Necrotizingcc163

2014:Volume4,Number3ApublicationoftheCentersforMedicare&MedicaidServices,OfficeofInformationProducts&DataAnalyticsTheHHS-HCCRiskAdjustmentModelforIndividualandSmallGroupMarketsundertheAffordableCareActJohnKautter,1GregoryC.
Pope,1MelvinIngber,1SaraFreeman,1LindseyPatterson,1MichaelCohen,2andPatriciaKeenan21RTIInternational2CentersforMedicare&MedicaidServicesAbstract:Beginningin2014,individualsandsmallbusinessesareabletopurchaseprivatehealthinsurancethroughcompetitiveMarketplaces.
TheAffordableCareAct(ACA)providesforaprogramofriskadjustmentintheindividualandsmallgroupmarketsin2014asMarketplacesareimplementedandnewmarketreformstakeeffect.
Thepurposeofriskadjustmentistolessenoreliminatetheinfluenceofriskselectiononthepremiumsthatplanscharge.
Theriskadjustmentmethodologyincludestheriskadjustmentmodelandtherisktransferformula.
ThisarticleisthesecondofthreeinthisissueoftheReviewthatdescribetheDepartmentofHealthandHumanServices(HHS)riskadjustmentmethodologyandfocusesontheriskadjustmentmodel.
Inourfirstcompanionarticle,wediscussthekeyissuesandchoicesindevelopingthemethodology.
Inthisarticle,wepresenttheriskadjustmentmodel,whichisnamedtheHHS-HierarchicalConditionCategories(HHS-HCC)riskadjustmentmodel.
WefirstsummarizetheHHS-HCCdiagnosticclassification,whichisthekeyelementoftheriskadjustmentmodel.
Thenthedataandmethods,results,andevaluationoftheriskadjustmentmodelarepresented.
Fifteenseparatemodelsaredeveloped.
Foreachagegroup(adult,child,andinfant),amodelisdevelopedforeachcostsharinglevel(platinum,gold,silver,andbronzemetallevels,aswellascatastrophicplans).
Evaluationoftheriskadjustmentmodelsshowsgoodpredictiveaccuracy,bothforindividualsandforgroups.
Lastly,thisarticleprovidesexamplesofhowthemodeloutputisusedtocalculateriskscores,whichareaninputintotherisktransferformula.
Ourthirdcompanionpaperdescribestherisktransferformula.
Keywords:riskadjustment,affordablecareact,ACA,riskscore,hierarchicalconditioncategories,HHS-HCCmodel,planliability,predicthealthcareexpenditures,healthinsurancemarketplacesISSN:2159-0354doi:http://dx.
doi.
org/10.
5600/mmrr.
004.
03.
a03Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E1MMRR2014:Volume4(3)Medicare&MedicaidResearchReview2014:Volume4,Number3MissionStatementMedicare&MedicaidResearchReviewisapeer-reviewed,onlinejournalreportingdataandresearchthatinformscurrentandfuturedirectionsoftheMedicare,Medicaid,andChildren'sHealthInsuranceprograms.
Thejournalseekstoexamineandevaluatehealthcarecoverage,qualityandaccesstocareforbeneficiaries,andpaymentforhealthservices.
http://www.
cms.
gov/MMRR/CentersforMedicare&MedicaidServicesMarilynTavennerAdministratorEditor-in-ChiefDavidM.
Bott,Ph.
D.
ThecompletelistofEditorialStaffandEditorialBoardmembersmaybefoundontheMMRRWebsite(clicklink):MMRREditorialStaffPageContact:mmrr-editors@cms.
hhs.
govPublishedbytheCentersforMedicare&MedicaidServices.
AllmaterialintheMedicare&MedicaidResearchReviewisinthepublicdomainandmaybeduplicatedwithoutpermission.
Citationtosourceisrequested.
IntroductionBeginningin2014,individualsandsmallbusinessesareabletopurchaseprivatehealthinsurancethroughcompetitiveMarketplaces.
Issuersmustfollowcertainrulestoparticipateinthemarkets,forexample,inregardtothepremiumstheycanchargeenrolleesandalsonotbeingallowedtorefuseinsurancetoanyoneorvaryenrolleepremiumsbasedontheirhealth.
EnrolleesinindividualmarkethealthplansthroughtheMarketplacesmaybeeligibletoreceivepremiumtaxcreditstomakehealthinsurancemoreaffordableandfinancialassistancetocovercostsharingforhealthcareservices.
ThisarticleisthesecondinaseriesofthreerelatedarticlesinthisissueofMedicare&MedicaidResearchReviewthatdescribetheDepartmentofHealthandHumanServices(HHS)-developedriskadjustmentmethodologyfortheindividualandsmallgroupmarketsestablishedbytheAffordableCareAct(ACA)of2010.
Theriskadjustmentmethodologyconsistsofariskadjustmentmodelandarisktransferformula.
Theriskadjustmentmodelusesanindividual'sdemographicsanddiagnosestodetermineariskscore,whichisarelativemeasureofhowcostlythatindividualisanticipatedtobe.
Therisktransferformulaaveragesallindividualriskscoresinariskadjustmentcoveredplan,makescertainadjustments,andcalculatesthefundstransferredbetweenplans.
Risktransfersareintendedtooffsettheeffectsofriskselectiononplancostswhilepreservingpremiumdifferencesduetofactorssuchasactuarialvaluedifferences.
Thisarticledescribestheriskadjustmentmodel.
Seeourcompanionarticle(Popeetal.
,2014)foradescriptionoftherisktransferformula.
Anothercompanionarticle(Kautter,Pope,andKeenan,2014)discussestheKautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E2MMRR2014:Volume4(3)keyissuesandchoicesindevelopingtheACAriskadjustmentmethodology.
1HHSwillusethisriskadjustmentmethodologywhenoperatingriskadjustmentonbehalfofastate.
In2014,theHHSmethodologywillbeusedinallstatesexceptone(Massachusetts),anditwillapplytoallnon-grandfatheredplans2bothinsideandoutsideoftheMarketplacesintheindividualandsmall-groupmarketsineachstate.
Theorganizationofthisarticleisasfollows.
WefirstsummarizetheHierarchicalConditionCategories(HCC)diagnosticclassificationusedfortheriskadjustmentmodel,whichwedesignatetheHHS-HCCdiagnosticclassificationtodistinguishitfromtheCentersforMedicare&MedicaidServices(CMS)HCC,orCMS-HCC,classificationusedinMedicareriskadjustment(Popeetal.
,2004).
Thenthedataandmethods,results,andevaluationfortheriskadjustmentmodelarepresented.
Finally,weprovideexamplesofhowthemodeloutputisusedtocalculateriskscores,whichareaninputintotherisktransferformula.
HHS-HCCDiagnosticClassificationThebasisoftheHHS-HCCriskadjustmentmodelisusinghealthplanenrolleediagnoses(anddemographics)topredictmedicalexpenditurerisk.
Toobtainaclinicallymeaningfulandstatisticallystablesystem,thetensofthousandsofICD-9-CMcodesusedtocapturediagnosesmustbegroupedintoasmallernumberoforganizedcategoriesthatproduceadiagnosticprofileofeachperson.
Thediagnosticclassificationiskeyindeterminingtheabilityofariskadjustmentmodel1Forgeneralbackgroundonriskadjustment,risktransfers("riskequalization"),andriskselection,seevandeVenandEllis(2000),vandeVenandSchut(2011),VandeVen(2011),andBreyer,Bundorf,andPauly(2012).
2GrandfatheredplansarethosethatwereinexistenceonMarch23,2010,andhavenotbeenchangedinwaysthatsubstantiallycutbenefitsorincreasecostsforenrollees.
GrandfatheredplansareexemptedfrommanyofthechangesrequiredundertheAffordableCareAct.
todistinguishhighfromlowcostindividuals.
Theclassificationalsodeterminesthesensitivityofthemodeltointentionalorunintentionalvariationsindiagnosticcoding,animportantconsiderationinreal-worldriskadjustment.
ThestartingpointfortheHHS-HCCswastheMedicareCMS-HCCs.
TheCMS-HCCshadtobeadaptedintotheHHS-HCCsforACAriskadjustmentforthreemainreasons:1.
PredictionYear—TheCMS-HCCriskadjustmentmodelusesbaseyeardiagnosesanddemographicinformationtopredictthenextyear'sspending.
TheHHS-HCCriskadjustmentmodelusescurrentyeardiagnosesanddemographicstopredictthecurrentyear'sspending.
Medicalconditionsmayhavedifferentimplicationsintermsofcurrentyearcostsandfuturecosts;selectionofHCCsfortheriskadjustmentmodelshouldreflectthosedifferences.
2.
Population—TheCMS-HCCsweredevelopedusingdatafromtheaged(age≥65)anddisabled(agenetworkselection,ratherthanrandomacuteeventsthatrepresentinsurancerisk.
Followinganextensivereviewprocess,weselected127HHS-HCCstobeincludedintheHHSriskadjustmentmodel(seeAppendixExhibitA1foralistingofthe127HHS-HCCs).
Finally,tobalancethecompetinggoalsofimprovingpredictivepowerandlimitingtheinfluenceofdiscretionarycoding,asubsetofHHS-HCCsintheriskadjustmentmodelweregroupedintolargeraggregates,inotherwords"grouping"clustersofHCCstogetherasasingleconditionwithasinglecoefficientthatcanonlybecountedonce.
Aftergrouping,thenumberofHCCfactorsincludedinthemodelwaseffectivelyreducedfrom127to100.
DataandMethodsInthissectionwedescribethedataandmethodsusedfordevelopmentoftheHHS-HCCriskadjustmentmodel.
Wefirstdiscussthechoiceofprospectiveversusconcurrentriskadjustment.
Wethendiscussthedefinitionanddatasourcefortheconcurrentmodelingsample.
Modelvariables,includingexpenditures,demographics,anddiagnosesaredefined.
Finally,themodelestimationandevaluationstrategiesarediscussed.
ModelTypeTheHHS-HCCriskadjustmentmodelisaconcurrentmodel.
Aconcurrentmodelusesdiagnosesfromatimeperiodtopredictcostinthatsameperiod.
Thisisincontrasttoaprospectivemodel,whichusesdiagnosesfromabaseperiodtopredictcostsinafutureperiod.
WhileaprospectivemodelisusedfortheMedicareAdvantageprogram,wedevelopedaconcurrentmodelfortheHHSriskadjustmentmethodologybecause,forimplementationin2014,prioryear(2013)diagnosesdatawillnotbeavailable.
Inaddition,unlikeMedicare,peoplemaymoveinandoutofenrollmentintheindividualandsmallgroupmarkets,soprioryeardiagnosticdatawillnotbeavailableforallenrolleesevenafter2014.
DataThecalibrationsamplefortheHHS-riskadjustmentmodelconsistsof2010TruvenMarketScanCommercialClaimsandEncounterdata.
TheMarketScandataisalarge,well-respected,widely-used,nationally-dispersedproprietarydatabasesourcedfromlargeemployersandhealthplans.
Employees,spouses,anddependentscoveredbyemployer-sponsoredprivatehealthinsuranceareincluded.
TheMarketScansampleincludesenrolleesfromall50statesandtheDistrictofColumbia.
AlthoughMarketScanKautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E4MMRR2014:Volume4(3)representsthelargeemployerratherthanthesmallgroup/individualmarket,weknowofnoevidencethattherelationshipbetweendiagnosesandrelativeexpendituresdifferssignificantlyinthetwomarkets,holdingconstantthegenerosityofplanbenefits(essentialhealthbenefitsandmetallevel).
Wecomparedtheage,sex,andregionaldistributionoftheMarketScansampletotheexpectedACAriskadjustmentpopulation(Trish,Damico,Claxton,Levitt,&Garfield,2011;Buettgens,Garrett,&Holahan,2010).
Wefoundthatoveralltheyaresimilar,althoughtheMarketScandatahasmorechildrenandfeweryoungadults,andmoresamplemembersintheSouthandfewerintheNortheastandWestthantheexpectedriskadjustmentpopulation.
3SampleAnenrolleeisincludedintheconcurrentmodelingsampleiftheenrolleehasatleastonemonthof2010enrollment,isenrolledinapreferredproviderorganization(PPO)orotherfee-for-service(FFS)healthplan,4hasnopaymentsmadeonacapitatedbasis,hasprescriptiondrugcoverage,andhasintegratedmentalhealth/substanceabusecoverage.
5Theprimarygoalsofthesampleselectioncriteriaweretoensurethat1)enrolleeshadcompleteexpenditureanddiagnosisdata,2)enrolleesincludedthoseentering(e.
g.
,newborns)andexiting(e.
g.
,decedents)enrollmentduringtheyear,and3)enrolleeshadhealthcarecoverage3Asdiscussedbelow,wedevelopseparatemodelsforadults,children,andinfants,whichavoidsanyinfluenceofthelargerproportionofchildrenintheMarketScandataonmodelparametervaluesforadults.
Weightingthecalibrationdatatoimprovecorrespondencewiththeriskadjustmentpopulationwillberevisitedinfuturerecalibrationsofthemodelasactualdataontheage-genderandothercharacteristicsoftheACAriskadjustmentpopulationbecomeavailable.
