabilisticRiskAnalysisDifficulttoCommunicatetoDecision

MakersandStakeholders?.........................................................................................................................24 

Page 7: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

vi

3.  ANOVERVIEWOFSOMEOFTHETECHNIQUESUSEDINPROBABILISTICRISKANALYSIS.........................................................................................................................................................26 

3.1.  WhatIstheGeneralConceptualApproachinProbabilisticRiskAnalysis?..........................26 3.2.  WhatLevelsandTypesofProbabilisticRiskAnalysesAreThereandHowAre

TheyUsed?.......................................................................................................................................................26 3.3.  WhatAreSomeSpecificAspectsofandIssuesRelatedtoMethodologyfor

ProbabilisticRiskAnalysis?......................................................................................................................29 3.3.1.  DevelopingaProbabilisticRiskAnalysisModel.......................................................29 3.3.2.  DealingWithDependenciesAmongProbabilisticInputs......................................29 3.3.3.  ConductingtheProbabilisticAnalysis...........................................................................30 

4.  SUMMARYANDRECOMMENDATIONS...................................................................................................33 

4.1.  ProbabilisticRiskAnalysisandRelatedAnalysesCanImprovetheDecision‐MakingProcessatEPA................................................................................................................................33 

4.2.  MajorChallengestoUsingProbabilisticRiskAnalysistoSupportDecisions......................34 4.3.  RecommendationsforEnhancedUtilizationofProbabilisticRiskAnalysisatEPA..........34 

GLOSSARY...............................................................................................................................................................37 

REFERENCES..........................................................................................................................................................42 

BIBLIOGRAPHY.....................................................................................................................................................48 

ProbabilisticRiskAnalysisMethodologyGeneral....................................................................................48 ProbabilisticRiskAnalysisandDecisionMaking.........................................................................................49 ProbabilisticRiskAnalysisMethodologySpecificAspects...................................................................49 SensitivityAnalysis...................................................................................................................................................50 CaseStudyExamplesofProbabilisticRiskAnalysisEPA(SeeAlsothePRACase

StudiesAppendix).........................................................................................................................................51 CaseStudyExamplesofProbabilisticRiskAnalysisOther..................................................................52 

APPENDIX:CASESTUDYEXAMPLESOFTHEAPPLICATIONOFPROBABILISTICRISKANALYSISINU.S.ENVIRONMENTALPROTECTIONAGENCYREGULATORYDECISIONMAKING............................................................................................................................................................54 

A.  OVERVIEW.......................................................................................................................................................55 

B.  INTRODUCTION.............................................................................................................................................55 

C.  OVERALLAPPROACHTOPROBABILISTICRISKANALYSISATTHEU.S.ENVIRONMENTALPROTECTIONAGENCY............................................................................................56 

C.1. U.S.EnvironmentalProtectionAgencyGuidanceandPoliciesonProbabilisticRiskAnalysis............................................................................................................................................................56 

C.2. CategorizingCaseStudies..........................................................................................................................56 

Group1CaseStudies..........................................................................................................................57 Group2CaseStudies..........................................................................................................................57 Group3CaseStudies..........................................................................................................................58 

Page 8: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

vii

D.  CASESTUDYSUMMARIES..........................................................................................................................63 

D.1.  Group1CaseStudies....................................................................................................................63 

CaseStudy1:SensitivityAnalysisofKeyVariablesinProbabilisticAssessmentofChildren’sExposuretoArsenicinChromatedCopperArsenatePressure‐TreatedWood........................................................................................................................................63 CaseStudy2:AssessmentoftheRelativeContributionofAtmosphericDepositiontoWatershedContamination...................................................................................64 

D.2.  Group2CaseStudies....................................................................................................................65 

CaseStudy3:ProbabilisticAssessmentofAnglingDurationUsedintheAssessmentofExposuretoHudsonRiverSedimentsviaConsumptionofContaminatedFish...............................................................................................................................65 CaseStudy4:ProbabilisticAnalysisofDietaryExposuretoPesticidesforUseinSettingToleranceLevels..............................................................................................................65 CaseStudy5:One‐DimensionalProbabilisticRiskAnalysisofExposuretoPolychlorinatedBiphenylsviaConsumptionofFishFromaContaminatedSedimentSite.........................................................................................................................................66 CaseStudy6:ProbabilisticSensitivityAnalysisofExpertElicitationofConcentration‐ResponseRelationshipBetweenFineParticulateMatterExposureandMortality.....................................................................................................................69 CaseStudy7:EnvironmentalMonitoringandAssessmentProgram:UsingProbabilisticSamplingtoEvaluatetheConditionoftheNation’sAquaticResources................................................................................................................................................70 

D.3.  Group3CaseStudies....................................................................................................................71 

CaseStudy8:Two‐DimensionalProbabilisticRiskAnalysisofCryptosporidiuminPublicWaterSupplies,WithBayesianApproachestoUncertaintyAnalysis...........................................................................................................................71 CaseStudy9:Two‐DimensionalProbabilisticModelofChildren’sExposuretoArsenicinChromatedCopperArsenatePressure‐TreatedWood...................................72 CaseStudy10:Two‐DimensionalProbabilisticExposureAssessmentofOzone......74 CaseStudy11:AnalysisofMicroenvironmentalExposurestoFineParticulateMatterforaPopulationLivinginPhiladelphia,Pennsylvania..........................................76 CaseStudy12:ProbabilisticAnalysisinCumulativeRiskAssessmentofOrganophosphorusPesticides........................................................................................................77 CaseStudy13:ProbabilisticEcologicalEffectsRiskAssessmentModelsforEvaluatingPesticideUse...................................................................................................................78 CaseStudy14:ExpertElicitationofConcentration‐ResponseRelationshipBetweenFineParticulateMatterExposureandMortality.................................................80 CaseStudy15:ExpertElicitationofSea‐LevelRiseResultingFromGlobalClimateChange......................................................................................................................................81 CaseStudy16:KnowledgeElicitationforaBayesianBeliefNetworkModelofStreamEcology......................................................................................................................................82 

CASESTUDYREFERENCES................................................................................................................................85 

Page 9: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

viii

List of Figures and Tables Figures

Figure1.GeneralPhasesoftheRiskAssessmentProcess.....................................................................................8Figure2.TieredApproachforRiskAssessment.......................................................................................................10Figure3.GraphicalDescriptionoftheLikelihood(Probability)ofRisk........................................................25Figure4.DiagrammaticComparisonofThreeAlternativeProbabilisticApproachesforthe

SameExposureAssessment...........................................................................................................................28 

AppendixFigures

FigureA‐1.MonteCarloCancerSummaryBasedonaOne‐DimensionalProbabilisticRiskAnalysisofExposuretoPolychlorinatedBiphenyls........................................................................72

FigureA‐2.MonteCarloNoncancerHazardIndexSummaryBasedonaOne‐DimensionalProbabilisticRiskAnalysisofExposuretoPolychlorinatedBiphenyls...................................73

FigureA‐3.UncertaintyDistributionModelResults...............................................................................................80

Tables

Table1.SelectedExamplesofEPAApplicationsofProbabilisticRiskAssessmentTechniques.........14

AppendixTable

TableA‐1.CaseStudyExamplesofEPAApplicationsofDeterministicandProbabilisticRiskAssessmentTechniques..................................................................................................................................59

Page 10: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

ix

List of Acronyms and Abbreviations 1‐DMCA One‐DimensionalMonteCarloAnalysis2‐DMCA Two‐DimensionalMonteCarloAnalysisAPEX AirPollutantsExposureModelBBN BayesianBeliefNetworkCAA CleanAirActCASAC CleanAirScientificAdvisoryCommitteeCCA ChromatedCopperArsenateCPSC ConsumerProductSafetyCommissionCRA CumulativeRiskAssessmentCSFII ContinuingSurveyofFoodIntakebyIndividualsCWA CleanWaterActDEEM DietaryExposureEvaluationModelDRA DeterministicRiskAssessmentDRES DietaryRiskEvaluationSystemERA EcologicalRiskAssessmentEMAP EnvironmentalMonitoringandAssessmentProgramEPA U.S.EnvironmentalProtectionAgencyFACA FederalAdvisoryCommitteeActFAQ FrequentlyAskedQuestionsFDA U.S.FoodandDrugAdministrationFFDCA FederalFood,Drug,andCosmeticActFIFRA FederalInsecticide,Fungicide,andRodenticideActFQPA FoodQualityProtectionActGAO GovernmentAccountabilityOfficeHHRA HumanHealthRiskAssessmentHI HazardIndexIPCC IntergovernmentalPanelonClimateChangeIRIS IntegratedRiskInformationSystemLHS LatinHypercubeSamplingLOAEL Lowest‐Observed‐Adverse‐EffectLevelLT Long‐TermLT2 Long‐Term2EnhancedSurfaceWaterTreatmentRuleMCA MonteCarloAnalysisMCS MonteCarloSimulationMEE MicroexposureEventAnalysisMOEs MarginsofExposureNAAQS NationalAmbientAirQualityStandardsNAS NationalAcademyofSciencesNERL NationalExposureResearchLaboratoryNOAEL No‐Observed‐Adverse‐EffectLevelNRC NationalResearchCouncilOAQPS OfficeofAirQualityPlanningandStandardsOAR OfficeofAirandRadiationOCSPP OfficeofChemicalSafetyandPollutionPreventionOERR OfficeofEmergencyandRemedialResponseOGWDW OfficeofGroundwaterandDrinkingWaterOMB OfficeofManagementandBudgetOP OrganophosphorusPesticide

Page 11: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

x

OPP OfficeofPesticideProgramsORD OfficeofResearchandDevelopmentOSA OfficeoftheScienceAdvisorOSTP OfficeofScienceandTechnologyPolicyOSWER OfficeofSolidWasteandEmergencyResponseOW OfficeofWaterPAH PolycyclicAromaticHydrocarbonPCB PolychlorinatedBiphenylPCCRARM Presidential/CongressionalCommissiononRiskAssessmentandRiskManagementPDP PesticideDataProgramPM ParticulateMatterPRA ProbabilisticRiskAssessmentRAF RiskAssessmentForumRfC ReferenceConcentration(Inhalation)RfD ReferenceDose(Oral)RIA RegulatoryImpactAnalysisRME ReasonableMaximumExposureSAB ScienceAdvisoryBoardSAP ScientificAdvisoryPanelSHEDS StochasticHumanExposureandDoseSimulationModelSHEDS‐PM StochasticHumanExposureandDoseSimulationModelforParticulateMatterSTPC ScienceandTechnologyPolicyCouncilTRIM.Expo TotalRiskIntegratedMethodology/ExposureModelUI UncertaintyIntervalUSDA U.S.DepartmentofAgricultureUSGS U.S.GeologicalSurveyWHO WorldHealthOrganization

Page 12: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

1

EXECUTIVE SUMMARY Probabilisticriskassessment(PRA),initssimplestform,isagroupoftechniquesthatincorporateuncertaintyandvariabilityintoriskassessments.Variabilityreferstotheinherentnaturalvariation,diversityandheterogeneityacrosstime,spaceorindividualswithinapopulationorlifestage,whileuncertaintyreferstoimperfectknowledgeoralackofpreciseknowledgeofthephysicalworld,eitherforspecificvaluesofinterestorinthedescriptionofthesystem(USEPA2011c).Variabilityanduncertaintyhavethepotentialtoresultinoverestimatesorunderestimatesofthepredictedrisk.

PRAprovidesestimatesoftherangeandlikelihoodofahazard,exposureorrisk,ratherthanasinglepointestimate.StakeholdersinsideandoutsideoftheAgencyhaverecommendedamorecompletecharacterizationofrisks,includinguncertaintiesandvariability,inprotectingmoresensitiveorvulnerablepopulationsandlifestages.PRAcanbeusedtosupportriskmanagementbyassessmentofimpactsofuncertaintiesoneachofthepotentialdecisionalternatives.

Numerousadvisorybodies,suchastheScienceAdvisoryBoard(SAB)andtheNationalResearchCouncil(NRC)oftheNationalAcademyofSciences(NAS),haverecommendedthatEPAincorporateprobabilisticanalysesintotheAgency’sdecision‐makingprocess.EPA’sRiskAssessmentForum(RAF)formedaTechnicalPanel,consistingofrepresentativesfromtheAgency’sprogramandregionaloffices,todevelopthiswhitepaperanditscompaniondocument,titledProbabilisticRiskAssessmenttoInformDecisionMaking:FrequentlyAskedQuestions(FAQ).TheRAFisrecommendingthedevelopmentofAgencyresources,suchasaclearinghouseofPRAcasestudies,bestpractices,resourcesandseminars,toraisegeneralknowledgeabouthowtheseprobabilistictoolscanbeused.

TheintendedgoalofthiswhitepaperistoexplainhowEPAcanuseprobabilisticmethodstoaddressdata,modelandscenariouncertaintyandvariabilitybycapitalizingonthewidearrayoftoolsandmethodsthatcomprisePRA.ThiswhitepaperdescribeswherePRAcanfacilitatemoreinformedriskmanagementdecisionmakingthroughbetterunderstandingofuncertaintyandvariabilityrelatedtoAgencydecisions.Theinformationcontainedinthisdocumentisintendedforbothriskanalystsandmanagersfacedwithdeterminingwhenandhowtoapplythesetoolstoinformtheirdecisions.Thisdocumentdoesnotprescribeaspecificapproachbut,rather,describesthevariousstagesandaspectsofanassessmentordecisionprocessinwhichprobabilisticassessmenttoolsmayaddvalue.

ProbabilisticRiskAssessment

PRAisananalyticalmethodologyusedtoincorporateinformationregardinguncertaintyand/orvariabilityintoanalysestoprovideinsightregardingthedegreeofcertaintyofariskestimateandhowtheriskestimatevariesamongdifferentmembersofanexposedpopulation,includingsensitivepopulationsorlifestages.Traditionalapproaches,suchasdeterministicanalyses,oftenreportrisksas“centraltendency,”“highend”(e.g.,90thpercentileorabove)or“maximumanticipatedexposure,”butPRAcanbeusedtodescribemorecompletelytheuncertaintysurroundingsuchestimatesandidentifythekeycontributorstovariabilityoruncertaintyinpredictedexposuresorriskestimates.Thisinformationthencanbeusedbydecisionmakerstoachieveascience‐basedlevelofsafety,tocomparetherisksrelatedtodifferentmanagementoptions,ortoinvestinresearchwiththegreatestimpactonriskestimateuncertainty.

Tosupportregulatorydecisionmaking,PRAcanprovideinformationtodecisionmakersonspecificquestionsrelatedtouncertaintyandvariability.Forexample,inthecontextofadecisionanalysisthathasbeenconducted,PRAcan:identify“tippingpoints”wherethedecisionwouldbedifferentif

Page 13: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

2

theriskestimatesweredifferent;estimatethedegreeofconfidenceinaparticulardecision;andhelptoestimatetrade‐offsrelatedtodifferentrisksormanagementoptions.PRAcanprovideuseful(evencritical)informationabouttheuncertaintiesandvariabilityinthedata,models,scenario,expertjudgmentsandvaluesincorporatedinriskassessmentstosupportdecisionmakingacrosstheAgency.

PRAisapplicabletobothhumanhealthriskassessment(HHRA)andecologicalriskassessment(ERA);however,therearedifferencesbetweenhowPRAisusedforthetwo.BothHHRAandERAhaveasimilarstructureandusethesameriskassessmentsteps,butHHRAfocusesonindividuals,asinglespecies,morbidityandmortality,butERAismoreconcernedwithmultiplepopulationsoforganisms(e.g.,individualspeciesoffishinariver)orecologicalintegrity(e.g.,willthetypesofspecieslivingintheriverchangeovertime).InERA,therealsoisarelianceonindicatorsofimpacts(e.g.,sentinelspeciesandothermetrics).

RiskAssessmentatEPA

PRAbeganplayinganincreasinglyimportantroleinAgencyriskassessmentsfollowingthe1997releaseofEPA’sPolicyforUseofProbabilisticAnalysisinRiskAssessmentattheU.S.EnvironmentalProtectionAgency(USEPA1997a)andpublicationoftheGuidingPrinciplesforMonte‐CarloAnalysis(USEPA1997b).PRAwasamajorfocusinanassociatedreviewofEPAriskassessmentpracticesbytheSAB(USEPA2007b).TheNRCrecommendedthatEPAadopta“tiered”approachforselectingthelevelofdetailusedinuncertaintyandvariabilityassessment(NRC2009).Furthermore,theNRCrecommendedthatadiscussionaboutthelevelofdetailusedforuncertaintyanalysisandvariabilityassessmentshouldbeanexplicitpartoftheplanning,scopingandproblemformulationstepintheriskassessmentprocess.BoththiswhitepaperandthecompanionFAQdocumenttakeintoaccountrecommendationsonriskassessmentprocessesdescribedintheNRC’sreportScienceandDecisions:AdvancingRiskAssessment(NRC2009)andEnvironmentalDecisionsintheFaceofUncertainty(IOM2013).

EPA’srecentriskassessmentpublications,includingthedocumenttitledFrameworkforHumanHealthRiskAssessmenttoInformDecisionMaking(UAEPA2014b)aswellasthiswhitepaper,emphasizetheimportanceofcommunicatingtheresultsofaPRAbecauseitprovidestherangeandlikelihoodestimatesforoneormoreaspectsofhazard,exposureorrisk,ratherthanasinglepointestimate.Riskassessorsareresponsibleforsharinginformationonprobabilisticresultssothatdecisionmakershaveaclearunderstandingofquantitativeassessmentsofuncertaintyandvariability,andhowthisinformationwillaffectthedecision.EffectivecommunicationbetweentheriskassessoranddecisionmakeriskeytopromoteunderstandinganduseoftheresultsfromthePRA.

PRAgenerallyrequiresmoreresourcesthanstandardAgencydefault‐baseddeterministicapproaches.Appropriatelytrainedstaffandtheavailabilityofadequatetools,methodsandguidanceareessentialfortheapplicationofPRA.Properapplicationofprobabilisticmethodsrequiresnotonlysoftwareanddata,butalsoguidanceandtrainingforanalystsusingthetools,andformanagersanddecisionmakerstaskedwithinterpretingandcommunicatingtheresults.Inmostcircumstances,probabilisticassessmentsmaytakemoretimeandefforttoconductthanconventionalapproaches,primarilybecauseofthecomprehensiveinclusionofavailableinformationonmodelinputs.Thepotentiallyhigherresourcecostsmaybeoffset,however,byamoreinformeddecisionthanwouldbeprovidedbyacomparabledeterministicanalysis.

ContentoftheWhitePaperandFrequentlyAskedQuestionsCompanionDocuments

ThesetwodocumentsdescribehowPRAcanbeappliedtoenhancethescientificfoundationofEPA’sdecisionmakingacrosstheAgency.ThiswhitepaperdescribesthechallengesfacedbyEPA

Page 14: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

3

decisionmakers,definesandexplainsthebasicprinciplesofprobabilisticanalysis,brieflyhighlightsinstanceswherethesetechniqueshavebeenimplementedinEPAdecisions,anddescribescriteriathatmaybeusefulindeterminingwhetherandhowtheapplicationofprobabilisticmethodsmaybeusefuland/orapplicabletodecisionmaking.Thiswhitepaperalsodescribescommonlyemployedmethodstoaddressuncertaintyandvariability,includingthoseusedintheconsiderationofuncertaintyinscenariosanduncertaintyinmodels.Additionally,itaddressesuncertaintyandvariabilityintheinputsandoutputsofmodelsandtheimpactoftheseuncertaintiesoneachofthepotentialmanagementoptions.Ageneraldescriptionoftherangeofmethodsfromsimpletocomplex,rapidtomoretimeconsumingandleasttomostresourceintensiveisprovided,aswellasusesofthesemethods.

Bothdocumentsaddressissuessuchasuncertaintyandvariability,theirrelevancetodecisionmakingandthePRAgoaltoprovidequantitativecharacterizationoftheuncertaintyandvariabilityinestimatesofhazard,exposure,orrisk.ThedifferencebetweenthewhitepaperandtheFAQsdocumentisthelevelofdetailprovidedaboutPRAconceptsandpractices,andtheintendedaudience(e.g.,riskassessorsvs.decisionmakers).DetailedexamplesofapplicationsofthesemethodsareprovidedintheAppendixofthiswhitepaper,whichistitled“CaseStudyExamplesoftheApplicationofProbabilisticRiskAnalysisinU.S.EnvironmentalProtectionAgencyDecisionMaking.”ThewhitepaperAppendixincludes16casestudies—11HHRAand5ERAexamples—thatillustratehowEPA’sprogramandregionalofficeshaveusedprobabilistictechniquesinriskassessment.Toaidindescribinghowthesetoolswereapplied,the16casestudiesaresubdividedamong3categoriesforpurposesofthisdocument.Group1includes2casestudiesdemonstratingpointestimate,includingsensitivityanalysis;Group2iscomprisedof5casestudiesdemonstratingprobabilisticriskanalysis,includingone‐dimensionalMonteCarloanalysisandprobabilisticsensitivityanalysis;andGroup3includes9casestudiesdemonstratingadvancedprobabilisticriskanalysis,includingtwo‐dimensionalMonteCarloanalysiswithmicroexposure(microenvironments)modeling,Bayesianstatistics,geostatisticsandexpertelicitation.

TheFAQdocumentprovidesanswerstocommonquestionsregardingPRA,includingkeyconceptssuchasscientificandinstitutionalmotivationsfortheuseofPRA,andchallengesintheapplicationofprobabilistictechniques.TheprincipalreasonforincludingPRAasanoptionintheriskassessor’stoolboxisitsabilitytosupporttherefinementandimprovementoftheinformationleadingtodecisionmakingbyincorporatingknownuncertainties.

Page 15: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

4

1. INTRODUCTION: RELEVANCE OF UNCERTAINTY TO DECISION MAKING: HOW PROBABILISTIC APPROACHES CAN HELP

1.1. EPA Decision Making TodiscusswhereprobabilisticapproachescanaidEPA’sdecisionmaking,itisimportanttogenerallydescribetheAgency’scurrentdecision‐makingprocessesandhowbetterunderstandingandimprovingelementswithintheseprocessescanclarifywhereprobabilisticapproachesmightprovidebenefits.TheenhanceduseofPRAandcharacterizationofuncertaintywouldallowEPAdecisionmakersopportunitiestouseamorerobustandtransparentprocess,whichmayallowgreaterresponsivenesstooutsidecommentsandrecommendations.SuchanapproachwouldsupporthigherqualityEPAassessmentsandimproveconfidenceinAgencydecisions.

Therearetwomajorareasinthedecision‐makingprocessthatmightbeimprovedwithPRA.Scientistscurrentlyaregenerallyfocusedonthefirstarea—theunderstandingofdata,modelandscenariouncertaintiesandvariability.Thesecondareaisonethathasnot,untilrecentlyandonlyinalimitedfashion,beenusedbyEPAdecisionmakers.Thisareaisformaldecisionanalysis.Withdecisionanalytictechniques,decisionmakerscanweightherelativeimportanceofriskinformationcomparedtootherinformationinmakingthedecision,understandhowuncertaintyaffectstherelativeattractivenessofpotentialdecisionalternatives,andassessoverallconfidenceinadecision.Inadditiontodata,modelandscenariouncertainty,thereisaseparatecategoryofuncertaintiesspecificallyassociatedwithhowthedecisioncriteriarelatetothedecisionalternatives.Althoughitisquiterelevanttoriskmanagementdecisions,thetopicanddecisionanalysisingeneralareoutsideofthescopeofthisreport.Thiswhitepaperfocusesontechnicalinformationthatwouldallowbetterunderstandingoftherelationshipsamongalternativedecisionsinassessingrisks.

1.2. The Role of Probabilistic Risk Analysis in Characterizing Uncertainty and Variability

Probabilisticanalysesincludetechniquesthatcanbeappliedformallytoaddressbothuncertaintyandvariability,typicallyarisingfromlimitationsofdata,modelsoradequatelyformulatingthescenariosusedinassessingrisks.Probabilityisusedinscience,business,economicsandotherfieldstoexamineexistingdataandestimatethechanceofanevent,fromhealtheffectstoraintomentalfatigue.Onecanuseprobability(chance)toquantifythefrequencyofoccurrenceorthedegreeofbeliefininformation.Forvariability,probabilitydistributionsareinterpretedasrepresentingtherelativefrequencyofagivenstateofthesystem(e.g.,thatthedataaredistributedinacertainway);foruncertainty,theyrepresentthedegreeofbelieforconfidencethatagivenstateofthesystemexists(e.g.,thatwehavetheappropriatedata;CullenandFrey1999).PRAoftenisdefinednarrowlytoindicateastatisticalorthoughtprocessusedtoanalyzeandevaluatethevariabilityofavailabledataortolookatuncertaintyacrossdatasets.

Forthepurposesofthisdocument,PRAisatermusedtodescribeaprocessthatemploysprobabilitytoincorporatevariabilityindatasetsand/ortheuncertaintyininformation(suchasdataormodels)intoanalysesthatsupportenvironmentalrisk‐baseddecisionmaking.PRAisusedherebroadlytoincludebothquantitativeandqualitativemethodsfordealingwithscenario,modelandinputuncertainty.Probabilistictechniquescanbeusedwithothertypesofanalysis,suchasbenefit‐costanalysis,regulatoryimpactanalysisandengineeringperformancestandards;thus,theycanbeusedforavarietyofapplicationsandbyexpertsinmanydisciplines.

Page 16: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

5

1.3. Goals and Intended Audience TheprimarygoalsofthiswhitepaperaretointroducePRA,describehowitcanbeusedtobetterinformandimprovethedecision‐makingprocess,andprovidecasestudieswhereithasbeenusedinhumanhealthandecologicalanalysesatEPA(seetheAppendixforthecasestudies).AsecondarygoalofthispaperistobridgecommunicationgapsregardingPRAamonganalystsofvariousdisciplines,betweentheseanalystsandAgencydecisionmakers,andamongaffectedstakeholders.Thiswhitepaperalsoisintendedtoserveasacommunicationtooltointroducekeyconceptsandbackgroundinformationonapproachestoriskanalysisthatincorporateuncertaintyandprovideamorecomprehensivetreatmentofvariability.Riskanalysts,decisionmakersandaffectedstakeholderscanbenefitfromunderstandingthepotentialusesofPRA.PRAandrelatedapproachescanbeusedtoidentifyadditionalresearchthatmayreduceuncertaintyandmorethoroughlycharacterizevariabilityinariskassessment.ThiswhitepaperexplainshowPRAcanenhancethedecision‐makingprocessesfacedbymanagersatEPAbybettercharacterizingdata,model,scenarioanddecisionuncertainties.

1.4. Overview of This Document ThiswhitepaperprovidesanoverviewofEPA’sinterestandexperienceinaddressinguncertaintyandvariabilityusingprobabilisticmethodsinriskassessment;identifieskeyquestionsaskedorfacedbyAgencydecisionmakers;demonstrateshowconventionaldeterministicapproachestoriskanalysismaynotanswerthesequestionsfully;providesexamplesofapplications;andshowshowandwhy“probabilisticriskanalysis”(broadlydefined)couldprovideaddedvalue,comparedtotraditionalmethods,withregardtoregulatorydecisionmakingbymorefullycharacterizingriskestimatesandexploringdecisionuncertainties.Forthepurposesofthiswhitepaper,PRAandrelatedtoolsforbothhumanhealthandecologicalassessmentsincludearangeofapproaches,fromstatisticaltools,suchassensitivityanalysis,tomulti‐dimensionalMonteCarlomodels,geospatialapproachesandexpertelicitation.KeypointsaddressedbythisdocumentincludedefinitionsandkeyconceptspertainingtoPRA,benefitsandchallengesofPRA,ageneralconceptualframeworkforPRA,conclusionsregardingproductsandinsightsobtainedfromPRA,andexampleswhereEPAhasusedPRAinhumanhealthandecologicalanalyses.AGlossaryandaBibliographyalsoareprovided.

1.5. What Are Common Challenges Facing EPA Risk Decision Makers?

EPAoperatesunderstatutoryandregulatoryconstraintsthatoftenlimitthetypesofcriteriathatcanbeconsidered(includingwhethertheuseofPRAisappropriate)andimposestricttimeframesinwhichdecisionsmustbemade.Typically,thedecisionbeginswithunderstanding(1)whoorwhatwillbeprotected;(2)therelationshipbetweenthedataanddecisionalternatives;and(3)theimpactofdata,modelanddecisionuncertaintiesrelatedtoeachdecisionalternative.Theseareamongtheconsiderationsoftheplanningandscopingandproblemformulationphasesofriskassessment(USEPA2014).EPAdecisionmakersneedtoconsidermultipledecisioncriteria,whichareinformedbyvaryingdegreesofconfidenceintheunderlyinginformation.Decisionmakersneedtobalancetheregulatory/statutoryrequirementsandtimeframes,resources(i.e.,expertise,costsoftheanalysis,reviewtimes,etc.)toconducttheassessment,managementoptions,andstakeholderswhileatthesametimekeepingriskassessmentanddecisionmakingseparate.

Uncertaintycanbeintroducedintoanyassessmentatanystepintheprocess,evenwhenusinghighlyaccuratedatawiththemostsophisticatedmodels. Uncertaintycanbereducedorbettercharacterizedthroughknowledge.Variabilityornaturalheterogeneityisinherentinnaturalsystemsandthereforecannotbereduced,butcanbeexaminedanddescribed.Uncertaintyindecisionsisunavoidablebecausereal‐worldsituationscannotbeperfectlymeasured,modeledor

Page 17: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

6

predicted.Asaresult,EPAdecisionmakersfacescientificallycomplexproblemsthatarecompoundedbyvaryinglevelsofuncertaintyandvariability.Ifuncertaintyandvariabilityhavenotbeenwellcharacterizedoracknowledged,potentialcomplicationsariseintheprocessofdecisionmaking.Increaseduncertaintycanmakeitmoredifficulttodetermine,withreasonableconfidence,thebalancepointbetweenthecostsofregulationandtheimplicationsforavoidingdamagesandproducingbenefits.Characterizationfacilitatedbyprobabilisticanalysescanprovideinsightintoweighingtherelativecostsandbenefitsofvaryinglevelsofregulationandalsocanassistinriskcommunicationactivities.

Decisionmakersoftenwanttoknowwhoisatriskandbyhowmuch,thetradeoffsbetweenalternativeactionsandthelikelyorpossibleconsequencesofdecisions.Tothisend,itisparticularlyusefulfordecisionmakerstounderstandthedistributionofriskacrosspotentiallyimpactedpopulationsandecologicalsystems.Itcanbeimportanttoknowthenumberofindividualsexperiencingdifferentmagnitudesofrisk,thedifferencesinriskmagnitudeexperiencedbyindividualsindifferentlifestagesorpopulationsortheprobabilityofaneventthatmayleadtounacceptablelevelsofrisk.Giventhelimitationsofdata,traditionalmethodsofriskanalysesarenotwellsuitedtoproducesuchestimates.Probabilisticanalyticalmethodsarecapableofaddressingtheseshortcomingsandcancontributetoamorethoroughrecognitionoftheimpactofdatagapsontheprojectedriskestimates.AlthoughPRAcanbeusedtocharacterizetheuncertaintyandvariabilityinsituationswithlimiteddata,currentlythereisnotextensiveexperienceusingPRAtocharacterizetherangeofeffectsordose‐responserelationshipsforpopulations,includingsensitivepopulationsandlifestages.

OtherchallengesfacingEPAdecisionmakersincludetheneedtoconsidermultipledecisioncriteria,whichareinformedbyvaryingdegreesofconfidenceintheunderlyinginformation,understandingtherelationshipbetweenandamongthosedecisioncriteria(includingmulti‐pollutantandmulti‐mediaeffects)andthedecisionalternatives,andthetimelinessofthedecisionmaking.Furthermore,evenwhenPRAisused,EPAdecisionmakersmustbemindfulofpotentialmisusesandobfuscationswhenconductingorpresentingPRAresults.DecisionmakersalsoneedtoconsidertheevolvingsciencebehindPRA.AstheuseofPRAincreasesdecisionmakerswillbecomemorefamiliarwiththetechniquesandtheirapplication.

Ariskassessmentprocessneedstoconsideruncertainties,variabilityandtherationaleorfactorsinfluencinghowtheymaybeaddressedbyadecisionmaker.Decisionmakersneedafoundationforestimatingthevalueofcollectingadditionalinformationtoallowforbetterinformeddecisions.Therearecostsassociatedwithignoringuncertainty(McConnell1997andToll1999),andafocusbydecisionmakersontheinformationprovidedbyuncertaintyanalysiscanstrengthentheirchoices.

1.6. What Are Key Uncertainty and Variability Questions Often Asked by Decision Makers?

Asdescribedabove,determiningthedecision‐makingcontextandspecificconcernsisacriticalfirststeptowarddevelopingausefulandresponsiveriskassessmentthatwillsupportthedecision.Forexample,theappropriatefocusandlevelofdetailoftheanalysisshouldbecommensuratewiththeneedsofthedecisionmakerandstakeholders,aswellastheappropriateuseofscience.Analysesoftenareconductedatalevelofdetaildictatedbytheissuebeingaddressed,thebreadthandqualityoftheavailableinformationuponwhichtobaseananalysis,andthesignificancesurroundingadecision.Theanalyticalprocesstendstobeiterative.Althoughaguidingsetofquestionsmayframetheinitialanalyses,additionalquestionscanarisethatfurtherdirectorevenreframetheanalyses.

Page 18: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

7

BasedonaseriesofdiscussionswithAgencydecisionmakersandriskassessors,sometypicalquestionsaboutuncertaintyandvariabilityrelevanttoriskanalysesincluding:

Factorsinfluencingdecisionuncertainty:

Wouldmydecisionbedifferentifthedataweredifferent,improvedorexpanded?Wouldadditionaldatacollectionandresearchlikelyleadtoadifferentdecision?Howlongwillittaketocollecttheinformation,howmuchwoulditcost,andwouldtheresultingdecisionbesignificantlyaltered?

Whataretheliabilitiesandconsequencesofmakingadecisionunderthecurrentlevelofknowledgeanduncertainty?

Howdothealternativesandtheirassociateduncertaintyandvariabilityaffectthetargetpopulationorlifestage?

Considerationsforevaluatingdataormethoduncertainty:

Howrepresentativeorconservativeistheestimateduetodataormethoduncertainty(alsoincorporatingvariability)?

Whatarethemajorgapsinknowledge,andwhatarethemajorassumptionsusedintheassessment?Howreasonablearetheassumptions?

Issuesarisingwhenaddressingvariability:

Canaprobabilisticapproach(e.g.,tobettercharacterizeuncertaintiesandvariability)beaccomplishedinatimelymanner?

Whatisthedesiredpercentileofthepopulationtobeprotected?Bychoosingthispercentile,whomaynotbeprotected?

Thequestionsthatariseconcerninguncertaintyandvariabilitychangedependingonthestageandnatureofthedecision‐makingprocessandanalysis.GeneralphasesoftheriskassessmentprocessareillustratedinFigure1.Forfurtherinformationontheprocessofdecisionmaking,wesuggestreferringtothedescriptionprovidedbyEPARegion3ontheMulti‐CriteriaIntegratedResource

Page 19: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

8

Figure 1. General Phases of the Risk Assessment Process. Risk assessment is an iterative process comprised of planning, scoping and problem formulation; analysis (e.g., hazard identification, dose‐response assessment and exposure assessment); interpretation and risk characterization; and risk communication. The highlighted boxes explain how PRA fits into the overall process. 

AssessmentInternetpageathttp://www.epa.gov/reg3esd1/data/mira.htm.TheutilityofvariouslevelsofanalysisandsophisticationinansweringthesequestionsisillustratedinthecasestudiesdescribedinSection1.10andpresentedintheAppendixofthiswhitepaper.ReferencestoexamplesbeyondtheseEPAcasestudiescanbefoundintheBibliography.Additionally,Lesteretal.(2007)identifiedmorethan20PRAapplicationcasestudies(includingEPAexamples)performedsince2000;thesecasestudyexamplesarecategorizedassite‐specificapplicationsandregionalriskassessments.

Page 20: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

9

1.7. Why Is the Implementation of Probabilistic Risk Analysis Important?

TheprincipalreasonfortheinclusionofPRAasanoptionintheriskassessor’stoolboxisPRA’sabilitytosupportrefinementandimprovementoftheinformationleadingtodecisionmakingbyincorporatingknownuncertainties.Beginningasearlyasthe1980s,expertscientificadvisorygroups,suchastheNationalResearchCouncil(NRC),recommendedthatriskanalysesincludeacleardiscussionoftheuncertaintiesinriskestimation(NRC1983).TheNRCstatedtheneedtodescribeuncertaintyandtocapturevariabilityinriskestimates(NRC1994).ThePresidential/CongressionalCommissiononRiskAssessmentandRiskManagement(PCCRARM)recommendedagainstarequirementorneedfora“brightline”orsingle‐numberlevelofrisk(PCCRARM1997).SeeSection2.4formoreinformationregardingthescientificcommunity’sopinionontheuseofPRA.

