Should patients with acute coronary disease be ... - BMJ Open

The updated GRACE risk score has better model discrimination and is easier to use than previous scores based on categorical variables. It is accurate in the ... Skiptomaincontent YouarehereHome Archive Volume4, Issue2 Shouldpatientswithacutecoronarydiseasebestratifiedformanagementaccordingtotheirrisk?Derivation,externalvalidationandoutcomesusingtheupdatedGRACEriskscore Emailalerts ArticleText Articlemenu ArticleText Articleinfo CitationTools Share RapidResponses Articlemetrics Alerts PDF XML CardiovascularmedicineResearch Shouldpatientswithacutecoronarydiseasebestratifiedformanagementaccordingtotheirrisk?Derivation,externalvalidationandoutcomesusingtheupdatedGRACEriskscore KeithAAFox1,GordonFitzGerald2,EtiennePuymirat3,4,5,6,WeiHuang2,KathrynCarruthers1,TabassomeSimon7,8,9,10,11,PierreCoste,JacquesMonsegu12,PhilippeGabrielSteg13,14,15,NicolasDanchin3,4,5,6,FredAnderson21CentreforCardiovascularScience,UniversityofEdinburgh,Edinburgh,,UK2UniversityofMassachusettsMedicalSchool,Worcester,Massachusetts,USA3DepartmentofCardiology,EuropeanHospitalofGeorgesPompidou,Paris,France4AssistancePubliquedesHôpitauxdeParis(AP-HP),Paris,France5UniversityParisDescartes,Paris,France,6INSERMU-970,Paris,France7DepartmentofPharmacology,AssistancePublique-HôpitauxdeParis,HôpitalStAntoine,UnitédeRechercheClinique(URCEST),Paris,France8INSERMU698,Paris,France9UniversitéPierreetMarieCurie-Paris06,Paris,France10HôpitalduHautLevêque,Pessac,Paris,France11UniversitéBordeauxSegalen,Bordeaux,France12DepartmentofCardiology,HôpitalduValdeGrâce,Paris,France13AssistancePublique-HôpitauxdeParis,HôpitalBichat,Paris,France14INSERMU698,Paris,France15UniversitéParisDiderot,Paris,FranceCorrespondencetoProfessorKeithAAFox;k.a.a.fox{at}ed.ac.uk Abstract ObjectivesRiskscoresarerecommendedinguidelinestofacilitatethemanagementofpatientswhopresentwithacutecoronarysyndromes(ACS).Internationally,suchscoresarenotsystematicallyusedbecausetheyarenoteasytoapplyandsomeriskindicatorsarenotavailableatfirstpresentation.WeaimedtoderiveandexternallyvalidateamoreaccurateversionoftheGlobalRegistryofAcuteCoronaryEvents(GRACE)riskscoreforpredictingtheriskofdeathordeath/myocardialinfarction(MI)bothacutelyandoverthelongerterm.Theriskscorewasdesignedtobesuitableforacuteandemergencyclinicalsettingsandusableinelectronicdevices. DesignandsettingTheGRACEriskscore(2.0)wasderivedin32 037patientsfromtheGRACEregistry(14countries,94hospitals)andvalidatedexternallyintheFrenchregistryofAcuteST-elevationandnon-ST-elevationMI(FAST-MI)2005. ParticipantsPatientspresentingwithST-elevationandnon-STelevationACSandwithlong-termoutcomes. OutcomemeasuresTheGRACEScore(2.0)predictstheriskofshort-termandlong-termmortality,anddeath/MI,overallandinhospitalsurvivors. ResultsForkeyindependentriskpredictorsofdeath(1 year),non-linearassociations(vslinear)werefoundforage(p<0.0005),systolicbloodpressure(p<0.0001),pulse(p<0.0001)andcreatinine(p<0.0001).Byemployingnon-linearalgorithms,therewasimprovedmodeldiscrimination,validatedexternally.UsingtheFAST-MI2005cohort,thecindicesfordeathexceeded0.82fortheoverallpopulationat1 yearandalsoat3 years.DiscriminationfordeathorMIwasslightlylowerthanfordeathalone(c=0.78).Similarresultswereobtainedforhospitalsurvivors,andwithsubstitutionsforcreatinineandKillipclass,themodelperformednearlyaswell. ConclusionsTheupdatedGRACEriskscorehasbetterdiscriminationandiseasiertousethanthepreviousscorebasedonlinearassociations.GRACERisk(2.0)performedequallywellacutelyandoverthelongertermandcanbeusedinavarietyofclinicalsettingstoaidmanagementdecisions. ACCIDENT&EMERGENCYMEDICINEThisisanOpenAccessarticledistributedinaccordancewiththeCreativeCommonsAttributionNonCommercial(CCBY-NC3.0)license,whichpermitsotherstodistribute,remix,adapt,builduponthisworknon-commercially,andlicensetheirderivativeworksondifferentterms,providedtheoriginalworkisproperlycitedandtheuseisnon-commercial.See:http://creativecommons.org/licenses/by-nc/3.0/ http://dx.doi.org/10.1136/bmjopen-2013-004425 StatisticsfromAltmetric.com ACCIDENT&EMERGENCYMEDICINEStrengthsandlimitationsofthisstudy TheGlobalRegistryofAcuteCoronaryEvents(GRACE)2.0riskscoreisderivedfromthelargestmultinationalregistryinacutecoronarysyndromes(ACS)andvalidatedinanentirelyindependentdatasetwithcomprehensivelong-termoutcomedata. Thisriskscoreemploysnon-linearfunctionsandismoreaccuratethantheoriginalversion.Itisnowvalidatedoverthelongerterm(to1and3 years)andwithsubstitutionspossibleforcreatininevaluesandKillipclass(performingalmostaswell). Thiselectronicriskscoreisdesignedtobeusedinmobileelectronicdevices(approximately30 