Table Of ContentContractNo.: 500-94-0062
MPRReferenceNo.: 8246
PreliminaryReport:The
ImpactofProspective
PaymentonMedicare
HomeHealthUse-
PromisingResultsfor
aFutureProgram
July22,1998
ValerieCheh
ChristopherTrenholm
RandallBrown
BarbaraPhillips
Submittedto: Submittedby:
HealthCareFinancingAdministration MathematicaPolicyResearch,Inc.
7500SecurityBlvd,(C-3-21-06) P.O.Box2393
Baltimore,MD 21244-1850 Princeton,NJ 08543-2393
(609)799-3535
ProjectOfficer:
AnnMeadow ProjectDirector:
ValerieCheh
CONTENTS
Chapter Pag^
EXECUTIVESUMMARY
ix
I THEPER-EPISODEHOMEHEALTHDEMONSTRATION
ANDEVALUATION
1
A. THEMEDICAREHOMEHEALTHBENEFIT 2
B. THEPER-EPISODEDEMONSTRATION 4
1. DemonstrationPaymentandIncentives 6
2. OtherDemonstrationProcedures 11
C. COMPONENTSOFANDAPPROACHTOTHEEVALUATION 13
1. AnalysisofAgencyDecisionsandOperations 14
2. AnalysisofProgramImpacts 16
D. GUIDETOTHERESTOFTHISREPORT 19
II DATAANDMETHODOLOGY 21
A. DATA 21
1. IdentifyingEpisodes 22
2. MedicareClaimsforOutcomesVariables 23
3. ControlVariables 23
4. AnalysisSample 29
5. SummaryStatisticsforControlVariables 30
B. METHODS 35
1. StatisticalModelsforEstimatingOverallImpacts 36
2. EstimatingSubgroupImpacts 38
3. HypothesisTestsfortheImpactEstimates 40
4. Weighting 41
5. DesignEffects 42
6. RobustnessChecks 43
iii
CONTENTS(continued)
Chapter Page
III IMPACTSONTHEUSEOFSERVICESINTHE
FIRST120DAYS 47
A. EXPECTEDEFFECTSOFPROSPECTIVEPAYMENT 47
B. DIDPROSPECTIVEPAYMENTAFFECTTHENUMBER
OFVISITSRENDERED? 48
C. DIDPROSPECTIVEPAYMENTAFFECTTHEEPISODE
COMPOSITION? 53
D. DIDPROSPECTIVEPAYMENTAFFECTTHEEPISODE
LENGTH? 56
E. DIDPROSPECTIVEPAYMENTAFFECTAGENCY
SUBGROUPSDIFFERENTLY? 60
F. ROBUSTNESSOFESTIMATEDIMPACTS 69
1. ComparisonofRegression-AdjustedandUnadjusted
DemonstrationImpacts 69
2. WeightingAgenciesbyShareofEpisodes 72
3. EffectsofOutliers 74
4. EffectsofCensoring 75
5. InfluenceofUnobservedVariables 75
6. SummaryofRobustnessChecks > 78
IV SUMMARYANDCONCLUSIONS 79
A. KEYFINDINGS ....79
B. POLICYIMPLICATIONS 80
C. LIMITATIONSOFTHEANALYSIS 82
D. CONCLUSION 84
REFERENCES 85
APPENDIXA: DATAQUALITY 87
APPENDIXB COEFFICIENTESTIMATESFROMTHE
: REGRESSIONONTOTALVISITSPEREPISODE 95
APPENDIXC CASEMIXADJUSTMENTSDURING
: THEDEMONSTRATION 101
iv
TABLES
Table Page
II.1 OUTCOMEVARIABLESDESCRIBINGMEDICARESERVICE
USEDURINGTHE120-DAYAT-RISKPERIOD 24
11.2 CONTROLVARIABLESFORMULTIVARIATEANALYSIS,
BYSOURCE 25
11.3 WEIGHTEDMEANSFOREXPLANATORYVARIABLESBY
TREATMENTSTATUS,ANDTESTSFORDIFFERENCESIN
TREATMENTANDCONTROLGROUPMEANS 31
III.1 THEIMPACTOFPER-EPISODEPAYMENTONTHENUMBEROF
VISITSINFIRST120DAYS 50
111.2 DECOMPOSITIONOFTHEIMPACTOFPER-EPISODEPAYMENT
