how do hospitals respond to price changes? files/16_dafny_how do hos… · 6 bruce steinwald and...

How Do Hospitals Respond to Price Changes

By LEEMORE S DAFNY

This paper examines hospital responses to changes in diagnosis-specific prices byexploiting a 1988 policy reform that generated large price changes for 43 percentof Medicare admissions I find hospitals responded primarily by ldquoupcodingrdquo pa-tients to diagnosis codes with the largest price increases This response wasparticularly strong among for-profit hospitals I find little evidence hospitals in-creased the volume of admissions differentially for diagnoses subject to the largestprice increases despite the financial incentive to do so Neither did they increaseintensity or quality of care in these diagnoses suggesting hospitals do not competefor patients at the diagnosis level (JEL H0 I0 L0)

The vast majority of US healthcare is pri-vately provided Yet until the 1980s the sectorwas largely immune from standard marketforces promoting efficiency in production Thecanonical healthcare market imperfectionsmdashinformational asymmetries between providersand consumers and an insurance-induced wedgebetween marginal out-of-pocket costs and pa-tient benefitsmdashwere exacerbated by a cost-plus-reimbursement system and primarily not-for-profit providers So long as providers couldalways earn nonnegative profits there was littlesupply-side incentive to cut costs and consum-ersrsquo incentives via co-payments and deductibleswere weak

In 1984 the federal government injectedmarket discipline into the system by implement-ing the Prospective Payment System (PPS) Un-der PPS hospitals receive a fixed payment foreach Medicare patient in a given diagnosis-related group (DRG) regardless of the actual

expenses the hospital incurs in caring for thepatient1 Other public and private insurers sub-sequently implemented similar payment schemeswresting price-setting control from providersand imposing ldquoyardstick competitionrdquo in whichproviders are rewarded based on performancerelative to the average (Andrei Shleifer 1985)

A large literature documents hospitalsrsquo re-sponses to the introduction of PPS but fewstudies have explored reactions to changes inthese prices Yet once the transition to a fixed-price regime is complete price levels constitutethe sole lever in the system and there remainseveral unanswered empirical questions regard-ing their effect In the face of a price increasefor a particular diagnosis or treatment will hos-pitals find ways to attract more such patientsWill they compete more vigorously for thesepatients by improving the quality of their carethereby dissipating some of the rents from theprice increase Will they game the system andshift patients to higher-paying diagnoses with-out altering any aspects of their care The an-swers to these questions are critical to ongoingpolicy decisions and can also provide valuableinsights into hospital industry conduct and theeffectiveness of fixed-price regulation moregenerally

The main challenge in studying these ques-tions is locating an exogenous source of varia-

Kellogg School of Management Northwestern Univer-sity 2001 Sheridan Road Evanston IL 60208 NationalBureau of Economic Research and Institute for PolicyResearch at Northwestern University (e-mail l-dafnykelloggnorthwesternedu) I am grateful to Jonathan Gru-ber James Poterba David M Cutler and Glenn Ellison forexcellent guidance I thank two anonymous referees JoshAngrist William Collins Joseph Doyle David DranoveMark Duggan David Levine Joseph Newhouse ScottStern Nancy Rose and seminar participants at severaluniversities for helpful comments Financial support fromthe National Science Foundation the National Bureau ofEconomic Research and the National Institute on Aging isgratefully acknowledged The MEDPAR data are confiden-tial and cannot be released

1 Hospitals receive additional payments for admissionswhose costs exceed the annual ldquooutlier thresholdrdquo set by theCenters for Medicare and Medicaid Services Total outlierpayments comprise roughly 8 percent of payments underPPS (68 Federal Register 34122)

1525

tion in DRG prices which are typically adjustedto reflect changes in hospital costs As a resultpositive associations between changes in priceand changes in spending or intensity of treat-ment likely reflect bilateral causality and do notconstitute a priori evidence that hospitals altertreatment patterns in response to price changesTo obtain unbiased estimates of hospital re-sponses to price changes this study exploits a1988 policy change that generated large pricechanges for 43 percent of Medicare admis-sions2 The policy change was the eliminationof ldquoage over 69rdquo and ldquoage under 70rdquo in thedescriptions of the DRGs to which patients maybe assigned Qualifiers that formerly read ldquowithcomplications or age over 69rdquo and ldquowithoutcomplications and age under 69rdquo now readldquowith complicationsrdquo or ldquowithout complica-tionsrdquo This seemingly innocuous modificationwhich is described in greater detail in Section Iactually led to substantial changes in prices forDRGs with these qualifiers

I consider both ldquonominalrdquo and ldquorealrdquo re-sponses to these price changes where nominalrefers to hospital coding practices and real re-fers to admissions volumes and intensity of careactually provided Because hospitals are respon-sible for coding patients to the appropriateDRGs raising prices for certain DRGs mayentice hospitals to ldquoupcoderdquo or switch patientsfrom lower-paying to higher-paying DRGsWhile upcoding does not affect real elements ofpatient care it inflates hospital reimbursementsThis was the primary response of hospitals tothe 1988 policy change Hospitals also demon-strated a keen awareness of risk-reward trade-offs in their upcoding which can lead to finesand criminal charges if detected Although thepolicy shock created a blanket incentive to in-crease upcoding in dozens of diagnoses hospi-tals upcoded most in those diagnoses where theincentive to do so was greatest The upcodingresponse was also strongest among for-profithospitals a finding that is consistent with priorresearch

The real responses to the policy change wereless finely tuned I find little evidence that hos-

pitals adjusted the intensity or quality of theircare in response to changes in diagnosis-specificprices where intensity is measured by totalcosts length of stay number of surgical proce-dures number of intensive-care-unit (ICU)days and quality by the in-hospital death rateRather hospitals spread the additional fundsacross all admissions I also find no conclusiveevidence that hospitals attracted more patientsin diagnoses subject to larger price increases

The remainder of the paper is organized intosix sections Section I describes PPS and priorrelated research and lays out the hypotheses Itest Section II offers a detailed explanation of the1988 reclassification followed by a discussionof the data in Section III Sections IV and Vexamine the nominal and real responses to pricechanges respectively and Section VI concludes

I Background

A A PPS Primer

The defining element of PPS is a reimburse-ment amount that is fixed regardless of a hos-pitalrsquos actual expenditures on a patient Thispayment does vary however by the patientrsquosmedical diagnosis Diagnoses are grouped intoapproximately 500 DRGs Each DRG is as-signed a weight (called a ldquoDRG weightrdquo) thatreflects the relative resource intensity of admis-sions within that group Reimbursement to hos-pital h for an admission in DRG d is given by

(1) Phd Ph 1 IMEh

1 DSHh DRG weightd

where Ph is a hospital-specific amount (inflatedannually by a congressionally approved ldquoupdatefactorrdquo) IMEh represents an adjustment for in-direct medical education (teaching) and DSHhadjusts payment levels to compensate hospitalswith a disproportionate share of indigent patients3

Most of the variation in Phd is due to theDRG weights which range between 009 (DRG448 for allergic reactions) to 228 (DRG 480 forliver transplants)4 The Health Care Financing

2 In 2000 Medicare beneficiaries accounted for 37 per-cent of hospital discharges and 31 percent of total revenues(Data Compendium Centers for Medicare and MedicaidServices and authorrsquos tabulations from the 2000 Survey ofHospitals by the American Hospital Association (AHA))