stoenterdata)andpresentstheriskofdeath(ordeath/myocardialinfarction)andrelativetotheentireACSpopulation. Thescoreisdesignedtoassistclinicalmanagementdecisionsandisnotasubstituteforindividualpatientclinicalassessment.However,itmayhelptoaddressthecurrent‘treatment-riskparadox’wherebylow-riskratherthanhigh-riskpatientsaremorelikelytoreceiveinterventionaltherapies. Additionalfactorsmayinfluenceoutcome,especiallyingeographicalpopulationsandhealthcaresystemsnotevaluatedinthemultinationalGRACEprogramme. Introduction Acutecoronarysyndromes(ACSs)compriseaheterogeneousspectrumofpatientswhoarecurrentlystratifiedformanagementmainlyonthebasisofECGcharacteristicsandbiomarkerresults.NationalInstituteforHealthandCareExcellence(NICE),ScottishIntercollegiateGuidelinesNetwork(SIGN),EuropeanSocietyofCardiology(ESC)andNorthAmericanguidelinesseparatepatientsintoST-elevationMIornon-ST-elevationACS,andtheyalsorecommenduseofariskscoresuchastheGlobalRegistryofAcuteCoronaryEvents(GRACE)score.1–4However,systematicriskstratificationisnotwidelyperformed,despitetheevidenceandtheguidelines. Whyshouldriskassessmentbeimportantforthetriageandmanagementofpatientswithacutecoronarydisease? Whetherapatientproceedstoanimmediate,urgentordelayedcoronaryangiographyandrevascularisationandwhichoftheacuteantithromboticregimensischosendependsonpatientriskcharacteristics.Evidencefromrandomisedtrialsandguidelinerecommendationsallsupporttheuseofdifferentstrategiesaccordingtoriskstatus.1–4 InthedevelopmentofNICEguideline94(http://www.nice.org/cg94),theguidelinestatesthatsinglevariables(eg,troponin)werenotasgoodasmultiplevariablesinpredictingoutcome.1NICEindependentlytestedallofthepublishedriskscores(GRACE,5,6TIMI,7PURSUIT,8PREDICT,9EMMACE,10SRI,11AMIS,12UA13riskscore)in64 312patientsfromtheMINAPdataset.Theyemployeda‘mini-GRACEscore’asmanyoftheMINAPpatientslackedcreatininevaluesandKillipclassification(substitutingahistoryofrenaldysfunctionandtheuseofdiuretics)andthisapproachalsodemonstratedgoodperformanceinanindependentassessment.14Thecstatisticswas0.825with95%confidencebounds0.82to0.83andthiswassuperiortotheperformanceoftheotherriskscores,andhencetherecommendationfromNICEtoemploytheGRACEriskscore.1However,theuseofsubstitutionsforcreatinineandforKillipclasshasnotbeenvalidatedinanindependentdatasetandthepredictionoflong-termoutcomehadnotbeentested.Inaddition,non-linearfunctionsforcontinuousvariablesandforKillipclassmayimprovemodeldiscriminationandcouldbeimplementedinhand-heldelectronicdevices. Resolvingthe‘treatment-riskparadox’ Weandothershaverevealedatreatment-riskparadoxinthemanagementofacutecoronarydisease.15,16Incontrasttotheevidenceandtheguidelinerecommendations,lowerriskratherthanhigherriskpatientsaremorelikelytoundergointerventionalproceduresandreceivemoreaggressiveantithromboticandothertherapies.15,16Thisphenomenonhasnowbeenreportedacrosswidelydifferenthealthcaresystemsanddifferentgeographicalsettings.Whyisthis?First,currenttreatmentdecisionsrelyonclinicalassessmentanditisdifficultforthecliniciantoweighuppotentialbenefitsagainstpotentialhazards,andhencelowerriskpatientsarecommonlyselectedformoreaggressivetreatment(anunintendedrisk-averseapproach).However,evidencedemonstratesthatevenexcludingthosewithcontraindications,higherriskcohortspotentiallyhavemoretogain.15 Whyareriskscoresnotmorewidelyused? Internationally,riskscoresarenotsystematicallyappliedforthemanagementofACSdespitetheevidenceandguidelinerecommendations.Severalfactorscontributetothis,includingthemisperceptionthatclinicianassessmentortheuseofindividualriskindicatorsissufficient.1,2,17Inaddition,themostaccurateriskscoreshavebeencumbersometocompute(eg,requiringlook-uptablesandmanyusearbitraryscoreresults).Finally,theparametersnecessaryfortheirimplementationmaynotbeavailableatthetimeofthepatient'sinitialpresentation. Whatthisstudyadds? WeaimedtodevelopandvalidatearevisedandmoreaccurateversionoftheGRACEriskscoresuitablefortheacuteandlong-termpredictionofrisk.InsteadofassumingthatcontinuousvariablessuchasageandthecategoricalvariableKillipclasswerelinearlyassociatedwithrisk,wetestedfornon-linearassociationsandincludedthemintherevisedpredictiontoolwhereappropriate.IncontrasttotheearlierversionoftheGRACEscore,whichrequiredthecomputationofanumericalscore(withoutabsoluterisks),wederivedandexternallyvalidatedanelectronicversionwithabsolutepercentagerisks.Thisissuitableforuseinhand-heldelectronicdevicesandsmartphones,andtheclinicalapplicabilityisbroadenedbyusingsubstitutionsforcreatinineandKillipclass.CreatininevaluesmayonlybeavailableafterhospitaladmissionandmanysettingsdonotroutinelyuseKillipclassforevaluatingheartfailuresymptoms.Thus,theaimofthisstudywastodevelopasimplifiedriskscoresuitableforapplicationsinavarietyofsettingsandtotesttheaccuracyoftherevisedGRACEriskpredictor(GRACEscore2.0)topredictearlyandlong-termrisk,asanaidtoclinicalmanagement. Methods