Improving risk stratification in non-ST-segment elevation myocardial infarction with combined assessment of GRACE and CRUSADE risk scores

Availableonlineat

ScienceDirect

www.sciencedirect.com

CLINICAL

RESEARCH

Improving

risk

stratification

in

non-ST-segment

elevation

myocardial

infarction

with

combined

assessment

of

GRACE

and

CRUSADE

risk

scores

L’utilisation

combinée

des

scores

GRACE

et

CRUSADE

pour

la

stratification

du

risque

d’infarctus

du

myocarde

Luis

Paiva

,

Rui

Providência

,

Sérgio

N.

Barra

,

Paulo

Dinis

,

Ana

C.

Faustino

,

Marco

Costa

,

Lino

Gonc

¸alves

a_{Coimbra’sHospitalandUniversityCentre,Coimbra,Portugal} b_{FacultyofMedicine,UniversityofCoimbra,Coimbra,Portugal} c_{PapworthHospital,Cambridge,England,UK}

Received1stApril2014;receivedinrevisedform12June2014;accepted24June2014 Availableonline11September2014

KEYWORDS Myocardialinfarction; Risk; GRACE; CRUSADE Summary

Background.—Riskassessmentisfundamentalinthemanagementofacutecoronarysyndromes (ACS),enablingestimationofprognosis.

Aims.—To evaluate whether the combined use of GRACE and CRUSADE risk stratification schemes inpatients with myocardialinfarctionoutperformseach ofthe scores individually intermsofmortalityandhaemorrhagicriskprediction.

Methods.—Observationalretrospectivesingle-centrecohortstudyincluding566consecutive patients admittedfor non-ST-segment elevationmyocardial infarction.The CRUSADE model increased GRACE discriminatory performance in predicting all-cause mortality, ascertained

Abbreviations:ACS,acutecoronarysyndrome;AUC,areaunderthecurve;CI,confidenceinterval;CRUSADE,canrapidriskstratification

ofunstableanginapatientssuppressadverseoutcomeswithearlyimplementationoftheAmericanCollegeofCardiology/AmericanHeart Associationguidelines;GRACE,GlobalRegistryofAcuteCoronaryEvents;HR,hazardratio;IDI,integrateddiscriminationimprovement;NRI, NetReclassificationIndex;NSTEMI,non-ST-elevationmyocardialinfarction;ROC,receiveroperatingcharacteristic.

∗_{Correspondingauthor.QuintadosVales,3041-801,S.MartinhodoBispo,Portugal.}

E-mailaddress:[email protected](L.Paiva).

http://dx.doi.org/10.1016/j.acvd.2014.06.008

byCoxregression,demonstratingCRUSADEindependentandadditivepredictivevalue,which was sustained throughoutfollow-up. The cohortwas divided into four different subgroups: G1 (GRACE<141; CRUSADE<41); G2 (GRACE<141; CRUSADE_≥41); G3 (GRACE_≥141; CRU-SADE<41);G4(GRACE≥141;CRUSADE≥41).

Results.—Outcomesandvariablesestimatingclinicalseverity,suchasadmissionKillip-Kimbal classandleftventricularsystolicdysfunction,deterioratedprogressivelythroughoutthe sub-groups(G1toG4).Survivalanalysisdifferentiated threeriskstrata (G1,lowestrisk;G2and G3,intermediaterisk;G4,highestrisk).TheGRACE+CRUSADEmodelrevealedhigher progno-sticperformance(areaunderthecurve[AUC]0.76)thanGRACEalone(AUC0.70)formortality prediction,furtherconfirmedbytheintegrateddiscriminationimprovementindex.Moreover, GRACE+CRUSADEcombinedriskassessmentseemedtobevaluableindelineatingbleedingrisk inthissetting,identifyingG4asaveryhigh-risksubgroup(hazardratio3.5;P<0.001). Conclusions.—CombinedriskstratificationwithGRACEandCRUSADEscorescanimprovethe individualdiscriminatorypowerofGRACEandCRUSADEmodelsinthepredictionofall-cause mortalityandbleeding.Thiscombinedassessmentisapracticalapproachthatispotentially advantageousintreatmentdecision-making.

©2014ElsevierMassonSAS.Allrightsreserved.

MOTSCLÉS Syndromecoronaire aigu; Risque; GRACE; CRUSADE Résumé

Contexte.—L’évaluationdesrisquesestfondamentaledanslagestiondessyndromes coronar-iensaigus,permettantl’estimationdupronostic.

Objectifs.—Lebutdenotreétudeétaitd’évaluerl’utilisationcombinéedesscoresGRACEet CRUSADEpourlastratificationdelamortalitéetdurisquehémorragiquedespatientsprisen chargepouruninfarctusaigudumyocardeencomparaisonàl’utilisationisoléedechacunde cesscores.

Méthodes.—Cohorterétrospectiveobservationnellemonocentriqueayantinclus566patients consécutifshospitaliséspourunsyndromecoronarienaigusanssus-décalagedusegmentST. LescoreCRUSADEaaugmentélepouvoirdiscriminantduscoreGRACEpourlaprédictionde lamortalitéglobale, enutilisantla régressiondeCox, cequidémontrelavaleur prédictive indépendanteetadditiveduscoreCRUSADE,laquelleétaitmaintenuetoutaulongdusuivi.La cohorteaétédiviséeen4sous-groupes:G1(GRACE<141;CRUSADE<41);G2(GRACE<141; CRUSADE≥41);G3(GRACE≥141;CRUSADE<41);G4(GRACE≥141;CRUSADE≥41).