Regulatoryscienceoftenrequiresselectionofalimitforacontaminant,yetthatlimitalwayscontainsuncertaintyastohowprotectiveitis.PRAandrelatedtoolsquantitativelydescribetheveryrealvariationsinnaturalsystemsandlivingorganisms,howtheyrespondtostressors,andtheuncertaintyinestimatingthoseresponses.

RiskcharacterizationbecameEPApolicyin1995(USEPA1995b),andtheprinciplesoftransparency,clarity,consistencyandreasonablenessareexplicatedinthe2000RiskCharacterizationHandbook(USEPA2000a).Transparency,clarity,consistencyandreasonablenesscriteriarequiredecisionmakerstodescribeandexplaintheuncertainties,variabilityandknowndatagapsintheriskanalysisandhowtheyaffecttheresultingdecision‐makingprocesses(USEPA1992,1995a,2000a).

TheuseofprobabilisticmethodsalsohasreceivedsupportfromsomedecisionmakerswithintheAgency,andthesemethodshavebeenincorporatedintoanumberofEPAdecisionstodate.Programoffices,suchastheOfficeofPesticidePrograms(OPP),OfficeofSolidWasteandEmergencyResponse(OSWER),OfficeofAirandRadiation(OAR),andOfficeofWater(OW),aswellastheOfficeofResearchandDevelopment(ORD),haveutilizedprobabilisticapproachesindifferentwaysandtovaryingextents,forbothhumanexposureandecologicalriskanalyses.Inaddition,OSWERhasprovidedexplicitguidanceontheuseofprobabilisticapproachesforexposureanalysis(USEPA2001).SomeprogramofficeshaveheldtrainingsessionsonMonteCarlosimulation(MCS)softwarethatisusedfrequentlyinprobabilisticanalyses.

TheNRCrecommendedthatEPAshouldadoptatieredapproachforselectingthelevelofdetailusedinuncertaintyandvariabilityassessment(NRC2009).Furthermore,NRCrecommendedthatadiscussionaboutthelevelofdetailusedforuncertaintyanalysisandvariabilityassessmentshouldbeanexplicitpartoftheplanning,scopingandproblemformulationstepintheriskassessmentprocess.ThewaythatPRAfitsintoagraduatedhierarchical(tiered)approachismorefullydescribedinSection2.10andillustratedinFigure2.

Whenitisbeneficialtorefineriskestimates,theuseofPRAcanhelpinthecharacterizationandcommunicationofuncertainty,variabilityandtheimpactofdatagapsinriskanalysesforassessors,decisionmakersandstakeholders(includingthetargetpopulationorlifestage).

Page 21: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

10

Figure 2. Tiered Approach for Risk Assessment. The applicability of a probabilistic approach depends on the needs of decision makers and stakeholders. Assessments that are high in complexity and regulatory significance benefit from the application of probabilistic techniques.  Source: Adapted from USEPA 2004a and WHO 2008. 

1.8. How Does EPA Typically Address Scientific Uncertainty and Variability?

Environmentalassessmentscanbecomplex,suchascoveringexposuretomultiplechemicalsinmultiplemediaforawide‐rangingpopulation.TheAgencyhasdevelopedsimplifiedapproachestocharacterizerisksassociatedwithsuchcomplexassessmentsthroughtheuseofpointestimatesformodelvariablesorparameters.Suchanapproachtypicallyproducespointestimatesofrisks(e.g.,10‐5oralifetimeprobabilityofcancerriskofoneindividualin100,000).Theseoftenarecalled“deterministic”assessments.Asaresultoftheuseofpointestimatesforvariablesinmodelalgorithms,deterministicriskresultsusuallyarereportedaswhatareassumedtobeeitheraverageorworst‐caseestimates.Theydonotcontainanyquantitativeestimateoftheuncertaintyinthatestimate,norreportwhatpercentileoftheexposedpopulationtheestimateapplies.ThemethodstypicallyusedinEPAriskassessmentsrelyonacombinationofpointvalueswithpotentiallyvaryinglevelsofconservatismandcertainty,yieldingapointestimateofexposureatsomepointintherangeofpossiblerisks.

Becauseuncertaintyisinherentinallriskassessments,itisimportantthattheriskassessmentprocessenablehandlinguncertaintiesinalogicalwaythatistransparentandscientificallydefensible,consistentwiththeAgency’sstatutorymissionandresponsivetotheneedsofdecisionmakers(NRC1994).Uncertaintyisafactorinbothecologicalandhumanhealthriskassessments.

Page 22: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

11

Forhumanhealthriskassessments,uncertaintiesariseforbothnoncancerandcancerendpoints.Thus,whendataaremissing,EPAoftenusesseveraloptionstoprovideboundariesonuncertaintyandvariabilityinanattempttoavoidriskunderestimation;attemptingtogiveasinglequantificationofhowmuchconfidencethereisintheriskestimatemaynotbeinformativeorfeasible.

Inexposureassessment,forexample,thepracticeatEPAistocollectnewdatawheretheyareneededandwheretimeandresourcesallow.Alternativeapproachestoaddressuncertaintyincludenarrowingthescopeoftheassessment;usingscreening‐leveldefaultassumptionsthatincludeupper‐endvaluesand/orcentraltendencyvaluesthataregenerallycombinedtogenerateriskestimatesthatfallwithinthehigherendofthepopulationriskrange(USEPA2004b);applyingmodelstoestimatemissingvalues;usingsurrogatedata(e.g.,dataonaparameterthatcomefromadifferentregionofthecountrythantheregionbeingassessed);orapplyingprofessionaljudgment.Theuseofindividualassumptionscanrangefromqualitative(e.g.,assumingthatoneissecuredtotheresidencelocationanddoesnotmovethroughtimeorspace)tomorequantitative(e.g.,usingthe95thpercentileofasampledistributionforaningestionrate).Thisapproachalsocanbeappliedtothepracticeofhazardidentificationanddose‐responseassessmentwhendataaremissing.Identifyingthesensitivityofexposureorriskestimatestokeyinputscanhelpfocuseffortstoreduceuncertaintybycollectingadditionaldata.

CurrentEPApracticestoaddressuncertaintyandvariabilityarefocusedontheevaluationofdata,model,andscenariouncertaintyandvariability.Inaddition,decisionmakersarefacedwithcombiningmanydifferentdecisioncriteriathatmaybeinformedbyscienceandPRAaswellasbyexpertjudgmentortheweightingofvaluestochooseadecisionalternative.Data,model,andscenariouncertaintiesandvariability(includingtheirprobabilitydistributions),aswellasexpertjudgment,canbeimportantconsiderationsintheselectionofonealternativeoveranother(Costanzaetal.1997;Morganetal.2009;StahlandCimorelli2005;Wrightetal.2002).

1.9. What Are the Limitations of Relying on Default-Based Deterministic Approaches?

Default‐baseddeterministicapproachesareappliedtodata,modelandscenariouncertainties.Deterministicriskassessment(DRA)oftenisconsideredatraditionalapproachtoriskanalysisbecauseoftheexistenceofestablishedguidanceandproceduresregardingitsuse,theeasewithwhichitcanbeperformed,anditslimiteddataandresourceneeds.TheuseofdefaultssupportingDRAprovidesaproceduralconsistencythatallowsforriskassessmentstobefeasibleandtractable.DecisionmakersandmembersofthepublictendtoberelativelyfamiliarwithDRA,andtheuseofsuchanapproachaddressesassessment‐relateduncertaintiesprimarilythroughtheincorporationofpredetermineddefaultvaluesandconservativeassumptions.Itaddressesvariabilitybycombininginputparametersintendedtoberepresentativeoftypicalorhigherendexposure(i.e.,consideredtobeconservativeassumptions).Theintentionoftenistoimplicitlyprovideamarginofsafety(i.e.,morelikelytooverestimateriskthanunderestimaterisk)orconstructascreening‐levelestimateofhigh‐endexposureandrisk(i.e.,anestimaterepresentativeofmorehighlyexposedandsusceptibleindividuals).

DRAprovidesanestimationoftheexposuresandresultingrisksthataddressesuncertaintiesandvariabilitiesinaqualitativemanner.ThemethodstypicallyusedinEPADRArelyonacombinationofpointvaluessomeconservativeandsometypicalyieldingapointestimateofexposurethatisatsomeunknownpointintherangeofpossiblerisks.AlthoughthisconservativebiasalignswiththepublichealthmissionofEPA(USEPA2004b),thedegreeofconservatismintheseriskestimates(andinanyconcomitantdecision)cannotbeestimatedwellorcommunicated(HattisandBurmaster1994).Typically,thisresultsinunquantifieduncertaintyinriskstatements.

Page 23: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

12

Quantitativeinformationregardingtheprecisionorpotentialsystematicerrorandthedistributionofexposures,effectsandresultingrisksacrossdifferentmembersofanexposedpopulationareusuallynotprovidedwithestimatesgeneratedusingdefaultapproaches.AlthoughDRAmaypresentqualitativeinformationregardingtherobustnessoftheestimates,theimpactofdataandmodellimitationsonthequalityoftheresultscannotbequantified.Relianceondeterministicallyderivedestimationsofriskcanresultindecisionmakingbasedsolelyonpointestimateswithanunknowndegreeofconservatism,whichcancomplicatethecomparisonofrisksormanagementoptions.

Inriskassessmentsofnoncancerendpoints,metricssuchasanoralreferencedose(RfD)andaninhalationreferenceconcentration(RfC)aretypicallyused.Theuseofconservativedefaultslonghasbeenthetargetofcriticism(Finkel1989)andhasledtothepresumptionbycriticsthatEPAassessmentsareoverlyconservativeandunrealistic.TheuseofPRAwouldbeadvantageousineliminatingasinglevalueandmightbelesslikelytoimplyundueprecisionandlessentheneedforconservativeassumptions,therebyreducingbiasintheestimate.Intheprobabilisticframework,aprobabilitydistributionwouldbeusedtoexpressthebeliefthatanyparticularvaluerepresentsthedoseorexposureconcentrationthatwouldposenoappreciableriskofadverseeffects(NRC2009).EPAisinvestigatingtheuseofPRAtoderiveriskvaluesforRfDandRfCinEPA’sIntegratedRiskInformationSystem(IRIS)Database(www.epa.gov/IRIS/).

EPAcommissionedawhitepaper(HattisandLynch2010)presentedattheHazardousAirPollutantWorkshop,2009,illustratingtheimplementationofprobabilisticmethodsindefiningRfDsandassessingthebenefitsforreducingexposuretotoxicantsthatactinpartthroughtraditionalindividualthresholdprocesses.TheuseofPRA,amongotherthings,makesprovisionforinteractionswithbackgroundpathologicalprocesses,asrecommendedbytheNRC(2009),andshowshowthesystemcaninformassessmentsfor“data‐poor”toxicants.

PRAmaybemoresuitablethanDRAforcomplexassessments,includingthoseofaggregateandcumulativeexposuresandtime‐dependentindividualexposure,doseandeffectsanalyses.Identificationandprioritizationofcontributorysourcesofuncertaintycanbedifficultandtimeconsumingwhenusingdeterministicmethods,leadingtodifficultiesinmodelevaluationandthesubsequentappraisalofriskestimates(CullenandFrey1999).Quantitativeanalysesofmodelsensitivitiesareessentialfortheprioritizationofkeyuncertaintiesacriticalprocessinidentifyingstepsfordatacollectionorresearchtoimproveexposureorriskestimates.

1.10. What Is EPA’s Experience with the Use of Probabilistic Risk Analysis?

EPA’sexperiencewithPRAhas,todate,primarilybeenlimitedtotheevaluationofdata,modelandscenariouncertainties.Toassistwiththegrowingnumberofprobabilisticanalysesofexposuredataintheseuncertaintyareas,EPAissuedGuidingPrinciplesforMonteCarloAnalysis(USEPA1997b).Givenadequatesupportingdataandcredibleassumptions,probabilisticanalysistechniques,suchasMonteCarloanalysis,canbeviablestatisticaltoolsforanalyzinguncertaintyandvariabilityinriskassessments.EPA’spolicyfortheuseofprobabilisticanalysisinriskassessment,releasedin1997,isinclusiveofhumanexposureandecologicalriskassessmentsanddoesnotruleoutprobabilistichealtheffectsanalyses(USEPA1997a).Subsequently,EPA’sSABandScientificAdvisoryPanel(SAP)havereviewedPRAapproachestorisksusedbyEPAofficessuchasOAR,OPPandothers.SeveralprogramshavedevelopedspecificguidanceontheuseofPRA,includingOPPandOSWER(USEPA1998a,2001).

ToillustratethepracticalapplicationofPRAtoproblemsrelevanttotheAgency,severalexamplecasestudiesarebrieflydescribedhere.TheAppendixtitledCaseStudyExamplesofApplicationof

Page 24: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

13

ProbabilisticRiskAnalysisinU.S.EnvironmentalProtectionAgencyRegulatoryDecisionMaking,discussestheseandothercasestudiesingreaterdetail,includingtheproceduresandoutcomes.TheAppendixincludes16casestudies—11HHRAand5ERAexamples—thatareintendedtoillustratehowsomeofEPA’sprogramsandofficescurrentlyutilizePRA.Toaidindescribinghowprobabilisticanalyseswereused,the16casestudiesaresubdividedamong3categoriesofPRAtools:Group1—pointestimate,includingsensitivityanalysis;Group2—probabilisticriskanalysis,includingone‐dimensionalMonteCarloanalysis(1‐DMCA)andprobabilisticsensitivityanalysis;andGroup3—advancedprobabilisticriskanalysis,includingtwo‐dimensionalMonteCarloanalysis(2‐DMCA)withmicroexposure(microenvironments)modeling,Bayesianstatistics,geostatisticsandexpertelicitation.

ItisusefultonotethattheNRC(2009)recommendedatieredapproachtoriskassessmentusingbothqualitativeandquantitative(deterministicandprobabilistic)tools,withthecomplexityoftheanalysisincreasingasprogressismadethroughthetiers.TheuseofPRAtoolstoaddressissuesofuncertaintyandvariabilityinatieredapproachisdescribedmorecompletelyinSection2.10andwasillustratedinFigure2.ThethreetiersillustratedinthatfigureapproximatelycorrespondtothethreegroupsofEPAcasestudiesdescribedintheAppendixthatprovideexamplesoftheuseofvariousPRAtools.

TableA‐1intheAppendixoffersasummaryofthe16casestudiesbasedonthetypeofriskassessment,thePRAtoolsusedintheassessment,andtheEPAprogramorregionalofficeresponsiblefortheassessment.Someoftheapproachesthatareprofiledinthesecasestudiescanbeusedintheplanningandscopingphasesofriskassessmentsandriskmanagement.Other,morecomplexPRAapproachesareusedtoanswermorespecificquestionsandprovidearicherdescriptionoftherisks.MoststudiesshowthatPRAcanimproveorexpandoninformationgeneratedbydeterministicmethods.Insomeofthecasestudies,theuseofmultiplePRAtoolsisillustrated.Forexample,CaseStudy1describestheuseofapointestimatesensitivityanalysistoidentifyexposurevariablescriticaltotheanalysissummarizedinCaseStudy9.Bothofthesecasestudiesfocusonchildren’sexposuretochromatedcopperarsenate(CCA)‐treatedwood.InCaseStudy9,anMCAwasusedasanexampleofatwo‐dimensional(i.e.,addressingbothvariabilityanduncertainty)probabilisticexposureassessment.

Overall,thecasestudiesillustratethattheAgencyalreadyhasappliedthescienceofPRAtoecologicalriskandhumanexposureestimationandhasbegunusingPRAtodescribehealtheffects.Someoftheapplicationshaveusedexisting“off‐the‐shelf”software,whereasothershaverequiredsignificanteffortandresources.Oncedeveloped,however,someofthemorecomplexmodelshavebeenusedmanytimesfordifferentassessments.Alloftheassessmentshavebeenvalidatedbyinternalandexternalpeerreview.Table1givessomehighlightsthecasestudiesfromdeterministictomorecomplexassessments,whicharedescribedinmoredetailintheAppendix.

Page 25: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

14

Table 1. Selected Examples of EPA Applications of Probabilistic Risk Assessment Techniques 

Case Study No. Description Group Type of Risk

Assessment Office/Region

2

Atmospheric Deposition to Watershed Contamination: The Office of Research and Development (ORD) developed an analysis of nitrogen, mercury and polycyclic aromatic hydrocarbons (PAHs) depositions toward watershed contamination in the Casco Bay Estuary in southwestern Maine.

Group 1: Point Estimate

Ecological ORD

5

Hudson River Polychlorinated Biphenyl (PCB)-Contaminated Sediment Site: Region 2 evaluated the variability in risks to anglers who consume recreationally caught fish contaminated with PCBs from sediment contamination in the Hudson River.

Group 2: 1-D Monte Carlo Analysis

Human Health Superfund/ Region 2

(New York)

7

Environmental Monitoring and Assessment Program (EMAP): ORD developed and the Office of Water (OW) applied probabilistic sampling techniques to evaluate the Nation’s aquatic resources under the Clean Water Act (CWA) Section 305(b).

Group 2: Probabilistic Sensitivity Analysis

Ecological ORD/OW

9

Chromated Copper Arsenate (CCA) Risk Assessment: ORD and the Office of Pesticide Programs (OPP) conducted a probabilistic assessment of children’s exposure (addressing both variability and uncertainty) to arsenic and chromium from contact with CCA-treated wood play sets and decks.

Group 3: 2-D Monte Carlo Analysis

Human Health ORD/OPP

13

Evaluating Ecological Effects of Pesticide Uses: OPP developed a probabilistic model, which evaluates acute mortality levels in generic and specific ecological species for user-defined pesticide uses and exposures.

Group 3: Probabilistic Analysis

Ecological OPP

14

Fine Particulate Matter Health Impacts: ORD and the Office of Air and Radiation (OAR) used expert elicitation to more completely characterize, both qualitatively and quantitatively, the uncertainties associated with the relationship between reduction in fine particulate matter (PM2.5) and benefits of reduced PM2.5-related mortality.

Group 3: Expert Elicitation

Human Health ORD/OAR

Page 26: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

15

2. PROBABILISTIC RISK ANALYSIS 2.1. What Are Uncertainty and Variability, and How Are They Relevant

to Decision Making? Theconceptsofuncertaintyandvariabilityareintroducedhere,andtherelevanceoftheseconceptstodecisionmakingisdiscussed.

2.1.1. Variability Variabilityreferstorealdifferencesovertime,spaceormembersofapopulationandisapropertyofthesystembeingstudied(e.g.,drinkingwaterconsumptionratesforeachofthemanyindividualadultresidentslivinginaspecificlocationordifferencesinbodylengthsorweightsforhumansorecologicalspecies)(CullenandFrey1999;USEPA2011c).Variabilitycanarisefrominherentlyrandomprocesses,suchasvariationsinwindspeedovertimeatagivenlocationorfromtruevariationacrossmembersofapopulationthat,inprinciple,couldbeexplained,butwhich,inpractice,maynotbeexplainableusingcurrentlyavailablemodelsordata(e.g.,therangeofleadlevelsinthebloodofchildren6yearsoldoryoungerfollowingaspecificdegreeofleadexposure).OfparticularinterestinbothHHRAandERAisinter‐individualvariability,whichtypicallyreferstodifferencesbetweenmembersofthesamepopulationineitherbehaviorrelatedtoexposure(e.g.,dietaryconsumptionratesforspecificfooditems),orbiokineticsrelatedtochemicaluptake(e.g.,gastrointestinaluptakeratesforleadfollowingintake)ortoxicresponse(e.g.,differencesamongindividualsorspeciesintheinternaldoseneededtoproduceaspecificamountofneurologicalimpairment).

Inter‐individualvariabilityisillustratedinCaseStudy5intheAppendix,whichassessesaPCB‐contaminatedsedimentsiteintheHudsonRiver.Inthiscasestudy,thequantificationofvariabilityisillustratedthroughtheuseofaPRAtool—1‐DMCA—todescribethevariabilityofexposureasafunctionofindividualexposurefactors(i.e.,youngchildren’sfishingestion).

2.1.2. Uncertainty Uncertaintyisthelackofknowledgeofthetruevalueofaquantityorrelationshipsamongquantities(USEPA2011c).Forexample,theremaybealackofinformationregardingthetruedistributionofvariabilitybetweenindividualsforconsumptionofcertainfooditems.Thereareanumberoftypesofuncertaintiesforbothriskanalysis.Thefollowingdescriptionsofthetypesofuncertainty(adaptedfromCullenandFrey1999)addressesuncertaintiesthatariseduringriskanalyses.Theseuncertaintiescanbeseparatedbroadlyintothreecategories:(1)scenariouncertainty;(2)modeluncertainty;and(3)inputorparameteruncertainty.Eachoftheseisexplainedintheparagraphsthatfollow.

Scenariouncertaintyreferstoerrors,typicallyofomission,resultingfromincorrectorincompletespecificationoftheriskscenariotobeevaluated.Theriskscenarioreferstoasetofassumptionsregardingthesituationtobeevaluated,suchas:(1)thespecificsourcesofchemicalemissionsorexposuretobeevaluated(e.g.,oneindustrialfacilityoraclusterofvariedfacilitiesimpactingthesamestudyarea);(2)thespecificreceptorpopulationsandassociatedexposurepathwaystobemodeled(e.g.,indoorinhalationexposure,track‐industorconsumptionofhome‐produceddietaryitems);and(3)activitiesbydifferentlifestagestobeconsidered(e.g.,exposureonlyathome,orconsiderationofworkplaceorcommutingexposure).Mis‐specificationoftheriskscenariocanresultinunderestimation,overestimationorothermischaracterizationofrisks.Underestimationmayoccurbecauseoftheexclusionofrelevantsituationsortheinclusionofirrelevantsituationswithrespecttoaparticularanalysis.Overestimationmayoccurbecauseoftheinclusionof

Page 27: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

16

unrealisticorirrelevantsituations(e.g.,assumingcontinuousexposuretoanintermittentairbornecontaminantsourceratherthanaccountingformobilitythroughouttheday).

Modeluncertaintyreferstolimitationsinthemathematicalmodelsortechniquesthataredevelopedtorepresentthesystemofinterestandoftenstemsfrom:(1)simplifyingassumptions;(2)exclusionofrelevantprocesses;(3)mis‐specificationofmodelboundaryconditions(e.g.,therangeofinputparameters);or(4)misapplicationofamodeldevelopedforotherpurposes.Modeluncertaintytypicallyariseswhentheriskmodelreliesonmissingorimproperlyformulatedprocesses,structuresorequations.SourcesofmodeluncertaintyaredefinedintheGlossary.

Inputorparameteruncertaintytypicallyreferstoerrorsincharacterizingtheempiricalvaluesusedasinputstothemodel(e.g.,engineering,physical,chemical,biologicalorbehavioralvariables).Inputuncertaintycanoriginatefromrandomorsystematicerrorsinvolvedinmeasuringaspecificphenomenon(e.g.,biomarkermeasurements,suchastheconcentrationofmercuryinhumanhair);statisticalsamplingerrorsassociatedwithsmallsamplesizes(e.g.,ifthedataarebasedonsamplesselectedwitharandom,representativesamplingdesign);theuseofsurrogatedatainsteadofdirectlymeasureddata;theabsenceofanempiricalbasisforcharacterizinganinput(e.g.,theabsenceofmeasurementsforfugitiveemissionsfromanindustrialfacility);ortheuseofsummarymeasuresofcentraltendencyratherthanindividualobservations.Nonlinearrandomprocessescanexhibitabehaviorthat,forsmallchangesininputvalues,producesalargevariationinresults.

InputorparameteruncertaintyisillustratedinCaseStudy3intheAppendixtitled“ProbabilisticAssessmentofAnglingDurationUsedintheAssessmentofExposuretoHudsonRiverSedimentsviaConsumptionofContaminatedFish.”Inthiscasestudy,aprobabilisticanalysisofoneparameterinanexposureassessment—thetimeanindividualspendsfishinginalargeriversystem—wasassessedusingsensitivityanalysis.Thisanalysiswasconductedbecausetherewasuncertaintythattheindividualexposuredurationbasedonresidencedurationmayunderestimatethetimespentfishing(i.e.,anglingduration).Thefulldistributionofthecalculatedvalueswasusedinconductingthe1‐DMCAforthefishconsumptionpathway,whichispresentedinCaseStudy5.

Decisionuncertaintyreferstoadecisionanalysisthatwouldincludenotonlytheimpactofscenario,modelandinputuncertaintiesontherelativeattractivenessofpotentialdecisionalternatives,butalsowouldincludethedegreetowhichspecificchoices(suchasselectinginputdata,models,andscenarios,andevenhowtheproblemordecisionanalysisisframed)impacttherelativeattractivenessofpotentialdecisionalternatives.Indecisionmaking,analystsusedatatorepresentdecisioncriteriathatdecisionmakersandotherstakeholdersbelievewillhelpthemtoanswertheirdecisionquestion(s).ThesequestionsmightincludewhichpolicyalternativebestmeetsAgencygoals(thatmustbearticulated)orwhichriskassessmentscenariobestdescribestheobservedeffects.Data,modelandscenariouncertaintieswillinfluencetheriskassessmentresultsandthose,inturn,willinfluencetheriskmanagementoptions.Decisionmakerswhounderstandtheuncertaintyassociatedwiththeirspecificchoicescanbemoreconfidentthatthedecisionwillproducetheresultsthattheyseek.Inaddition,thesedecisionmakerswillbeabletodefendtheirdecisionsbetterandexplainhowthedecisionmeetsAgencyandstakeholdergoals.

Whilethisisbeyondthescopeofthisdocument,StahlandCimorelli(2005and2012)illustratehowuncertaintythroughoutthedecisionmakingprocesscanbeassessed.Thesecasestudiesexploredtheassessmentofozonemonitoringnetworksandairqualitymanagementpoliciesthatseektominimizetheadverseimpactsfromozone,fineparticulatematterandairtoxicssimultaneously.Thesecasestudiesdemonstratetheimportanceandfeasibilityofbetterunderstandingtheuncertaintyintroducedbyspecificchoices(e.g.,selectinginputdata,models,andscenarios)whenmakingpublicpolicydecisions.

Page 28: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

17

2.2. When Is Probabilistic Risk Analysis Applicable or Useful? PRAmaybeparticularlyuseful,forexample,inthefollowing(Cooke1991;CullenandFrey1999;NRC2009;USEPA2001):

Whenascreening‐levelDRAindicatesthatrisksarepossiblyhigherthanalevelofconcernandamorerefinedassessmentisneeded.

Whentheconsequencesofusingpointestimatesofriskareunacceptablyhigh.

Whensignificantequityorenvironmentaljusticeissuesareraisedbyinter‐individualvariability.

Toestimatethevalueofcollectingadditionalinformationtoreduceuncertainty.

Toidentifypromisingcriticalcontrolpointsandlevelswhenevaluatingmanagementoptions.

Torankexposurepathways,sites,contaminantsandsoonforthepurposesofprioritizingmodeldevelopmentorfurtherresearch.

Whencombiningexpertjudgmentsonthesignificanceofthedata.

Whenexploringtheimpactoftheprobabilitydistributionsofstakeholderanddecision‐makervaluesontheattractivenessofpotentialdecisionalternatives(Fischhoff1995;Illing1999;KunreutherandSlovic1996;USEPA2000b).

Whenexploringtheimpactoftheprobabilitydistributionsofthedata,modelandscenariouncertainties,andvariabilitytogethertocomparepotentialdecisionalternatives.

PRAmayaddminimalvaluetotheassessmentinthefollowingtypesofsituations(CullenandFrey1999;USEPA1997a):

Whenascreening‐leveldeterministicriskassessmentindicatesthatrisksarenegligible,presumingthattheassessmentisknowntobeconservativeenoughtoproduceoverestimatesofrisk.

Whenthecostofavertingtheexposureandriskissmallerthanthecostofaprobabilisticanalysis.

Whenthereislittleuncertaintyorvariabilityintheanalysis(thisisararesituation).

2.3. How Can Probabilistic Risk Analysis Be Incorporated Into Assessments?

AsillustratedintheaccompanyingcasestudiesintheAppendix,probabilisticapproachescanbeincorporatedintoanystageofariskassessment,fromproblemformulationorplanningandscopingtotheanalysisofalternativedecisions.Insomesituations,PRAcanbeusedselectivelyforcertaincomponentsofanassessment.Itiscommoninassessmentsthatsomemodelinputsareknownwithhighconfidence(i.e.,basedonsite‐specificmeasurements),whereasvaluesforotherinputsarelesscertain(i.e.,basedonsurrogatedatacollectedforadifferentpurpose).Forexample,anexposuremodelermaydeterminethatrelevantairqualitymonitoringdataexists,butthereisalackofdetailedinformationonhumanactivitypatternsindifferentmicroenvironments.Thus,anassessmentofthevariabilityinexposuretoairbornepollutantsmightbebasedondirectuseofthemonitoringdata,whereasassessmentofuncertaintyandvariabilityintheinhalationexposurecomponentmightbebasedonstatisticalanalysisofsurrogatedataoruseofexpertjudgment.Theuncertaintiesarelikelytobelargerforthelatterthantheformercomponentoftheassessment;

Page 29: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

18

effortstocharacterizeuncertaintiesassociatedwithpollutantexposureswouldfocusonthelatter.PRAalsodealswithdependencyissues;adescriptionoftheseissuesisavailableinSection3.3.2.

2.4. What Are the Scientific Community’s Views on Probabilistic Risk Analysis, and What Is the Institutional Support for Its Use in Performing Assessments?

TheNRCandIOMrecentlyemphasizedtheirlong‐standingadvocacyforPRA(NRC2007aandb;IOM2013).Datingfromits1983RiskAssessmentintheFederalGovernment:ManagingtheProcess(NRC1983)—whichfirstformalizedtheriskassessmentparadigm—throughreportsreleasedfromthelate1980sthroughtheearly2000s,variousNRCpanelshavemaintainedconsistentlythatbecauseriskanalysisinvolvessubstantialuncertainties,theseuncertaintiesshouldbeevaluatedwithinariskassessment.Thesepanelsnotedthat:

1. Whenevaluatingthetotalpopulationrisk,EPAshouldconsiderthedistributionofexposureandsensitivityofresponseinthepopulation(NRC1989).

2. Whenassessinghumanexposuretoairpollutants,EPAshouldpresentmodelresultsalongwithestimateduncertainties(NRC1991).

3. WhenconductingERA,EPAshoulddiscussthoroughlyuncertaintyandvariabilitywithintheassessment(NRC1993).

4. “Uncertaintyanalysisistheonlywaytocombatthe‘falsesenseofcertainty,’whichiscausedbyarefusaltoacknowledgeand[attemptto]quantifytheuncertaintyinriskpredictions,”asstatedintheNRCreport,ScienceandJudgmentinRiskAssessment(NRC1994).

5. EPA’sestimationofhealthbenefitswasnotwhollycrediblebecauseEPAfailedtodealformallywithuncertaintiesinitsanalyses(NRC2002).

6. EPAshouldadopta“tiered”approachforselectingthelevelofdetailusedinuncertaintyandvariabilityassessment.Furthermore,theNRCrecommendedthatadiscussionofthelevelofdetailusedforuncertaintyanalysisandvariabilityassessmentshouldbeanexplicitpartoftheplanning,scopingandproblemformulationphaseoftheriskassessmentprocess(NRC2009).

7. EPAshoulddevelopmethodstosystematicallydescribeandaccountforuncertaintiesindecision‐relevantfactorsinadditiontoestimatesofhealthriskinitsdecision‐makingprocess(IOM2013).

AskedtorecommendimprovementstotheAgency’sHHRApractices,EPA’sSABechoedtheNRC’ssentimentsandurgedtheAgencytocharacterizeuncertaintyandvariabilitymorefullyandsystematicallyandtoreplacesingle‐pointuncertaintyfactorswithasetofdistributionsusingprobabilisticmethods(ParkinandMorgan2007).ThekeyprinciplesofriskassessmentcitedbytheOfficeofScienceandTechnologyPolicy(OSTP)andtheOfficeofManagementandBudget(OMB)include“explicit”characterizationoftheuncertaintiesinriskjudgments;theyproceedtocitetheNationalAcademyofScience’s(NAS)2007recommendationtoaddressthe“variabilityofeffectsacrosspotentiallyaffectedpopulations”(OSTP/OMB2007).

Page 30: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

19

2.5. Additional Advantages of Using Probabilistic Risk Analysis and How It Can Provide More Comprehensive, Rigorous Scientific Information in Support of Regulatory Decisions.

ExternalstakeholderspreviouslyhaveusedtheAdministrativeProcedureActandtheDataQualityActtochallengetheAgencyforalackoftransparencyandconsistencyorfornotfullyanalyzingandcharacterizingtheuncertaintiesinriskassessmentsordecisions(Fisheretal.2006).ThemorecompleteimplementationofPRAandrelatedapproachestodealwithuncertaintiesindecisionmakingwouldaddressstakeholderconcernsinregardtocharacterizinguncertainties.

Theresultsofanyassessment,includingPRA,aredependentontheunderlyingmethodsandassumptions.Accompaniedbytheappropriatedocumentation,PRAmaycommunicateamorerobustrepresentationofrisksandcorrespondinguncertainties.Thischaracterizationmaybeintheformofarangeofpossibleestimatesasopposedtothemoretraditionallypresentedsingle‐pointvalues.Dependingontheuseoftheassessment,rangescanbederivedforvariabilityanduncertainty(oracombinationofthetwo)inbothmodelinputsandresultingestimationsofrisk.

PRAquantifieshowexposures,effectsandrisksdifferamonghumanpopulationsorlifestagesortargetecologicalorganisms.PRAalsoprovidesanestimationofthedegreeofconfidencewithwhichtheseestimatesmaybemade,giventhecurrentuncertaintyinscientificknowledgeandavailabledata.A2007NRCpanelstatedthattheobjectiveofPRAsisnottodecide“howmuchevidenceissufficient”toadoptanalternativebut,rather,todescribethescientificbasesofproposedalternativessothatscientificandpolicyconsiderationsmaybemorefullyevaluated(NRC2007a).EPA’sSABsimilarlynotedthatPRAsprovidemore“valueofinformation”throughaquantitativeassessmentofuncertaintyandclarifythescienceunderlyingAgencydecisions(USEPA2007b).

TheSABarticulatedanumberofadvantagesforEPAdecisionmakersfromtheutilizationofprobabilisticmethods(ParkinandMorgan2007):

AprobabilisticreferencedosecouldhelpreducethepotentiallyinaccurateimplicationofzeroriskbelowtheRfD.

Byunderstandingandexplicitlyaccountingforuncertaintiesunderlyingadecision,EPAcanestimateformallythevalueofgatheringmoreinformation.Bydoingso,theAgencycanbetterprioritizeitsinformationneedsbyinvestinginareasthatyieldthegreatestinformationvalue.

StrategicuseofPRAwouldallowEPAtosendtheappropriatesignaltotheintellectualmarketplace,therebyencouraginganalyststogatherdataanddevelopmethodologiesnecessaryforassessinguncertainties.

2.6. What Are the Challenges to Implementation of Probabilistic Analyses?

Currently,EPAisusingPRAinavarietyofprogramstosupportdecisions,butchallengesremainregardingtheexpandeduseofthesetoolswithintheAgency.Thechallengesinclude:

AlackofunderstandingofthevalueofPRAfordecisionmaking.PRAhelpstoimprovetherigorofthedecision‐makingprocessbyallowingdecisionmakerstoexploretheimpactsofuncertaintyandvariabilityonthedecisionchoices.

Aclearinstitutionalunderstandingofhowtoincorporatetheresultsofprobabilisticanalysesintodecisionmakingislacking.

Page 31: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

20

PRAtypicallyrequiresadifferentskillsetthanusedincurrentevaluations,andlimitedresources(staff,time,trainingormethods)toconductPRAareavailable.

Communicatingprobabilisticanalysisresultsandtheimpactofthoseresultsonthedecision/policyoptionscanbecomplex.

Communicationwithstakeholdersisoftendifficultandresultsintheappearanceofregulatorydelaysduethenecessityofanalyzingnumerousscenariosusingvariousmodels.

PRAcomplicatesdecisionmakingandriskcommunicationininstanceswhereamorecomprehensivecharacterizationoftheuncertaintiesleadstoadecreaseinclarityregardinghowtoestimateriskforthescenariounderconsideration.ThesechallengesarediscussedinmoredetailinSections2.7through2.13.