GRACEriskscore TheGRACEregistrywasdesignedtoreflectanunbiasedpopulationofpatientswithACSandwasundertakenover10 years,in94hospitalsand14countries.5,6,18–20Thedesignhasbeenreportedpreviously.18,20 In-hospitalandupto6 monthsoutcomesandriskscoreswerederivedbasedonindependentpredictorsofoutcome.Thesehavebeendescribedpreviously(STsegmentdeviation,age,heartrate,systolicbloodpressure,creatinine,Killipclass,cardiacarrestatadmissionandelevatedbiomarkersofnecrosis).5,6TheGRACEriskscorewasderivedfromtheoriginalpopulationof26 267patients(11 389forhospitalscoreforpatientsenrolledthrough31March2001;21 688wereusedtoderivethe6-monthriskscoreforpatientsenrolledthrough30September2002)withsuspectedACS,validatedprospectivelyinafurthersetof22 122patientsandvalidatedexternally.5 Riskcharacteristicsofpopulationsmayevolveovertime(asmanagementchanges)anditisappropriatethattheGRACEscoreshouldbetestedinamorerecentcohortofpatientswithACSandwithextendedfollow-up.21 TheoriginalGRACEscorewasestimatedinhospitalriskofdeathorthecombinationofdeathorMIandthesameoutcomesupto6 monthspostdischarge.ThenewversionoftheGRACEriskscorefor1-yearoutcomeswasderivedinthemorerecentdatasetof32 037patientsfromtheGRACEregistryenrolledbetweenJanuary2002andDecember2007.Forthree3-yearmortality,theUKcohortof1274patientswithlong-termfollow-upwasemployed.Thecharacteristicsofthisstudypopulationhavebeenreportedpreviously.20Thealgorithmemployedthesameindependentpredictorsofoutcomeasoriginallyderivedandreported,butnon-linearassociationswereincorporatedtoimprovemodeldiscrimination.Inaddition,asimplifiedversionoftheriskscorewasdevelopedwithsubstitutionsforcreatinine(historyofrenaldysfunction)andsubstitutionsforKillipclass(diureticusage).Aspreviouslyvalidated,aparsimoniousmodelofonlyeightfactorsconveyedmorethan90%ofthepredictiveaccuracyofthecompletemultivariablemodel.5,6 ConsistencyofestimatesindifferentGRACEriskmodels TheGRACEriskscoreV.2.0containsslightlymorepreciseestimatesofV.1.0hospital6and6-monthdeath5probabilities.Insteadofconvertingmodelestimatestoapointsystem,andusingintervalsforcontinuousvariablessuchasage,asinV.1.0,V.2.0directlyutilisesmodelestimatesthemselvestocomputecumulativerisk(see:http://www.outcomes-umassmed.org/grace/files/GRACE_RiskModel_Coefficients.pdf). BecauseGRACEmodelswerederivedindifferentpatientpopulationsfromdifferentstudyperiods,differencesincumulativerateestimatesforthesameintervalexist.The1-yeardeathmodelcontainsthemostrecentandlargestpatientpopulations.Therefore,6-monthand3-yeardeathmodelswerestandardisedtoconformtoestimatedKaplan-Meiercumulativeratesforthe1-yearmodel.TherevisedV.2.06 monthcumulativeestimatesnowconformtoV.2.01-yearmodelestimatesasof6 months,andthe1-yearestimatesfortheV.2.03-yearmodelasof1 yearalsoconformtoV.2.01-yearestimatesforthe1-yearmodel. Externalvalidation TheupdatedGRACEriskscorewasvalidatedbytestingthealgorithminitsfullversionandsimplifiedversioninanentirelyseparateregistrypopulation,theFrenchregistryofAcuteST-elevationandnon-ST-elevationMyocardialInfarction(FAST-MI).22–24FAST-MI2005isanationwideFrenchregistryconductedovera1-monthperiodattheendof2005anditincluded3059patientswithSTsegmentelevationmyocardialinfarction(STEMI)ornon-STEMIfrom223centres.Follow-upwasconductedbyaresearchteamfromtheSociétéFrançaisedeCardiologieandinvestigators.22,23Sequentially,theyconsulteddeathregistrydata,wrotetofamilydoctorsand/orcardiologistsandwrotetopatients.Inmanyinstances,writtencontactwasfollowedbytelephoneinterviews.22,23AllvariablesrequiredtocalculatethenewGRACEriskscorewereavailablein2959ofthe3059patients(96.7%ofthefullcohort).TheGRACEalgorithmwasappliedtothe2959patientsusinglogisticregressionandthecstatisticscalculatedformortalityat1 year,mortalityat3 yearsandthenforthesubsetsofpatientswithSTEMIandnon-STEMI.Inaddition,cstatisticswerecalculatedfordeathorMI.Thesameanalyseswerethenrepeatedforhospitalsurvivorsonly(n=2806).Inaddition,goodnessoffitwastestedusingtheHosmer-Lemeshowtest.Likewise,thesimplifiedscorewastestedinthe3035patientsinwhomallvariablesneededforitscalculationwereavailable. Statistics TheKaplan-Meiermethodwasusedtoestimate1-yearand3-yearoutcomerates. CoxmultipleregressionmodelswerefittedtooutcomesofdeathanddeathorMIwithin1and3 