ContentslistsavailableatScienceDirect
Journal
of
Health
Economics
jo u rn al h om ep age :w w w . e l s e vi e r . c o m / l o c a t e / e c o n b a s e
Does
price
deregulation
in
a
competitive
hospital
market
damage
quality?
Anne-Fleur
Roos
a,∗,
Owen
O’Donnell
b,
Frederik
T.
Schut
c,
Eddy
Van
Doorslaer
d,
Raf
Van
Gestel
e,
Marco
Varkevisser
caNetherlandsBureauofEconomicPolicyAnalysis(CPB)&ErasmusSchoolofHealthPolicy&Management(ESHPM),Erasmus
UniversityRotterdam(EUR),Netherlands
bErasmusSchoolofEconomics(ESE)&ESHPM,EUR,TinbergenInstitute(TI),Netherlands cESHPM,EUR,Netherlands
dESHPM&ESE,EUR,TI,Netherlands eESHPM&ESE,EUR,Netherlands
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received17April2018
Receivedinrevisedform14April2020 Accepted18April2020
Availableonline23May2020 JELclassification: I11 L14 L15 Keywords: Healthcare Hospital Competition Quality Contracting
a
b
s
t
r
a
c
t
Regulatorsmaybehesitanttopermitpricecompetitioninhealthcaremarketsbecauseof itspotentialtodamagequality.Weassesswhetherthisfeariswellfoundedbyexamining areformthatpermittedDutchhealthinsurerstofreelynegotiatepriceswithhospitals. Unlikepreviousresearchonhospitalcompetitionthathasreliedonqualityindicatorsfor urgenttreatments,wetakeadvantageofaplausibleabsenceofselectionbiastoidentifythe effectonthequalityofelectiveproceduresthatshouldbemorepriceresponsive.Usingdata onalladmissionsforhipreplacementstoDutchhospitalsandadifference-in-differences comparisonbetweenmoreandlessconcentratedmarkets,wefindnoevidencethatprice deregulationinacompetitiveenvironmentreducesqualitymeasuredbyhipreplacement readmissionrates.
©2020ElsevierB.V.Allrightsreserved.
1. Introduction
Competitionbetweenhealthcareprovidersis
increas-inglyencouragedwiththeaimofimprovingqualityofcare
whileslowingthegrowthofhealthspending.Whenprices
areregulated,providersareforcedtocompeteonquality
toattractpatientsorcontractswithinsurers.Whenprices
∗ Correspondingauthor.
E-mailaddresses:a.f.roos@cpb.nl(A.F.Roos),odonnell@ese.eur.nl (O.O’Donnell),schut@eshpm.eur.nl(F.T.Schut),vandoorslaer@ese.eur.nl (E.VanDoorslaer),vangestel@ese.eur.nl(R.VanGestel),
varkevisser@eshpm.eur.nl(M.Varkevisser).
areunregulated,theeffectofcompetitiononqualityisless
clear.Ifdemandismoreresponsivetopricethantoquality,
thentheoptimalcompetitivestrategywillinvolvedriving
downthepriceandsacrificingquality(Gaynor,2006).This
isaplausiblescenariowheninformationonqualityispoor.
Fearthatcompetitionwithunregulatedpriceswillbe
dam-agingtoqualitymaymakeregulatorswaryaboutallowing
healthcareproviderstocompeteonprice.However,itisnot
clearwhetherthisfeariswellfounded.Providersmaynot
adopttheprofitmaximizingcompetitivestrategies.
Not-for-profitgoals,asocialmissionandintrinsicmotivation
mayleadthemtomaintainqualityevenifthismeans
for-goingopportunitiestogain a competitiveadvantage by
https://doi.org/10.1016/j.jhealeco.2020.102328 0167-6296/©2020ElsevierB.V.Allrightsreserved.
cuttingpricesattheexpenseofquality.Whetherquality
suffers in competitivehealthcare marketswith
unregu-latedpricesisanempiricalquestion.Evidencetoanswer
itissparse.
Thispaperexaminestheimpactofpricederegulation
on the quality of hospital care delivered in the Dutch
healthcare market in which insurers compete for
cus-tomersandhospitalscompeteforcontractswithinsurers.
Weestimatetheeffectofmovingfromfinancinghospitals
throughprospectiveglobalbudgetstoallowinginsurers
andhospitalstofreelynegotiateprocedure-specificprices
in contractsfor the deliveryof DRG-type products.We
identifytheeffectofthispricederegulationbyexploiting
variationinitsconsequences acrosshospitals
differenti-atedby theconcentration of themarket in which they
operate.Assumingthatfreeinsurer-hospitalnegotiationof
pricescreatesgreatercompetitivepressurewherethe
mar-ketislessconcentrated,thedifference-in-differences(DID)
betweenmoreandlessconcentratedmarketscanidentify
alowerboundontheeffectofderegulatingpricesinamore
competitiveenvironment.
Weestimatetheeffectofpermittingpricecompetition
onan elective procedurequality indicator– unplanned
readmissionafternon-acutehipreplacement.Higher
read-missionratesfollowinghipreplacementhavebeenshown
toberelated tosuboptimal quality(Rosen et al.,2013;
Mokhtaret al.,2012).Theinstitutional context and our
empiricalstrategyfacilitateidentificationwithoutrunning
muchriskoftheselectionbiasthatmostotherstudiesof
competitioninhealthcaremarketshaveavoidedonlyby
estimatingeffects onindicatorsof thequalityofurgent
treatments,suchasmortalityafteracutemyocardial
infarc-tion(AMI)(e.g.Kessler&Geppert, 2005; Propperetal.,
2008;Cooperetal.,2011;Mutteretal.,2011;Romano& Balan,2011;Gaynoretal.,2013).1Thisrestrictionof
atten-tiontourgenttreatmentsleavesadearthofevidenceonthe
effectofcompetitionontreatments,suchaselective
surg-eries,thathospitalsdirectlycompeteforandthedemand
forwhichpotentiallyexhibitsmuchgreaterresponsiveness
topriceandquality(BevanandSkellern, 2011;Gravelle
etal.,2014;Collaetal.,2016;Skellern,2019).Thereform
weexploit permittedprice competitionbut leftpatient
choiceof provider effectively unconstrained. There was
noavailableinformationonhipreplacementreadmission
rates,andsopatientscouldnotselecta hospitalonthe
basisofthisoutcome.Toidentifytheeffect,weseparate
hospitalsintotwobroad(treatment/comparison)groups
accordingtotheconcentrationofthemarketinwhichthey
operate.Iftherewere anyselectioncorrelated withthe
outcome,itwouldmostlikelyinvolveswitchingbetween
1 TheurgencyofAMItreatmentgreatlyreducestheriskofselectionbias
sincepatientsaresimplytakentothenearesthospital.Thereislittleorno opportunityfordifficult-to-treatpatientsselectinghospitalsthatdeviate fromtheaverageinbothqualityandexposuretocompetition.Andthere islittlescopeforthosehospitalstocherrypicktheeasiercases.However, thisempiricalstrategyidentifiestheimpactofcompetitiononquality onlyinsofarasthepressuretocompeteinthedeliveryoftreatments thatarepriceand/orqualityresponsiveaffectsthegeneralmanagement ofahospitalandthisfeedsthroughtotreatments,suchasAMI,thatare largelyshieldedfromcompetition.
neighboringhospitalsthatbelongtothesamegroup.This
wouldnotinduceselectionbias.Baselinepatient(casemix)
characteristicsaresimilaracrossthetreatmentand
com-parison groups, changes in these characteristicsdo not
differbetweenthegroupsandconditioningonthese
char-acteristicshaslittleimpactontheestimates.
Wefindnoeffectonqualitydespiteexamininga
situ-ationinwhichpricederegulationhadthegreatestscope
todamagequality–anelectiveprocedurewithlittle
infor-mationonitsquality,potentiallyleaving thedemandof
insurer-purchasersmoreresponsivetopricethanto
qual-ity. Overa five-yearperiodafterpricederegulation,the
changeinthe90-dayhipreplacementreadmissionrates
of hospitalsin less concentrated markets didnot differ
significantlyfromthatofhospitalsinmoreconcentrated
marketsthatwereexposedtolesscompetitivepressure.
Theinsignificantpointestimateissmallinmagnitude–less
than1percentofthepre-reformreadmissionrate–and
reasonablypreciselyestimated.Failuretorejectthenullof
noeffectisrobusttoalternativedefinitionsofthemarket,
tocomparinghospitalsattheextremesofmarket
concen-tration,tousingthe30-day(insteadof90-day)readmission
rate,todroppingthemostruralhospitalsandtousing
read-missionafterkneereplacementasthequalitymeasure.In
theyearimmediatelyafterpricederegulation,wefinda
marginallysignificantnegativeeffectonthereadmission
rate,whichisfollowedbyinsignificantpositivepoint
esti-matesin later years.This hintsata positive immediate
impactonqualitythatisnotsustainedwhenhospitals
con-tinuetobeexposedtocompetitivepressuresarisingfrom
pricederegulation.Overall,this studyfindsnoevidence
thatpricederegulationinamorecompetitivehealthcare
marketdamagesquality,evenwheninformationonquality
ispoor.