ONTHENUMBEROFVISITSINFIRST120DAYS 52
111.3 THEIMPACTOFPER-EPISODEPAYMENTONTHETYPES
OFCARERENDEREDINFIRST120DAYS 54
111.4 THEIMPACTOFPER-EPISODEPAYMENTONEPISODELENGTH 58
111.5 IMPACTOFPER-EPISODEPAYMENTONTHEUSEOF
SERVICES,BYWHETHERTHEAGENCYISFOR-PROFIT
ORNONPROFIT 63
111.6 IMPACTOFPER-EPISODEPAYMENTONTHEUSEOF
SERVICES,BYWHETHERTHEAGENCYHADAHIGH-USE
ORLOW-USEPRIORPRACTICEPATTERN 64
111.7 IMPACTOFPER-EPISODEPAYMENTONTHEUSEOF
SERVICES,BYWHETHERTHEAGENCYISSMALLOR
LARGESIZE 65
111.8 IMPACTOFPER-EPISODEPAYMENTONTHEUSEOF
SERVICES,BYWHETHERTHEAGENCYISHOSPITAL-
BASEDORFREESTANDING 66
111.9 UNADJUSTEDESTIMATESOFTHEIMPACTOFPER-EPISODE
PAYMENTONTHENUMBEROFVISITSINFIRST120DAYS 70
V
TABLES(continued)
Table Page
III.1 ROBUSTNESSCHECKSFORTHEPRINCIPALFINDINGSONTHE
IMPACTOFPER-EPISODERATESETTINGDURINGTHEFIRST
120DAYS 73
vi
FIGURES
Figure Page
III.1 EPISODELENGTHBYTREATMENTSTATUS(AGENCIES
WEIGHTEDEQUALLY) 59
III.2 DISTRIBUTIONOFTHEPERCENTAGECHANGEINTHENUMBER
OFVISITSPEREPISODE 77
vii
EXECUTIVESUMMARY
Aspartofitsongoingefforttostudymethodsofprovidingmorecost-effectivecare,theHealth
CareFinancingAdministration(HCFA)hasrecentlyimplementedthePer-EpisodeHomeHealth
ProspectivePaymentDemonstration. Underthedemonstration,participatinghomehealthagencies
arepaidafixed,lump-sumpaymentforthefirst 120daysofeachepisodeofcareprovidedto
Medicare beneficiaries and a predetermined rate for each visit thereafter. This method of
compensationdifferssubstantiallyfromthecurrentmethodofMedicarereimbursementforhome
healthservices,underwhichagenciesarereimbursedforactualcostsincurred,uptoaspecificlimit.
Byallowingagenciestoretainmostofanysurpluspaymentsovercost,prospectivepaymentgives
agenciesafinancialincentivetoprovidehomehealthcareinamorecost-efficientmannerthan
undertraditionalcost-basedreimbursement.
Ninety-oneagenciesinfivestatesenteredthethree-yeardemonstrationatthestartoftheir1996
fiscalyear. Priortothestartofthedemonstration,theparticipatingagencieswererandomlyassigned
toeitherthetreatmentgroup(whichispaidunderthedemonstration'sprospectivepaymentmethod)
oracontrolgroup(whichcontinuestobepaidunderMedicare'snormalmethodofcost-based
reimbursement). Thepaymentstreatmentgroupagenciesreceiveforthefirst120daysofapatient
episodearebasedoneachagency'sowncostsinthefiscalyearimmediatelyprecedingitsentryinto
thedemonstration,adjustedforchangesinitscasemix. Whileeachagencyis"atrisk"duringthe
first 120daysafteradmissionforallhomehealthvisitsthepatientneeds,HCFAreimburses
treatmentagenciesforupto99percentoffiscal-yearlossesuptotheSection223paymentlimits.'
ProfitsinexcessofspecifiedlimitsmustbesharedwithHCFA.
