Geographic variation in inpatient costs for Acute Myocardial Infarction care: Insights from Italy

(1)

ContentslistsavailableatScienceDirect

Health

Policy

jo u rn al h om ep a g e :w w w . e l s e v i e r . c o m / l o c a t e / h e a l t h p o l

Geographic

variation

in

inpatient

costs

for

Acute

Myocardial

Infarction

care:

Insights

from

Italy

Chiara

Seghieri

a,_∗

_,

_Paolo

_Berta

b

_,

_Sabina

_Nuti

a

a_Institute_of_Management_and_Department_EMbeDS,_Scuola_Superiore_Sant’Anna,_Piazza_Martiri_della_Libertà_33,₅₆₁₂₇_Pisa,_Italy b_Department_of_Quantitative_Methods,_University_of_Milano_Bicocca_Via_Bicocca_degli_Arcimboldi_8,₂₀₁₂₆_Milano,_Italy

a

r

t

i

c

l

e

i

n

f

o

Articlehistory:

Received24December2017

Receivedinrevisedform18January2019 Accepted19January2019

Keywords: Geographicvariation Healthcareexpenditures Acutemyocardialinfarction Governancemodels Recycleprediction

a

b

s

t

r

a

c

t

Geographicvariationsinhealthcareexpenditureshavebeenwidelyreportedwithinandbetween

coun-tries.Nevertheless,empiricalevidenceontheroleoforganizationalfactorsandcaresystemsinexplaining

thesevariationsisstillneeded.Thispaperaimsatassessingtheregionaldifferencesinhospitalspending

forpatientshospitalizedforAcuteMyocardialInfarction(AMI)inTuscanyandLombardyregions(Italy),

whichrankhighintermsofcarequalityandthathavebeen,atleastuntil2016,characterizedbyquite

differentgovernancesystems.Generalizedlinearmodelsareperformedtoestimateindex,30-dayand

one-yearhospitalizationspendingadjustedforbaselinecovariates.Atwo-partmodelisusedto

esti-mate31–365dayexpenditure.AdjustedhospitalspendingforAMIpatientsweresignificantlyhigher

inLombardycomparedwithTuscany.InLombardy,patientsexperiencedhigherre-hospitalizationsin

the31–365daysandlongerlengthofstaysthaninTuscany.Ontheotherhand,nosignificantregional

differencesinadjustedmortalityratesatbothacuteandlongerphaseswerefound.

Comparingtworegionalhealthcaresystemswhichmainlydifferinboththereimbursementsystems

andthelevelofintegrationbetweenhospitalandcommunityservicesprovidesinsightsintofactors

potentiallycontributingtoregionalvariationsinspendingand,therefore,inareasforefficiency

improve-ment.

BY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

Thegeographicvariationinhealthcareacrossandwithin

coun-triesiswelldocumented,andatalllevelssuchasbetweendifferent

healthcareservicesandbetweenbothlargeandsmallareas[1,2].

Partofthisgeographicvariationisexpectedandoftenlinkedto

dif-ferencesinpopulationhealthandneeds(suchdemographicsand

healthstatus).However,someofthisvariationmaybeunwarranted

anddrivenbynon-clinicalfactors,suchasprofessionaldecision

making,theavailability anddistributionof resources,physician

incentives,financingandreimbursementmodels[2].Themain

pol-icychallengeisthusnottoeliminatevariationaltogetherbutto

identifyareasof variationwhere populationcharacteristicsand

healthoutcomesarecomparablebuthigher-than-averagecosts

suggestthepotentialforimprovedefficiency.Evidencefromthe

Dartmouthresearchgroupsuggestedthatevenafteradjustingfor

differencesinpopulationcharacteristicsandinpricelevels,

high-treatmentintensity regions weregenerally not associated with

∗Correspondingauthor.

E-mailaddress:[email protected](C.Seghieri).

betteroutcomesbutwithsignificantlyhigherspending.Theresults

showedthatifalltheseregionshadreducedcarevolumestothe

levelobservedinlow-spendingregionswithasimilarlevelofcare

quality, approximately30%in spendingcouldhave beensaved.

Higherefficiencycouldthusbeachievedbyidentifyingand

reduc-ingunwarrantedvariationwithoutproducingworseoutcomes,on

average,orreductionsinthequalityofcare[3–6].

Despitesimilarvariationswerecorroboratedbyalargebody

of literature in US [7,8] and more recently in other advanced

economies[9–13],thereisstilllittleagreementaboutthecauses

andtheclinicalimplicationsofthesereportedvariations.

Numer-ous possiblefactorsof varyingimpactcoulddrivevariationsin

spending,themajorityofthestudiesfoundthatthehighest

per-centage ofvariationsisduetosupply-sidefactorsincludingthe

typesofservicesprovided[8],paymentdifferentials[14]physician

behavior[15].However,otherstudiesshownthatcostvariations

werenotduetosystemicinefficienciesbutmostlytopopulation

healthstatus[16,17].

Additionally,mostfindingsinUSconfirmedthatgreaterhealth

carespendingdoesnotnecessarilycorrelatewithbettercareor

betterhealth[8,18–20].However,indistinctiontotheDartmouth

https://doi.org/10.1016/j.healthpol.2019.01.010

(2)

research,otherstudiesinUSshowedthatmorespendingand

uti-lizationarelikelytoleadtobetteroutcomes[21–25].

Otherrecent studies in Europealso reported mixed results.

Whilesomeresearcheshaveshownthatwithincountryvariations

inhealthexpendituresaremainlyexplainedbymedicalneed[26]

andthatexpendituresarelikelytobepositivelyrelatedtohealth

outcomes[27],otherstudies,althoughconfirmingsignificant

vari-ationsincostsandoutcomesacrossandwithincountriesandfor

differentconditions,didnotfoundanyclearcost-qualitytradeoff

pattern[28,29,13,30].

Regardless the magnitude and direction of the association

betweenhealthcarecostsandoutcomes,itseemsthatthereisa

largeproportionofvariationinspendingandutilizationamongand

withincountrieswhichmightreflectdifferencesininefficiencies.

Furtherresearchisthereforeneededtogainabetterunderstanding

oftheunderlyingdriversofvariationsandtoidentifytherelated

healthpolicyimplications.

Inthis context,thepresent studyaimsat analysingregional

variationinhospitalspendingforpatientswithAcuteMyocardial

Infarction(AMI) inItaly. Using specific econometricmodelsfor

healthcarecosts,wecompareone-yeartreatmentsand

Diagnostic-RelatedGroup(DRG)-basedexpendituresofpatientshospitalized

forAMIinLombardyandTuscanyregionswhichareboth

character-izedbysimilarhigh-performinghealthcaresystemsbut,however

differ in terms of institutional settings. In particular, Tuscany

healthcaresystemisbasedonastrongintegrationbetweenthe

var-ioushealthcareproviderstofavourtheefficientuseofbothacute

andpost-acuteservices,andonplanning,targetsettingand

com-petitivebenchmarkingonqualityofcarethatplayacrucialrolein

theregionalgovernance.Lombardyhasadopteduntil2016a

quasi-marketsystemmadeupofbothpublicandprivateprovidersand

basedonaprospectivepaymentsystemtoreimburse

hospitaliza-tions[31–34].ThedistinctdifferencesbetweenLombardy’squasi

marketgovernancemodelandtheintegratedcaremodelof

Tus-canyofferavaluableopportunitytoexploreregionalvariationsin

healthcarespending.