3 This simplified formula appears in David M Cutler(1995)

4 The range for DRG weights is given for 1985ndash1996

1526 THE AMERICAN ECONOMIC REVIEW DECEMBER 2005

Administration (HCFA) uses hospital cost data(obtained by deflating charges using annualcostcharge ratios) to recalibrate the weightsannually raising weights for DRGs that expe-rience relative increases in average costs andreducing weights for DRGs with relative de-creases in average costs5 The average DRGweight per hospital admission has risen substan-tially over time from 113 in 1984 to 136 in19966 This phenomenon has been termedldquoDRG creeprdquo as patients are increasingly codedinto DRGs with higher weights A 1-percentincrease in the average case weight is associatedwith an additional $930 million in annual Medi-care payments to hospitals7 To reduce the costof DRG creep HCFA often sets the annualupdate factor below the average growth in costs

Although the implementation of PPS elimi-nated marginal reimbursement for services ren-dered (within a given DRG hospitals are notcompensated more when they spend more on apatient) economists have noted that averagepayment incentives remain If Phd is low rela-tive to actual costs in DRG d hospitals have anincentive to reduce the intensity of care and thenumber of admissions in that DRG Due to theregular recalibrations described above it is dif-ficult to identify hospital responses to changesin average payment incentives (hereafter DRGprices or weights) When costs increase DRGprices increase Thus the coefficient onDRG price in a regression of costs (or someother measure of intensity of care) on DRGprice would suffer from a strong upward biasTo obtain an unbiased estimate of this coeffi-cient exogenous variation in payment levels isrequired This variation is provided by the nat-ural experiment described in Section II

B Hypotheses and Prior Research

Because more than 80 percent of hospitals arenot-for-profit or government-owned economists

typically assume that hospitals maximize an ob-jective function with nonnegative weights on pa-tient care (often called ldquointensityrdquo or quality) andprofits (see for example David Dranove 1988Burton A Weisbrod 2005) Although not-for-profit and government hospitals are also subjectto a nondistribution constraint limiting theirability to distribute profits to stakeholders un-der most scenarios they too will pursueprofit-increasing activities in order to financetheir missions Therefore a price increase for aparticular diagnosis constitutes an incentive toldquoproducerdquo more such diagnoses

Under PPS there are two principal means forproducing more coding existing patients intothe diagnosis subject to the price increase andattracting additional patients in that diagnosisthrough quality or intensity improvements(loosely defined to include better care patientamenities stronger referral networks etc)8 Irefer to these as ldquonominalrdquo and ldquorealrdquo re-sponses respectively as the former are essen-tially accounting tricks and the latter are realaspects of treatment

Nominal ResponsesmdashThe coding of patientconditions is performed by administrative staffwho use specialized software hospital chartsand diagnosis codes provided by physicians tomap patient conditions into DRGs Hospitalscan pursue a variety of approaches to facilitateupcoding into higher-paying DRGs First phy-sicians often rely on administrative or nursingstaff to identify diagnosis codes and these staffmembers can be trained to err on the side ofmore lucrative diagnoses Second hospitalsmay try to persuade physicians to alter theirdiagnoses in order to increase hospital reve-nues9 Finally administrative staff may falsifypatient records andor assign DRGs incorrectly

5 HCFA is now known as The Centers for Medicare andMedicaid Services However I refer to HCFA throughoutthis paper as this was the agencyrsquos acronym during thestudy period

6 Bruce Steinwald and Laura A Dummit (1989) au-thorrsquos calculations The original 1984 weights were con-structed so that the average DRG weight for hospitalscalled the case-mix index would equal 1

7 ldquoProgram Informationrdquo Centers for Medicare andMedicaid Services June 2002

8 Note Medicare beneficiaries face the same out-of-pocket costs at all hospitals for covered stays hence hospi-tals cannot reduce prices to attract these patients Theexception is Medicare HMO enrollees who accounted forfewer than 3 percent of Medicare beneficiaries during thestudy period (Lauren A Murray and Franklin J Eppig2002) Reimbursement for these patients is not determinedby PPS and co-payments may differ across hospitals

9 For example one resident I interviewed described aninteraction with hospital coding personnel in which she wasasked to reconsider her diagnosis of ldquourinary tract infec-tionrdquo and replace it with ldquosepticemiardquo (which occurs whenthe bacteria enter the bloodstream) as the hospital is

1527VOL 95 NO 5 DAFNY HOW DO HOSPITALS RESPOND TO PRICE CHANGES

to inflate reimbursements Hospitals engagingin these practices face substantial penalties ifaudits reveal systematic efforts to defraud theMedicare program10

When deciding whether to upcode a particu-lar patient or group of patients hospitals tradeoff the added revenue (less any change in treat-ment costs) against the increased risk of detec-tion plus the cost of upcoding In its purestform upcoding implies no effect whatsoever onthe amount of care received by patients sotreatment costs are unchanged Ceteris paribusa price increase for a given DRG thereforeincreases the incentive to upcode patients intothat DRG The policy change I study involvesDRGs with particularly low upcoding costs andrisks of detection In these DRGs simply cod-ing complications such as hypotension (lowblood pressure) onto a patientrsquos chart results ina substantially higher price although the pri-mary diagnosis remains the same (Comparethis scenario to upcoding a patient sufferingfrom bronchitis to the heart transplant DRG)Section IV examines how increases in the pricespaid for complications affected the reported in-cidence of complications

The subject of upcoding has generated a sub-stantial literature not least because the rapidrise in the average case weight is the singlelargest source of increased hospital spending byMedicare Another concern is that the strategyused to reduce the resulting financial burden(smaller increases in the annual update factor)punishes all providers uniformly regardless ofthe extent to which they upcode In additionupcoding may result in adverse health conse-quences due to corrupted medical records Rob-ert F Coulam and Gary L Gaumer (1991)review the upcoding literature through 1990concluding that there is evidence of upcodingduring the first few years of PPS but theamount of the case-mix increase attributable tothis practice is unknown There are two general

empirical approaches to estimating the magni-tude of upcoding detailed chart review andcomparisons of case-mix trends over time andacross hospitals

Grace M Carter et al (1990) created theldquogold standardrdquo in chart review to estimate therole of upcoding in the case-mix increase be-tween 1986 and 1987 they sent a nationallyrepresentative sample of discharge records from1986 and 1987 to an expert coding group (calledthe ldquoSuperPROrdquo) which regularly reviews sam-ples of discharges to enforce coding accuracyThey find that one-third of the case-mix in-crease was due to upcoding although the stan-dard error of this estimate is large Morerecently Bruce M Psaty et al (1999) useddetailed chart review to estimate that upcodingis responsible for over one-third of admissionsassigned to the heart failure DRG (DRG 127)

Most of the nonmedical analyses of case-mixincreases (eg Steinwald and Dummit 1989)are descriptive focusing on which types of hos-pitals exhibit faster case-mix growth (large ur-ban and teaching hospitals) and when theseincreases occur (there is a big jump in the firstyear a hospital is paid under PPS) Becausethese studies use data from the transition toPPS the results are difficult to interpret patientseverity changed dramatically due to changes inpatient composition following the implementa-tion of PPS