2. Competitionandhealthcarequalitywith unregulatedprices:theoryandevidence
Whenpricesareunregulated,theimpactofcompetition
onqualitydependsonhowitaffectstheresponsivenessof
demandtoqualityrelativetoitsresponsivenesstoprice.If
consumers,orinsurerspurchasingontheirbehalf,observe
pricesbuthaveonlyimperfectinformationonquality,then
competitionmightbeexpectedtoraisethepricesensitivity
relativetothequalitysensitivityofdemand,andsoreduce
quality(Kranton,2003).Gaynor(2006)makesthis
argu-mentusinganamended versionoftheDorfman-Steiner
condition(DorfmanandSteiner,1954):=p
d εz
εp,wherezis
quality,pisprice,disthemarginalcostofquality,εz is
theelasticityofdemandwithrespecttoqualityandεpis
theelasticitywithrespecttoprice.2Ifcompetitionexerts
downwardpressureonthepricerelativetothemarginal
costand/orraisesthemagnitudeofthepriceelasticity
rel-2Although Dorfman and Steiner (1954) model a monopolist’s
behaviour, Dranoveand Satterthwaite (2000) show thatthe model providesanapproximationtothebehaviourofanoligopolisticor monop-olisticallycompetitivefirmifwethinkofthedemandfunctionasareduced form.Hence,themodelhasrelevanceforimperfectlycompetitive health-caremarkets(Gaynor,2006;Gaynoretal.,2015).
ativetothequalityelasticity,thenitwillreducequality (Gaynoret al.,2015).3 However,ifqualityissufficiently
observable,thencompetitioncouldconceivablyraisethe
quality elasticity relative to theprice elasticity.Quality
wouldincrease,providedpricedoesnotfallrelativetothe
marginalcostofquality.Theeffectofcompetitionon
qual-itywithanunregulatedpriceisthereforeambiguous.It
dependsoncharacteristicsofthemarket,theobservability
ofqualityandtheobjectivefunctionsofthedemand-side
and supply-sideagents –insurersand hospitals
respec-tively(Gaynoretal.,2015).
Evidence on theeffect of competition on healthcare
quality when prices are unregulated is scarce.4 This is
mainlybecauseonlyafewcountriespermitfreeprice
nego-tiationinhealthcaremarketsanddataontheperformance
ofprivatehealthcareprovidersaretypicallynotaccessible.
UsingdatafromSouthernCalifornia,Gowrisankaranand
Town (2003)findthatincreasedcompetitionfor Health
Maintenance Organization (HMO) patientsis correlated
with reduced risk-adjusted hospital mortality for both
pneumonia and AMI. Consistent with this, Sari (2002)
findsthatlowerhospitalmarketconcentrationin16US
states is associated with fewer hospital complications.
However, the internal validity of these studies can be
doubted becauseof endogeneity problems (Gaynor and
Town,2012),andtheirexternalvalidityislimitedbecause
theHMOmarketsstudiedareveryparticulartotheUS
hos-pitalmarketinthe1990s.
Thefewstudiesthatexploitapolicychangeto
iden-tifythequalityeffectofsomeformofpricederegulation
arestrongerwithrespecttointernalvaliditybutalso
dif-ficulttogeneralizebecausethefindings areobtainedin
specificsettingswithaparticulardesignofprice
competi-tion.Subjecttothiscaveat,thesestudiesgenerallyfindthat
permittinggreaterpricecompetitiondoesdamage
hospi-talquality.Volppetal.(2003)compareAMImortalityrates
ofNewJerseyhospitalsbeforeandafterthederegulation
ofpricesin1992withthoseofNewYorkhospitalswhere
there wasno deregulation. The mortality rateof
unin-suredAMIpatientsincreasedinNewJerseyrelativetoNew
York.However,coincidenttothereforminNewJerseyand
potentiallyconfoundingitseffect,hospitalpriceswerealso
pressuredthroughrapidgrowthoflarge-volumebuyers,
suchasHMOs,andtherewerelargereductionsinsubsidies
forhospitalcareofuninsuredpatients.
Theswitchfromfixedbudgetsthathospitalsreceived
directlyfromthenationalgovernmenttocontracts
hospi-3Itisunlikelythathospitalsdeliberatelysetouttolowerqualityof
care.Studiesthatinvestigatethecompetition-qualityrelationshipoften arguethatinresponsetocompetitivepressurehospitalsmaycutservices thataffectqualityoutcomes(Propperetal.,2008;Bloometal.,2015). GaynorandTown(2012)showthat,forthepurposeofmodeling,itdoes notmatterwhetherhospitalsareassumedtochoosequalitydirectlyor indirectlythrougheffortexerted.
4Thereismoreevidenceontheimpactofcompetitiononhealthcare
qualitywhenpricesareregulated.Findingsaremixed.Somestudiesfind thatcompetitionimprovesqualityinthiscontext(KesslerandMcClellan, 2000;KesslerandGeppert,2005;Cooperetal.,2011;Gaynoretal.,2013; Gaynoretal.,2016;GobillonandMilcent,2017).Othersfindevidenceof thecontrary(Moscellietal.,2019;Skellern,2019),whileonestudyfinds noeffectatall(Bertaetal.,2016).
talsnegotiatedwithpurchasingorganizationsintheBritish
NationalHealthServicein1991hasbeenusedtoestimate
thequalityeffectofahighlyregulatedformofprice
compe-tition(Propperetal.,2004;Propperetal.,2008).Contracts
werewrittenforblocksofservices,includingaccidentand
emergencyproceduresandnotfordefinedproducts,such
asDRGs.Hospitalsweremandatedtosetpriceequal to
averagecost,hadtopublishthesepricesandwerenot
per-mittedtocarrysurplusesorlossesacrossfinancialyears.A
limiteddegreeofpricecompetitionwaspossibleatthe
spe-cialtylevelbecauseofthedifficultyofcheckingadherence
tothepricingruleatthatlevel(Propperetal.,2008).The
evidencesuggeststhateventhisregulatedformofprice
competitionhadanegativeimpactonquality(measured
byAMImortalityrates),whichisattributedtohospitals’
incentivestocompeteonpriceratherthanqualitywhen
theavailableinformation onthelatter ispoor(Propper
etal.,2004;Propperetal.,2008).
Ifthehighlyregulatedformofpricecompetition
per-mittedin theUKcoulddamagequality,thenonemight
anticipatethatallowinghospitalstofreelynegotiateprices
withpurchasers would be seriouslydetrimental to the
quality of care deliveredin more competitive markets.
The2005reform oftheDutchhospital marketprovides
anopportunitytotestthis conjecture. Sincethere were
nopublishedoutcome indicators of qualityavailable to
patientsandhealthinsurersbeforeandafterthereform,
theriskofanegativeimpactonqualitywassubstantial.
Particularlyin morecompetitivehospitalmarkets,price
deregulationmayhaveraisedthemagnitudeoftheprice
elasticityrelativetothequalityelasticityandsomayhave
reducedquality.
3. PricederegulationintheDutchhospitalcare market
Comprehensive health insurance in the Netherlands
withverylimitedcostsharingleavespatientsinsensitiveto
priceandplausiblymoreconcernedaboutquality.5
How-ever,asDutchhealthinsurerscompeteonthepricesofthe
packagestheyoffer,theyarelikelytobemoresensitive
tothepricesofhealthcareproductsthanpatients.6 After
pricederegulationin2005,insurershad anincentiveto
pushpriceslowerinnegotiationswithhospitalsandwere
possiblymoreconcernedaboutpricethanquality.Thiswas
strengthenedbythelackofoutcome-basedquality
indica-tors–onlyalimitedsetofstructure-andprocess-based
indicatorswasavailableatthetime.Ifbetterinformation
5 Inthespecificcontextofourstudy,pricesofhipreplacementsexceed
thedeductible,whichistheonlyformofcost-sharing.Forthis proce-dure,out-of-pocketcoststopatientsare,therefore,invarianttotheprices chargedbyhospitals.
6 Thescenarioweexamineissimilartothatcapturedbyatwo-stage
modelinwhichinsurers(ManagedCareOrganizations)andhospitals firstnegotiateprices,andthenpatients,whoareexposedtolittlecost sharing,selectahospital(Cappsetal.,2003;TownandVistnes,2001; Gowrisankaranetal.,2015).TheestimatesGowrisankaranetal.(2015) obtainfromsuchamodelimplythattheinsurersaremuchmoreprice sensitivethantheinsuredpatients,buttheinsurersarelesspricesensitive thanpatientswouldbeiftheywereuninsured.
onqualityhadbeenavailable,theninsurers,presumably,
wouldhavebeenmoreresponsivetoit.
AllDutchhospitalsareprivatenonprofitfoundations
facinga legallybindingnon-distributionconstraint
pro-hibitingthemfromdistributinganynetearnings.Before
2005,thehospitalswerefinancedbyaprospective
bud-getingsystemwithregulatedperdiemratesthatproduced
relativelystablerevenueflowsknownatthebeginningof
eachyear.From2005,hospitals’revenuesbecame
contin-gentoncontractssecuredwithindividualhealthinsurers.