4Otherfee-for-servicehealthplansinclude,forexample,indemnity,consumer-directed,andhigh-deductiblehealthplans.
5Additionally,motherswithbundlednewbornclaims,andnewbornswithnobirthrecords,wereexcluded.
comparabletotheessentialhealthbenefitsundertheACA.
ExpendituresTheHHS-HCCriskadjustmentmodelpredictshealthcareexpendituresforwhichplansareliable,whichexcludeenrolleecostsharing.
Thisistermedaplanliabilityriskadjustmentmodel,whichhasbeenusedinotherpaymentsystems,suchasMedicarePartCandPartD(Popeetal.
,2004;Kautter,Ingber,Pope,&Freeman,2012).
Weconsideredpredictingtotalexpendituresandthenadjustingtoplanliabilitywithamultiplicativeplanactuarialvaluefactor.
However,thisapproachmaynotaccuratelycaptureplanliabilitylevelsduetothenon-linearrelationshipofplanliabilitytototalexpenditures.
Althoughalternativeplancostsharingdesignsexist,wedefineastandardbenefit(planliabilitycostsharing)designforeachcostsharinglevel(platinum,gold,silver,andbronzemetallevels,aswellascatastrophicplans6)usingthefollowingelements.
Planliabilityiszeropercentoftotalexpendituresbelowthedeductible,oneminusthecoinsurancepercentageoftotalexpendituresbetweenthedeductibleandtheout-of-pocketlimit,andonehundredpercentoftotalexpendituresabovetheout-of-pocketlimit.
Thus,thestandardbenefitforeachmetalleveliscompletelyspecifiedbyadeductible,coinsurancerate,andout-of-pocketmaximum.
Usingthe2010MarketScaninpatient,outpatient,anddrugservicesfiles,wesummedtotalpayments(submittedchargesminusnon-coveredchargesminuspricingdiscounts),whichincludeenrolleecostsharing.
Wethentrendedthe2010expendituresto2014byapplyingaconstantannualgrowthrate.
Onceexpendituresweretrended,thestandardbenefitdesignparameters6Whiletechnicallymetallevels(platinum,gold,silver,bronze),andcatastrophicplansdiffer,forpurposesofthisarticle,referencestometallevelswillincludecatastrophicplans.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E5MMRR2014:Volume4(3)(deductibles,coinsurancerates,out-of-pocketlimits)wereappliedtosimulateplanliabilityexpendituresforeachmetallevel.
Planliabilityexpenditureswerethenannualizedbydividingthembythefractionofmonthsin2010thateachbeneficiaryisenrolledintheplan(i.
e.
,bytheeligibilityfraction).
Annualizedexpendituresarethe"permemberpermonth"amountmultipliedby12.
Annualizedexpenditureswerenottruncated.
Finally,planliabilityexpenditureswereconvertedtorelativeplanliabilityexpenditures,whicharedefinedasplanliabilityexpendituresdividedbyadenominator.
Arelativeplanliabilityexpenditureof1.
0correspondstotheaverageplanliabilityexpenditureforthecalibrationsample.
Thedenominatorwascalculatedasfollows.
Fortheentirecalibrationsample,wecalculatedthemeanplanliabilityforeachmetallevelandthentookaweightedaverageofthesemeans,wheretheweightswerebasedonaforecasteddistributionofenrollmentin2014acrossthefivemetallevels.
Goingforward,weusetheterm"planliability"tomean"relativeplanliability.
"Inshort,wesimulatedplanliabilityexpendituresforeachmetaltierfromtotalexpendituresforeachsamplemember(thatis,weapplieddifferentbenefitstructurestothesamesample).
Analternativeapproachwouldhavebeentomodelactualplanliability(payments)forenrolleesinMarketScanplansgroupedintoACAmetaltiersbytheplans'actualactuarialvalues.
However,MarketScanprovidessufficientplanbenefitinformationtocalculateplanactuarialvalueforonlyasmallfractionofitssample.
Also,groupingplansbyactuarialvaluewouldhaveledtodifferentsamplesofindividualsforeachmetallevelmodelestimation,whichwouldhavereducedsamplesizesforeachmodelandledtodifferencesinunmeasuredfactorsacrossmetallevelsamples.
Simulatingplanliabilityonthefullsampleforeachmetalalsomeansthat(asintended)themodelestimatesdonotreflectdifferentialinduceddemand(moralhazard)acrossmetals.
Forthisreason,induceddemandisaccountedforintherisktransferformula,asdiscussedinourcompanionarticle.
DemographicsandDiagnosesTheHHS-HCCriskadjustmentmodeluses2010beneficiarydemographicsanddiagnosestopredict2010(trendedto2014)planliabilityexpendituresforeachbeneficiary.
Thedemographicfactorsemployedareageandsex.
Ageismeasuredasofthelastmonthofenrollment,whichingeneralresultsininfantsaged0havingbeenbornin2010.
7Agerangesweredeterminedbytheagedistributionofthecommercialpopulation,aswellasconsiderationofpost2014marketreformrulesfortheindividualandsmallgroupmarkets.
Thereare18age/sexcategoriesforadultsand8age/sexcategoriesforchildren.
Howageandsexareincorporatedintotheinfantmodelisdescribedbelow.
Adultsaredefinedasages21+,childrenareages2–20,andinfantsareages0–1.
Theagecategoriesforadultmaleandfemaleareages21–24,25–29,30–34,35–39,40–44,45–49,50–54,55–59,and60+.
Theagecategoriesforchildrenmaleandfemaleareages2–4,5–9,10–14,and15–20.
ICD-9-CMperson-leveldiagnosesfrom2010wereusedtocreatediagnosisgroups(HCCs)foreachbeneficiaryinthesample.
OnlydiagnosiscodesfromsourcesallowableforriskadjustmentwhenHHSisoperatingonbehalfofastateareincludedinthediagnosis-levelfile.
Thegoal7Morespecifically,MarketScanincludesageonthefirstdayofenrollmentforthatmonth,andthisishowageismeasured.
Notethatifageforaninfantismeasuredaszeroandtheinfanthasnobirthrecords(inthe2010MarketScandatabase),weexcludedtheinfantfromthesample.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E6MMRR2014:Volume4(3)oftherestrictionsonsourceofdiagnosesistoimprovethequality,accuracy,andauditabilityofdiagnosesusedforriskadjustment.
Forexample,clinicallaboratorydiagnoses,whichinclude"ruleouts"anddiagnosesnotverifiedbyaclinician,wereexcluded.
Allowablediagnosesincludethosefrominpatienthospitalclaims,outpatientfacilityclaims(hospitaloutpatient,ruralhealthclinic,federallyqualifiedhealthcenter,andcommunitymentalhealthclinic),andprofessionalclaims(diagnosesaregenerallynotavailableonprescriptiondrugclaims,includingfortheMarketScandata).
Inaddition,diagnosesfromoutpatientfacilityclaimsandprofessionalclaimsarerestrictedtothosewithatleastoneCPT/HCPCSprocedurecode8correspondinggenerallytoface-to-faceencounterswithaclinician.
SubpopulationsDuetotheinherentclinicalandcostdifferencesintheadult,child,andinfantpopulations,wedevelopedseparateriskadjustmentmodelsforeachgroup.
Theadultandchildmodelshavesimilarspecifications,withage/sexdemographiccategoriesandHCCs(individualHCCsandaggregateHCCgroupings)predictingannualizedplanliabilityexpenditures.
However,infantshavelowfrequenciesformostHCCsleadingtounstableparameterestimatesinanadditivemodel.
Becauseofthis,theinfantmodelutilizesacategoricalapproachinwhichinfantsareassignedabirthmaturity(bylengthofgestationandbirthweight)orAge1category,andadiseaseseveritycategory(basedonHCCsotherthanbirthmaturity).
TherearefourAge0birthmaturitycategories—Extremely8CPTistheCurrentProceduralTerminologymaintainedbytheAmericanMedicalAssociation,andHCPCSistheHealthcareCommonProcedureCodingSystemmaintainedbytheCentersforMedicareandMedicaidServices.
Immature;Immature;Premature/Multiples;Term—andasingleAge1Maturitycategory.
AgezeroinfantsareassignedtooneofthefourbirthmaturitycategoriesandageoneinfantsareassignedtotheAge1Maturitycategory.
Thereare5diseaseseveritycategoriesbasedontheclinicalseverityandassociatedcostsofthenon-maturityHCCs:SeverityLevel5(HighestSeverity)toSeverityLevel1(LowestSeverity).
9Examplesofseveritylevelassignmentsare:Level5—HCC137(HypoplasticLeftHeartSyndromeandOtherSevereCongenitalHeartDisorders);Level4—HCC127(Cardio-RespiratoryFailureandShock,IncludingRespiratoryDistressSyndromes);Level3—HCC45(IntestinalObstruction);Level2—HCC69(AcquiredHemolyticAnemia,IncludingHemolyticDiseaseofNewborn);and,Level1—HCC37(ChronicHepatitis).
Allinfants(age0or1)areassignedtoadiseaseseveritycategorybasedonthesinglehighestseveritylevelofanyoftheirnon-maturityHCCs.
HCCsnotappropriatelydiagnosedforinfants—suchaspregnancyandpsychiatricHCCs—wereexcludedfromtheinfantdiseaseseveritycategories.
InfantswithnoseverityHCCsareassignedtoLevel1.
Whencross-classified,the5maturitycategoriesand5severitycategoriesdefine25mutually-exclusivecategories.
Eachinfantisassignedto1ofthe25categories.
Finally,therearetwoadditivetermsforsex,foragezeromalesandageonemales.
109InassigningHCCstoinfantseveritylevels,theHCChierarchiesaremaintained.
IftwoHCCsareinahierarchicalrelationship,thehigher-rankingHCCisassignedtothesameorahigherseveritylevelthanthelower-rankingHCC.
10Maleinfantshavehighercoststhanfemaleinfantsduetoincreasedmorbidityandneonatalmortality.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E7MMRR2014:Volume4(3)ModelEstimationAllriskadjustmentmodelsareestimatedbyweightedleastsquaresregression.
11Thedependentvariableisannualized,simulated,planliabilityexpenditures,andtheweightistheperson-specific,sampleeligibilityfraction.
Annualizationandweighting—whichareequivalentonanannualbasistopredictingpermemberpermonthexpendituresweightingbythenumberofmonthseachindividualiseligibleforthesample—appropriatelyadjustsformonthsofenrolleeeligibilityinthesample.
Independentvariablesfortheadultmodelinclude18age/sexdemographiccategories,114HCCdiagnosisgroups,and16diseaseinteractions(discussedbelow),andforthechildmodel,8age/sexdemographiccategoriesand119HCCdiagnosisgroups.
Fortheinfantmodel,independentvariablesinclude25categoriesdefinedbybirthmaturityforage0,age1,anddiagnosticseverity,and2age/sexdemographicadditiveterms.
Ineachadultandchildregressionmodel,weincludeabinaryindicatorvariableforeachindividualHCCthatisnotincludedinanaggregateHCCgrouping.
Inaddition,weincludeabinaryindicatorforeachaggregateHCCgrouping.
Inthelattercase,itindicateswhetherornottheenrolleehadatleastoneHCCintheaggregateHCCgrouping.
Inaddition,weimposecoefficientconstraintstoensurethattheprinciplethathigher-clinically-rankedHCCsinanHCChierarchyhaveatleastaslargeincrementalpredictedexpendituresaslower-rankedHCCsismet.
Constraintsgenerallyhavetheeffectofaveragingtwoormoregroupstogetherwhen,unconstrained,thereisaviolationofclinicallogic.
11Weinvestigatedvariousnon-linearapproachestomodelestimationthatmighthavebeenbetterabletoaccountforthenon-linearitiesinplanliability.
However,thesemodelssufferfromseveralimportantshortcomings,includingcomplexity,lackoftransparency,andnotpredictingmeanexpendituresaccuratelyforalldiagnosticanddemographicsubgroups,orevenfortheoverallsample.
Weconcludedthat,evaluatedagainstabroadrangeofcriteriaforreal-worldriskadjustment,weightedleastsquaresisthepreferableestimationmethod.
DiseaseInteractionsFortheadultmodels,theinclusionofdiseaseinteractiontermsbetterreflectedplanliabilityacrossmetallevelsandimprovedmodelperformance.
12Basedonempiricalfindings,aswellasclinicalreview,wedevelopedasetofeightdiagnosticmarkersofsevereillness:HCC2(Septicemia,Sepsis,SystemicInflammatoryResponseSyndrome/Shock);HCC42(Peritonitis/GastrointestinalPerforation/NecrotizingEntercolitis);HCC120(SeizureDisordersandConvulsions);HCC122(Non-TraumaticComa,BrainCompression/AnoxicDamage);HCC125(RespiratorDependence/TracheostomyStatus);HCC126(RespiratoryArrest);HCC127(Cardio-RespiratoryFailureandShock,IncludingRespiratoryDistressSyndromes);andHCC156(PulmonaryEmbolismandDeepVeinThrombosis).
Asevereillnessindicatorvariablewasdefinedashavingatleastoneoftheeightdiagnosticmarkersofsevereillness.
13ThesevereillnessindicatorwasinteractedwithindividualHCCsandaggregateHCCgroupings.