WefocusoncomparisonofAMIcareforseveralreasons

sup-portedbythescientificliterature[35,36],includingthefactthat

AMIis one of the most commonhospital admission; that it is

anacuteconditionwhichisnotrelatedtothepatients’choiceof

providers,patientsareinfactgenerallytakentothenearesthospital

forimmediatetreatment,and,althoughthereisagrowing

interna-tionalconsensusonwhichtreatmentsshouldbeadministeredto

AMIpatients,geographicvariationsinqualityandtreatment

inten-sityarewidelydocumented[13].Moreover,thehospitalizationfor

AMIisawell-definedindexevent,whichislikelytobesimilarly

definedacrossandwithincountries.

2. ThehealthcaresystemsofLombardyandTuscany

Regions

The Italian National Healthcare System (NHS) follows the

Beveridge model [37], providing universal healthcare coverage

throughouttheItalianStateasasinglepayer.Itentitlesallcitizens,

regardlessoftheirsocialstatus,toequalaccesstoessential

health-careservices.In1992,asystemreformtransferredadministrative

andorganizationalresponsibilitiesandtasksfromthecentral

gov-ernmenttotheadministrationsofthe21regionsinItaly.These

regionsnowhavesignificantautonomyontherevenuesideand

inorganizingservicesdesignedtomeettheneedsoftheir

respec-tivepopulations[38,34].Acrossyears,thedecentralizedstructure

oftheItalianNHShasledtotheemergenceofarangeof

differ-entgovernancemodelsattheregionallevel[33],whichofferrich

opportunitiesforcomparativestudyoftheimpactofdifferent

insti-Table1

DescriptionoftheTuscanyandLombardyhealthcaresystems.

TuscanyRegion LombardyRegion

No.Inhabitants(million) 3,7 10

Populationdensity(inh./km2₎ _163.24 _419,11

Lifeexpectancy(years) 85 85

No.ofpublichospitals 40 98

No.ofprivatehospitals 0 102

tutionalsettingsoncostsandoutcomes(i.e.onthevaluecreation

process[39]).

Amongthe20Italianregions,bothLombardyandTuscanyare

amongwealthiestregionsinItalyandtheirpopulationaresimilar

intermsofhealthconditionsandlifestyles.AsshowninTable1,

Tuscanycountsabout3,7inhabitantswhileLombardyabout10

mil-lions.Lifeexpectancyatbirthisof85yearsinbothregionsandthe

percentageofpeoplereportingofbeingingoodhealthconditions

are71.5%and71.2%inTuscanyandLombardyrespectively(Source:

http://dati.istat.it).Thetworegionsarebothcharacterizedbyhigh

qualityhealthcaresystems[40,41]butwithdifferentapproachesto

theorganisationofhealthcaredeliveryandtothefinancingmodel.

TheLombardyhealthcaresystemcomprisesapproximately200

hospitalsgenerating2milliondischargesannually;18billionEuros

aredevotedtohealthcareexpenditures(80%oftheregionalbudget)

everyyear.Theregionalhealthcaresystemwasreformedin1997

becomingaquasi-marketcharacterizedbyaseparationbetween

healthcarepurchasersandproviders,andwhereprivateproviders

deliverthe30%ofthehospitalizationsperyearandcompetewith

thepublicproviders.Patients,therefore,arefreetochoosewhere

to behospitalized whether in any privateor public accredited

providersAnessi-Pessinaetal.[31,33,42,43].

DifferentlyfromLombardy,thehealthcaresysteminTuscany

isbasedona strongintegrationbetweenthevarioushealthcare

providerstofavourtheefficientuseofbothacuteandpost-acute

services.Planning,targetsettingandcompetitivebenchmarking

playcrucialrolesintheregionalgovernance[40]ofTuscany.The

systemcomprisesabout40publicprovidersand34healthdistricts

that areincharge oforganize anddeliverservicesof territorial

healthnetworks,socialcareandsocialintegration.Startingfrom

2016,theRegionalHealthcareSystemhasbeenreformedby

aggre-gatingtheexisting12LocalHealthAuthorities(LHAs)intothree

majornewLocalHealthAuthoritieswhilekeepingtheexistingfour

TeachingHospitals.Morethan95%ofhospitalbedsarepublicand

eachoftheLHAsarefinancedbytheregionaladministrationunder

aglobalbudgetwithaweightedcapitationsystem,thatis,funds

areallocatedtoprovidersthroughanadministrativeprocessnot

directlylinkedtopatientvolumes.Ontheotherhand,budgetcost

controlandqualitycaretargetsareappliedtoteachinghospitals.

Eventhoughpatientshaveafreechoiceofhealthcareprovider,the

absenceofaclearlinkbetweenvolumesandreimbursementmeans

thathospitalsdonothavethesameincentivetocompeteformarket

sharethattheydoinDRG-basedfinancingsystems.Indeed,

hospi-talsinTuscany,aremorefocusedonoutcomeandappropriateness

targetsthataremonitoredthroughabenchmarkingapproachand

throughpublicdisclosureofproviders’performancelevels(http://

performance.sssup.it/netval).

WithregardtotheorganizationofcareforAMI,inItalythere

hasbeenarisinginAMIcarenetworkcoverageacrossthewhole

territory,withagrowingnumberofPercutaneousCoronary

Inter-vention (PCI)promptlydelivered [44]. Thisholds also for both

TuscanyandLombardyregions,inwhichrevascularizationservices

arepromptlyaccessedbytheregionalpopulationsexceptforfew

municipalitieswhicharemorethan60mindistantfromthe

near-estcarenetwork[44].Moreover,inthelastyears,thehealthcare

(3)

inte-gratednetworkofcardiachealthservices,followingthelogicof

Hub&Spokemodel[45],whereasaccesstotreatmentofAMIin

Lombardyisassuredthroughthecoordinationamonghospitalsand

emergencyservices,especiallyforthetreatmentofSTEMIpatients

[46,47].

3. Methods

3.1. Datasources

Weconductedaretrospectivecohortstudyusingdatafrom

hos-pitaldischargerecordsfromTuscanyandLombardy.Theanalysis

includedallpatientsdischargedwithaprimarydiagnosisofAMI

(InternationalClassificationofDeceases,9threvision,Clinical

Mod-ification-ICD-9-CMcodes:410.xx)betweenJanuary1,2012,and

December31,2013,fromallhospitalsinTuscanyandLombardy.

Admissionslastinglessthantwodayswereexcluded,aswell

ashospitalizationsofpatientsnotresidentinthetworegions.The

analysisconcernedpatientsolderthan18yearsandyoungerthan

100years.Recordswithadiagnosticcode410.9AMIofunspecified

site,andadiagnosticcode410.×2subsequentepisodeofcarefor

dischargedpatientswerealsoexcludedfromtheanalysis.