RESEARCHQUESTIONSANDMETHODOLOGY
Inthisreport,weexaminetheavailabledatafrom(roughly)thefirstyearofthedemonstration
totesthypothesesaboutthepossibleeffectsofprospectivepaymentonpatients'useofMedicare-
coveredservices.Giventhelimiteddatacurrentlyavailable,thispreliminaryreportfocusesonlyon
homehealthusetakingplaceduringthe"at-risk"period(first120days)ofahomehealthepisode.
Asmoredatabecomeavailable,futurereportswillprovideamorecompleteinvestigationof
demonstrationimpactsonhomehealthuseandotheroutcomes.Here,wetesthypothesesconcerning
theimpactsofthedemonstrationon(1)thenumberofvisitsprovided,bothintotalandbytype;(2)
theper-episodemixofservicesprovided;and(3)thelengthofan episodeofcare. Inaddition,we
alsotestwhethertheseoutcomesdifferedbetweensubgroupsofagenciesdefinedbytheirfor-profit
status,size,andotherkeycharacteristics.
Theanalysisisbasedonapproximately51,000homehealthepisodestakingplacein85ofthe
demonstrationagencies(6ofthe91 agencieswereexcludedbecausetheydroppedoutofthe
'TheSection223paymentlimitsarecost-per-visitpaymentlimitsthatapplytoallagenciesin
theMedicareprogram.
ix
demonstrationorhadinadequatedata). Alladmissionsoccurringbetweenanagency'sstartdateand
August 1996areincluded. Medicareclaimsfilesprovidedthedataontheoutcomevariables
describingtheuseofservicesduringthefirst120daysofhomehealthepisodes. Datacollectedat
admission for case-mix adjustment and from prior Medicare claims provided measures of
preadmissioncharacteristicsofpatientsadmittedtoagenciesparticipatinginthedemonstration.
Dataonagencycharacteristicswereobtainedfromtheagencycostreportsandthedemonstration
implementationcontractor.
Ordinaryleastsquaresmodelsandlogisticmodelswereusedtoestimateprogrameffects,
controllingforpreexistingdifferencesbetweentreatmentandcontrolagenciesinpatientandagency
characteristics. Thisapproachprovedcrucialtoobtainingvalidimpactestimatessince,despitethe
randomizationofparticipatingagencies,therewereseveralsignificantdifferencesbetweentreatment
andcontrolagenciesasidefromthemethodofpayment. Observationsareweightedsothateach
agencyisrepresentedequallyintheanalysis. Standarderrorsofimpactestimateswerecalculated
usingspecialsoftwaredesignedtoaccountfortheeffectsofsampleclusteringandweighting,soas
toavoidoverstatingtheprecisionoftheestimates. Analysesoftherobustnessofourregression
estimatesshowedthattheywerenotsensitivetotheweightingscheme,statisticalmethods,ormodel
specificationsused.
FINDINGS
VisitsperEpisodeFell17Percent
Wefindstrongevidencethatprospectivepaymentreducedthetotalnumberofvisitsper
episode,withlargeandstatisticallysignificanteffectsonskillednursing,homehealthaide,and
medicalsocialworkervisits.Inaddition,wefindrelativelylargebutstatistically insignificant
declinesinthenumberofvisitsbyphysicalandoccupationaltherapists. Overall,weestimatethat
prospectivepaymentledtreatmentagenciestoprovideanaverageof7.8fewervisitsthancontrol
agenciesduringthefirst120daysofeachepisode,adeclineof17percentrelativetothemeanfor
controlgroupagencies(seeFigure1). Thisincludesanestimatedper-episodereductionofabout4
skillednursingvisits(down18percent),2.7homehealthaidevisits(dovm18percent),0.6physical
therapyvisits(down 10percent),and0.3medicalsocialworkervisits(down37percent). The
effectsonskillednursingandaidevisitsdominatetheeffectsonoverallvisitsbecausetheyaccount
for83percentofallvisitsduringthefirst120daysofanepisode.
X
FIGURE1
AVERAGENUMBEROFVISITSPEREPISODE
BYTREATMENTANDCONTROLAGENCIES
NumberofVisits
45
TotalVisits SkilledNursing HomeHealth
Visits AideVisits
Treatment Control
Source:MedicareClaimsData.
Note: Treatmentgroupmeanisadjustedtoaccountfor
preexistingdifferencesbetweentreatmentandcontrol
agencies.