14Thediseaseinteractionsthatmetminimumsamplesizeandincrementalpredictedexpenditurethresholdswereincludedinthemodel.
Theincrementalpredictedexpendituresforthediseaseinteractionswerecategorizedintomediumandhighcostcategories.
Foreachcategory,weincludedabinaryindicatorvariableintheregressionmodelforwhetherornottheenrolleehadatleastonediseaseinteractioninthecategory.
Finally,ahierarchywasimposedsuchthatifanenrolleewasinthehighcostdiseaseinteractioncategory,he/shewasexcludedfromthemediumcostcategory.
12Diseaseinteractionswereempiricallyunimportantinthechildmodelandwerenotincluded.
Theinfantmodelisacategoricalmodel.
13Thediagnosticmarkersofsevereillnessarealsoincludedinthemodelnotinteractedwithotherdiagnoses(HCCs).
14Whenweexaminedacomprehensivesetofinteractions,highfrequency,highincrementalexpenditurediseaseinteractionstendedtoincludesevereillnesses.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E8MMRR2014:Volume4(3)Insum,apersoncanhave,atmost,onediseaseinteractioncoefficient/incrementalpredictedexpenditure.
Thisconstraintwasimposedbecauseclinicalreasoningandempiricalevidenceindicatedthatasingleoneofthediagnosticmarkerssufficedtodistinguishthemostseverelyillpatientsamongthosewiththeunderlyinginteracteddiagnoses.
PredictedPlanLiabilityExpendituresForanenrolleeinagivenmetallevelplan,thetotalpredictedplanliabilityexpendituresisthesumoftheincrementalpredictedplanliabilityexpenditures(coefficients)fromtherelevantmetallevelmodel.
Foradultsandchildren,thisisthesumoftheage/sex,HCC,anddiseaseinteractioncoefficients.
15Forinfants,thisisthesumofthematurity/disease-severitycategoryandadditivesexcoefficients.
Recallthatplanliabilityexpenditureswereconvertedtorelativeplanliabilityexpenditures,resultinginarelativeplanliabilityexpenditureof1.
0fortheaverageplanliabilityexpenditureinthecalibrationsample.
Converting"actual"planliabilityexpenditurestorelativesautomaticallyconverts"predicted"planliabilityexpenditurestorelatives.
Goingforward,weusetheterm"predictedplanliability"tomean"predictedrelativeplanliability.
"ModelEvaluationThepredictiveaccuracyofariskadjustmentmodelforindividualsistypicallyjudgedbythepercentageofvariationinindividualexpendituresexplainedbythemodel(asmeasuredbytheR-squaredstatistic).
TotesttheperformanceoftheHHS-HCCriskadjustmentmodelsforsubgroups,wecalculatetheexpenditureratioofpredictedtoactualweightedmeanplanliabilityexpenditures,whichiscommonlytermedthe"predictiveratio.
"15Thechildriskadjustmentmodelsdonothavediseaseinteractions.
Ifpredictionisperfect,meanpredictedwillequalmeanactualexpenditures,andthepredictiveratiois1.
00.
Asaruleofthumb,predictiveratioswithamarginoferrorof10percentineitherdirection(0.
90≤predictiveratio≤1.
10)indicatereasonablyaccurateprediction(Kautteretal.
,2012).
ResultsSampleExclusionsAsshowninExhibit1,the2010dataincluded45,239,752enrollees.
Afterallexclusionarycriteriawereimposed,theconcurrentsamplecomprised20,040,566enrollees,whichis44.
3percentoftheoriginalsample.
Adults,children,andinfantscomprise,respectively,71.
0percent,27.
1percent,and1.
9percentoftheconcurrentsample.
PlanLiabilityExpendituresMeansimulatedplanliabilityexpenditures(annualized,weighted)perenrolledbeneficiaryrangesfrom1.
369(36.
9%higherthanaverage)fortheplatinumcostsharinglevelto0.
877(12.
3%lowerthanaverage)forthecatastrophiccostsharinglevel(showninExhibit2,decomposedbyadult/child/infantaswellasbymetallevel).
Themedianrangesfrom0.
216forplatinumto0.
000forsilver,bronze,andcatastrophic.
16Thepercentageofindividualswith$0planliabilityincreasesfrom16.
9percentforplatinumto83.
2percentforcatastrophic.
Meanplanliabilityexpendituresarehighestforinfants(e.
g.
,2.
232forsilvermeanplanliability),whichisnotsurprisinggiveninfantshavecostsrelatedtohospitalizationatbirthandcanhavesevereandexpensiveconditionsthatdonot16Everyenrolleewillhaveapositiveplanliabilityriskscore,regardlessofwhetherhe/shehasapositiveplanliabilityexpenditure(theoneexceptionisforchildrenages2–9withoutariskadjustmentmodelHHS-HCCandenrolledinacatastrophicplan—theseenrolleeswillhaveaplanliabilityriskscoreof0—seesectionbelow"ChildRiskAdjustmentModels"andExhibit5andAppendixExhibitA2).
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E9MMRR2014:Volume4(3)Exhibit1.
ExclusionstoCreateHHS-HCCRiskAdjustmentConcurrentModelingSample1CategoryEnrolleesPercentEnrollees,beforeExclusionsPercentEnrollees,afterExclusionsEligiblein2010,beforeexclusions45,239,752100.
0—Exclusions1:notPPOorotherFFSplan6,088,38213.
5—anycapitatedservices1,910,9944.
2—nomentalhealth/substanceabusecoverage15,714,41834.
7—noprescriptiondrugcoverage10,498,69323.
2—motherswithbundlednewbornclaims32,1580.
1—newbornswithnobirthclaims79,5510.
2—Concurrentsample20,040,56644.
3100.
0adultsample(age21–64)14,220,50331.
471.
0childsample(age2–20)5,439,64512.
027.
1infantsample(age0–1)380,4180.
81.
9NOTE:1Exclusionsnotmutuallyexclusive.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
Exhibit2.
DistributionofRelativePlanLiabilityExpenditures1byMetalTierandAgeGroupPlatinumGoldSilverBronzeCatastrophicAdult(age21+)Mean1.
6621.
4971.
3291.
1481.
097Median0.
3040.
2160.
0000.
0000.
000%with$022.
032.
953.
172.
079.
2Child(age2–20)Mean0.
5320.
4540.
3570.
2730.
252Median0.
0870.
0070.
0000.
0000.
000%with$027.
248.
576.
190.
193.
6Infant(age0–1)Mean2.
7062.
5182.
2321.
9181.
842Median0.
7140.
5960.
2570.
0000.
000%with$05.
210.
031.
770.
483.
0NOTES:1Expendituresare2010expenditurestrendedto2014.
Expendituresincludeinpatient,outpatient,andprescriptionexpenditures.
Totalexpendituresincludealloftheseexpenditures.
Simulatedplanliabilityexpendituresreflectstandardizedbenefitdesignsbymetallevel.
Expendituresareannualizedbydividingbytheeligibilityfraction,andexpendituresstatisticsareweightedbythissameeligibilityfraction.
Planliabilityexpendituresareconvertedtorelativeplanliabilityexpenditures.
Arelativeplanliabilityexpenditureof1.
0representstheaverageplanliabilityexpenditureinthecalibrationsample(adult+child+infant).
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
occurinadultsorchildren.
Adultshaveclosetofourtimeshighermeanplanliabilityexpendituresthanchildren(e.
g.
,1.
329vs.
0.
357fortheSilvermetallevel),which,again,isnotsurprisinggivenKautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E10MMRR2014:Volume4(3)thattheonsetofmostchronicconditionsarehighlycorrelatedwithage.
HHS-HCCsAsshowninExhibit3,intheadultconcurrentmodelingsample,only19.
2percentofenrolleeshaveatleastoneHCC,withthevastmajority(79.
2percent)ofthesehavingonlyoneHCC.
Thisresultdoesnotsuggest,however,thattheHCCsareunimportantintheriskadjustmentmodel.
Tothecontrary,whileaminorityoftheadultsamplehasHCCs,themajorityofexpenditurescorrespondtoenrolleeswithHCCs.
Dependingonmetallevel,thepercentageofadultexpenditurescorrespondingtoenrolleeswithatleastoneHCCrangesfrom63.
4percent(platinum)to75.
9percent(catastrophic).
Healthcareexpendituresareconcentratedinasmallproportionofenrolleeswithseriousmedicalproblems,whilethemajorityofthecommercialpopulationisrelativelyhealthy.
Finally,thereissubstantialvariationbyagegroupinthenumberofHCCs,with19.
2percentoftheadultsamplehavingatleastoneHCC,butonly9.
1percentofthechildsample.
AlmosthalfoftheinfantsamplehasatleastoneHCC,whichistobeexpectedgivenapproximatelyhalfofthatsamplearenewbornswithassociatedbirthmaturityHCCs.
AdultRiskAdjustmentModelsThemodelforeachofthemetallevelsiscalibratedonthesameadultconcurrentsample.
Eachmodelincludesthesameindependentvariables:18age-sexcells,114HCCs,17and16diseaseinteractionterms.
Predictedplanliabilityforeachenrolleeisthesumofoneage-sexcoefficient,fromzerotomanyHCCcoefficients(individualHCCsandaggregate17BecauseofHCCgroupings,theeffectivenumberofHHS-HCCsfortheadultriskadjustmentmodelis91.
Exhibit3.
DistributionofHHS-HCCConcurrentSamplebyNumberofPaymentHHS-HCCs1%ofPlanLiabilityExpendituresCountofHCCsEnrollees%ofEnrolleesPlatinumGoldSilverBronzeCatastrophicAdult(age21–64)011,492,63580.
836.
634.
931.
225.
824.
112,160,22015.
232.
132.
032.
833.
433.
52+567,6484.
031.
433.
036.
040.
842.
4Child(age2–20)04,942,58690.
952.
048.
641.
031.
928.
91446,3088.
228.
029.
031.
633.
333.
72+50,7510.
920.
022.
327.
434.
837.
4Infant(age0–1)0209,11655.
015.
313.
89.
85.
74.
71148,66339.
128.
627.
625.
420.
018.
32+22,6396.
056.
158.
664.
874.
377.
0NOTES:1HHS-HCCsisDepartmentofHealthandHumanServices(HHS)-HierarchicalConditionCategories(HCCs).
HHS-HCCsarebasedonICD9-CMdiagnosiscodesfromvalidsources(includinginpatient,hospitaloutpatient,andphysician).
Thereare264HHS-HCCs,amongwhich127HHS-HCCsareusedfortheriskadjustmentmodels.
These127HHS-HCCsincorporate23.
8%(3,439)ofthe14,445ICD9-CMcodes.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E11MMRR2014:Volume4(3)HCCgroupings)subjecttoHCChierarchiesandconstraints/groups,andzerooronesevereillnessdiseaseinteractionterm.
Themodelcoefficientsrepresenttheincremental,nottotal,predictedplanliabilityexpendituresofeachriskmarkerinthemodel,giventheotherriskmarkerscharacterizinganindividual.
Thedependentvariableforeachmodelistheannualizedplanliabilityexpendituressimulatedaccordingtoastandardcostsharingdesignforthatmetallevel.
Exhibit4showsselectedresultsfortheadultriskadjustmentmodelsbymetallevel(forthefullresults,seeAppendixExhibitA1).
ThemodelR-squaresrangebetween36percentfortheplatinummodelto35percentforthecatastrophicmodel.
Thesamplesizeforeachmodelis14,220,503,witheachage/sexcategoryhavingbetween0.
5millionand1millionobservations.
Givensuchlargesamplesizes,allcoefficientsarestatisticallysignificantatconventionalsignificancelevels.
Theage/sexdemographiccoefficientsaremonotonicallyincreasingwithage,andhigherforfemalesineveryagegroup,butespeciallyinthelatterchild-bearingyears(ages35–44).
Thesearethetotalpredictedplanliabilitiesforenrolleeswithout(model)HCCs.
Inaddition,foreachage/sexcategory,theage/sexcoefficientsaredecreasingfromplatinumtocatastrophic.
Forexample,forfemalesage55–59,theagecoefficientdecreasesbymorethanhalf,from1.
054fortheplatinummodelto0.
443forthecatastrophicmodel.
Thelowercoefficientreflectsthehigherenrolleecostsharingand,thus,lowerplanliability,movingfromtheplatinumtocatastrophicplans.
Exhibit4.
SelectedIncrementalRelativePlanLiabilityResultsFromtheHHS-HCCRiskAdjustmentModels—Adultage21+(forfullresults,seeAppendixExhibitA1)R-squared=0.
36020.
35530.
35240.
35050.
3496PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCata-strophicPlanLiabilityHCCNumberVariableLabelCountEstimateEstimateEstimateEstimateEstimateDemographics,MaleAgerange21–24538,6480.
2580.
2080.
1410.
0780.
062Agerange25–29606,6080.
2780.
2230.
1500.
0810.
064Agerange30–34687,8320.
3380.
2740.
1870.
1010.
079Agerange35–39745,6990.
4130.
3390.
2400.
1400.
113Agerange40–44796,8280.
4870.
4040.
2930.
1760.
145Agerange45–49858,8620.
5810.
4870.
3650.
2310.
195Agerange50–54884,0860.
7370.
6260.
4840.
3160.
269Agerange55–59821,6120.