yearsofhospitaladmission.24ThesameeightfactorsusedintheoriginalGRACEriskscoreswereused.6Themethodofrestrictedcubicsplines25employsasmoothpolynomialfunctionandwasusedtotestforpossiblenon-linearassociationsbetweenoutcomesandage,creatinine,pulseandsystolicbloodpressure.Also,KillipclassusingfourcategorieswascomparedwithlinearKillipclass.Associationsthatimprovedmodellikelihoodattheα=0.05levelwereretainedinfinalmodels.Suchassociationswerealsoplottedandexaminedforclinicalplausibility. ModelperformancewasevaluatedusingtheMay-Hosmergoodnessoffittest,26andHarrell'scindexformodeldiscrimination.27Apredictiontoolbasedonthesemodelsusespointestimatesandbaselinesurvivaltoarriveatpredictedoutcomesforagivenpatient'scovariateexperience.28Plotsofestimatedmodeleventprobabilitiesfornon-linearcovariateswereproducedusingbaselinesurvivalestimatesandriskfactorparameterestimates(ontheloghazardscale),evaluatedatcovariatemeans.Theseplotsdescribetheshapeoftheassociationbetweenthenon-linearfactorsandoutcomes,buttheydonotsubstituteforenteringallofthepatient'sriskfactorinformationintotherisktool. Results Patientcharacteristics Forthe32 037patientsfromtheGRACEregistry(table1),therewere2422deathswithin365 daysofinitialadmission,andcompletecovariatedata.Thedistributionofdeathswasasfollows:1275inhospital(53%),983deathsafterdischargewithin180 daysofadmission(41%),164deathsfrom181–365 daysafteradmission(7%).Theestimated365 daycumulativedeathrateis9.3%usingtheKaplan-Meiermethod. Viewthistable:Viewinline Viewpopup Table 1CharacteristicsonadmissionoftheGRACEpatientswithACSusedin1-yeardeathmodelandtheFAST-MIpatients Forthe3-yearmodelderivedfrom1274patientsfromtheUK,therewere261deaths:59in-hospital(23%),51afterdischargewithin180 daysofadmission(20%)and151intheremaining2.5 yearssinceadmission(58%).Theestimated3-yearcumulativedeathrateis20.5%. Performanceofthemodelusingnon-linearfunctions AnalyseswereundertakenfirstusingcategoricalvariablesandlinearassociationsforcontinuousvariablesandKillipclass(asintheoriginaldescriptionoftheGRACEriskscore),5,6andthenusingnon-linearassociationsforage,heartrate,systolicbloodpressureandcreatinine(figure1A–D).Differenceswereobservedbetweenthenon-linearandthelinearmodelwiththeformermorelikelytoclassifypatientsasathighrisk(datanotshown). Downloadfigure Openinnewtab Downloadpowerpoint Figure 1Non-linearassociationsforthe1 yearmortalitymodelwerefoundforfourcontinuousmeasures:systolicbloodpressure(A),pulse(B),age(C)andcreatinine(figure1D);p<0.001vslinearforeachcomparison. Non-linearassociationsforthe1-yearmortalitymodelwerefoundforallfourcontinuousmeasures:systolicbloodpressure,pulse,ageandcreatinine(p<0.001vslinear).Therestrictedcubicspline(polynomialcurve)functionsforageandsystolicbloodpressurehadthreeknots(‘inflectionpoints’)atthe10th,50thand90thcentilesoftheirdistributions,andfourknotsatthe5th,35th,65thand95thcentilesofpulseandcreatininedistributions.HRestimatesarereportedforselectedintervals,toprovideasenseofhowassociationschangeovercovariateranges(table2).Killipclassismodelledasfourdistinctgroups(p<0.001vslinearclass).The1-yeardeath/MImodelhassimilarnon-linearassociations,whilethe3-yeardeathmodelhasfourknotcubicsplineassociationsforsystolicbloodpressureandpulseandlinearassociationsfortheremainingfactors.Alsoshownareestimatesforthesubstitutefactorsofrenalinsufficiencyanddiuretics,whichcanbeusedtoreplacecreatinineandKillipwhentheyareunavailable.Samplesizesincreasesomewhatformodelsusingthesubstitutefactors,andmodeldiscriminationisonlyslightlydiminished. Viewthistable:Viewinline Viewpopup Table 2SummaryofCoxregressionmodels Thegoodnessoffittestispartlyafunctionofsamplesizewithlargersamplesizesincreasingthechancethatasmalldifferencebetweenobservedandexpectednumbersofdeathwillbedetected.Thiswasobserved,withdifferencesmainlyinthe9thriskdecile,(themodelpredicteda3-yearriskof17%,estimatedobserveddeath19.5%).Thelargestdifferenceintheremainingdecilesis1.2%. Basedontherelativemodelχ2values,ageisthemostimportantfactorinallthreemodels,followedbysystolicbloodpressure,creatinineandKillipclassinthe1-yearmodel(allhavesimilarχ2values),creatinineandKillipclassinthe1-yeardeath/MImodel,andsystolicbloodpressureandpulseinthe3-yeardeathmodel.Allmodelsshowgooddiscrimination(cindices≥0.74),althoughcombiningMIwithdeathinthe1 yearmodelreducesmodeldiscrimination,becausedeathandMIarenotinterchangeablewithrespecttopatientriskprofiles. Externalvalidationofthenon-linearGRACEriskscoreintheFAST-MI2005registry ThecharacteristicsoftheFAST-MI2005registryarereflectiveoftherangeofpatientspresentingwithACS(meanage66.9 