2.7. How Can Probabilistic Risk Analysis Support Specific Regulatory Decision Making?

Decisionmakerssometimesperceivethatthebinarynatureofregulatorydecisions(e.g.,Doesanexposureexceedareferencedoseornot?DoemissionscomplywithAgencystandardsornot?)precludestheuseofariskrangedevelopedthroughPRA.Generally,itisnecessarytoexplaintherationaleunderlyingaparticulardecision.PRA’sprimarypurposeistoprovideinformationtoenhancetheabilitytomaketransparentdecisionsbasedonthebestavailablescience.Byconductingasensitivityanalysisoftheinfluenceoftheuncertaintyonthedecision‐makingprocess,itcanbedeterminedhoworifPRAcanhelptoimprovetheprocess.

PRAcanprovideinformationtodecisionmakersonspecificquestionsrelatedtouncertaintyandvariability.Forquestionsofuncertaintyandtominimizethelikelihoodofunintendedconsequences,PRAcanhelptoprovidethefollowingtypesofinformation:

Characterizationoftheuncertaintyinestimates(i.e.,Whatisthedegreeofconfidenceintheestimate?).Couldthepredictionbeoffbyafactorof2,afactorof10orafactorof1,000?

Criticalparametersandassumptionsthatmostaffectorinfluenceadecisionandtheriskassessment.

“Tippingpoints”wherethedecisionwouldbealterediftheriskestimatesweredifferent,orifadifferentassumptionwasvalid.

Estimatethelikelihoodthatvaluesforcriticalparameterswilloccurortestthevalidityofassumptions.

Estimatethedegreeofconfidenceinaparticulardecisionand/orthelikelihoodofspecificdecisionerrors.

Thepossibilityofalternativeoutcomeswithadditionalinformation,orestimatetradeoffsrelatedtodifferentrisksorrisk‐managementdecisions.

Theimpactofadditionalinformationondecisionmaking,consideringthecostandtimetoobtaintheinformationandtheresultingchangeindecision(i.e.,thevalueoftheinformation).

Fortheconsiderationofvariability,PRAcanhelptoprovidethefollowingtypesofinformationforexposures:

Explicitlydefinedexposuresforvariouspopulationsorlifestages(i.e.,Whoarewetryingtoprotect?).Thatis,willtheregulatoryactionkeep50percent,90percent,99.9percentorsomeotherfractionofthepopulationbelowaspecifiedexposure,doseorrisktarget?

Page 32: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

21

Variabilityintheexposures,amongvariouspopulationsorlifestages,andinformationonthepercentileofthepopulationthatisbeingevaluatedintheriskassessment(e.g.,variationsinthenumberoflitersofwaterperkilogram[kg]bodyweightperdayconsumedbythepopulation).Thisinformationishelpfulinaddressingcomments:

OntheconservatismofEPA’sriskassessments;

Concernsaboutwhethertheirparticularexposureswereevaluatedintheriskassessment;

Whomorwhatisbeingprotectedbyimplementingadecision;and

Whetherandwhatadditionalresearchmaybeneededtoreduceuncertainty.

PRAhelpstoinformdecisionsbycharacterizingthealternativesavailabletothedecisionmakerandtheuncertaintyheorshefaces,andbyprovidingevaluationmeasuresofoutcomes.Uncertaintiesoftenarerepresentedasprobabilitiesorprobabilitydistributionsnumericallyoringraphs.Aspartofadecisionanalysis,stakeholderscanmorefullyexaminehowuncertaintiesinfluencethepreferenceamongalternatives.

2.8. Does Probabilistic Risk Analysis Require More Resources Than Default-Based Deterministic Approaches?

PRAgenerallycanbeexpectedtorequiremoreresourcesthanstandardAgencydefault‐baseddeterministicapproaches.ThereisextensiveexperiencewithinEPAinconductingandreviewingDRA.Theseassessmentstendtofollowstandardizedmethodsthatminimizetheeffortrequiredtoconductthemandtocommunicatetheresults.Probabilisticassessmentsoftenentailamoredetailedanalysis,andasaresult,theseassessmentsrequiresubstantiallymoreresources,includingtimeandeffort,thandodeterministicapproaches.

Appropriatelytrainedstaffandtheavailabilityofadequatetools,methodsandguidanceareessentialfortheapplicationofPRA.Properapplicationofprobabilisticmethodsrequiresnotonlysoftwareanddata,butalsoguidanceandtrainingforanalystsusingthetoolsandformanagersanddecisionmakerstaskedwithinterpretingandcommunicatingtheresults.

Anupfrontincreaseinresourcesneededtoconductaprobabilisticassessmentcanbeexpected,butdevelopmentofstandardizedapproachesand/ormethodscanleadtotheroutineincorporationofPRAinAgencyapproaches(e.g.,OPP’suseoftheDietaryExposureEvaluationModel[DEEM;http://www.epa.gov/pesticides/science/deem/],aprobabilisticdietaryexposuremodel).Theinitialand,insomecases,ongoingresourcecost(e.g.,fordevelopmentofsite‐specificmodelsforsiteassessments)maybeoffsetbyamoreinformeddecisionthanacomparabledeterministicanalysis.Probabilisticmethodsareusefulforidentifyingeffectivemanagementoptionsandprioritizingadditionaldatacollectionorresearchaimedatimprovingriskestimation,ultimatelyresultingindecisionsthatenableimprovedenvironmentalprotectionwhilesimultaneouslyconservingmoreresources.

2.9. Does Probabilistic Risk Analysis Require More Data Than Conventional Approaches?

TherearedifferencesofopinionwithinthetechnicalcommunityastowhetherPRArequiresmoredatathanothertypesofanalyses.AlthoughsomeemphaticallybelievethatPRArequiresmoredata,othersarguethatprobabilisticassessmentsmakebetteruseofalloftheavailabledataandinformation.StahlandCimorelli(2005)discusswhenandhowmuchdataarenecessaryforadecision.PRAcanbenefitfrommoredatathanmightbeusedinaDRA.Forexample,whereDRA

Page 33: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

22

mightemployselectedpointestimates(e.g.,themeanor95thpercentilevalues)fromavailabledatasetsforuseinmodelinputs,PRAfacilitatestheuseoffrequency‐weighteddatadistributions,allowingforamorecomprehensiveconsiderationoftheavailabledata.Inmanycases,thedatathatwereusedtodevelopthepresumptive95thpercentilecanbeemployedinthedevelopmentofprobabilisticdistributions.

RestrictionofPRAtoprincipallydata‐richsituationsmaypreventitsbroaderapplicationwhereitismostuseful.BecausePRAincorporatesinformationondataquality,variabilityanduncertaintyintoriskmodels,theinfluenceofthesefactorsonthecharacterizationofriskcanbecomeagreaterfocusofdiscussionanddebate.

AkeybenefitofusingPRAisitsabilitytorevealthelimitationsaswellasthestrengthsofdatathatoftenaremaskedbyadeterministicapproach.Indoingso,PRAcanhelptoinformresearchagendas,aswellassupportregulatorydecisionmaking,basedonthestateofthebestavailablescience.Insummary,PRAtypicallyrequiresmoretimefordevelopinginputassumptionsthanaDRA,butwhenincorporatedintotherelevantstepsoftheriskassessmentprocess,PRAcandemonstrateaddedbenefits.Insomecases,PRAcanprovideadditionalinterpretationsthatcompensatefortheextraeffortrequiredtoconductaPRA.

2.10. Can Probabilistic Risk Analysis Be Used to Screen Risks or Only in Complex or Refined Assessments?

Probabilisticmethodstypicallyarenotnecessarywheretraditionaldefault‐baseddeterministicmethodsareadequateforscreeningrisks.Suchmethodsarerelativelylowcost,intendedtoproduceconservativelybiasedestimates,andusefulforidentifyingsituationsinwhichrisksaresolowthatnofurtheractionisneeded.Theapplicationofprobabilisticmethodscanbetargetedtosituationsinwhichascreeningapproachindicatesthatariskmaybeofconcernorwhenthecostofmanagingtheriskishigh,creatinganeedforinformationtohelpinformdecisionmaking.PRAfitsdirectlyintoagraduatedhierarchicalapproachtoriskanalysis.Thistieredapproach,depictedinFigure2,isaprocessforasystematicinformedprogressiontoincreasinglymorecomplexriskassessmentmethods,dependingonthedecision‐makingcontextandneed.Highertiersreflectincreasingcomplexityandoftenwillrequiremoretimeandresources.Ananalysismighttypicallystartatalowertierandonlyprogresstoahighertierifthereisaneedforamoresophisticatedassessmentcommensuratewiththeimportanceoftheproblem.Highertiersalsoreflectincreasingcharacterizationofvariabilityand/oruncertaintyintheriskestimate,whichmaybeimportantforrisk‐managementdecisions.ThecasestudiesdescribedintheAppendixarepresentedinthreegroupsthatgenerallycorrespondtothetiersidentifiedinFigure2.Group1casestudiesarepointestimate(sensitivityanalysis)examples(Tier1);Group2casestudiesincludemostmoderate‐complexityPRAexamples(Tier2);andGroup3casestudiesareadvanced(highcomplexity)PRAexamples(Tier3).

ThetieredapproachinFigure2depictsacontinuumfromscreeninglevelpointestimatethatisdonewithlittledataandconservativeassumptionstoPRAthatrequiresanextensivedatasetandmorerealistic(lessconservative)assumptions.Inbetween,therecanbeawidevarietyoftiersofincreasingcomplexity,ortheremaybeonlyafewreasonablechoicesbetweenscreeningmethodsandhighlyrefinedanalyses(USEPA2004a).Asimilarfour‐tieredapproachforcharacterizingthevariabilityand/oruncertaintyintheestimatedexposureorriskanalysis(WHO2008)hasbeenadaptedbyEPAintheriskandexposureassessmentsconductedfortheNationalAmbientAirQualityStandards(NAAQS).

PRAalsocouldbeusedtoexaminemorefullytheexistingdefault‐basedmethodsbasedonthecurrentstateofinformationandknowledgetodetermineifsuchmethodsaretrulyconservative

Page 34: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

23

andadequateforscreening(e.g.,indose‐responseanalysesdealingwithhazardcharacterization)(Swartoutetal.1998;Hattisetal.2002).

Theuseofaspectrumofdatashouldbeemployedbothindeterminingscreeningrisksandinmorecomplexassessments.ForHHRA,datafromhuman,animal,mechanisticandotherstudiesshouldbeusedtodevelopaprobabilisticcharacterizationofcancerandnoncancerrisksandtoidentifyuncertainties.TheNRCrecommendedthatEPAfacilitatethisapproachbyredefiningRfDandRfCwithintheprobabilisticframeworktotakeintoaccounttheprobabilityofharm(NRC2009).ItislikelythatbothDRAandPRAwillbepartofthisframework.

2.11. Does Probabilistic Risk Analysis Present Unique Challenges to Model Evaluation?

Theconceptof“validation”ofmodelsusedforregulatorydecisionmakinghasbeenatopicofintensediscussion.Inarecentreportontheuseofmodelsinenvironmentalregulatorydecisionmaking,theNRCrecommendedusingthenotionofmodel“evaluation”ratherthan“validation,”suggestingthatuseofaprocessthatencompassestheentirelifecycleofthemodelandincorporatesthespectrumofinterestedpartiesintheapplicationofthemodeloftenextendsbeyondthemodelbuilderanddecisionmaker.Suchaprocesscanbedesignedtoensurethatjudgmentofthemodelapplicationisbasednotonlyonitspredictivevaluedeterminedfromcomparisonwithhistoricaldata,butalsoonitscomprehensiveness,rigorindevelopment,transparencyandinterpretability(NRC2007b).

Modelevaluationisimportantinallriskassessments.InthecaseofPRA,thereisanadditionalquestionastothevalidityoftheassumptionsregardingprobabilityandfrequencydistributionsformodelinputsandtheirdependencies.Probabilisticinformationcanbeaccountedforduringevaluationanalysesbyconsideringtherangeofuncertaintyinthemodelpredictionandwhethersucharangeoverlapswiththe“true”valuebasedonindependentdata.Thus,probabilisticinformationcanaidincharacterizingtheprecisionofthemodelpredictionsandwhetherapredictionissignificantlydifferentfromabenchmarkofinterest.Forexample,comparisonsofprobabilisticmodelresultsandmonitoringdatawereperformedformultiplemodelsindevelopingthecumulativepesticideexposuremodel.ConcurrentPRAmodelevaluationsusingaBayesiananalysisalsohavebeenpublished(Clyde2000).

Whenriskassessorsdevelopmodelsofrisk,theyrelyontwopredominantstatisticalmethods.Bothmethodsarisefromaxiomsofprobability,buteachappliestheseaxiomsdifferently.Underthefrequentistapproach,onedevelopsandevaluatesamodelbytestingwhetherthemodel—asappliedtotheobservations—conformstoidealizeddistributions.UndertheBayesianapproach,onedevelopsandevaluatesamodelbytestingwhich—amongalternativemodels—bestyieldstheunderlyingdistributiondescribingthedata.Thepracticaldifferencesbetweenthesetwoapproachescanperhapsbestbeappreciatedwhenconsideringthestructuraluncertaintyinmodels(Section3.3.3).BecauseBayesiansestimatemodelparameterswiththeexpectationthattheseparameters—orevenmodelstructures—willbeupdatedasnewdatabecomeavailable,theyhavedevelopedformaltechniquestoprovideuncertaintyboundsaroundtheseparameterestimates,selectmodelsthatbestexplainthegivendata,orcombinetheresultsofalternativemodels.

2.12. How Do You Communicate the Results of Probabilistic Risk Analysis?

Effectivecommunicationmakesiteasierforregulatorsandstakeholderstounderstandthedecisioncriteriadrivingthedecision‐makingprocess.Inotherwords,communicationofPRAresultswithinthedecision‐makingcontextfacilitatesunderstanding.Thespecificapproachesforreportingresults

Page 35: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

24

fromPRAvarydependingontheassessmentobjectiveandtheintendedaudience.Beyondthebasic1997principlesandthepolicyfromthesameyear(USEPA1997aandb),theRiskAssessmentGuidanceforSuperfund:VolumeIII—PartA,ProcessforConductingProbabilisticRiskAssessmentalsoprovidessomeguidanceonthequalityandcriteriaforacceptanceaswellascommunicationbasics(USEPA2001).TherehavebeenlimitedstudiesofhowinformationfromPRAregardinguncertaintyandvariabilitycanorshouldbecommunicatedtokeyaudiences,suchasdecisionmakersandstakeholders(e.g.,MorganandHenrion1990;Bloometal.1993;Krupnicketal.2006).Amongtheanalystcommunity,thereoftenisaninterestinvisualizationofthestructureofascenarioandmodelusinginfluencediagramsanddepictionoftheuncertaintyandvariabilityinmodelinputsandoutputsusingprobabilitydistributionsintheformofcumulativedensityfunctionsorprobabilitydistributionfunctions(Figure3).Sensitivityofthemodeloutputtouncertaintyandvariabilityinmodelinputscanbedepictedusinggraphicaltools.

Insomecases,thesegraphicalmethodscanbeusefulforthoselessfamiliarwithPRA,butinmanycasesthereisaneedtotranslatethequantitativeresultsintoamessagethatextractsthekeyinsightswithoutburdeningthedecisionmakerwithobscuretechnicaldetails.Inthisregard,theuseofrangesofvaluesforaparticularmetricofdecision‐makingrelevance(e.g.,therangeofuncertaintyassociatedwithaparticularestimateofrisk)maybeadequate.ThepresentationofPRAresultstoadecisionmakermaybeconductedbestasaninteractivediscussion,inwhichaprincipalmessageisconveyed,followedbyexplorationofissuessuchasthesource,qualityanddegreeofconfidenceassociatedwiththeinformation.ThereisaneedforthedevelopmentofrecommendationsandacommunicationplanregardinghowtocommunicatetheresultsofPRAtodecisionmakersandstakeholders,buildingontheexperienceofvariousprogramsandregions.

2.13. Are the Results of Probabilistic Risk Analysis Difficult to Communicate to Decision Makers and Stakeholders?

Researchhasshownthattheabilityofdecisionmakerstodealwithconceptsofprobabilityanduncertaintyvaries.Bloometal.(1993)surveyedagroupofseniormanagersatEPAandfoundthatmanycouldinterpretinformationaboutuncertaintyifitwascommunicatedinamannerresponsivetodecision‐makerinterests,capabilitiesandneeds.Inamorerecentsurveyofex‐EPAofficials,Krupnicketal.(2006)concludedthatmosthaddifficultyunderstandinginformationonuncertaintywithconventionalscientificpresentationapproaches.ThefindingsofthesestudieshighlighttheneedforpracticalstrategiesforthecommunicationofresultsofPRAanduncertaintyinformationbetweenriskanalystsanddecisionmakers,aswellasbetweendecisionmakersandotherstakeholders.TheOfficeofEmergencyandRemedialResponse(OERR)hascompiledguidancetoassistanalystsandmanagersinunderstandingandcommunicatingtheresultsofPRA(USEPA2001).

Riskanalystsneedtofocusonhowtouseuncertaintyanalysistocharacterizehowconfidentdecisionmakersshouldbeintheirchoices.AsWilson(2000)explained,“…uncertaintyisthebaneofanydecisionmaker’sexistence.Thus,anyonewhowantstoinformdecisionsusingscientificinformationneedstoassurethattheiranalysestransformuncertaintyintoconfidenceinconclusions.”Hence,althoughenvironmentalriskassessmentsarecomplicatedanditiseasytogetlostinthedetails,presentinganddiscussingtheseresultswithinthecontextofthedecisionfacilitatesunderstanding.Thetranslationofuncertaintyintoconfidencestatementsforcesa“top‐down”perspectivethatpromotesaccountingforwhetherandhowuncertaintiesaffectchoices(Tolletal.1997).

Page 36: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

25

 Figure 3. Graphical Description of the Likelihood (Probability) of Risk. Hypothetical fitted data distribution with upper and lower confidence intervals are depicted for the output of a 2‐D MCA model. 

Example of 2-D MCA Output Graphical

Example 2-D MCA Output

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.0E-06 1.0E-05 1.0E-04 1.0E-03

Risk

Cum

ulat

ive P

roba

bility

Lower confidence bound

Best Estimate

Upper confidence bound

Page 37: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

26

3. AN OVERVIEW OF SOME OF THE TECHNIQUES USED IN PROBABILISTIC RISK ANALYSIS

3.1. What Is the General Conceptual Approach in Probabilistic Risk Analysis?

PRAincludesseveralmajorsteps,whichparalleltheacceptedenvironmentalhealthriskassessmentprocess.Theseinclude:(1)problemand/ordecisioncriteriaidentification;(2)gatheringinformation;(3)interpretingtheinformation;(4)selectingandapplyingmodelsandmethodsforquantifyingvariabilityand/oruncertainty;(5)quantifyinginter‐individualorpopulationuncertaintyandvariabilityinmetricsrelevanttodecisionmaking;(6)sensitivityanalysistoidentifykeysourcesofvariabilityanduncertainty;and(7)interpretingandreportingresults.

Problemformulationentailsidentifyingtheassessmentendpointsorissuesthatarerelevanttothedecision‐makingprocessandstakeholders,andthatcanbeaddressedinascientificassessmentprocess.Followingproblemformulation,informationisneededfromstakeholdersandexpertsregardingthescenariostoevaluate.Basedonthescenariosandassessmentendpoints,theanalystsselectordevelopmodels,whichinturnleadstoidentificationofmodelinputdatarequirementsandacquisitionofdataorotherinformation(e.g.,expertjudgmentencodedastheresultofaformalelicitationprocess)thatcanbeusedtoquantifyinputstothemodels.Thedataorotherinformationformodelinputsisinterpretedintheprocessofdevelopingprobabilitydistributionstorepresentvariability,uncertaintyorbothforaparticularinput.Thus,steps(1)through(4)listedabovearehighlyinteractiveanditerativeinthatthedatainputrequirementsandhowinformationistobeinterpreteddependonthemodelformulation,whichdependsonthescenarioandthatinturndependsontheassessmentobjective.Theassessmentobjectivemayhavetoberefineddependingontheavailabilityofinformation.

Onceascenario,modelandinputsarespecified,themodeloutputisestimated.AcommonapproachistouseMonteCarloAnalysis(MCA)orotherprobabilisticmethodstogeneratesamplesfromtheprobabilitydistributionsofeachmodelinput,runthemodelbasedononerandomvaluefromeachprobabilisticinput,andproduceonecorrespondingestimateofthemodeloutputs.Thisprocessisrepeatedtypicallyhundredsorthousandsoftimestocreateasyntheticstatisticalsampleofmodeloutputs.Theseoutputdataareinterpretedasaprobabilitydistributionoftheoutputofinterest.Sensitivityanalysiscanbeperformedtodeterminewhichmodelinputdistributionsaremosthighlyassociatedwiththerangeofvariationinthemodeloutputs.Theresultsmaybereportedinawidevarietyofformsdependingontheintendedaudience,rangingfromqualitativesummariestotables,graphsanddiagrams.

DetailedintroductionstoPRAmethodologyareavailableelsewhere,suchasAngandTang(1984),CullenandFrey(1999),EPA(2001),andMorganandHenrion(1990).AfewkeyaspectsofPRAmethodologyarebrieflymentionedhere.ReaderswhoseekmoredetailshouldconsultthesereferencesandseetheBibliographyforadditionalreferences.

3.2. What Levels and Types of Probabilistic Risk Analyses Are There and How Are They Used?

Therearemultiplelevelsandtypesofanalysisusedtoconductriskassessments(illustratedinFigure2andTable1,respectively).Graduatedapproachestoanalysisarewidelyrecognized(e.g.,USEPA1997a,2001;WHO2008).Theideaofagraduatedapproachistochoosealevelofdetailandrefinementforananalysisthatisappropriatetotheassessmentobjective,dataquality,informationavailableandimportanceofthedecision(e.g.,resourceimplications).

Page 38: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

27

Asdiscussedinsection1.8,thereisavarietyofapproachestoriskassessmentthatdifferintheircomplexityandthemannerinwhichtheyaddressuncertaintyandvariability.InDRAonedoesnotformallycharacterizeuncertaintyorvariabilitybutrathertypicallyreliesonusingdefault‐basedassumptionsandfactorstogenerateasingleestimateofrisk.InPRAthereisavarietyofapproachestoexplicitlyaddressorcharacterizeuncertaintyorvariabilityinriskestimatesandthesedifferintermsofhowtheyaccomplishthis,thedataused,andtheoverallcomplexity.Someexamplesare:

Sensitivityanalysis

MonteCarloanalysisofvariabilityinexposuredata

Humanhealthorecologicaleffectsdata

MonteCarloanalysisofuncertainty

“Cumulative”PRAmulti‐pathwayormulti‐chemical

Two‐dimensionalPRAofuncertaintyandvariability

Decisionuncertaintyanalysis

Geospatialanalysis

Expertelicitation

TheDRAapproachesdescribedinSection1.8areexamplesoflowerlevelsinagraduatedapproachtoanalysis.Riskatthelowerlevelsofanalysisisassessedbyconservative,boundingassumptions.Iftheriskestimateisfoundtobeverylowdespitetheuseofconservativeassumptions,thenthereexistsagreatdealofcertaintythattheactualriskstothepopulationofinterestforthegivenscenarioarebelowthelevelofconcernandnofurtherinterventionisrequired,assumingthatthescenarioandmodelspecificationsarecorrect.WhenaconservativeDRAindicatesthatariskmaybehigh,itispossiblethattheriskestimateisbiasedandtheactualriskmaybelower.Insuchasituation,dependingontheresourceimplicationsofthedecision,itmaybeappropriatetoproceedwithamorerefinedorhigherlevelofanalysis.Therelativecostsofinterventionversusfurtheranalysisshouldbeconsideredwhendecidingwhethertoproceedwithadecisionbasedonalowerlevelanalysisortoescalatetoahigherlevelofanalysis.Insomedeterministicassessments(e.g.,ecologicalrisks),theassumptionsarenotwellassuredofconservatism,andtheestimatedrisksmightbebiasedtoappearlowerthantheunseenactualrisk.

AmorerefinedanalysiscouldinvolvetheapplicationofDRAmethods,butwithalternativesetsofassumptionsintendedtocharacterizecentraltendencyandreasonableupperboundsofexposure,effectsandriskestimates,suchthattheestimatescouldbeforanactualindividualinthepopulationofinterestratherthanahypotheticalmaximallyexposedindividual.Suchanalysesarenotlikelytoprovidequantificationregardingtheproportionofthepopulationatorbelowaparticularexposureorrisklevelofconcern,uncertaintiesforanygivenpercentileoftheexposedpopulation,orprioritiesamonginputassumptionswithrespecttotheircontributionstouncertaintyandvariabilityintheestimates.

Tomorefullyanswerthequestionsoftenaskedbydecisionmakers,theanalysiscanbefurtherrefinedbyincorporatingquantitativecomparisonsofalternativemodelingstrategies(torepresentstructuraluncertaintiesassociatedwithscenariosormodels),quantifyingrangesofuncertaintyandvariabilityinmodeloutputs,andprovidingthecorrespondingrangesformodeloutputsofinterest.Whenperformingprobabilisticanalyses,choicesaremaderegardingwhethertofocusonthequantificationofvariabilityonly,uncertaintyonly,bothvariabilityanduncertaintytogether(representingarandomlyselectedindividual),orvariabilityanduncertaintyindependently(e.g.,in

Page 39: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

28

atwo‐dimensionaldepictionofprobabilitybandsforestimatesofinter‐individualvariability;seeFigure4).Thesimultaneousbutdistinctpropagationofuncertaintyandvariabilityinatwo‐dimensionalframeworkenablesquantificationofuncertaintyintheriskforanypercentileofthepopulation.Forexample,onecouldestimatetherangeofuncertaintyintheriskfacedbythemedianmemberofthepopulationorthe95thpercentilememberofthepopulation.Suchinformationcanbeusedbyadecisionmakertogaugetheconfidencethatshouldbeplacedinanyparticularestimateofrisk,aswellastodeterminewhetheradditionaldatacollectionorinformationmightbeusefultoreducetheuncertaintyintheestimates.TheOPPassessmentofChromatedCopperArsenate‐treatedwoodusedsuchanapproach.(SeeCaseStudy9intheAppendix.)

Figure 4. Diagrammatic Comparison of Three Alternative Probabilistic Approaches for the Same Exposure Assessment. In Option 1 (one dimensional Monte Carlo analysis), only variability is quantified. In Option 2 (one dimensional Monte Carlo analysis), both uncertainty and variability are combined. In Option 3 (two dimensional Monte Carlo analysis), variability and uncertainty are analyzed separately. Source: WHO 2008. 

Whenconductingananalysisforthefirsttime,itmaynotbeknownorclear,priortoanalysis,whichcomponentsofthemodelorwhichmodelinputscontributethemosttotheestimatedriskoritsuncertaintyandvariability.Asaresultofcompletingananalysis,however,theanalystoftengainsinsightintothestrengthsandweaknessesofthemodelsandinputinformation.Probabilisticanalysisandsensitivityanalysiscanbeusedtogethertoidentifythekeysourcesofquantifieduncertaintyinthemodeloutputstoinformdecisionsregardingprioritiesforadditionaldatacollection.Ideally,timeshouldbeallowedforcollectingsuchinformationandrefiningtheanalysistoarriveatamorerepresentativeandrobustestimateofuncertaintyandvariabilityinrisk.Thus,thenotionofiterationindevelopingandimprovingananalysisiswidelyrecommended.

0%

20%

40%

60%

80%

100%

% o

f spe

cies

100 1000 104 105

95% Conf. IntervalMedianEC50

2D

MLE

1DPerc

entil

e co

nsum

er

Exposure (g/day)

1 10 100 1000

Option 1. Only variability

Option 3. Separating variability & uncertainty (2D MC)

Option 2. Combining variability & uncertainty (1D MC)

0%

20%

40%

60%

80%

100%

% o

f spe

cies

100 1000 104 105

95% Conf. IntervalMedianEC50

2D

MLE

1DPerc

entil

e co

nsum

er

Exposure (g/day)

1 10 100 1000

Option 1. Only variability

Option 3. Separating variability & uncertainty (2D MC)

Option 2. Combining variability & uncertainty (1D MC)

Page 40: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

29

Thenotionofiterationcanbeappliedbroadlytotheriskassessmentframework.Forexample,afirstefforttoperformananalysismayleadtoinsightthattheassessmentquestionsmightbeimpossibletoaddress,orthatthereareadditionalassessmentquestionsthatmaybeequallyormoreimportant.Thus,iterationcanincludereconsiderationoftheinitialassessmentquestionsandthecorrespondingimplicationsfordefinitionofscenarios,selectionofmodelsandprioritiesforobtainingdataformodelinputs.Alternatively,inatime‐limiteddecisionenvironment,probabilisticandsensitivityanalysesmayofferinsightintotheeffectofmanagementoptionsonriskestimates.

3.3. What Are Some Specific Aspects of and Issues Related to Methodology for Probabilistic Risk Analysis?

ThissectionbrieflydescribesafewkeyaspectsofPRA,modeldevelopmentandassociateduncertainties.DetailedintroductionstoPRAmethodologyareavailableelsewhere,suchasAngandTang(1984),MorganandHenrion(1990),CullenandFrey(1999)andEPA(2001).Formoredetailedinformation,consultthesereferencesandseetheBibliographyforadditionalsources.

3.3.1. Developing a Probabilistic Risk Analysis Model TherearetwokeyissuesthatshouldbeconsideredindevelopingaPRAmodel;asdiscussedbelow.

StructuralUncertaintyinScenarios

Apotentiallykeysourceofuncertaintyinananalysisisthescenario,whichincludesspecificationofpollutantsources,transportpathways,exposureroutes,timingandlocations,geographicextentandrelatedissues.Thereisnoformalizedmethodologyfordealingquantitativelywithuncertaintyandvariabilityinscenarios.Decisionsregardingwhattoincludeorexcludefromascenariocouldberecastashypothesesregardingwhichagents,pathways,microenvironments,etc.,contributesignificantlytotheoverallexposureandriskofinterest.Inpractice,however,theuseofqualitativemethodstoframeanassessmenttendstobemorecommon,giventheabsenceofaformalquantitativemethodology.

CoupledModels

Forsource‐to‐outcomeriskassessments,itoftenisnecessarytoworkwithmultiplemodels,eachofwhichrepresentsadifferentcomponentofascenario.Forexample,theremaybeseparatemodelsforemissions,airquality,exposure,doseandeffects.Suchmodelsmayhavedifferentspatialandtemporalscales.Whenconductinganintegratedassessment,theremaybesignificantchallengesandbarrierstocouplingsuchmodelsintoonecoherentframework.Sometimes,thecouplingisdonedynamicallyinasoftwareenvironment.Inothercases,theoutputofonemodelmightbeprocessedmanuallytopreparetheinformationforinputtothenextmodel.Furthermore,theremaybefeedbackbetweencomponentsofthescenario(e.g.,poorairqualitymightaffecthumanactivity,which,inturn,couldaffectbothemissionsandexposures)thatareincompletelycapturedornotincluded.Thus,thecouplingofmultiplemodelscanbeapotentiallysignificantsourceofstructuraluncertainty(Özkaynak2009).

3.3.2. Dealing With Dependencies Among Probabilistic Inputs Whenrepresentingtwoormoreinputstoamodelasprobabilitydistributions,thequestionarisesastowhetheritisreasonabletoassumethatthedistributionsarestatisticallyindependent.Ifthereisadependence,itcouldbeassimpleasalinearcorrelationbetweentwoinputs,oritcouldbemorecomplicated,suchasnonlinearornonmonotonicrelationships.Dependenciestypicallyarenotimportantiftheriskestimateorothermodeloutputissensitivetooneornoneoftheprobabilisticinputsthatmighthaveinterdependence.Furthermore,dependenciestypicallyarenotofpracticalimportanceiftheyareweak.Whendependenciesexistandmightsignificantlyinfluence

Page 41: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

30

theriskestimate,theycanbetakenintoaccountusingavarietyofstatisticalsimulationmethodsor,perhapsmoreappropriately,bymodelingthedependenceanalyticallywherepossible.DetailsonmethodsforassessingtheimportanceofpossibledependenciesandofquantifyingthemwhenneededaredescribedinFersonetal.(2004and2009).

Forsometypesofmodels,suchasairqualitymodels,itisnotpossibletointroduceaprobabilitydistributiontooneinput(e.g.,ambienttemperatureataparticularlocation)withoutaffectingvariablesatotherlocationsortimes(e.g.,temperaturesinotherlocationsatthesametimesortemporaltrendsintemperature).Insuchcases,itisbettertoproducean“ensemble”ofalternativetemperaturefields,eachofwhichisinternallyconsistent.Individualmembersofanensembleusuallyarenotinterpretedasrepresentingaprobabilitysample;however,comparisonofmultipleensemblesofmeteorologicalconditions,forexample,canprovideinsightintonaturalsourcesofvariabilityinambientconcentrations.

3.3.3. Conducting the Probabilistic Analysis

QuantifyingUncertaintyandVariabilityinModelInputsandParameters

Afterthemodelsareselectedordevelopedtosimulateascenarioofinterest,attentiontypicallyturnstothedevelopmentofinputdataforthemodel.Thereisasubstantialamountofliteratureregardingtheapplicationofstatisticalmethodsforquantifyinguncertaintyandvariabilityinmodelinputsandparametersbasedonempiricaldata(e.g.,AngandTang1984;CullenandFrey1999;MorganandHenrion1990;USEPA2001).Forexample,acommonlyusedmethodforquantifyingvariabilityinamodelinputistoobtainasampleofdata,selectatypeofparametricprobabilitydistributionmodeltofittothedata(e.g.,normal,lognormalorotherform),estimatetheparametersofthedistributionbasedonthedata,critiquethegoodness‐of‐fitusinggraphical(e.g.,probabilityplot)andstatistical(e.g.,Anderson‐Darling,Chi‐SquareorKolmogorov‐Smirnovtests)methodsandchooseapreferredfitteddistribution.Thismethodologycanbeadjustedtoaccommodatevarioustypesofdata,suchasdatathataresamplesfrommixturesofdistributionsorthatcontainnon‐detected(censored)values.Uncertaintiescanbeestimatedbasedonconfidenceintervalsforstatisticsofinterest,suchasmeanvalues,ortheparametersoffrequencydistributionsforvariability.Varioustextsandguidancedocuments,bothAgencyandprogrammatic,describetheseapproaches,includingtheGuidingPrinciplesforMonteCarloAnalysis(USEPA1997b).

Themostcommonmethodforestimatingaprobabilitydistributionintheoutputofamodel,basedonprobabilitydistributionsspecifiedformodelinputs,isMCS(CullenandFrey1999;MorganandHenrion1990).MCSispopularbecauseitisveryflexible.MCScanbeusedwithawidevarietyofprobabilitydistributionsaswellasdifferenttypesofmodels.ThemainchallengeforMCSisthatitrequiresrepetitivemodelcalculationstoconstructasetofpseudo‐randomnumbersformodelinputsandthecorrespondingestimatesformodeloutputsofinterest.TherearealternativestoMCSthataresimilarbutmorecomputationallyefficient,suchasLatinHypercubeSampling(LHS).TechniquesareavailableforsimulatingcorrelationsbetweeninputsinbothMCSandLHS.Formodelswithverysimplefunctionalforms,itmaybepossibletouseexactorapproximateanalyticalcalculations,butsuchsituationsareencounteredinfrequentlyinpractice.Theremaybesituationsinwhichthedatadonotconformtoawell‐definedprobabilitydistribution.Insuchcases,algorithms(suchasMarkovChainMonteCarlo)canestimateaprobabilitydistributionbycalculatingamathematicalformdescribingthepatternofobserveddata.Thisform,calledthelikelihoodfunction,isakeycomponentofBayesianinferenceand,therefore,servesasthebasisforsomeoftheanalyticalapproachestouncertaintyandvariabilitydescribedbelow.

Theuseofempiricaldatapresumesthatthedatacomprisearepresentative,randomsample.Ifknownbiasesorotherdataqualityproblemsexist,orifthereisascarcityorabsenceofrelevantdata,thennaïverelianceonavailableempiricaldataislikelytoresultinmisleadinginferencesin

Page 42: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

31

theanalysis.Alternatively,estimatesofuncertaintyandvariabilitycanbeencoded,usingformalprotocols,basedonelicitationofexpertjudgment(e.g.,MorganandHenrion1990,USEPA2011a).Elicitationofexpertjudgmentforsubjectiveprobabilitydistributionsisusedinsituationswherethereareinsufficientdatatosupportastatisticalanalysisofuncertainty,butinwhichthereissufficientknowledgeonthepartofexpertstomakeaninferenceregardinguncertainty.Forexample,EPAconductedanexpertelicitationstudyontheconcentration‐responserelationshipbetweentheannualaverageambientlessthan2.5micrometer(µm)diameterparticulatematter(PM2.5)exposureandannualmortality(IEC2006;seealsoCaseStudies6and14intheAppendix).Subjectiveprobabilitydistributionsthatarebasedonexpertjudgmentcanbe“updated”withnewdataastheybecomeavailableusingBayesianstatisticalmethods.