Résultats.— Lesévénementsetvariablesquiévaluaientlasévéritéclinique,commelaclasse Killip-Kimbal à l’admission et la dysfonction systolique du ventricule gauche étaient plus fréquentsde manièrelinéaire enfonctiondes sous-groupes(G1—G4).L’analysedela survie amontré3groupesderisque(G1,risquebas;G2etG3,risqueintermédiaire;G4,risqueplus élevé).LemodèleGRACE+CRUSADEamontréuneperformancepronostiquesupérieure(AUC 0,76)auscoreGRACEutilisédemanièreisolé(AUC0,70)pourlaprédictiondelamortalité, ce quiaété confirméparl’amélioration de l’index dela discrimination intégrée. Deplus, l’évaluationcombinéedesscoresGRACE+CRUSADEsembleavoirunevaleuradditionnellepour laprédictionderisquedesaignementetpermetd’identifierlegroupeG4commeétantàrisque trèsélevé(HR3,5;p=0,001).

Conclusion.—L’utilisation combinée des scores GRACE et CRUSADE pourrait améliorer leur pouvoir discriminant en comparaisonàleurutilisation isolée pour la prédiction de la mor-talitéglobaleainsiquedurisquehémorragique.Cettenouvelleapprochesembleapporterdes avantagesdanslapratiquequotidienneetorienterlapriseenchargethérapeutique.

©2014ElsevierMassonSAS.Tousdroitsréservés.

Background

Risk assessment is fundamental in acute coronary syn-drome (ACS) management,enabling estimation of patient prognosis—a key issue for communicating with patients and relatives, and for therapeutic decision-making. Cur-rent recommendations propose an aggressive treatment approachforhigh-risk non-ST-elevationmyocardial infarc-tion (NSTEMI), including more potent antithrombotic

therapies and a rapid invasive strategy [1,2]. Conversely, lower-risk casesmaydo wellwithless aggressivemedical treatment and a more selective invasive strategy. Thus, it is essential to assess ischaemic risk on an individual basis,preferablyusingquantitativeriskscoringsystemssuch as the Global Registry of Acute Coronary Events (GRACE) model [3], useofwhich isfavoured overother riskscores in the latest guidelines update [1,2]. However, with the greateruseofmorepotentantithromboticdrugsandearly

revascularization,bleedingoccursmorefrequentlyandhas becomearelevantclinicalproblemintheACSsetting, mak-inghaemorrhagicriskassessmentanecessarytooltoguide treatment strategies. The can rapid risk stratification of unstable angina patients suppress adverse outcomes with early implementationof the AmericanCollegeof Cardiol-ogy/American HeartAssociationguidelines (CRUSADE)risk score[4]isoneofthemostpopularbleedingriskalgorithms, consistingofseveralrecognizedpredictorsofhaemorrhage

[5,6].Asbleedingresultsnotonlyinanimmediatethreatbut alsoinincreasedriskofadverseoutcomesduringfollow-up

[7],itremainstobedeterminedifACSriskassessmentwith combinedischaemicandbleedingriskassessmentwillprove advantageous.Ouraimwastoestablishtheappropriateness ofthecombineduseofGRACEandCRUSADErisk stratifica-tionin NSTEMIpatients andtoevaluatepotential gainsin outcomeprediction,comparedwiththeseparateuseofthe traditionalrisk-scoringsystems.

Methods

Patient

selection

Thiswasanobservationalretrospectivesingle-centrecohort study includingallpatients consecutivelyadmitted toour University Hospital’s AcuteCardiac Care Unit witha final diagnosisofmyocardialinfarctionbetween1December2006 and31May2008.Myocardialinfarctionwasdefined accord-ingtotherecentlyupdateddefinition[8],excludingpatients withunstableanginaandthosewithmyocardialinjury (ele-vated cardiac biomarkers) without evidence of ischaemia (i.e. symptoms, electrocardiogram, imaging modalities). Furthermore,onlyNSTEMIcaseswereconsidered,withthe finalstudycohortincludingatotalof566patients.

Data

collection

and

patient

follow-up

Demographicandclinicalfeatureswerecollectedat admis-sionandduringhospitalization.Theelectrocardiogramand analyticalassessment(includingcompletebloodcountand biochemicalandclotting tests)were performedaccording to the Acute Cardiac Care Unit standards: at admission

and then at least daily, according to patient’s clinical evaluation.TroponinImeasurementsweretakenat admis-sion, between 12 and 24hours after admission and daily thereafter.The measurementoftroponinIwasperformed withthechemiluminescenttechnique(OrthoClinical Diag-nostics VITROS® _{Troponin I ES Assay; Johnson & Johnson}

Ltd., Maidenhead,UK). The lowerdetection limitfor this assayis0.012ng/mL.The 99thpercentileupperreference limit is 0.034ng/mL, with a reported imprecision of 10% coefficientofvariation.Results>0.034ng/mLwere consid-ered positive. Creatinine clearance was estimated using themodificationofdietinrenaldiseaseequation[9].The reference for coronary angiography and potential percu-taneous myocardial revascularization was an individually tailoreddecision,involvingtheAcuteCardiacCareUnitand theinterventionalcardiologist’sclinicaljudgment,in accor-dancewiththeEuropeanSocietyofCardiologyguidelinesfor myocardialinfarctionmanagement[1].Finally,left ventri-cularejectionfractionwasobtainedfromthepredischarge transthoracicechocardiogram,inaccordancewithEuropean AssociationofCardiovascularImagingstandards[10].

Patientswerefollowedfor 21.1±7.5monthsafter dis-charge by means of patient’s clinical records, routine visits,consultationoftheNationalHealthSystemUserCard databaseandtelephonecallsuntiltheendofa2-yearperiod afterdischarge,andwheneverclinicalfileswereconsidered insufficient.