Atthattime,therewerefivehealthinsurancecompanies,
plusajointpurchasingcooperativeofsmallerinsurers.7
Contractsarewrittenforproductsdefinedbyapurposively
developedproductclassificationsystem–Diagnosisand
TreatmentCombinations (DTCs)– akintoDRGs.8 These
productsbearnorelationtotheoutputparametersofthe
pre-reformhospitalbudgets(e.g.number ofadmissions
andhospitaldays)andtheproduct-specificpricesarenot
relatedtotheregulatedperdiemratesthatwereusedtoset
thesebudgets.Forallproducts,hospitalsandinsurersare
permittedtonegotiateovervolumewhiletakingaccountof
quality,althoughintheperiodwestudytheattentionthat
couldbepaidtoqualitywashamperedbythelackof
out-comeindicators.9Forsomeproducts,freeinsurer-hospital
negotiationofpriceswasallowed.In2005,thiswas
permit-tedforasubsetofproductsthataccountedforabout10%
ofhospitalrevenues.Thissubsetincludednon-acutehip
replacements–theprocedurewefocuson.Overtime,the
numberofproductsforwhichfreepricesettingis
permit-tedhasincreased.Anextensionin2008tooktheshareof
hospitalrevenuesobtainedfromproductswithnegotiated
pricesto20%.Thefractionwasfurtherincreasedto34%in
2009and70%in2012.
Afterpermittingpricecompetition,theprice
elastic-ityoftheinsurers’ demandisgreater(inmagnitude)in
lessconcentrated,morecompetitivemarkets,andprices
couldbepusheddownfurtherinthesemarkets.
Accord-ingtotheDorfman-Steinercondition,qualitywouldthen
suffer in more competitive markets unless there was
a sufficient countervailing increase in the quality
elas-ticity. This increase would occur only if quality was
sufficientlyobservablesuchthatinsurerscouldmonitorit
andthenewcontractingarrangementsgavethemgreater
motivation and scope to pressure hospitals for quality
improvements.10
7 Thefourlargestcompaniesaccountfor90percentofthemarket.
Mar-ketconcentrationbyregionisoftenevenhigher,whichisduetothe factthatthesecompaniestypicallyevolvedfromformerregionalsickness funds(Halbersmaetal.,2010).
8 TheDTCsystemismorecomprehensivethanDRGs.Itincludes
out-patientconsultationsandtheremunerationofmedicalspecialists.There wereabout29,000DTCsintheperiodweexamine.
9 Duringtheperiodcoveredbyouranalysis,almostallcontractual
agreementsthatincludedqualityimprovementinitiativeswereframedin termsofstructureandprocessindicatorsratherthanoutcomeindicators (SchutandVandeVen,2011).
10 Notethat,atthetimeofthecontractingreform,medical
special-istswerepaidfee-for-serviceatregulatedproduct-specificremuneration ratesdeterminedbythepredictedtimerequiredforeachprocedureand afixedpaymentperhour.Whilespecialists,likehospitals,hadafinancial interestinattractingmorepatients,theircompensationwas,therefore,
Becauseofthehighoverallnumberofproducts(DTCs),
insurersandhospitalsoftennegotiateoverclustersof
prod-ucts.Thiswasnotthecaseforhigh-volumeproductslike
non-emergencyhipreplacementsduringtheperiod
cov-ered by this study.11 At that time, contracts, including
those agreedafternegotiationoverprices,werewritten
foraperiodofoneyear.Thegoalofthecontractingreform
wastomakeinsurers,actingaspurchasingagentsfortheir
customers,moreresponsivetoprice,volumeandquality.
Insurerswereallowedtocontracthospitalsselectively,
giv-ingthemleveragetonegotiatelowerpricesand,possibly,
alsotoobtainbetterqualitytotheextentthatthiscouldbe
specifiedbythelimitedqualityindicatorsavailableatthe
time.
Inacompetitiveinsurancemarket,lowerpriceswould
feedthroughtolowerpremiums(Ho,2009).Inthisrespect,
an important complement tothe 2005 insurer-hospital
contractingreformwasa2006reformofthehealth
insur-ancemarketintendedtoincreasepricecompetitionamong
insurersbymandatingcitizenstopurchaseabasichealth
insurancepackagefromprivateinsurers(SchutandVande
Ven,2011).Thelogicofthesetwinreformsisthat–asin
thetwo-stagemodelofhospitalcompetition(Cappsetal.,
2003;TownandVistnes,2001;Gowrisankaranetal.,2015)
–insurerswouldcompeteforcustomersonpremiums,as
wellasthescopeandqualityoftheirprovidernetworks,
while hospitalswould competeonpriceand qualityfor
inclusion inthosenetworks.Infact,prior to2010–the
last yearcovered by ouranalysis – there was very
lit-tleimplementationofselectivecontracting.Bythatyear,
onlyoneinsurerhadenteredintoselectivecontractsand
thesedidnotcoverhipreplacements(SchutandVande
Ven,2011).Since2010,thenumberofinsurerscontracting
withrestrictedprovidernetworkshasincreased.12But
dur-ingthepost-reformperiodwestudy(2006–2010),insurers
reliedonthethreatofselectivecontracting,ratherthanits
implementation,togivethembargainingpowerin
negoti-ationswithhospitals.
Someinsurerstriedtosteertheircustomerstohospitals
identifiedontheirwebsites as‘preferredproviders’that
wereclaimedtooffergoodqualitycare,althoughchoice
wasnotrestrictedtothesehospitals.Oneinducementwas
towaivethedeductibleiftheinsuredsoughttreatment
at apreferredprovider. Theavailableevidencesuggests
thatthesesoftchannelingpolicieshadlittleinfluenceon
patientchoiceduringtheperiodwestudy(VanderGeest
andVarkevisser,2016,2019),andwearenotawareofany
attempttousethemtospecificallytargethipreplacement
notdirectlyaffectedbytheoutcomeofanypricenegotiation.The qual-ityoftreatmentdeliveredbythespecialistscouldbeaffectedindirectly throughthevolumeofprocedures–negotiatedsimultaneoustoprice– theywerecalledupontoperform.Quality–specifically,thereadmission ratesweexamine–couldalsohavebeenaffectedbypricenegotiation throughconsequentpressuresonnon-specialistmedicalstaff,attention paidtoaftercareandhospitalpurchasingofnon-manpowerinputs.
11Thishasbeenconfirmedbytheauthorsthroughinterviewswith
rep-resentativesofinsurersandhospitalswhowereinvolvedincontract negotiationsduringtheperiodcoveredbythisstudy.
12Thisisoftendonebystipulatingminimumvolumestandards,butonly
foralimitedsetoftreatments,suchascomplexcancersurgery,thatdoes notincludehipreplacements.
patients.Additionally,asurveyofpatientswhohad
non-emergencyhospitalcareintheperiod2008-10revealed
thatonly7percenthadsearchedforqualityinformation
(VanderGeestandVarkevisser,2012).Hence,thereform
leftpatients’choiceofhospital(forhipreplacementsand
otherprocedures)effectivelyunfetteredduringtheperiod
studied.
The complexityof healthcare andits stochastic
rela-tionshipwithhealthoutcomesmakesmeasurementofits
qualityinherentlydifficult.In2008,theDutchHealthCare
Inspectoratelaunchedaninitiativewiththeambitionof
developingacomprehensivesetofqualityindicatorsfor10
procedures,includinghipreplacement.Butthiswas
con-finedtoaminorityof33hospitalsandplanstoextendit
werenotimplementedbecauseoflackofcooperationfrom
hospitalsandfailuretoagreeonastandardizedsetof
qual-ityindicators(SchutandVandeVen,2011).Afterthisfailed
attemptatselfregulation,onlyin2013didthegovernment
establishaQualityInstitutetodisseminateuniform
qual-ityindicatorsanddefinelegallyenforcedstandards.Hence,
uptotheendofourperiodofanalysisin2010,no
mean-ingful, comparable information aboutthequality ofhip
replacements,suchashospital-specificreadmissionrates,
wasavailable.
Thelackofinformationwouldbeexpectedtoresultin
hospitalsexposedtogreatercompetitivepressureshifting
effortfrom maintainingpoorlyobserved quality to
cut-tingcostsinordertobecomemorepricecompetitiveonce
priceswerederegulated(Propperetal.,2008).Ontheother
hand,thenewcontractsinvolvedhospitalsandinsurers
negotiatingforthefirsttimeoverthedeliveryofspecific
proceduresakintoDRGs.Hospitalsexposedtomore
com-petitionmighthaveexpendedgreatereffortonensuring
thattheywerenotpenalizedforpoorperformanceonthe
availableproxiesforqualitytakenintoaccountinfuture
contract negotiations.Ifthis motivationwassufficiently
strong,thenexposuretogreatercompetitivepressureafter
thecontractingreformcouldevenhaveraisedquality.