years±SD14.4 years,31%women,53%STEMI,47%non-STEMI,historyofcoronaryarterydisease30%,historyofstroke5%,documenteddiabetesmellitus24%,documentedhypertension57%,currentsmoking30%anddocumentedhypercholesterolaemia47.5%).TheFAST-MI2005registryhasexcellentcompletenessoffollow-up(3-yearfollow-up98%complete).Overallsurvivalwas79%andinfarct-freesurvivalwas73%. UsingtheFAST-MI2005cohortof2959patients,c-statisticsfordeathexceeded0.82fortheoverallpopulationat1 yearandalsoat3 years(table3).DiscriminationfordeathinthemodelwashigherintheST-elevationMIpopulation(c=0.84)at1 yearcomparedtothenon-STEMIpopulation(c=0.80).DiscriminationfordeathorMIwasslightlylowerthanfordeathalone(c=0.78)bothat1and3 years.Similarfigureswereobtainedforhospitalsurvivors(seetables3and4). Viewthistable:Viewinline Viewpopup Table 3ThefullGRACEriskscoretestedinFAST-MI2005 Viewthistable:Viewinline Viewpopup Table 4ThesimplifiedGRACEriskscore,withsubstitutionsforKillipandcreatinine(n=3035),testedinFAST-MI2005 Thec-statisticsfor3-yeardeathwerecalculatedusingthesameapproachforthewholeACSpopulationandat3 yearswere0.82fordeathand0.75fordeathorMI. ThecindicesusingthesimplifiedGRACEmodelwithsubstitutionsforKillipclassandserumcreatininewereavailablefor99.2%ofpatients;thesewere0.82forboth1and3-yearmodels). Insummary,useofnon-linearfunctionsforcontinuousvariablesimprovedmodelperformanceovertheoriginalGRACEriskscoreusinglinearfunctions.Theexternalvalidationdemonstratedgoodmodeldiscriminationat1and3 yearsforbothdeathanddeathorMI,andinsubtypesofMI,ST-elevationandnon-ST-elevationMI.Thishasnotbeentestedpreviously.Theriskscoreperformssimilarlywhenconsideringonlythesurvivorsofhospitalisation.Thesimplifiedriskscoreusingahistoryofrenaldysfunctioninplaceofcreatininevalues,andusingofdiureticsinplaceofKillipclass,performedalmostaswellasthefullGRACEscore. Discussion ThisstudyaimedtodevelopanimprovedversionoftheGRACEriskpredictor(GRACEscore2.0)incorporatingnon-linearassociationsbetweencontinuousriskfactorsandoutcomesinaformatsuitableforeaseofuseinhand-heldelectronicdevicesandsmartphones(figure2).Inaddition,therevisedGRACEriskscoreallowsreadilyavailablesubstitutionsformissingvariablesatthetimeoffirstpatientpresentation(creatinine,Killipscore),andthisallowsthehealthcareprofessionaltoriskscoreamorecompleterangeofpatientshospitalisedwithACS.Thescoreisnotdependentonkeyvariables—itallowsflexibilityinlightofdataavailability.Further,theGRACEscorehadnotbeentestedforpredictiveaccuracybeyond6 months,andthesimplifiedversionoftheriskscorewithsubstitutionsforcreatinineandforKillipclasshadnotbeentestedinanindependentpopulation.Akeyfindingisthatmodellikelihoodusingindividualnon-linearfunctionsforheartrate,systolicbloodpressure,ageandcreatininewassignificantlyimprovedoveramodelusinglinearfunctionsforthesefactors.Inbrief,themodelwithnon-linearfunctionsmatchesobserveddatamoreclosely.Further,theupdatedGRACEriskscoredemonstratedsimilarhighmodeldiscriminationat1and3 yearsashadpreviouslybeendemonstratedforin-hospitaloutcomesandoutcomesto6 months.Inaddition,thereducedversionoftheGRACEriskscore,withsubstitutionsforcreatinineandKillipclass(withahistoryofrenaldysfunctionanduseofdiuretics,respectively),performsnearlyaswellasthemodelwithoriginalfactors. Downloadfigure Openinnewtab Downloadpowerpoint Figure 2IllustrationoftheGlobalRegistryofAcuteCoronaryEvents(GRACE)Score2.0onamobiledevice(suitableforuseiniOS,androidorwebversions).Leftpanel:valuesforpercentageriskofdeathordeath/myocardialinfarction(ornumericalGRACEScore).Remainingpanelsshowtheindividualpatientresultsasaverticalcolumnsuperimposedontheentireacutecoronarysyndromedistributioncurve(greencolumn=lowriskillustration,yellowcolumn=mediumriskandredcolumn=highrisk).34Forfurtherinformationseehttp://www.gracescore.co.ukandhttp://www.outcomes.org/grace. Whataretheimplications? Inadiverserangeofhospitalsin14countriesworldwide,withon-siteangiographicfacilities,thefrequencyofcatheterisationsandpercutaneouscoronaryinterventions(PCI)exhibitedaparadoxicalpattern,wherebymostinterventionswereperformedinlow-riskratherthanhigh-riskpatients(the‘treatment-riskparadox’).15,16 Tocounterthecriticismthatnotallhigh-riskpatientswillbesuitableforrevascularisation,weundertookfurtheranalysesinapreviouspublicationaccordingtothefrequencyofangiography(hospitalswithon-siteangiographicfacilitiesweredividedintotertilesaccordingtotherateofcoronaryangiography).15Hospitalswithahighrateofcoronaryangiographyperformedsubstantiallymoreinterventionsinhigherriskpatientsthanthoseperformedinlowratehospitals,despiteasimilarrangeofriskstopatients,demonstratingthatthesepatientswereamenabletotheinterventionprocedures.15 