Risk

assessment

We tested and compared the prognostic performance of GRACE[3]andCRUSADE[4]riskstratificationmodelsinthis cohort, throughevaluation of their overall discriminative performanceandcalibrationinthepredictionofall-cause mortalityduringtheindexevent,follow-upandin-hospital bleeding, respectively. The traditional risk categories of GRACEandCRUSADEscoresaredepictedinSupplementary Table1.TheGRACEscoreforin-hospitalmortality(GRACEIH)

is more commonly used in clinical practice than the 6-monthpostdischargeGRACEscore,becausetheformermay guiderevascularization timing in NSTEMI (i.e. patients at high ischaemic risk [GRACE ≥ 141] should be considered for an early invasive strategy) [1,11]. Subsequently, the

Table1 CohortdistributionaccordingtoGRACEandCRUSADEriskclasses.

Low-riskclass Intermediate-riskclass High-riskclass

GRACEIH 77(13.6) 135(23.9) 354(62.5)

GRACE6M 155(27.3) 219(38.7) 192(34.0)

CRUSADE 222(39.2) 126(22.3) 218(38.5)

Cross-tabulationofCRUSADEandGRACEriskcategories

GRACE

Low-riskclass Intermediate-riskclass High-riskclass CRUSADE

Low-riskclass 58(26.1) 83(37.4) 81(36.5)

Intermediate-riskclass 12(9.5) 28(22.2) 86(68.3)

High-riskclass 6(2.8) 28(12.8) 184(84.4)

cohort was divided into four different groups according tothe presence of at least one high-risk category (using in-hospital GRACE and CRUSADE cut-offs used in clinical practice)[1,2]:group1(G1:GRACE<141non-high-riskclass; CRUSADE<41non-high-riskclass),group2(G2:GRACE<141, non-high-risk class; CRUSADE≥41, high-risk class); group 3 (G3: GRACE≥141, high-risk class; CRUSADE<41, non-high-riskclass); group4(G4:GRACE≥141,high-risk class; CRUSADE≥41,high-riskclass).Eachgroupwasevaluatedin termsofbaselinecharacteristicsandstudyendpoints.

Inthisstudy,majorbleedingwasdefinedinaccordance with the CRUSADE investigators [4]: intracranial hae-morrhage,documentedretroperitonealbleed,haematocrit drop≥12%(frombaseline),any redblood cell transfusion whenbaselinehaematocritwas≥28%oranyredbloodcell transfusionwhenbaselinehaematocritwas<28%with wit-nessedbleed.

Study

endpoints

The primaryoutcome measures werein-hospitalall-cause mortality, all-cause mortality during follow-up and in-hospitalmajorbleeding.

Statistical

analysis

Statistical analyses were done usingSPSS® _software,

ver-sion 17.0 (StataCorp LP, College Station, Texas, USA). Whenneeded,baselinecharacteristicsweredescribedwith means±standarddeviationsforcontinuousandcountsand proportionsforcategoricaldata.TheKolmogorov—Smirnov testwasusedtotestthenormaldistributionofcontinuous variables.TheChi2_{testandStudent’sttestwereusedfor}

quantitativeandnominalcomparisonsbetweentwogroups, andnon-parametricequivalenttestswereusedwhen appro-priate. Regression estimation techniques were applied to replacemissingvalueswheneverthenumberofmissing val-ueswasnegligible,otherwisecaseswithmissingvalueswere omitted.Pvalues<0.05(two-sided)wereconsidered statis-ticallysignificant.

Univariateanalysiswasperformedtoevaluatethe poten-tialassociationbetweeneachpreviouslydefinedmyocardial infarction group and the study endpoints. Cox regression wasusedtoevaluatethepredictivevalueoftheCRUSADE score compared with the GRACE algorithm for follow-up mortality. The analysis of variance (or equivalent non-parametric test, when necessary) wasused to determine differencesamongthepredefinedmyocardialinfarction sub-groupmeans.Discrimination,measuredintermsofthearea under the receiver operating characteristic (ROC) curve (AUC),wasperformedtoassessthepredictivepowerofthe GRACE score in-hospital (GRACEIH) and 6 months

postdis-charge (GRACE6M) for in-hospital and follow-up mortality,

respectively, and of the CRUSADE model for in-hospital majorbleeding.Finally,thecombinedGRACEandCRUSADE model (GRACE+CRUSADE) was tested for in-hospital and follow-upmortalitiesandmajorbleeding.Other measures ofincrementalvalue have been proposed,whichexamine theextenttowhichamodelreclassifiessubjects,suchasthe netreclassificationindex(NRI)andtheintegrated discrim-inationimprovement(IDI)[12].TheNRImethod,described byPencinaetal.[13],statesthatapositiveandsignificant

NRItranslatesanetoverallsuccessfulreclassificationof sub-jectsintoamoreappropriateriskcategory.TheIDI,which may be seen as a continuous form of the NRI, assesses improvementinriskdiscriminationbyestimatingthechange in the difference of the average of predicted probabili-tiesofaneventbetweenthosewithandwithouttheevent underconsideration[14];itisamoreappropriatemeasure of riskreclassification whencomparingscores with differ-entrisk categorization(e.g. GRACEstratifiespatientsinto three risk strata and GRACE+CRUSADE stratifies patients into four risk categories). Calibration of each score was also assessed using the Hosmer—Lemeshow test. Finally, Kaplan—Meiercurveswereconstructedtoevaluatesurvival during follow-upaccording toeach predefinedmyocardial infarctiongroup.