A recent study by Elaine M Silverman andJonathan S Skinner (2004) presents strong ev-idence of post-transition-era upcoding for pneu-monia and respiratory infections between 1989and 1996 Focusing on the share of patients withthese diagnoses who are assigned to the mostexpensive DRG possible Silverman and Skin-ner document large increases in upcoding de-spite a downward trend in mortality rates Theauthors find that for-profit hospitals upcode themost and not-for-profit hospitals are morelikely to engage in upcoding when area marketshare of for-profit hospitals is higher This find-ing suggests that practices of competitors mayaffect upcoding indirectly through pressure onhospital profits or directly via the disseminationof upcoding practices11 Silverman and Skinner

ldquounderpaidrdquo and ldquoneeds the funds to provide care for theuninsuredrdquo

10 Agencies known as ldquoPeer Review Organizationsrdquo reg-ularly audit DRG assignments CMS works with the Officeof the Inspector General the Federal Bureau of Investiga-tions and the US Attorneyrsquos Office to levy fines recoverfunds and prosecute providers who defraud the Medicareprogram The qui tam provision of the Federal Civil FalseClaims Act protects and rewards whistle-blowers

11 Several recent studies document this indirect channeleg Mark G Duggan (2002) who finds that not-for-profithospitals respond more strongly to financial incentives to


also find that hospitals under financial distressupcode less than financially sound institutions

The upcoding analysis I perform improvesupon prior research in three ways First I ex-ploit a discrete change in upcoding incentiveswhich enables me to cleanly separate upcodingfrom time trends that may be associated withomitted factors such as the severity of patientsrsquoconditions Second I exploit differences acrossdiagnoses in the magnitude of the change inupcoding incentives to see whether hospitalsrespond to upcoding incentives on the marginupcoding even more when the payoff is greaterThird I expand the scale of previous work byanalyzing nearly 7 million Medicare admissionsto 186 DRGs and 5352 hospitals nationwide

In addition to estimating upcoding responsesfor all hospitals financed under PPS I estimateresponses separately for different subsamples ofhospitals Due to heterogeneity in objectivefunctions and endowments hospitals may re-spond differently to the same price changes Forexample hospitals placing a higher weight onprofits should upcode more while hospitals fac-ing a greater penalty (real or perceived mone-tary or otherwise) or a higher probability ofdetection should upcode less All things equalhospitals experiencing financial distress shouldbe more willing to risk detection Size may alsoenter into upcoding decisions as larger hospi-tals can spread the costs of training their codingpersonnel across more admissions Finallythere are important regional differences in hos-pital behavior although there are few theoreti-cal explanations for this phenomenon apartfrom ldquocultural normsrdquo In Section IV B I strat-ify the hospital sample by a variety of thesecharacteristics in order to explore potential dif-ferences in responses

Real ResponsesmdashBecause Medicare patientsface the same out-of-pocket costs for coveredstays at all hospitals they are free to select theirpreferred provider and may consider such fac-tors as location quality of care and physicianrecommendations12 In order to attract patientsin diagnoses subject to price increases hospitals

may increase the intensity or quality of treat-ment (hereafter ldquointensityrdquo) provided to suchpatients In Section V I estimate the elasticityof intensity with respect to price I instrumentfor price using price changes that were exog-enously or ldquomechanicallyrdquo imposed by the newpolicy independently of hospitalsrsquo upcodingresponses

The few studies that investigate the effect ofDRG-level price changes on intensity levels allfind a positive relationship where intensity ismeasured by length of stay number of surgicalprocedures andor death rates (Cutler 19901995 Boyd H Gilman 2000)13 Thus all evi-dence to date suggests that a ldquoflypaper effectrdquooperates in the hospital industry additional in-come is allocated to the clinical area in which itis earned rather than spread across a broadrange of activities However because all ofthese studies utilize data from a transition to aprospective payment system they face the for-midable challenge of separating two simulta-neous changes in incentives the elimination ofmarginal reimbursement and changes in theaverage level of payments for each DRG

Cutler (1990) examines the transition to PPSin Massachusetts finding that length of stay andnumber of procedures per patient declined themost in DRGs subject to the largest price re-ductions Despite finding an elasticity of inten-sity with respect to price of 02 Cutler does notfind corresponding increases in volume Cutler(1995) studies the impact of PPS on adversemedical outcomes again finding an intensityresponse reductions in average price levels areassociated with a compression of mortality ratesinto the immediate post-discharge period al-though there is no change in mortality at oneyear post-discharge Both papers assume thateliminating the marginal reimbursement incen-tive affected all DRGs equally when in factthere was substantial variation in the degree ofldquofat to trimrdquo To the extent that price reductions

treat indigent patients in markets with greater for-profitpenetration

12 Co-payments may vary across hospitals for Medicarebeneficiaries enrolled in HMOs However fewer than 3

percent of Medicare beneficiaries were enrolled in HMOsduring the study period

13 Most prior research on real responses examines theeffect of hospital-wide financial shocks on overall qualityas measured by the average length of stay and inpatientmortality rates These studies which generally exploitMedicare shocks to hospital finances find a positive effectof reimbursement on quality (eg Jack Hadley et al 1989Douglas Staiger and Gaumer 1992)


were correlated with this excess (the very goalof the price-setting process) the intensity re-sponses to price changes will be overstated Moregenerally the elasticity estimate will be biased byany omitted factor influencing both price and in-tensity changes during the transition to PPS14 Thesame critique pertains to Gilman (2000) who in-vestigates the impact of a 1994 reform to Medic-aid DRGs for HIV diagnoses in New York

By investigating responses to changes in av-erage payment levels (ie prices) in the post-implementation period I circumvent both thischallenge and the concern that transitory re-sponses are driving previous results I also ex-plore the effect of price on DRG volume Ifhospitals increase intensity and patient demandis elastic with respect to this intensity admis-sions volumes should rise in DRGs with priceincreases For example if the price for prosta-tectomy rises and hospitals invest in nerve-sparing surgical techniques the number ofpatients electing to undergo this procedure islikely to increase Alternatively hospitals mayoffer free screening tests for prostate cancer andincrease demand through the detection of newcases (a common practice)15

As with upcoding responses there are manyreasons that real responses may differ acrosshospitals For example for-profit hospitals may

respond more to the financial incentive to attractpatients in lucrative DRGs Differences acrossDRGs are another possible source of variationin responses Patient demand for planned orelective admissions may be more sensitive tochanges in intensity than demand for urgentcare When a hospitalization is anticipated apatient can ldquoshop aroundrdquo soliciting advice andinformation directly from the hospital as wellas from physicians and friends The elasticity ofdemand with respect to quality might thereforebe larger for such admissions raising hospitalsrsquoincentives to increase quality in the face of priceincreases I explore differences in intensity andvolume responses across hospitals and admis-sion types in Section V B

II A Price Shock The Elimination of the AgeCriterion

Although there were 473 individual DRGcodes in 1987 40 percent of these codes be-longed to a ldquopairrdquo of codes that shared the samemain diagnosis Within each pair the codeswere distinguished by age restrictions and pres-ence of complications (CC) For example thedescription for DRG 138 was ldquocardiac arrhyth-mia and conduction disorders age 69 andorCCrdquo while that for DRG 139 was ldquocardiacarrhythmia and conduction disorders withoutCCrdquo Accordingly the DRG weight for the topcode in each pair exceeded that for the bottomcode There were 95 such pairs of codes and283 ldquosinglerdquo codes