IfapatienthadmorethanoneadmissionforAMIduringany

studyyear,werandomlyselectedoneofthoseadmissionsforeach

studyyear[48,49]. StartingwiththeAMIhospitalization (index

admission),eachpatientwasthenfollowedforoneyeartomeasure

variationsinhospitalizationvolumesandspendingacrossTuscany

andLombardy,bothattheinitialAMIepisodeandduringthe

post-hospitalizationperiod.

The study observationwas divided into fourperiods: index

admission(1), allhospitalizationsupto30daysfromtheindex

admission(0–30)(2),fromdays31to365(3),andoverall1-year

hospitalizations(4).The0–30dayswindowcanbeconsideredas

theacutephaseofAMImanagementwhenpatientsshouldreceive

mostoftheguideline-basedadmissiontherapies(i.e.acute

reperfu-siontherapy),andre-hospitalizationsinthisperiodmaybestrongly

linkedtotheadmissionepisodeof care.Ontheotherhand,

re-hospitalizationoncetheacutephasehaspassed(i.e.fromday31

to365)shouldalsodependontheprimarycareandcommunity

supportprovidedtopatients.

Foreachpatient,hospitalizationcostsatindexandpost-index

periodswereidentifiedusingtheDRGreimbursementregistered

intheHDRs.Inordertotakeintoaccountoftheinter-regional

differencesinspendingassociatedwithdifferencesinDRGtariffs,

expenditureswerecalculatedfortheAMIpatientsofeachregion

byapplyingtheLombardyDRGtariffsystemtothe

correspond-ingDRGsinTuscany.Data hadalreadybeenanonymizedatthe

RegionalHealthInformationSystemOffice,whereeachpatientwas

assignedauniqueidentifier.Thisidentifierpreventsthepatient’s

identityandothersensitivedatafrombeingtraced.Thestudywas

carriedoutincompliancewithItalianlawonprivacy,andapproval

byanEthicsCommitteewasnotrequired.

3.2. Methodologicalapproach

The major methodological issues that arise when analysing

healthcarecostsarethatthesenon-negativedataoftenexhibit

sub-stantialpositiveskewness,“heavy”uppertailandamassatzero

fornon-users, asinourstudy,forthosepatientswhowerenot

hospitalizedinthe31–365observationperiod[50–52].

Theskewnessofthecostdistributionswasapproachedusing

GeneralizedLinearModels(GLMs).Inparticular,toestimate

hospi-talizationspendingatthepatient’sindexadmission,at30days,and

at1yearfollowup;threedifferentGMLsweretested,specifying

severalfamilies(Gamma,Poisson,InverseGaussian)andalinklog

function.ThemodifiedParktest[53]andHosmer-Lemeshowtest

[54]wereappliedtochoosethespecificationwiththebettermodel

fitting. For the three dependent variables here considered, the

testsindicatedthattheGamma-Loglinkmodelwasadequate

com-paredtoothertypesofdistributions.Differently,atwo-partmodel

[50,55,56]wasusedtoestimatehospitalization costsat31–365

follow-upperiod(conditionalonsurvivingthefirst30days).This

allowstocontrolforthezero-inflationaffectingthehospitalization

costsinthisobservationperiod.Two-partmodeliscomposedby

tworelatedequation:inthefirstpartofthemodel(Eq.(2)inthis

paper)theprobabilitytoobserveapatientintheperiod31–365

daysismodelledviaprobitregression.Inthesecondpartofthe

model(Eq.(3)inthispaper),thelevelofhospitalexpenditureforthe

patientshospitalizedintheperiod31–365daysismodelledusinga

GLMwithGamma-Loglinkspecification.Allconsideredmodelsare

adjustedforage,sex,ST-segmentAMIandselectedcomorbiditiesat

baselinewhichwereretrieved,foreachpatient,atboththeindex

andinallhospitalizationsoneyearpriortotheindex.Adummy

variableindicatingthepatient’sregionofresidence(1=Lombardy,

0=Tuscany)wasalsoincludedasacovariateinthemodels,inorder

tomeasurethedifferencesbetweenthetworegions.

Thiseconometricapproachcanbeformulatedasfollow:

Model1:Dependentvariables:costatIndex,0–30dayspost

AMI,andat1year

E(Cosi|X)=exp(X/!)

Wherei=1,....,naretheLombardyandTuscanypatients,Xisthe

matrixofthecovariatesand!_is_the_vector_of_the_{coefficients.}_Using

similarnotation,Model2forthedependentvariableexpressing

costat31–365dayspostAMIcanbeformulatedasfollow:

E(CostiX) =Pr (Costi>0X) *E (Costi>0,X) (2)

Whichiscomposedby:

Pr (Costi>0X) = exp(Xϕ) 1+exp(Xϕ) (3) E(Costi>0,X) =exp

!

X#

"

(4)

Asanticipated,differentlyfromModel1,whenweestimatethe

costfor31–365days,weneedtocontrolfortheprobabilitytobe

aliveafter30days.Forthisreason,thetwo-partmodelapproachas

describedbyEq.(2)isadopted,whereEq.(3)estimatesthe

prob-abilitytobealiveat30daysandEq.(4)estimatestheexpenditure

adjustedfortheprobabilitytobealiveat30days.

DifferencesinadjustedcostsbetweenthetwoItalianregions

wereestimatedusingrecycledpredictions.Thismethodavoidsthe

problemofthecovariatesimbalanceaffectingnonlinear

retrans-formationsinGLM[56],andconsistsin1000bootstrapreplications

usedtoestimate95%-bootstrap-percentile-intervals[56,57]ofthe

differencebetweentheexpenditureinTuscanyandLombardy.

Moreover, although the study wasnot designed to test the

impactofspendingonhealthoutcomes,wecomparedmortality

ratesbetweenthetwo regionsafteradjustingfor age,sex,race,

ST-segmentAMI,andcomorbiditiesatbaselinebyusingalogistic

regression.Theavailabledatadidnotallowustoanalyse

possi-bledifferencesinnon-clinicaloutcomessuchaspatients’reported

outcomesorsatisfaction.

Finally,asensitiveanalysiswasperformedaimedatchecking

for potentialselection biasdue tothemainconfounders andto

potentialdifferencesincodingpracticesbetweenthetworegions,

we replicatedtheanalysisbyfirst matchingthepatientsofthe

tworegionsonage,sexandcomorbiditiesusingaCoarsenedExact

Matching(CEM)non-parametricapproach[32,58,59].Theglobal

balancebetweenthepatientsbelongingtothetworegionswas

(4)

Table2

SampleCharacteristicsCareProvidedtopatientsadmittedforAcuteMyocardialInfarction.