8630.
7360.
5800.
3930.
339Agerange60+830,1191.
0280.
8800.
7040.
4870.
424Demographics,FemaleAgerange21–24569,0870.
4330.
3500.
2210.
1010.
072Agerange25–29674,0340.
5480.
4480.
3010.
1560.
120(Continued)Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E12MMRR2014:Volume4(3)Exhibit4Continued.
SelectedIncrementalRelativePlanLiabilityResultsFromtheHHS-HCCRiskAdjustmentModels—Adultage21+(forfullresults,seeAppendixExhibitA1)PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCata-strophicPlanLiabilityHCCNumberVariableLabelCountEstimateEstimateEstimateEstimateEstimateAgerange30–34749,9380.
6560.
5460.
3960.
2430.
203Agerange35–39798,4750.
7600.
6410.
4900.
3340.
293Agerange40–44863,2560.
8390.
7130.
5540.
3840.
338Agerange45–49954,6590.
8780.
7470.
5830.
4020.
352Agerange50–54991,7821.
0130.
8690.
6950.
4860.
427Agerange55–59931,2701.
0540.
9050.
7260.
5070.
443Agerange60+917,7081.
1560.
9900.
7980.
5590.
489Top10HCCsbyCountHCC021DiabeteswithoutComplication645,5951.
3311.
1991.
1201.
0000.
957HCC088MajorDepressiveandBipolarDisorders401,3771.
8701.
6981.
6011.
4761.
436HCC161Asthma364,0191.
0980.
9780.
9040.
8100.
780HCC020DiabeteswithChronicComplications159,9611.
3311.
1991.
1201.
0000.
957HCC160ChronicObstructivePulmonaryDisease,IncludingBronchiectasis155,4941.
0980.
9780.
9040.
8100.
780HCC012Breast(Age50+)andProstateCancer,Benign/UncertainBrainTumors,andOtherCancersandTumors145,4033.
5093.
2943.
1943.
1413.
121HCC142SpecifiedHeartArrhythmias122,3003.
3633.
1933.
1123.
0633.
046HCC130CongestiveHeartFailure102,1633.
7903.
6483.
5873.
5913.
594(Continued)Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E13MMRR2014:Volume4(3)Exhibit4Continued.
SelectedIncrementalRelativePlanLiabilityResultsFromtheHHS-HCCRiskAdjustmentModels—Adultage21+(forfullresults,seeAppendixExhibitA1)PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCata-strophicPlanLiabilityHCCNumberVariableLabelCountEstimateEstimateEstimateEstimateEstimateHCC056RheumatoidArthritisandSpecifiedAutoimmuneDisorders100,0323.
4143.
1353.
0092.
9872.
982HCC209CompletedPregnancywithNoorMinorComplications82,0773.
7783.
2853.
1342.
9312.
906NOTES:1.
N=14,220,503.
2.
Meanplanliabilityexpendituresforplatinum,gold,silver,bronze,andcatastrophic,respectively:1.
6531.
489,1.
321,1.
142,and1.
091.
3.
Expendituresare2010expenditurestrendedto2014.
Expendituresincludeinpatient,outpatient,andprescriptionexpenditures.
Totalexpendituresincludealloftheseexpenditures.
Simulatedplanliabilityexpendituresreflectstandardizedbenefitdesignsbymetallevel.
Planliabilityexpendituresareconvertedtorelativeplanliabilityexpenditures.
Arelativeplanliabilityexpenditureof1.
0representsthemeanfortheoverallcalibrationsample(adult+child+infant).
Expendituresareannualizedbydividingbytheeligibilityfraction,andregressionmodelsareweightedbythissameeligibilityfraction.
4.
HHS-HCCsistheacronymforDepartmentofHealthandHumanServices(HHS)-HierarchicalConditionCategories(HCCs).
5.
Allcoefficientestimatesarestatisticallysignificantatthe5%levelorlower.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
Fortheadultsilvermodel,HCCcoefficientsrangefrom0.
521(HCC113,CerebralPalsy,exceptQuadriplegic)to78.
175(HCC41,IntestineTransplantStatus/Complications).
ForthefivemostprevalentHCCs,thecoefficientsare1.
120(HCC21,DiabeteswithoutComplications),1.
601(HCC88,MajorDepressiveandBipolarDisorders),0.
904(HCC161,Asthma),1.
120(HCC20,DiabeteswithComplications),and0.
904(HCC160,ChronicObstructivePulmonaryDisease,includingBronchiectasis).
18Asforthediseaseinteractions,thesevereillnesshighcostandmediumcostcategorycoefficientsare12.
427and2.
714,respectively.
Theseamountsareaddedtothepredictedplanliabilityofindividualswho18ThediabetesHCCsweregroupedintoasinglecluster(aggregateHCCgrouping)withthesamecoefficient.
Thus,diabeteswithandwithoutcomplicationshavethesamecoefficient.
havebothaqualifyingunderlyingdisorderandoneofthediagnosticmarkersofsevereillness.
HCCcoefficientsdecreasebymetallevelwhenmovingfromtheplatinummodeltothecatastrophicmodel,buttypicallynotbyasubstantialamount,withthemajoritydecreasingbylessthanhalfthesampleaverageexpenditure(i.
e.
,bylessthan0.
500).
Forexample,thecoefficientfor"HCC130,CongestiveHeartFailure"decreasesonlyfrom3.
790fortheplatinummodelto3.
594forthecatastrophicmodel.
19ThedifferencesintheHCCcoefficientsacrossmetallevelsarenotaspronouncedas19SomeHCCs—thoseassociatedwithlowerexpenditures—doshowlargercoefficientchangesacrossmetals.
Forexample,thecoefficientofthediabetesgroup(HCCs19–21)fallsfrom1.
331inthesimulatedplatinumplanto0.
957inthesimulatedcatastrophicplan.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E14MMRR2014:Volume4(3)thedifferencesintheage/sexcoefficients.
Thisoccursbecausetheage-sexcoefficientsrepresenttheentirepredictedliabilityforpersonswithoutHCCs,whoarerelativelyhealthy.
Theplan'sliabilityfortheirlowerexpendituresisgreatlyreducedbytheincreaseinthedeductibleacrossthesimulatedmetallevelplans.
Incontrast,muchofthespendingforpersonswithHCCs,especiallythemoreexpensiveones,occursabovetheplandeductibleandevenabovetheplanout-of-pocketmaximum,andthusislessaffectedbythechangeincostsharingwhenmovingacrossmetallevels.
Theupshotisthatpredictedplanliability,andhencetheriskscore,aremorestable(proportionately)acrossmetallevelsforverysickindividuals,whilepredictedplanliability/riskscoreforhealthyindividualsismuchlowerinthebronzeorcatastrophicplansthanintheplatinumorgoldplans.
20,21Inotherwords,planswillincurasignificantliabilityforverysickpeopleeveniftheyhavehigherlower-endcostsharing;buttheirproportionateliabilityforrelativelyhealthypeoplewillbemuchlower.
ChildRiskAdjustmentModelsEachofthefivemetallevelmodelsiscalibratedonthesamechildconcurrentsample.
Eachmodelincludesthesameindependentvariables:eightage-sexcellsand119HCCs.
22Diseaseinteractionswereempiricallyunimportantforthechildmodeland20Allindividuals,includingverysickones,receiveanage-sexcoefficientaspartoftheirpredictedplanliability.
Thus,theirpredictionsaresubjecttothesameabsolutechangesinplanliabilitywhenmovingacrossmetallevels.
However,becauseHCCcoefficientscomprisethelargestportionofthepredictedliabilityofverysickindividuals,proportionately(percentage-wise)theirtotalpredictionislessaffectedbymetallevel.
21Thesevereillnessdiseaseinteractioncoefficientsarefairlystableacrossmetals,butriseslightlywithgreatercostsharing.
Thismayoccurbecausetheindividualdisease(HCC)andaggregatedisease(HCC)groupingcoefficientsdeclineacrossmetals,andthesevereillnessinteractionsarepickingupmoreofthecostsoftheveryexpensivepeopleinthemetalswithhighercostsharing.
22BecauseofaggregateHCCgroupings,theeffectivenumberofHHS-HCCsforthechildriskadjustmentmodelis100.
werenotincluded.
Thedependentvariableforeachmodelistheannualizedplanliabilityexpendituressimulatedaccordingtoastandardcostsharingdesignforthatmetallevel.
Predictedplanliabilityforeachchildisthesumofoneage-sexcoefficientandzerotomanyHCCcoefficients,eachofwhichrepresentsanincrementalexpenditure.
23Exhibit5showsselectedresultsforthechildriskadjustmentmodelsbymetallevel(forthefullresults,seeAppendixExhibitA2).
ThemodelR-squaresforeachofthe5metallevelsrangebetween31percentfortheplatinummodelto30percentforthecatastrophicmodel.
TheseR-squaresareapproximately5percentagepointslowerthantheR-squaresfortheadultmodels.
Thiscanbeexplainedpartiallybynotingthatlessthan10percentofthechildsamplehasanyHCCs,whicharethemainpredictorsofindividualvariationinplanliabilityexpenditures.
Thesamplesizeforeachmodelis5,439,645,witheachage/sexcategoryhavingbetween362,777and921,236observations.
Givensuchlargesamplesizes,exceptfortheyoungestage/sexcategories(age2–4,age5–9)forthelowestmetallevels(bronze,catastrophic),allcoefficientsarestatisticallysignificantatconventionalsignificancelevels.
Theage/sexdemographiccoefficientshaveaU-shapedpattern,unlikethemonotonicallyincreasingcoefficientsofadults.
Forexample,formalesinthesilvermodel,theage/sexcoefficientsare0.
106forage2–4,0.
064forage5–9,0.
110forage10–14,and0.
191forage15–20.
Femalechildrenarelessexpensivethanmalechildrenuntilages15–20,whichisperhapswhenreproductivehealthexpensesbegintobecomemorepronounced.
Similartotheadultmodel,theage/sexcoefficientsdecreasefromplatinumtocatastrophic.
2423Theriskscoreforeachchildisthesumofhis/herrelativecoefficients.
Seeabovefordetails.
24Thezerocoefficientsforages2–9inthecatastrophicmodelindicatethatthemodelpredictsnegligibleexpendituresabovethedeductibleforchildrenoftheseageswithoutanyoftheriskadjustmentmodelHCCs.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E15MMRR2014:Volume4(3)Forthechildsilvermodel,HCCcoefficientsrangefrom0.
354(HCC161,Asthma;andHCC160,ChronicObstructivePulmonaryDisease,includingBronchiectasis)to106.
991(HCC41,IntestineTransplantStatus/Complications).
ForthefivemostprevalentHCCs,thecoefficientsare0.
354(HCC161,Asthma),1.
453(HCC88,MajorDepressiveandBipolarDisorders),1.
882(HCC120,SeizureDisordersandConvulsions),2.
198(HCC21,DiabeteswithoutComplication),and1.
372(HCC102,AutisticDisorder).
ThreeofthefivemostprevalentHCCsarethesameintheadultandchildsamples.
However,theincrementalpredictedexpendituresaremarkedlydifferent,illustratingtheclinicalandcostdifferencesamongthetwopopulations,whichwereamajorreasonfordevelopingseparateadultandchildmodels.
Thechildsilvermodelcoefficientfor"HCC161,Asthma"islessthanhalftheadultcoefficient(0.
354vs.
0.
904);thechildcoefficientfor"HCC21,DiabeteswithoutComplications"isalmostdoubletheadultcoefficient,perhapsreflectingthegreaterExhibit5.
SelectedIncrementalRelativePlanLiabilityResultsfromtheHHS-HCCRiskAdjustmentModels—Childage2–20(forfullresults,seeAppendixExhibitA2)R-squared=0.
30670.
30240.
29930.
29620.
2950PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCata-strophicPlanLiabilityHCCNumberVariableLabelCountEstimateEstimateEstimateEstimateEstimateDemographics,MaleAgerange2–4380,8410.
2830.
2090.
1060.
0190.
000Agerange5–9688,4990.
1960.
1400.
0640.
0050.
000Agerange10–14749,9820.
2460.
1890.
1100.
0470.
033Agerange15–20955,9720.
3360.
2730.
1910.
1140.
095Demographics,FemaleAgerange2–4362,7770.
2330.
1650.
0710.
0190.
000Agerange5–9660,7170.
1650.
1130.
0480.
0050.
000Agerange10–14719,6210.
2230.
1680.
0950.
0420.
031Agerange15–20921,2360.
3790.
3040.
1980.
1010.
077Top10HCCsbyCountHCC161Asthma260,4350.
5210.
4580.
3540.
2150.
175HCC088MajorDepressiveandBipolarDisorders67,7381.
7791.
5911.
4531.
2521.
188HCC120SeizureDisordersandConvulsions30,3662.
1882.
0121.
8821.
7021.
644HCC021DiabeteswithoutComplication14,0422.
6292.
3542.
1981.
9041.
799(Continued)Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E16MMRR2014:Volume4(3)Exhibit5Continued.
SelectedIncrementalRelativePlanLiabilityResultsfromtheHHS-HCCRiskAdjustmentModels—Childage2–20(forfullresults,seeAppendixExhibitA2)PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCata-strophicPlanLiabilityHCCNumberVariableLabelCountEstimateEstimateEstimateEstimateEstimateHCC102AutisticDisorder12,3551.