Itispossibletoestimatethe‘deficit’inthefrequencyofrevascularisationbasedontheactualdifferencesbetweenhighrateandlowratehospitalsobservedintheGRACEprogramme.Fromtheoverallpopulation,37.8%ofpatientswereintheGRACEhigh-riskgroup,36.1%intheGRACEmediumriskgroupand26.1%intheGRACElowerriskgroup(categoriesaccordingtotheESCguidelines).3Aspreviouslyreported,15individualsinthehighestthirdoftheGRACEriskscorehadcatheterisationperformedin51%andPCIorcoronaryarterybypassgrafting(CABG)in31.4%ofpatients,whereasthoseinthemediumGRACEriskgrouphadcatheterisationperformedin68%andPCIorCABGin42.9%andthoseinthelowerriskgrouphadcatheterisationin72%andPCIorCABGin47.6%. Takingtheperformanceofhospitalsthatwereinthehighestthirdfortherateofcoronaryangiography(theyperformedPCIandCABGin60.2%ofthepresentingpopulation),itispossibletocalculatethedeficitcomparedwiththehospitalswiththelowestrateofangiographyandrevascularisation.ThecalculationassumesthatthelowperformancehospitalsincreasedtheirrateofPCIandCABGtothesameextentaswasobservedinthehighestthirdofhospitals.Thisprojectionisbasedonobservedperformancedatafortherateofangiography.ThecalculationassumesthatnomorePCIorCABGwasperformedthanwasobservedinthehighratehospitals.Inbrief,700morepatientsper10 000wouldundergorevascularisationifthesamepatientspresentedtohighperformancehospitals. Theimpactofrevascularisationsonoutcomescanbeestimatedfromthepooledanalysisofalltherandomisedtrialswherepatientswererandomisedtoaninterventionalstrategyasaroutine,ortoaselectivestrategybasedonsymptomsandischaemia.29WepreviouslyreportedthiscombinedanalysisbasedonindividualpatientdatafromtheFRISC-2,30RITA-331andICTUS32trials,andtheabsolutereductionincardiovasculardeathsandMIswas11.1per100patientsinthehighestriskgroupand4per100inthemediumriskgroup,over5 years.29–32Thus,basedontheimpactofasystematicinterventionalstrategyintherandomisedtrials,therewouldbebetween30and80fewercardiovasculardeathsorMIsforeach10 000patientswithnon-ST-elevationACS.Theestimateisconservativeasitexcludestheimpactonmediumriskpatientsandthenumberwouldbehigherifthetopquintileofperformancewasusedasthereferencestandardratherthanthetoptertile.Thus,consistentwiththeguidelinerecommendations,asystematicapproachforevaluatingriskhasthepotentialtoincreasetherateofrevascularisationinhigh-riskpatientswithoutcontraindications.Basedonthecombinedanalysisofalltherandomisedtrialswithlong-termoutcomes,thisrisk-relatedstrategyhasthepotentialtoreducethefrequencyofcardiovasculardeathandMIoverthelongerterm.The‘High’,‘Medium’and‘Low’riskcategoriesmayhelpguidepracticedecisionsandtheycorrespondwithcategoriesusedbytheESCguidelines.3However,formorepreciseestimates,theGRACEriskscorealsoprovidesthenumericalriskofdeath(ordeath/MI)atvarioustimepoints. Strengthsandlimitations RecognisingthatpopulationcharacteristicsmaydifferincomparisonwiththatoftheoriginallyderivedGRACEmodel,weemployedthemostrecentGRACEdatasetinthisstudyandexternallyvalidatedtheriskmodelinanentirelyseparatedataset(FAST-MI2005withyearlyfollow-upto2010).WehavepreviouslyreportedthattheGRACEriskpredictionisnotsubjecttosignificantchangeovertime.33Thepurposeofproviding1and3-yearriskestimateswastoaidtheclinicianregardingsecondaryprevention.34Theriskpredictionestimatesat3 yearswereconsistentwiththoseat1 year(althoughthe3-yeardataderivefromasmallerdataset). TheGRACEprogrammeisthelargestmultinationalprogrammeinacutecoronaryarterydiseaseandwasdesignedtoensurethattheincludedpatientswerereflectiveofthebroadspectrumofpatientspresentingwithACS,aswellasoftherangeofhospitalsinclinicalpractice.Thesitesweretrained,auditedandqualitycontrolmeasureswereenactedthroughoutthestudy.UseoftheUKcohortallowedestimationoflong-termoutcomes(aspreviouslyreported)withcompletemortalitydatato5 years.20TheexternalvalidationoftheupdatedriskscorewasperformedintheFAST-MI2005registrywithinclusionofthefullspectrumofhospitalsadmittingpatientswithACSandexcellentcompletenessoffollow-up. AlthoughtheupdatedGRACEriskscoreprovidesareliableestimateforstratifyingpatientsbothacutelyandinthelongterm,additionalfactorscontributetolongertermrisk.Furtherrefinementoftheriskscoreforlong-termoutcomesmayrequiretheinclusionofadditionalriskfactorsandbiomarkerstoincreaseprecision,butthecurrentriskscorediscriminationallowsseparationofpatientsintobroadcategoriesrelevantfordecisionsonclinicalmanagement.Futurestudieswilldeterminetheimpactofriskscoringstrategiesinvariouspopulationsincludingthefrailandelderly. Conclusions TheupdatedGRACEriskscorehasbettermodeldiscriminationandiseasiertousethanpreviousscoresbasedoncategoricalvariables.Itisaccurateintheacutephaseandoverthelongertermandcanbeusedinavarietyofclinicalsettingstoaidmanagementdecisions. References↵UnstableanginaandNSTEMItheearlymanagementofunstableanginaandnonST-segment-elevationmyocardialinfarction.NICEguideline94March2010.http://www.nice.co.uk/guidance/CG94↵ScottishIntercollegiateGuidelinesNetwork.Acutecoronarysyndromes:anationalclinicalguideline.