Patientscharacteristics Lombardy(N=30,225) Tuscany(N=13,187) P-value

Mean(SD) Mean(SD)

Age 70.325 (13.404) 71.940 (13.329) <0.001

Male 0.647 (0.478) 0.637 (0.4808) 0.068

Lipidmetabolismdisorder 0.093 (0.290) 0.212 (0.408) <0.001

Cancer 0.031 (0.174) 0.036 (0.185) 0.027

Diabetes 0.132 (0.338) 0.212 (0.408) <0.001

Hypertensivediseases 0.209 (0.406) 0.411 (0.4925) <0.001

Otherformsofischemicheartdiseases 0.248 (0.431) 0.350 (0.477) <0.001

PreviousHeartfailure 0.173 (0.3773) 0.268 (0.442) <0.001

Conductiondisordersandcardiacdysrhythmias 0.182 (0.3857) 0.209 (0.406)

Cerebrovasculardiseases 0.054 (0.225 0.073 (0.260) <0.001

Vasculardiseases 0.039 (0.192) 0.060 (0.237) <0.001

Chronicobstructivepulmonarydisease 0.040 (0.195) 0.066 (0.249) <0.001

Chronicnephropathies 0.070 (0.254) 0.116 (0.320) <0.001

N-STEMI 0.513 (0.499) 0.557 (0.496) <0.001

Treatmentsatindexadmission

PCI 0.624 (0.484) 0.594 (0.491) <0.001

CABG 0.019 (0.135) 0.009 (0.093) <0.001

Patientoutcomesa

30daymortality 0.077 (0.001) 0.075 (0.002) 0.753

1yearmortality 0.167 (0.001) 0.157 (0.002) 0.055

PatientspendinginEuros(D)b

Index 7,324 (4,336) 6,610 (3,204) <0.001

30day 9,033 (6.032) 8,037 (5,227) <0.001

31-365dayc _3,668 _(7,236) _3,044 _(6,537) _<0.001

1year 12,419 (9,908) 10,854 (8,682) <0.001

a_Represents_{risk-adjusted}_percentage_(standard_error).

b_Spending_is_adjusted_for_regional_differences_in_DRG-tariff_systems. c _Conditional_on_surviving_the_first₃₀_days.

4. Results

Weidentified13,187patientsinTuscanyand30,225in

Lom-bardy who were hospitalized for AMI in the years 2012 and

2013.AsshowninTable2,Tuscanpatientsweresignificantlyolder

andshowedmorecomorbiditiesonaveragethantheLombardy

patients(p<0.001).Theseresultscouldexplainthehigher

signifi-cantpercentageofNSTEMI(non-STsegmentelevationmyocardial

infarction)patientsinTuscanycomparedtoLombardy,with56%

and 51% respectively (p<0.001). Otherstudies have foundthat

patientshospitalizedforSTEMI(STsegmentelevationmyocardial

infarction)weremorelikelytobeyounger,male,andwereless

likelytohaveapriorhistoryofseveralcomorbidities[60].

Additionally, Lombardypatientsweremore likely toreceive

invasivetreatmentssuchasPCIorCoronaryArteryBypassGraft

(CABG) at theindex admission(p<0.001). Aftercontrolling for

patientcharacteristics,adjustedmortalityratesattheacutephase

(within30daysoftheAMI)wassimilarinthetworegions(7.7%

and7.5%inLombardyandTuscanyrespectively)whereas,inthe

longer-term(oneyear),themortalityofAMIpatientsinLombardy

wasaboutonepercentagepointshigherthaninTuscany(adjusted

1yearmortalityrates,16.7%inLombardyand15.7%inTuscany)

However,theadjustedratesatbothobservationperiodsdidnot

differsignificantlyacrossthetworegions.

Unadjusted mean hospital spending, standardized for the

regionaldifferencesintheDRGtariffsystems,weresignificantly

higherinLombardycomparedtoTuscanyacrossalltheobservation

periods.

Morespecifically,totalone-yearspendingperpatientwas14%

higherinLombardythanTuscany(p<0.001).Theregionalgapin

spending waslower in theacutephase of AMIwithLombardy

exceedingTuscanyby11%and12% (p<0.001) attheindexand

within30daysrespectively.Ontheotherhand,thelongerterm

post-admissionperiod hadthegreatestinterregionalgapinthe

spending.Thatis,between31and365days,spendingfor

hospi-talreadmissionwasonaverage20%higherinLombardycompared

withTuscany(p<0.001).Totalexpendituresinthispost-acutecare

phaseaccountedforrespectively27%and26%ofthetotal1year

spendinginLombardyandTuscany.

Atotalof41%oftheLombardypatientswerereadmittedatleast

onceinthe31–365daysaftertheindexadmissioncomparedwith

37%oftheTuscanypatients.Inbothregions,thehighestvolumes

ofhospitalizationswereforcardiovascularinterventions,although

therateofrehospitalizationforrevascularizations(eitherCABGor

PCI)washigherinLombardycomparedtoTuscany(Fig.1).Indeed,

almost9%and2%ofthereadmissionsinLombardywerefor

per-cutaneouscardiovascularprocedureswithadrug-elutingstentand

barestentrespectively,comparedto5%and1%inTuscany,whereas

1%ofthereadmissionswereforCABGinLombardycomparedto

0.5%inTuscany

Table3describestheregionaltrendsintheLengthofStay(LOS)

togetherwiththepercentageofhospitalizationwithLOS

exceed-inganestablishedthreshold(long-stayoutliers).TheaverageLOS

attheindexforLombardywastwodayslongerthaninTuscany,

whereasinthepost-acutecare(31–365dayspostadmission),the

averagelengthofstaywasabout10and7daysforLombardyand

Tuscany,respectively.BesideshavingonaveragelongerLOS

com-paredwithpatientsinTuscany(p<0.001),patientsinLombardy

werealsolikelytoexperiencegreaterlengthsofstaythantheDRG

specificLOSthresholdbothatindexandinthe1-yearfollow-up

(p<0.001).Indeed,theproportionsofhospitalizationsclassifiedas

long-stayoutliersinLombardywerealmosttwicethoseof

Tus-cany,thusleadingtohigherhospitalspendingforthesameDRGin

LombardycomparedtoTuscany.

Theresultspresentedabovewereconfirmedbyfindingsfrom

themultipleregressionofhospitalspendingforAMI(Fig.2),which

showedthatalsoafteradjustingforcomorbiditiesandfor

differ-encesinDRGtariffs,adjustedhospitalspendingwassignificantly

higherinLombardythanTuscanyacrossallobservationalperiods.

(5)

Fig.1. Re-hospitalizationinthe31–365daysfollowingtheindexadmissionperpatientwithAcuteMyocardialInfarction(AMI)forthemostcommonsurgicalinterventions.

Table3

Analysisofthelengthofstay.

Lombardy Tuscany P-value

Index

AverageLengthofStay(SD) 8.987 (6.979) 6.898 (5.239) <0.001

%hospitalizationoverDRGthreshold(SD) 20.206% (40.154) 9.559% (29.404)

Upto30daysaftertheindexadmission

31to365daysaftertheindexadmission

1yearaftertheindexadmission

(6)

patienthospitalizedinLombardywasonaverage2081euroshigher

thanthatforAMIpatientshospitalizedinTuscany(p<0.001).On

theotherhand,themeandifferencesbetweenLombardyand

Tus-canywere774and1197eurosforthehospitalizationcostsatindex

admissionandinthefirst30days(p<0.001)respectively.