Results

Cohort

characteristics

The cohort included 566 patients with a mean age of 70.4±12.3years(range31—92years),61.3%ofwhomwere men. Patients’ baseline clinical, analytical and imaging characteristics are shown in Supplementary Table 1. The cohort distribution according toGRACE and CRUSADE risk values/categoriesaregiveninTable1.In270(47.7%)cases therewasoverallconcordancebetweenGRACEIH and

CRU-SADEriskcategories;184(32.5%)patientswereclassifiedas high-riskbybothGRACEandCRUSADEriskmodels(Table1).

Risk

score

performance

ThediscriminationperformancesofGRACEIH(forin-hospital

mortality), GRACE6M (for follow-up mortality) and

CRU-SADE(for majorbleeding) were tested inourcohort, and theirdiscriminationperformancesaredisplayedinTable2. All scores showed good calibration, as demonstrated by Hosmer—LemeshowtestPvalues>0.05.

Table2 Cohortriskmodelperformances,usingas con-tinuous variables and as risk score categories (low-, intermediate-andhigh-riskclasses).

AUC(95%CI) P Continuousvariable GRACEIH 0.81(0.75—0.88) <0.001 GRACE6M 0.78(0.73—0.83) <0.001 CRUSADE 0.70(0.62—0.77) <0.001 Categoricalvariable GRACEIH 0.70(0.64—0.76) <0.001 GRACE6M 0.74(0.69—0.83) <0.001 CRUSADE 0.69(0.62—0.76) <0.001 GRACE+CRUSADEIH 0.76(0.70—0.82) <0.001 GRACE+CRUSADE6M 0.78(0.73—0.83) <0.001

GRACE+CRUSADEbleed 0.66(0.59—0.74) <0.001

AUC:areaunderthecurve;CI:confidenceinterval;CRUSADE: in-hospital major bleeding; GRACE6M: 6-month postdischarge

mortality;GRACEIH:in-hospitalmortality;GRACE+CRUSADE6M:

follow-up mortality; GRACE+CRUSADEbleed: in-hospital major

Table3 Cohortsubgroupanalysis. Subgroups P G1(GRACE<141; CRUSADE<41) G2(GRACE<141; CRUSADE≥41) G3(GRACE≥141; CRUSADE<41) G4(GRACE≥141; CRUSADE≥41) (n=173;30.6%) (n=53;9.4%) (n=168;29.6%) (n=172;30.4%) Age(years) 59.7±11.1 69.2±9.7 74.3±8.6 78.3±8.3 <0.001 Men 139(80.3) 15(28.3) 119(70.8) 74(43.0) <0.001 Killip-Kimballclass (admission) 1.0±0.4 1.1±0.7 1.2±0.7 1.8±0.9 <0.001

Systolicarterialpressure (mmHg) 140.9±23.8 148.4±34.1 127.9±20.0 128.0±27.0 <0.001 Heartrate(bpm) 72.8±15.4 76.5±17.0 75.9±16.7 84.6±22.7 <0.001 Haemoglobin(g/dL) 14.6±1.6 12.2±1.5 13.5±1.7 11.9±1.9 <0.001 Haematocrit(%) 43.8±4.7 36.6±4.7 40.4±5.0 35.7±5.9 <0.001 Glycaemia(mmol/L) 7.4±3.6 8.6±5.1 8.1±3.8 10.5±5.5 <0.001 HaemoglobinA1c(%) 6.2±1.3 6.3±1.6 6.4±1.3 7.1±1.8 <0.001

Serumcreatinine(mol/L) 83.6±20.0 168.8±167.4 89.9±28.9 204.1±174.8 <0.001 Creatinineclearancea (mL/min) 78.5±26.2 43.6±22.9 68.0±25.9 45.4±26.7 <0.001 MaximumtroponinI (ng/mL) 12.9±20.5 16.7±34.5 20.9±25.6 25.2±46.1 0.006 NT-proBNP(pg/mL) 857.3±1131.9 3173.9±3497.9 3382.3±3745.2 19785.8±26710.7 <0.001 GRACEscore <0.001 Inhospital 111.6±21.1 127.4±17.0 166.9±21.0 195.7±31.6 <0.001 6-month 92.7±19.2 110.9±15.6 138.0±17.1 161.91±23.2 CRUSADEscore 27.9±12.9 49.0±9.5 32.73±9.8 50.9±14.5 <0.001 LVEF<40%b_{[3,4] 10(5.8) 8(15.1) 35(20.8) 61(35.5) <0.001} Myocardial revascularization 96(55.5) 25(47.2) 75(44.6) 46(26.7) <0.001 Percutaneouscoronary revascularization 83(48.0) 21(39.6) 69(41.1) 39(22.7) <0.001 Surgicalcoronary revascularization 13(7.5) 4(7.5) 6(3.6) 7(4.1) 0.062 Three-vesselcoronary disease 14(8.1) 4(7.5) 13(7.7) 24(13.9) 0.061 Inhospitalmortality 0(0) 2(3.8) 11(6.5) 24(14.0) <0.001 Follow-upmortality 5(3.4) 7(13.2) 25(14.9) 64(37.2) <0.001 Majorbleeding 5(2.9) 8(15.1) 8(4.8) 31(18.0) <0.001 Reinfarction 12(6.9) 6(11.3) 23(13.7) 35(20.3) <0.001 Heartfailure hospitalization 18(10.4) 8(15.1) 29(17.3) 57(33.1) <0.001

Dataarenumber(%)ormean±standarddeviation.bpm:beatsperminute;LVEF:leftventricularejectionfraction;NT-proBNP:N-terminal pro-brainnatriureticpeptide.

a _{Clearanceofcreatinineasperthemodifieddietinrenaldiseaseequation.} b _{Predischargetransthoracicechocardiogram.}

Study

endpoint

analysis

The in-hospitalmortalityrate was6.7% (n=38)and19.5% of patients (n=103) died during follow-up. Patients who reached the primary endpoints were older and had sev-eralworseclinicalandanalyticalfindings,andhigherGRACE andCRUSADEscores(SupplementaryTable2).Predictorsof deathintheunivariateanalysisareshowninSupplementary Table3,showingmyocardial revascularization(either per-cutaneousor surgical),whichwasassociatedwithalower risk of in-hospital mortality (hazard ratio [HR] 0.20, 95% confidenceinterval[CI]0.11—0.34;P<0.001)andfollow-up

mortality(HR0.24,95%CI0.14—0.41;P<0.001).Moreover, majorbleedingwasalsorelatedtofollow-upmortality(HR 2.27,95%CI1.16—4.44;P=0.015).In-hospitalmajor bleed-ingoccurredin52(9.2%)patients.