(93)Edinburgh,UK:ScottishIntercollegiateGuidelinesNetwork,2007.↵TheTaskForcefortheManagementofAcuteCoronarysyndromesoftheEuropeanSocietyofCardiology.ESCguidelinesforthemanagementofacutecoronarysyndromesinpatientspresentingwithoutpersistentST-segmentelevation.EHeartJ2011;32:2999–3054.OpenUrlFREEFullText↵TheTaskForcefortheManagementofunstableangina/nonSTelevationmyocardialinfarctionoftheAmericanCollegeofCardiologyFoundation/AmericanHeartAssociation.ACCF/AHAfocusedupdateoftheguidelineforthemanagementofpatientswithunstableangina/nonSTelevationmyocardialinfarction:areportoftheAmericanCollegeofCardiologyFoundation/AmericanHeartAssociationTaskForceonPracticeGuidelines.Circulation2012;126:875–910.OpenUrlFREEFullText↵FoxKAA,DabbousOH,GoldbergRJ,etal.PredictionofriskofdeathandmyocardialinfarctioninthesixmonthsafterpresentationwithACS:aprospective,multinational,observationalstudy(GRACE).BMJ2006;333:1091–4.OpenUrlAbstract/FREEFullText↵GrangerCB,GoldbergRJ,DabbousO,etal.PredictorsofhospitalmortalityintheGlobalRegistryofAcuteCoronaryEvents.ArchInternMed2003;63:2345–53.OpenUrl↵AntmanEM,CohenM,BerninkPJ,etal.TheTIMIriskscoreforunstableangina/nonSTelevationMI:amethodforprognosticationandtherapeuticdecisionmaking.JAMA2000;284:835–42.OpenUrlCrossRefPubMedWebofScience↵BoersmaE,PieperKS,SteyerbergEW,etal.PredictorsofoutcomeinpatientswithacutecoronarysyndromeswithoutpersistentSTsegmentelevation.Resultsfromaninternationaltrialof9461patients.ThePURSUITInvestigators.Circulation2000;101:2557–67.OpenUrlAbstract/FREEFullText↵JacobsDRJr.,KroenkeC,CrowR,etal.PREDICT:asimpleriskscoreforclinicalseverityandlong-termprognosisafterhospitalisationforacutemyocardialinfarctionorunstableangina:theMinnesotaheartsurvey.Circulation1999;100:599–607.OpenUrlAbstract/FREEFullText↵GaleCP,MandaSO,WestonCF,etal.EvaluationofriskscoresforriskstratificationofacutecoronarysyndromesintheMyocardialInfarctionNationalAuditProject(MINAP)Database.Heart2008;95:221–7.OpenUrl↵MorrowDA,AntmanEM,GiuglianoRP,etal.AsimpleriskindexforrapidinitialtriageofpatientswithSTelevationmyocardialinfarction:anInTIMEIIsubstudy.Lancet2001;358:1571–5.OpenUrlCrossRefPubMedWebofScience↵KurzDJ,BernsteinA,HuntK,etal.Simplepointofcareriskstratificationinacutecoronarysyndromes:theAMISmodel.Heart2009;95:662–8.OpenUrlAbstract/FREEFullText↵PiomboAC,GagliardiJA,GuettaJ,etal.Anewscoringsystemtostratifyriskinunstableangina.BMCCardiovascDisord2003;3:8.OpenUrlCrossRefPubMed↵SimmsAD,ReynoldsS,PieperK,etal.EvaluationoftheNICEmini-GRACEriskscoresforacutemyocardialinfarctionusingtheMyocardialIschaemiaNationalAuditProject(MINAP)2003–2009:NationalInstituteforCardiovascularOutcomesResearch(NICOR).Heart2013;99:35–40.OpenUrlAbstract/FREEFullText↵FoxKAA,AndersonFAJr.,DabbousOH,etal.Interventioninacutecoronarysyndromes:dopatientsundergointerventiononthebasisoftheirriskcharacteristics?Theglobalregistryofacutecoronaryevents(GRACE).Heart2007;93:177–82.OpenUrlAbstract/FREEFullText↵YanAT,YanRT,TanM,etal.Riskscoresforriskstratificationinacutecoronarysyndromes:usefulbutsimplerisnotnecessarilybetter.EurHeartJ2007;28:1072–8.OpenUrlAbstract/FREEFullText↵StegPG,FitzGeraldG,FoxKAA.Riskstratificationinnon-ST-segmentelevationacutecoronarysyndromes:troponinisnotenough.AmJMed2009;122:107–8.OpenUrlCrossRefPubMedWebofScience↵FoxKAA,EagleKA,GoreJM,etal.TheGlobalRegistryofAcuteCoronaryEvents1999to2009—GRACE.Heart2010;96:1095–101.OpenUrlAbstract/FREEFullText↵FoxKAA,AndersonF,GoodmanS,etal.Timecourseofeventsinacutecoronarysyndromes:implicationsforclinicalpractice.TheGRACEregistry.NatClinPractCardiovascMed2008;5:580–9.OpenUrlCrossRefPubMedWebofScience↵FoxKAA,CarruthersKF,DunbarDR,etal.Underestimatedandunder-recognized:thelateconsequencesofacutecoronarysyndrome(GRACEUK—BelgianStudy).EurHeartJ2010;31:2755–64.OpenUrlAbstract/FREEFullText↵PieperKS,GoreJM,FitzgerlandG,etal.Validityofarisk-predictiontoolforhospitalmortality:theGlobalRegistryofAcuteCoronaryEvents.AmHeartJ2009;157:1097–105.OpenUrlCrossRefPubMedWebofScience↵CambouJ-P,SimonT,MulakG,etal.TheFrenchregistryofAcuteSTelevationornonSTelevationMyocardialInfarction(FAST-MI):studydesignandbaselinecharacteristics.ArchMalCoeurVaiss2007;100:524–34.OpenUrlPubMedWebofScience↵DanchinN,FauchierL,MarijonE,etal.Impactofearlystatintherapyondevelopmentofatrialfibrillationattheacutestageofmyocardialinfarction:datafromtheFAST-MIregister.Heart2010;96:1809–14.OpenUrlAbstract/FREEFullText↵HarrellFEJr..Regressionmodelingstrategies:withapplicationstolinearmodels,logisticregressionandsurvivalanalysis.1stedn.NewYork:Springer,2001.