In 1987 separate analyses by HCFA and theProspective Payment Assessment Commission(ProPAC) revealed that ldquoin all but a few casesgrouping patients who are over 69 with the CCpatients is inappropriaterdquo (52 Federal Register18877)16 The ProPAC analysis found that hos-pital charges for uncomplicated patients over 69were only 4 percent higher than for uncompli-cated patients under 70 while average chargesfor patients with a CC were 30 percent higherthan for patients without a CC In order tominimize the variation in resource intensitywithin DRGs and to reimburse hospitals moreaccurately for these diagnoses HCFA elimi-

14 Cutlerrsquos methodology for calculating the change inaverage payment incentives following the implementationof PPS likely yields upward-biased elasticity estimatesCutler defines the change in average price as the differencebetween the 1988 PPS price and the price that Medicarewould have paid in 1988 were cost-plus reimbursement stillin effect To estimate this latter figure he inflates 1984 costsfor each DRG by the overall cost-growth rate for 55- to64-year-olds However DRGs with disproportionatelystronger cost growth between 1984 and 1988 receivedweight increases yielding higher 1988 PPS prices and gen-erating the concern that the positive relationship betweenprice changes and intensity levels may be spurious Thepossibility that these estimated price changes are not exog-enous is reinforced by the use of hospital-specific prices inthe specifications The average price changes are thereforerelated to hospitalsrsquo pre-PPS DRG-specific costs hospitalswith high costs faced price reductions when transitioning tonational payment standards These hospitals may have hadldquomore fat to trimrdquo in terms of intensity provision

15 To my knowledge there are no prior studies thatexamine the effect of DRG prices on volume There iscompelling evidence however that inpatient utilization ingeneral increases with insurance coverage which affectsboth out-of-pocket expenses and hospital reimbursements(see Frank R Lichtenberg 2002 on the effects of Medicareand Dafny and Jonathan Gruber 2005 on Medicaid)

16 ProPAC now incorporated into MedPAC (MedicarePayment Advisory Commission) was an independent fed-eral agency that reported to Congress on all PPS matters


nated the age over 69under 70 criterion begin-ning in fiscal year (FY) 198817 The agencyrecalibrated the weights for all DRGs to reflectthe new classification system This recalibrationresulted in large increases in the weights for topcodes within DRG pairs and moderate declinesfor bottom codes

Table 1 lists the three most commonly codedpairs and their DRG weights before and afterthe policy change18 These examples are fairlyrepresentative of the change overall Using1987 admissions from a 20-percent sample ofMedicare discharge data as weights theweighted average increase in the top code for allDRG pairs was 113 percent while the weightedaverage decrease in the bottom code was 62percent This increase in the ldquospreadrdquo betweenthe weights for the top and bottom codes in eachpair strengthened the incentive for hospitals tocode complications on patientsrsquo records In Sec-tion IV I exploit variation across DRG pairs inthe magnitude of these increases to examinewhether hospitals responded to differences inupcoding incentives

To estimate the elasticity of intensity withrespect to price I exploit recalibration errorsmade by HCFA when adjusting the weights toreflect the new classifications As the analysis inSection V reveals even if hospitals had notreported higher complication rates followingthe policy change the average price for admis-sions to DRG pairs would have risen by at least26 percent on average This ldquomechanical ef-fectrdquo varied across DRG pairs as well rangingfrom a reduction of $1232 to an increase of$3090 per admission I investigate whetherhospitals adjusted their intensity of care andadmissions volumes in response to these exog-enous price changes

HCFA subsequently acknowledged that mis-takes in their recalibrations had led to higherDRG weights In 1989 the agency published an(unfortunately flawed) estimate of the contribu-tion of recalibration mistakes to the large in-crease in average DRG weight between 1986and 1988 (54 Federal Register 169) HCFAconcluded that 093 percentage points could beattributed to faulty recalibration of DRGweights for 1988 and an additional 029 per-centage points to errors in 198719 These

17 HCFArsquos fiscal year begins in October of the precedingcalendar year

18 The large volume increase for the bottom code in eachpair is due to the new requirement that uncomplicated patientsover 69 be switched from the top to the bottom code

19 The original notice for the policy change states thatthe goal of the recalibration is to ensure no overall

TABLE 1mdashEXAMPLES OF POLICY CHANGE

DRGcode

Description in 1987(Description in 1988)

1987weight

1988weight

Percent changein weight

1987 volume(20-percent

sample)


sample)Percent change

in volume

96 Bronchitis and asthma age 69 andor CC(bronchitis and asthma age 17 withCC)

08446 09804 16 44989 42314 6

97 Bronchitis and asthma age 18ndash69 withoutCC (bronchitis and asthma age 17without CC)

07091 07151 1 4611 10512 128

138 Cardiac arrhythmia and conduction disordersage 69 andor CC (cardiac arrhythmiaand conduction disorders with CC)

08136 08535 5 45080 35233 22

139 Cardiac arrhythmia and conduction disordersage 70 without CC (cardiac arrhythmiaand conduction disorders without CC)

06514 05912 9 4182 16829 302

296 Nutritional and misc metabolic disordersage 69 andor CC (nutritional andmisc metabolic disorders age 17 withCC)

08271 09259 12 45903 38805 15

297 Nutritional and misc metabolic disordersage 18ndash69 without CC (nutritional andmisc metabolic disorders age 17without CC)

06984 05791 17 2033 12363 508

Notes Of the 95 DRG pairs these three occur most frequently in the 1987 20-percent MedPAR sample DRG weights are from the FederalRegister


estimates motivated an across-the-board reduc-tion of 122 percent in all DRG weights begin-ning in 1990 Because this reduction applieduniformly to all DRGs the relative effectsacross DRG pairs were unabated

III Data

My primary data sources are the 20-percentMedicare Provider Analysis and Review (Med-PAR) files (FY85ndashFY91) the annual tables ofDRG weights published in the Federal Register(FY85ndashFY91) the Medicare Cost Reports(FY85ndashFY91) and the Annual Survey of Hos-pitals by the American Hospital Association(1987) The MedPAR files contain data on allhospitalizations of Medicare enrollees includ-ing select patient demographics DRG codemeasures of intensity of care (eg length ofstay and number of surgeries) and hospitalidentification number20 The data span the threeyears before and after the policy change

The MedPAR discharge records are matchedto DRG weights from the Federal Register andhospital characteristics from the Annual Surveyof Hospitals and the Medicare Cost Reports for1987 the year preceding the policy change21

Due to the poor quality of hospital financialdata the debtasset ratio from the MedicareCost Reports is among the best measures offinancial distress I also construct two additionalfinancial distress measures Medicare ldquobiterdquo(the fraction of a hospitalrsquos discharges reim-bursed by Medicare) and Medicaid ldquobiterdquo (sim-ilarly defined) Table A1 in the Appendixpresents descriptive statistics for these meas-ures together with other hospital characteristicsthat may be associated with responses to theshock (ownership status region teaching statusnumber of beds and service offerings) Becauseprice varies at the hospital and DRG level the

individual discharge records are aggregated toform DRG-year or hospital-year cells Descrip-tive statistics for these cells are reported inTable 2