More-over,inthetwo-partmodel,comparedwithTuscany,Lombardy

incurred onaverage 941 eurosextra (p<0.001) in

hospitaliza-tionexpenditureinthelonger-term(31–365days)post-admission

period.

ResultsoftheCEManalysisreportedalowerpost-matchingL1

valuewhichdecreasesfrom0.51to0.28indicatingahigh-quality

match(astrongreductionintheunbalancingofthecovariates).

AveragehospitalspendingofthematchedsamplesofAMIpatients

ofLombardyand Tuscanywerefoundtobesignificantly

differ-ent.Thismeansthatevenafterremovingpotentialselectionbias

duetodifferentclinicalanddemographiccharacteristicsofthetwo

AMIcohorts,hospitalspendingremainssignificantlyhigherin

Lom-bardyregioncomparedtoTuscanyacrossallthestudyobservation

periods(p<0.001).

Aftermatchingthetwocohorts,significantregionaldifferences

werealsoobservedforbothLOSandvolumeofrevascularizations

(eitherCABGorPCI)inthe31–365observation,whichremained

higherforLombardypatients(p<0.001)comparedtoTuscanyones.

Onthecontrary,wedidnotobserveanydifferenceintermsof

mor-talitybetweenTuscanyandLombardyinalltheobservationperiods

(p>0.10)

5. Discussion

Theliteratureongeographicvariationhaswidelyprovedthat

thereareavarietyoffactorsthatcontributetoexplainregional

variationsinhealthcarespending andutilization.Someofthese

factorssuchasthedifferentpatients’needsandpreferencesare

“warranted”,whereas,theultimatechallengeforpolicymakersis

reducingregionalsvariationsdueto“unwarranted”factorswhich

ultimatelyreflectsysteminefficiencies.

Inthisstudy,weanalysetheextentofregionalvariationin

hos-pitalspendingforAMIpatientsacrosstwoItalianregions,Tuscany

andLombardy,bothwithhighqualityhealthcaresystemsbutwith

differentregionalgovernancemodelsforfinancinganddelivering

healthcare.

Inlinewithpreviousstudiesreportingthatdifferencesinprices

andpatienthealthstatusaccountforonlyapartofthevariationin

spending[6,17,61],wefoundthat,comparedwithTuscany,

Lom-bardypresented significantly higheradjusted spending for AMI

patientsatboththeacutephase(within30daysoftheAMI

hos-pitalization)andinthelonger-term(31–365days)post-admission

periods.Morespecifically,afteradjustingfordifferencesinpatients’

case-mixand regionalDRG tariffs,the averageannual hospital

costperAMIpatientstillsignificantlydifferedbetweenthetwo

regionsanditwasonaverage2081euroshigherinLombardy

com-paredtoTuscany.Ontheotherhand,aftercontrollingforpatients’

characteristics, themortality rates at acute (within 30 days of

theAMI)andlonger-termpostadmissionperiod(1year)ofthe

AMIdidnotsignificantlydifferbetweenthetworegions.Results

werealsoreinforcedbytheCEManalysiswhichshowsthatthein

matchedsamplesLombardyreportedsignificantlyhigherAMI

hos-pitalspendingcomparedtoTuscanywhereasthetworegionsdid

notsignificantlydifferintermsofAMImortalityacrossthestudy

periodsBeyondtheseobserveddifferencesinregionalspending,

thestudyattemptstocontributetothediscussiononthe

poten-tialdriversatsystemlevelofvariationsinhospitalspendingacross

regionalhealthcaresystemswithcomparablehealthoutcomesand

populations.

WefoundthattheaverageLOSwashigherinLombardy

com-paredtoTuscany,withapercentagedifferencebetweenLombardy

andTuscanyrangingbetween30%attheindexat36%at1year

follow-up.Moreover,theproportionofhospitalizationswithlonger

thanastatedLOSoutlierthresholdwasalmostdoubleinLombardy

withrespecttoTuscanyacrossallthestudyperiodsandthiswas

truealsointhematchedcohorts.

Lombardy’shigherpercentageoflong-stayoutliers,especially

in the post-acute phase, could, for instance, suggest that this

regionalhealthcaresystemmightlackeffectivetransitional

pro-gramsbetweentheacutehospitalsectorandcommunityservices

fortheAMIpatients,especiallyforthosewhoaremoreinneedof

coordinateddischargeandongoingsupport[62].

Furthermore,between31and365days,AMIpatientsin

Lom-bardyweremorelikelytobereadmittedcomparedtoTuscany.In

particular,higherratesofrepeatedPCIwereobservedinthe31–365

daysfortheLombardypatientswithrespecttotheTuscanyones

inboththematchedandnotmatchedcohorts.Intheliterature,

activity-relatedfundingwasfoundtoprovideincentivesforgreater

volumes[63,64]andthereforetopotentiallyleadtodifferentlocal

practicepatterns.Indeed,inthis context,wemighthypothesize

adifferentprofessionals’choiceonrevascularizationmodalityin

patientswithmulti-vesseldisease.PCIMulti-vesselcaninfactbe

performedeitheratthetimeoftheindexhospitalizationorat

sepa-ratesubsequenthospitalizations.Althoughtheidentificationofthe

optimaltimingofastagedprocedureinAMIpatientsstillrequired

furtherinvestigation,clearlythestrategyofproceduresperformed

inrepeatedhospitalizationsleadstoincreasedhospitalcostswith

respecttotreatothervesselsatonestage[65,66].

Variationsinproceduresratesandintheuseofresources(e.g.

differentLOSforsimilarpatients)especiallyinthepost-acutephase

mightbealsoexplainedbyregionaldifferencesinother

supply-sidefactorssuchasprovider’scapacity(e.g.hospitalandlong-term

carebedsperthousand,numberofcardiacfacilities)[67,3,68].As

anexample,cardiachospitalbedsin2013were2.29per100,000

inhabitantsinLombardyregioncomparedwith1.53per100,000

inhabitantsinTuscany(Source:Ministryofhealth-http://www.

dati.salute.gov.it/dati/homeDataset.jsp).Theavailabledatadidnot

allowustoanalysedifferencesinout-of-hospitalresourcecapacity.

Indeed, these results seem to confirm that, even between

regionalhealthcaresystemsdominatedbyhighstandardsofcare,

there can exist both different type of organizational structures

withinandbetweencaresettingsanddifferentlevelsof

profes-sionals’discretionoverthecareprovidedforsimilarpatients.

Wethereforehypothesizethatsystem-levelfactors,suchasthe

differentreimbursementsystems,andthedifferentorganization

andlevelofintegrationbetweenacutehospitalprovidersand

com-munitycareservicesinthetworegions,mightcontributetoexplain

thesupply-sideofvariationinexpenditureand,moregenerally,the

valueformoneythatisproduced.