StructuralUncertaintyinModels

Theremaybesituationsinwhichitprovesusefultoevaluatenotjusttheuncertaintiesininputsandparametervalues,butalsouncertaintiesregardingwhetheramodeladequatelycaptures—inahypothesized,mathematical,structuredform—therelationshipunderinvestigation.Aqualitativeapproachtoevaluatingthestructuraluncertaintyinamodelincludesdescribingthecriticalassumptionswithinamodel,thedocumentationofamodelorthemodelquality,andhowthemodelfitsthepurposeoftheassessment.Quantitativeapproachestoevaluatingstructuraluncertaintyinmodelsaremanifold.Theseincludeparameterizationofageneralmodelthatcanbereducedtoalternativefunctionalforms(e.g.,MorganandHenrion1990),enumerationofalternativemodelsinaprobabilitytree(e.g.,Evansetal.1994),comparingalternativemodelsbyevaluatinglikelihoodfunctions(e.g.,Royall1997;BurnhamandAnderson2002),poolingresultsofmodelalternativesusingBayesianmodelaveraging(e.g.,Hoetingetal.1999)ortestingthecausalrelationshipswithinalternativemodelsusingBayesianNetworks(Pearl2009).

SensitivityAnalysis:IdentifyingtheMostImportantModelInputs

Probabilisticmethodstypicallyfocusonhowuncertaintyorvariabilityinamodelinputaffect[orresultin]withrespecttouncertaintyorvariabilityinamodeloutput.Afteraprobabilisticanalysisiscompleted,sensitivityanalysistypicallytakestheperspectiveoflookingbacktoevaluatehowmuchofthevariationinthemodeloutputisattributabletoindividualmodelinputs(e.g.,FreyandPatil2002;Mokhtarietal.2006;Saltellietal.2004).

Therearemanytypesofsensitivityanalysismethods,includingsimpletechniquesthatinvolvechangingthevalueofoneinputatatimeandassessingtheeffectonanoutput,andstatisticalmethodsthatevaluatewhichofmanysimultaneouslyvaryinginputscontributethemosttothevarianceofthemodeloutput.Sensitivityanalysiscananswerthefollowingkeyquestions:

Whatistheimpactofchangesininputvaluesonmodeloutput?

Howcanvariationinoutputvaluesbeapportionedamongmodelinputs?

Whataretherangesofinputsassociatedwithbestorworstoutcomes?

Whatarethekeycontrollablesourcesofvariability?

Whatarethecriticallimits(e.g.,theemissionreductiontarget)?

Whatarethekeycontributorstotheoutputuncertainty?

Thus,sensitivityanalysiscanbeusedtoinformdecisionmaking.

Iteration

Therearetwomajortypesofiterationinriskassessmentmodeling.Oneisiterativerefinementofthetypeofanalysis,perhapsstartingwitharelativelysimpleDRAasascreeningstepinaninitial

Page 43: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

32

levelofanalysisandproceedingtomorerefinedtypesofassessmentsasneededinsubsequentlevelsofanalysis.ExamplesofmorerefinedlevelsofassessmentincludeapplicationofsensitivityanalysistoDRA;theuseofprobabilisticmethodstoquantifyvariabilityonly,uncertaintyonly,orcombinedvariabilityanduncertainty(torepresentarandomlyselectedindividual);ortheuseoftwo‐dimensionalprobabilisticmethodsfordistinguishingandsimultaneouslycharacterizingbothuncertaintyandvariability.

Theothertypeofiterationoccurswithinaparticularlevelandincludesiterativeeffortstoformulateamodel,obtaindataandevaluatethemodeltoprioritizedataneeds.Forexample,amodelmayrequirealargenumberofinputassumptions.Toprioritizeeffortsofspecifyingdistributionsforuncertaintyandvariabilityformodelinputs,itisusefultodeterminewhichmodelinputsarethemostinfluentialwithrespecttotheassessmentendpoint.Therefore,sensitivitycanbeusedbasedonpreliminaryassessmentsofrangesordistributionsforeachmodelinputtodeterminewhichinputsarethemostimportanttotheassessment.Refinedeffortstocharacterizedistributionsthencanbeprioritizedtothemostimportantinputs.

Page 44: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

33

4. SUMMARY AND RECOMMENDATIONS 4.1. Probabilistic Risk Analysis and Related Analyses Can Improve

the Decision-Making Process at EPA PRAcanprovideuseful(evencritical)informationabouttheuncertaintiesandvariabilityinthedata,models,scenario,expertjudgmentsandvaluesincorporatedinriskassessmentstosupportdecisionmakingacrosstheAgency.AsdiscussedinthispaperPRAisananalyticalmethodologycapableofincorporatinginformationregardinguncertaintyand/orvariabilityinriskanalysestoprovideinsightonthedegreeofcertaintyofariskestimateandhowtheriskestimatevarieswithintheexposedpopulation.TraditionalapproachessuchasDRA,oftenreportrisksusingdescriptorssuchas“centraltendency,”“highend”(e.g.,90thpercentileorabove)or“maximumanticipatedexposure”.BycontrastPRAcanbeusedtodescribemorecompletelytheuncertaintysurroundingsuchestimates,aswellastoidentifythekeycontributorstouncertaintyandvariabilityinpredictedexposuresorriskestimates.Thisinformationthencanbeusedbydecisionmakerstoweighalternatives,ortomakedecisionsonwhethertocollectadditionaldata,ortoconductadditionalresearchinordertoreducetheuncertaintyandfurthercharacterizevariabilitywithintheexposedpopulation.Informationonuncertaintiesandvariabilityinexposureandresponsecanultimatelyimprovetheriskestimates.

PRAcanbeusedtoobtaininsightonwhetheronemanagementalternativeismorelikelytoreduceriskscomparedtoanother.Inaddition,PRAcanfacilitatethedevelopmentofmodelingscenariosandthesimultaneousconsiderationofmultiplemodelalternatives.Probabilisticmethodsofferanumberoftoolsdesignedtoincreaseconfidenceindecisionmakingthroughtheincorporationofinputuncertaintyandvariabilitycharacterizationandprioritizationinriskanalyses.Forexample,onePRAtool,sensitivityanalysescanbeusedtoidentifyinfluentialknowledgegapsintheestimationofrisk;thisimprovestransparencyinthepresentationoftheseuncertaintiesandimprovestheabilitytocommunicatethemostrelevantinformationmoreclearlytodecisionmakersandstakeholders.PRAallowsonetoinvestigatepotentialchangesindecisionsthatcouldresultfromthecollectionofadditionalinformation.However,theadditionalresources(e.g.,time,costs,orexpertise)toundertakeneedtobeweighedagainstthepotentialimprovementsinthedecisionmakingprocess.Ultimately,PRAmayenhancethescientificfoundationoftheEPA’sapproachtodecisionmaking.

Thevarioustoolsandmethodsdiscussedinthiswhitepapercanbeutilizedatallstagesofriskanalysisandalsocanaidthedecision‐makingprocessby,forexample,characterizinginter‐individualvariabilityanduncertainties.

PRAandrelatedmethodsareemployedinvaryingdegreesacrosstheAgency.BasicguidanceexistsatEPAontheuseandacceptabilityofPRAforriskestimation,butimplementationvariesgreatlywithinprograms,officesandregions.TheuseofMonteCarloorotherprobability‐basedtechniquestoderivearangeofpossibleoutputsfromuncertaininputsisafairlywell‐developedapproachwithinEPA.Althoughhighlysophisticatedhumanexposureassessmentandecologicalriskapplicationshavebeendeveloped,theuseofPRAmodelstoassesshumanhealtheffectsanddose‐responserelationshipshasbeenmorelimitedattheAgency.

TheevaluationoftheapplicationofPRAtechniquesunderspecificlawsandregulationsvariesbyprogram,officeandregion.Movingforward,itisimportanttobroadendiscussionsbetweenriskassessorsandriskmanagersregardinghowPRAtoolscanbeusedtosupportspecificdecisionsandhowtheycanbeusedwithintheregulatoryframeworkusedbyprograms,offices,andregionstomakedecisions.Thiscanbeaccomplishedbyexpandingthedialoguebetweenassessorsand

Page 45: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

34

managesatalllevelsregardinghowthePRAtoolshavebeenusedandhowtheyhaveenhanceddecisionmaking.

IncreaseduseofPRAandconsistentapplicationofPRAtoolsinsupportofEPAdecisionmakingrequiresenhancedinternalcapacityforconductingtheseassessments,aswellasimprovedinterpretationandcommunicationofsuchinformationinthecontextofdecisions.ImprovementsofAgencycapacitycouldbeaccomplishedthroughsharingofexperiences,knowledgeandtrainingandincreasedavailabilityoftoolsandmethods.

4.2. Major Challenges to Using Probabilistic Risk Analysis to Support Decisions

ThechallengesforEPAaretwo‐fold.AsanAgencyresponsibleforprotectinghumanhealthandtheenvironment,EPAmakesregulatoryandpolicydecisions,eveninthepresenceofconflictingstakeholderpositionsandtheinevitableuncertaintiesinthescience.ThefirstchallengeforEPAistodeterminehowtoconductitsdecision‐makingresponsibilities,weighingdeterminationsofwhatconstitutestoomuchuncertaintytomakeadecision,againstpotentialadverseconsequencesofpostponingdecisions.

Thesecondchallenge,isthatalthoughcurrentPRAtechniquesareavailablethatwouldhelptoinformEPAdecision‐makingprocesses,researchandguidanceareneededtoimprovethesemethodsforamorecompleteimplementationofPRAinHHRAandERA.Inparticular,additionalguidanceisneededtohelpanalystsanddecisionmakersbetterunderstandhowtoincorporatePRAapproachesintothedecision‐makingprocess.Thisincludes,guidanceonwhichstatisticaltoolstouseandwhentousethem,andhowprobabilisticinformationcanhelptoinformthescientificbasisofdecisions.BothDRAandPRAaswellasappropriatestatisticalmethodsmaybeusefulatanystageoftheriskanalysisanddecision‐makingprocess,fromplanningandscopingtocharacterizingandcommunicatinguncertainty.

AsnotedinSection3.3,therearesignificantchallengesinproperlyaccountingforuncertaintyandvariabilitywhenmultiplemodelsarecoupledtogethertorepresentthesource‐to‐outcomecontinuum.Moreover,thecouplingofmultiplemodelsmightneedtoinvolveinputsandcorrespondinguncertaintiesthatareincorporatedintomorethanonemodel,potentiallyresultingincomplexdependencies.Integrativeresearchoncoupledmodeluncertaintieswillbequitevaluable.

Theremaybemismatchesinthetemporalandspatialresolutionofeachmodelthatconfoundtheabilitytopropagateuncertaintyandvariabilityfromonemodeltoanother.Forsomemodels,thekeyuncertaintiesmaybeassociatedwithinputs,whereasforothermodels,thekeyuncertaintiesmaybeassociatedwithstructureorparameterizationalternatives.Modelintegrationandharmonizationactivitieswillbeimportanttoaddressingthesetechnicalissues.

4.3. Recommendations for Enhanced Utilization of Probabilistic Risk Analysis at EPA

Someexamplesofareaswhereneworupdatedguidancewouldbehelpfularethese:

IdentificationofdifferenttypesofinformationrequiredforthevariousAgencydecision‐makingprocesses,suchasdataanalysis,tools,models,anduseofexperts.

Useofprobabilisticapproachestoevaluatehealtheffectsdata.

UseofprobabilisticapproachesforERA.

Page 46: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

35

Integratingprobabilisticexposureandriskestimatesandcommunicatinguncertaintyandvariability.

Inordertosupportthedevelopmentofguidanceontheseorrelatedtopics,followingstudiesorresearcharerecommended:

TheuseofPRAmodelstoevaluatetoxicitydatahasbeenverylimited.Scientific,technicalandpolicy‐baseddiscussionsareneededinthisarea.

Additionalresearchonformalmethodsfortreatingmodeluncertaintieswillbevaluable.

Somestepstoimproveimplementationincludethese:

InformingdecisionmakersabouttheadvantagesanddisadvantagesofusingPRAtechniquesintheirdecision‐makingprocessesthroughlectures,webinarsandcommunicationsregardingthetechniquesandtheiruseinEPA.

IncorporatingadiscussionofPRAtoolsduringPlanningandScopingforHHRAsandERAs.

ContinuingthedialoguebetweenassessorsandmanagersonhowtousePRAwithintheregulatorydecisionmakingprocess.

ConductingmeetingsanddiscussionsofPRAtechniquesandtheirapplicationwithbothmanagersandassessorstoaidinprovidinggreaterconsistencyandtransparencyinEPA’sriskassessmentandriskmanagementprocessandindevelopingEPA’sinternalcapacity.

Developinga“CommunityofPractice”forfurtherdiscussionregardingtheapplicationofPRAtechniquesandtheuseofthesetoolsindecisionmaking.

Riskassessorsandriskmanagersneedinformationandtrainingsothattheycanbetterutilizethesetools.Educationandexperiencewillgeneratefamiliaritywiththesetools,whichwillhelpanalystsanddecisionmakersbetterunderstandandconsidermorefullyutilizingthesetechniqueswithintheirregulatoryprograms.Increasedtrainingisneededtofacilitateunderstandingonalllevelsandmayincludethefollowing:

ProvidingintroductoryaswellasadvancedtrainingtoallEPAoffices.

TrainingriskassessorsandriskmanagersinthePRAtechniquessothattheycanlearnaboutthevarioustoolsavailable,theirapplications,softwareandreviewconsiderations,andresourcesforadditionalinformation(e.g.,expertsandsupportserviceswithintheAgency).

Providingeasilyavailable,flexible,modulartrainingforalllevelsofexperiencetofamiliarizeEPAemployeeswiththemenuoftoolsandtheircapacities.

Providingliveandrecordedseminarsandwebinarsforintroductoryandsupplementaleducation,aswellasperiodic,centralizedhands‐ontrainingsessionsdemonstratinghowtoutilizesoftwareprograms.

TrainingiscriticalbothforanimprovedunderstandingbutalsotobuildincreasedcapacityintheAgencyandexplicitstepscouldincludethese:

Demonstrating,throughinformationalopportunitiesandresourcelibraries,thevarioustoolsandmethodsthatcanbeusedatallstagesofriskanalysistoaidthedecision‐makingprocessbycharacterizinginter‐individualvariabilityanduncertainties.

Promotingthesharingofexperience,knowledge,modelsandbestpracticesviameetingsofriskassessorsandmanagers;electronicexchanges,suchastheEPAPortalEnvironmental

Page 47: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

36

ScienceConnector(https://ssoprod.epa.gov/sso/jsp/obloginESCNew.jsp);andmoredetaileddiscussionsofthecasestudies.

AsEPAworkstowardthemoreintegratedevaluationofenvironmentalproblems,thiswillincludenotjusttheimprovedunderstandingofsinglepollutants/singlemedia,butmulti‐pollutant,multi‐mediaandmulti‐receptoranalysiswithinadecisionanalyticframework.EPAisbeginningtobuildsuchintegratedcapabilityintoanalyticaltoolslikePRA(BabendreierandCastleton2005;Stahletal.2011).

TheRAFwillbetakingaleadershiprolethroughtheUncertaintyandVariabilityWorkgrouptomorefullyevaluatetheapplicationanduseofPRAtoolsandbroadeningthedialoguebetweenassessorsandmanagers.UpdatesontheprogressofthisTechnicalPanelwillbeprovidedontheRAFwebpageat:www.epa.gov/raf.

Page 48: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

37

GLOSSARY Analysis.Examinationofanythingcomplextounderstanditsnatureortodetermineitsessentialfeatures(WHO2004).

Assessment.Adeterminationorappraisalofpossibleconsequencesresultingfromananalysisofdata(2011b).

Assessmentendpoint.Anexplicitexpressionoftheenvironmentalvaluethatistobeprotected,operationallydefinedbyanecologicalentityanditsattributes.Forexample,salmonarevaluedecologicalentities;reproductionandageclassstructurearesomeoftheirimportantattributes.Together,salmon“reproductionandageclassstructure”formanassessmentendpoint(USEPA1998b).

Bayesianprobability.Anapproachtoprobability,representingapersonaldegreeofbeliefthatavalueofrandomvariablewillbeobserved.Alternatively,theuseofprobabilitymeasurestocharacterizethedegreeofuncertainty(Gelmanetal.2004).

BayesianAnalysis.Bayesiananalysisisamethodofstatisticalinferenceinwhichtheknowledgeofprioreventsisusedtopredictfutureevents(USEPA2011b).

Correlation.Anestimateofthedegreetowhichtwosetsofvariablesvarytogether,withnodistinctionbetweendependentandindependentvariables.Correlationreferstoabroadclassofstatisticalrelationshipsinvolvingdependence(USEPA2012).

Criticalcontrolpoint.Acontrollablevariablethatcanbeadjustedtoreduceexposureandrisk.Forexample,acriticalcontrolpointmightbetheemissionratefromaparticularemissionsource.Theconceptofcriticalcontrolpointisfromthehazardassessmentandcriticalcontrolpointconceptforriskmanagementthatisusedinspaceandfoodsafetyapplications,amongothers(USEPA2006c).

Criticallimit.Anumericalvalueofacriticalcontrolpointatorbelowwhichriskisconsideredtobeacceptable.Acriterionthatseparatesacceptabilityfromunacceptability(USEPA2006c).

Deterministic.Amethodologyrelyingonpoint(i.e.,exact)valuesasinputstoestimaterisk;thisobviatesquantitativeestimatesofuncertaintyandvariability.Resultsalsoarepresentedaspointvalues.Uncertaintyandvariabilitymaybediscussedqualitativelyorsemi‐quantitativelybymultipledeterministicriskestimates(USEPA2006b).

Deterministicriskassessment(DRA).Riskevaluationinvolvingthecalculationandexpressionofriskasasinglenumericalvalueor“singlepoint”estimateofrisk,withuncertaintyandvariabilitydiscussedqualitatively(USEPA2012).

Ecologicalriskassessment.Theprocessthatevaluatesthelikelihoodthatadverseecologicaleffectsmayoccurorareoccurringasaresultofexposuretooneormorestressors(USEPA1998b).

Ecosystem.Thebioticcommunityandabioticenvironmentwithinaspecifiedlocationinspaceandtime(USEPA1998b).

Ensemble.Amethodforpredictivemodelingbasedonmultiplemeasuresofthesameeventovertime(e.g.,theamountofcarbondioxidepresentintheatmosphereatselectedtimepoints).Thecollectionofdatainputisknownasanensembleandcanbeusedtodevelopaquantificationofpredictionvariabilitywithinthemodel.Ensemblemodelingisusedmostcommonlyinatmosphericpredictioninforecasting,althoughensemblemodelinghasbeenappliedtobiologicalsystemstobetterquantifyrisksofeventsorperturbationswithinbiologicalsystems(FuentesandFoley2012).

Environment.Thesumofallexternalconditionsaffectingthelife,developmentandsurvivalofanorganism(USEPA2010a).

Page 49: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

38

Expertelicitation.Asystematicprocessofformalizingandquantifying,typicallyinprobabilisticterms,expertjudgmentsaboutuncertainquantities(USEPA2011a).

Frequentist(orfrequency)probability.Aviewofprobabilitythatconcernsitselfwiththefrequencywith which an event occurs givenalongsequenceofidenticalandindependenttrials(USEPA1997b).

Hazardidentification.Theriskassessmentprocessofdeterminingwhetherexposuretoastressorcancauseanincreaseintheincidenceorseverityofaparticularadverseeffect,andwhetheranadverseeffectislikelytooccur(USEPA2012).

Humanhealthriskassessment.1.Theprocesstoestimatethenatureandprobabilityofadversehealtheffectsinhumanswhomaybeexposedtochemicalsincontaminatedenvironmentalmedia,noworinthefuture(USEPA2010b).2.Theevaluationofscientificinformationonthehazardouspropertiesofenvironmentalagents(hazardcharacterization),thedose‐responserelationship(dose‐responseassessment),andtheextentofhumanexposuretothoseagents(exposureassessment).Theproductoftheriskassessmentisastatementregardingtheprobabilitythatpopulationsorindividualssoexposedwillbeharmedandtowhatdegree(riskcharacterization)(USEPA2006a).

Inputs.Quantitiesthatareappliedtoamodel(WHO2008).

LikelihoodFunction.Anapproachtomodelingexposureinwhichlong‐termexposureofanindividualissimulatedasthesumofseparateshort‐termexposureevents(USEPA2001).

Microenvironment.Well‐definedsurroundingssuchasthehome,office,automobile,kitchen,store,etc.,thatcanbetreatedashomogenous(orwellcharacterized)intheconcentrationsofachemicalorotheragent(USEPA1992).

Microexposureevent(MEE)analysis.Anapproachtomodelingexposureinwhichlong‐termexposureofanindividualissimulatedasthesumofseparateshort‐termexposureevents(USEPA2001).

Model.Amathematicalrepresentationofanaturalsystemintendedtomimicthebehavioroftherealsystem,allowingdescriptionofempiricaldata,andpredictionsaboutuntestedstatesofthesystem(USEPA2006b).

Modelboundaries.1.Decisionsregardingthetime,space,numberofchemicals,etc.,usedinguidingmodelingofthesystem.Riskscanbeunderstatedoroverstatedifthemodelboundaryismis‐specified.Forexample,ifastudyareaisdefinedtobetoolargeandincludesasignificantnumberoflow‐exposureareas,thenapopulation‐levelriskdistributioncanbedilutedbyincludinglessexposedindividuals,whichcan,inturn,resultinarisk‐baseddecisionthatdoesnotprotectsufficientlythemostexposedindividualsinthestudyarea.2.Designatedareasofcompetenceofthemodel,includingtime,space,pathogens,pathways,exposedpopulations,andacceptablerangesofvaluesforeachinputandjointlyamongallinputsforwhichthemodelmeetsdataqualityobjectives(WHO2008).

Modeling.Developmentofamathematicalorphysicalrepresentationofasystemortheorythataccountsforallorsomeofitsknownproperties.Modelsoftenareusedtotesttheeffectofchangesofcomponentsontheoverallperformanceofthesystem(USEPA2010a).

Modeluncertainty(sourcesof):

Modelstructure.Asetofassumptionsandinferenceoptionsuponwhichamodelisbased,includingunderlyingtheoryaswellasspecificfunctionalrelationships(WHO2008).

Page 50: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

39

Modeldetail.Levelofsimplicityordetailassociatedwiththefunctionalrelationshipsassumedinthemodelcomparedtotheactualbutunknownrelationshipsinthesystembeingmodeled(WHO2008).

Extrapolation.Useofmodelsoutsideoftheparameterspaceusedintheirderivationmayresultinerroneouspredictions.Forexample,athresholdforhealtheffectsmayexistatexposurelevelsbelowthosecoveredbyaparticularepidemiologicalstudy.Ifthatstudyisusedinmodelinghealtheffectsatthoselowerlevels(anditisassumedthatthelevelofresponseseeninthestudyholdsforlowerlevelsofexposure),thendiseaseincidencemaybeoverestimated(USEPA2007a).

MonteCarloanalysis(MCA)orsimulation(MCS).Arepeatedrandomsamplingfromthedistributionofvaluesforeachoftheparametersinagenericexposureorriskequationtoderiveanestimateofthedistributionofexposuresorrisksinthepopulation(USEPA2006b).

One‐dimensionalMonteCarloanalysis(1‐DMCA).Anumericalmethodofsimulatingadistributionforanendpointofconcernasafunctionofprobabilitydistributionsthatcharacterizevariabilityoruncertainty.Distributionsusedtocharacterizevariabilityaredistinguishedfromdistributionsusedtocharacterizeuncertainty(WHO2008).

Parameter.Aquantityusedtocalibrateorspecifyamodel,suchas‘parameters’ofaprobabilitymodel(e.g.,meanandstandarddeviationforanormaldistribution).Parametervaluesoftenareselectedbyfittingamodeltoacalibrationdataset(WHO2008).

Probability.Afrequentistapproachconsidersthefrequencywithwhichsamplesareobtainedwithinaspecifiedrangeorforaspecifiedcategory(e.g.,theprobabilitythatanaverageindividualwithaparticularmeandosewilldevelopanillness)(WHO2008).

Probabilisticriskanalysis(PRA).Calculationandexpressionofhealthrisksusingmultipleriskdescriptorstoprovidethelikelihoodofvariousrisklevels.Probabilisticriskresultsapproximateafullrangeofpossibleoutcomesandthelikelihoodofeach,whichoftenispresentedasafrequencydistributiongraph,thusallowinguncertaintyorvariabilitytobeexpressedquantitatively(USEPA2012).

Problemformulation.Theinitialstageofariskassessmentwherethepurposeoftheassessmentisarticulated,exposureandriskscenariosareconsidered,aconceptualmodelisdeveloped,andaplanforanalyzingandcharacterizingriskisdetermined(USEPA2004a).

Referenceconcentration(RfC).Anestimate(withuncertaintyspanningapproximatelyanorderofmagnitude)ofacontinuousinhalationexposuretothehumanpopulation(includingsensitivesubgroups)thatislikelytobewithoutanappreciableriskofdeleteriouseffectsduringalifetime.ItcanbederivedfromaNo‐Observed‐Adverse‐EffectLevel(NOAEL),Lowest‐Observed‐Adverse‐EffectLevel(LOAEL),orbenchmarkconcentration,withuncertaintyfactorsgenerallyappliedtoreflectlimitationsofthedataused.ItisgenerallyusedinEPA’snoncancerhealthassessments(USEPA2007a).

Referencedose(RfD).Anestimate(withuncertaintyspanningapproximatelyanorderofmagnitude)ofadailyoralexposuretothehumanpopulation(includingsensitivesubgroups)thatislikelytobewithoutanappreciableriskofdeleteriouseffectsduringalifetime.ItcanbederivedfromaNOAEL,LOAELorbenchmarkdose,withuncertaintyfactorsgenerallyappliedtoreflectlimitationsofthedataused.ItistypicallyusedinEPA’snoncancerhealthassessments(USEPA2011c).

Risk.1.Riskincludesconsiderationofexposuretothepossibilityofanadverseoutcome,thefrequencywithwhichoneormoretypesofadverseoutcomesmayoccur,andtheseverityor

Page 51: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

40

consequencesoftheadverseoutcomesifsuchoccur.2.Thepotentialforrealizationofunwanted,adverseconsequencestohumanlife,health,propertyortheenvironment.3.Theprobabilityofadverseeffectsresultingfromexposuretoanenvironmentalagentormixtureofagents.4.Thecombinedanswersto:Whatcangowrong?Howlikelyisit?Whataretheconsequences?(USEPA2011c).

Riskanalysis.Aprocessforidentifying,characterizing,controllingandcommunicatingrisksinsituationswhereanorganism,system,subpopulationorpopulationcouldbeexposedtoahazard.Riskanalysisisaprocessthatincludesriskassessment,riskmanagementandriskcommunication(WHO2008).

Riskassessment.1.Aprocessintendedtocalculateorestimatetherisktoagiventargetorganism,system,subpopulationorpopulation,includingtheidentificationofattendantuncertaintiesfollowingexposuretoaparticularagent,takingintoaccounttheinherentcharacteristicsoftheagentofconcern,aswellasthecharacteristicsofthespecifictargetsystem(WHO2008).2.Theevaluationofscientificinformationonthehazardouspropertiesofenvironmentalagents(hazardcharacterization),thedose‐responserelationship(dose‐responseassessment),andtheextentofhumanexposuretothoseagents(exposureassessment)(NRC1983).Theproductoftheriskassessmentisastatementregardingtheprobabilitythatpopulationsorindividualssoexposedwillbeharmedandtowhatdegree(riskcharacterization;USEPA2000a).3.Qualitativeandquantitativeevaluationoftheriskposedtohumanhealthortheenvironmentbytheactualorpotentialpresenceoruseofspecificpollutants(USEPA2012).

Risk‐baseddecisionmaking.Aprocessthroughwhichdecisionsaremadeaccordingtotheriskeachposedtohumanhealthandtheenvironment(USEPA2012).

Riskmanagement.Adecision‐makingprocessthattakesintoaccountenvironmentallaws;regulations;andpolitical,social,economic,engineeringandscientificinformation,includingariskassessment,toweighpolicyalternativesassociatedwithahazard(USEPA2011c).

Scenario.Asetoffacts,assumptionsandinferencesabouthowexposuretakesplacethataidstheexposureassessorinevaluating,estimatingorquantifyingexposures(USEPA1992).Scenariosmightincludeidentificationofpollutants,pathways,exposureroutesandmodesofaction,amongothers.

Sensitivityanalysis.Theprocessofchangingonevariablewhileleavingtheothersconstanttodetermineitseffectontheoutput.Thisprocedurefixeseachuncertainquantityatitscrediblelowerandupperbounds(holdingallothersattheirnominalvalues,suchasmedians)andcomputestheresultsofeachcombinationofvalues.Theresultshelptoidentifythevariablesthathavethegreatesteffectonexposureestimatesandhelpfocusfurtherinformation‐gatheringefforts(USEPA2011b).

Tieredapproach.Referstovarioushierarchicaltiers(levels)ofcomplexityandrefinementfordifferenttypesofmodelingapproachesthatcanbeusedinriskassessment.Adeterministicriskassessmentwithconservativeassumptionsisanexampleofalowerleveltypeofanalysis(Tier0)thatcanbeusedtodeterminewhetherexposuresandrisksarebelowlevelsofconcern.Examplesofprogressivelyhigherlevelsincludetheuseofdeterministicriskassessmentcoupledwithsensitivityanalysis(Tier1),theuseofprobabilistictechniquestocharacterizeeithervariabilityoruncertaintyonly(Tier2),andtheuseoftwo‐dimensionalprobabilistictechniquestodistinguishbetweenbutsimultaneouslycharacterizebothvariabilityanduncertainty(Tier3)(USEPA2004aandWHO2008).

Two‐dimensionalMonteCarloanalysis(2‐DMCA).Anadvancednumericalmodelingtechniquethatusestwostagesofrandomsampling,alsocallednestedloops,todistinguishbetween

Page 52: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

41

variabilityanduncertaintyinexposureandtoxicityvariables.Thefirststage,oftencalledtheinnerloop,involvesacomplete1‐DMCAsimulationofvariabilityinrisk.Inthesecondstage,oftencalledtheouterloop,parametersoftheprobabilitydistributionsareredefinedtoreflectuncertainty.Theseloopsarerepeatedmanytimesresultinginmultipleriskdistributions,fromwhichconfidenceintervalsarecalculatedtorepresentuncertaintyinthepopulationdistributionofrisk(WHO2008).

Uncertainty.Uncertaintyoccursbecauseofalackofknowledge.Itisnotthesameasvariability.Forexample,ariskassessormaybeverycertainthatdifferentpeopledrinkdifferentamountsofwaterbutmaybeuncertainabouthowmuchvariabilitythereisinwaterintakeswithinthepopulation.Uncertaintyoftencanbereducedbycollectingmoreandbetterdata,whereasvariabilityisaninherentpropertyofthepopulationbeingevaluated.Variabilitycanbebettercharacterizedwithmoredatabutitcannotbereducedoreliminated.Effortstoclearlydistinguishbetweenvariabilityanduncertaintyareimportantforbothriskassessmentandriskcharacterization,althoughtheybothmaybeincorporatedintoanassessment(USEPA2011c).

Uncertaintyanalysis.Adetailedexaminationofthesystematicandrandomerrorsofameasurementorestimate;ananalyticalprocesstoprovideinformationregardinguncertainty(USEPA2006b).

Valueofinformation.Ananalysisthatinvolvesestimatingthevaluethatnewinformationcanhavetoariskmanagerbeforetheinformationisactuallyobtained.Itisameasureoftheimportanceofuncertaintyintermsoftheexpectedimprovementinariskmanagementdecisionthatmightcomefrombetterinformation(USEPA2001).

Variability.Referstotrueheterogeneityordiversity,asexemplifiedinnaturalvariation.Forexample,amongapopulationthatdrinkswaterfromthesamesourceandwiththesamecontaminantconcentration,therisksfromconsumingthewatermayvary.Thismayresultfromdifferencesinexposure(e.g.,differentpeopledrinkingdifferentamountsofwaterandhavingdifferentbodyweights,exposurefrequenciesandexposuredurations),aswellasdifferencesinresponse(e.g.,geneticdifferencesinresistancetoachemicaldose).Thoseinherentdifferencesarereferredtoasvariability.Differencesamongindividualsinapopulationarereferredtoasinter‐individualvariability,anddifferencesforoneindividualovertimearereferredtoasintra‐individualvariability(USEPA2011c).

Page 53: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

42

REFERENCES Ang,A.H‐S.,andW.H.Tang.1984.ProbabilityConceptsinEngineeringPlanningandDesign,Volume2:Decision,Risk,andReliability.NewYork:JohnWiley&Sons.

Babendreier,J.E.,andK.J.Castleton.2005.“InvestigatingUncertaintyandSensitivityinIntegrated,MultimediaEnvironmentalModels:ToolsforFRAMES‐3MRA.”EnvironmentalModeling&Software20:1043–1055.

Bloom,D.L.,D.M.Byrne,andJ.M.Andreson.1993.CommunicatingRisktoSeniorEPAPolicy‐Makers:AFocusGroupStudy.PreparedbyBloomResearchandtheOfficeofAirQualityPlanningandStandards.ResearchTrianglePark,NC:USEPA.

Burnham,K.P.,andD.R.Anderson.2002.ModelSelectionandInference:APracticalInformation‐TheoreticApproach.2nded.NewYork:Springer‐Verlag.

Clyde,M.,P.Guttorp,andE.Sullivan.2000.EffectsofAmbientFineandCoarseParticlesonMortalityinPhoenix,Arizona.NationalResearchCenterforStatisticsandtheEnvironmentTechnicalReportSeries,No.40.

Cooke,R.M.1991.ExpertsinUncertainty:OpinionandSubjectiveProbabilityinScience.NewYork:OxfordUniversityPress.

Costanza,R.,S.O.Funtowicz,andJ.R.Ravetz.1997.“AssessingandCommunicatingDataQualityinPolicy‐RelevantResearch.”InFrontiersinEcologicalEconomics:TransdisciplinaryEssays,byRobertCostanza,283–308.Cheltenham,UK:E.Elgar.

Cullen,A.C.,andH.C.Frey.1999.ProbabilisticExposureAssessment:AHandbookforDealingwithVariabilityandUncertaintyinModelsandInputs.NewYork:PlenumPress.

Evans,J.S.,J.D.Graham,G.M.Gray,andR.L.Sielken.1994.“ADistributionalApproachtoCharacterizingLow‐DoseCancerRisk.”RiskAnalysis14(1):25–34.

Ferson,S.,R.Nelsen,J.Hajagos,D.Berleant,J.Zhang,W.T.Tucker,L.Ginzburg,andW.L.Oberkampf.2004.DependenceinProbabilisticModeling,Dempster‐ShaferTheory,andProbabilityBoundsAnalysis.SAND20043072.Albuquerque,NM:SandiaNationalLaboratories.www.ramas.com/depend.pdf.

Ferson,S.,W.L.Oberkampt,andL.Ginzburg.2009.“ValidationofImpreciseProbabilityModels.”InternationalJournalofReliabilityandSafety3(1–3):3–22.

Finkel,A.M.1989.“IsRiskAssessmentReallyTooConservative?:RevisingtheRevisionists.”JournalofEnvironmentalLaw14:427–67.

Fisher,L.,P.Pascual,andW.Wagner.2006.MappingtheRoleofModelsinU.S.andE.U.RiskRegulation:ALegalBackgroundPaper.Washington,D.C.:TheAssociationofUnionConstructorsWorkshop.

Fischhoff,B.1995.“RankingRisks.”Risk:Health,Safety&Environment6:191–202.

Frey,H.C.,andS.R.Patil.2002.“IdentificationandReviewofSensitivityAnalysisMethods.”RiskAnalysis22(3):553–78.

Fuentes,M.andK.Foley.2012.“EnsembleModels.”InAbdel,H.andWalterW.El‐Shaarawai(eds.)EncyclopediaofEnvironmetrics.Indianapolis:JohnWileyandSons.

Gelman,A.,J.B.Carlin,H.S.Stern,andD.B.Rubin.2004.BayesianDataAnalysis.2nded.BocaRaton,FL:ChapmanandHall/CRC.

Page 54: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

43

Hattis,D.,S.Baird,andR.Goble.2002.“AStrawManProposalforaQuantitativeDefinitionoftheRfD.”DrugandChemicalToxicology25(4):403–36.

Hattis,D.B.,andD.E.Burmaster.1994.“AssessmentofVariabilityandUncertaintyDistributionsforPracticalRiskAnalyses.”RiskAnalysis14(5):713–30.