IV The Nominal Response More DRG Creep

Although DRG creep was known to be apervasive problem by 1987 HCFArsquos policychange nevertheless increased the reward forupcoding The increase in prices for the topcodes in DRG pairs together with the decreasein prices for the bottom codes provided a strongincentive to continue using the top code for allolder patients (not just those with CC) and touse it more frequently for younger patientsFigure 1 charts the overall fraction of patients inDRG pairs assigned to top codes between 1985and 1991 broken down by age group The shareof older patients in top codes falls sharply fromnearly 100 percent in 1985ndash1987 to 70 percentin 1988 and creeps steadily upward thereafterThe trend in complications between 1988 and1991 exactly matches that for young patientswhose complication rate increases steadilythroughout the study period with the exceptionof a plateau between 1987 and 1988

Figure 1 suggests that time-series identifica-tion is insufficient for examining the upcodingresponse to the policy change While it is pos-sible that the increase in the share of youngpatients coded with complications between1988 and 1991 is a lagged response to the policychange it could also be a continuation of apreexisting trend For older patients the data donot reveal what share of admissions were codedwith complications prior to the policy changeso it is impossible to discern whether there wasan abrupt increase in the reported complicationrate Furthermore upcoding among the old islikely to be even greater than upcoding amongthe young as hospitals with sharp increases inthe complication rate of older patients can arguethat they failed to code these complications inthe pre-1988 era because there was no payofffor doing so22 Fortunately differences across

change in reimbursement to hospitals that is the averagenational DRG weight should have been constant whetherthe 1987 or the 1988 classification system (called theGROUPER program) was employed on a given set ofdischarge records

20 The data include all categories of eligibles but ex-clude admissions to enrollees in Medicare HMOs (see foot-note 8)

21 The Cost Reports also contain an indicator for whethera hospital is paid under the PPS system I omit exempthospitals from my sample For hospitals with missing AHAdata in 1987 I use data from the nearest year possible

22 HCFArsquos audit agencies should have had access to thecomplication rates for older patients throughout the studyperiod simple manipulations of the GROUPER programsfrom 1985ndash1987 could produce these rates in the pre-periodUnfortunately these programs have reportedly been erasedfrom HCFArsquos records


DRG pairs in the policy-induced incentive toupcode as measured by changes in ldquospreadrdquoenable me to calculate lower-bound estimates ofthe upcoding response in both age groups

A Aggregate Analysis

The dependent variable for this analysis isfractionpt the share of admissions to pair p inyear t that is assigned to the top code in thatpair The independent variable of interestspreadpt is defined as

(2) spreadpt DRG weight in top codept

DRG weight in bottom codept

eg spreadDRG 1381391988 weightDRG 1381988 weightDRG 1391988 08535 05912 02623spreadpt is simply a measure of the upcodingincentive in pair p at time t Between 1987and 1988 mean spread increased by 020 approx-imately $87523 However there was substantial

variation in spread changes across pairs asdemonstrated by the frequency distribution ofspreadp88-87 in Figure 2 Due to the recalibra-tion procedure the largest spread increases oc-curred in DRG pairs in which complications areparticularly costly to treat andor in which theshare of older uncomplicated patients is high

To examine the effect of the change in spreadon fraction I estimate the specification

(3) fractionpt pairp yeart

spreadp88-87 post pt

where p indexes DRG pairs t indexes yearspost is an indicator for the years following thepolicy change (1988ndash1991) and the dimensionsof the coefficient vectors are (1 93) (1 6) and (1 1)24 captures the averageimpact of the policy reform on all pairs while captures the marginal effect of differential up-coding incentives 0 signifies that hospitals

23 This dollar amount is based on Ph for urban hospitalsin 2001 which was $4376

24 Of the 95 DRG pairs in 1991 two pairs are droppedbecause the age criterion was eliminated one year early forthese pairs

TABLE 2mdashDESCRIPTIVE STATISTICS

Unit of observation

DRG pair-year Hospital-year

N Mean SD N Mean SD

Price (DRG weight) 650 112 062 36651 127 (019)Admissions per cell 650 10624 (15013) 36651 373 (389)Nominal responses

Fraction(young) in top code 650 066 (014)Fraction(old) in top code 650 085 (015)

Real responsesMean cost ($) 650 9489 (6230) 36169 12272 (5692)Mean LOS (days) 650 937 (332) 36651 881 (221)Mean surgeries 650 115 (069) 35897 121 (055)Mean ICU days 650 051 (065) 28226 081 (059)Death rate 650 006 (006) 34992 006 (002)Mean admissions 650 31806 (25822) 36651 778 (538)

Instrumentsspread 650 020 (016)spread post 650 012 (016) ln(Laspeyres price) 650 003 (006) ln(Laspeyres price) post 650 001 (005)Share CC 36651 009 (003)Share CC post 36651 005 (005)

Notes The table reports weighted means and standard deviations for DRG pair-year cells and hospital-year cells which areconstructed by aggregating individual-level data in the 20-percent MedPAR sample Cells are weighted by the number ofindividual admissions in the 20-percent MedPAR sample with the exception of admissions per cell Costs for all years areconverted to 2001 dollars using the CPI for hospital services


FIGURE 1 FRACTION OF ADMISSIONS IN TOP CODES BY AGE GROUP

Notes Sample includes all admissions to DRG pairs in hospitals financed under PPSAuthorrsquos tabulations from the 20-percent MedPAR sample

FIGURE 2 DISTRIBUTION OF CHANGES IN SPREAD 1987ndash1988

Notes Spread is the difference in DRG weights for the top and bottom codes in a DRG pairChanges in spread are converted to 2001 dollars using the base price for urban hospitals in2001 ($4376)


upcoded more in pairs where the incentive to doso increased more25

The estimation results for equation (3) arereported separately by age group in Table 3 Inall specifications observations are weighted bythe number of admissions in their respectivepair-year cells Using grouped data producesconservative standard errors which I furthercorrect for heteroskedasticity For older pa-tients I include fraction(young)p87 post as anestimate of the underlying complication rate ineach DRG pair where fraction(young)pt refersto the fraction of young patients assigned to thetop code in pair p and year t This seems areasonable baseline assumption given a corre-lation coefficient of 094 for young and oldcomplication rates in the post period

The coefficient estimates reveal that upcod-ing is sensitive to changes in spread even aftercontrolling (via the year dummies) for the largeaverage increase in spread between 1987 and1988 Thus the decline in fraction for old pa-tients was least in those pairs subject to the larg-est increase in spread while the increase infraction for young patients was greatest in thosepairs subject to the largest increase in spreadAs hypothesized the upcoding response wasgreater for older patients the coefficient esti-mates imply a spread-induced increase of 0022in the fraction of old patients coded with CC ascompared to 0015 for younger patients

Figures 3A and 3B illustrate these responsesfor young and old patients respectively bygraphing fraction for pairs in the top and bottomquartiles of spread changes Interestingly theaggregate plateau in fraction for young patientsbetween 1987 and 1988 appears to be com-prised of an increase in fraction for those codeswith large spread increases and a decrease infraction for those codes with small spread in-creases A plausible explanation is that hospitalswere concerned that a rise in complications forall young patients would attract the attention ofregulators so they chose to upcode in thosediagnoses with the greatest payoff and down-code in those with the least payoff Fig-ure 3B shows a similar response for older pa-tients the post-shock fraction was higher in thetop quartile of spread increases even though the

pre-shock fraction for young patients in theseDRGs (the ldquobaselinerdquo) was lower