6731.
5001.
3721.
1771.
112HCC138MajorCongenitalHeart/CirculatoryDisorders11,2172.
2572.
1432.
0181.
8701.
828HCC103PervasiveDevelopmentalDisorders,ExceptAutisticDisorder9,8520.
9630.
8500.
7230.
5110.
441HCC139AtrialandVentricularSeptalDefects,PatentDuctusArteriosus,andOtherCongenitalHeart/CirculatoryDisorders9,0171.
4111.
3191.
2061.
0781.
047HCC062Congenital/DevelopmentalSkeletalandConnectiveTissueDisorders6,9781.
5361.
4101.
3111.
2111.
183HCC030Adrenal,Pituitary,andOtherSignificantEndocrineDisorders6,9746.
1775.
8675.
6965.
6425.
625NOTES:1.
N=5,439,645.
2.
Meanplanliabilityexpendituresforplatinum,gold,silver,bronze,andcatastrophic,respectively:0.
532,0.
454,0.
357,0.
273,and0.
252.
3.
Expendituresare2010expenditurestrendedto2014.
Expendituresincludeinpatient,outpatient,andprescriptionexpenditures.
Totalexpendituresincludealloftheseexpenditures.
Simulatedplanliabilityexpendituresreflectstandardizedbenefitdesignsbymetallevel.
Planliabilityexpendituresareconvertedtorelativeplanliabilityexpenditures.
Arelativeplanliabilityexpenditureof1.
0representsthemeanfortheoverallcalibrationsample(adult+child+infant).
Expendituresareannualizedbydividingbytheeligibilityfraction,andregressionmodelsareweightedbythissameeligibilityfraction.
4.
HHS-HCCsistheacronymforDepartmentofHealthandHumanServices(HHS)-HierarchicalConditionCategories(HCCs).
5.
Allnon-zerocoefficientestimatesarestatisticalysignificantatthe5%levelorlower.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E17MMRR2014:Volume4(3)severityofTypeIversusTypeIIdiabetes(2.
198vs.
1.
120);andthechildcoefficientfor"HCC88,MajorDepressiveandBipolarDisorders"isrelativelysimilarinmagnitudetotheadultcoefficient(1.
453vs.
1.
601).
Someothernotablyhigherchildversusadultsilvercoefficientsare:"HCC112QuadriplegicCerebralPalsy"(5.
223childvs.
1.
681adult);"HCC159CysticFibrosis"(12.
743childvs.
9.
957adult);and"HCC102AutisticDisorder"(1.
372childvs.
0.
974adult).
Finally,liketheadultmodel,theHCCcoefficientsinthechildmodeldecreasewhenmovingfromtheplatinummodeltothecatastrophicmodel,butoftennotbyasubstantialamount.
InfantRiskAdjustmentModelsAsdescribedpreviously,theinfantmodelutilizesacategoricalapproachinwhichinfantsareassignedabirthmaturity(bylengthofgestationandbirthweight)orAge1category,andadiseaseseveritycategory(basedonHCCsotherthanbirthmaturity).
Exhibit6showstheestimatedinfantriskadjustmentmodelsbymetallevel.
ThemodelR-squaresare29percentacrossthefivemetallevelsintheinfantmodel,whichareslightlylowerthanthechildmodelR-squares.
Thesamplesizeforeachmodelis380,418,with90percentofobservationsinthe"TermxSeverityLevel1"category(n=121,841)orthe"Age1xSeverityLevel1"category(n=219,105).
Theremainingcategories(exceptfortheMaleAdditiveterms)eachhavefewerthan10,000observations.
Infact,samplesizesforahandfulofcategoriesarelessthan100,whichrequiredcoefficientconstraintstoimprovestatisticalprecision.
Predictedplanliabilityforeachinfantisthecoefficientofhisorhersinglecategory[(maturity)x(diseaseseverity)]plus,ifmale,thecoefficientoftheAge0orAge1MaleAdditiveTerm.
2525Theriskscoreforeachinfantisthesumofhis/herrelativecoefficients.
Seeabovefordetails.
Fortheinfantsilvermodel,predictedplanliabilityforage0femaleinfantsrangesfrom391.
387forthe"ExtremelyImmaturexSeverityLevel5"category,to0.
998forthe"TermxSeverityLevel1"category.
Thus,thepredictedplanliabilityforanextremelyimmatureinfantwiththehighestdiseaseseveritylevelisalmost400timesthepredictedplanliabilityforaterminfantwiththelowestdiseaseseveritylevel.
Forage1femaleinfants,predictedplanliabilityrangesfrom61.
217forthe"Age1xSeverityLevel5"categoryto0.
333forthe"Age1xSeverityLevel1"category.
The"Age0,Male"and"Age1,Male"AdditiveTermsare0.
574and0.
094,respectively.
Withineachmaturitylevel,predictedplanliabilityisincreasinginseverity(orisequalwhensmallsamplesizesrequireseveritylevelstobecombinedinestimation).
Also,forage0infants,withineachseveritylevel,predictedplanliabilityincreaseswithgreaterimmaturity.
Theinfantmodelpredictedplanliability,the(maturity)x(diseaseseverity)coefficients,decreasewithgreaterplanenrolleecostsharing(movingfromplatinumtocatastrophicplans).
But,proportionately,thereductionismuchlargerforthelessexpensivecategories.
Forexample,the(Term)x(SeverityLevel5)predictedplanliabilityfallsonlyfrom132.
588(platinum)to130.
292(catastrophic).
Butthe(Term)x(SeverityLevel1)predictedplanliabilityfallsfrom1.
661(platinum)to0.
188(catastrophic).
Thiscanbeexplainedbythelargedifferenceindeductiblesinthestandardbenefitdesignsusedtosimulateplanliabilityexpenditures,whichhaveamuchlargerproportionateeffectonthelower-expenditurecategories.
EvaluationInevaluatingthemodels'performancewelookatbothexplanatorypowerattheindividuallevelandunder-andover-predictionforsubgroupsofthepopulation.
WeevaluatemodelpredictiveaccuracyusingourMarketScancalibrationsample.
WhileKautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E18MMRR2014:Volume4(3)Exhibit6.
HHS-HCCRiskAdjustmentModels—Infant(age0–1)RelativePlanLiabilityResultsR-squared=0.
29160.
28930.
28840.
28850.
2885PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCata-strophicPlanLiabilityVariableCountEstimateEstimateEstimateEstimateEstimateAGE0(allage0infantsareassignedtoexactly1ofthese20mutually-exclusivecategories)ExtremelyImmature*SeverityLevel5178393.
816392.
281391.
387391.
399391.
407ExtremelyImmature*SeverityLevel4513225.
037223.
380222.
424222.
371222.
365ExtremelyImmature*SeverityLevel35560.
36359.
23258.
53258.
24758.
181C1ExtremelyImmature*SeverityLevel2260.
36359.
23258.
53258.
24758.
181C1ExtremelyImmature*SeverityLevel112160.
36359.
23258.
53258.
24758.
181C1Immature*SeverityLevel5144207.
274205.
589204.
615204.
629204.
644Immature*SeverityLevel41,63889.
69488.
10587.
18887.
16987.
178Immature*SeverityLevel324345.
71544.
30543.
50343.
39443.
379Immature*SeverityLevel26933.
58532.
24731.
44931.
22131.
163C2Immature*SeverityLevel11,26433.
58532.
24731.
44931.
22131.
163C2Premature/Multiples*SeverityLevel5213173.
696172.
095171.
169171.
111171.
108Premature/Multiples*SeverityLevel42,20534.
41732.
98132.
15531.
96031.
925Premature/Multiples*SeverityLevel363418.
50217.
38216.
69416.
31116.
200Premature/Multiples*SeverityLevel23719.
3628.
5337.
9677.
4117.
241Premature/Multiples*SeverityLevel19,1896.
7636.
1445.
5994.
9614.
771Term*SeverityLevel5377132.
588131.
294130.
511130.
346130.
292Term*SeverityLevel44,14620.
28319.
22218.
56018.
08217.
951Term*SeverityLevel33,8186.
9156.
2865.
7655.
0924.
866Term*SeverityLevel23,4403.
8253.
3932.
9252.
1891.
951Term*SeverityLevel1121,8411.
6611.
4490.
9980.
3390.
188AGE1(allage1infantsareassignedtoexactly1ofthese5mutually-exclusivecategories)Age1*SeverityLevel543262.
38561.
65761.
21761.
13061.
108Age1*SeverityLevel42,50910.
85510.
3349.
9889.
7479.
686Age1*SeverityLevel33,6383.
6333.
2993.
0072.
6922.
608(Continued)Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E19MMRR2014:Volume4(3)Exhibit6Continued.
HHS-HCCRiskAdjustmentModels—Infant(age0–1)PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCata-strophicPlanLiabilityVariableCountEstimateEstimateEstimateEstimateEstimateAge1*SeverityLevel24,2732.
1771.
9301.
6651.
3201.
223Age1*SeverityLevel1219,1050.
6310.
5310.
3330.
1710.
137AGE0MaleAdditiveTerm(allage0maleshavethistermaddedtotheirassociatedage0categorycoefficient)Age0Male77,6420.
6290.
5870.
5740.
5330.
504AGE1MaleAdditiveTerm(allage1maleshavethistermaddedtotheirassociatedage1categorycoefficient)Age1Male117,6660.
1170.
1020.
0940.
0650.
054NOTES:1.
N=380,418.
2.
Meanplanliabilityexpendituresforplatinum,gold,silver,bronze,andcatastrophic,respectively:2.
706,2.
518,2.
232,1.
918,and1.
842.
3.
Expendituresare2010expenditurestrendedto2014.
Expendituresincludeinpatient,outpatient,andprescriptionexpenditures.
Totalexpendituresincludealloftheseexpenditures.
Simulatedplanliabilityexpendituresreflectstandardizedbenefitdesignsbymetallevel.
Planliabilityexpendituresareconvertedtorelativeplanliabilityexpenditures.
Arelativeplanliabilityexpenditureof1.
0representsthemeanfortheoverallcalibrationsample(adult+child+infant).
Expendituresareannualizedbydividingbytheeligibilityfraction,andregressionmodelsareweightedbythissameeligibilityfraction.
4.
HHS-HCCsisDepartmentofHealthandHumanServices(HHS)-HierarchicalConditionCategories(HCCs).
5.
Regressionmodelcoefficientconstraintswereappliedasfollows:C1:TheExtremelyImmatureinteractionsforSeverityLevels3and2wereconstrainedtoSeverityLevel1.
C2:TheImmatureinteractionforSeverityLevel2wasconstrainedtoSeverityLevel1.
6.
Allcoefficientestimatesarestatisticallysignificantatthe5%levelorlower,exceptfor:i)age1maleadditivecoefficientforallmodels,ii)term*severitylevel1coefficientforthecatastrophicmodel,whichisstatisticallysignificantonlyatthe6%level(p-value=0.
0536).
7.
Severitylevel5isthehighestseveritylevel,andseveritylevel1isthelowest.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
webelievethattheevaluationresultsfromthisverylargeandnationallydisperseddatabaseareinformativeandrepresentativeonaverage,ourevaluationresultsdonotnecessarilygeneralizeperfectlytoeachindividualstate'sACAriskadjustmentpopulationorplans.
Toevaluatethepredictiveaccuracyofthemodelsforindividuals,weexaminethemodels'R-squaredstatistics.
Thesewerebetween35and36percentfortheadultmodels,between30and31percentforthechildmodels,and29percentfortheinfantmodels(Exhibits4,5,&6).
Incomparison,thepredictivepowerofdemographic-onlymodelsisrelativelylow,generallylessthan2percent.
Addinginformationaboutdiagnosessubstantiallyimprovesthepredictivepowerofthemodels.
Further,thepredictivepoweroftheconcurrentdiagnosis-basedmodelspresentedheresubstantiallyexceedsthepredictiveabilityforindividualsofprospectivediagnosis-basedmodels(e.
g.
,theMedicareCMS-HCCriskadjustmentmodel),whichtypicallyhaveR-squaredstatisticsof10–15percent.
TheR-squaredstatisticsoftheHHS-HCCmodelsarewithintherangeofR-squaredstatisticsofotherconcurrentmodelspredictingexpendituresforcommercialinsuranceenrollees(Winkelman&Mehmud,2007).
However,althoughpredictiveaccuracyisanimportantgoalinmodeldevelopment,theHHS-HCCKautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E20MMRR2014:Volume4(3)modelsarenotdevelopedpurelytomaximizethevalueoftheR-squaredstatistic.
Instead,theHHS-HCCmodelsareintendedtobalancehighpredictiveabilitywithlowersensitivitytodiscretionarydiagnosticcoding.
ThelatterisprimarilyachievedbyincludingonlyasubsetoflessdiscretionaryHCCsthatidentifychronicorsystematicconditionssubjecttoinsuranceriskselectionratherthanbeingrandomacuteevents.
Inaddition,HCCsthatprimarilyrepresentcomplicationsoforpoorqualityofcare(e.
g.
,pressureulcers)areexcluded.
Itisalsoimportanttoassessaggregatepredictiveaccuracyfordefinedsubgroupsofhealthplanenrollees.