Targetpolicieswhichfavouroutcomegoalsandfacilitate

con-tinuityofcarebetweenhospitalandpost-hospitalizationsettings

canincentivizehospitalstowardsamoreefficientuseoftheirinput

resources (i.e.hospital beds), allowto efficiently reallocate the

resourceswhereservicesmayprovidemorevalueforpatientsand

toguaranteeinganappropriatesupporttopatientsfromonehealth

caresettingtoanother[69].

TheperformanceevaluationsystemadoptedinTuscany[40,70]

allowsphysicianstobemoreawareoftheimpactoftheirchoices

ontheuseofresourcesthankstotheanalysisofadministrative

datathroughthepatients’clinicalpathperspective.Additionally,

benchmarkingandpublicdisclosureofoutcomeresultsand the

participationofphysiciansinmeeting,stronglyenhancedbythe

regionalgovernment,aimedatidentifyingandsharingbest

prac-ticehelpingTuscanclinicianstouniformbehaviorsatregionallevel

puttingreputationtowork[71].

Asmatter of fact,in 2016 theLombardyRegion hasstarted

(7)

overcom-ingthequasimarketgovernancesysteminfavouroftheintegration

betweenhospital and primary care, reducing readmissionsand

increasingtheappropriateness.Moreover,Lombardydecidedtobe

partofanetworkof10ItalianRegionsthatadoptaunique

per-formanceevaluationsystem,basedontargetsetting,competitive

benchmarking,andpublicdisclosureofdata[75].Thisisaimedat

improvingqualitydeliveredandatcontainingtheexpenditure.

Thestudysuffersfromsomerelevantlimitationsmanyofwhich

related to the data and design of the study. First, as with all

observationaldata,weareunabletomakeinferencesabout

causa-tion,therefore,effectestimatesshouldbecautiouslyinterpreted.

Second,aswithotherstudiesusingadministrativedata[72,73],

informationwaslackingonpotentialconfoundingvariablessuch

asAMIseverity,lifestylefactors,treatmentadherence,and

proce-duralcharacteristics.Moreover,thetypeofdatadonotallowto

carefullyevaluatetheappropriatenessofstrategiesbothadopted

inthetworegionsfortheAMItreatmentbothattheindexandinthe

31–365hospitalizations.However,theuseofadministrativedata

hasbeenwidelyadoptedbyhealthcareagenciesandother

stake-holderstomeasurehospitalperformance.Thisisbecausesuchdata,

comparedtoclinicalregistries,areeasilyaccessible,relatively

inex-pensivetouse,andenableinformationtobecollectedontheentire

populationofconcern[74].

Othervariablesthat areimportanttothecomparisonsbeing

mademighthavebeingomittedorunmeasuredinthisstudysuchas

informationonAMIpatientsthatmaydiebeforereachingthe

hos-pital,whichcouldprovideindicationonwhetherhospitalpatient

populationsdifferacrossthetworegions.

Finally,atthisstageoftheanalysis,wefocusedourstudyonthe

treatmentofAMIonlyandwerecognizedthatfindingsmightnot

extendtotherestofhealthcare.Also,wewereonlyableto

con-siderhospitalizationspendingandcouldnotincludenon-clinical

outcomes.

Despitetheselimitations,ourstudyhasseveralstrengths.Tothe

bestofourknowledge,thisisthefirststudythathasanalysed

geo-graphicaldifferencesinhealthcarespendingbetweentworegional

healthcaresystemsinItaly,inalongertimeframethanthe

arbi-trary30daysoftheacutecarephase.Inaddition,comparedtothe

otherstudies,theopportunitytocomparetheregionalhealthcare

systemsinItalyallowsustoanalysegeographicvariationbetween

tworegionsthataresimilarintermsofbothsocio-economicfigures

andcitizens’standardoflivingbutdifferintermsofthegovernance

andfundingmodelsoftheirhealthcaresystems.Thismight

facili-tatetheidentificationofsupply-sidefactorsthatcouldpotentially

explainsysteminefficiencies.

6. Conclusions

Health systems arelooking forways toshifttovalue-based

healthcare.Ourstudysuggeststheimportanceofconsideringthe

impactofthegovernancesystemasoneofthepossible

determi-nantsofgeographicvariationwhencomparingdifferentregional

healthcaresystems.Althoughthedifferentregionalorganizational

modelsmaynotleadalonetoreductionsinunwarrantedvariations

inspending,thechoicetointroducequasi-marketmechanismsto

promote bothefficiencyand efficacyseemstobelessfavorable

thentheintroductionofgovernancebasedonplanning,target

set-tingandcompetitivebenchmarkingonqualityofcare,moreoverif

partneredwithastrongintegrationbetweenacuteandpost-acute

services.

Conflictofinterest

Theauthorsdonothaveanyconflictsofinteresttodeclare

Acknowledgements

Wewouldliketothanktheregionalandlocalrepresentatives

ofbothregionsandalltheresearchersandcolleagues(inparticular

fromtheMeS-LabandtheDartmouthInstitute)fortheirvaluable

support.WewouldalsoliketothanktheScientificCommittee,Prof.

VeronicaGrembiandalltheparticipantsofthe5thHealth

Econo-metricsWorkshop-Bari(Italy)andTheWennbergInternational

Collaborativecommunityfortheircommentsandsuggestions.

References

[1]Paul-ShaheenP,ClarkJD,WilliamsD.Smallareaanalysis:areviewand analy-sisofthenorthAmericanliterature.JournalofHealthPolitics,PolicyandLaw 1987;12(4):741–809.

[2]Wennberg JE. Time to tackle unwarranted variations in practice. BMJ 2011;342:d1513.

[3]FisherES,WennbergDE,StukelTA,GottliebDJ,LucasFL,PinderEL.The impli-cationsofregionalvariationsinmedicarespending.Part1:thecontent,quality, andaccessibilityofcare.AnnalsofInternalMedicine2003;138(4):273–87. [4]FisherES,WennbergDE,StukelTA,GottliebDJ,LucasFL,PinderEL.The

impli-cationsofregionalvariationsinmedicarespending.Part2:healthoutcomes andsatisfactionwithcare.AnnalsofInternalMedicine2003;138(4):288–98. [5]SkinnerJ,FisherE.Reflectionsongeographicvariationsinushealthcare.

Dart-mouthInstituteforHealthPolicyandClinicalPractice2013;12(May). [6]WennbergJE,FisherES,SkinnerJS.Geographyandthedebateovermedicare

reform.HealthAffairs2002:W96–114.Jul-Dec;SupplWebExclusives. [7]MedicarePaymentAdvisoryCommission(MedPAC),FebruaryRegional

varia-tioninMedicarePartA,PartB,andPartDspendingandserviceuse;2017. [8]InstituteofMedicine.Interimreportofthecommitteeongeographic

varia-tioninhealthcarespendingandpromotionofhigh-valuecare:preliminary committeeobservations.Washington,DC:TheNationalAcademiesPress;2013. [9]CoralloAN,CroxfordR,GoodmanDC,BryanEL,SrivastavaD,StukelTA.A sys-tematicreviewofmedicalpracticevariationinOECDcountries.HealthPolicy 2014;114(1):5–14.