The main threat to the validity of this analysisis the possibility that changes in spread arecorrelated with omitted factors which may bedriving the observed changes in fraction iethat the changes in spread are not exogenousTo help rule out this possibility I regress thechange in fractionpt for young patients between1985ndash1986 1986ndash1987 and 1987ndash1988 onspreadd88-87 and DRG fixed effects I findcoefficients (robust standard errors) of 0002(0013) 0017 (0015) and 0084 (0034) re-spectively These results indicate that thechanges in spread between 1987 and 1988are not correlated with preexisting trends infractionpt As another robustness check I rees-timated equation (3) including individual DRGtrends The coefficient estimates on spread

25 An alternative specification using spreadp88-87 postas an instrument for spreadpt yields similar results

TABLE 3mdashEFFECT OF POLICY CHANGE ON UPCODING

(N 650)

Fraction(young)mean 066

Fraction(old)mean 085

spread88-87 post 0077 0108(0016) (0015)

Fraction(young)87 post 0731(0020)

Year dummies1986 0044 0000

(0008) (0005)1987 0077 0011

(0008) (0005)1988 0058 0813

(0011) (0014)1989 0097 0780

(0009) (0014)1990 0115 0764

(0009) (0014)1991 0128 0748

(0010) (0014)Adj R-squared 0948 0960

Notes The table reports estimates of the effect of changes inspread between 1987 and 1988 on the fraction of patientscoded with complications (equation 3 in the text) ldquoPostrdquo isan indicator for the post-1987 period ldquoyoungrdquo refers toMedicare beneficiaries under 70 and ldquooldrdquo refers to bene-ficiaries aged 70 Regressions include fixed effects forDRG pairs The weighted mean of the dependent variable isreported at the top of each column The unit of observationis the DRG pair-year All observations are weighted by thenumber of admissions in the 20-percent MedPAR sampleThe sum of the weights is 19 million (young) and 50million (old) Robust standard errors are reported in paren-theses Signifies p 005 signifies p 001 sig-nifies p 0001


FIGURE 3 FRACTION OF ADMISSIONS IN TOP CODES BY AGE GROUP AND QUARTILE OF

SPREAD CHANGE

Notes DRG pairs experiencing spread changes under $598 are in quartile 1 DRG pairsexperiencing spread changes over $1375 are in quartile 4


change little and both remain statistically sig-nificant with p 005 Thus the identifyingassumption is that changes in fraction andspread are not correlated with an omitted factorthat first appeared in 1988

The spread-related upcoding alone translatesinto a price increase of 07 percent and 09percent for young and old patients admitted toDRG pairs respectively26 The estimate foryoung patients rises to 15 percent if the jumpbetween 1987 and 1989 is included althoughthis is an upper-bound estimate due to the po-tential role of confounding factors Given aver-age operating margins of 1 to 2 percent in thehospital industry these figures are large Theestimates imply increased annual payments of$330 to $425 million27 This range is very con-servative as it excludes the effect of any aver-age increase in upcoding of older patients acrossall DRG pairs and older patients account for 70percent of Medicare admissions

HCFArsquos 1990 across-the-board reduction inDRG weights decreased annual payments by$113 billion However while this reductionaffected all hospitals equally the rewards fromupcoding accrued only to those hospitals engag-ing in it The following section investigates therelationship between hospital characteristicsand upcoding responses

B Hospital Analysis

To determine whether individual hospitalsresponded differently to the changes in upcod-ing incentives I estimate equation (3) sepa-rately for subsets of hospitals For example Icompare the results obtained using data solelyfrom teaching hospitals with those obtained us-ing the sample of nonteaching hospitals I alsoconsider stratifications by ownership type (for-profit not-for-profit government) financial sta-tus region size and market-level Herfindahlindex28 Hospitals with missing data for any

of these characteristics are omitted from thisanalysis

Table 4 presents estimates of and byhospital ownership type financial status andregion Figure 4 plots the from these specifi-cations For both young and old patients thereare no statistically significant differences in theresponse to spread across the hospital groupsThe discussion here therefore focuses on theresults for young patients for whom the yearcoefficients are relevant

The main finding is that for-profit hospitalsupcoded more than government or not-for-profithospitals following the 1988 reform Figure 4Aillustrates that upcoding trends were the samefor all three ownership types until 1987 butthereafter the trend for for-profits diverged sub-stantially By 1991 the fraction of young pa-tients with complications had risen by 018 infor-profit hospitals compared with 013 forthe other two groups Given a universal mean of065 in 1987 these figures are extremely largeFor-profitsrsquo choice to globally code more pa-tients with complications rather than to manifestgreater sensitivity to spread is consistent with thereward system implemented by the nationrsquos larg-est for-profit hospital chain ColumbiaHCA (nowHCA) A former employee reported that hospitalmanagers were specifically rewarded for upcodingpatients into the more-remunerative ldquowith compli-cationsrdquo codes (Lucette Lagnado 1997)

Hospitals with high debtasset ratios (Figure4B) and hospitals in the South (Figure 4C) alsoexhibited very large increases in fraction(young)although the graphs illustrate that these trendspredated the policy change Moreover thestrong presence of for-profits in the South andthe tendency of for-profits to be highly lever-aged suggests that for-profit ownership is drivingthe large fraction gains in these subsamples aswell All other hospital characteristics were notassociated with changes in upcoding proclivity

V Real Responses

To investigate real responses to price in-creases I create a dataset of mean intensitylevels and price for each DRG pair and year

26 These estimates are calculated using the averagespread in 1988 (045) together with the average weights forDRG pairs in 1987 (105 for young patients 113 for olderpatients)

27 Dollar figures are calculated using PPS expendituresin 2000

28 The Herfindahl index is calculated as the sum ofsquared market shares for all hospitals within a healthservice area as defined by the National Health Planning and

Resources Development Act of 1974 (and reported in theAHA data)


(N 650) Descriptive statistics for these vari-ables are in Table 2 Although the eliminationof the age criterion resulted in large pricechanges for individual DRGs it would not beinformative to investigate whether intensity lev-els rose (fell) for patients admitted to the top(bottom) code of DRG pairs because the com-position of patients admitted to each codechanged as a result of the policy reform Topcodes which were formerly assigned to allolder patients as well as to young patients withCC are now intended to be used exclusively forpatients with CC young or old A finding thataverage intensity of care increased in top codeswould not reveal whether hospitals increasedintensity of care for patients with CC the onlypatients for whom a price increase was enactedFurthermore policy-induced upcoding frombottom to top codes exacerbates the problem ofcompositional changes within each DRG code29 Itherefore combine data from the top and bottom

codes in order to keep the reference populationconstant before and after the policy reform

To instrument for price I exploit differencesacross DRG pairs in the magnitude of recalibra-tion mistakes made by HCFA I isolate thismechanical effect of the policy change becausehospitals may respond differently to exogenousprice changes than to the endogenous pricechanges produced via upcoding (Recall that pureupcoding implies no change in patient care)


To estimate the mechanical effect of the pol-icy change I create a Laspeyres price index for1988 using the 1987 volumes of young patientsin each code as the weights eg

(4) Laspeyres priceDRG 1381391988

priceDRG 1381988 NDRG 1381987


NDRG 1381987 NDRG 1391987

This fixed-weight index approximates the aver-age price hospitals would have earned for an

29 If the sample were restricted to patients under 70 thefirst of these compositional problems would not applyHowever the second would bias any intensity responseestimated using individual DRGs as the unit of observation