Existingevidenceonthemarketresponsetothe2005
contracting reform is limited. Qualitative analyses
con-cludethatpriceratherthanqualityhasbeentheprimary
focusofcontractnegotiations(Meijeretal.,2010;Ruwaard
etal.,2014;SchutandVandeVen,2011).Thisis
unsurpris-inggiventhedearthofinformationavailableonquality.The
DutchHealthcareAuthorityreportsthatpricesofproducts
(DTCs)thatwereinthefree-pricingsegmentfrom2005
declinedinrealtermsandrelativetotheregulatedpricesof
otherproductsupto2008(NZa,2009).Between2006and
2009,nominalpricesincreasedby2.7%,onaverage,inthe
regulatedsegmentofproductscomparedwithanincrease
ofonly1.2%inthefree-pricingsegment(SchutandVande
Ven,2011).Thereisnoevidencethathospitalsoffsetlower
priceincreasesbyincreasingservicevolumeinthe
free-pricingsegment(Krabbe-Alkemadeetal.,2017;Schutand
VandeVen,2011).Krabbe-Alkemadeetal.(2017)found
thatpermittingpricecompetitionledtolowertotal
hospi-talcosts.
Theeffectofthepricederegulationbroughtaboutbythe
2005reformonhospitalqualityhasnotpreviouslybeen
estimated.Afewstudieslookattherelationshipbetween
priceand qualityvariation orbetweenhospital
concen-trationand qualityafterprices wereliberalized.Heijink
etal.(2013)findonlylimitedvariationinhospitalquality
andnorelationshipbetweencontractpricesandqualityfor
cataracttreatment.Croesetal.(2017)findanegative
rela-tionshipbetweenhospitalmarketshareandqualityscores
fortwoofthethreediagnosticgroupsstudied.Bijlsmaetal.
(2013)findthathospitalconcentrationisassociatedwith
variousprocessindicatorsofquality,butbothpositiveand
negativerelationshipsarefoundandthereisno
relation-shipbetweenhospitalconcentrationandanyofanumber
of outcome indicators examined.None of these studies
haveadesigncapableofidentifyingthecausaleffecton
qualityofpricederegulation.
4. Dataandmeasures
4.1. Sources
We usecomprehensive, hospital-level datafrom the
NationalMedicalRegistryonpatientdischargesfromall
Dutchhospitalsfor the years 2001–2010.For each
dis-charge, we observe the patient’s gender, age,zip code,
primary/secondary/tertiarydiagnoses(ICD-9CM),
admis-sion period, the admission hospital code (but not its
name)andprocedurescarriedout.Proceduresare
classi-fiedaccordingtoamethodthatisbasedon,andforthe
proceduresexaminedequivalentto,theInternational
Clas-sificationofProceduresinMedicine(WHO-FIC,2017).For
mostoftheanalyses,werestrictattentiontopatients
dis-chargedafteranon-acutehipreplacement(seebelowfor
detailsoftheselectioncriteria).13Asarobustnesstest,we
examinepatientsdischargedafternon-acuteknee
replace-ment.Weconstructahospital-levelpanelwhichincludes
informationonqualityofcareandpatientcasemix,and
supplementthiswithanindexofsocioeconomicstatusthat
isaveragedoverallthenon-acutehipreplacementpatients
ofahospitalinagivenyear.Thisindexisconstructedby
theNetherlandsInstituteforSocialResearchfromthe
edu-cation,incomeandlabormarketstatusofresidentsofazip
codearea(SCP,2017).
4.2. Qualitymeasures
We use theunplanned 90-day readmission rate
fol-lowing non-acute hipreplacement as our main quality
indicator. Thisis preferred totheposthip replacement
mortality rate because the latter was very low in the
periodstudied.14 Werestrict attentiontoallunplanned
readmissions,includingemergencyreadmissions,because
plannedreadmissions(e.g.forascheduledprocedure)are
notgenerallyasignalofqualityofcare.Higherunplanned
readmissionrateshavebeenshowntoberelatedto
sub-optimalqualityoftreatmentgenerally(e.g.Mokhtaretal.,
13 Alsosometimesreferredtoasplannedhipreplacementsorelective
hipreplacements.Hence,thesedonotincludehipfracturesoracutehip replacements.
14 UsingtheCausesofDeathRegisterprovidedbyDutchHospitalData
andStatisticsNetherlands,wecalculateawithinhospitalmortalityrate of0.25percentanda30-daymortalityrateof0.30percentfollowing non-acutehipreplacementintheperiod2001–2010.
2012;Rosenetal.,2013)andspecificallyforhip
replace-ments(e.g.Clementetal.,2013;Avrametal.,2014;Saucedo
etal.,2014;Kurtzetal.,2016).Allunplannedreadmissions
areattributedtotheoriginaltreatmenthospitalusingthe
anonymizedhospitalcodes.
Unplannedreadmissionsfollowing jointreplacement
aredetermined,inpart,bythequalityandsafetyofthe
initialhospitalstay,transitionalcareservicesandpost
dis-charge support(Friebel et al., 2017). Widespread belief
thatreadmissionsareindicativeofpoorqualitytreatment
isreflectedinthefactthatfinancialpenaltiesforexcess
readmissions(includingforhipreplacements)havebeen
imposedonhospitalsinboththeUSandtheUKsince2012
(JoyntandJha,2012).Consistentwiththis,inourdata,four
ofthetopfivereasons(identifiedfromdiagnosticcodes)for
hipreplacementpatientstobereadmittedwithin90-days
arerelated tocomplications, infections orinflammatory
reactionsduetoprostheticimplants.15Thereisno
consen-susonwhethera90-dayor30-dayfollow-upwindowto
defineorthopedicreadmissionsprovidesthebetter
indica-torofquality(Ramkumaretal.,2015).Sincethetwoare
highlycorrelatedforhipreplacementsinourdata(r=0.85,
p<0.01atthestartofourstudyperiodin2001),itshould
make little differencewhich is used. Complicationsare
alsothemainreasontobereadmittedwithin30-days.We
examinerobustnesstousing30-dayreadmissions.
Infor-mationonhipreplacementreadmissionrateswasnotin
thepublicdomainoravailabletohealthinsurersduring
theperiodofanalysis,andsothisindicatorisunlikelyto
havebeensubjecttomanipulationbyhospitals.
4.3. Sampleselection
Wehave abalanced panelof 103hospitalsobserved
from2001to2010,yielding1,030hospital-year
observa-tions.Patientlevelsampleinclusionandexclusioncriteria
arebasedonthosedefinedinthetechnicalspecifications
of the US Agency for Healthcare Research and Quality
(AHRQ)InpatientQualityIndicator#14(AHRQQIVersion
5.0;IQI#14),whichmeasuresthehipreplacement
mor-tality rate.The total populationincludes all discharged
patients aged 18 or older in any year between 2001
and 2010 with any procedure code that indicates
par-tialor fullhip replacementand anyprincipal diagnosis
codethatindicatesosteoarthrosisofthepelvicregionor
thigh,andforwhichallnecessaryinformationwaspresent
(n=153,208).16Forouranalysis,wedropthosewithany
listeddiagnosiscodes indicating hipfracture and those
withcodesindicatingpregnancy,childbirthorpuerperium
(n=793).Wealsoexcludethosewhotransfertoanother
hospital(n=264) because it is impossibleto determine
whetherreadmissioninsuchcasesindicatessub-optimal
qualityofthetreatmentreceivedinthefirstorthe
sec-15 Consequently,itisunlikelythatrestrictingattentiontounplanned
readmissionsjudged(bysomecriteria)toberelatedtotheindexhip replacementadmissionwouldhavemuchaffectonourestimates.
16 SeeAppendix1fortherelevantprocedurecodesandICD-9CM
diag-nosiscodes.
ondhospital.17Patientswhodiedinthehospital(n=85)
are also dropped.18 After imposing all these exclusion
restrictions,weareleftwith152,066(99.3%of)discharges
followingnon-acutehipreplacementduringourperiodof
analysis.Ofthese,8.0percentwerereadmittedtoahospital
within90daysforanyreasonthatwasnotplanned.19
4.4. Measuresofhospitalmarketstructure
Wemeasureconcentrationatthehospitallevelusing
theHerfindahl-HirschmanIndex(HHI)basedonthe
num-berofhospitalbeds:HHIh=
Nh
i=1
m2
i,wheremiisthepercent
marketshareofhospitalithatlieswithinafixedradius
of hospitalhand Nh isthetotal numberof hospitalsin
thatmarket.Somehospitalshavemultiplelocationsthat
donotallliewithinthesamemarketdefinedbydistance.20
Appendix2 explainshow wecalculatetheHHIinthese
cases.For ourbaselineanalysis,weusea30 kilometers
(by road)fixedradiusbecause patientstravel, on
aver-age,for20minutestogettothehospitaloftheirchoice
(Varkevisseretal.,2010;Varkevisseretal.,2012;Beukers
etal.,2014)andallDutchhospitalsdohipreplacements
(Roosetal.,2019).Butsincevariationaroundthemean
traveltime ishigh(Varkevisseret al.,2010; Varkevisser
etal.,2012;Beukersetal.,2014),weexaminesensitivityto
fixingtheradiusat20,40and50kilometerstodefinethe
market.