Hattis,D.,andM.K.Lynch.2010.AlternativestoPollutant‐by‐PollutantDose‐ResponseEstimationforAirToxics(FinalReport).EPA‐W‐05‐022,WA3‐80.Washington,D.C.:USEPA.

Hoeting,J.A.,D.Madigan,A.E.Raftery,andC.T.Volinsky.1999.“BayesianModelAveraging:ATutorial.”StatisticalScience14(4):382–417.

IEC(IndustrialEconomics,Inc.).2006.ExpandedExpertJudgmentAssessmentoftheConcentration‐ResponseRelationshipBetweenPM2.5ExposureandMortality(FinalReport).ResearchTrianglePark,NC:USEPA.http://www.epa.gov/ttn/ecas/regdata/Uncertainty/pm_ee_report.pdf.

InstituteofMedicine.2013.EnvironmentalDecisionsintheFaceofUncertainty.Washington,D.C.:TheNationalAcademiesPress.

Illing,H.P.A.1999.“AreSocietalJudgmentsBeingIncorporatedIntotheUncertaintyFactorsUsedinToxicologicalRiskAssessment?”RegulatoryToxicologyandPharmacology29(3):300–308.

Krupnick,A.,R.Morgenstern,M.Batz,P.Nelson,D.Burtraw,J.Shih,andM.McWilliams.2006.NotaSureThing:MakingRegulatoryChoicesUnderUncertainty.Washington,D.C.:ResourcesfortheFuture.http://www.rff.org/rff/Documents/RFF‐Rpt‐RegulatoryChoices.pdf.

Kunreuther,H.C.,andP.Slovic.1996.“Science,ValuesandRisk.”AnnalsoftheAmericanAcademyofPoliticalandSocialScience545:116–25.

Lester,R.R.,L.C.Green,andI.Linkov.2007.“Site‐SpecificApplicationsofProbabilisticHealthRiskAssessment:ReviewoftheLiteratureSince2000.”RiskAnalysis27:635–58.

McConnell,K.E.1997.“UsingCost‐BenefitAnalysisintheManagementofContaminatedSediments(AppendixE).”InContaminatedSedimentsinPortsandWaterways:CleanupStrategiesandTechnologies,bytheCommissiononEngineeringandTechnicalSystemsoftheNationalResearchCouncil,239–56.Washington,D.C.:TheNationalAcademiesPress.

Mokhtari,A.,H.C.Frey,andJ.Zheng.2006.“EvaluationandRecommendationofSensitivityAnalysisMethodsforApplicationtoStochasticHumanExposureandDoseSimulation(SHEDS)Models.”JournalofExposureScienceandEnvironmentalEpidemiology16(6):491–506.

Morgan,G.,H.Dowlatabadi,M.Henrion,D.Keith,R.Lempert,S.McBrid,M.Small,andT.Wilbanks,eds.2009.BestPracticeApproachesforCharacterizing,Communicating,andIncorporatingScientificUncertaintyinDecisionMaking.FinalReport.Washington,D.C.:U.S.ClimateChangeScienceProgram,NationalOceanicandAtmosphericAdministration.http://library.globalchange.gov/sap‐5‐2‐best‐practice‐approaches‐for‐characterizing‐communicating‐and‐incorporating‐scientific‐uncertainty‐in‐decisionmaking.

Morgan,M.G.,andM.Henrion.1990.Uncertainty:AGuidetoDealingwithUncertaintyinQuantitativeRiskandPolicyAnalysis.NewYork:CambridgeUniversityPress.

NRC(NationalResearchCouncil).1983.RiskAssessmentintheFederalGovernment:ManagingtheProcess.CommitteeontheInstitutionalMeansforAssessmentofRiskstoPublicHealth.Washington,D.C.:TheNationalAcademiesPress.

NRC.1989.RiskAssessmentintheFederalGovernment:ManagingtheProcess.Washington,D.C.:TheNationalAcademiesPress.

Page 55: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

44

NRC.1991.RethinkingtheOzoneProbleminUrbanandRegionalAirPollution.Washington,D.C.:TheNationalAcademiesPress.

NRC.1993.IssuesinRiskAssessment.Washington,D.C.:TheNationalAcademiesPress.

NRC.1994.ScienceandJudgmentinRiskAssessment.Washington,D.C.:TheNationalAcademiesPress.

NRC.2002.EstimatingthePublicHealthBenefitsofProposedAirPollutionRegulations.Washington,D.C.:TheNationalAcademiesPress.

NRC.2007a.ScientificReviewoftheProposedRiskAssessmentBulletinFromtheOfficeofManagementandBudget.Washington,D.C.:TheNationalAcademiesPress.

NRC.2007b.ModelsinEnvironmentalRegulatoryDecisionMaking.Washington,D.C.:TheNationalAcademiesPress.

NRC.2009.ScienceandDecisions:AdvancingRiskAssessment.Washington,D.C.:TheNationalAcademiesPress.

OSTP(OfficeofScienceandTechnologyPolicy)/OMB(OfficeofManagementandBudget).2007.Memorandum(M‐07‐24)fortheHeadsofExecutiveDepartmentsandAgencies;fromSusanE.Dudley,Administrator,OfficeofInformationandRegulatoryAffairs,OfficeofManagementandBudget,toSharonL.Hays,AssociateDirectorandDeputyDirectorforScience,OfficeofScienceandTechnologyPolicy:UpdatedPrinciplesforRiskAnalysis.September19,2007.

ÖzkaynakH.,H.C.Frey,J.Burke,andR.W.Pinder.2009.“AnalysisofCoupledModelUncertaintiesinSource‐to‐DoseModelingofHumanExposurestoAmbientAirPollution:APM2.5Case‐Study.”AtmosphericEnvironment43:1641–49.

Parkin,R.T.,andM.G.Morgan.2007.ConsultationonEnhancingRiskAssessmentPracticesandUpdatingEPA’sExposureGuidelines.EPA/SAB‐07/003.Washington,D.C.:USEPA.USEPA.2007.LetterFromM.G.MorganandR.T.Parkin,ScienceAdvisoryBoard,toS.Johnson,U.S.EnvironmentalProtectionAgency.February28,2007.EPA/SAB‐07/003.http://yosemite.epa.gov/sab/sabproduct.nsf/55E1B2C78C6085EB8525729C00573A3E/$File/sab‐07‐003.pdf.

PCCRARM(Presidential/CongressionalCommissiononRiskAssessmentandRiskManagement).1997.FrameworkforEnvironmentalHealthRiskManagementFinalReport,Volume1.Washington,D.C.:PCCRARM.http://www.riskworld.com/nreports/1997/risk‐rpt/pdf/EPAJAN.PDF.

Pearl,J.2009.Causality:Models,Reasoning,andInference.2nded.NewYork:CambridgeUniversityPress.

Royall,R.M.1997.StatisticalEvidence:ALikelihoodParadigm.London:Chapman&Hall.

Saltelli,A.,S.Tarantola,F.Campolongo,andM.Rattoet.2004.SensitivityAnalysisinPractice.NewYork:JohnWiley&Sons.

Stahl,C.H.,andA.J.Cimorelli.2005.“HowMuchUncertaintyIsTooMuchandHowDoWeKnow?ACaseExampleoftheAssessmentofOzoneMonitorNetworkOptions.”RiskAnalysis25(5):1109–20.

Stahl,C.H.,andA.J.Cimorelli.2012.“TheDemonstrationoftheNecessityandFeasibilityofUsingaClumsyDecisionAnalyticApproachinWickedEnvironmentalProblems.”IntegratedEnvironmentalAssessmentandManagement9(1):17.

Page 56: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

45

Stahl,C.H.,A.J.Cimorelli,C.Mazzarella,andB.Jenkins.2011.“TowardSustainability:ACaseStudyDemonstratingTransdisciplinaryLearning.”IntegratedEnvironmentalAssessmentandManagement7(3):483–98.

Swartout,J.C.,P.S.Price,M.L.Dourson,H.L.Carlson‐Lynch,andR.E.Keenan.1998.“ProbabilisticFrameworkfortheReferenceDose(ProbabilisticRfD).”RiskAnalysis18(3):271–82.

Toll,J.,S.Pavlou,D.Lee,L.Zaragosa,andP.Shelley.1997.“UsingDecisionAnalysisintheManagementofContaminatedSediments(AppendixE).”InContaminatedSedimentsinPortsandWaterways:CleanupStrategiesandTechnologies,bytheCommissiononEngineeringandTechnicalSystemsoftheNationalResearchCouncil,257–84.Washington,D.C.:TheNationalAcademiesPress.

Toll,J.E.1999.“ElementsofEnvironmentalProblem‐Solving.”HumanandEcologicalRiskAssessment5(2):275–80.

USEPA(U.S.EnvironmentalProtectionAgency).1992.GuidelinesforExposureAssessment.EPA/600/Z‐92/001.Washington,D.C.:RiskAssessmentForum,USEPA.http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=15263.

USEPA.1995a.GuidanceforRiskCharacterization.Washington,D.C.:SciencePolicyCouncil,USEPA.http://www.epa.gov/spc/pdfs/rcguide.pdf.

USEPA.1995b.PolicyforRiskCharacterization.Washington,D.C.:USEPA.http://www.epa.gov/oswer/riskassessment/pdf/1995_0521_risk_characterization_program.pdf.

USEPA.1997a.PolicyforUseofProbabilisticAnalysisinRiskAssessmentattheU.S.EnvironmentalProtectionAgency.Washington,D.C.:USEPA.http://www.epa.gov/spc/pdfs/probpol.pdf.

USEPA.1997b.GuidingPrinciplesforMonteCarloAnalysis.EPA/630/R‐97/001.Washington,D.C.:USEPA.http://www.epa.gov/raf/publications/pdfs/montecar.pdf.

USEPA.1998a.GuidanceforSubmissionofProbabilisticHumanHealthExposureAssessmentstotheOfficeofPesticidePrograms(Draft).Washington,D.C.:USEPA.http://www.epa.gov/fedrgstr/EPA‐PEST/1998/November/Day‐05/6021.pdf.

USEPA.1998b.GuidelinesforEcologicalRiskAssessment.EPA/630/R‐95/002F.Washington,D.C.:RiskAssessmentForum,USEPA.http://www.epa.gov/raf/publications/pdfs/ECOTXTBX.PDF.

USEPA.2000a.EPASciencePolicyCouncilHandbook:RiskCharacterization.EPA/100/B‐00/002.Washington,D.C.:USEPA.http://www.epa.gov/spc/pdfs/rchandbk.pdf.

USEPA.2000b.TowardIntegratingEnvironmentalDecision‐Making.EPA/SAB/EC‐00/011.Washington,D.C.:ScienceAdvisoryBoard,USEPA.

USEPA.2001.RiskAssessmentGuidanceforSuperfund:VolumeIII—PartA,ProcessforConductingProbabilisticRiskAssessment.EPA/540/R‐02/002.Washington,D.C.:USEPA.http://www.epa.gov/oswer/riskassessment/rags3adt/pdf/rags3adt_complete.pdf.

USEPA.2004a.AirToxicsRiskAssessmentReferenceLibrary.VolumeI:TechnicalResourceManual.EPA/453/K‐04/001A.ResearchTrianglePark,NC:USEPA.http://www.epa.gov/ttn/fera/data/risk/vol_1/title_page_volume_1.pdf.

USEPA.2004b.AnExaminationofEPARiskAssessmentPrinciplesandPractices.EPA/100/B‐04/001.Washington,D.C.:USEPA.http://www.epa.gov/osa/pdfs/ratf‐final.pdf.

USEPA.2006a.AFrameworkforAssessingHealthRisksofEnvironmentalExposurestoChildren.EPA/600/R‐05/093F.Washington,D.C.:USEPA.http://ofmpub.epa.gov/eims/eimscomm.getfile?p_download_id=459047.

Page 57: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

46

USEPA.2006b.AirToxicsRiskAssessmentReferenceLibrary,Volume3:Community‐ScaleAssessment.EPA/452/K‐06/001C.Washington,D.C.:USEPA.http://www.epa.gov/ttn/fera/risk_atra_vol3.html.

USEPA.2006c.HazardAnalysisCriticalControlPointStrategiesforDistributionSystemMonitoring,HazardAssessmentandControl.Washington,D.C.:USEPA.http://www.epa.gov/ogwdw/disinfection/tcr/pdfs/issuepaper_tcr_haccp‐strategies.pdf.

USEPA.2007a.Concepts,MethodsandDataSourcesforCumulativeHealthRiskAssessmentofMultipleChemicals,ExposuresandEffects:AResourceDocument.EPA/600/R‐06/013F.Washington,D.C.:USEPA.http://cfpub.epa.gov/ncea/risk/recordisplay.cfm?deid=190187.

USEPA.2007b.LetterFromM.G.MorganandR.T.Parkin,ScienceAdvisoryBoard,toS.Johnson,U.S.EnvironmentalProtectionAgency.February28,2007.EPA/SAB‐07/003.http://yosemite.epa.gov/sab/sabproduct.nsf/55E1B2C78C6085EB8525729C00573A3E/$File/sab‐07‐003.pdf.

USEPA.2010a.EnvironmentalMonitoringandAssessmentProgramMasterGlossary.LastmodifiedNovember8.http://www.epa.gov/emap

2/html/pubs/docs/resdocs/mglossary.html#ee.

USEPA.2010b.HumanHealthRiskAssessmentOnondagaLake,LakeBottomSubsite:SedimentConsolidationArea,Camillus,NY.Washington,D.C.:USEPA.http://www.epa.gov/region2/superfund/npl/onondagalake/healthriskassess.pdf.

USEPA.2011a.ExpertElicitationTaskForceWhitePaper.Washington,D.C.:USEPA.http://www.epa.gov/stpc/pdfs/ee‐white‐paper‐final.pdf.

USEPA.2011b.ExposureFactorsHandbook:2011Edition.EPA/600/R‐09/052F.Washington,D.C.:NationalCenterforEnvironmentalAssessment,OfficeofResearchandDevelopment,USEPA.http://cfpub.epa.gov/ncea/risk/recordisplay.cfm?deid=236252.

USEPA.2011c.IntegratedRiskInformationSystemGlossary.LastmodifiedAugust21.http://ofmpub.epa.gov/sor_internet/registry/termreg/searchandretrieve/glossariesandkeywordlists/search.do?details=&glossaryName=IRIS%20Glossary.

USEPA.2012.WasteandCleanupRiskAssessmentGlossary.OfficeofSolidWasteandEmergencyResponse,USEPA.LastmodifiedDecember24.http://www.epa.gov/oswer/riskassessment/glossary.htm.

USEPA.2014a.ProbabilisticRiskAssessmenttoInformDecisionMaking:FrequentlyAskedQuestions.EPA/100/R‐09/001B.Washington,D.C.:RiskAssessmentForum,OfficeoftheScienceAdvisor,USEPA.

USEPA.2014b.FrameworkforHumanHealthRiskAssessmenttoInformDecisionMaking.EPA/100/R‐14/001.Washington,D.C.:RiskAssessmentForum,OfficeoftheScienceAdvisor,USEPA.

Wilson,J.D.2000.“RiskManagementCannotAbideUncertainty.”InJ.H.ExnerandM.L.Trehy(eds.)SymposiapaperspresentedbeforetheDivisionofEnvironmentalChemistry,AmericanChemicalSociety,August20–24,2000.PreprintsofExtendedAbstracts40(2):268–69.Washington,D.C.:AmericanChemicalSociety.

WHO(WorldHealthOrganization).2004.InternationalProgrammeonChemicalSafetyRiskAssessmentTerminology.Geneva:WHO.http://www.who.int/ipcs/methods/harmonization/areas/ipcsterminologyparts1and2.pdf.

Page 58: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

47

WHO.2008.UncertaintyandDataQualityinExposureAssessment.Geneva:WHO.http://www.who.int/ipcs/publications/methods/harmonization/exposure_assessment.pdf.

Wright,G.,F.Bolger,andG.Rowe.2002.“AnEmpiricalTestoftheRelativeValidityofExpertandLayJudgmentsofRisk.”RiskAnalysis22(6):1107–22.

Page 59: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

48

BIBLIOGRAPHY

Probabilistic Risk Analysis MethodologyGeneral Cullen,A.C.,andH.C.Frey.1999.TheUseofProbabilisticTechniquesinExposureAssessment:AHandbookforDealingwithVariabilityandUncertaintyinModelsandInputs.NewYork:Plenum.

Dakins,M.E.,J.E.Toll,M.J.Small,andK.P.Brand.1996.“Risk‐BasedEnvironmentalRemediation:BayesianMonteCarloAnalysisandtheExpectedValueofSampleInformation.”RiskAnalysis16(1):67–79.

Ferson,S.,andV.Kreinovich.2006.ModelingCorrelationandDependenceAmongIntervals.UniversityofTexasatElPasoDepartmentalTechnicalReports(CS),Paper131.http://digitalcommons.utep.edu/cs_techrep/131.

Finkel,A.M.1990.ConfrontingUncertaintyinRiskManagement:AGuideforDecision‐Makers.Washington,D.C.:ResourcesfortheFuture.

Frey,H.C.,D.Crawford‐Brown,J.Zheng,andD.Loughlin.2003.HierarchyofMethodstoCharacterizeUncertainty:StateofScienceofMethodsforDescribingandQuantifyingUncertainty(Draft).ResearchTrianglePark,NC:USEPA.

Frey,C.,J.Penman,L.Hanle,S.Monni,andS.Ogle.2006.Chapter3,“Uncertainties.”InS.Eggleston,L.Buendia,K.Miwa,T.Ngara,andK.Tanabe(eds.)2006IPCCGuidelinesforNationalGreenhouseGasInventories,VolumeI:GeneralGuidanceandReporting.Kanagawa,Japan:NationalGreenhouseGasInventoriesProgramme,Inter‐GovernmentalPanelonClimateChange.

Hattis,D.1996.“HumanInterindividualVariabilityinSusceptibilitytoToxicEffects—FromAnnoyingDetailtoaCentralDeterminantofRisk.”Toxicology111:5–14.

Hattis,D.,andE.Anderson.1999.“WhatShouldBetheImplicationsofUncertainty,Variability,andInherent‘Biases’/‘Conservatism’forRiskManagementDecisionMaking?”RiskAnalysis19:95–107.

Hattis,D.,andM.K.Lynch.2007.“EmpiricallyObservedDistributionsofPharmacokineticandPharmacodynamicVariabilityinHumans—ImplicationsfortheDerivationofSinglePointComponentUncertaintyFactorsProvidingEquivalentProtectionasExistingRfDs.”InJ.C.LipscombandE.V.Ohanian(eds.)ToxicokineticsinRiskAssessment,69–93.NewYork:InformaHealthcareUSA,Inc.

Hattis,D.2009.“FailuretoCommunicate—Comment:Regulator/RiskPerspectiveonBurzalaandMazzuchi’sPaper.”InR.M.Cooke(ed.).UncertaintyModelinginDoseResponse:BenchTestingEnvironmentalToxicity,153–59.NewYork:JohnWiley&Sons,Inc.

Loughlin,D.,H.C.Frey,K.Hanisak,andA.Eyth.2003.ImplementationRequirementsfortheDevelopmentofaSensitivity/UncertaintyAnalysisToolforMIMS(Draft).ResearchTrianglePark,NC:USEPA.

Morgan,M.G.,andM.Henrion.1990.Uncertainty:AGuidetoDealingwithUncertaintyinQuantitativeRiskandPolicyAnalysis.NewYork:CambridgeUniversityPress.

NARSTO(NorthAmericanResearchStrategyforTroposphericOzone).2005.ImprovingEmissionInventoriesforEffectiveAirQualityManagementAcrossNorthAmerica,ANARSTOAssessment.NARSTO‐05‐001.Pasco,WA:NARSTO.

NRC(NationalResearchCouncil).1994.ScienceandJudgmentinRiskAssessment.Washington,D.C.:TheNationalAcademiesPress.

Page 60: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

49

USEPA(U.S.EnvironmentalProtectionAgency).1997.GuidingPrinciplesforMonteCarloAnalysis.EPA/630/R‐97/001.Washington,D.C.:USEPA.http://www.epa.gov/raf/publications/pdfs/montecar.pdf.

Probabilistic Risk Analysis and Decision Making Bernardini,A.,andF.Tonon.2010.BoundingUncertaintyinCivilEngineering:TheoreticalBackground.NewYork:Springer.

Bloom,D.L.,D.M.Byrne,andJ.M.Andreson.1993.CommunicatingRisktoSeniorEPAPolicy‐Makers:AFocusGroupStudy.ResearchTrianglePark,NC:USEPA.

Krupnick,A.,R.Morgenstern,M.Batz,P.Nelson,D.Burtraw,J.‐S.Shih,andM.McWilliams.2006.NotaSureThing:MakingRegulatoryChoicesUnderUncertainty.Washington,D.C.:ResourcesfortheFuture.

McGarity,T.O.,andW.E.Wagner.2003.“LegalAspectsoftheRegulatoryUseofEnvironmentalModeling.”EnvironmentalLawReporter33:10751–74.

NRC.1996.UnderstandingRisk:InformingDecisionsinaDemocraticSociety.Washington,D.C.:TheNationalAcademiesPress.

NRC.2009.ScienceandDecisions:AdvancingRiskAssessment.Washington,D.C.:TheNationalAcademiesPress.

Pascual,P.,N.Steiber,andE.Sunderland.2003.DraftGuidanceontheDevelopment,Evaluation,andApplicationofRegulatoryEnvironmentalModels.Washington,D.C.:USEPA.

Thompson,K.M.,andJ.D.Graham.1996.“GoingBeyondtheSingleNumber:UsingProbabilisticRiskAssessmenttoImproveRiskManagement.”HumanandEcologicalRiskAssessment2(4):1008–34.

USEPA.1997.MemofromFredHansen.UseofProbabilisticTechniques(includingMonteCarloAnalysis)inRiskAssessment,datedMay15,1997.Washington,D.C.:USEPA.

Vick,S.G.2002.DegreesofBelief:SubjectiveProbabilityandEngineeringJudgment.Reston,VA:ASCEPress.

Probabilistic Risk Analysis MethodologySpecific Aspects Abdel‐Aziz,A.,andH.C.Frey.2003.“DevelopmentofHourlyProbabilisticUtilityNOxEmissionInventoriesUsingTimeSeriesTechniques:PartI—UnivariateApproach.”AtmosphericEnvironment37:5379–89.

Abdel‐Aziz,A.,andH.C.Frey.2003.“DevelopmentofHourlyProbabilisticUtilityNOxEmissionInventoriesUsingTimeSeriesTechniques:PartII—MultivariateApproach.”AtmosphericEnvironment37:5391–401.

Abdel‐Aziz,A.,andH.C.Frey.2004.“PropagationofUncertaintyinHourlyUtilityNOxEmissionsThroughaPhotochemicalGridAirQualityModel:ACaseStudyfortheCharlotte,NCModelingDomain.”EnvironmentalScience&Technology38(7):2153–60.

Casman,E.A.,M.G.Morgan,andH.Dowlatabadi.1999.“MixedLevelsofUncertaintyinComplexPolicyModels.”RiskAnalysis19(1):33–42.

Page 61: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

50

Crawford‐Brown,D.J.,andK.G.Brown.1987.“AFrameworkforAssessingtheRationalityofJudgmentsinCarcinogenicityHazardIdentification.”Risk:Health,SafetyandEnvironment8:307–37.

Draper,D.1995.“AssessmentandPropagationofModelUncertainty.”JournaloftheRoyalStatisticalSociety57:45–97.

Evans,J.S.,J.D.Graham,G.M.Gray,andR.L.Sielken.1994.“ADistributionalApproachtoCharacterizingLow‐DoseCancerRisk.”RiskAnalysis14(1):25–34.

Frey,H.C.,andD.E.Burmaster.1999.“MethodsforCharacterizingVariabilityandUncertainty:ComparisonofBootstrapSimulationandLikelihood‐BasedApproaches.”RiskAnalysis19(1):109–30.

Frey,H.C.,andD.S.Rhodes.1996.“Characterizing,Simulating,andAnalyzingVariabilityandUncertainty:AnIllustrationofMethodsUsinganAirToxicsEmissionsExample.”HumanandEcologicalRiskAssessment:AnInternationalJournal2(4):762–97.

Frey,H.C.,andY.Zhao.2004.“QuantificationofVariabilityandUncertaintyforAirToxicEmissionInventoriesWithCensoredEmissionFactorData.”EnvironmentalScience&Technology38(22):6094–100.

Landis,W.G.2005.RegionalScaleEcologicalRiskAssessmentUsingtheRelativeRiskModel.BocaRaton,FL:CRCPress.

Maginnis,C.M.2006.AScreeningLevelIntegratedEcologicalandHumanHealthRiskAssessmentforLakeWhatcom,WhatcomCounty,Washington.Master’sThesis,WesternWashingtonUniversity.

Shlyakhter,A.I.1994.“AnImprovedFrameworkforUncertaintyAnalysis:AccountingforUnsuspectedErrors.”RiskAnalysis14:441–47.

Zheng,J.,andH.C.Frey.2004.“QuantificationofVariabilityandUncertaintyUsingMixtureDistributions:EvaluationofSampleSize,MixingWeightsandSeparationbetweenComponents.”RiskAnalysis24(3):553–71.

Zheng,J.,andH.C.Frey.2005.“QuantitativeAnalysisofVariabilityandUncertaintywithKnownMeasurementError:MethodologyandCaseStudy.”RiskAnalysis25(3):663–76.

Sensitivity Analysis Agro,K.E.,C.A.Bradley,andN.Mittmann.1997.“SensitivityAnalysisinHealthEconomicandPharmacoeconomicStudies—anAppraisaloftheLiterature.”Pharmacoeconomics11(1):75–88.

Cukier,R.I.,H.B.Levine,andK.E.Shuler.1978.“NonlinearSensitivityAnalysisofMulti‐parameterModelSystems.”JournalofChemicalPhysics26(1):1–42.

Frey,H.C.,andR.Patil.2002.“IdentificationandReviewofSensitivityAnalysisMethods.”RiskAnalysis22(3):553–77.

Frey,H.C.,A.Mokhtari,andJ.Zheng.2004.RecommendedPracticeRegardingSelection,Application,andInterpretationofSensitivityAnalysisMethodsAppliedtoFoodSafetyProcessRiskModels.Washington,D.C.:U.S.DepartmentofAgriculture.http://www.ce.ncsu.edu/risk/Phase3Final.pdf

Helton,J.C.,andF.J.Davis.2002.“IllustrationofSampling‐BasedMethodsforSensitivityAnalysis.”ReliabilityEngineering&SystemSafety81(1):23–69.

Page 62: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

51

Mokhtari,A.,andH.C.Frey.2005.“RecommendedPracticeRegardingSelectionofSensitivityAnalysisMethodsAppliedtoMicrobialFoodSafetyProcessRiskModels.”HumanandEcologicalRiskAssessment11(3):591–605.

Mokhtari,A.,H.C.Frey,andJ.Zheng.2006.“EvaluationandRecommendationofSensitivityAnalysisMethodsforApplicationtoEPA‐StochasticHumanExposureandDoseSimulation(SHEDS)Models.”JournalofExposureScienceandEnvironmentalEpidemiology16(6):491–506.

Saltelli,A.,K.Chan,andM.Scott.2000.SensitivityAnalysis,ProbabilityandStatisticsSeries.NewYork:JohnWiley&Sons.

Sobol,I.M.1993.“SensitivityEstimatesforNonlinearMathematicalModels.”MathematicalModelingandComputation1(4):407–14.

Xue,J.,V.G.Zartarian,H.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.2006.“AProbabilisticArsenicExposureAssessmentforChildrenWhoContactChromatedCopperArsenate(CCA)‐TreatedPlaysetsandDecks,Part2:SensitivityandUncertaintyAnalyses.”RiskAnalysis26(2):533–41.

Case Study Examples of Probabilistic Risk AnalysisEPA (See Also the PRA Case Studies Appendix) Babendreier,J.E.,andK.L.Castleton.2005.“InvestigatingUncertaintyandSensitivityinIntegratedMultimediaEnvironmentalModels:ToolsforFRAMES‐3MRA.”EnvironmentalModeling&Software20:1043–55.

Blancato,J.N.,F.W.Power,R.N.Brown,andC.C.Dary.2006.ExposureRelatedDoseEstimatingModel(ERDEM):APhysiologically‐BasedPharmacokineticandPharmacodynamic(PBPK/PD)ModelforAssessingHumanExposureandRisk.EPA/600/R‐06/061.ResearchTrianglePark,NC:USEPA.http://www.epa.gov/heasd/products/erdem/237edrb05‐Report.pdf

Burke,J.2005.EPAStochasticHumanExposureandDoseSimulationforParticulateMatter(SHEDS‐PM)User’sGuide.ResearchTrianglePark,NC:USEPA.

Frey,H.C.2003.EvaluationofanApproximateAnalyticalProcedureforCalculatingUncertaintyintheGreenhouseGasVersionoftheMulti‐ScaleMotorVehicleandEquipmentEmissionsSystem.AnnArbor,MI:USEPA.

Frey,H.C.,andY.Zhao.2003.DevelopmentofProbabilisticEmissionInventoriesofBenzene,FormaldehydeandChromiumfortheHoustonDomain.ResearchTrianglePark,NC:USEPA.

Frey,H.C.,R.Bharvirkar,andJ.Zheng.1999.QuantitativeAnalysisofVariabilityandUncertaintyinEmissionsEstimation.ResearchTrianglePark,NC:USEPA.

Frey,H.C.,J.Zheng,Y.Zhao,S.Li,andY.Zhu.2002.TechnicalDocumentationoftheAuvToolSoftwareforAnalysisofVariabilityandUncertainty.ResearchTrianglePark,NC:USEPA.

ÖzkaynakH.,H.C.Frey,J.Burke,andR.W.Pinder.2009.“AnalysisofCoupledModelUncertaintiesinSource‐to‐DoseModelingofHumanExposurestoAmbientAirPollution:APM2.5Case‐Study.”AtmosphericEnvironment43:1641–49.

Zartarian,V.G.,J.Xue,H.A.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.2005.AProbabilisticExposureAssessmentforChildrenWhoContactCCA‐TreatedPlaysetsandDecksUsingtheStochasticHumanExposureandDoseSimulationModelfortheWoodPreservativeScenario(SHEDS‐WOOD),FinalReport.EPA/600/X‐05/009.Washington,D.C.:USEPA.http://www.epa.gov/heasd/sheds/CCA_all.pdf

Page 63: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

52

Zartarian,V.G.,J.Xue,H.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.2006.“AProbabilisticArsenicExposureAssessmentforChildrenWhoContactCCA‐TreatedPlaysetsandDecks,Part1:ModelMethodology,VariabilityResults,andModelEvaluation.”RiskAnalysis26(2):515–32.

Zheng,J.,andH.C.Frey.2002.AuvToolUser’sGuide.ResearchTrianglePark,NC:USEPA.

Case Study Examples of Probabilistic Risk AnalysisOther

Bloyd,C.,J.Camp,G.Conzelmann,J.Formento,J.Molburg,J.Shannon,M.Henrion,etal.1996.TrackingandAnalysisFramework(TAF)ModelDocumentationandUser’sGuide.ANL/DIS/TM‐36.Washington,D.C.:ArgonneNationalLaboratory.

Frey,H.C.,andS.Li.2003.“QuantificationofVariabilityandUncertaintyinAP‐42EmissionFactors:CaseStudiesforNaturalGas‐FueledEngines.”JournaloftheAir&WasteManagementAssociation53(12):1436–47.

Hanna,S.R.,Z.Lu,H.C.Frey,N.Wheeler,J.Vukovich,S.Arunachalam,M.Fernau,andD.A.Hansen.2001.“UncertaintiesinPredictedOzoneConcentrationsDuetoInputUncertaintiesfortheUAM‐VPhotochemicalGridModelAppliedtotheJuly1995OTAGDomain.”AtmosphericEnvironment35(5):891–903.

Munns,W.R.,Jr.,H.A.Walker,andJ.F.Paul.1989.“AnEcologicalRiskAssessmentFrameworkforExaminingtheImpactsofOceanicDisposalStatusandTrendsProgram.”InOceans‘89,Volume2:OceanPollution,bytheMarineTechnologySocietyandtheOceanicEngineeringSociety(U.S.),664–69.IEEEPublicationNo.89,CH2780‐5.NewYork:IEEEPress.

Munns,W.R.,Jr.,H.A.Walker,J.F.Paul,andJ.H.Gentile.1996.“AProspectiveAssessmentofEcologicalRiskstoUpperWaterColumnPopulationsfromOceanDisposalatthe106‐MileDumpsite.”JournalofMarineEnvironmentalEngineering3:279–97.

Nocito,J.A.,H.A.Walker,J.F.Paul,andC.Menzie.1988.“ApplicationofaRiskAssessmentforMarineDisposalofSewageSludgeatMidshelfandOffshelfSites.”InProceedingsofthe7thInternationalOceanDisposalSymposium,644–63.Ottawa:EnvironmentCanada.

Nocito,J.A.,H.A.Walker,J.F.Paul,andC.A.Menzie.1989.“ApplicationofaRiskAssessmentFrameworkforMarineDisposalofSewageSludgeatMidshelfandOffshelfSites.”InG.W.SuterIIandM.A.Lewis(eds.).AquaticToxicologyandEnvironmentalFate:EleventhVolume,101–20.ASTMSTP1007.Philadelphia,PA:AmericanSocietyforTestingandMaterials.

Paul,J.F.,H.A.Walker,andV.J.Bierman,Jr.1983.“ProbabilisticApproachfortheDeterminationofthePotentialAreaofInfluenceforWasteDisposedatthe106‐MileOceanDisposalSite.”InJ.B.Pearce,D.C.Miller,andC.Berman(eds.).106‐MileWasteDisposalSiteCharacterizationUpdateReport,A–1toA–8.NOAATechnicalMemorandumNMFS‐F/NEC‐26.WoodsHole,MA:NationalMarineFisheriesService.

Paul,J.F.,V.J.Bierman,Jr.,W.R.Davis,G.L.Hoffman,W.R.Munns,C.E.Pesch,P.F.Rogerson,andS.C.Schimmel.1988.“TheApplicationofaHazardAssessmentResearchStrategytotheOceanDisposalofaDredgedMaterial:ExposureAssessmentComponent.”InD.A.WolfeandT.P.O’Connor(eds.).OceanicProcessesinMarinePollution,Volume5,123–35.Melbourne,FL:KreigerPublishingCo.

Prager,J.C.,V.J.Bierman,Jr.,J.F.Paul,andJ.S.Bonner.1986.“SamplingtheOceansforPollution:ARiskAssessmentApproachtoEvaluatingLow‐LevelRadioactiveWasteDisposalatSea.”DangerousPropertiesofIndustrialMaterialsReport6(3):2–26.

Page 64: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

53

Walker,H.A.,J.F.Paul,J.A.Nocito,andJ.H.Gentile.1988.“EcologicalandHumanHealthRisksforSewageSludgeDisposalatthe106‐MileSite.”InProceedingsofthe7thInternationalOceanDisposalSymposium,625–43.Ottawa:EnvironmentCanada.

Page 65: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

54

APPENDIX: CASE STUDY EXAMPLES OF THE APPLICATION OF PROBABILISTIC RISK ANALYSIS IN U.S.

ENVIRONMENTAL PROTECTION AGENCY REGULATORY DECISION MAKING

Page 66: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

55

A. OVERVIEW ThisAppendixfocusesonexamplesofhowprobabilisticriskanalysis(PRA)approacheshavebeenusedatEPAtoinformregulatorydecisions.TheAppendixwaspreparedbyrepresentativesfromvariousEPAprogramofficesandregionscurrentlyinvolvedinthedevelopmentandapplicationofPRAtechniques.TheTechnicalPanelselectedthecasestudyexamplesbasedonthemembers’knowledgeofthespecificPRAprocedures,thetypesoftechniquesdemonstrated,theavailabilitytothereaderthroughtheInternetandtheconditionofhavingbeenpeerreviewed;theyalsowereselectedtobeillustrativeofaspectrumofPRAusedatEPA.ThecasestudiesarenotdesignedtoprovideanexhaustivediscussionofthewidevarietyofapplicationsofPRAusedwithintheAgency,buttohighlightspecificexamplesreflectingtherangeofapproachescurrentlyappliedwithinEPA.