TABLE 4mdashEFFECT OF POLICY CHANGE ON UPCODING OF YOUNG BY HOSPITAL CHARACTERISTICS

(N 650)

By ownership type By financial state By region

For-profitNot-for-

profit Government DistressedNot

distressed Northeast Midwest South West

spread88-87 post 0071 0080 0058 0082 0074 0083 0062 0082 0079(0024) (0017) (0018) (0109) (0017) (0016) (0018) (0017) (0024)

Year fixed effects1986 0038 0047 0040 0059 0041 0095 0027 0033 0035

(0009) (0009) (0008) (0008) (0009) (0008) (0009) (0009) (0012)1987 0081 0077 0078 0099 0073 0123 0054 0078 0052

(0010) (0008) (0008) (0008) (0008) (0007) (0009) (0008) (0012)1988 0080 0055 0065 0083 0054 0104 0036 0063 0024

(0012) (0011) (0012) (0011) (0011) (0010) (0010) (0012) (0014)1989 0140 0094 0104 0127 0094 0131 0075 0111 0067

(0011) (0009) (0009) (0009) (0009) (0009) (0010) (0009) (0012)1990 0147 0114 0120 0144 0112 0148 0092 0132 0080

(0011) (0009) (0010) (0009) (0010) (0008) (0009) (0010) (0013)1991 0179 0123 0136 0161 0124 0159 0103 0148 0091

(0011) (0011) (0010) (0010) (0010) (0010) (0011) (0010) (0014)89 87 0059 0016 0027 0027 0022 0007 0021 0033 0015

(0010) (0009) (0008) (0009) (0009) (0008) (0010) (0008) (0014)Adj R-squared 0914 0946 0927 0933 0947 0939 0940 0940 0915

Notes The table reports estimates of equation (3) in the text separately for subsets of hospitals The specification is analogous to that used incolumn 1 Table 3 Regressions include fixed effects for DRG pairs ldquoYoungrdquo refers to Medicare beneficiaries under 70 ldquoDistressedrdquo denoteshospitals with 1987 debtasset ratios at the seventy-fifth percentile or above The unit of observation is the DRG pair-year All observations areweighted by the number of admissions in the 20-percent MedPAR sample Hospitals with missing values for any of the hospital characteristicsare dropped The sum of the weights is 145 million Robust standard errors are reported in parentheses Signifies p 005 signifies p 001 signifies p 0001


admission to a given DRG pair in 1988 had thefraction of patients with CC remained constantat the 1987 fraction for young patients (Be-cause the fraction of old patients with CC can-not be ascertained in 1987 the fraction foryoung patients must proxy for this measure)The mechanical effect of the policy change canthen be estimated as the difference between theLaspeyres price in 1988 and the actual price in1987 Figure 5 graphs the distribution of thesechanges translated into 2001 dollars The meanchange is $102 per admission with a standarddeviation of $448 and an interquartile range of($131) to $262 These figures are small relativeto the average price per admission ($4376) butlarge relative to average operating margins forhospitals Note also that this formula underesti-mates mechanical price changes because theshare of old patients with CC is likely to begreater than that of young patients

Assuming the measurement error inln(Laspeyres price) is small its coefficientin the following first-stage regression shouldequal 1

(5) ln pricept pairp yeart

1lnLaspeyres pricep88-87 post

pairp year pt

1 may differ from 1 if upcoding or post-1988 price recalibrations are correlated withln(Laspeyres price) The inclusion of individ-ual pair time trends should mitigate these po-tential biases and indeed the estimate of 1 0925 (0093) As a robustness check I subtractan estimate of the component of ln(Laspeyresprice)p88-87 that is due to lagged cost growth30

The coefficient on this revised instrument is1120 (0119)

In the second stage I use five dependentvariables to measure intensity total costs (totalcharges from MedPAR deflated by annual costcharge ratios from the Cost Reports and con-verted to 2001 dollars using the hospitalservices CPI) length of stay number of surger-ies number of ICU days and in-hospital deaths

30 Because HCFA operates with a two-year data lagwhen recalibrating prices I calculate this component usingthe estimated coefficients from ln(Laspeyres price)p88-87 ln(price)p86-85

FIGURE 4 EFFECT OF POLICY CHANGE ON UPCODING OF

YOUNG BY HOSPITAL CHARACTERISTICS

Notes The figures above plot the year coefficients fromTable 4 ldquoYoungrdquo refers to Medicare beneficiaries under 70


All variables are normalized by the number ofadmissions in the relevant cell (ie average costper patient in DRG pair 138139 in 1987) Thefirst four measures are strong indicators of hos-pital expenditures on behalf of patients31 Deathrate is clearly an important albeit limited indi-cator of quality of care Although these mea-sures are commonly used in the healtheconomics literature they are imperfect One ofthe most common measures length of staycould be correlated positively or negativelywith quality of care Better care may enable apatient to leave sooner on the other hand hos-pitals may discharge patients too early in orderto cut costs (The latter was of greater concernin the 1980s as lengths of stay fell dramatically

in response to PPS) However the consistencyof the results across the variables suggests thatthe findings are robust

Table 5 presents estimates of 2 from thereduced-form regressions

(6) lnintensity or volumept pairp

yeart 2lnLaspeyres pricep88-87

post pairp year pt

This specification also controls for pair-specifictrends in the dependent variable throughout thestudy period ensuring that 2 does not reflectpreexisting trends in intensity or volume Ta-ble 5 offers little evidence that hospitals alteredtheir treatment policies or increased their admis-sions differentially in DRG pairs subject tolarger mechanical price changes Two of the sixpoint estimates indicate a negative response(costs and ICU days) and all are imprecisely

31 Total charges deflated by hospital costcharge ratiosshould be positively correlated with the services provided topatients indeed this is the measure HCFA uses to calculateDRG weights so that diagnosis groups with higher averagecharges are reimbursed more than diagnosis groups withlower average charges

FIGURE 5 DISTRIBUTION OF CHANGES IN LASPEYRES PRICE 1987ndash1988

Notes The Laspeyres price is defined as the average DRG weight for a given pair and yearholding constant the fraction of patients coded with CC at the 1987 fraction for young patientsin that DRG pair ldquoYoungrdquo refers to Medicare beneficiaries under age 70 Changes inLaspeyres price are converted to 2001 dollars using the base price for urban hospitals in 2001($4376)


estimated Given a precisely estimated first-stage coefficient of nearly 1 the correspondingIV estimates (21) and standard errors areroughly the same Due to the wide confidenceintervals intensity and volume responses can-not be ruled out but the evidence for theseresponses is underwhelming The results are notaffected by the addition of a control variable forthe severity of patients treated in each pair-year32

To obtain upper bounds for the intensityelasticities I estimated OLS regressions ofln(intensity) on ln(price) pair and year fixedeffects and pair-specific trends These esti-mated elasticities also reported in Table 5 areupward-biased due to the price recalibrationmethod33 Notwithstanding this bias the pointestimates are extremely small For example theOLS estimate indicates that only 7 cents ofevery additional dollar of reimbursement within

a DRG is spent on care for patients in that DRGThe elasticity of length of stay with respect toprice (018) is similar to the estimate reported inCutler (1990) (023) but the elasticity of sur-geries is much smaller (017 as compared to023) Overall Table 5 suggests that the fly-paper effect is weak in this sector