[10]McPhersonK,GonG,ScottM,OECDHealthWorkingPapersNo.61International variationsinaselectednumberofsurgicalprocedures.Paris:OECDPublishing; 2013.

[11]OECD.Geographicvariationsinhealthcare:whatdoweknowandwhatcan Bedonetoimprovehealthsystemperformance?OECDPublishing;2014. [12]Wennberg International Consortium (WIC) (2016). http://

wennbergcollaborative.org/index.php.

[13]HeijinkRichard,EngelfrietPeter,RehnbergClas,KittelsenSverre,Hakkinen Unto.Awindowongeographicvariationinhealthcare:insightsfrom Euro-HOPE.HealthEconomics2015;24(Suppl.2):164–77,http://dx.doi.org/10.1002/ hec.3287.

[14]Medicare Payment Advisory Commission (MedPAC), January Report to congress:regionalvariationinmedicareservicesuse;2011.

[15]CutlerD,SkinnerJ,SternAD,WennbergD.Physicianbeliefsandpatient prefer-ences:anewlookatregionalvariationinhealthcarespending(No.w19320). NationalBureauofEconomicResearch;2013.

[16]ZuckermanS,WaidmannT,BerensonR, etal. Clarifyingsourcesof geo-graphicdifferencesinMedicarespending.NewEnglandJournalofMedicine 2010;363(July(1)):54–62.

[17]ReschovskyJD,HadleyJ,RomanoPS.Geographicvariationinfee-for-service medicarebeneficiaries’medicalcostsislargelyexplainedbydiseaseburden. MedCareResRev2013;70(5):542–63.

[18]YasaitisL,FisherES,SkinnerJS,ChandraA.Hospitalqualityandintensityof spending:isthereanassociation?HealthAffairs(Millwood)2009;28:w566–72. [19]SirovichBE,GottliebDJ,WelchHG,FisherES.Regionalvariationsinhealth careintensityandphysicianperceptionsofqualityofcare.AnnalsofInternal Medicine2006;144:641–9.

[20]BaickerK,ChandraA.Medicarespending,thephysicianworkforce,and ben-eficiaries’qualityofcare.HealthAffairsWebExclusives2004(January–June): W4-184–W4-197.

[21]OngMK,MangioneCM,RomanoPS,ZhouQ,AuerbachAD,ChunA,etal. Look-ingforward,lookingback:assessingvariationsinhospitalresourceuseand outcomesforelderlypatientswithheartfailure.Circulation:Cardiovascular QualityandOutcomes2009;2:548–57.

[22]BarnatoAE,ChangCCH,FarrellMH,LaveJR,RobertsMS,AngusDC.Issurvival betterathospitalswithhigher“end-of-life”treatmentintensity?MedicalCare 2010;48(2):125.

[23]RomleyJA,JenaAB,GoldmanDP.Hospitalspendingandinpatientmortality: evidencefromCalifornia:anobservationalstudy.AnnalsofInternalMedicine 2011;154:160–7.

[24]SilberJH,KaestnerR,Even-ShoshanO,WangY,BresslerLJ.Aggressive treat-mentstyleandsurgicaloutcomes.HealthServicesResearch2010;25:1872–92. [25]DoyleJrJJ,GravesJA,GruberJ,KleinerSA.Measuringreturnstohospital care:evidencefromambulancereferralpatterns.JournalofPoliticalEconomy 2015;123(1):170–214.

[26]GöpffarthD,KopetschT,SchmitzH.Determinantsofregionalvariationin healthexpendituresinGermany.HealthEconomics2016;25(7):801–15.

(8)

[27]SchreyöggJ,StargardtT,TiemannO.Costsandqualityofhospitalsindifferent healthcaresystems:amulti-levelapproachwithpropensityscorematching. HealthEconomics2011;20(1):85–100.Jan.

[28]GutackerN,BojkeC,DaidoneS,DevlinNJ,ParkinD,StreetA.Truly ineffi-cientorprovidingbetterqualityofcare?Analysingtherelationshipbetween risk-adjustedhospitalcostsandpatients’healthoutcomes.HealthEconomics 2013;22(Aug(8)):931–47.

[29]HäkkinenU,RosenqvistG,PeltolaaM,KapiainenS,RättöH,CotsF,etal. Qual-ity,cost,andtheirtrade-offintreatingAMIandStrokepatientsinEuropean hospitals.HealthPolicy2014.

[30]KittelsenSAC,AnthunKS,GoudeF,HuitfeldtIMS,H ¨akkinenU,KruseM,etal. CostsandqualityatthehospitallevelinNordiccountries.HealthEconomics 2015;24(Suppl.2):140–63.

[31]BrennaE.Quasi-marketandcost-containmentinBeveridgesystems:the Lom-bardymodelofItaly.HealthPolicy2011;103(2):209–18.

[32]BertaP,BossiM,VerzilloS.%Cem:asasmacrotoperformcoarsenedexact matching.JournalofStatisticalComputationandSimulation2017;87(2):1–12. [33]BertaP,SeghieriC,VittadiniG.Comparinghealthoutcomesamong hospi-tals:theexperienceoftheLombardyRegion.HealthCareManagementScience 2013;16(3):245–57.

[34]NutiS,VolaF,BoniniA,VainieriM.Makinggovernanceworkinthehealthcare sector:evidencefroma‘naturalexperiment’inItaly.HealthEconomics,Policy andLaw2016;11(01):17–38.

[35]Likosky DS, Zhou W,Malenka DJ, Borden WB, Nallamothu BK, Skinner JS. Growthin medicareexpenditures forpatients with acutemyocardial infarction: acomparisonof1998through1999 and2008.JAMAInternal Medicine 2013;173(22):2055–61,http://dx.doi.org/10.1001/jamainternmed. 2013.10789.Dec9–23.

[36]ChandraA.Economicsmeetsthegeographyofmedicine;2013. [37]BeveridgeW.Socialinsuranceandalliedservices.London:HMSO;1942. [38]FerréF,DeBelvisAG,ValerioL,LonghiS,LazzariA,FattoreG,etal.Italy:health

systemreview.HealthSystemsinTransition2014;16(4):1–168. [39]GrayM.Valuebasedhealthcare.BMJ2017;356:j437.

[40]NutiS,DeRosisS,BoncianiM,MuranteAM.Re-thinkinghealthcare perfor-manceevaluationsystemstowardsthepeople-centerednessapproach:their pathways,theirexperience,theirevaluation.HealthcarePapers2017. [41]PianoNazionaleEsiti(NationalOutcomeProgram)(2015).http://95.110.213.

190/PNEed15/index.php.