ThisAppendixisintendedtoserveasaresourceformanagersfacedwithdecisionsregardingwhentoapplyPRAtechniquestoinformenvironmentaldecisions,andforexposureandriskassessorswhomaynotbefamiliarwiththewidevarietyofavailablePRAapproaches.ThedocumentoutlinescategoriesofPRAsclassifiedbythecomplexityofanalysistoaidthedecision‐makingprocess.ThisapproachidentifiesvariousPRAtools,whichincludetechniquesrangingfromasimplesensitivityanalysis(e.g.,identificationofkeyexposureparametersordatavisualization)requiringlimitedtime,resourcesandexpertisetodevelop(Group1);toprobabilisticapproaches,includingMonteCarloanalysis,thatprovidetoolsforevaluatingvariabilityanduncertaintyseparatelyandthatrequiremoreresourcesandspecializedexpertise(Group2);tosophisticatedtechniquesofexpertelicitationthatgenerallyrequiresignificantinvestmentofemployeetime,additionalexpertiseandexternalpeerreview(Group3).

ThecasestudiesinthisAppendixusedPRAtechniqueswithinthisrankedframeworktoprovideadditionalinformationformanagers.ThecasestudysummariesareprovidedinaformatdesignedtohighlighthowtheresultsofthePRAswereconsideredindecisionmaking.Thesesummariesincludespecificinformationontheconductoftheanalysesasanaidindeterminingwhattoolsmightbeappropriatetodevelopspecificexposureorriskassessmentsforothersites.

Thecasestudiesrangefromexamplesoflessresource‐intensiveanalyses,whichmightassistinidentifyingkeyexposureparametersortheneedformoredata,tomoredetailedandresource‐intensiveapproaches.Examplesofapplicationsinhumanhealthandecologicalriskassessmentincludetheexposureofchildrentochromatedcopperarsenate(CCA)‐treatedwood,therelationbetweenparticulatesinairandhealth,dietaryexposurestopesticides,modelingsealevelchange,samplingwatersheds,andmodelingbirdandanimalexposures.

B. INTRODUCTION Historically,EPAhasuseddeterministicriskassessments,orpointestimatesofrisk,toevaluatecancerrisksandnoncancerhealthhazardstohigh‐endexposedindividuals(90thpercentileorhigher)andtheaverageexposedindividual(50thpercentile)and,whereappropriate,risksandhazardstopopulations,asrequiredbyspecificenvironmentallaws(USEPA1992a).Theuseofdefaultvaluesforexposureparametersinriskassessmentsprovidesaproceduralconsistencythatallowsriskassessmentstobefeasibleandtractable(USEPA2004).ThemethodstypicallyusedinEPAdeterministicriskassessments(DRA)relyonacombinationofpointvalues―someconservativeandsometypical―yieldingapointestimateofexposurethatisatsomeunknownpointintherangeofpossiblerisks(USEPA2004).

ThisAppendixpresentscasestudiesofPRAconductedbyEPAoverthepast10to15years.TableA‐1summarizesthecasestudiesbytitle,techniquedemonstrated,classificationasahumanhealthriskassessment(HHRA)orecologicalriskassessment(ERA),andtheprogramorregional

Page 67: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

56

officeresponsiblefordevelopingthecasestudies.ThisAppendix,providesa“snapshot”oftheutilizationofPRAacrossvariousprogramsinEPA.

C. OVERALL APPROACH TO PROBABILISTIC RISK ANALYSIS AT THE U.S. ENVIRONMENTAL PROTECTION AGENCY

C.1. U.S. Environmental Protection Agency Guidance and Policies on Probabilistic Risk Analysis

ThecasestudiespresentedherebuildontheprinciplesofPRAoutlinedinEPA’s1997PolicyforUseofProbabilisticAnalysisinRiskAssessmentattheU.S.EnvironmentalProtectionAgency(USEPA1997a)andGuidingPrinciplesforMonteCarloAnalysis(USEPA1997b),aswellassubsequentguidancedocumentsondevelopingandusingPRA.GuidancehasbeendevelopedfortheAgencyandindividualprograms.SpecificdocumentsthatrefertotheuseofPRAincludetheRiskAssessmentGuidanceforSuperfund:VolumeIII(USEPA2001);RiskAssessmentForum(RAF)FrameworkforEcologicalRiskAssessment(USEPA1992b);GuidelinesforEcologicalRiskAssessment(USEPA1998);GuidanceforRiskCharacterization(USEPA1995a);PolicyonEvaluatingHealthRiskstoChildren(USEPA1995b);PolicyforUseofProbabilisticAnalysisinRiskAssessment(USEPA1997a);GuidanceonCumulativeRiskAssessment,Part1:PlanningandScoping(USEPA1997c);andRiskCharacterizationHandbook(USEPA2000a);andFrameworkforHumanHealthRiskAssessmenttoInformDecisionMaking(USEPA2014).

Asshownintheindividualcasestudies,therangeandscopeofthePRAwilldependontheoverallobjectivesofthedecisionthattheanalysiswillinform.TheGuidingPrinciplesforMonteCarloAnalysis(USEPA1997b)layoutthegeneralapproachthatshouldbetakeninallcases,beginningwithdefiningtheproblemandscopeoftheassessmenttoselectingthebesttoolsandapproach.TheGuidingPrinciplesalsodescribetheprocessofestimatingandcharacterizingvariabilityanduncertaintyaroundriskestimates.StahlandCimorelli(2005)andtheRiskAssessmentGuidanceforSuperfund:VolumeIII(USEPA2001)highlighttheimportanceofcommunicationbetweentheriskassessorandmanager.StahlandCimorelli(2005)andJamieson(1996)indicatethatitisimportanttodeterminewhetheraparticularlevelofuncertaintyisacceptableornot.Theauthorsalsosuggestthatthisdecisiondependsoncontext,valuesandregulatorypolicy.TheRiskAssessmentGuidanceforSuperfund:VolumeIII(Chapter2andAppendixFinUSEPA2001)describesaprocessfordeterminingtheappropriatelevelofPRAusingarankedapproachfromthelessresource‐andtime‐intensiveapproachestomoresophisticatedanalyses.Furthermore,theRiskAssessmentGuidanceforSuperfund:VolumeIIIoutlinesaprocessfordevelopingaPRAworkplanandachecklistforPRAreviewers(Chapter2andAppendixFinUSEPA2001).ThisguidancealsoprovidesinformationregardinghowtocommunicatePRAresultstodecisionmakersandstakeholders(Chapter6inUSEPA2001).

C.2. Categorizing Case Studies

Therankedapproachusedforcategorizationisaprocessforasystematic,informedprogressiontoincreasinglycomplexriskassessmentmethodsofPRA,whichisoutlinedintheRiskAssessmentGuidanceforSuperfund(USEPA2001).TheuseofcategoriesprovidesaframeworkforevaluatingthevarioustechniquesofPRA.Highercategoriesreflectincreasingcomplexityandoftenwillrequiremoretimeandresources.Highercategoriesalsoreflectincreasingcharacterizationofvariabilityanduncertaintyintheriskestimate,whichmaybeimportantformakingspecificriskmanagementdecisions.Centraltotheapproachisasystematic,informedprogressionusingan

Page 68: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

57

iterativeprocessofevaluation,deliberation,datacollection,planningandscoping,development,andupdatestotheworkplanandcommunication.Allofthesestepsfocusondeciding:

1. Whetherornottheriskassessment,initscurrentstate(e.g.,DRA)issufficienttosupportdecisions(i.e.,aclearpathtoexitingtheprocessisavailableateachstep).

2. Iftheassessmentisdeterminedtobeinsufficient,whetherornotprogressiontoahigherlevelofcomplexity(orrefinementofthecurrentanalyses)wouldprovideasufficientbenefittowarranttheadditionaleffortofperformingaPRA.

ThisAppendixgroupscasestudiesaccordingtolevelofeffortandcomplexityoftheanalysisandtheincreasingsophisticationofthemethodsused(TableA‐1).Althougheachgroupgenerallyrepresentsincreasingeffortandcost,thismaynotalwaysbetrue.Thegroupsalsoareintendedtoreflecttheprogressionfromsimpletocomplexanalysisthatisdeterminedbytheinteractiveplanningandscopingeffortsoftheriskassessorsandmanagers.Theuseofparticulartermstodescribethegroups,including“tiers,”wasavoidedduetospecificprogrammaticandregulatoryconnotations.

Group 1 Case Studies Assessmentswithinthisgrouptypicallyinvolveasensitivityanalysisandserveasaninitialscreeningstepintheriskassessment.Sensitivityanalysesidentifyimportantparametersintheassessmentwhereadditionalinvestigationmaybehelpful(KurowickaandCooke2006).Sensitivityanalysiscanbesimpleorinvolvemorecomplexmathematicalandstatisticaltechniques,suchascorrelationandregressionanalysis,todeterminewhichfactorsinariskmodelcontributemosttothevarianceintheriskestimate.

Withinthesensitivityanalyses,arangeoftechniquesisavailable:simple,“back‐of‐the‐envelope”calculations,wheretheriskparametersareevaluatedusingarangeofexposureparameterstodeterminetheparameterthatcontributesmostsignificantlytotherisk(CaseStudy1);analysestoranktherelativecontributionsofvariablestotheoverallrisk(CaseStudy2);anddatavisualizationusinggraphicaltechniquestoarraythedataorMonteCarlosimulations(e.g.,scatterplots).

Moresophisticatedanalysesmayincludesensitivityratios(e.g.,elasticity);sensitivityscores(e.g.,weightedsensitivityratios);correlationcoefficientorcoefficientofdetermination;r2(e.g.,Pearsonproductmoment,Spearmanrank);normalizedmultipleregressioncoefficients;andgoodness‐of‐fittestsforsubsetsoftheriskdistribution(USEPA2001).

Thesensitivityanalysestypicallyrequireminimalresourcesandtime.ResultsofthesensitivityanalysesareusefulinidentifyingkeyparameterswhereadditionalGroup2orGroup3analysesmaybeappropriate.Sensitivityanalysesalsoarehelpfulinidentifyingkeyparameterswhereadditionalresearchwillhavethehighestimpactontheriskassessment.

Group 2 Case Studies Casestudieswithinthisgroupincludeamoresophisticatedapplicationofprobabilistictools,includingPRAofspecificexposureparameters(CaseStudies3and4),one‐dimensionalanalyses(CaseStudy5)andprobabilisticsensitivityanalysis(CaseStudies6and7).

TheGroup2casestudiesrequirelargertimecommitmentsfordevelopment,specializedexpertiseandadditionalanalysisofexposureparameterdatasources.Dependingonthenatureoftheanalysis,peerinvolvementorpeerreviewmaybeappropriatetoevaluatetheproductsoftheanalysis.

Page 69: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

58

Group 3 Case Studies Assessmentswithinthisgrouparethemostresource‐andtime‐intensiveanalysesofthethreecategories.Riskanalysesincludetwo‐dimensionalMonteCarloanalysis(2‐DMCA)thatevaluatesmodelvariabilityanduncertainty(CaseStudies8,9and10);microexposureeventanalysis(MEE),inwhichlong‐termexposureofanindividualissimulatedasthesumofseparateshort‐termexposureevents(CaseStudy11);andprobabilisticanalysis(CaseStudies12and13).

Othertypesofanalyseswithinthisgroupincludetheexpertelicitationmethodthatisasystematicprocessofformalizingandquantifying,intermsofprobabilities,experts’judgmentsaboutuncertainquantities(CaseStudies14and15);Bayesianstatistics,whichisaspecializedbranchofstatisticsthatviewstheprobabilityofaneventoccurringasthedegreeofbelieforconfidenceinthatoccurrence(CaseStudy16);andgeostatisticalanalysis,whichisanotherspecializedbranchofstatisticsthatexplicitlytakesintoaccountthegeo‐referencedcontextofthedataandtheinformation(e.g.,attributes)attachedtothedata.

TheGroup3analysesrequireadditionaltimeandexpertiseintheplanningandanalysisoftheassessment.Withinthisgroup,thelevelofexpertiseandresourcecommitmentsmayvarywiththetechniques.Expertelicitation,forexample,requiressignificantlymoretimeforplanning,identificationofexpertsandmeetings,whencomparedwiththeothertechniques.

 

Page 70: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

59

Table A‐1. Case Study Examples of EPA Applications of Deterministic and Probabilistic Risk Assessment Techniques  

Case Study Number

Title and Case Study Description Type of Risk Assessment

Office/Region

Group 1: Point Estimate—Sensitivity Analysis

1

Sensitivity Analysis of Key Variables in Probabilistic Assessment of Children’s Exposure to Arsenic in Chromated Copper Arsenate (CCA) Pressure-Treated Wood. This case study demonstrates use of a point estimate sensitivity analysis to identify exposure variables critical to the analysis summarized in Case Study 9. The sensitivity analysis identified critical areas for future research and data collection and better characterized the amount of dislodgeable residue that exists on the wood surface.

Human Health

Office of Research and Development (ORD) and Office of Pesticide Programs (OPP)

2

Assessment of the Relative Contribution of Atmospheric Deposition to Watershed Contamination. An example of a workbook that demonstrates how “back-of-the-envelope” analysis of potential exposure rates can be used to target resources to identify other inputs before further analysis of air inputs in watershed contamination. Identification of key variables aided in identifying uncertainties and data gaps to target resource expenditures for further analysis. A case study example of the application of this technique also is identified.

Ecological ORD

Group 2: Probabilistic Risk Analysis, One-Dimensional Monte Carlo Analysis (1-D MCA) and Probabilistic Sensitivity Analysis

Group 2: Probabilistic Risk Analysis

3

Probabilistic Assessment of Angling Duration Used in the Assessment of Exposure to Hudson River Sediments via Consumption of Contaminated Fish. A probabilistic analysis of one parameter in an exposure assessment―the time an individual fishes in a large river system. Development of site-specific information regarding exposure, with an existing data set for this geographic area, was needed to represent this exposed population. This information was used in the one-dimensional PRA described in Case Study 5.

Human Health

Superfund/ Region 2 (New York)

4

Probabilistic Analysis of Dietary Exposure to Pesticides for Use in Setting Tolerance Levels. The probabilistic Dietary Exposure Evaluation Model (DEEM) provides more accurate information on the range and probability of possible exposures.

Human Health OPP

Page 71: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

60

Table A‐1. Case Study Examples of EPA Applications of Deterministic and Probabilistic Risk Assessment Techniques  

Case Study Number

Title and Case Study Description Type of Risk Assessment

Office/Region

Group 2: One-Dimensional Monte Carlo Analysis (1-D MCA)

5

One-Dimensional Probabilistic Risk Analysis of Exposures to Polychlorinated Biphenyls (PCBs) via Consumption of Fish From a Contaminated Sediment Site. An example of a one-dimensional PRA (1-D MCA) of the variability of exposure as a function of the variability of individual exposure factors to evaluate the risks to anglers who consume recreationally caught fish from a PCB-contaminated river.

Human Health

Superfund/ Region 2

(New York)

Group 2: Probabilistic Sensitivity Analysis

6

Probabilistic Sensitivity Analysis of Knowledge Elicitation of the Concentration-Response Relationship Between PM2.5 Exposure and Mortality. An example of how the probabilistic analysis tools can be used to conduct a probabilistic sensitivity analysis following an expert elicitation (Group 3) presented in Case Study 14.

Human Health Office of Air

and Radiation (OAR)

7

Environmental Monitoring and Assessment Program (EMAP): Using Probabilistic Sampling Techniques to Evaluate the Nation’s Ecological Resources. A probability-based sampling program designed to provide unbiased estimates of the condition of an aquatic resource over a large geographic area based on a small number of samples.

Ecological ORD

Group 3: Advanced Probabilistic Risk Analysis―Two-Dimensional Monte Carlo Analysis (2-D MCA) Including Microexposure Modeling, Bayesian Statistics, Geostatistics and Expert Elicitation

Group 3: Two-Dimensional Probabilistic Risk Analysis

8

Two-Dimensional Probabilistic Risk Analysis of Cryptosporidium in Public Water Supplies, With Bayesian Approaches to Uncertainty Analysis. An analysis of the variability in the occurrence of Cryptosporidium in raw water supplies and in the treatment efficiency, as well as the uncertainty in these inputs. This case study includes an analysis of the dose-response relationship for Cryptosporidium infection.

Human Health Office of

Water (OW)

9

Two-Dimensional Probabilistic Model of Children’s Exposure to Arsenic in Chromated Copper Arsenate (CCA) Pressure-Treated Wood. A two-dimensional model that addresses both variability and uncertainty in the exposures of children to CCA pressure-treated wood. The analysis was built on the sensitivity analysis described in Case Study 2.

Human Health OPP/ORD

Page 72: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

61

Table A‐1. Case Study Examples of EPA Applications of Deterministic and Probabilistic Risk Assessment Techniques  

Case Study Number

Title and Case Study Description Type of Risk Assessment

Office/Region

10

Two-Dimensional Probabilistic Exposure Assessment of Ozone. A probabilistic exposure assessment that addresses short-term exposures to ozone. Population exposure to ambient ozone levels was evaluated using EPA’s Air Pollutants Exposure (APEX) model, also referred to as the Total Risk Integrated Methodology/Exposure (TRIM.Expo) model.

Human Health OAR

Group 3: Microexposure Event Analysis

11

Analysis of Microenvironmental Exposures to Fine Particulate Matter (PM2.5) for a Population Living in Philadelphia, Pennsylvania. A microexposure event analysis to simulate individual exposures to PM2.5 in specific microenvironments, including the outdoors, indoor residences, offices, schools, stores and a vehicle.

Human Health Region 3

(Philadelphia) and ORD

Group 3: Probabilistic Analysis

12

Probabilistic Analysis in Cumulative Risk Assessment of Organophosphorus Pesticides. A probabilistic computer software program used to integrate various pathways, while simultaneously incorporating the time dimensions of the input data to calculate margins of exposure.

Human Health OPP

13

Probabilistic Ecological Effects Risk Assessment Models for Evaluating Pesticide Uses. A multimedia exposure/ effects model that evaluates acute mortality levels in generic or specific avian species over a user-defined exposure window.

Ecological OPP

Group 3: Expert Elicitation and Bayesian Belief Network

14

Expert Elicitation of Concentration-Response Relationship Between Fine Particulate Matter (PM2.5) Exposure and Mortality. A knowledge elicitation used to derive probabilistic estimates of the uncertainty in one element of a cost-benefit analysis used to support the PM2.5 regulations.

Human Health ORD/ OAR

15

Expert Elicitation of Sea-Level Change Resulting From Global Climate Change. An example of a PRA that describes the probability of sea level rise and parameters that predict sea level change.

Ecological

Office of Policy,

Planning, and Evaluation

Page 73: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

62

Table A‐1. Case Study Examples of EPA Applications of Deterministic and Probabilistic Risk Assessment Techniques  

Case Study Number

Title and Case Study Description Type of Risk Assessment

Office/Region

16

Knowledge Elicitation for Bayesian Belief Network Model of Stream Ecology. An example of a Bayesian belief network model of the effect of increased fine-sediment load in a stream on macroinvertebrate populations.

Ecological ORD

Page 74: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

63

D. CASE STUDY SUMMARIES D.1. Group 1 Case Studies

Case Study 1: Sensitivity Analysis of Key Variables in Probabilistic Assessment of Children’s Exposure to Arsenic in Chromated Copper Arsenate Pressure-Treated Wood ThiscasestudyprovidesanexampleoftheapplicationofsensitivityanalysistoidentifyimportantvariablesforpopulationexposurevariabilityforaGroup2assessment(CaseStudy9)andtoindicateareasforfurtherresearch.Specifically,EPA’sOfficeofResearchandDevelopment(ORD),incollaborationwiththeOfficeofPesticidePrograms(OPP),usedsensitivityanalysestoidentifythekeyvariablesinchildren’sexposuretoCCA‐treatedwood.

Approach.Thesensitivityanalysesusedtwoapproaches.Thefirstapproachestimatedbaselineexposurebyrunningtheexposuremodelwitheachinputvariablesettoitsmedian(50thpercentile)value.Next,alternativeexposureestimatesweremadebysettingeachinputtoits25thor75thpercentilevaluewhileholdingallotherinputsattheirmedianvalues.Theratiooftheexposureestimatecalculatedwhenaninputwasestimatedatits25thor75thpercentiletotheexposureestimatecalculatedwhentheinputwasatitsmedianvalueprovidedameasureofthatinput’simportancetotheoverallexposureassessment.Thesecondapproachappliedamultiplestepwiseregressionanalysistothedatapointsgeneratedfromthefirstapproach.Thecorrelationbetweentheinputvariablesandtheexposureestimatesprovidedanalternativemeasureoftheinputvariable’srelativeimportanceintheexposureassessment.Thesetwoapproacheswereusedintandemtoidentifythecriticalinputstotheexposureassessmentmodel.

ResultsofAnalysis.Thetwosensitivityanalysestogetheridentifiedsixcriticalinputvariablesthatmostinfluencedtheexposureassessment.Thecriticalinputvariableswere:woodsurfaceresidue‐to‐skintransferefficiency,woodsurfaceresiduelevels,fractionofhandsurfaceareamouthedpermouthingevent,averagefractionofnonresidentialoutdoortimespentplayingonaCCA‐treatedplayset,frequencyofhandwashingandfrequencyofhand‐to‐mouthactivity.

ManagementConsiderations.Theresultsofthesensitivityanalyseswereusedtoidentifythemostimportantinputparametersinthetreatedwoodriskassessments.Theprocessalsoidentifiedcriticalareasforfutureresearch.Inparticular,theassessmentpointedtoaneedtocollectdataontheamountofdislodgeableresiduethatistransferredfromthewoodsurfacetoachild’shanduponcontact,andtobettercharacterizetheamountofdislodgeableresiduethatexistsonthewoodsurface.

SelectedReferences.ThefinalreportontheprobabilisticexposureassessmentofCCA‐treatedwood:

Zartarian,V.G.,J.Xue,H.A.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.AProbabilisticExposureAssessmentforChildrenWhoContactCCA‐TreatedPlaysetsandDecksUsingtheStochasticHumanExposureandDoseSimulationModelfortheWoodPreservativeScenario(SHEDS‐WOOD),FinalReport.EPA/600/X‐05/009.Washington,D.C.:USEPA.

Seealso:Xue,J.,V.G.Zartarian,H.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.2006.“AProbabilisticExposureAssessmentforChildrenWhoContactChromatedCopperArsenate(CCA)‐TreatedPlaysetsandDecks,Part2:SensitivityandUncertaintyAnalyses.”RiskAnalysis26:533–41.

Page 75: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

64

Case Study 2: Assessment of the Relative Contribution of Atmospheric Deposition to Watershed Contamination Watershedcontaminationcanresultfromseveraldifferentsources,includingthedirectreleaseofpollutionintoawaterbody,inputfromupstreamwaterbodiesanddepositionfromairbornesources.Effortstocontrolwaterbodycontaminationbeginwithananalysisoftheenvironmentalsourcestoidentifytheparametersthatprovidethegreatestcontributionanddeterminewheremitigationand/oranalysisresourcesshouldbedirected.

Approach.Thiscasestudyprovidesanexampleofa“back‐of‐the‐envelope”deterministicanalysisofthecontributionofairdepositiontooverallwatershednitrogen?Nutrient?contaminationtoidentifyuncertaintiesand/ordatagaps,aswellastotargetresourceexpenditures.NitrogeninputshavebeenstudiedinseveraleastandGulfCoastestuariesduetoconcernsabouteutrophication.Nitrogenfromatmosphericdepositionisestimatedtobeashighas10to40percentofthetotalinputofnitrogentomanyoftheseestuaries,andperhapshigherinafewcases.Forawatershedthathasnotbeenstudiedyet,aback‐of‐the‐envelopecalculationcouldbepreparedbasedoninformationaboutthenitrogendepositionratesmeasuredinasimilararea.Toestimatethedepositionloaddirectlytothewaterbody,onewouldmultiplythenitrogendepositionratebytheareaofthewaterbody.Theanalystthencouldestimatethenitrogenloadfromothersources(e.g.,pointsourcedischargesandrunoff)toestimateatotalnitrogenloadforthewaterbody.Theestimateofloadingduetoatmosphericdepositionthencouldbedividedbythetotalnitrogenloadforthewaterbodytoestimatethepercentofcontributiondirectlytothewaterbodyfromatmosphericdeposition.

TheMay2003reportbytheCascoBayAirDepositionStudyTeamtitledEstimatingPollutantLoadingFromAtmosphericDepositionUsingCascoBay,MaineasaCaseStudyisananalysisusingthemethodologydescribedabove.TheCascoBayEstuary,locatedinsouthwesternMaine,isusedasacasestudy.ThepaperalsoincludestheresultsofafieldairdepositionmonitoringprogramconductedinCascoBayfrom1998to2000andfavorablycomparestheestimatesdevelopedfortherateofdepositionofnitrogen,mercuryandpolycyclicaromatichydrocarbons(PAHs)tothefieldmonitoringresults.Theestimationapproachisausefulstartingpointforunderstandingthesourcesofpollutantsenteringwaterbodiesthatcannotbeaccountedforthroughrunofforpointsourcedischarges.

ResultsofAnalysis.TheapproachoutlinedabovewasappliedtotheCascoBayEstuaryinMaine.Resources,toolsandstrategiesforpollutionabatementcanbeeffectivelytargetedatprioritysourcesifestuariesaretobeprotected.Understandingthesourcesandannualloadingofcontaminantstoanestuaryfacilitatesgoodwaterqualitymanagementbydefiningtherangeofcontrolsofbothairandwaterpollutionneededtoachieveadesiredresult.Thecostofconductingmonitoringtodetermineatmosphericloadingtoawaterbodycanbeprohibitivelyhigh.Also,collectionofmonitoringdataisalong‐termundertakingbecauseaminimumof3yearsofdataisadvisableto“smoothout”inter‐annualvariability.Theestimationtechniquesdescribedinthispapercanserveasausefulandinexpensive“first‐cut”atunderstandingtheimportanceoftheatmosphereasapollutionsourceandcanhelptoidentifyareaswherefieldmeasurementsareneededtoguidefuturemanagementdecisions.

ManagementConsiderations.Ifareviewofinformationonairdepositionavailablefortheanalysisindicatesawiderangeofpotentialdepositionrates,furtherstudyofthisinputwouldleadtobettercharacterizationoftheaircontributiontooverallcontamination.Iftheback‐of‐theenvelopeanalysissuggeststhatairdepositionisverysmallrelativetootherinputs,thenresourcesshouldbetargetedatstudyingorreducingotherinputsbeforeproceedingwithfurtheranalysisoftheairinputs.

Page 76: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

65

SelectedReferences.Theback‐of‐the‐envelopecalculationisoutlinedinFrequentlyAskedQuestionsaboutAtmosphericDeposition:AHandbookforWatershedManagers.http://www.epa.gov/air/oaqps/gr8water/handbook/airdep_sept.pdf.

FurtheranalysisisavailableinDepositionofAirPollutantstotheGreatWaters—ThirdReporttoCongress.http://www.epa.gov/air/oaqps/gr8water/3rdrpt/index.html.

TheCascoBayEstuaryexampleisavailableathttp://epa.gov/owow/airdeposition/cascobay.pdf.

D.2. Group 2 Case Studies

Case Study 3: Probabilistic Assessment of Angling Duration Used in the Assessment of Exposure to Hudson River Sediments via Consumption of Contaminated Fish InassessingthehealthimpactofcontaminatedSuperfundsites,exposuredurationtypicallyisassumedtobethesameasthelengthoftimethatanindividuallivesinaspecificarea(i.e.,residenceduration).InconductingtheHHRAfortheHudsonRiverPolychlorinatedBiphenyl(PCB)SuperfundSite,however,therewasconcernthatexposuredurationbasedonresidencedurationmayunderestimatethetimespentfishing(i.e.,anglingduration).

RiskAnalysis.Anindividualmaymovefromoneresidencetoanotherandcontinuetofishinthesamelocation,oranindividualmaychoosetostopfishingirrespectiveofthelocationofhisorherhome.EPARegion2developedasite‐specificdistributionofanglingdurationusingthefishingpatternsreportedinaNewYorkState‐wideanglingsurvey(Connellyetal.1990)andmigrationdataforthefivecountiessurroundingmorethan40milesoftheUpperHudsonRivercollectedaspartoftheU.S.Census.

ResultsofAnalysis.The50thand95thpercentilevaluesfromthedistributionofanglingdurationswerehigherthanthedefaultvaluesbasedonresidencedurationusingstandarddefaultexposureassumptionsforresidentialscenarios.Thesevalueswereusedasabaseforthecentraltendencyandreasonablemaximumexposurepointestimates,respectively,inthedeterministicassessment.

ManagementConsiderations.Theinformationprovidedinthisanalysiswasusedinthepointestimateanalysis.ThefulldistributionwasusedinconductingaGroup2PRAforthefishconsumptionpathway,whichispresentedasCaseStudy5.

SelectedReferences.ThefinalriskassessmentwasreleasedinNovember2000andisavailableathttp://www.epa.gov/hudson/reports.htm.

Furtherinformation,includingEPA’sJanuary2002responsetocommentsontheriskassessment,isavailableathttp://www.epa.gov/hudson/ResponsivenessSummary.pdf.

Case Study 4: Probabilistic Analysis of Dietary Exposure to Pesticides for Use in Setting Tolerance Levels UndertheFederalFood,Drug,andCosmeticAct(FFDCA),EPAmayauthorizeatoleranceorexemptionfromtherequirementofatolerancetoallowapesticideresidueinfood,onlyiftheAgencydeterminesthatsuchresidueswouldbe“safe.”Thisdeterminationismadebyestimatingexposuretothepesticideandcomparingtheestimatedexposuretoatoxicologicalbenchmarkdose.Until1998,theOPPusedasoftwareprogramcalledtheDietaryRiskEvaluationSystem(DRES)toconductitsacutedietaryriskassessmentsforpesticideresiduesinfoods.AcuteassessmentsconductedwithDRESassumedthat100percentofagivencropwithregistereduseofapesticide

Page 77: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

66

wastreatedwiththatpesticideandallsuchtreatedcropitemscontainedpesticideresiduesatthemaximumlegal(tolerance)level,matchingthistoareasonablyhighconsumptionvalue(aroundthe95thpercentile).TheresultingDRESacuteriskestimateswereconsidered“high‐end”or“bounding”estimates.Itwasnotpossible,however,toknowwherethepesticideexposureestimatesfromtheDRESsoftwarefitintheoveralldistributionofexposuresduetothelimitsofthetoolsbeingused.

Toaddressthesedeficiencies,OPPdevelopedanacuteprobabilisticdietaryexposureguidancetouseamodelthatwouldestimatetheexposuretopesticidesinthefoodsupply.Ratherthanthecrude“high‐end,”single‐pointestimatesprovidedbydeterministicassessments,theprobabilisticDietaryExposureEvaluationModel(DEEM)providesspecificinformationabouttherangeandprobabilityofpossibleexposures.Dependingonthecharacterizationoftheinput,thiscouldincludethe95thpercentileregulation—generallyforlowertiersthatdonotincludethepercentofcroptreated—tothe99.9thpercentileforthemorerefinedassessments,whichwouldincludethepercentofcroptreatedinformation.

ProbabilisticAnalysis.Thiscasestudyprovidesanexampleofaone‐dimensionalPRAofdietaryexposuretopesticides(Group2).TheDEEMgeneratesacute,probabilisticdietaryexposureassessmentsusingdataon:(1)thedistributionofdailyconsumptionofspecificcommodities(e.g.,wheat,cornandapples)byspecificindividuals;and(2)thedistributionofconcentrationsofaspecificpesticideinthosefoodcommodities.DataoncommodityconsumptionarecollectedbytheU.S.DepartmentofAgriculture(USDA)initsContinuingSurveyofFoodIntakebyIndividuals(CSFII).Pesticideresidueconcentrationsonfoodcommoditiesaregenerallyobtainedfromcropfieldtrials,USDA’sPesticideDataProgram(PDP),U.S.FoodandDrugAdministration(FDA)monitoringdata,ormarketbasketsurveys.Usingthesedata,DEEMisabletocalculateanestimateoftherisktothegeneralU.S.population,inadditionto26populationsubgroups,including5subgroupsforinfantsandchildren(infantslessthan1,children1to2,children3to5,youths6to12andteens13to19yearsofage).

ResultsofAnalysis.DEEMhasbeenusedinriskassessmentstosupporttolerancelevelsforseveralpesticides(e.g.,phosalone)andaspartofcumulativeriskassessmentsfororganophosphoruscompounds(seeCaseStudy12)andotherpesticides.

ManagementConsiderations.UsingtheDRES,decisionswerebeingmadewithoutacompleterepresentationofthedistributionofriskamongthepopulationandwithoutfullknowledgeofwhereinthedistributionofrisktheDRESriskestimatelay.Thiswasofconcernnotonlyforregulatorsinterestedinpublichealthprotection,butalsoforthepesticideregistrantswhocouldarguethattheAgencywasarbitrarilyselectingthelevelatwhichtoregulate.FormostcasesreviewedbyOPPtodate,estimatedexposureatthe99.9thpercentilecalculatedbyDEEMprobabilistictechniquesissignificantlylowerthanexposurecalculatedusingDRES‐typedeterministicassumptionsattheunknownpercentile.

SelectedReferences.AlinktotheDEEMmodelisavailableathttp://www.epa.gov/pesticides/science/deem/index.html.

Case Study 5: One-Dimensional Probabilistic Risk Analysis of Exposure to Polychlorinated Biphenyls via Consumption of Fish From a Contaminated Sediment Site EPARegion2conductedapreliminarydeterministicHHRAattheHudsonRiverPCBsSuperfundsite.TheDRAdemonstratedthatconsumptionofrecreationallycaughtfishprovidedthehighest

Page 78: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

67

exposureamongrelevantexposurepathways,whichresultedincancerrisksandnoncancerhealthhazardsthatexceededregulatorybenchmarks.

ProbabilisticAnalysis.Becauseofthesize,complexityandhighlevelofpublicinterestinthissite,EPARegion2implementedaGroup2probabilisticassessmenttocharacterizethevariabilityinrisksassociatedwiththefishconsumptionexposurepathway.Theanalysiswasaone‐dimensionalMonteCarloanalysis(1‐DMCA)ofthevariabilityofexposureasafunctionofthevariabilityofindividualexposurefactors.Uncertaintywasassessedusingsensitivityanalysisoftheinputvariables.Datatocharacterizethedistributionsofexposureparametersweredrawnfromthepublishedliterature(e.g.,fishconsumptionrate)orfromexistingdatabases,suchastheU.S.Censusdata(e.g.,anglingduration,seeCaseStudy3).Mathematicalmodelsoftheenvironmentalfate,transportandbioaccumulationofPCBsintheHudsonRiverpreviouslydevelopedwereusedtoforecastchangesinPCBconcentrationovertime.

ResultsofAnalysis.TheresultsofthePRAwereconsistentwiththedeterministicresults.Forthecentraltendencyindividual,pointestimateswerenearthemedian(50thpercentile).Forthereasonablemaximumexposure(RME)individual,pointestimatevalueswereatorabovethe95thpercentileoftheprobabilisticanalysis.TheDRAandPRAwerethesubjectofaformalpeerreviewbyapanelofindependentexperts.

TheMonteCarlo‐basedcasescenarioistheonefromwhichpointestimateexposurefactorsforfishingestionweredrawn;thus,thepointestimateRMEsandtheMonteCarlo‐basedcaseestimatescanbecompared.Similarly,thepointestimatecentraltendency(average)andtheMonteCarlo‐basedcasemidpoint(50thpercentile)arecomparable.Forcancerrisk,thepointestimateRMEforfishingestion(1x10‐3)fallsapproximatelyatthe95thpercentilefromtheMonteCarlo‐basedcaseanalysis.Thepointestimatecentraltendencyvalue(3x10‐5)andtheMonteCarlo‐basedcase50thpercentilevalue(6x10‐5)aresimilar.Fornoncancerhealthhazards,thepointestimateRMEforfishingestion(104forayoungchild1to6yearsofage)fallsbetweenthe95thand99thpercentilesoftheMonteCarlo‐basedcase.Thepointestimatecentraltendencyhazardindex(HI;12forayoungchild)isapproximatelyequaltothe50thpercentileoftheMonteCarlo‐basedcaseHIof11.

FiguresA‐1andA‐2provideacomparisonofresultsfromtheprobabilisticanalysiswiththatoftheDRAforcancerrisksandnoncancerhealthhazards.FiguresA‐1andA‐2plotpercentilesfor72combinationsofexposurevariables(e.g.,distributionsfromcreelanglersurveys’residenceduration,fishinglocationsandcookinglosses)ofthenoncancerHIvaluesandthecancerrisks,respectively.Ineachofthesefigures,thevariabilityofcancerriskornoncancerHIsforanglerswithintheexposedpopulationisplottedonthey‐axisforparticularpercentileswithinthepopulation.Thisvariabilityisafunctionofthevariationsinfishconsumptionrates,fishingduration,differencesinfishspeciesingestedandsoforth.TheuncertaintyintheestimatesisindicatedbytherangeofeithercancerriskornoncancerHIvaluesplottedonthex‐axis.Thisuncertaintyisafunctionofthe72combinationsoftheexposurefactorinputsexaminedinthesensitivityanalysis.Thisanalysisprovidesasemi‐quantitativeconfidenceintervalforthecancerrisksandHIvaluesatanyparticulatepercentile.Asthesefiguresshow,theintervalsspansomewhatlessthantwoordersofmagnitude(e.g.,<100‐fold).Theverticallinesindicatethedeterministicendpoints.