Subgroup

analysis

The cohort was divided into four different subgroups, accordingtoGRACE and/or CRUSADEhigh-risk categoriza-tion.Eachgroup’sbaselinecharacteristicsandoutcomesare showninTable3.

The subgroups G1, G3 and G4 each included a sim-ilar number of cases. Patients’ average age, admission Killip-Kimbal class and left ventricular systolic dysfunc-tion increased across the four groups, with G3 (high-risk GRACE/low-risk CRUSADE) and G4 (high-risk GRACE/high-riskCRUSADE)showingthehighestvalues.Admissionsystolic blood pressure was reduced and peak troponin I and N-terminal pro-brainnatriuretic peptide values werehigher inthegroupswithGRACE≥141(G3andG4).

InthosegroupswithhigherCRUSADEscores,suchasG2 (low-risk GRACE/high-riskCRUSADE) and G4,patients had lowerhaemoglobinandhaematocritandhighercreatinine andadmissionglycaemiavalues.

The lowest-risk patients (G1) (low-risk GRACE/low-risk CRUSADE)weresignificantlymoreoftenrevascularizedthan patients in higher-risk groups, especially compared with G4(highest-riskgroup).Intermsofrevascularization strat-egy,the proportion of patients referred for percutaneous coronary intervention or surgery was balanced between groups (Table 3). Complex coronary disease (three-vessel disease)wasmorefrequentinG4and,simultaneously,fewer revascularization options (percutaneous or surgical) were consideredsuitableforthehighest-riskgroup(G4). Follow-upreinfarctionandheartfailurehospitalizationincreasingly occurredthroughouttherisksubgroups(G1toG4),withG4 showingthehighestrates.

Overall, the subgroup analysis allowed the identifica-tion of a low-risk class (G1) and a high-risk class (G4) plus two intermediate-risk classes (G2 and G3). Although G1 and G2 included patients with GRACE<141, G2 (CRU-SADE≥41) patients had a poorer outcome than those in G1(CRUSADE<41). Moreover,while both G3 and G4 com-prisedpatients withGRACE≥141, G4 (CRUSADE≥41)had theworstprognosis.

Subgroup

mortality

analysis

Deathsin-hospitalandduringfollow-upoccurredmore fre-quently in groups with high-risk GRACE values (≥141; G3 andG4).However,deathwasmorecommonly observedin G2(GRACE<141,CRUSADE≥41)thaninthelowest-risk cat-egory(G1).

TheKaplan—MeieranalysisisshowninFig.1.Thesurvival curveofG1(lowrisk) separatesearlyfromthe other sub-groupcurves.TheG2andG3(intermediaterisk)curveswere similarduringtheaveragefollow-up,displayingamuch bet-tersurvivalcoursethanthosepatientsclassifiedinG4(high risk).Differencesinsurvivalbetweengroupsweresustained andcumulativethroughoutfollow-up.

As previously mentioned, revascularization was asso-ciated with higher follow-up and in-hospital survival. In the subgroup analysis, revascularizationimpacted on sur-vivalonly in the groups withGRACE>140: G3 (in-hospital mortality, HR 0.23, 95% CI 0.03—2.07, P=0.15; follow-up mortality, HR 0.20, 95% CI 0.07—0.59, P=0.002); and G4 (in-hospital mortality, HR 0.42, 95% CI 0.11—0.63,

P=0.003; follow-upmortality, HR0.35, 95%CI0.16—0.75,

P=0.006).

Thetime-to-eventmodel(Coxregression)revealedthat theprognosticvalueoftheCRUSADEscore(HR1.03,95%CI 1.01—1.04;P<0.001)wasindependentandadditivetothat

Figure1. Survivalanalysisaccordingtopredefinedsubgroups.

oftheGRACE6Mscore(HR1.02,95%CI1.01—1.03;P<0.001)

forthepredictionofmortalityduringfollow-up.

The ROC curve comparison (Table 2) between GRACE (low-, intermediate- and high-risk categories), CRU-SADE (low-, intermediate- and high-risk categories) and

GRACE+CRUSADE combined (G1, G2, G3 and G4) showed

thatthecombinationhadhigherdiscriminatoryperformance forin-hospitalandfollow-upmortalityprediction,although differences were not statistically significant (P>0.05). However, IDI (Table 4) confirmed that GRACE+CRUSADE improved risk reclassification for both in-hospital and follow-up mortalities (GRACE+CRUSADEIH, relative IDI

17.1%;GRACE+CRUSADE6M,relativeIDI11.5%).

Subgroup

bleeding

analysis

Majorbleedingratewasincreasedinthesubgroupswitha high-risk CRUSADEcategory(G2,G4).Inthegroupswitha low-risk CRUSADEscore (G1andG3),bleedingwas signifi-cantlymorefrequentinpatientswithGRACE≥141(G3)than inpatientswithGRACE<141(G1).Moreover,majorbleeding riskwas4.5timeslowerinG1thanintheothersubgroups (HR [G2—G4] 4.5, 95% CI 1.8—11.7; P=0.001), establish-ing G1 asa very low bleeding risk. Additionally, patients in G4hadaveryhigh risk ofmajorbleeding(HR3.5, 95% CI 2.0—6.3; P<0.001) compared with the remaining sub-groups,includingG2,whichalreadycomprisedpatientswith CRUSADE>41(highbleedingrisk).