To protect privacy, hospitals are anonymized in the
patient-level data and we are not allowed to attach a
continuousmeasureofHHItoapatientrecordsincethis
couldrevealthehospitalused.TheHHI ofeachhospital
was thereforeconstructed in a databasenot containing
patient-leveldata.Next,anindicatorofwhethertheHHI
of each hospital is under2500was derived –to
deter-mine location in a less concentrated market – and this
was then linked to the patient-level dataset using the
hospital identifier code byStatistics Netherlands.There
is no objective HHI threshold for defining a
competi-tivemarket.Ourchoiceof2500,whichcorrespondstoa
marketcomprising fourequallysizedhospitals,isbased
on theUS FederalTrade Commission(2010) horizontal
mergerguidelines.21Giventhegreatercompetitive
pres-17Whiletransfertoanotherhospitalcoulditselfbeanindicatoroflow
quality,theextremelysmallnumberoftransfersrulesoutanalysisofsuch anindicator.Italsomeansthatdroppingthesecaseswillhaveanegligible impactonourestimates.
18Giventheverylowwithinhospitalmortalityratefollowingnon-acute
hipreplacement,anyselectionbiasarisingfromexcludingthosewhodie islikelytohaveanegligibleimpactontheestimates(Laudicellaetal., 2013;Fischeretal.,2014).
19AsshownbyFigureA2intheAppendix,totalvolumesaggregated
overthetwogroupsdroppedin2006andthenslightlyincreasedagain. Thedifferenceinvolumesbetweenthetreatmentandcomparisongroups isrelativelystableovertime.
20Ofthe103hospitals,4hadmorethanonelocationwithintheperiod
thatwestudy.
21TheEuropeanCommission(2004)guidelinesontheassessementof
horizontalmergersdonotincludeanHHI-basedmarketclassification. Theymentionaslightlylowerthreshold(HHI>2000)asaninitialindicator ofalackofcompetition.
sureonhospitalsoperatinginlessconcentratedmarkets
approximatedbyHHI<2500,itishypothesizedthatthey
wouldbeaffectedmorebythe2005reformthatpermitted
competitionthroughfreenegotiationofpricesthanwere
hospitalsoperatinginmoreconcentratedmarkets(i.e.with
HHI≥2500).22Overtheperiod2001–2010,thereisanearly
evensplitbetweenhospitalswithHHI≥2500(n=52)and
hospitalswithHHI<2500(n=51).
5. Empiricalstrategy
Identificationofthequalityeffectofpermitting
compe-titionthroughfreenegotiationofpriceisdifficultgiventhat
allhospitalsareexposedtopricederegulationafter2005.
Intherelatedliterature,identificationfroman
across-the-boardpolicychangeistypicallydonethroughcomparison
ofthebefore-and-afterchangeintheoutcomeinless
con-centrated markets with the respective change in more
concentrated markets(e.g. Propperet al., 2008; Cooper
etal.,2011;Gaynoretal.,2013).Whilethis
difference-in-differences(DID)strategydoesnotidentifytheeffectofthe
policyperse,itmaybethoughtcapableofidentifyingthe
differentialeffectofthegreatercompetitiongeneratedby
thepolicyinlessconcentratedmarketscomparedwiththe
weakercompetitioninducedinmoreconcentrated
mar-kets,providedtheoutcomesofhospitalsoperatinginmore
and in less concentrated markets would have followed
commontrendsifthepolicyhadnotbeenimplemented.
Fricke(2017)pointsoutthatthestatedcommontrends
assumptionisinsufficienttoidentifythedifferentialeffect
oftreatmentintensities.Iftheassumptiondoeshold,then
the differential effect is identified only under the
fur-therassumptionofhomogeneityintheeffectofmarginal
changes in treatmentacross, in this context, more and
lessconcentratedmarkets.Thishomogeneityassumption
islikelytobeimplausiblegiventhathospitalsarenot
ran-domlyassignedtomarkets.ApplyingFricke’slogictothe
situationstudiedhere,whatcanplausiblybeidentifiedis
alowerbound(inmagnitude)ontheeffectofprice
dereg-ulationinlessconcentratedmarketscomparedwiththe
continuationofpriceregulationinthosemarkets.23Two
assumptionsarenecessarytoobtainthispartial
identifi-cation.First,there mustbecommontrendsacrossmore
andlessconcentratedmarketsunderthecounterfactualof
nopolicychange.Second,thepolicyeffect(relativetono
change) inlessconcentrated marketsmustbethesame
signbutofgreatermagnitudethantheeffectinmore
con-centratedmarkets.24Intuitively,ifpricederegulationhas
anyeffectinmoreconcentratedmarkets,thentakingthe
22Asacheckforfurtherheterogeneityinhospitals’responsestoprice
deregulation,afinercategorizationofhospitalsbyHHIisalsoused.
23Fricke(2017)considersidentificationoftheeffectsoftwo(ormore)
distincttreatmentsusingaDIDstrategy.Thetreatmentscanbe distin-guishedbydegreeofintensity.Toapplythesetuptothisstudy,onemust thinkofthetreatmentsnotaspricederegulation,butratherasthe inten-sityofcompetitionarisingfromthispolicy.Weareinterestedinthe qualityeffectofmoreintensecompetitionandidentifythisby (effec-tively)interactingasinglepolicychange(pricederegulation)withmarket concentrationtoobtainvariationincompetition(treatment)intensity.
24Thatis,monotonicityintreatmentintensity,wheretreatmentisthe
competitivepressureinducedbythepricederegulation.
DIDbetweenmoreandlessconcentratedmarketscannot
pointidentifytheeffectinthelatter.Butitwillgivealower
boundontheeffectinthelessconcentratedmarkets
pro-videdthattheeffectisgreaterinmagnitudebuthasthe
samesignastheeffectinmoreconcentratedmarkets(and
commontrendshold).
Hospitals with an HHI below 2500 form our
treat-ment group, while those with an HHI of at least 2500
belongtothecomparisongroup.Onlyhospitalswithan
HHIeitheralwaysbelow2500oralwaysabove2500during
the2001–2010periodareusedintheanalysis.Hence,no
hospitalcanswitchfromthetreatmentgrouptothe
com-parisongrouporviceversa,andthecompositionofeach
groupisheldconstantbyconstruction.Twenty-one
hospi-talsoutofatotalof103areexcludedbecausetheyfailto
meetthiscriterion.Thisismainlybecauseofmerger
activ-itybetween2001and2010.Oftheremaining82hospitals,
43haveHHI<2500andareinthetreatmentgroup.
Inourmainanalysis,weusedatafrom2001to2004to
capturetheperiodbeforepricederegulationanddatafrom
2006to2010forthepost-reformperiod.Weexcludedata
from2005,asthepolicywasimplementedonFebruary1
ofthatyear.
Weestimatethefollowingfixedeffectsmodelbyleast
squares:
RRht=˛+ı1 (HHIh<2500)×POSTt+Xht
+uh+t+εht (1)
whereRRhtistheunplanned90-dayreadmissionrate
(per-cent)fornon-acutehipreplacementsathospitalhinyear
t,1()istheindicatorfunction,POSTt isabinaryindicator
equalto1forthepost-reformperiod(2006-2010),Xhtisa
vectorofhospitalcharacteristicsthatvaryovertimebutare
plausiblynotaffectedbypricederegulation,uhisahospital
fixedeffect,tisayeareffectandεhtisarandomerrorterm.
ThecovariatesconsistoftheCharlsonindexofcomorbidity
(Quanetal.,2011;Quanetal.,2005)averagedovera
hos-pital’snon-acutehipreplacementpatientsinayear,the
percentageofthesepatientsaged65+,40–60and18–39
years,thepercentagefemale,thepercentagedischargedto
askillednursingfacility,25andthemeanzipcode-specific
socioeconomicscoreofthepatients. Theseindicatorsof
casemixareincludedtoincreaseefficiencyandtoallowfor
anychangeinthecompositionofhipreplacementpatients
thatdiffersbetweenhospitalsin lessand more
concen-tratedmarketswithoutbeingcaused bythedifferential
effectofpricederegulation.Wehavearguedthatthereis
littleornoreasontoexpectthereformtohavecausedhip
replacementpatientstoselectdifferenthospitalsor
hos-pitalstohaveselecteddifferentpatients,andwereturnto
thispointattheendofthissection.
Table1 presentsmeansofthe covariatesbeforeand
afterthereformforthetreatmentandcomparisongroups.
Priortothereform,therearesomesignificantdifferences
inthecharacteristicsofthepatientsacrossthetwogroups.
25 Weassignedpatientsreportedtobedischargedto‘otherhealthcare
organization’toanursinghomedischargesincearecent(unpublished) StatisticsNetherlandsstudyreportsthataround70%ofthesepatientsare transferredtoanursinghomefacility.
Table1
Meansofcovariatesbyperiodandtreatmentgroup.