[42]Anessi-PessinaE,Cantu `E,JommiC.Phasingoutmarketmechanismsinthe italiannationalhealthservice.PublicMoney&Management2004:309–16. [43]BertaP,MartiniG,MosconeF,VittadiniG.Theassociationbetweenasymmetric

information,hospitalcompetitionandqualityofhealthcare:evidencefrom Italy.JournaloftheRoyalStatisticalSociety:SeriesA(StatisticsinSociety) 2016;179(4):907–26.

[44]BertiS.Infarto,lareteIMAfa“90”,Sole24OreSanità;2014.p.14–20.Ottobre. [45]RegioneToscana,Availableonline:D.G.R.n.1380del27–12;2016http://www.

regione.toscana.it/bancadati/atti/?redirect=/bancadati/atti/DettaglioAttiG. xml%3fcodprat=2016DG00000001707.

[46]MarzegalliM,FontanaG,SesanaG,GriecoN,LombardiF,ElenaC,etal.Le retidell’emergenzainCardiologia:l’esperienzalombarda.GiornaleItalianodi Cardiologia2008;9(Suppl.1–10).

[47]RegioneLombardia,decreton.10446,15OttobreDirezioneSanitaria, Deter-minazioniinmeritoalla “reteperiltrattamento deipazienticoninfarto miocardicocontrattoSTelevato(STEMI)”;2009.

[48]KrumholzHM,NormandS,KeenanP,etal.Hospital30-dayacutemyocardial infarctionreadmissionmeasure:methodology.ReportpreparedfortheCenters forMedicare&MedicaidServices;2008.

[49]KrumholzHM,LinZ,KeenanPS,ChenJ,RossJS,DryeEE,etal.Relationship between hospital readmission and mortality rates for patients hospital-ized withacutemyocardialinfarction,heartfailure,orpneumonia.JAMA 2013;309(6):587–93.

[50]Mullahy J.Much adoabouttwo: reconsideringretransformation andthe two-part model in health econometrics. Journal of Health Economics 1998;17(3):247–81.

[51]MullahyJ.Econometricmodelingofhealthcarecostsandexpenditures:a surveyofanalyticalissuesandrelatedpolicyconsiderations.MedicineCare 2009;47(7Supplement1):S104–8.

[52]ManningW.Dealingwithskeweddataoncostsandexpenditures.In:Theelgar companiontohealtheconomics.Cheltenham:Elgar;2006.

[53]ManningWG,MullahyJ.Estimatinglogmodels:totransformornotto trans-form?JournalofHealthEconomics2001;20(4):461–94.

[54]HosmerJR,LemeshowS.Appliedlogisticregression.JohnWiley&Sons;2004. [55]TianL,HuangJ.Atwo-partmodelforcensoredmedicalcostdata.Statisticsin

Medicine2007;26(23):4273–92.

[56]GlickHA,DoshiJA,SonnadSS,PolskyD.Economicevaluationinclinicaltrials. OUPOxford;2014.

[57]BasuA,PolskyD,ManningWG.Estimatingtreatmenteffectsonhealthcare costsunderexogeneity:isthereamagicbullet?HealthServicesandOutcomes ResearchMethodology2011;11(1-2):1–26.

[58]IacusSM,KingG,PorroG.Multivariatematchingmethodsthatare mono-tonicimbalancebounding.JournaloftheAmericanStatisticalAssociation 2011;106(493):345–61.

[59]IacusSM,KingG,PorroG.Causalinferencewithoutbalancechecking: coars-enedexactmatching.PoliticalAnalysis2012;20(1):1–24.

[60]McManusDD,GoreJ,YarzebskiJ,etal.Recenttrendsintheincidence, treat-ment,andoutcomesofpatientswithstemiandnstemi.TheAmericanJournal ofMedicine2011;124(1):40–7.

[61]Gottlieb DJ, Zhou W, Song Y, Andrews KG, Skinner JS, Sutherland JM. Prices don’t drive regional medicarespending variations. Health Affairs 2010;29(3):10–377.

[62]CostaAP,PossJW,PeirceT,HirdesJP.Acutecareinpatientswithlong-term delayed-discharge:evidencefromaCanadianhealthregion.BMCHealth Ser-vicesResearch2012;12(172).

[63]WeeksWB,ParaponarisA,VentelouB.Geographicvariationinratesofcommon surgicalproceduresinFrancein2008–2010,andcomparisontotheUSand Britain.HealthPolicy2014;118(2):215–21.

[64]Emanuel EJ, Fuchs VR. The perfect storm of overutilization. JAMA 2008;299(23):2789–91.

[65]AmericanCollegeofCardiology,availableonlineatAmericancollegeof cardiol-ogythemanagementofMVDinSTEMI:thescienceandartofdecision-making inSTEMI;2018https://www.acc.org/latest-in-cardiology/articles/2018/02/07/ 07/45/the-management-of-mvd-in-stemi.

[66]LiZ,ZhouY,XuQ,ChenX.Stagedversusone-timecompleterevascularization withpercutaneouscoronaryinterventioninSTEMIpatientswith multives-seldisease:asystematicreviewandmeta-analysis.PLoSOne2017;12(1), e0169406.

[67]CongressionalBudgetOffice.Geographicvariationinhealthcarespending,Vol. 2008.Washington,DC:CongressoftheUnitedStates,CBO;2008.

[68]WennbergDE,DickensJrJ,SouleDN,KellettJrM,MalenkaD,RobbJ,etal.The relationshipbetweenthesupplyofcardiaccatheterizationlaboratories, cardi-ologistsandtheuseofinvasivecardiacproceduresinnorthernNewEngand. JournalofHealthServicesResearch&Policy1997;2(2):75–80.

[69]NutiS,VainieriM,BoniniA.Disinvestmentforreallocation:aprocesstoidentify priorityinhealthcare.HealthPolicy2010;95:137–43.

[70]NutiS,NotoG,VolaF,VainieriM,ManagementDecisionLet’splaythepatients’ music:anewgenerationofperformancemeasurementsystemsinhealthcare; 2018.

[71]BevanG,EvansA,NutiS.Reputationscount:whybenchmarkingperformance isimprovinghealthcareacrosstheworld.HealthEconomics,PolicyandLaw 2018.

[72]KhawajaFJ,ShahND,LennonRJ,SlusserJP,AlkatibAA,RihalCS,etal. Fac-torsassociatedwith30-dayread-missionratesafterpercutaneouscoronary intervention.ArchivesofInternalMedicine2012;172(2):112–7.

[73]LindenauerPK,LaguT,RothbergMB,AvruninJ,PekowPS,WangY,etal.Income inequalityand30dayoutcomesafteracutemyocardialinfarction,heartfailure, andpneumonia:retrospectivecohortstudy.BMJ2013;346:f521.

[74]AustinPC,TuJV.Comparingclinicaldatawithadministrativedataforproducing acutemyocardialinfarctionreportcards.JournaloftheRoyalStatisticalSociety 2006;169(1):115–26.

[75]NutiS,BiniB,RuggieriTG,PiaggesiA,RicciL.Bridgingthegapbetween the-oryandpracticeinintegratedcare:thecaseofthediabeticfootpathwayin Tuscany.InternationalJournalofIntegratedCare2016;16(2):1–14.