Discussion

The combined risk assessment with GRACE and CRUSADE modelsimprovedtheoverallriskstratificationprovidedby eachscoreindividuallyintheNSTEMIsetting.Thiscombined evaluationenabledthedistinctionofavery—low-riskgroup (GRACE<141, CRUSADE<41) and a very—high-risk group (GRACE≥141, CRUSADE≥41) from the other cases with anintermediate-riskpattern(presenceofeitherGRACEor

Table4 Integrated discrimination improvement comparing GRACE+CRUSADE model with GRACE and CRUSADE risk categories.

GRACE+CRUSADEIH GRACEIH GRACE+CRUSADE6M GRACE6M

Averageofestimated probabilitiesofanevent 0.119 0.103 0.350 0.327 Averageofestimated probabilitiesofa non-event 0.064 0.065 0.173 0.181

Cross-tabulationforIDIand relativeIDIcalculation

GRACE+CRUSADE IDI,0.017;relativeIDI,17.07% IDI,0.031;relativeIDI,11.45%

GRACE6M: 6-month postdischarge mortality; GRACEIH: in-hospital mortality; GRACE+CRUSADE6M: follow-up mortality;

GRACE+CRUSADEIH:in-hospitalmortality;IDI:integrateddiscriminationimprovement.

CRUSADEhigh-riskclasses).CRUSADEsignificantlyenhanced theprognosticperformanceoftheGRACEscore.

There isconsiderable variabilityin patient characteris-ticsandoutcomesacrosstheACSspectrum,andasystematic assessmentoftheprobabilityofadverseeventsby quanti-tative riskmodelscanhelptoguidetreatment strategies. Although severalACS risk prediction tools have been pro-posedinrecentyears,themostrobustonesforevaluating ischaemic and bleeding risk are the GRACE and CRUSADE scores, respectively [1,2]. These risk algorithms are

rec-ommended by contemporary guidelines and have been

incorporatedintoclinicalpracticewithpotential improve-ments in decision-making. However, some concerns have alsobeen raised concerningthe‘treatment—risk’ paradox incurrent internationalpractice[14],inwhich higher-risk patients are less likely toreceive more aggressive treat-ment than lower-risk cases. Notwithstanding, the early versusdelayedtimingofinterventioninpatientswithacute coronary syndromes (TIMACS) trial [11,15] showed early coronaryangiographytobeadvantageousin patientswith GRACE>140,similartowhatwasobserved forST-segment elevation myocardial infarction in each GRACE category

[16]. However, there are no other studies evaluating the impactofACSriskscoresinothertreatmentmodalities,such asantithromboticoranticoagulationtherapies.

Major bleeding is one of the most common serious

adverse events in patients admitted with an ACS [4]. In thisclinicalsetting,thereisastrongrelationshipbetween bleedingandmortality,evenwhenthehaemorrhageisnot consideredtobesevere.Majorbleedingisassociatedwith a60%increaseinhospitaldeath[17]andafivefoldincrease in1-yearmortality[7].The bleeding-mortalityinteraction seemstobeattributabletomorethanthespecificbleeding episode. Asignificant haemorrhage maylead tocomplete cessationofantithrombotictherapyandpotentialischaemic recurrences. Moreover, the advancing ACS therapies are increasinglyofferedtohigher-riskpatients(e.g.theelderly, thosewithco-morbidities)whoalsohaveanincreasedrisk ofbleedingcomplications.Therefore,ACSriskstratification needstobereliableinoutliningthepatient’sriskprofile.We believe that ischaemic andbleeding risksshould be eval-uatedsimultaneously. Bleedinghas an impactbeyond the indexeventandACSmanagementismuchmorethantotal ischaemic burden. Besides, ischaemia and bleeding share

overlappingriskfactors(e.g.olderage,diabetes,renal dys-function),anditisnotuncommontofindanACSpatientwho isathighriskofdeath/ischaemicrecurrencesandis simulta-neouslyatincreasedriskofdismalbleedingcomplications. Inourcohort,we foundconcordance betweenGRACEand CRUSADEriskcategoriesinapproximately50%ofcases,with nearlyone-third of patientspresenting concurrentGRACE andCRUSADEhigh-risk categories. Patientmanagementin thesecasesischallengingandweoughttounderstandmore aboutACSriskprofilesandrelatedoutcomes.

We assessed the strength of the combined

evalua-tion by GRACE and CRUSADE models using a cut-off

(GRACE≥141) related to the optimal revascularization timinginNSTEMIandthehigh-riskbleedingcategory (CRU-SADE≥41).Throughthesedivisioncriteria,whichareeasily obtainable, we sought to define four different NSTEMI risk profiles: low ischaemic and bleeding risk patients (GRACE<141; CRUSADE<41);high ischaemic and bleeding riskcases(GRACE≥141;CRUSADE≥41);andtoidentify sig-nificantdifferencesbetweentheintermediate-riskprofiles (GRACE<141; CRUSADE≥41/GRACE≥141; CRUSADE<41) andthe former subgroups. Ourresults suggest that these subgroups are very different from each other in terms of patient characteristics and outcomes. Several clinical variables, such as admission Killip-Kimbal class and left ventriclesystolic dysfunction,and outcomes deteriorated significantlythroughoutthesubgroups(G1toG4).Patients withGRACE≥141wereexpectedtohaveapooreroutcome. However,thosewithaGRACEscore<141butwithahigh-risk CRUSADEscorehadaworseprognosisthanpatientsinthe G1group(lowest-riskgroup).Similarly,patientsinG3 (high-riskGRACEclass)didnothaveaworseclinicalpicturethan thatobservedinG4(highest-riskgroup).