Patients’characteristics Pre-reform Post-Reform Change(Post-Pre)
Proportiondischargedto skillednursingfacility
ComparisonGroup 0.048 0.059 0.011 {0.061} {0.084} (0.009) TreatmentGroup 0.075 0.087 0.012 {0.096} {0.133} (0.013) Difference(T-C) 0.03*** 0.001 (0.009) (0.016) Proportionfemale ComparisonGroup 0.69 0.68 −0.01* {0.05} {0.06} (0.01) TreatmentGroup 0.72 0.69 −0.03*** {0.06} {0.07} (0.01) Difference(T-C) 0.03*** −0.02* (0.01) (0.01) Meanage ComparisonGroup 69.85 70.36 0.51* {19.23} {22.53} (0.24) TreatmentGroup 70.60 70.50 −0.11 {25.09} {33.50} (0.34) Difference(T-C) 0.75*** −0.61 (0.25) (0.42)
MeanCharlsonScore (comorbidity) ComparisonGroup 0.0018 0.0009 −0.0009* {0.01} {0.00} (0.0005) TreatmentGroup 0.0039 0.0017 −0.0022 {0.02} {0.00} (0.0016) Difference(T-C) 0.0021 −0.0014 (0.0016) (0.0016)
Meansocioeconomicscore
ComparisonGroup 6.81{0.53} 6.88{0.55} 0.06 (0.06) TreatmentGroup 7.46 7.49 0.03 {0.48} {0.48} (0.06) Difference(T-C) 0.65*** −0.04 (0.06) (0.08)
Numberofhospitals ComparisonGroup 39 39 39
TreatmentGroup 43 43 43
Numberofpatients ComparisonGroup 28613 33096 61709
TreatmentGroup 25528 26851 52379
Notes:Pre-/post-reformcellentriesareobtainedbyfirstcomputingthemeanacrossallnon-acutehipreplacementpatientsdischargedfromeachhospital andthentakingthesimpleaverageofthesemeansacrossallhospitalswithinagroupandperiod.Figuresincurlybracketsarestandarddeviationsacross hospitals.Figuresinparenthesesarestandarderrorsoftheestimatedchangeinthemean.Hospitalsandpatientsareselectedusingthecriteriadescribed intheData,sampleselectionsection.Thesocioeconomicscoreisincreasinginsocioeconomicstatusandrangesfrom0to10.TheCharlsonscore(Quan etal.,2011)rangesfrom0to9,withhigherbeingmoresevere.***Significantatthe1percentlevel.**Significantatthe5percentlevel.
Butthedifferencesarerathersmall.Significancereflects
thelargesamplesize.Thetreatmentgrouphasaslightly
higherproportionoffemales,itspatientsareabout1year
older and they have a higher socioeconomic status as
wellasagreater propensitytobeadmittedtoa skilled
nursingfacilityafterdischarge,onaverage.Thereareno
pre-reformdifferences in comorbidity measuredby the
Charlsonindex.Thecharacteristicsofthepatientschange
relativelylittlebetweenthetwoperiodsforbothgroups.
Noneofthedifference-in-differencesofthese
character-isticsaresignificantlydifferentfromzero,indicatingthat
therewasnodifferentialchangeinthecompositionofthe
groupswithrespecttotheseobservables.
Conditionalonthecovariates(Xht),ifintheabsenceof
pricederegulationin2005theaveragereadmissionrateof
hospitalsinlessconcentratedmarketswouldhavechanged
in2006-10 by asmuch as thechangethat would have
occurredinhospitalsoperatinginmoreconcentrated
mar-ketsandifpermittingpricecompetitionhadalargereffect
inlessconcentratedmarkets,thentheparameterıin(1)is
alowerbound(inmagnitude)ontheaverageeffectofthe
pricederegulationinamorecompetitivemarket
environ-mentonthereadmissionrateamongthehospitalsinthe
lessconcentratedmarkets.
Fig.1supportstheplausibilityofthecommontrends
assumption. Between 2001 and 2005, the trend in the
readmissionrate,andindeeditslevel,isverysimilarfor
hospitalsoperatinginmoreandlessconcentratedmarkets.
Estimationofamodelsimilarto(1)usingdatafrom2001to
2004onlyandallowingtheyeareffectstodifferbetween
hospitalslocatedinmore(HHI≥2500) andless
concen-tratedmarketsrevealsnoevidenceofdifferentialtrendsin
theperiodimmediatelyprecedingthereform(Appendix3,
TableA1column(i)).
Agapopensupinthereadmissionratesin2006
imme-diatelyafterthecontractingreform.Thereadmissionrate
fallsinthelessconcentratedmarkets,whileitcontinuesto
riseinthemoreconcentratedmarkets.Takenatfacevalue,
this wouldsuggestthathospitalsthatexperiencedprice
deregulationinamorecompetitiveenvironmentraisedthe
qualityofthecaretheydelivered.However,thedivergence
sim-Fig.1.90-dayreadmissionrateafternon-acutehipreplacement.
Notes:HospitalsandpatientsselectedbycriteriadescribedintheData,sampleselectionsection.HHI(Herfindahl-HirschmanIndex)calculatedwitha radiusof30kmdefiningtheboundaryofamarket.
ilarinmoreandinlessconcentratedmarkets.Overthefull
five-yearpost-reformperiod,thefiguresuggeststhatprice
deregulationinamorecompetitiveenvironmentdidnot
consistentlylowerorraisethequalityofhipreplacements.
MotivatedbyFig.1,andbecausehospitalsandinsurers
maynothavefullyadjustedtothenewcontracting
condi-tionsimmediatelyafterpricesbecamefreelynegotiable,
we estimatea secondmodel that allows thetreatment
effecttovaryinthepost-reformperiod:
RRht=˛+
k=06,10ık1(HHIh<2500)×YEARkt
+Xht+uh+t+εht (2)
whereYEAR06t=1(YEARkt=1)iftheyearis2006(20k).
Underthesameassumptionsaboutcommontrendsand
differentialeffectsinmoreandlessconcentratedmarkets,
␦06givesalowerboundontheaverageeffectofprice
dereg-ulationinamorecompetitivemarketenvironmentin2006,
etc.
Market concentration is generally considered to be
potentially endogenous because performance may feed
backintostructureandunobservableattributesmay
influ-enceboth quality and patientchoice ofhospital (Evans
etal.,1993).Theempiricalstrategyadoptedandthe
insti-tutionalcontextinwhichthisstudyisconductedminimize
the threats to identification from these two potential
sourcesofendogeneity.Hospitalfixedeffectsdealwithany
timeinvariantcorrelatedunobservables.Weavoidusing
timevaryinginformationonmarketconcentrationby
cat-egorizingeachhospitalintooneoftwogroupsaccording
towhetheritsHHIisalwaysbelow2500oralwaysabove
2500.Hospitalsthatcrossthisthresholdovertheperiod
ofanalysisaredropped.26TheHHIarecalculatedfrombed
numbers,ratherthanpatientflows,andsoendogeneityof
26Asnotedearlier,horizontalmergersamonghospitals aremainly
responsibleforthresholdcrossings.
thismeasureofmarketstructuretoperformanceisnota
majorconcern.
Wedeliberatelychooseanelectiveprocedureto
mea-surequalityinorder toobtainevidenceontheeffectof
competitiononatreatmentthatislikelytoexhibitmuch
greaterdemandelasticitywithrespecttopriceand
qual-itythanisthecasewithacutetreatments (e.g.forAMI)
thathavebeenthefocusofpreviousreseach(e.g.Cooper
etal., 2011; Gaynoretal., 2013).Ifgreater competition
doespotentiallyreducequalitybecausedemandismore
responsivetopricethantoquality,thenwewouldexpect
toobservethisforanelectiveprocedure.Therearethree
reasonswhythis studyisnot particularlyvulnerableto
selectionbiasarisingfrompatientchoicedespiteitsfocus
onan electiveprocedure. First,we eliminate correlated
timeinvariantunobservabledifferencesinpatient
compo-sitionacrosshospitalswithfixedeffects.Onlyifthereform
weretochangeunobservablepatientcharacteristics
dif-ferentiallyacrossthetreatment andcomparison groups
wouldtherebeanypotentialbias.Thelackofany
substan-tialorsignificantdifference-in-differencesinobservable
covariates(Table1)suggeststhattheremaybelittlereason
toworryaboutpotentiallycorrelatedtimevarying
unob-servables.Second,aspreviouslymentioned,patientsand
insurerslackedinformationonhospitalquality,including
readmissionratesforhipreplacements,beforeandafter
thereform.Therewaslimitedscopeforselectionon
qual-ity.Third,incontrasttotheUKhealthcaremarketreforms
that have beenthesubject of many previousstudies,27
thereformweexaminedidnotchangeopportunitiesfor
27 Thesestudieseitheruserichdataorintrumentstodealwithtime
vary-ingpatientselection.Skellern(2019)controlsforrisk-adjustedPatient ReportedOutcomeMeasures(PROMs),whileGaynoretal.(2013),Cooper etal.(2011)andMoscellietal.(2016)instrumenthospitalchoiceusing GP/patient-hospitaldistances.Cooperetal.(2011)donotreject exogene-ityofmarketstructureandMoscellietal.(2016)findthatinstrumenting hasverylittleimpactontheestimates,relativetocontrollingforarichset ofpatientcovariates.