ManagementConsiderations.Earlyandcontinuedinvolvementofthecommunityimprovedpublicacceptanceoftheresults.Inaddition,carefulconsiderationofthemethodsusedtopresenttheprobabilisticresultstothepublicleadtogreaterunderstandingofthefindings.

SelectedReferences.ThefinalriskassessmentwasreleasedinNovember2000andisavailableathttp://www.epa.gov/hudson/reports.htm.

Page 79: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

68

Furtherinformation,includingEPA’sJanuary2002responsetocommentsontheriskassessment,isavailableathttp://www.epa.gov/hudson/ResponsivenessSummary.pdf.

 Figure A‐1. Monte Carlo Cancer Summary Based on a One‐Dimensional Probabilistic Risk Analysis of Exposure to Polychlorinated Biphenyls. The estimated cancer rate was calculated based on the consumption of fish from a contaminated sediment site. Source: USEPA 2000b. 

  Figure A‐2. Monte Carlo Noncancer Hazard Index Summary Based on a One‐Dimensional Probabilistic Risk Analysis of Exposure to Polychlorinated Biphenyls. The incremental individual hazard index (HI) was calculated based on the consumption of fish from a contaminated sediment site. Source: USEPA 2000b. 

Fraction of Anglers

With HI ≤

Indicated Value

Fraction of Anglers

With Risk ≤

Indicated Value

Page 80: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

69

Case Study 6: Probabilistic Sensitivity Analysis of Expert Elicitation of Concentration-Response Relationship Between Fine Particulate Matter Exposure and Mortality In2002,theNationalResearchCouncil(NRC)recommendedthatEPAimproveitscharacterizationofuncertaintyinthebenefitsassessmentforproposedregulationsofairpollutants.NRCrecommendedthatprobabilitydistributionsforkeysourcesofuncertaintybedevelopedusingavailableempiricaldataorthroughformalelicitationofexpertjudgments.Inresponsetothisrecommendation,EPAconductedanexpertelicitationevaluationoftheconcentration‐responserelationshipbetweenfineparticulatematter(PM2.5)exposureandmortality,akeycomponentofthebenefitsassessmentofthePM2.5regulation.FurtherinformationontheexpertelicitationprocedureandresultsisprovidedinCaseStudy14.Toevaluatethedegreetowhichtheresultsoftheassessmentdependedonthejudgmentsofindividualexpertsoronthemethodsofexpertelicitation,aprobabilisticsensitivityanalysiswasperformedontheresults.

ProbabilisticAnalysis.Theexpertelicitationprocedureusedcarefullyconstructedinterviewstoelicitfromeachof12expertsanestimateoftheprobabilisticdistributionfortheaverageexpecteddecreaseinU.S.annual,adult,all‐causemortalityassociatedwitha1microgrampercubicmeter(μg/m3)decreaseinannualaveragePM2.5levels.Thiscasestudyprovidesanexampleoftheuseofprobabilisticsensitivityanalysis(Group2)asoneelementoftheoverallassessment.Forthesensitivityanalysis,asimplifiedhealthbenefitsanalysiswasconductedtoassessthesensitivityoftheresultstotheresponsesofindividualexpertsandtothreefactorsinthestudydesign:(1)theuseofparametricornonparametricapproachesbyexpertstocharacterizetheiruncertaintyinthePM2.5mortalitycoefficient;(2)participationinthePre‐ElicitationWorkshop;and(3)allowingexpertstochangetheirjudgmentsafterthePost‐ElicitationWorkshop.Theindividualquantitativeexpertjudgmentswereusedtoestimateadistributionofbenefits,intheformofthenumberofdeathsavoided,associatedwithareductioninambient,annualaveragePM2.5concentrationsfrom12to11μg/m3.The12individualdistributionsofestimatedavoideddeathswerepooledusingequalweightstocreateasingleoveralldistributionreflectinginputfromeachexpert.Thisdistributionservedasthebaselineforthesensitivityanalysis,whichcomparedthemeansandstandarddeviationsofthebaselinedistributionwithseveralvariants.

ResultsofAnalysis.Thefirstanalysisexaminedthesensitivityofthemeanandstandarddeviationoftheoverallmortalitydistributiontotheremovalofindividualexperts’distributions.Ingeneral,theresultssuggestedafairlyequaldivisionbetweenthoseexpertswhoseremovalshiftedthedistributionmeanupandthosewhoshifteditdown.Therewererelativelymodestimpactsofindividualexperts.Thestandarddeviationofthecombineddistributionalsowasnotaffectedstronglybytheremovalofindividualexperts.Thesecondanalysisevaluatedwhethertheuseofparametricornonparametricapproachesaffectedtheoverallresults.Theresultssuggestedthattheuseofparametricdistributionsledtodistributionswithsimilarorslightlyincreaseduncertaintycomparedwithdistributionsprovidedbyexpertswhoofferedpercentilesofanonparametricdistribution.ThelastanalysisevaluatedwhetherparticipationinthePre‐orPost‐ElicitationWorkshopsaffectedtheresults.Participationineitherworkshopdidnotappeartohaveasignificanteffectonexperts’judgmentsbasedonmeasuresofchangeinthebaselinedistribution.Overall,thesensitivityanalysesdemonstratedthattheassessmentwasrobust,withlittledependenceonindividualexperts’judgmentsoronthespecificelicitationmethodsevaluated.

ManagementConsiderations.ThesensitivityanalysisdemonstratedtherobustnessofthePM2.5expertelicitation‐basedassessmentbyshowingthatthepanelofexpertswasgenerallywellbalancedandthatalternativeelicitationmethodswouldnothavemarkedlyalteredtheoverallresults.

Page 81: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

70

SelectedReferences.ThedetailsofthisanalysisareprovidedintheIndustrialEconomics,Inc.,documenttitled:ExpandedExpertJudgmentAssessmentoftheConcentration‐ResponseRelationshipBetweenPM2.5ExposureandMortality,FinalReport,September21,2006.Thisdocumentisavailableathttp://www.epa.gov/ttn/ecas/regdata/Uncertainty/pm_ee_report.pdf.

Theexpertelicitationassessment,alongwiththeRegulatoryImpactAnalysis(RIA)ofthePM2.5standard,isavailableathttp://www.epa.gov/ttn/ecas/ria.html.

Case Study 7: Environmental Monitoring and Assessment Program: Using Probabilistic Sampling to Evaluate the Condition of the Nation’s Aquatic Resources Monitoringisakeytoolusedtoidentifythelocationswheretheenvironmentisinahealthybiologicalconditionandrequiresprotection,andwhereenvironmentalproblemsareoccurringandneedremediation.Mostmonitoring,however,isnotperformedinawaythatallowsforstatisticallyvalidassessmentsofwaterqualityconditionsinunmonitoredwaters(USGAO2000).StatesthuscannotadequatelymeasurethestatusandtrendsinwaterqualityasrequiredbytheCleanWaterAct(CWA)Section305(b).

TheEnvironmentalMonitoringandAssessmentProgram’s(EMAP)focushasbeentodevelopunbiasedstatisticalsurveydesignframeworksandsensitiveindicatorsthatallowtheconditionofaquaticecosystemstobeassessedatstate,regionalandnationalscales.AcornerstoneofEMAPhasbeentheuseofprobabilisticsamplingtoallowrepresentative,unbiased,cost‐effectiveconditionassessmentsforaquaticresourcescoveringlargeareas.EMAP’sstatisticalsurveymethodsareveryefficient,requiringrelativelyfewsamplelocationstomakevalidscientificstatementsabouttheconditionofaquaticresourcesoverlargeareas(e.g.,theconditionofallofthewadeablestreamsinthewesternUnitedStates).

ProbabilisticAnalysis.Thisresearchprogramprovidesmultiplecasestudiesusingprobabilisticsamplingdesignsfordifferentaquaticresources(estuaries,streamsandrivers).AnEMAPprobability‐basedsamplingprogramdeliversanunbiasedestimateoftheconditionofanaquaticresourceoveralargegeographicareafromasmallnumberofsamples.Theprincipalcharacteristicsofaprobabilisticsamplingdesignare:thepopulationbeingsampledisunambiguouslydescribed;everyelementinthepopulationhastheopportunitytobesampledwithaknownprobability;andsampleselectionisconductedbyarandomprocess.Thisapproachallowsstatisticalconfidencelevelstobeplacedontheestimatesandprovidesthepotentialtodetectstatisticallysignificantchangesandtrendsinconditionwithrepeatedsampling.Inaddition,thisapproachpermitstheaggregationofdatacollectedfromsmallerareastopredicttheconditionofalargegeographicarea.

TheEMAPdesignframeworkallowstheselectionofunbiased,representativesamplingsitesandspecifiestheinformationtobecollectedatthesesites.Thevalidityoftheoverallinferencerestsonthedesignandsubsequentanalysistoproduceregionallyrepresentativeinformation.TheEMAPusestheapproachoutlinedintheEPA’sGeneralizedRandomTessellationStratifiedSpatially‐BalancedSurveyDesignsforAquaticResources(Olsen2012).Thespatiallybalancedaspectspreadsoutthesamplinglocationsgeographically,butstillensuresthateachelementhasanequalchanceofbeingselected.

ResultsofAnalysis.DatacollectedusingtheEMAPapproachhasallowedtheAgencytomakescientificallydefensibleassessmentsoftheecologicalconditionoflargegeographicareasforreportingtoCongressunderCWA305(b).TheEMAPapproachhasbeenusedtoprovidethefirstreportsontheconditionofthenation’sestuaries,streams,riversandlakes,anditisscheduledtobe

Page 82: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

71

usedforwetlands.EMAPfindingshavebeenincludedinEPA’sReportontheEnvironmentandtheHeinzCenter’sTheStateoftheNation’sEcosystems.DatacollectedthroughanEMAPapproachimprovetheabilitytoassessecologicalprogressinenvironmentalprotectionandrestoration,andprovidevaluableinformationfordecisionmakersandthepublic.TheuseofprobabilisticanalysismethodsallowsmeaningfulassessmentandregionalcomparisonsofaquaticecosystemconditionsacrosstheUnitedStates.Finally,theprobabilisticapproachprovidesscientificcredibilityforthemonitoringnetworkandaidsinidentifyingdatagaps.

ManagementConsiderations.UseofanEMAPapproachaddressescriticismsfromtheGovernmentAccountabilityOffice(GAO),theNationalAcademyofSciences(NAS),theHeinzCenter(anonprofitenvironmentalpolicyinstitution),andothersthatnotedthenationlackedthedatatomakescientificallyvalidcharacterizationsofwaterqualityregionallyandacrosstheUnitedStates.Theprogramprovidescost‐effective,scientificallydefensibleandrepresentativedata,topermittheevaluationofquantifiabletrendsinecosystemcondition,tosupportperformance‐basedmanagementandfacilitatebetterpublicdecisionsregardingecosystemmanagement.EMAP’sapproachnowhasbeenadoptedbyEPA’sOfficeofWater(OW)tocollectdataontheconditionofallthenation’saquaticresources.OW,OfficeofAirandRadiation(OAR)andOfficeofChemicalSafetyandPollutionPrevention(OCSPP;formerlytheOfficeofPrevention,Pesticides,andToxicSubstances)nowhaveenvironmentalaccountabilityendpointsusingEMAPapproachesintheirAgencyperformancegoals.

SelectedReferences.GeneralinformationconcerningEMAPisavailableathttp://www.epa.gov/emap/index.html.

InformationonEMAPmonitoringdesignsisavailableathttp://www.epa.gov/nheerl/arm/designpages/monitdesign/monitoring_design_info.htm.

EPA’sGeneralizedRandomTessellationStratifiedSpatially‐BalancedSurveyDesignsforAquaticResourcesdocumentisavailableathttp://www.epa.gov/nheerl/arm/documents/presents/grts_ss.pdf.

USGAO(U.S.GovernmentAccountabilityOffice).2000.WaterQuality:KeyEPAandStateDecisionsLimitedbyInconsistentandIncompleteData.GAO/RCED‐00‐54.Washington,D.C.:USGAO.http://www.environmental‐auditing.org/Portals/0/AuditFiles/useng00ar_ft_key_epa.pdf.

USEPA(U.S.EnvironmentalProtectionAgency).2002.EMAPResearchStrategy.ResearchTrianglePark,NC:EnvironmentalMonitoringandAssessmentProgram,NationalHealthandEnvironmentalEffectsResearchLaboratory(NHEERL),USEPA.http://www.epa.gov/nheerl/emap/files/emap_research_strategy.pdf.

D.3. Group 3 Case Studies

Case Study 8: Two-Dimensional Probabilistic Risk Analysis of Cryptosporidium in Public Water Supplies, With Bayesian Approaches to Uncertainty Analysis ProbabilisticassessmentoftheoccurrenceandhealtheffectsassociatedwithCryptosporidiumbacteriainpublicdrinkingwatersupplieswasusedtosupporttheeconomicanalysisofthefinalLong‐Term2EnhancedSurfaceWaterTreatmentRule(LT2).EPA’sOfficeofGroundWaterandDrinkingWater(OGWDW)conductedthisanalysisandestablishedabaselinediseaseburdenattributabletoCryptosporidiuminpublicwatersuppliesthatusesurfacewatersources.Next,itmodeledsourcewatermonitoringandfinishedwaterimprovementsthatwillberealizedasaresult

Page 83: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

72

oftheLT2.Post‐RuleriskisestimatedandtheLT2’shealthbenefitistheresultofsubtractingthisfromthebaselinediseaseburden.

ProbabilisticAnalysis.ProbabilisticassessmentwasusedforthisanalysisasameansofaddressingthevariabilityintheoccurrenceofCryptosporidiuminrawwatersupplies,thevariabilityinthetreatmentefficiency,andtheuncertaintyintheseinputsandinthedose‐responserelationshipforCryptosporidiuminfection.ThiscasestudyprovidesanexampleofaPRAthatevaluatesbothvariabilityanduncertaintyatthesametimeandisreferredtoasatwo‐dimensionalPRA.Theanalysisalsoincludedprobabilistictreatmentsofuncertaindose‐responseandoccurrenceparameters.MarkovChainMonteCarlosamplesofparametersetsfilledthisfunction.ThisBayesianapproach(treatingtheunknownparametersasrandomvariables)differsfromclassicaltreatments,whichwouldregardtheparametersasunknown,butfixed(Group3:AdvancedPRA).Theriskanalysisusedexistingdatasets(e.g.,theoccurrenceofCryptosporidiumandtreatmentefficacy)toinformthevariabilityoftheseinputs.UncertaintydistributionswerecharacterizedbasedonprofessionaljudgmentorbyanalyzingdatausingBayesianstatisticaltechniques.

ResultsofAnalysis.TheriskanalysisidentifiedtheCryptosporidiumdose‐responserelationshipasthemostcriticalmodelparametersintheassessment,followedbytheoccurrenceofthepathogenandtreatmentefficiency.BysimulatingimplementationoftheRuleusingimprecise,biasedmeasurementmethods,theassessmentprovidedestimatesofthenumberofpublicwatersupplysystemsthatwouldrequirecorrectiveactionandthenatureoftheactionslikelytobeimplemented.Thisinformationaffordedarealisticmeasureofthebenefits(inreduceddiseaseburden)expectedwiththeLT2.InresponsetoScienceAdvisoryBoard(SAB)comments,additionalCryptosporidiumdose‐responsemodelswereaddedtomorefullyreflectuncertaintyinthiselementoftheassessment.

ManagementConsiderations.TheLT2underwentexternalpeerreview,reviewbyEPA’sSABandintensereviewbytheOfficeofManagementandBudget(OMB).Occurrenceanddose‐responsecomponentsoftheriskanalysismodelwerecommunicatedtostakeholdersduringtheRule’sFederalAdvisoryCommitteeAct(FACA)process.DuetotherigoroftheanalysisandthesignedFACA“AgreementinPrinciple,”theOMBreviewwasstraightforward.

SelectedReferences.ThefinalassessmentofoccurrenceandexposuretoCryptosporidiumwasreleasedinDecember2005andisavailableathttp://www.epa.gov/safewater/disinfection/lt2/regulations.html.

Case Study 9: Two-Dimensional Probabilistic Model of Children’s Exposure to Arsenic in Chromated Copper Arsenate Pressure-Treated Wood ProbabilisticmodelsweredevelopedinresponsetoEPA’sOctober2001FederalInsecticide,Fungicide,andRodenticideAct(FIFRA)ScientificAdvisoryPanel(SAP)recommendationstouseprobabilisticmodelingtoestimatechildren’sexposurestoarsenicinCCA‐treatedplaysetsandhomedecks.

ProbabilisticAnalysis.EPA’sORD,incollaborationwiththeOfficeofPesticidePrograms(OPP),developedandappliedaprobabilisticexposureassessmentofchildren’sexposuretoarsenicandchromiumfromcontactwithCCA‐treatedwoodplaysetsanddecks.Thiscasestudyprovidesanexampleoftheuseoftwo‐dimensional(i.e.,addressingbothvariabilityanduncertainty)probabilisticexposureassessment(Group3:AdvancedPRA).Thetwo‐dimensionalassessmentemployedamodificationofORD’sStochasticHumanExposureandDoseSimulation(SHEDS)model

Page 84: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

73

tosimulatechildren’sexposuretoarsenicandchromiumfromCCA‐treatedwoodplaysetsanddecks,aswellasthesurroundingsoil.StafffrombothORDandOPPcollaboratedinthedevelopmentoftheSHEDS‐Woodmodel.

ResultsofAnalysis.AdraftoftheprobabilisticexposureassessmentreceivedSAPreviewinDecember2003;thefinalreportwasreleasedin2005.Theresultsoftheprobabilisticexposureassessmentwereconsistentwithorintherangeoftheresultsofdeterministicexposureassessmentsconductedbyseveralotherorganizations.Themodelresultswereusedtocompareexposuresunderavarietyofscenarios,includingcoldversuswarmweatheractivitypatterns,useofwoodsealantstoreducetheavailabilityofcontaminantsonthesurfaceofthewood,anddifferenthand‐washingfrequencies.Themodelingofalternativemitigationscenariosindicatedthattheuseofsealantscouldresultinthegreatestexposurereduction,whileincreasedfrequencyofhandwashingalsocouldreduceexposure.

OPPusedtheSHEDS‐Woodmodelexposureresultsinitsprobabilisticchildren’sriskassessmentforCCA(USEPA2008).Thisincludedrecommendationsforriskreduction(useofsealantsandcarefulattentiontochildren’shandwashing)tohomeownerswithexistingCCAwoodstructures.Inaddition,theexposureassessmentwasusedtoidentifyareasforfurtherresearch,including:theefficacyofwoodsealantsinreducingdislodgeablecontaminantresidues,thefrequencywithwhichchildrenplayonoraroundCCAwood,andtransferefficiencyandresidueconcentrations.Tobettercharacterizetheefficacyofsealantsinreducingexposure,EPAandtheConsumerProductSafetyCommission(CPSC)conducteda2‐yearstudyofhowdislodgeablecontaminantresiduelevelschangedwiththeuseofvarioustypesofcommerciallyavailablewoodsealants.

ManagementConsiderations.TheOPPusedSHEDSresultsdirectlyinitsfinalriskassessmentforchildrenplayingonCCA‐treatedplaygroundequipmentanddecks.Themodelenhancedriskassessmentandmanagementdecisionsandprioritizeddataneedsrelatedtowoodpreservatives.Themodelingproductwaspivotalintheriskmanagementandre‐registrationeligibilitydecisionsforCCA,andinadvisingthepublichowtominimizehealthrisksfromexistingtreatedwoodstructures.IndustryalsoisusingSHEDStoestimateexposurestoCCAandotherwoodpreservatives.SomestatesareusingtheriskassessmentasguidanceinsettingtheirregulationsforCCA‐relatedplaygroundequipment.

SelectedReferences.ThefinalprobabilisticriskassessmentbasedontheSHEDS‐Woodexposureassessmentisavailableathttp://www.epa.gov/oppad001/reregistration/cca/final_cca_factsheet.htm.

ThemodelresultswereincludedinthefinalreportontheprobabilisticexposureassessmentofCCA‐treatedwoodsurfaces:Zartarian,V.G.,J.Xue,H.A.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.2006.AProbabilisticExposureAssessmentforChildrenWhoContactCCA‐TreatedPlaysetsandDecksUsingtheStochasticHumanExposureandDoseSimulationModelfortheWoodPreservativeScenario(SHEDS‐Wood),FinalReport.EPA/600/X‐05/009.Washington,D.C.:USEPA.

ResultsofthesealantstudieswerereleasedinJanuary2007andareavailableathttp://www.epa.gov/oppad001/reregistration/cca/index.htm#reviews.

Theresultsoftheanalysiswerepublishedas:Zartarian,V.G.,J.Xue,H.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.2006.“AProbabilisticArsenicExposureAssessmentforChildrenwhoContactCAA‐TreatedPlaysetsandDecks,Part1:ModelMethodology,VariabilityResults,andModelEvaluation.”RiskAnalysis26:515–31.

Moreinformationontheanalysiscanbefoundbyconsultingthefollowingresource:

Page 85: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

74

USEPA(U.S.EnvironmentalProtectionAgency).2008.CaseStudyExamplesoftheApplicationofProbabilisticRiskAnalysisinU.S.EnvironmentalProtectionAgencyRegulatoryDecision‐Making(InReview).Washington,D.C.:RiskAssessmentForum,USEPA

Case Study 10: Two-Dimensional Probabilistic Exposure Assessment of Ozone AspartofEPA’srecentreviewoftheozoneNationalAmbientAirQualityStandards(NAAQS),theOfficeofAirQualityPlanningandStandards(OAQPS)conducteddetailedprobabilisticexposureandriskassessmentstoevaluatepotentialalternativestandardsforozone.Atdifferentstagesofthisreview,theCleanAirScientificAdvisoryCommittee(CASAC)OzonePanel(anindependentscientificreviewcommitteeofEPA’sSAB)andthepublicreviewedandprovidedcommentsonanalysesanddocumentspreparedbyEPA.Ascopeandmethodsplanfortheexposureandriskassessmentswasdevelopedin2005(USEPA2005).ThisplanwasintendedtofacilitateconsultationwiththeCASAC,aswellaspublicreview,andtoobtainadviceontheoverallscope,approachesandkeyissuesinadvanceofthecompletionoftheanalyses.Thiscasestudydescribestheprobabilisticexposureassessment,whichaddressesshort‐termexposurestoozone.TheexposureestimateswereusedasaninputtotheHHRAforlungfunctiondecrementsinallchildrenandasthmaticschool‐agedchildrenbasedonexposure‐responserelationshipsderivedfromcontrolledhumanexposurestudies.

ProbabilisticAnalysis.PopulationexposuretoambientozonelevelswasevaluatedusingEPA’sAirPollutantsExposure(APEX)model,alsoreferredtoastheTotalRiskIntegratedMethodology/Exposure(TRIM.Expo)model.Exposureestimatesweredevelopedforrecentozonelevels,basedon2002to2004airqualitydata,andforozonelevelssimulatedtojustmeettheexisting0.08ppm,8‐hourozoneNAAQSandseveralalternativeozonestandards,basedonadjustingthe2002to2004airqualitydata.Exposureestimatesweremodeledfor12urbanareaslocatedthroughouttheUnitedStatesforthegeneralpopulation,allschool‐agechildrenandasthmaticschool‐agechildren.Thisexposureassessmentisdescribedinatechnicalreport(USEPA2007b).TheexposuremodelAPEXisdocumentedinauser’sguideandtechnicaldocument(USEPA2006).AMonteCarloapproachwasusedtoproducequantitativeestimatesoftheuncertaintyintheAPEXmodelresults,basedonestimatesoftheuncertaintiesforthemostimportantmodelinputs.Thequantificationofmodelinputuncertainties,adiscussionofmodelstructureuncertainties,andtheresultsoftheuncertaintyanalysisaredocumentedinLangstaff(2007).

ResultsofAnalysis.UncertaintyintheAPEXmodelpredictionsresultsfromuncertaintiesinthespatialinterpolationofmeasuredconcentrations,themicroenvironmentmodelsandparameters,people’sactivitypatterns,andtoalesserextent,modelstructure.Thepredominantsourcesofuncertaintyappeartobethehumanactivitypatterninformationandthespatialinterpolationofambientconcentrationsfrommonitoringsitestootherlocations.Theprimarypolicy‐relevantfindingwasthattheuncertaintyintheexposureassessmentissmallenoughtolendconfidencetotheuseofthemodelresultsforthepurposeofinformingtheAdministrator’sdecisionontheozonestandard.

FigureA‐3illustratestheuncertaintydistributionforonemodelresult,thepercentofchildrenwithexposuresabove0.08ppm,8‐hourwhileatmoderateexertion.Thepointestimateof20percentistheresultfromtheAPEXsimulationusingthebestestimatesofthemodelinputs.ThecorrespondingresultfromtheMonteCarlosimulationsrangesfrom17to26percent,witha95percentuncertaintyinterval(UI)of19to24percent.NotethattheUIsarenotsymmetricbecausethedistributionsareskewed.

Page 86: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

75

ManagementConsiderations.Theexposureanalysisalsoprovidedinformationonthefrequencywithwhichpopulationexposuresexceededseveralpotentialhealtheffectbenchmarklevelsthatwereidentifiedbasedontheevaluationofhealtheffectsinclinicalstudies.

TheexposureandriskassessmentsaresummarizedinChapters4and5,respectively,oftheOzoneStaffPaper(USEPA2007a).TheexposureestimatesoverthesepotentialhealtheffectbenchmarkswerepartofthebasisfortheAdministrator’sMarch27,2008,decisiontorevisetheozoneNAAQSfrom0.08to0.075ppm,8‐houraverage(seethefinalrulefortheozoneNAAQS1).

Figure A‐3. Uncertainty Distribution Model Results. The estimated percentage of children with 8‐hour exposures above 0.08 ppm at moderate exertion (the point estimate is 20%). 

SelectedReferences.Moreinformationontheanalysiscanbefoundbyconsultingthefollowingresources:

Langstaff,J.E.2007.AnalysisofUncertaintyinOzonePopulationExposureModeling.OfficeofAirQualityPlanningandStandardsStaffMemorandumtoOzoneNAAQSReviewDocket.OAR‐2005‐0172.http://www.epa.gov/ttn/naaqs/standards/ozone/s_ozone_cr_td.html

USEPA(U.S.EnvironmentalProtectionAgency).2005.OzoneHealthAssessmentPlan:ScopeandMethodsforExposureAnalysisandRiskAssessment.ResearchTrianglePark,NC:OfficeofAirQualityPlanningandStandards,USEPA.http://www.epa.gov/ttn/naaqs/standards/ozone/s_o3_cr_pd.html

USEPA.2006.TotalRiskIntegratedMethodology(TRIM)—AirPollutantsExposureModelDocumentation(TRIM.Expo/APEX,Version4)VolumeI:User’sGuide;VolumeII:TechnicalSupportDocument.ResearchTrianglePark,NC:OfficeofAirQualityPlanningandStandards,USEPA.June2006.http://www.epa.gov/ttn/fera/human_apex.html

USEPA.2007a.ReviewofNationalAmbientAirQualityStandardsforOzone:PolicyAssessmentofScientificandTechnicalInformation—OAQPSStaffPaper.ResearchTrianglePark,NC:OfficeofAirQualityPlanningandStandards,USEPA.http://www.epa.gov/ttn/naaqs/standards/ozone/s_ozone_cr_sp.html

1NationalAmbientAirQualityStandardsforOzone,FinalRule.73Fed.Reg.16436(Mar.27,2008).

0

50

100

150

200

250

300

350

400

450

17% 18% 19% 20% 21% 22% 23% 24% 25% 26%

Percent of children

Num

ber o

f Mon

te C

arlo

sim

ulat

ions

Page 87: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

76

USEPA.2007b.OzonePopulationExposureAnalysisforSelectedUrbanAreas.ResearchTrianglePark,NC:OfficeofAirQualityPlanningandStandards,USEPA.http://www.epa.gov/ttn/naaqs/standards/ozone/s_ozone_cr_td.html

Case Study 11: Analysis of Microenvironmental Exposures to Fine Particulate Matter for a Population Living in Philadelphia, Pennsylvania ThiscasestudyusedtheStochasticHumanExposureandDoseSimulationmodelforParticulateMatter(SHEDS‐PM)developedbyEPA’sNationalExposureResearchLaboratory(NERL)toprepareaprobabilisticassessmentofpopulationexposuretoPM2.5inPhiladelphia,Pennsylvania.Thiscasestudysimulationwaspreparedtoaccomplishthreegoals:(1)estimatethecontributionofPM2.5ofambient(outdoor)origintototalPM2.5exposure;(2)determinethemajorfactorsthatinfluencepersonalexposuretoPM2.5;and(3)identifyfactorsthatcontributethegreatestuncertaintytomodelpredictions.

ProbabilisticAnalysis.Thetwo‐dimensionalprobabilisticassessmentusedamicroexposureeventtechniquetosimulateindividualexposurestoPM2.5inspecificmicroenvironments(outdoors,indoorresidence,office,school,store,restaurantorbar,andinavehicle).Thepopulationforthesimulationwasgeneratedusingdemographicdataatthecensus‐tractlevelfromtheU.S.Census.Characteristicsofthesimulatedindividualswereselectedrandomlytomatchthedemographicproportionswithinthecensustractforgender,age,employmentstatusandhousingtype.TheassessmentusedspatiallyandtemporallyinterpolatedambientPM2.5measurementsfrom1992to1993and1990U.S.CensusdataforeachcensustractinPhiladelphia.Thiscasestudyprovidesanexampleofbothtwo‐dimensional(variabilityanduncertainty)probabilisticassessmentandmicroexposureeventassessment(Group3:AdvancedPRA).

ResultsofAnalysis.ResultsoftheanalysisshowedthathumanactivitypatternsdidnothaveasstronganinfluenceonambientPM2.5exposuresaswasobservedforexposuretoindoorPM2.5

sources.ExposuretoPM2.5ofambientorigincontributedasignificantpercentofthedailytotalPM2.5

exposures,especiallyforthesegmentofthepopulationwithoutexposuretoenvironmentaltobaccosmokeintheresidence.DevelopmentoftheSHEDS‐PMmodelusingthePhiladelphiaPM2.5casestudyalsoprovidedusefulinsightsintodataneedsforimprovinginputsintotheSHEDS‐PMmodel,reducinguncertaintyandfurtherrefinementofthemodelstructure.Someoftheimportantdataneedsidentifiedfromtheapplicationofthemodelinclude:dailyPM2.5measurementsovermultipleseasonsandacrossmultiplesiteswithinanurbanarea,improvedcapabilityofdispersionmodelstopredictambientPM2.5concentrationsatgreaterspatialresolutionandovera1‐yeartimeperiod,measurementstudiestobettercharacterizethephysicalfactorsgoverninginfiltrationofambientPM2.5intoresidentialmicroenvironments,furtherinformationonparticle‐generatingsourceswithintheresidence,anddatafortheotherindoormicroenvironmentsnotspecifiedinthemodel.

ManagementConsiderations.TheapplicationoftheSHEDS‐PMmodeltothePhiladelphiapopulationgaveinsightsintodataneedsandareasformodelrefinement.ThecontinueddevelopmentandevaluationoftheSHEDS‐PMpopulationexposuremodelarebeingconductedaspartofORD’sefforttodevelopasource‐to‐dosemodelingsystemforPMandairtoxics.Thistypeofexposureanddosemodelingsystemisconsideredtobeimportantforthescientificandpolicyevaluationofthecriticalpathways,aswellastheexposurefactorsandsourcetypesinfluencinghumanexposurestoavarietyofenvironmentalpollutants,includingPM.

Page 88: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

77

SelectedReferences.Theresultsoftheanalysiswerepublishedin:

Burke,J.,M.Zufall,andH.Özkaynak.2001.“APopulationExposureModelforParticulateMatter:CaseStudyResultsforPM2.5inPhiladelphia,PA.”JournalofExposureAnalysisandEnvironmentalEpidemiology11(6):470–89.

Georgepoulos,P.G.,S.W.Wang,V.M.Vyas,Q.Sun,J.Burke,R.Vedantham,T.McCurdy,andH.Özkaynak.2005.“ASource‐to‐DoseAssessmentofPopulationExposuretoFinePMandOzoneinPhiladelphia,PA,DuringaSummer1999Episode.”JournalofExposureAnalysisandEnvironmentalEpidemiology15(5):439–57.

Case Study 12: Probabilistic Analysis in Cumulative Risk Assessment of Organophosphorus Pesticides In1996,CongressenactedtheFoodQualityProtectionAct(FQPA),whichrequiresEPAtoconsider“availableevidenceconcerningthecumulativeeffectsoninfantsandchildrenofsuchresiduesandothersubstancesthathaveacommonmechanismoftoxicity”whensettingpesticidetolerances(i.e.,themaximumamountofpesticideresiduelegallyallowedtoremainonfoodproducts).FQPAalsomandatedthatEPAcompletelyreassessthesafetyofallexistingpesticidetolerances(thoseineffectsinceAugust1996)toensurethattheyaresupportedbycurrentscientificdataandmeetcurrentsafetystandards.Becauseorganophosphoruspesticides(OPs)wereassignedpriorityfortolerancereassessment,thesepesticideswerethefirst“commonmechanism”groupidentifiedbyEPA’sOPP.Theultimategoalassociatedwiththiscumulativeriskassessment(CRA)wastoprovideabasisforthedecisionmakertoestablishsafetolerancelevelsforthisgroupofpesticides,whilemeetingtheFQPAstandardforprotectinginfantsandchildren.

ProbabilisticAnalysis.Thiscasestudyprovidesanexampleofanadvancedprobabilisticassessment(Group3).In2006,EPAanalyzedexposuresto30OPsthroughfoodconsumption,drinkingwaterintake,andexposureviapesticideapplication.Distributionsofhumanexposurefactors,suchasbreathingrates,bodyweightandsurfaceareasusedintheassessment,camefromtheAgency’sExposureFactorsHandbook(USEPA1997d).EPAusedCalendex,aprobabilisticcomputersoftwareprogram(availableathttp://www.epa.gov/pesticides/science/deem/)tointegratevariouspathways,whilesimultaneouslyincorporatingthetimedimensionsoftheinputdata.Basedontheresultsoftheexposureassessment,EPAcalculatedmarginsofexposure(MOEs)forthetotalcumulativeriskfromallpathwaysforeachagegroup(infantlessthan1;children1–2,3–5,6–12;youth13–19;andadults20–49and50+yearsofage).

ThefoodcomponentoftheOPsCRAwashighlyrefined,asitwasbasedonresiduemonitoringdatafromtheUSDA’sPDPandsupplementedwithinformationfromtheFDA’sSurveillanceMonitoringProgramsandTotalDietStudy.TheresiduedatawerecombinedwithactualconsumptiondatafromUSDA’sContinuingSurveyofFoodIntakesbyIndividuals(CSFII)usingprobabilistictechniques.TheCRAevaluateddrinkingwaterexposuresonaregionalbasis.TheassessmentfocusedonareaswherecombinedOPexposureislikelytobehighestwithineachregion.Primarily,theanalysisusedprobabilisticmodelingtoestimatetheco‐occurrenceofOPresiduesinwater.Monitoringdatawerenotavailablewithenoughconsistencytobethesolebasisfortheassessment;however,theywereusedtocorroboratethemodelingresults.DatasourcesforthewatercomponentoftheassessmentincludedUSDAAgriculturalUsageReportsforFieldCrops,FruitsandVegetables;USDATypicalPlantingandHarvestingDatesforFieldCropsandFreshMarketandProcessingVegetables;localsourcesforrefinements;andmonitoringstudiesfromtheU.S.GeologicalSurvey(USGS)andothersources.Finally,exposureviatheoral,dermalandinhalationroutesresultingfromapplicationsofOPsinandaroundhomes,schools,officesandotherpublicareaswereassessedprobabilisticallyforeachofthesevenregions.Thedatasourcesforthispartof

Page 89: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

78

theassessmentincludedinformationfromsurveysandtaskforces,specialstudiesandreportsfrompublishedscientificliterature,EPA’sExposureFactorsHandbook(USEPA1997d),andothersources.