TheCRUSADEmodelincreasedthediscriminatory perfor-manceof GRACE in the prediction of all-cause mortality, ascertained by a time-to-event model (Cox regression),

showing CRUSADE to have an independent and additive

predictive value that is sustained throughout follow-up. This improved performance of the GRACE+CRUSADE model was demonstrated by survival curves (Fig. 1) that clearly differentiate three strata (G1, lowest-risk curve; G2 and G3, intermediate-risk curve; G4, highest-riskcurve).The GRACE+CRUSADEprognosticperformance was measured using ROC analysis, establishing a higher

combinedrisk model AUC(GRACE+CRUSADEIH, AUC0.76;

GRACE+CRUSADE6M, AUC 0.78) than GRACE categories

(GRACEIH,AUC0.70;GRACE6M,AUC0.74)forbothin-hospital

and follow-up mortalities, although the difference was not statistically significant. As AUC cannot always mea-sureaclinicallymeaningfulimprovementinreclassification, an extended statistical evaluation with IDI documenteda successfully improvementin reclassification, strongly sug-gesting that the combinedrisk model would be clinically valuable.

Myocardialrevascularizationwasassociatedwith follow-up and in-hospital survival advantage. In the subgroup analysis, revascularization benefits were only evident in thosewithGRACE>140, mostlyG4(highest-riskpatients). Yet,thehigherrateofrevascularizationwasseeninthe low-estriskgroup(G1)(‘treatment—risk’paradox).Ourresults seemedtoindicatethatcasesof lowerischaemicriskand highbleedinghazard(G3)mightbebettermanagedwitha moreconservativerevascularizationapproach.

Another key section of this study was to assess the strengthofGRACE+CRUSADEinmajorbleedingprediction andtodeterminewhetherthecombinedriskassessmentis ofgreatervaluethanCRUSADEevaluationonly.Asexpected, G2 and G4 (CRUSADE ≥ 41) had a higher rate of bleed-ing.Notwithstanding,theadditionoftheGRACEalgorithm madeitpossibletodifferentiatebleedingrisk profiles:G3 hadhigherbleedingratesthanG1(althoughbothhad CRU-SADE<41)andtheG4subgroup hadanincreasedbleeding hazardcomparedwithG2 (Table 3).Remarkably, patients inG4hadbleedingrisk thatwas3.5 timeshigherthanall remaininggroups.

Currently, ACS guidelines do not advise on tailoring medicaltreatment inNSTEMI, assumingthat allischaemic cases will derive similar and potential benefit from sev-eraltreatmentmodalities (i.e.anticoagulation andnewer antiplatelet regimens), unless contraindicated. Neverthe-less, this four-group approach may potentially alter a patient’s usual management regarding preload doses or other treatment options, to ensure that they gain most advantage from them. In the future, it might be suit-abletouse amoreconservativemanagement strategyfor patientswithlowischaemicburden(e.g.nopreloadingdose or weaker antiplatelet regimens [clopidogrel versus tica-grelor/prasugrel]). Nevertheless, as previously discussed, G2did notbehave likea low-risk group,showinga worse

ischaemic prognosis compared with G1 (although both

had GRACE<141), and could possibly benefit more from antithrombotic therapies and preloading. Moreover, con-cerning thegroups at highest ischaemic risk(G3 andG4), oneshould focuson the bleedingrisk. G4 had thehigher incidenceofbleedingeventsaswellastheworstlong-term prognosis.Becausebleedingisstronglyassociatedwith mor-talityand asbleedingevents occurearly in thecourse of myocardial infarction, it would be appropriate to refrain fromgivingpreloadingdosesorcombinedantiplatelet thera-piesorstrongeranticoagulationregimenstothesepatients. Presently, we do not fully understand the heterogeneous group of NSTEMI or its best management, and upcoming ACS recommendations need to address the possibility of patient-tailoredtherapy,regardingthespecificitiesofeach risk group and the aetiological nature of the myocardial infarction.

Inourview, thisstudy increasedcomprehensionof the ACS risk profiles, their characteristics and, more impor-tantly, their outcomes. They study supports the concept that ACS patients should be seen as a whole and that bleeding should be prevented intensely due to its huge prognosticimpact. Importantly,combinedassessmentwith GRACE+CRUSADEsignificantlyimprovedthediscriminatory powerof bothGRACEandCRUSADEwhenusedseparately. Withoutnewriskassessmenttools,thiscombinedand prac-tical approach might be a step forward in the future managementofNSTEMI.

Study

limitations

Thiswasasingle-centrecase-controlstudy,whichwas ret-rospectiveinnature,withasmall-to-moderatesamplesize. Becausepatientinclusionbeganin2006,thisstudysample maynotrepresentthestateoftheartinACSmanagement, asithaschangedimportantlyinthepastdecaderegarding medicaltreatmentandrevascularizationprocedures. More-over, the relatively small number of patients included in somegroups(e.g.G2,n=53)mayhavelimited interpreta-tionofresults.Nonetheless,itisourimpressionthatlarger groups, suchas in the case of G2, would ratherenhance thosedifferencesfound,suchasthoseconcerning haemor-rhagicevents.Anotherlimitationofthisstudywastheuseof riskmodelsthatsharedoverlappingriskvariables(e.g.heart rate,systolicbloodpressure,Killip-Kimbalclass,creatinine clearance[SupplementaryTable4]),which couldhaveled toaredundantclinicalassessment.