Table2
Unplanned90-dayand30-dayhipreplacementreadmissionratesbyperiodandtreatmentgroup.
Outcome Pre-reform Post-Reform Change(Post-Pre)
90-dayreadmissionrate
ComparisonGroup 0.0768 0.0827 0.0059* {0.0316} {0.0290} (0.0036) TreatmentGroup 0.0766 0.0813 0.0047 {0.0321} {0.0326} (0.0037) Difference(T-C) −0.0003 −0.0012 (0.0036) (0.0051)
30-dayreadmissionrate
ComparisonGroup 0.0419 0.0482 0.0062** {0.0227} {0.0200} (0.0025) TreatmentGroup 0.0439 0.0493 0.0054* {0.0237} {0.0239} (0.0027) Difference(T-C) 0.0019 −0.0008 (0.0026) (0.0037)
Numberofhospitals ComparisonGroup 39 39 39
TreatmentGroup 43 43 43
Numberofpatients ComparisonGroup 28613 33096 61709
TreatmentGroup 25528 26851 52379
Notes:Tablegivesthesimplemeanreadmissionrateaveragedoverallhospitalsinthetreatment(HHI<2500)groupandthecomparison(HHI≥>2500) group.Figuresincurlybracketsarestandarddeviationsacrosshospitals.Robuststandarderrorsinparentheses.Hospitalsandpatientsselectedbycriteria describedintheData,sampleselectionsection.*Significantatthe10percentlevel.
patientchoice.Asexplainedabove,patientshaddefacto
freechoiceofproviderbeforeandafterthereform.
Hospital-initiatedselectionofpatientsispotentiallyof
greaterconcern.Afterpriceswerederegulated,hospitals
operatinginmorecompetitivemarketscouldpossiblyhave
hadtheincentivetodrivedowncosts;e.g.bycherrypicking
morestraightforwardcasessothattighterbudgetswould
notimpingeonquality.However,becauseweidentifyfrom
comparisonacrosshospitalscategorizedbybroadranges
ofHHI,anycherrypickingwouldonlybiasourestimates
inthehighlyunlikelysituationthatpatientswereshunted
longdistances.Morelikelyisthatahospitalwouldrefer
apatientwhoisatgreaterriskofreadmissiontoa
neigh-boringhospital,whichwillmostprobablybeinthesame
treatmentorcomparisongroup.So,whilethecasemixof
individualhospitalsmaychangeduetopatientselectionin
responsetothereform,itisratherunlikelythatthiswould
changethecompositionofthegroups,andthecomparisons
inTable1againsupportthis.
6. Results
6.1. Mainestimates
Prior to the reform, there was no difference in the
90-dayreadmissionratebetweenthetreatment(low
mar-ketconcentration,HHI<2500)groupandthecomparison
(highmarket concentration) group(Table2, toppanel).
Thegroupsarebalancedontheoutcomeatbaseline.Post
reform,thereadmissionrateincreased(10%significance)
by0.59 percentage points (pp) or7.8% in the
compari-songroupandbyonlyslightlyless(andnotsignificantly)
inthetreatmentgroup.Consequently, thesimple
(non-parametric)DIDestimateisnegative,whichwouldindicate
thatpricederegulationinacompetitivemarket
environ-mentledtolowerreadmissionrates(i.e.higherquality).
Buttheestimateisverysmallinmagnitude–1.6%ofthe
treatmentgrouppre-reformrate–andnotatallcloseto
beingsignificantlydifferentfromzero.
TheconditionalDIDestimateobtainedfrommodel(1)
andgiveninthefirstcolumnofthetoppanelofTable3is
positive,butitisevensmallerinmagnitudethanthesimple
DIDestimate,anditalsolacksanysignificance.Thefactthat
conditioningonobservablesdoesnotmarkedlychangethe
estimatefurtherindicatesthatthereislikelytobelittlebias
fromcorrelatedtimevaryingunobservables.The
insignif-icant conditionalDIDpoint estimateis only0.7% ofthe
readmissionrateinthetreatmentgrouphospitalspriorto
thereform.Wecanruleoutaneffectgreaterthan11%ofthe
pre-reformreadmissionratewith95%confidence.Subject
totheusualcaveatthatfailuretorejectthenull
hypoth-esisdoesnotnecessarilyimplythatthereisnoeffect,the
magnitudeandprecisionoftheestimatedonotgivecause
tobelievethatpricederegulationinacompetitivemarket
ledtosubstantial,orevenany,deteriorationinthequality
oftreatment.Thisinferenceissubjecttothefurthercaveat
thatourempiricalstrategydeliversonlyalowerbound(in
magnitude)estimateoftheeffectofpricederegulationin
acompetivesetting.
ConsistentwithwhatisobservedinFig.1,the
condi-tionalDIDestimatesinthefirstcolumnofthebottompanel
ofTable3suggestthatpricederegulationinless
concen-tratedmarketsmayhavereducedthe90-dayreadmission
rate byat least1.5 percentage points (19%) in thefirst
year(2006)afterthereformbuthadnoeffectintheyears
thereafter(2007-2010).Asisapparentfromthefigure,the
negativeeffectin2006,whichissignificantonlyatthe10%
level,isdrivenbyafallinthereadmissionrateofthe
hos-pitalsoperatingin lessconcentrated marketsanda rise
inthereadmissionrateofhospitalsinmoreconcentrated
markets.Itwouldbedifficulttoattributethesedivergent
movementstoapositive effectofpricederegulationon
quality.
For all post-reform years after2006, the point
esti-matesarepositive.Overlookingthefactsthatnoneofthe
estimatesareremotelysignificantandtheyaresmallin
magnitude,onemightventuretoexplainthispatternas
arising froman initialpost-reform shock (in 2006)
Table3
Estimatedeffectsofpricederegulationonreadmissionratesafterhip(knee)replacement.
Hipreplacement KneeReplacement
Specification (1) (2) (3) (4) (5)
90-dayreadmission 90-dayreadmission 90-dayreadmission 30-dayreadmission 90-dayreadmission radius30 radius20 radius40 radius30 radius30 Model(1) ␦ 0.0005 0.0063 0.0016 0.0003 0.0007 (0.0048) (0.0055) (0.0057) (0.0037) (0.0153) R2 0.49 0.461 0.458 0.437 0.375 Model(2) ␦6 −0.0147* 0.0007 −0.0056 −0.0059 −0.0315 (0.0076) (0.0097) (0.0093) (0.0054) (0.0199) ␦7 0.0077 0.0089 0.0062 0.0072 0.0080 (0.0086) (0.0099) (0.0098) (0.0064) (0.0249) ␦8 0.0039 0.0085 0.0012 0.0028 0.0260 (0.0085) (0.0101) (0.0097) (0.0066) (0.0215) ␦9 0.0003 0.0018 0.0026 −0.0011 −0.0147 (0.0068) (0.0074) (0.0077) (0.0058) (0.0195) ␦10 0.0081 0.0122 0.0038 −0.0015 0.0220 (0.0074) (0.0087) (0.0088) (0.0058) (0.020) R2 0.4990 0.4630 0.4600 0.4410 0.3880 NTreatmentHospitals 43 13 29 43 34 NComparisonHospitals 39 32 15 39 32 Nhospitalsxyears 738 495 396 738 5974 Npatients 114408 80077 74888 114408 18231 Nreadmittedpatients 9064 6489 6043 5241 1183
Notes:ToppanelgivesOLSestimatesof␦fromregression(1).SecondpanelgivesOLSestimatesof␦6until␦10fromregression(2).Allestimatesobtained fromregressionscontaininghospitalandyearfixedeffectsandcovariatesidentifiedinTable1.FullestimatesinAppendix3;Tables2and3.Robuststandard errorsinparentheses.HospitalsandpatientsselectedbycriteriadescribedintheData,sampleselectionsection.RadiusXindicatesthattheestimatesare basedontreatment/comparisongroupsformedonthebasisofaHHIcalculatedwitharadiusofXkmdefiningtheboundaryofamarket.Thesamplesize fallsastheradiusisreducedbecausemorehospitalscrosstheHHIthresholdof2500usedtodefinethetreatment/comparisongroupsduringtheestimation period.**Significantatthe5percentlevel.*Significantatthe10percentlevel.
hospitalsthatareexposedtomorecompetitivemarkets
areoperatingwithunregulatedprices.Inconsistentwith
thishypothesis,thepointestimatesdonotmonotonically
increaseinmagnitudeastimesincethereformlengthens.
Still,thelargest(buthighlyinsignificant)pointestimate
isinthelastofthepost-reformyearsanditis10%ofthe
pre-reformreadmissionrateofthetreatmentgroup.This
placesafinalcaveatontheconclusionthattherewasno
negativeimpactonquality.