Thisanalysisevaluateswhetherthemodelpredictsliabilityaccuratelyforplansenrollingdifferenttypesofpeople,andwhetheroncethemodelisimplemented,planshaveanyincentivestoavoidorenrollcertaintypesofindividuals,forexample,thosewithhighhealthcarecostsorcertainmedicalconditions.
Inthecalibrationsample,themodelspredictmeanplanliabilityexpendituresperfectly(predictiveratio=1.
00)foreachoftheagegroupsubpopulations(adult,child,infant)foreachlevelofplancostsharing(platinum,gold,silver,bronze,catastrophic).
Notonlythat,predictionisperfectforeachoftheincludeddemographic(age/sexcategories)anddiagnosticfactors(HCCdiagnosisgroups)foreachsubpopulation.
Thisisexpected,giventhespecificationandstatisticaltechniquesusedtoestimatethemodel.
However,giventheirclinicalandcostdifferences,predictingaccuratelyonaverageforthesesubpopulationsisimportant.
Forexample,themodelaccountsfortheveryhighincrementalhealthcarecostsofchildrenwithhemophilia(45.
551—relativeincrementalplanliabilityestimateinchildsilvermodel).
Basingrisktransferpaymentsandchargesonaccurateestimatesofthedifferentialcostsbysubpopulationwillhelpensurethatplansintheindividualandsmallgroupmarketsreceiveadequatepaymentstotreatenrolleeswithhighexpectedcosts.
Wealsotestedthepredictiveaccuracyofthemodelsusingenrolleegroupssortedintopredictedexpenditurepercentileranges(0–40%,40–80%,80–100%,top10%,top5%,top1%).
Thissetofratiosdetermineswhetherthemodelpredictionsareaccurateatvariouslevelsofpredictedexpenditures;thatis,itdetermineswhetherexpendituresthemodelpredictstobelowareinfactlowonaverage,andwhetherexpendituresthemodelpredictstobehighareinfacthighonaverage.
Wechosethissetofpercentileranges(whichwerefertosimplyas"percentiles")notonlytocovertheentirerangeofpredictedexpenditures,buttoemphasizethehigherpercentilesthatcapturethesmallproportionofhigh-costindividualsinwhichmostmedicalexpendituresareconcentrated.
Accuratemodelpredictionisespeciallycriticalforthesehigh-costcases.
Fortheadultsample,Exhibit7presentspredictiveratiosforpercentilesofenrolleescreatedbysortingpredictedplanliabilityexpenditures.
Theadultplatinummodelpredictswellforthesepredictedexpendituregroups.
Thereislessthana10percentpredictionerrorineitherdirectionforeachofthesegroups,rangingfromlower-costtovery-high-costindividuals.
Thelowerpercentiles,0–40%and40–80%,aresomewhatunder-predicted,whereasthehighestpercentiles(80–100%,top10%,top5%,top1%)aresomewhatover-predicted.
Theadultmodelsperformadequatelyacrossallmetallevels,doingespeciallywellforthecriticalhighestpercentiles.
Forexample,forthe80–100%percentile,thepredictiveratiosrangefrom1.
04(platinum)to1.
07(catastrophic).
Themeanactualplanliabilityexpendituresforenrolleesinthe80–100%percentilerangefrom5.
012to3.
944acrossmetaltiers,whichrepresents,respectively,Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E21MMRR2014:Volume4(3)Exhibit7.
PredictiveRatiosbyPercentilesofPredictedExpenditures—AdultModelsPercentiles(sortedbypredicted$)0–40%40–80%80–100%top10%top5%top1%PlatinumPredicted$0.
4670.
9275.
2188.
28012.
57231.
630Actual$0.
5170.
9885.
0127.
88611.
86030.
531PredictiveRatio0.
900.
941.
041.
051.
061.
04%ofOverallActual$11.
824.
264.
050.
838.
119.
1GoldPredicted$0.
3850.
7914.
8477.
79411.
99830.
813Actual$0.
4370.
8574.
6287.
36811.
24129.
658PredictiveRatio0.
880.
921.
051.
061.
071.
04%ofOverallActual$11.
123.
365.
652.
740.
120.
5SilverPredicted$0.
2740.
6254.
5717.
47311.
63430.
337Actual$0.
3300.
6934.
3397.
03510.
85929.
120PredictiveRatio0.
830.
901.
051.
061.
071.
04%ofOverallActual$9.
521.
369.
356.
743.
722.
7BronzePredicted$0.
1600.
4314.
2967.
20611.
39630.
188Actual$0.
2270.
5054.
0356.
75210.
61828.
983PredictiveRatio0.
710.
851.
061.
071.
071.
04%ofOverallActual$7.
517.
974.
662.
949.
426.
2CatastrophicPredicted$0.
1300.
3764.
2167.
13111.
32830.
148Actual$0.
2000.
4523.
9446.
67110.
54528.
947PredictiveRatio0.
650.
831.
071.
071.
071.
04%ofOverallActual$6.
916.
876.
365.
151.
427.
4NOTES:1.
Predicted$aremeanrelativepredictedannualizedplanliabilityexpendituresforpercentilegroup.
2.
Actual$aremeanrelativeactualannualizedplanliabilityexpendituresforpercentilegroup.
3.
Predictiveratioispredicted$dividedbyactual$.
4.
%ofoverallactual$isweightedsumofactual$forpercentilegroupdividedbyweightedsumofactual$acrossentireadultsample,foreachmetaltier.
5.
Foragivenmodel,percentilesaresortedbypredicted$forthatmodel.
6.
Adultsareage21+.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E22MMRR2014:Volume4(3)64.
0percentto76.
3percentofoverallmeanactualplanliabilityexpenditures.
Sincemostofthedollarsareinthehighestpercentiles,itismostimportantforthemodeltoperformwellforthesehighcostsubgroups.
Theadultmodelsperformlesswellforthelowestpercentiles,especiallyforthelowermetallevels.
Forexample,forthe0–40%percentile,thepredictiveratioforthecatastrophicmodelisonly0.
65.
However,theenrolleescomprisingthe0–40%percentilerepresentonly6.
9percentofoverallactualexpendituresforthecatastrophicmetallevel.
Moreover,theabsoluteamountoftheunder-prediction,0.
130forpredictedexpendituresversus0.
200foractualexpendituresforadifferenceof0.
070,issmall.
Thepredictiveratioislow,inpart,becausethedenominatoroftheratio,0.
200(1/5oftheaveragepredictedexpendituresforthecalibrationsample),issmallfortheselow-costbeneficiaries,magnifyingtheabsolutepredictionerrorwhenexpressedasaratio.
Forthecatastrophicmetal,asfortheothermetals,theHHS-HCCmodelpredictsawiderangeofplanliabilitiesacrossgroups,from0.
130to30.
148(0–40%percentilevs.
top1%percentile),correspondingtoasimilarrangeofactualplanliabilitiesrangingfrom0.
200to28.
948.
Thepredictiveratiosforthechildmodels(Exhibit8)exhibitthesamequalitativepatternsasfortheadultmodels,exceptthatthepredictiveratiosdenotelesspredictiveaccuracy.
Forthechildplatinummodel,thereislessthana20percenterrorforeachpercentile(exceptforthetop1%percentile).
Liketheadultmodels,thechildmodelperformslesswellforthelowestpercentiles,especiallyforthelowermetallevels.
However,itisimportanttoconsidertheamountofactual(relative)dollarsthesepercentilesrepresent.
Forexample,forthecatastrophicmodel,whilethe0–40%percentilehasapredictiveratioof0.
08,theabsolutedifferenceofpredictedandactual(relative)expendituresisonly0.
049(predictedexpenditures0.
004;actualexpenditures0.
053),andonly8.
4percentofoverallexpendituresofthecatastrophicmetallevelisincurredbythelowestpercentilegroup.
Finally,theinfantmodelsperformquiteaccuratelyonthepredictiveratiosforpredictedexpenditurepercentiles(Exhibit9).
Ingeneral,thereisa5percentpredictionerrororsmalleracrossallpercentilesandallmetallevels.
Thetwoexceptionsarethe40–80%percentileforthebronzemodel(predictiveratio=0.
90)andthe0–40%percentileforthecatastrophicmodel(predictiveratio=0.
80).
Butagain,thedollaramountsoftheunder-predictionsaremodestandthesepercentilescompriseasmallshareoftotalactualexpenditures,7.
3percentforthe40–80%percentileforbronze,and4.
2percentforthe0–40%percentileforcatastrophic.
RiskScoreCalculationBelowweprovideseveralexamplesofhowempiricalriskadjustmentmodeloutputisappliedtocalculateanindividual's"planliabilityriskscore(PLRS)".
WethendefinetheplanaveragePLRS,whichisusedinthecalculationoftransferpaymentsandcharges.
IntheHHSmethodology,theriskscoreforanenrolleeisdefinedasthepredictedrelativeplanliabilityexpenditurefortheenrolleebasedontheHHS-HCCriskadjustmentmodelfortheenrollee'splanmetallevel.
Thepredictedrelativeplanliabilityexpendituresarecalculatedasfollows.
Foranadult(age21+),itisthesumoftheage/sex,HCC,anddiseaseinteractionriskfactorsinAppendixExhibitA1;forachild,itisthesumoftheage/sexandHCCriskfactorsinAppendixExhibitA2;andforinfants,itisthesumoftheappropriatematurity/disease-severitycategoryandage/sexAdditiveTerminExhibit6.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E23MMRR2014:Volume4(3)Exhibit8.
PredictiveRatiosbyPercentilesofPredictedExpenditures—ChildModelsPercentiles(sortedbypredicted$)0–40%40–80%80–100%top10%top5%top1%PlatinumPredicted$0.
2000.
3021.
6322.
8014.
81713.
928Actual$0.
2430.
3391.
4772.
4554.
08711.
049PredictiveRatio0.
820.
891.
101.
141.
181.
26%ofOverallActual$18.
225.
456.
448.
541.
122.
3GoldPredicted$0.
1440.
2381.
4872.
5894.
51413.
467Actual$0.
1870.
2751.
3312.
2423.
77610.
502PredictiveRatio0.
770.
871.
121.
161.
201.
28%ofOverallActual$16.
424.
159.
551.
844.
424.
9SilverPredicted$0.
0690.
1511.
3252.
3774.
26413.
155Actual$0.
1140.
1881.
1652.
0203.
51410.
176PredictiveRatio0.
610.
801.
141.
181.
211.
29%ofOverallActual$12.
721.
066.
359.
552.
630.
7BronzePredicted$0.
0140.
0761.
1752.
1574.
00512.
955Actual$0.
0660.
1140.
9951.
7813.
2229.
955PredictiveRatio0.
210.
661.
181.
211.
241.
30%ofOverallActual$9.
716.
873.
668.
463.
139.
2CatastrophicPredicted$0.
0040.
0581.
1342.
0953.
93112.
897Actual$0.
0530.
0970.
9511.
7153.
1399.
889PredictiveRatio0.
080.
601.
191.
221.
251.
30%ofOverallActual$8.
415.
476.
271.
366.
642.
2NOTES:1.
Predicted$aremeanrelativepredictedannualizedplanliabilityexpendituresforpercentilegroup.
2.
Actual$aremeanrelativeactualannualizedplanliabilityexpendituresforpercentilegroup.
3.
Predictiveratioispredicted$dividedbyactual$.
4.
%ofoverallactual$isweightedsumofactual$forpercentilegroupdividedbyweightedsumofactual$acrossentirechildsample,foreachmetaltier.
5.
Foragivenmodel,percentilesaresortedbypredicted$forthatmodel.
6.
Childrenareages2–20.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
Basedonlowerincomeorcertainotherqualifyingfactors,someenrolleesinMarketplaceplanswillbeeligibleforreducedcostsharinginadditiontopremiumsubsidies.
Anadjustmentwillbemadetotheriskscoreforenrolleesinindividualmarketcost-sharingplanvariationsinKautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E24MMRR2014:Volume4(3)Exhibit9.
PredictiveRatiosbyPercentilesofPredictedExpenditures—InfantModelsPercentiles(sortedbypredicted$)0–40%40–80%80–100%top10%top5%top1%PlatinumPredicted$0.
6671.
24612.
56820.
73238.
300123.
514Actual$0.
6751.
28112.
46120.
73838.
209123.
716PredictiveRatio0.
990.
971.
011.
001.
001.
00%ofOverallActual$12.
217.
070.
965.
757.
736.
2GoldPredicted$0.
5631.
09012.
27620.
73237.
294122.
116Actual$0.
5701.
12712.
16420.
71337.
202122.
314PredictiveRatio0.
990.
971.
011.
001.
001.
00%ofOverallActual$11.
016.
272.
867.
960.
338.
5SilverPredicted$0.
3630.
75911.
33919.
21236.
663121.
304Actual$0.
3690.
79711.
23219.
20936.
571121.
500PredictiveRatio0.
980.
951.
011.
001.
001.
00%ofOverallActual$8.
112.
779.
375.
066.
943.
2BronzePredicted$0.
1910.
35410.
79118.
76736.
307121.
218Actual$0.
1940.
39210.
69518.
76536.
218121.
415PredictiveRatio0.
980.
901.
011.
001.
001.