Although this study attempts to improve myocardial infarctionriskstratificationwithahighlypractical compo-nent, itcannotbe extrapolatedtoother populations.Our resultswarrantfurthervalidationinlargerandindependent cohorts before drawing any definite clinical applicability fromthesedata.

Conclusions

AcombinedriskstratificationstrategywithboththeGRACE and CRUSADE models enables a moreaccurate prediction of all-cause mortality and bleeding risk in patients with NSTEMI. The two scores complement each other in the prognosticationofthesepatients,potentiallyallowingmore accurate identification of patients who will benefit from more aggressive therapies and those who aresuited to a moreconservativeapproach.

Disclosure

of

interest

Theauthorsdeclarethattheyhavenoconflictsofinterest concerningthisarticle.

Acknowledgments

We thank Dr. Ziad Khoueiry, from the University Hospital of Montpellier, for his advice during the revision of the manuscript.Allauthorsmadeasubstantialcontributionto studyconceptionanddesignandtodataacquisition,analysis

and interpretation. The authors revised the article drafts andapprovedthefinalversion.

Appendix

A.

Supplementary

data

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/ 10.1016/j.acvd.2014.06.008.

References

[1]HammCW,BassandJP, AgewallS, etal. ESC Guidelinesfor the management of acute coronary syndromes in patients presentingwithoutpersistentST-segmentelevation:theTask Forceforthemanagementofacutecoronarysyndromes(ACS) inpatients presenting without persistent ST-segment eleva-tionoftheEuropeanSocietyofCardiology (ESC). EurHeart J2011;32:2999—3054.

[2]Wright RS, Anderson JL, Adams CD, et al. 2011 ACCF/AHA focused update of the guidelines for the management of patients with unstable angina/non-ST-elevation myocardial infarction (updating the 2007 Guideline): a report of the American Collegeof Cardiology Foundation/American Heart Association Task Force on Practice Guidelines. Circulation 2011;123:2022—60.

[3]FoxKA,EagleKA,GoreJM,StegPG,AndersonFA.TheGlobal Registry of Acute Coronary Events, 1999 to 2009—GRACE. Heart2010;96:1095—101.

[4]HoekstraJW,PollackJrCV,RoeMT,etal.Improvingthecare ofpatientswithnon-ST-elevationacutecoronarysyndromes intheemergency department:theCRUSADEinitiative.Acad EmergMed2002;9:1146—55.

[5]GharacholouSM,AlexanderKP,ChenAY,etal.Implicationsand reasonsforthelackofuseofreperfusiontherapyinpatients withST-segmentelevationmyocardialinfarction:findingsfrom theCRUSADEinitiative.AmHeartJ2010;159:757—63.

[6]KadakiaMB,DesaiNR,AlexanderKP,et al.Useof anticoag-ulantagents and risk of bleeding among patients admitted withmyocardialinfarction:a reportfromtheNCDRACTION

Registry—GWTG(NationalCardiovascularDataRegistryAcute Coronary Treatment and Intervention Outcomes Network Registry—get with the guidelines). JACC Cardiovasc Interv 2010;3:1166—77.

[7]SegevA, StraussBH, Tan M,Constance C,Langer A, Good-man SG. Predictors and 1-year outcome of major bleeding inpatientswithnon-ST-elevationacutecoronarysyndromes: insightsfromtheCanadianacutecoronarysyndromeregistries. AmHeartJ2005;150:690—4.

[8]ThygesenK,AlpertJS,JaffeAS,etal.Thirduniversaldefinition ofmyocardialinfarction.EurHeartJ2012;33:2551—67.

[9]LambEJ,TomsonCR,RoderickPJ.Estimatingkidneyfunction inadultsusingformulae.AnnClinBiochem2005;42:321—45.

[10]Mor-AviV,Lang RM,Badano LP,et al. Currentand evolving echocardiographic techniques for the quantitative evalua-tionofcardiacmechanics: ASE/EAEconsensusstatementon methodologyandindicationsendorsedbytheJapaneseSociety ofEchocardiography.EurJEchocardiogr2011;12:167—205.

[11]MehtaSR,GrangerCB,BodenWE,etal.Earlyversusdelayed invasiveinterventioninacutecoronarysyndromes.NEnglJ Med2009;360:2165—75.

[12]CookNR.Useandmisuseofthereceiveroperating character-isticcurveinriskprediction.Circulation2007;115:928—35.

[13]PencinaMJ,D’AgostinoRB,OverallC.asameasureof discrimi-nationinsurvivalanalysis:modelspecificpopulationvalueand confidenceintervalestimation.StatMed2004;23:2109—23.

[14]PencinaMJ,D’AgostinoSrRB,D’AgostinoJrRB,VasanRS. Eval-uatingtheaddedpredictiveabilityofanewmarker:fromarea undertheROCcurvetoreclassificationandbeyond.StatMed 2008;27:157—72[discussion207—12].

[15]FoxKA, Anderson JrFA, DabbousOH, etal. Interventionin acutecoronarysyndromes:dopatientsundergointervention onthebasisoftheirriskcharacteristics?TheGlobalRegistryof AcuteCoronaryEvents(GRACE).Heart2007;93:177—82.

[16]PuymiratE,TaldirG,AissaouiN,etal.Useofinvasivestrategy innon-ST-segmentelevationmyocardialinfarctionisamajor determinantofimprovedlong-termsurvival:FAST-MI(French RegistryofAcuteCoronarySyndrome).JACCCardiovascInterv 2012;5:893—902.

[17]Moscucci M, Fox KA, Cannon CP, et al. Predictors ofmajor bleedinginacutecoronarysyndromes:theGlobalRegistryof AcuteCoronaryEvents(GRACE).EurHeartJ2003;24:1815—23.