6.2. Robustness
6.2.1. Marketdefinition
ThemainestimatesaregeneratedonthebasisofHHIs
calculatedwitharadiusof30kmusedtodefinethe
bound-aryofthemarketaroundahospital.Tocheckrobustness,
we recalculatetheHHI using a radiusof 20 and 40km
todefineamarket,recategorizehospitalsintothe
treat-mentandcomparisonsgroupsonthebasisoftherevised
indexandthenre-estimatemodels(1)and(2)ineachcase.
Estimatesaregivenintheappropriatelylabelledcolumns
of Table3.Witha radiusof30km, hospitalsareevenly
splitbetweenthetreatmentandcomparisongroups.Asthe
radiusiswidened,morehospitalsgetputintothe
treat-ment groupbecausetheHHI decreasesastheareathat
definesthemarketincreases.
Irrespective of the radius used, thetreatment effect
averagedoverthefiveyearsofthepost-reformperiodis
insignificant.Whentheradiusisincreased,thepoint
esti-mateoftheeffectin2006(frommodel(2))continuesto
benegativebutsignificanceislost.28Whentheradiusis
reduced,thispointestimateturnspositiveandagainitloses
significance.Themarginallysignificantnegativepoint
esti-mateobtainedfor2006usingthebaselineradiusof30km
isclearlynotrobust.Thereisnosignificanteffectafter2007
irrespectiveofthegeographic radiususedtodefinethe
market.Overall,irrespectiveoftheradiususedtodefine
a hospital market,there is noclearevidence that price
deregulationinamorecompetitivemarketenvironment
consistentlyimpactedonthereadmissionrate.
One might be concerned that exposure to different
degreesofmarketconcentrationisconfoundedby
differ-ences(inresponsetothereform)betweenurbanandrural
locations.Thisis unlikelytobea well-founded concern
in thecontext of the Netherlands, which is the second
mostdenselypopulatedcountryintheOECDandoneof
themost urbanized.29 Highlydeveloped and integrated
transportnetworksfurtherlimitthescopeforanymarked
and consequential urban-rural division in the country.
Onlythreehospitalsinoursamplearenotwithina30km
radiusofacitywithatleast50,000inhabitants,andonly
28 Extendingtheradiusfurtherto50kmproducesthesamegeneral
pat-tern:nosignificanteffectaverageoverallpost-reformyearsandanegative butnotatallsignificantpointestimatefor2006.SeeAppendix3TableA2 andA3,column(iv).
29 DutchpopulationdensityisalmosttwicethatoftheUKandmore
than14timesgreaterthanthatoftheUS(IndexMundi,2020).Morethan 75percentofthepopulationlivesinpredominantlyurbanregions(OECD, 2018).
oneofthesethreehospitalsremainsinoursampleafter
selectinghospitalsthatarepersistentlyinthesamegroup
definedbyHHIaboveorbelow2500.30Theestimatesare
robusttoexcludingthis hospital(seeAppendix3, Table
A2andA3,column(v)).
6.2.2. Curtailedpost-reformperiod
In 2008, there were a number of policy changes in
the Dutch hospital industry. As mentioned above, the
HealthCareInspectoratemadeanunsuccessfulattemptto
developandimplementacomprehensivesetofuniform
quality indicators across hospitals. Free price
negotia-tionwasextendedtomoreproceduresthatbroughtthe
share of hospital revenues derived from products with
unregulatedpricesto20%.Themethodofpaying
special-istsalsochangedin2008,suchthattheirannualincome
becamecompletelyactivity-based,whichgeneratedstrong
incentivestoincreaseproduction(SchutandVarkevisser,
2013).Theirremunerationrates,however,werestill
reg-ulatedand,therefore,notaffectedbytheoutcomeofthe
insurer-hospitalpricenegotiations.Entrytothehospital
marketbecameeasierfromthatyearbecausegovernment
approval forthe construction ofnew hospitalbuildings
(oradditionalcapacity)wasnolongerrequired.Thiswas
accompaniedbyagradualincreaseinthefinancialriskfor
hospitalsbecausereimbursementofcapitalcostswasno
longer(fully)assured.
Although the year-specific estimates in the bottom
panelofTable3givenoindicationofanysubstantialchange
from2008thatmaybeduetoconfounding,weexamine
therobustnessofourfindingstocurtailingthepost-reform
periodto2006–2007.Doingsogivesaninsignificant
neg-ativepointestimateoftheeffectaveragedoverthesetwo
yearsthatissmallinmagnitude.31 Thisgives noreason
tosuspectthataneffectofpricederegulationonthehip
replacementreadmissionrateisbeingconfoundedbyother
policychangesthatoccurredwithinourperiodofanalysis.
6.2.3. 30-dayreadmissionrate
Sincetheoreticalandempiricalgroundsfor
unambigu-ouslypreferringthe90-daytothe30-dayreadmissionrate
asanindicatorofthequalityofcarearelacking,wecheck
robustnesstousingtheshorterperiod.Pre-reformtrendsin
30-dayreadmissionratesarereasonablyparallelbetween
thetreatmentandcomparisongroups,althoughthereis
somedivergencein2004(seeFig.2).Thehypothesisthat
yeareffectsinthe30-dayreadmissionrateareequalfor
thetreatmentandcomparisongrouphospitalsinthe
pre-treatmentperiod isnot rejected(Appendix3, TableA1;
column(ii)),whichlendsplausibilitytothecommontrends
identificationassumptionforthisoutcomealso.32
Theestimatedeffect onthe30-dayreadmission rate
overthefullpost-reformperiodgiveninthetoppanelof
column(4)ofTable3isabouthalfthemagnitudeofthe
30 AllthreehospitalsareinthehighHHIgroup.
31 Theseresultsareavailablefromtheauthorsuponrequest. 32 Forbothgroups,thetrendsinthe30-dayratedisplaygreatervolatility
thanthoseforthe90-dayrate,whichisduetothesubstantiallylowerrate ofreadmissionsovertheshorterperiod.Thisgreaternoiseinthe30-day rateisonegoodreasonforrelyingmoreonthe90-dayrate.
baselineestimatedeffectonthe90-dayrateincolumn(1),
whichisduetothelowermeanrateofreadmissionsover
theshorterperiod.Theestimatedeffectonthe30-dayrate
isalsonotatallsignificant,andsotherecontinuestobeno
evidencethatpricederegulationinmorecompetitive
mar-ketsaffectedthequalityofcare.Theyear-specificestimates
arenegativein3ofthe5post-reformyears,buttheyare
alwayssmallinmagnitudeandneverclosetosignificance.
6.2.4. Kneereplacements
Tofurtherassessthecredibilityofourfindingofanull
effect,wereplicatetheanalysisforanindicatorofthe
qual-ityofasecondelectiveprocedure–readmissionafterknee
replacement.Withinthesubsetofproductsforwhichfree
pricenegotiationwasallowedfrom2005,knee
replace-ment is theonlyotherprocedurewithsufficientlyhigh
volumeandforwhichareliablequalityindicatorcanbe
constructed.Kneereplacementsandhipreplacementsare
commonlyperformedbythesamespecialty,whichmakes
thisalocaltestofrobustness.Theupsideisthatconsistency
acrossthetwoprocedureswouldlendalotofcredibilityto
theevaluationoftheeffectonthequalityofthetreatment
performedbyaparticularspeciality.We usethe90-day
readmissionrateasthequalityindicator.
Thereadmissionrateafterkneereplacementdisplays
greatervariationfromyeartoyearthantherateforhip
replacements(see AppendixFig.A1),which reflectsthe
muchsmallernumberofpatientsundergoingknee
replace-ment.Wecannotrejectthatthetimeeffectsarecommon
betweenthetreatmentandcomparisongroupsinthe
pre-reformperiod(seeAppendixTableA1,column(iii)).The
estimatesoftheeffectonthekneereplacement
readmis-sionratearehighlyconsistentwiththosefortheeffecton
readmissionafterhipreplacement(Table3.Column(5)).
Theestimatedeffectoverthefullpost-reformperiod(top
panel) isveryclosein (small)magnitudetothe
respec-tiveestimateforhipreplacements,onlyitislessprecise
reflectingthesmallersample.Aswithhipreplacements,
theestimatedeffectisnegativein2006.Thereafter, itis
positive,exceptfor2009.Overalltheyearsandforeach
year,theestimatedeffectisneverclosetosignificance.
Irre-spectiveofwhetherreadmissionrateafterhipreplacement
orafterkneereplacementisusedtoindicatequality,there
isnoevidencethatpricederegulationinamore
competi-tivemarketaffectedthequalityofcare.
6.3. Heterogeneity
WhiletheHHIthresholdof2500isbasedonUSantitrust
guidelines,itissomewhatarbitrary,andevenmoresoin
a European context. Further,a binaryclassification will
miss any variation in the response to the intensity of
competitivepressureinducedbypricederegulationacross
finerdegreesofmarketconcentration.Thenulleffectwe
findcouldpossiblyarisefromdifferentialresponsesatthe
extremesofthemarketconcentrationdistributionbeing
diluted by similar responses either side, but closer to,
the2500threshold.Totestforthis,wecategorize