ThePercentReceivingOccupationalTherapyDeclined
Theproportionofpatientsreceivingoccupationaltherapydeclinedbyaboutone-thirdfromthe
controlgroupmeanof12.6percent,fallingby4.4percentagepoints. However,theproportion
receivingotherserviceswasrelativelyunaffected. Thereductioninoccupationaltherapymaybe
duetofewerpatientsreceivingassessmentvisitsfromoccupationaltherapists. Theprogrameffect
onthereceiptofanyskillednursingvisits,adropof2.6percentagepoints,wasalsostatistically
significant,butsmall. Thisreductionmaybeduetoagencieshavingsomecasesopenedbyphysical
therapists,whentherapyistheprimaryservicethepatientneeds. Fortheotherservices-aidecare
(receivedby46percentofpatients),physicaltherapy(42percent),speechtherapy(3percent),and
medicalsocialworkervisits(29percent)"therewerenodiscernibleeffects.
xi
ProspectivePaymentHadNoMajorEffectontheMixofVisitsProvided
Prospective paymentappeared to have no effectontheproportion ofvisitsperepisode
accountedforbyanyparticularspeciaUy. Estimatedeffectsontheproportionsofepisodevisitsthat
wereforskillednursing,therapies(physical,occupation,andspeechcombined),aidecare,and
medicalsocialserviceswereeachsmallandstatisticallyinsignificant,suggestingthattheimpacts
ofthedemonstrationfellroughlyproportionallyacrossthemajorvisittypes.
EpisodesWereShortenedby14Percent
Consistentwiththeimpactsonthenumberofvisitsprovided,prospectivepaymentreducedthe
lengthofepisodeswithintheat-riskperiodbyabout10days,or14percentofthemeanforcontrol
groupagencies(70days). Moreover,prospectivepaymentreducedtheprobabilitythatanepisode
exceeds120daysbyabout30percent(10percentagepoints)relativetothecontrolgroupmeanof
35percent. Thiseffect,whichaccountsforthemajorityofthetotalreductioninaverageepisode
length,suggeststhattheproportionoflonghomehealthepisodesmaybesignificantlyreduced. The
findingalsosuggeststhat,whenserviceuseduringtheperiodafter120daysisexamined,thetotal
reductioninvisitsduetoprospectivepaymentmaybesubstantiallylarger. However,itwillalsobe
necessarytoassesswhethertreatmentgrouppatientshavemorereadmissions.
ProgramImpactsVariedLittlewithAgencyCharacteristics
Wetestedwhetherimpactsonthefollowingoutcomesvariedwithagencycharacteristics: (1)
numberofvisitsperepisode,(2)theproportionofvisitsthatwereskilled nursingvisits,(3)the
proportionofvisitsthatwereaidevisits,and(4)episodelength. Agencycharacteristicsusedto
define subgroups included whether agencies were for-profit, their size, whether they were
freestandingorhospital-based,andtheirpredemonstrationpracticepattern(thatis,whetherthe
agency provided more or less than the average number ofvisits per episode during the
predemonstrationyear,forthecasemixserved).
Wefindonlyoneimportantdifferenceinimpactsacrosstheagencysubgroups. Theimpacton
totalvisitswassignificantlygreaterforagencieswithhigh-usepriorpracticepatternsthanforthose
withlow-usepatterns. High-useagenciescutvisitsperepisodebynearly11visits(about20percent
ofthecontrolgroupmean),comparedtoadropofonlyabout5visits(14percent)foragencieswith
low-usepracticepatterns.
Agencysizeandauspices(freestandingorhospital-based)havenoapparentinfluenceonthe
estimatedeffectsforanyoftheoutcomesexamined,buteffectsonthemixofvisitsmaydifferby
agencies' for-profitstatus. For-profitagenciessignificantlyreducedtheproportionofskilled
nursingvisitsperepisode,butnonprofitagenciesdidnot. Effectsonotheroutcomesweresimilar
forfor-profitandnonprofitagencies. Giventhatfor-profitagenciestendedtohavehigher-use
practicepatternsinthepredemonstrationperiod,programeffectsonvisitsperepisodemaybegreater
onaverageforfor-profitagenciesthanfornonprofitones.
xii