00%ofOverallActual$4.
97.
387.
885.
377.
150.
2CatastrophicPredicted$0.
1470.
24810.
63818.
63236.
199121.
194Actual$0.
1830.
24710.
54618.
62936.
113121.
391PredictiveRatio0.
801.
011.
011.
001.
001.
00%ofOverallActual$4.
25.
790.
288.
280.
152.
3NOTES:1.
Predicted$aremeanrelativepredictedannualizedplanliabilityexpendituresforpercentilegroup.
2.
Actual$aremeanrelativeactualannualizedplanliabilityexpendituresforpercentilegroup.
3.
Predictiveratioispredicted$dividedbyactual$.
4.
%ofoverallactual$isweightedsumofactual$forpercentilegroupdividedbyweightedsumofactual$acrossentireinfantsample,foreachmetaltier.
.
5.
Foragivenmodel,percentilesaresortedbypredicted$forthatmodel.
6.
Infantsareage0–1.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
Marketplaces(PatientProtectionandAffordableCareAct,2013).
Individualswhoqualifyforcostsharingreductionsmayutilizehealthcareservicesatahigherratethanwouldbethecaseintheabsenceofcostsharingreductions.
TheadjustmentforinduceddemandduetocostsharingKautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E25MMRR2014:Volume4(3)reductionswillbemultiplicativeandappliedtotheriskscore.
26Becausepremiumsforallcost-sharingreductionplanvariationsarerequiredtobethesame,despitetheincreasedactuarialvalueofcoverage,weaccountfortheinduceddemandassociatedwithcost-sharingplanvariationsaspartoftheriskadjustmentmodelandnotaspartoftherisktransferformula.
Exhibit10providesillustrativeexamplesofthePLRScalculation,assumingasilvermetallevelplan.
Enrollee1ismaleandaged56,withtwochronicconditions,diabeteswithcomplicationsandcongestiveheartfailure.
Predictedrelativeincrementalplanliabilityexpendituresfortheseriskfactorsintheadultsilvermodelare0.
580,1.
120,and26Forsilverplanvariantrecipientswiththe94percentand87percentplanvariations,theinducedutilizationfactorin2014is1.
12;forzerocostsharingrecipientsingold,silver,andbronzeplans,theinducedutilizationfactorin2014is1.
07,1.
12,and1.
15,respectively;otherwise,theinducedutilizationfactorin2014is1.
00(PatientProtectionandAffordableCareAct,2013).
3.
587,respectively.
Therefore,hispredictedrelativeplanliabilityexpenditureis5.
287,andsincehedoesnothavecostsharingreductions(inducedutilizationfactoris1.
00),hisPLRSis5.
287.
Enrollee2isfemaleandaged11withasthma.
Herpredictedrelativeplanliabilityexpendituresfromthechildsilvermodelis0.
449(0.
095+0.
354).
However,sheisalsoazerocostsharingrecipient,sohertotalpredictedexpendituresismultipliedbyherinducedutilizationfactor1.
12,resultinginaPLRSof0.
503.
Enrollee3ismaleandaged0,withatermbirthandseveritylevel1.
Hispredictedplanliabilityexpenditurefromtheinfantsilvermodelis1.
572(0.
574+0.
998),andsincehedoesn'thavecostsharingreductions,itishisPLRSaswell.
Finally,theplanaveragePLRS,whichisusedinthecalculationoftransferpaymentsandcharges,isdefinedastheplan'sweightedaverageofindividualPLRSs,wheretheweightsareenrollmentmonths.
WhentheplanaveragePLRSiscalculated,allExhibit10.
PlanLiabilityRiskScoresforSilverMetalLevelPlan—IllustrativeExamplesPredictedRelativeInducedPlanLiabilityPlanLiabilityDemandRiskScoreExpendituresFactorEnrollee1Age56andMale0.
580DiabeteswithComplications1.
120CongestiveHeartFailure3.
587Total5.
2871.
005.
287Enrollee2Age11andFemale0.
095Asthma0.
354Total0.
4491.
120.
503Enrollee3Age0andMale0.
574TermandSeverityLevel10.
998Total1.
5721.
001.
572NOTE:PlanliabilityriskscoreequalspredictedrelativeplanliabilityexpendituresbasedontheHHS-HCCriskadjustmentmodelfortheenrollee'splanmetallevel,multipliedbytheinduceddemandfactorduetocostsharingreductions.
SOURCE:Author'sanalysisofAppendixExhibitsA1–A2,Exhibit6,andthePatientProtectionandAffordableCareAct(2013).
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E26MMRR2014:Volume4(3)planenrolleesarecountedinthenumerator,butonlybillableplanenrollees(parentsandthethreeoldestchildren)arecountedinthedenominator(fordetails,seeourcompanionarticleontherisktransferformula).
ConclusionAsdiscussedinourcompanionoverviewarticle,thekeyprogramgoaloftheACAriskadjustmentmethodologydevelopedbyHHSistocompensatehealthinsuranceplansfordifferencesinenrolleehealthmixsothatplanpremiumsreflectdifferencesinscopeofcoverageandotherplanfactors,butnotdifferencesinhealthstatus.
Thisarticlediscusseshowwedevelopedanempiricalriskadjustmentmodelusingdemographicanddiagnosticinformationfromplanenrolleesandplanactuarialvalue(metaltier)todetermineariskscorethatreflectsexpectedplanliabilityforenrolleemedicalexpenditures.
ThisarticleshowsthattheHHSriskadjustmentmodeltakesintoaccountthenewpopulationandgenerosityofcoverage(actuarialvaluelevel)inanumberofways.
WeusedprivateclaimsdatatodeveloptheHHS-HCCdiagnosticclassification,whichisthekeycomponentoftheriskadjustmentmodel.
Wedevelopedfifteenseparateconcurrentplanliabilityriskadjustmentmodelsreflectingthreeagegroups(adult,child,andinfant),andfiveactuarialvaluetiers(platinum,gold,silver,bronze,andcatastrophic).
Evaluationofthemodelsshowedgoodpredictiveaccuracy,bothforindividualsandforgroups.
Thisarticlealsoprovidesseveralexamplesofhowtocalculateriskscores.
Anenrollee's"planliabilityriskscore"isarelativemeasureoftheactuarialrisktotheplanfortheenrollee.
Itreflectsthehealthstatusrisktotheplanoftheenrollee,theactuarialvalueoftheplan,andtheinduceddemandoftheenrolleeduetoplanvariationcostsharingreductions.
Planaverageriskscoresarethencalculatedfromtheenrolleeriskscoresandusedasaninputintherisktransferformula.
InacompanionarticleinthisissueoftheMedicare&MedicaidResearchReview,wediscusstherisktransferformula.
Wedescribehowtheriskscoreattheplanleveliscombinedwithfactorsforaplan'sallowablepremiumrating,actuarialvalue,induceddemand,geographiccost,marketshare,andthestatewideaveragepremiuminaformulathatcalculatesbalancedtransfersamongplans.
Thenwediscusshoweachplanfactorisdetermined,aswellashowthefactorsrelatetoeachotherinthetransferformula.
DisclaimerTheauthorshavebeenrequestedtoreportanyfundingsourcesandotheraffiliationsthatmayrepresentaconflictofinterest.
Theauthorsreportedthattherearenoconflictofinterestsources.
ThisstudywasfundedbytheCentersforMedicare&MedicaidServices.
TheviewsexpressedarethoseoftheauthorsandarenotnecessarilythoseoftheCentersforMedicare&MedicaidServices.
CorrespondenceJohnKautter,Ph.
D.
,RTIInternational,1440MainStreet,Suite310,WalthamMassachusetts02451,jkautter@rti.
org,Tel.
(781)434-1723Fax.
(781)434-1701.
AcknowledgmentWewouldliketothankseveralpeoplefortheircontributionstothisarticle.
TheseincludeJohnBertko,RichardKronick,andothersfromtheHHS"3Rs"advisorygroup;RTI'sclinicianpanel,whichincludedJohnAyanian,BruceLandon,MarkSchuster,ThomasStorch,andotherclinicians;andRTIcomputerprogrammersArnoldBragg,HelenMargulis,andAleksandraPetrovic.
ReferencesBreyer,F.
,Bundorf,K.
,&Pauly,M.
V.
(2012).
HealthCareSpendingRisk,HealthInsurance,andPaymenttoHealthPlansinM.
V.
Pauly,T.
G.
McGuire,andP.
P.
Barros,eds.
HandbookKautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E27MMRR2014:Volume4(3)ofHealthEconomics,Volume2.
Amsterdam:NorthHolland,Elsevier.
Buettgens,M.
,Garrett,B.
,&Holahan,J.
(2010,December).
AmericaUndertheAffordableCareAct.
UrbanInstituteandRobertWoodJohnsonFoundation,1–16.
Kautter,J.
,Ingber,M.
,Pope,G.
C.
,&Freeman,S.
(2012).
ImprovementsinMedicarePartDRiskAdjustment:BeneficiaryAccessandPaymentAccuracy.
MedicalCare,50,1102–1108.
PubMedhttp://dx.
doi.
org/10.
1097/MLR.
0b013e318269eb20Kautter,J.
,Pope,G.
,&Keenan,P.
(2014).
AffordableCareActRiskAdjustment:Overview,Context,andChallenges.
Medicare&MedicaidResearchReview4(3).
PatientProtectionandAffordableCareAct;HHSNoticeofBenefitandPaymentParametersfor2014,78Fed.
Reg.
47(March11,2013).
Pope,G.
C,Bachofer,H.
,Pearlman,A.
,Kautter,J.
,Hunter,E.
,Miller,D.
,&Keenan,P.
(2014).
RiskTransferFormulaforIndividualandSmallGroupMarketsUndertheAffordableCareAct.
Medicare&MedicaidResearchReview4(3).
Pope,G.
C.
,Kautter,J.
,Ellis,R.
P.
,Ash,A.
S.
,Ayanian,J.
Z.
,Iezzoni,L.
I.
Robst,J.
(2004).
RiskAdjustmentforMedicareCapitationPaymentsUsingtheCMS-HCCModel.
HealthCareFinancingReview,25(4),119–141.
PubMedTrish,E.
,Damico,A.
,Claxton,G.
,Levitt,L.
,&Garfield,R.
(2011,March).
AProfileofHealthInsuranceExchangeEnrollees(PublicationNo.
8147).
TheKaiserFamilyFoundation.
VandeVen,W.
P.
M.
M.
(2011).
RiskAdjustmentandRiskEqualization:WhatNeedstoBeDoneHealthEconomics,Policy,andLaw,6,147–156.
PubMedhttp://dx.
doi.
org/10.
1017/S1744133110000319vandeVen,W.
P.
M.
M.
,&Schut,F.
(2011).
GuaranteedAccesstoAffordableCoverageinIndividualHealthInsuranceMarkets.
inS.
GliedandP.
C.
Smitheds.
TheOxfordHandbookofHealthEconomics.
Oxford,UnitedKingdom:OxfordUniversityPress.
vandeVen,W.
P.
M.
M.
,&Ellis,R.
P.
(2000).
RiskAdjustmentinCompetitiveHealthPlanMarketsinAnthonyJ.
CulyerandJ.
P.
Newhouseeds.
HandbookofHealthEconomics,Volume1A.
Amsterdam:North-Holland,Elsevier.
Winkelman,R.
,&Mehmud,S.
(2007).
AComparativeAnalysisofClaims-BasedToolsforHealthRiskAssessment.
Schaumberg,IL:SocietyofActuaries.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E28MMRR2014:Volume4(3)AppendixExhibitA1.
HHS-HCCRiskAdjustmentModels—Adult(age21+)R-squared=0.
36020.
35530.
35240.
35050.
3496PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCatastrophicPlanLiabilityHCCNumberVariableLabelCountEstimateEstimateEstimateEstimateEstimateAgerange21–24Male538,6480.
2580.
2080.
1410.
0780.
062Agerange25–29Male606,6080.
2780.
2230.
1500.
0810.
064Agerange30–34Male687,8320.
3380.
2740.
1870.
1010.
079Agerange35–39Male745,6990.
4130.
3390.
2400.
1400.
113Agerange40–44Male796,8280.
4870.
4040.
2930.
1760.
145Agerange45–49Male858,8620.
5810.
4870.
3650.
2310.
195Agerange50–54Male884,0860.
7370.
6260.
4840.
3160.
269Agerange55–59Male821,6120.
8630.
7360.
5800.
3930.
339Agerange60+Male830,1191.
0280.
8800.
7040.
4870.
424Agerange21–24Female569,0870.
4330.
3500.
2210.
1010.
072Agerange25–29Female674,0340.
5480.
4480.
3010.
1560.
120Agerange30–34Female749,9380.
6560.
5460.
3960.
2430.
203Agerange35–39Female798,4750.
7600.
6410.
4900.
3340.
293Agerange40–44Female863,2560.
8390.
7130.
5540.
3840.
338Agerange45–49Female954,6590.
8780.
7470.
5830.
4020.
352Agerange50–54Female991,7821.
0130.
8690.
6950.
4860.
427Agerange55–59Female931,2701.
0540.
9050.
7260.
5070.
443Agerange60+Female917,7081.
1560.
9900.
7980.
5590.
489HCC001HIV/AIDS20,9365.