ResultsofAnalysis.TheOPsCRApresentedpotentialriskfromsingle‐day(acute)exposuresacross1yearandfromaseriesof21‐dayrollingaveragesacrosstheyear.MOEsatthe99.9thpercentilewereapproximately100orgreaterforallpopulationsforthe21‐dayaverageresults.Theonlyexceptionisabriefperiod(roughly2weeks)inwhichdrinkingwaterexposures(identifiedfromtheExposureFactorsHandbook,USEPA1997d)attributedtophorateuseonsugarcaneresultedinMOEsnear80forchildrenages1to2years.Generally,exposuresthroughthefoodpathwaydominatedtotalMOEs,andexposuresthroughdrinkingwaterweresubstantiallylowerthroughoutmostoftheyear.Residentialexposuresweresubstantiallysmallerthanexposuresthroughbothfoodanddrinkingwater.

TheOPsCRAwasveryresourceintensive.Workbeganin1997withthepreparationofguidancedocumentsandthedevelopmentofaCRAmethodology.Over2to3years,morethan25peoplespent50to100percentoftheirtimeworkingontheassessment,withupto50peopleworkingontheCRAatcriticalperiods.EPAhasspentapproximately$1milliononthisassessment(e.g.,forcomputers,modelsandcontractorsupport).

ManagementConsiderations.TheOPsCRAwasalandmarkdemonstrationofthefeasibilityofaregulatory‐levelassessmentoftheriskfrommultiplechemicals.Uponcompletion,EPApresentedtheCRAatnumerouspublictechnicalbriefingsandFIFRASAPmeetings,andmadeallofthedatainputsavailabletothepublic.TheOPP’ssubstantialefforttocommunicatemethodologies,approachesandresultstothestakeholdersaidedinthesuccessoftheOPsCRA.ThestakeholdersexpressedappreciationforthetransparentnatureoftheOPsCRAandrecognizedtheinnovationandhardworkthatwentintodevelopingtheassessments.

SelectedReferences.The2006assessmentandrelateddocumentsareavailableathttp://www.epa.gov/pesticides/cumulative/common_mech_groups.htm#op.

USEPA(U.S.EnvironmentalProtectionAgency).1997d.ExposureFactorsHandbook.Washington,D.C.:NationalCenterforEnvironmentalAssessment,USEPA.http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=12464.

Case Study 13: Probabilistic Ecological Effects Risk Assessment Models for Evaluating Pesticide Use AspartoftheprocessofdevelopingandimplementingaprobabilisticapproachforERA,anillustrativecasewascompletedin1996.ThiscaseinvolvedbothDRAandPRAfortheeffectsofahypotheticalchemicalX(ChemX)onbirdsandaquaticspecies.FollowingtherecommendationsoftheSAPandinresponsetoissuesraisedbyOPPriskmanagers,theAgencybegananinitiativetorefinetheERAprocessforevaluatingtheeffectsofpesticidestoterrestrialandaquaticspecieswithinthecontextofFIFRA,themainstatutoryauthorityforregulatingpesticidesatthefederallevel.ThekeygoalsandobjectivesofEPA’sinitiativewereto:

Incorporateprobabilistictoolsandmethodstoprovideanestimateonthemagnitudeandprobabilityofeffects.

Buildonexistingdatarequirementsforregistration.

Utilize,whereverpossible,existingdatabasesandcreatenewonesfromexistingdatasourcestominimizetheneedtogenerateadditionaldata.

Focusadditionaldatarequirementsonreducinguncertaintyinkeyareas.

Page 90: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

79

Afterproposingafour‐levelriskassessmentscheme,withhigherlevelsreflectingmorerefinedriskassessmenttechniques,theAgencydevelopedpilotmodelsforbothterrestrialandaquaticspecies.Refinedriskassessmentmodels(LevelII)thenweredevelopedandusedinagenericchemicalcasestudythatwaspresentedtotheSAPin2001.

ProbabilisticAnalysis.Thiscasestudydescribesanadvancedprobabilisticmodelfortheecologicaleffectsofpesticides(Group3).TheterrestrialLevelIImodel(version2.0)isamultimediaexposure/effectsmodelthatevaluatesacutemortalitylevelsingenericorspecificavianspeciesoverauser‐definedexposurewindow.Thespatialscaleisatthefieldlevel,whichincludesthefieldandsurroundingarea.Bothareasareassumedtomeetthehabitatrequirementsforeachspecies,andcontaminationofedgeoradjacenthabitatfromdriftisassumedtobezero.ForeachindividualbirdconsideredinarunoftheLevelIImodel,arandomselectionofvaluesismadeforthemajorexposureinputparameterstoestimateanexternaloraldoseequivalentforthatindividual.Theestimateddoseequivalentiscomparedtoarandomlyassignedtolerancefortheindividualpreselectedfromthedose‐responsedistribution.Ifthedoseisgreaterthanthetolerance,theindividualisscored“dead,”ifnot,theindividualisscored“notdead.”Aftermultipleiterationsofindividuals,aprobabilitydensityfunctionofpercentmortalityisgenerated.

FromMay29to31,1996,theAgencypresentedtwoERAcasestudiestotheSAPforreviewandcomment.AlthoughrecognizingandgenerallyreaffirmingtheutilityofEPA’scurrentdeterministicassessmentprocess,theSAPofferedanumberofsuggestionsforimprovement.Foremostamongtheirsuggestionswasarecommendationtomovebeyondtheexistingdeterministicassessmentapproachbydevelopingthetoolsandmethodologiesnecessarytoconductaprobabilisticassessmentofeffects.Suchanassessmentwouldestimatethemagnitudeandprobabilityoftheexpectedimpactanddefinethelevelofcertaintyandvariationinvolvedintheestimate;riskmanagerswithinEPA’sOPPalsohadrequestedthisinformationinthepast.

TheaquaticLevelIImodelisatwo‐dimensionalMonteCarloriskmodelconsistingofthreemaincomponents:(1)exposure,(2)effectsand(3)risk.TheexposurescenariosusedatLevelIIareintendedtoprovideestimatesforvulnerableaquatichabitatsacrossawiderangeofgeographicalconditionsunderwhichapesticideproductisused.TheLevelIIriskevaluationprocessyieldsestimatesoflikelihoodandmagnitudeofeffectsforacuteexposures.Fortheestimateofacuterisks,adistributionofestimatedexposureandadistributionoflethaleffectsarecombinedthrougha2‐DMCAtoobtainadistributionofjointprobabilityfunctions.Fortheestimateofchronicrisks,adistributionofexposureconcentrationsiscomparedtoachronicmeasurementendpoint.Theriskanalysisforchroniceffectsprovidesinformationonlyontheprobabilitythatthechronicendpointassessedwillbeexceeded,notonthemagnitudeofthechroniceffectexpected.

ResultsofAnalysis.AspartoftheprocessofdevelopingandimplementingaprobabilisticapproachforERA,acasestudywascompleted.ThecasestudyinvolvedbothDRAsandPRAsforeffectsofChemXonbirdsandaquaticspecies.ThedeterministicscreenconductedonChemXconcludedqualitativelythatitcouldposeahighrisktobothfreshwaterfishandinvertebratesandshowedthatPRAwaswarranted.Basedontheprobabilisticanalysis,itwasconcludedthattheuseofChemXwasexpectedtoinfrequentlyresultinsignificantfreshwaterfishmortalitiesbutroutinelyresultinreducedgrowthandotherchroniceffectsinexposedfish.Substantialmortalitiesandchroniceffectstosensitiveaquaticinvertebrateswerepredictedtooccurroutinelyafterpeakexposures.

ManagementConsiderations.Initsreviewofthecasestudy,theFIFRASAPcongratulatedtheAgencyontheeffortmadetoconductPRAofpesticideeffectsinecosystems.Thepanelcommentedthattheapproachhadprogressedgreatlyfromearlierefforts,andthattheintricacyofthemodelswassurprisinglygood,giventhetimeintervalinwhichtheAgencyhadtocompletethetask.

Page 91: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

80

Followingthecasestudy,EPArefinedthemodelsbasedontheSAPcomments.Inaddition,theterrestrialLevelIImodelwasrefinedtoincludedermalandinhalationexposure.

SelectedReferences.Anoverviewofthemodelsisavailableathttp://www.epa.gov/oppefed1/ecorisk/fifrasap/rra_exec_sum.htm#Terrestrial.

Case Study 14: Expert Elicitation of Concentration-Response Relationship Between Fine Particulate Matter Exposure and Mortality In2002,theNRCrecommendedthatEPAimproveitscharacterizationofuncertaintyinthebenefitsassessmentforproposedregulationsofairpollutants.NRCrecommendedthatprobabilitydistributionsforkeysourcesofuncertaintybedevelopedusingavailableempiricaldataorthroughformalelicitationofexpertjudgments.AkeycomponentofEPA’sapproachforassessingpotentialhealthbenefitsassociatedwithairqualityregulationstargetingemissionsofPM2.5anditsprecursorsistheeffectofchangesinambientPM2.5levelsonmortality.Avoidedprematuredeathsconstitute,onamonetarybasis,between85and95percentofthemonetizedbenefitsreportedinEPA’sretrospectiveandprospectiveSection812Abenefit‐costanalysesoftheCleanAirAct(CAA;USEPA1997eand1999)andinRegulatoryImpactAnalysis(RIA)forrulessuchastheHeavyDutyDieselEngine/FuelRule(USEPA2000c)andtheNon‐RoadDieselEngineRule(USEPA2004).InresponsetotheNRCrecommendation,EPAconductedanexpertelicitationevaluationoftheconcentration‐responserelationshipbetweenPM2.5exposureandmortality.

ProbabilisticAnalysis.Thiscasestudyprovidesanexampleoftheuseofexpertelicitation(Group3)toderiveprobabilisticestimatesoftheuncertaintyinoneelementofacost‐benefitanalysis.Expertelicitationusescarefullystructuredinterviewstoelicitfromeachexpertabestestimateofthetruevalueforanoutcomeorvariableofinterest,aswellastheiruncertaintyaboutthetruevalue.Thisuncertaintyisexpressedasasubjectiveprobabilisticdistributionofvaluesandreflectseachexpert’sinterpretationoftheoryandempiricalevidencefromrelevantdisciplines,aswellastheirbeliefsaboutwhatisknownandnotknownaboutthesubjectofthestudy.ForthePM2.5expertelicitation,theprocessconsistedofdevelopmentofanelicitationprotocol,selectionofexperts,developmentofabriefingbook,conductofelicitationinterviews,theuseofexpertinputpriortoandfollowingindividualelicitationofjudgmentsandtheexpertjudgmentsthemselves.Theelicitationinvolvedpersonalinterviewswith12healthexpertswhohadconductedresearchontherelationshipbetweenPM2.5exposuresandmortality.

ThemainquantitativequestionaskedeachexperttoprovideaprobabilisticdistributionfortheaverageexpecteddecreaseinU.S.annual,adultandall‐causemortalityassociatedwitha1μg/m3decreaseinannualaveragePM2.5levels.Whenansweringthemainquantitativequestion,eachexpertwasinstructedtoconsiderthatthetotalmortalityeffectofa1μg/m3decreaseinambientannualaveragePM2.5mayreflectreductionsinbothshort‐termpeakandlong‐termaverageexposurestoPM2.5.Eachexpertwasaskedtoaggregatetheeffectsofbothtypesofchangesintheiranswers.Theexpertsweregiventheoptiontointegratetheirjudgmentsaboutthelikelihoodofacausalrelationshiporthresholdintheconcentration‐responsefunctionintotheirowndistributionsortoprovideadistribution“conditionalon”oneorbothofthesefactors.

ResultsofAnalysis.TheprojectteamdevelopedtheinterviewprotocolbetweenOctober2004andJanuary2006.DevelopmentoftheprotocolwasinformedbyanApril2005symposiumheldbytheprojectteam,wherenumeroushealthscientistsandanalystsprovidedfeedback;detailedpretestingwithindependentEPAscientistsinNovember2005;anddiscussionwiththeparticipatingexpertsatapre‐elicitationworkshopinJanuary2006.TheelicitationinterviewswereconductedbetweenJanuaryandApril2006.Followingtheinterviews,theexpertsreconvenedforapost‐elicitation

Page 92: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

81

workshopinJune2006,inwhichtheprojectteamanonymouslysharedtheresultsofallexpertswiththegroup.

ThemedianestimatesforthePM2.5mortalityrelationshipweregenerallysimilartoestimatesderivedfromthetwoepidemiologicalstudiesmostoftenusedinbenefitsassessment.However,inalmostallcases,thespreadoftheuncertaintydistributionselicitedfromtheexpertsexceededthestatisticaluncertaintyboundsreportedbythemostinfluentialepidemiologicstudies,suggestingthattheexpertelicitationprocesswassuccessfulindevelopingmorecomprehensiveestimatesofuncertaintyforthePM2.5mortalityrelationship.TheuncertaintydistributionsforPM2.5

concentration‐responseresultingfromtheexpertelicitationprocesswereusedintheRIAfortherevisedNAAQSstandardforPM2.5(promulgatedonSeptember21,2006).BecausetheNAAQSareexclusivelyhealth‐basedstandards,thisRIAplayednopartinEPA’sdeterminationtorevisethePM2.5NAAQS.BenefitsestimatesintheRIAwerepresentedasrangesandincludedadditionalinformationonthequantifieduncertaintydistributionssurroundingthepointsontheranges,derivedfrombothepidemiologicalstudiesandtheexpertelicitationresults.OMB’sreviewoftheRIAwascompletedinMarch2007.

ManagementConsiderations.TheNAAQSareexclusivelyhealth‐basedstandards,sotheseanalyseswerenotusedinanymannerbyEPAindeterminingwhethertorevisetheNAAQSforPM2.5.Moreover,therequesttoengageintheexpertelicitationdidnotcomefromtheCASAC,theofficialpeerreviewbodyfortheNAAQS;adecisiontoconducttheanalysesdoesnotreflectCASACadvicethatsuchanalysesinformNAAQSdeterminations.TheanalysesaddressedcommentsfromtheNRCthatrecommendedthatprobabilitydistributionsforkeysourcesofuncertaintybeaddressed.TheanalyseswereusedinEPA’sretrospectiveandprospectiveSection812Abenefit‐costanalysesoftheCAA(USEPA1997eand1999)andinRIAsforrulessuchastheHeavyDutyDieselEngine/FuelRule(USEPA2000c)andtheNon‐RoadDieselEngineRule(USEPA2004).InresponsetotheNRCrecommendation,EPAconductedanexpertelicitationevaluationoftheconcentration‐responserelationshipbetweenPM2.5exposureandmortality.

SelectedReference.Theassessmentisavailableathttp://www.epa.gov/ttn/ecas/ria.html.

Case Study 15: Expert Elicitation of Sea-Level Rise Resulting From Global Climate Change TheUnitedNationsFrameworkConventiononClimateChangerequiresnationstoimplementmeasuresforadaptingtorisingsealevelandothereffectsofchangingclimate.Todecideonanappropriateresponse,coastalplannersandengineersweighthecostofthesemeasuresagainstthelikelycostoffailingtoprepare,whichdependsontheprobabilityofthesearisingaparticularamount.TheU.S.NationalAcademyofEngineeringrecommendedthatassessmentsofsealevelriseshouldprovideprobabilityestimates.Coastalengineersregularlyemployprobabilityinformationwhendesigningstructuresforfloods,andcourtsuseprobabilitiestodeterminethevalueoflandexpropriatedbyregulations.This1995casestudydescribesthedevelopmentofaprobabilitydistributionforsealevelrise,usingmodelsemployedbypreviousassessments,aswellastheexpertopinionsof20climateandglaciologyreviewersabouttheprobabilitydistributionsforparticularmodelcoefficients.

ProbabilisticAnalysis.Thiscasestudyprovidesanexamplebothofananalysisdescribingtheprobabilityofsealevelrise,aswellasanexpertelicitationofthelikelihoodofparticularmodelsandprobabilitydistributionsofthecoefficientsusedbythosemodelstopredictfuturesealevelrise(Group3).Theassessmentoftheprobabilityofsealevelriseusedexistingmodelsdescribingthechangeinfivecomponentsofsealevelriseassociatedwithgreenhousegas‐relatedclimatechange(thermalexpansion,smallglaciers,polarprecipitation,meltingandicedischargefromGreenland

Page 93: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

82

andicedischargefromAntarctica).Toprovideastartingpointfortheexpertelicitation,initialprobabilitydistributionswereassignedtoeachmodelcoefficientbasedonthepublishedliterature.

Aftertheinitialprobabilisticassessmentwascompleted,thedraftreportwascirculatedtoexpertreviewersconsideredmostqualifiedtorenderjudgmentsonparticularprocessesforrevisedestimatesofthelikelihoodofparticularmodelsandthemodelcoefficients’probabilitydistributions.Expertsrepresentingbothextremesofclimatechangescience(thosewhopredictedtrivialconsequencesandthosewhopredictedcatastrophiceffects;individualswhosethoughtshadbeenexcludedfrompreviousassessments)wereincluded.Theexpertswereaskedtoprovidesubjectiveassessmentsoftheprobabilitiesofvariousmodelsandmodelcoefficients.Thesesubjectiveprobabilityestimateswereusedinplaceoftheinitialprobabilitiesinthefinalmodelsimulations.Differentrevieweropinionswerenotcombinedtoproduceasingleprobabilitydistributionforeachparameter;instead,eachreviewer’sopinionswereusedinindependentiterationsofthesimulation.Thegroupofsimulationsresultedintheprobabilitydistributionofsealevelrise.

ResultsofAnalysis.Theanalysis,completedwithabudgetof$100,000,providedaprobabilisticestimateofsealevelriseforusebycoastalengineersandregulators.Theresultssuggestedthatthereisa65percentchancethatthesealevelwillrise1millimeter(mm)peryearmorerapidlyinthenext30yearsthanithasbeenrisinginthelastcentury.Undertheassumptionthatnonclimaticfactorsdonotchange,theprojectionssuggestedthatthereisa50percentchancethattheglobalsealevelwillrise45centimeters(cm),anda1percentchanceofa112cmrisebytheyear2100.Themedianpredictionofsealevelrisewassimilartothemidpointestimateof48cmpublishedbytheIntergovernmentalPanelonClimateChange(IPCC)shortlythereafter(IPCC1996).Thereportalsofounda1percentchanceofa4to5meterriseoverthenext2centuries.

ManagementConsiderations.Therearetworeports(USEPA1995c;TitusandNarayanan1996)thatdiscussseveralusesoftheresultsofthisstudy.Byprovidingaprobabilisticrepresentationofsealevelrise,theassessmentallowscoastalresidentstomakedecisionswithrecognitionoftheuncertaintyassociatedwithpredictedchange.Rollingeasementsthatvestwhenthesearisestoaparticularlevelcanbeproperlyvaluedinboth“arms‐length”transactionsalesorwhencalculatingtheallowabledeductionforacharitablecontributionoftheeasementtoaconservancy.Cost‐benefitassessmentsofalternativeinfrastructuredesigns—whichalreadyconsiderfloodprobabilities—alsocanconsidersealevelriseuncertainty.Assessmentsofthebenefitsofpreventinganaccelerationofsealevelrisecanincludemorereadilylow‐probabilityoutcomes,whichcanprovideabetterassessmentofthetrueriskwhenthedamagefunctionisnonlinear,whichoftenisthecase.

SelectedReferences.

USEPA(U.S.EnvironmentalProtectionAgency).1995c.TheProbabilityofSeaLevelRise.EPA/230/R‐95/008.Washington,D.C.:ClimateChangeDivision,USEPA.http://nepis.epa.gov/Exe/ZyPURL.cgi?Dockey=20011G1O.txt

IPCC(IntergovernmentalPanelonClimateChange).1996.ClimateChange1995:TheScienceofClimateChange.ContributionofWorkingGroupItotheSecondAssessmentoftheIntergovernmentalPanelonClimateChange.Cambridge:CambridgeUniversityPress.

Titus,J.G.,andV.Narayanan.1996.“TheRiskofSeaLevelRise.”ClimaticChange33(2):151–212.

Case Study 16: Knowledge Elicitation for a Bayesian Belief Network Model of Stream Ecology Theidentificationofthecausalpathwaysleadingtostreamimpairmentisacentralchallengetounderstandingecologicalrelationships.Bayesianbeliefnetworks(BBNs)areapromisingtoolfor

Page 94: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

83

modelingpresumedcausalrelationships,providingamodelingstructurewithinwhichdifferentfactorsdescribingtheecosystemcanbecausallylinkedandcalculatinguncertaintiesexpressedforeachlinkage.

BBNscanbeusedtomodelcomplexsystemsthatinvolveseveralinterdependentorinterrelatedvariables.Ingeneral,aBBNisamodelthatevaluatessituationswheresomeinformationalreadyisknown,andincomingdataareuncertainorpartiallyunavailable.Theinformationisdepictedwithinfluencediagramsthatpresentasimplevisualrepresentationofadecisionproblem,forwhichquantitativeestimatesofeffectprobabilitiesareassigned.Assuch,BBNshavethepotentialforrepresentingecologicalknowledgeanduncertaintyinaformatthatisusefulforpredictingoutcomesfrommanagementactionsorfordiagnosingthecausesofobservedconditions.Generally,specificationofaBBNcanbeperformedusingavailableexperimentaldata,literaturereviewinformation(secondarydata)andexpertelicitation.AttemptstospecifyaBBNforthelinkagebetweenfinesedimentloadandmacroinvertebratepopulationsusingdatafromliteraturereviewshavefailedbecauseoftheabsenceofconsistentconceptualmodelsandthelackofquantitativedataorsummarystatisticsneededforthemodel.Inlightofthesedeficiencies,aneffortwasmadetouseexpertelicitationtospecifyaBBNfortherelationshipbetweenfinesedimentloadresultingfromhumanactivityandpopulationsofmacroinvertebrates.ThegoalsofthiseffortweretoexaminewhetherBBNsmightbeusefulformodelingstreamimpairmentandtoassesswhetherexpertopinioncouldbeelicitedtomaketheBBNapproachusefulasamanagementtool.

ProbabilisticAnalysis.ThiscasestudyprovidesanexampleofexpertelicitationinthedevelopmentofaBBNmodeloftheeffectofincreasedfinesedimentloadinastreamonmacroinvertebratepopulations(Group3).Forthepurposeofthisstudy,astreamsetting(aMidwestern,low‐gradientstream)andascenarioofimpairment(introductionofexcessfinesediment)wereused.Fivestreamecologistswithexperienceinthespecifiedgeographicsettingwereinvitedtoparticipateinanelicitationworkshop.Aninitialmodelwasdepictedusinginfluencediagrams,withthegoalsofstructuringandspecifyingthemodelusingexpertelicitation.Theecologistswereguidedthroughaknowledgeelicitationsessioninwhichtheydefinedfactorsthatdescribedrelevantchemical,physicalandbiologicalaspectsoftheecosystem.Theecologiststhendescribedhowthesefactorswereconnectedandwereaskedtoprovidesubjective,quantitativeestimatesofhowdifferentattributesofthemacroinvertebrateassemblagewouldchangeinresponsetoincreasedlevelsoffinesediment.Elicitedinputwasusedtorestructurethemodeldiagramandtodevelopprobabilisticestimatesoftherelationshipsamongthevariables.

ResultsofAnalysis.Theelicitedinputwascompiledandpresentedintermsofthemodelasstructuredbythestreamecologistsandtheirmodelspecifications.TheresultswerepresentedbothasrevisedinfluencediagramsandwithBayesianprobabilistictermsrepresentingtheelicitedinput.Thestudyyieldedseveralimportantlessons.Amongthesewerethattheelicitationprocesstakestime(includinganinitialsessionbyteleconferenceaswellasa3‐dayworkshop),definingascenariowithanappropriatedegreeofdetailiscriticalandelicitationofcomplexecologicalrelationshipsisfeasible.

ManagementConsiderations.Thestudywasconsideredsuccessfulforseveralreasons.First,thefeasibilityoftheelicitationapproachtobuildingstreamecosystemmodelswasdemonstrated.Thestudyalsoresultedinthedevelopmentofnewgraphicaltechniquestoperformtheelicitation.TheelicitedinputwasinterpretedintermsofpredictivedistributionstosupportfittingacompleteBayesianmodel.Furthermore,thestudywassuccessfulinbringingtogetheragroupofexpertsinaparticularsubjectareaforthepurposeofsharinginformationandlearningaboutexpertelicitationinsupportofmodelbuilding.Theexerciseprovidedinsightsintohowbesttoadaptknowledgeelicitationmethodstoecologicalquestionsandinformedtheassembledstreamecologistsontheelicitationprocessandonthepotentialbenefitsofthismodelingapproach.Theexplicit

Page 95: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

84

quantificationofuncertaintyinthemodelnotonlyenhancestheutilityofthemodelpredictions,butalsocanhelpguidefutureresearch.

SelectedReferences.

Black,P.,T.Stockton,L.Yuan,D.Allan,W.Dodds,L.Johnson,M.Palmer,B.Wallace,andA.Stewart.2005.“UsingKnowledgeElicitationtoInformaBayesianBeliefNetworkModelofaStreamEcosystem.”Eos,Transactions,AmericanGeophysicalUnion86(18),JointAssemblySupplement,Abstract#NB41E‐05.

Yuan,L.2005.“TI:ABayesianApproachforCombiningDataSetstoImproveEstimatesofTaxonOptima.”Eos,Transactions,AmericanGeophysicalUnion86(18),JointAssemblySupplement,Abstract#NB41E‐04.

Page 96: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

85

CASE STUDY REFERENCES Black,P.,T.Stockton,L.Yuan,D.Allan,W.Dodds,L.Johnson,M.Palmer,B.Wallace,andA.Stewart.2005.“UsingKnowledgeElicitationtoInformaBayesianBeliefNetworkModelofaStreamEcosystem.”Eos,Transactions,AmericanGeophysicalUnion86(18),JointAssemblySupplement,Abstract#NB41E‐05.

Burke,J.,M.Zufall,andH.Özkaynak.2001.“APopulationExposureModelforParticulateMatter:CaseStudyResultsforPM2.5inPhiladelphia,PA.”JournalofExposureAnalysisandEnvironmentalEpidemiology11(6):470–89.

Connelly,N.A.,T.L.Brown,andB.A.Knuth.1990.NewYorkStatewideAnglerSurvey1988.Albany,NY:BureauofFisheries,NewYorkDepartmentofEnvironmentalConservation.

Georgepoulos,P.G.,S.W.Wang,V.M.Vyas,Q.Sun,J.Burke,R.Vedantham,T.McCurdy,andH.Özkaynak.2005.“ASource‐to‐DoseAssessmentofPopulationExposuretoFinePMandOzoneinPhiladelphia,PA,DuringaSummer1999Episode.”JournalofExposureAnalysisandEnvironmentalEpidemiology15(5):439–57.

IPCC(IntergovernmentalPanelonClimateChange).1996.ClimateChange1995:TheScienceofClimateChange.ContributionofWorkingGroupItotheSecondAssessmentoftheIPPC.Cambridge:CambridgeUniversityPress.

Jamieson,D.1996.“ScientificUncertaintyandthePoliticalProcess.”AnnalsoftheAmericanAcademyofPoliticalandSocialScience545:35–43.

Kurowicka,D.,andR.Cooke.2006.UncertaintyAnalysisWithHighDimensionalDependentModeling.WileySeriesinProbabilityandStatistics.NewYork:JohnWiley&Sons.

Langstaff,J.E.2007.AnalysisofUncertaintyinOzonePopulationExposureModeling.OfficeofAirQualityPlanningandStandardsStaffMemorandumtoOzoneNAAQSReviewDocket.OAR‐2005‐0172.Washington,D.C.:USEPA.http://www.epa.gov/ttn/naaqs/standards/ozone/s_ozone_cr_td.html.

OlsenAR,KincaidTM,PaytonQ.2012.Spatiallybalancedsurveydesignsfornaturalresources.In:GitzenRA,MillspaughJJ,CooperAB,LichtDS(eds)DesignandAnalysisofLong‐TermEcologicalMonitoringStudies.CambridgeUniversityPress,Cambridge,UK,pp126‐150

Stahl,C.H.,andA.J.Cimorelli.2005.“HowMuchUncertaintyIsTooMuchandHowDoWeKnow?ACaseExampleoftheAssessmentofOzoneMonitorNetworkOptions.”RiskAnalysis25(5):1109–20.

Titus,J.G.andV.Narayanan.1996.“TheRiskofSeaLevelRise:ADelphicMonteCarloAnalysisinWhichTwentyResearchersSpecifySubjectiveProbabilityDistributionsforModelCoefficientsWithinTheirRespectiveAreasofExpertise.”ClimaticChange33(2):151–212.

USEPA(U.S.EnvironmentalProtectionAgency).1992a.GuidelinesforExposureAssessment.EPA/600/Z‐92/001.Washington,D.C.:RiskAssessmentForum,USEPA.http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=15263.

USEPA.1992b.FrameworkforEcologicalRiskAssessment.EPA/630/R‐92/001.Washington,D.C.:RiskAssessmentForum,USEPA.http://www.epa.gov/raf/publications/framework‐eco‐risk‐assessment.htm.

USEPA.1995a.GuidanceforRiskCharacterization.Washington,D.C.:SciencePolicyCouncil,USEPA.http://www.epa.gov/spc/pdfs/rcguide.pdf.

Page 97: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

86

USEPA.1995b.PolicyonEvaluatingHealthRiskstoChildren.Washington,D.C.:SciencePolicyCouncil,USEPA.http://www.epa.gov/spc/pdfs/memohlth.pdf.

USEPA.1995c.TheProbabilityofSeaLevelRise.EPA/230/R‐95/008.Washington,D.C.:OfficeofPolicy,Planning,andEvaluation,USEPA.http://nepis.epa.gov/Exe/ZyPURL.cgi?Dockey=20011G1O.txt.

USEPA.1997a.PolicyforUseofProbabilisticAnalysisinRiskAssessmentattheU.S.EnvironmentalProtectionAgency.Washington,D.C.:SciencePolicyCouncil,USEPA.http://www.epa.gov/spc/pdfs/probpol.pdf.

USEPA.1997b.GuidingPrinciplesforMonteCarloAnalysis.EPA/630/R‐97/001.Washington,D.C.:RiskAssessmentForum,USEPA.http://www.epa.gov/raf/publications/pdfs/montecar.pdf.

USEPA.1997c.GuidanceonCumulativeRiskAssessment.Part1:PlanningandScoping.Washington,D.C.:SciencePolicyCouncil,USEPA.http://www.epa.gov/spc/pdfs/cumrisk2.pdf.

USEPA.1997d.ExposureFactorsHandbook.Washington,D.C.:NationalCenterforEnvironmentalAssessment,USEPA.http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=12464.

USEPA.1997e.FinalReporttoCongressonBenefitsandCostsoftheCleanAirAct,1970to1990.EPA/410/R‐97/002.Cincinnati,OH:OfficeofAirandRadiation,USEPA.

USEPA.1998.GuidelinesforEcologicalRiskAssessment.EPA/630/R‐95/002F.Washington,D.C.:RiskAssessmentForum,USEPA.http://www.epa.gov/raf/publications/pdfs/ECOTXTBX.PDF.

USEPA.1999.TheBenefitsandCostsoftheCleanAirAct1990to2010:EPAReporttoCongress.EPA/410/R‐99/001.Washington,D.C.:OfficeofAirandRadiation,USEPA.http://www.epa.gov/oar/sect812/1990‐2010/chap1130.pdf.

USEPA.2000a.EPASciencePolicyCouncilHandbook:RiskCharacterization.EPA/100/B‐00/002.Washington,D.C.:SciencePolicyCouncil,USEPA.http://www.epa.gov/spc/pdfs/rchandbk.pdf.

USEPA.2000b.Phase2Report:FurtherSiteCharacterizationandAnalysis.Volume2F—RevisedHumanHealthRiskAssessment,HudsonRiverPCBsReassessmentRI/FS.NewYork:Region2,USEPA.http://www.epa.gov/hudson/revisedhhra‐text.pdf.

USEPA.2000c.FinalHeavyDutyEngine/DieselFuelRule:AirQualityEstimation,SelectedHealthandWelfareBenefitsMethods,andBenefitAnalysisResults.ResearchTrianglePark,NC:OfficeofAirQualityPlanningandStandards,USEPA.http://www.epa.gov/ttnecas1/regdata/Benefits/tsdhddv8.pdf.

USEPA.2001.RiskAssessmentGuidanceforSuperfund:VolumeIII—PartA,ProcessforConductingProbabilisticRiskAssessment.EPA/540/R‐02/002.Washington,D.C.:OfficeofEmergencyandRemedialResponse,USEPA.http://www.epa.gov/oswer/riskassessment/rags3adt/pdf/rags3adt_complete.pdf.

USEPA.2002.EMAPResearchStrategy.EPA620/R‐02/002.ResearchTrianglePark,NC:EnvironmentalMonitoringandAssessmentProgram,NHEERL,USEPA.http://www.epa.gov/nheerl/emap/files/emap_research_strategy.pdf.

USEPA.2004.AnExaminationofEPARiskAssessmentPrinciplesandPractices.EPA/100/B‐04/001.Washington,D.C.:OfficeoftheScienceAdvisor,USEPA.http://www.epa.gov/osa/pdfs/ratf‐final.pdf.

USEPA.2005.OzoneHealthAssessmentPlan:ScopeandMethodsforExposureAnalysisandRiskAssessment.ResearchTrianglePark,NC:OfficeofAirQualityPlanningandStandards,USEPA.http://www.epa.gov/ttn/naaqs/standards/ozone/data/o3_health_assessment_plan_april05.pdf.

Page 98: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

87

USEPA.2006.TotalRiskIntegratedMethodology(TRIM)—AirPollutantsExposureModelDocumentation(TRIM.Expo/APEX,Version4)VolumeI:User’sGuide;VolumeII:TechnicalSupportDocument.ResearchTrianglePark,NC:OfficeofAirQualityPlanningandStandards,USEPA.http://www.epa.gov/ttn/fera/human_apex.html.

USEPA.2007a.ReviewofNationalAmbientAirQualityStandardsforOzone:PolicyAssessmentofScientificandTechnicalInformation—OAQPSStaffPaper.ResearchTrianglePark,NC:OfficeofAirQualityPlanningandStandards,USEPA.http://www.epa.gov/ttn/naaqs/standards/ozone/data/2007_07_ozone_staff_paper.pdf.

USEPA.2007b.OzonePopulationExposureAnalysisforSelectedUrbanAreas.EPA/452/R‐07‐010.ResearchTrianglePark,NC:OfficeofAirQualityPlanningandStandards,USEPA.http://www.epa.gov/ttn/naaqs/standards/ozone/data/2007_07_o3_exposure_tsd.pdf.

USEPA.2014.FrameworkforHumanHealthRiskAssessmenttoInformDecisionMaking.EPA/100/R‐14/001.Washington,D.C.:RiskAssessmentForum,OfficeoftheScienceAdvisor,USEPA.http://www.epa.gov/raf/frameworkhhra.htm

USGAO(U.S.GovernmentAccountabilityOffice).2000.WaterQuality:KeyEPAandStateDecisionsLimitedbyInconsistentandIncompleteData.GAO/RCED‐00‐54.Washington,D.C.:USGAO.http://www.environmental‐auditing.org/Portals/0/AuditFiles/useng00ar_ft_key_epa.pdf.

Xue,J.,V.G.Zartarian,H.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.2006.“AProbabilisticArsenicExposureAssessmentforChildrenWhoContactChromatedCopperArsenate(CCA)‐TreatedPlaysetsandDecks,Part2:SensitivityandUncertaintyAnalyses.”RiskAnalysis26(2):533–41.

Yuan,L.2005.“TI:ABayesianApproachforCombiningDataSetstoImproveEstimatesofTaxonOptima.”Eos,Transactions,AmericanGeophysicalUnion86(18),JointAssemblySupplement,Abstract#NB41E‐04.

Zartarian,V.G.,J.Xue,H.A.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.2006.AProbabilisticExposureAssessmentforChildrenWhoContactCCA‐TreatedPlaysetsandDecksUsingtheStochasticHumanExposureandDoseSimulationModelfortheWoodPreservativeScenario(SHEDS‐Wood),FinalReport.EPA/600/X‐05/009.Washington,D.C.:USEPA.

Zartarian,V.G.,J.Xue,H.A.Özkaynak,W.Dang,G.Glen,L.Smith,andC.Stallings.2006.“AProbabilisticArsenicExposureAssessmentforChildrenWhoContactChromatedCopperArsenate(CAA)‐TreatedPlaysetsandDecks,Part1:ModelMethodology,VariabilityResults,andModelEvaluation.”RiskAnalysis26(2):515–31.

Page 99: Risk Assessment Forum White Paper: Probabilistic Risk Assessment

What Did You Think?

We strive to constantly provide the highest level of value for you. Please take a few minutes to tell us about your experience using this product.

To be taken to a short consumer satisfaction survey, please click here or copy and paste the following URL into your browser:

https://www.surveymonkey.com/r/OSAconsumerfdbck?product=Probabilistic_Risk_Assessment_Methods_and_Case_Studies

Thank you for your feedback.

Sincerely,

Office of the Science Advisor United States Environmental Protection Agency www.epa.gov/[email protected]