↵SimonT,VerstuyftC,KrauseMM,etal.Geneticdeterminantsofresponsetoclopidogrelandcardiovascularevents.NEnglJMed2009;360:363–75.OpenUrlCrossRefPubMedWebofScience↵MayS,HosmerDW.AcautionarynoteontheuseoftheGronnesbyandBorgangoodness-of-fittestfortheCoxproportionalhazardsmodel.LifetimeDataAnal2004;10:283–91.OpenUrlCrossRefPubMedWebofScience↵HarrellFEJr.,CaliffRM,PryorDB,etal.Evaluatingtheyieldofmedicaltests.JAMA1982;247:2543–6.OpenUrlCrossRefPubMedWebofScience↵HosmerDWJr.,LemeshowS,MayS.Appliedsurvivalanalysis:regressionmodellingoftimetoeventdata.2ndedn.Hoboken,NJ:Wiley-Blackwell,2008.↵FoxKAA,ClaytonTC,DammanP,etal.Long-termoutcomeofaroutineversusselectiveinvasivestrategyinpatientswithnon-ST-segmentelevationacutecoronarysyndrome:ameta-analysisofindividualpatientdata.JAmCollCardiol2010;55:2435–45.OpenUrlCrossRefPubMedWebofScience↵LagerqvistB,HustedS,KontnyF,etal.5-yearoutcomesintheFRISC-IIrandomisedtrialofaninvasiveversusanon-invasivestrategyinnonSTelevationacutecoronarysyndrome:afollow-upstudy.Lancet2006;368:998–1004.OpenUrlCrossRefPubMedWebofScience↵FoxKAA,Poole-WilsonP,ClaytonTC,etal.LongtermimpactofaninterventionalstrategyinnonSTelevationacutecoronarysyndrome:5yearoutcomeoftheBHFRITA3study.Lancet2005;366:914–20.OpenUrlCrossRefPubMedWebofScience↵DammanP,HirschA,WindhausenF,etal.5-yearclinicaloutcomesintheICTUS(InvasiveversusConservativeTreatmentinUnstablecoronarySyndromes)trial:arandomisedcomparisonofanearlyinvasiveversusselectiveinvasivemanagementinpatientswithnonSTsegmentelevationacutecoronarysyndrome.JAmCollCardiol2010;55:865–6.OpenUrlPubMed↵KarenSP,JoelMG,GordonF,etal.Validityofarisk-predictiontoolforhospitalmortality.AmHeartJ2009;157:1097–105.OpenUrlCrossRefPubMedWebofScience↵WebsitesupportforuseoftheGRACEriskscore:http://www.gracescore.co.ukandhttp://www.outcomes.org/grace.ViewAbstract Footnotes ContributorsKAAFinitiatedtheprogrammeofwork,performedtheanalysesinconjunctionwithcoauthorsandwroteandrevisedthemanuscript.GFledtheworkofderivingtherevisedriskscore,inconjunctionwithFAandWH.KCanalysedandinterpretedthedata.ND,inconjunctionwithEP,TS,PC,JMandP-GS,performedtheworkontheFrenchregistryofAcuteST-elevationandnon-ST-elevationMI(FASTMI)dataset.Allauthorscontributedtotherevisionsofthemanuscriptandtheinterpretationofthefindings. FundingThisworkwassupportedbytheBritishHeartFoundation(CH92010),anawardfromtheChiefScientistScotland(CZG/2/455),andanunrestrictedgrantfromAstraZenecatotheUniversityofEdinburgh.TheFrenchregistryofAcuteST-elevationandnon-ST-elevationMI(FAST-MI)registryisaregistryoftheFrenchSocietyofCardiology,supportedbyunrestrictedgrantsfromPfizerandServier.AdditionalsupportforFAST-MIwasobtainedfromaresearchgrantfromtheFrenchNationalHealthInsuranceSystem(CaisseNationaled’AssuranceMaladie). CompetinginterestsNone. EthicsapprovalEthicsapprovalwasobtainedinalloftheparticipatingcentresoftheGRACEstudy.5,6 ProvenanceandpeerreviewNotcommissioned;externallypeerreviewed. DatasharingstatementTheGRACEriskscorehasbeenmadeavailabletodownloadwithoutcost(http://www.gracescore.co.ukandhttp://www.outcomes.org/grace). RequestPermissions IfyouwishtoreuseanyorallofthisarticlepleaseusethelinkbelowwhichwilltakeyoutotheCopyrightClearanceCenter’sRightsLinkservice.Youwillbeabletogetaquickpriceandinstantpermissiontoreusethecontentinmanydifferentways. Copyrightinformation: PublishedbytheBMJPublishingGroupLimited.Forpermissiontouse(wherenotalreadygrantedunderalicence)pleasegotohttp://group.bmj.com/group/rights-licensing/permissionsThisisanOpenAccessarticledistributedinaccordancewiththeCreativeCommonsAttributionNonCommercial(CCBY-NC3.0)license,whichpermitsotherstodistribute,remix,adapt,builduponthisworknon-commercially,andlicensetheirderivativeworksondifferentterms,providedtheoriginalworkisproperlycitedandtheuseisnon-commercial.See:http://creativecommons.org/licenses/by-nc/3.0/ ReadthefulltextordownloadthePDF: Subscribe Login LoginviaInstitutionLoginviaOpenAthens Loginusingyourusernameandpassword ForpersonalaccountsORmanagersofinstitutionalaccounts Username* Password* Forgotyourlogindetails?Registeranewaccount?Forgotyourusernameorpassword?