4854.
9724.
7404.
7404.
749HCC002Septicemia,Sepsis,SystemicInflammatoryResponseSyndrome/Shock24,60513.
69613.
50613.
42913.
50313.
529(Continued)Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E29MMRR2014:Volume4(3)ExhibitA1Continued.
HHS-HCCRiskAdjustmentModels—Adult(age21+)PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCatastrophicPlanLiabilityHCCNumberVariableLabelCountEstimateEstimateEstimateEstimateEstimateHCC003CentralNervousSystemInfections,ExceptViralMeningitis4,9887.
2777.
1407.
0837.
1177.
129HCC004ViralorUnspecifiedMeningitis3,0294.
9964.
7304.
6214.
5624.
550HCC006OpportunisticInfections5,3679.
6729.
5499.
5019.
5089.
511HCC008MetastaticCancer32,33625.
17524.
62724.
37624.
49124.
526HCC009Lung,Brain,andOtherSevereCancers,IncludingPediatricAcuteLymphoidLeukemia25,03411.
79111.
37711.
19111.
22411.
235HCC010Non-Hodgkin`sLymphomasandOtherCancersandTumors25,8766.
4326.
1506.
0185.
9835.
970HCC011Colorectal,Breast(AgeCC163AspirationandSpecifiedBacterialPneumoniasandOtherSevereLungInfections11,5849.
0528.
9348.
8838.
9138.
924HCC183KidneyTransplantStatus8,40510.
94410.
57610.
43210.
46310.
482(Continued)Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E35MMRR2014:Volume4(3)ExhibitA1Continued.
HHS-HCCRiskAdjustmentModels—Adult(age21+)PlatinumPlanLiabilityGoldPlanLiabilitySilverPlanLiabilityBronzePlanLiabilityCatastrophicPlanLiabilityHCCNumberVariableLabelCountEstimateEstimateEstimateEstimateEstimateHCC184EndStageRenalDisease10,82437.
71437.
35637.
19337.
35237.
403HCC187ChronicKidneyDisease,Stage53,7562.
1892.
0481.
9951.
9901.
992G16HCC188ChronicKidneyDisease,Severe(Stage4)6,1112.
1892.
0481.
9951.
9901.
992G16HCC203EctopicandMolarPregnancy,ExceptwithRenalFailure,Shock,orEmbolism5,0501.
3771.
2191.
1200.
9120.
828G17HCC204MiscarriagewithComplications1,7961.
3771.
2191.
1200.
9120.
828G17HCC205MiscarriagewithNoorMinorComplications30,1961.
3771.
2191.
1200.
9120.
828G17HCC207CompletedPregnancyWithMajorComplications6,0233.
7783.
2853.
1342.
9312.
906G18HCC208CompletedPregnancyWithComplications74,5763.
7783.
2853.
1342.
9312.
906G18HCC209CompletedPregnancywithNoorMinorComplications82,0773.
7783.
2853.
1342.
9312.
906G18HCC217ChronicUlcerofSkin,ExceptPressure36,1612.
5152.
3712.
3132.
3042.
304HCC226HipFracturesandPathologicalVertebralorHumerusFractures3,8809.
7889.
5709.
4809.
5219.
536HCC227PathologicalFractures,ExceptofVertebrae,Hip,orHumerus3,3291.
9271.
8051.
7351.
6481.
620HCC242ExtremelyImmatureNewborns,BirthweightCC1633,6842.
4982.
6482.
7142.
8132.
841INT2SevereillnessindicatorxHCC2532,3502.
4982.
6482.
7142.
8132.
841INT2SevereillnessindicatorxaggregateHCCgroupingG32,6022.
4982.
6482.
7142.
8132.
841INT2NOTES:1.
N=14,220,5032.
Meanplanliabilityexpendituresforplatinum,gold,silver,bronze,andcatastrophic,respectively:1.
6531.
489,1.
321,1.
142,and1.
091.
3.
Expendituresare2010expenditurestrendedto2014.
Expendituresincludeinpatient,outpatient,andprescriptiondrugexpenditures.
Totalexpendituresincludealloftheseexpenditures.
Simulatedplanliabilityexpendituresreflectstandardizedbenefitdesignsbymetallevel.
Planliabilityexpendituresareconvertedtorelativeplanliabilityexpenditures.
Arelativeplanliabilityexpenditureof1.
0representsthemeanfortheoverallcalibrationsample(adult+child+infant).
Expendituresareannualizedbydividingbytheeligibilityfraction,andregressionmodelsareweightedbythissameeligibilityfraction.
4.
HHS-HCCsisDepartmentofHealthandHumanServices(HHS)-HierarchicalConditionCategories(HCCs).
5.
Regressionmodelcoefficientconstraintswereappliedasfollows:1.
ConstraintsG1–G18indicateaggregateHHS-HCCgroupingconstraints.
2.
ConstraintsH1–H2indicateHHS-HCChierarchyviolationconstraints.
3.
ConstraintsINT1–INT2indicateaggregatehierarchicalseverityindicatorinteractionsgroupedbycost.
6.
AnaggregateHCCgroupingisasetofHCCsthatareeffectivelytreatedasanindividualHCC.
AnenrolleecanhaveatmostoneHCCcoefficient/incrementalexpenditurewithinanaggregateHCCgrouping.
7.
SevereIllnessxHCCinteractiongroupsINT1andINT2arehierarchicalandeachareeffectivelytreatedasindividualinteractions.
Anenrolleecanhaveatmostonediseaseinteractioncoefficient/incrementalexpenditure.
8.
Allcoefficientestimatesarestatisticalysignificantatthe5%levelorlower.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E38MMRR2014:Volume4(3)ExhibitA2.
HHS-HCCRiskAdjustmentModels—Child(age2–20)R-squared=0.
30670.
30240.
29930.
29620.
2950PlatinumPlanLiabilityGoldPlanLiabilitySliverPlanLiabilityBronzePlanLiabilityCatastrophicPlanLiabilityVariableLabelCountEstimateEstimateEstimateEstimateEstimateAgerange2–4Male380,8410.
2830.
2090.
1060.
0190.
000A1,A3Agerange5–9Male688,4990.
1960.
1400.
0640.
0050.
000A2,A4Agerange10–14Male749,9820.
2460.
1890.
1100.
0470.
033Agerange15–20Male955,9720.
3360.
2730.
1910.
1140.
095Agerange2–4Female362,7770.
2330.
1650.
0710.
0190.
000A1,A5Agerange5–9Female660,7170.
1650.
1130.
0480.
0050.
000A2,A6Agerange10–14Female719,6210.
2230.
1680.
0950.
0420.
031Agerange15–20Female921,2360.
3790.
3040.
1980.
1010.
077HCC001HIV/AIDS2562.
9562.
6132.
4212.
2282.
166HCC002Septicemia,Sepsis,SystemicInflammatoryResponseSyndrome/Shock1,94317.
30917.
14217.
06117.
08117.
088HCC003CentralNervousSystemInfections,ExceptViralMeningitis91112.
63612.
40912.
29612.
31312.
319HCC004ViralorUnspecifiedMeningitis9093.
2023.
0042.
8962.
7502.
702HCC006OpportunisticInfections47520.
35820.
26220.
22220.
20120.
189HCC008MetastaticCancer88834.
79134.
47734.
30734.
30634.
300HCC009Lung,Brain,andOtherSevereCancers,IncludingPediatricAcuteLymphoidLeukemia3,15611.
93911.
61811.
43611.
35811.
334HCC010Non-Hodgkin`sLymphomasandOtherCancersandTumors1,3189.
3549.
0718.
9088.
8068.
774HCC011Colorectal,Breast(AgeCC163AspirationandSpecifiedBacterialPneumoniasandOtherSevereLungInfections2,05710.
73010.
61510.
54910.
56610.
571HCC183KidneyTransplantStatus35318.
93318.
47618.
26418.
27918.
289S1HCC184EndStageRenalDisease14243.
15842.
81642.
65942.
77542.
808HCC187ChronicKidneyDisease,Stage55911.
75411.
58111.
47211.
37411.
340G16HCC188ChronicKidneyDisease,Severe(Stage4)9011.
75411.
58111.
47211.
37411.
340G16(Continued)Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E44MMRR2014:Volume4(3)ExhibitA2Continued.
HHS-HCCRiskAdjustmentModels—Child(age2–20)PlatinumPlanLiabilityGoldPlanLiabilitySliverPlanLiabilityBronzePlanLiabilityCatastrophicPlanLiabilityVariableLabelCountEstimateEstimateEstimateEstimateEstimateHCC203EctopicandMolarPregnancy,ExceptwithRenalFailure,Shock,orEmbolism3081.
1911.
0420.
9170.
6740.
590G17HCC204MiscarriagewithComplications1101.
1911.
0420.
9170.
6740.
590G17HCC205MiscarriagewithNoorMinorComplications1,4771.
1911.
0420.
9170.
6740.
590G17HCC207CompletedPregnancyWithMajorComplications3083.
4192.
9562.
7782.
4982.
437G18HCC208CompletedPregnancyWithComplications2,8543.
4192.
9562.
7782.
4982.
437G18HCC209CompletedPregnancywithNoorMinorComplications5,1743.
4192.
9562.
7782.
4982.
437G18HCC217ChronicUlcerofSkin,ExceptPressure1,5541.
5701.
4791.
3941.
3141.
289HCC226HipFracturesandPathologicalVertebralorHumerusFractures2977.
3897.
1747.
0226.
8826.
842HCC227PathologicalFractures,ExceptofVertebrae,Hip,orHumerus6742.
3532.
2442.
1281.
9651.
912HCC242ExtremelyImmatureNewborns,Birthweight<500GramsN/AN/AN/AN/AN/AN/AHCC243ExtremelyImmatureNewborns,IncludingBirthweight500–749GramsN/AN/AN/AN/AN/AN/AHCC244ExtremelyImmatureNewborns,IncludingBirthweight750–999GramsN/AN/AN/AN/AN/AN/AHCC245PrematureNewborns,IncludingBirthweight1000–1499GramsN/AN/AN/AN/AN/AN/AHCC246PrematureNewborns,IncludingBirthweight1500–1999GramsN/AN/AN/AN/AN/AN/AHCC247PrematureNewborns,IncludingBirthweight2000–2499GramsN/AN/AN/AN/AN/AN/A(Continued)Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E45MMRR2014:Volume4(3)ExhibitA2Continued.
HHS-HCCRiskAdjustmentModels—Child(age2–20)PlatinumPlanLiabilityGoldPlanLiabilitySliverPlanLiabilityBronzePlanLiabilityCatastrophicPlanLiabilityVariableLabelCountEstimateEstimateEstimateEstimateEstimateHCC248OtherPremature,LowBirthweight,Malnourished,orMultipleBirthNewbornsN/AN/AN/AN/AN/AN/AHCC249TermorPost-TermSingletonNewborn,NormalorHighBirthweightN/AN/AN/AN/AN/AN/AHCC251StemCell,IncludingBoneMarrow,TransplantStatus/Complications32430.
55830.
48530.
46630.
52230.
538HCC253ArtificialOpeningsforFeedingorElimination2,00614.
41014.
24714.
19714.
34014.
383HCC254AmputationStatus,LowerLimb/AmputationComplications9510.
1749.
9379.
7999.
6889.
641H6NOTES:1.
N=5,439,6452.
Meanplanliabilityexpendituresforplatinum,gold,silver,bronze,andcatastrophic,respectively:0.
532,0.
454,0.
357,0.
273,and0.
252.
3.
Expendituresare2010expenditurestrendedto2014.
Expendituresincludeinpatient,outpatient,andprescriptiondrugexpenditures.
Totalexpendituresincludealloftheseexpenditures.
Simulatedplanliabilityexpendituresreflectstandardizedbenefitdesignsbymetallevel.
Planliabilityexpendituresareconvertedtorelativeplanliabilityexpenditures.
Arelativeplanliabilityexpenditureof1.
0representsthemeanfortheoverallcalibrationsample(adult+child+infant).
Expendituresareannualizedbydividingbytheeligibilityfraction,andregressionmodelsareweightedbythissameeligibilityfraction.
4.
HHS-HCCsisDepartmentofHealthandHumanServices(HHS)-HierarchicalConditionCategories(HCCs).
5.
Regressionmodelcoefficientconstraintswereappliedasfollows:1.
ConstraintsG1–G18indicateHHS-HCCgroupconstraints.
2.
ConstraintsH1–H7indicateHHS-HCChierarchyviolationconstraints.
3.
ConstraintS1indicatesHHS-HCCspecifiedconstraint.
4.
ConstraintsA1–A6indicateage/sexcellconstraints.
A1andA2areforbronzeplansonlyandA3–A6areforcatastrophicplanonly.
6.
AnaggregateHCCgroupingisasetofHCCsthatareeffectivelytreatedasanindividualHCC.
AnenrolleecanhaveatmostoneHCCcoefficient/incrementalexpenditurewithinanaggregateHCCgrouping.
7.
Allnon-zerocoefficientestimatesarestatisticallysignificantatthe5%levelorlower.
SOURCE:Authors'analysisof2010MarketScanCommercialClaimsandEncountersDatabase.
Kautter,J.
,Pope,G.
C.
,Ingber,M.
,etal.
E46