B Responses by Hospital and DRG Type

The aggregate analysis captures the averageintensity and volume responses across all ad-missions but masks potentially different re-sponses across hospitals To determine whetherindividual hospitals responded differently I es-timate equation (6) separately for the varioushospital subsamples Due to the large volume ofcoefficients generated by these models tablesare not included here Out of 126 regressions (6dependent variables 21 subsamples) 2 is sta-tistically significant at p 005 only once andin this case it indicates a negative impact ofprice on mortality in hospitals with fewer than100 beds34

The point estimates suggest that for-profithospitals decreased costs and length of stay

32 To estimate patient severity I computed the Charlsoncomorbidity index for each admission and calculated pair-year means This index reflects the likelihood of one-yearmortality based on 19 categories of comorbidity that arecaptured in the diagnosis codes on each patientrsquos record

33 One manifestation of this bias is the positive estimatedelasticity of death rate with respect to price the explanationfor this paradoxical result is simply that DRG pairs thatexperience increases in death rates receive higher reim-bursements because in-hospital care for the dying is costly 34 Tables are available upon request

TABLE 5mdashREAL RESPONSES TO CHANGES IN DRG PRICES

(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937

ln(surgeries)mean 115

ln(ICU days)mean 051

ln(death rate)mean 006

ln(volume)mean 31806

Reduced form ln(Laspeyres price) post 0207 0073 0009 0642 0381 0245

(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)

Parametric tests of H0 IV estimate x H1 IV estimate x (p-values are reported)a

x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001

OLS estimateln(price) 0074 0180 0166 0106 0151 NA

(0053) (0049) (0068) (0136) (0134)

Notes The top panel of the table reports results from reduced-form regressions of ln(intensity) or ln(volume) on the change in ln(Laspeyresprice) between 1987 and 1988 multiplied by an indicator for the post-1987 period (ldquopostrdquo) Intensity is measured by average costs length ofstay number of surgeries and the in-hospital death rate The second and third panels report IV and OLS estimates respectively of the elasticityof DRG-pair intensity and DRG-pair volume with respect to DRG-pair price All regressions include year fixed effects DRG-pair fixed effectsand DRG-pair trends The unlogged weighted mean of the dependent variable is reported at the top of each column The unit of observationis the DRG pair-year All observations are weighted by the number of admissions in the 20-percent MedPAR sample The sum of the weightsis 69 million Robust standard errors are reported in parentheses

a For ln(death rate) the tests are H0 IV estimate x H1 IV estimate x for x 05 and x 10 Signifies p 005 signifies p 001 signifies p 0001


more than hospitals of other ownership typesalthough an F-test does not reject equality of 2across the groups This pattern together with alarger estimated elasticity of volume with re-spect to price [0418 (0318)] is consistent withthe extensive upcoding by for-profits docu-mented in the previous section Financially dis-tressed hospitals also exhibit more negative costand length-of-stay elasticities though again thestandard errors are too large to reject a zeroresponse overall or equality with financially sta-ble hospitals Finally there is no evidence thatelasticities are larger in more competitive mar-kets as measured by the 1987 Herfindahl indi-ces in each hospitalrsquos Health Service AreaRather the elasticity of mortality with respect toprice is negative and significant at p 010 forhospitals in the least-competitive markets[105 (063)]

There are several potential explanations forthe scant evidence of real responses to the veryreal price increases described in Section VFirst hospitals may be unable to select differentintensity levels for each diagnosis (ie intensityis ldquolumpyrdquo across diagnoses) New technologiesor practice patterns once put in place may bedifficult to apply to only a select group of pa-tients Second patients may respond to a hos-pitalrsquos overall choice of intensity rather thanintensity at the diagnosis level providing littleincentive for hospitals to allocate funds pre-cisely where they are earned35 Third if inten-sity choices are not initially in equilibrium ahospital may allocate new funds earned in cer-tain diagnoses to overdue investments in otherareas

To address these possibilities I first reesti-mated (5) and (6) using Medicarersquos Major Di-agnostic Categories (MDCs) in place of DRGpairs36 There were 25 MDCs in 1987 16 ofwhich contained one or more DRG pairs Ifhospitals alter intensity at this level of aggrega-tion and patients in turn respond then intensityand volume responses should be positive andsizeable The point estimates again suggestsmall or negative responses and the standard

errors are of course larger due to the decline inthe number of observations (from 650 to 112)

Next I consider the possibility that hospitalsmay have responded only in diagnoses in whichpatients are likely to be quality-elastic All ad-missions in the MedPAR files are assigned toone of five categories emergency (admittedthrough the ER 44 percent of admissions in1987) urgent (first available bed 29 percent)elective (23 percent) newborn (01 percent)unknown (4 percent) I assigned each DRG tothe group accounting for the plurality of itsadmissions in 1987 and then estimated bothstages of the intensity analysis separately bygroup Again I find no conclusive evidence ofreal responses in any group The intensity re-sponses do appear to be strongest (and correctlysigned) in the elective diagnoses but they can-not be statistically distinguished from the re-sponses in other categories

Thus in contrast to previous studies I findlittle evidence of intensity responses to pricechanges at the diagnosis level In addition I donot find conclusive evidence that hospitals wereldquopushingrdquo lucrative admissions during this timeperiod

C Where Did the Money Go

Given the lack of intensity responses at theDRG level the question arises where did themoney go One possibility is that hospitalsspread the funds across all admissions To in-vestigate this hypothesis I aggregate the indi-vidual data into hospital-year cells Therelationship of interest is the elasticity of hos-pital intensity with respect to hospital pricewhich can be estimated from

(7) lnintensityht hospitalh

yeart lnpriceht ht

However there are two sources of bias in theOLS estimate of (1) the DRG recalibrationmethod and (2) the omission of an annualhospital-level measure of patient severity37 Aswith the previous analyses the policy change

35 Note that patients themselves need not have detailedknowledge of intensity levels their primary care physiciansand specialists may provide referrals based on their assess-ments of intensity

36 Tables are available upon request

37 Because true severity is difficult to capture with dis-charge data this bias remains even if measures such as theCharlson index are included as controls


can be used to identify but variation at thehospital level is required Because hospitalswith a large fraction of admissions in the ldquowithCCrdquo DRGs benefited the most from the policychange the interaction between this measureand a dummy for the post-reform years canserve as an instrument for average price in equa-tion (7)38

In constructing this instrument I use the 1987share of Medicare patients who are young (un-der 70) and coded with CC (hereafter calledshare CC) I select the pre-shock year becausecontemporaneous share CC would be affectedby post-shock upcoding responses and I useyoung patients because the data do not indicatewhether old patients had CC before the policychange This measure should be correlated withthe mechanical component of the hospital-levelprice increase hospitals with a large share CCin 1987 enjoyed larger increases in their averageDRG price independently of their upcoding re-sponse to the policy change

Table 6 gives the results from the first-stageregression of ln(price) on share CC post

(8) ln priceht hospitalh yeart

1shareCCh postt ht

where hospitalh is a vector of hospital fixedeffects All hospitals that appear in the Med-PAR sample in 1987 are included in this (un-balanced) panel The mean (standard deviation)of share CC in 1987 is 0086 (0043) and 1 is0233 A two-standard-deviation increase inshare CC is therefore associated with a 2-percent increase in the average price paid to ahospital following the policy change To illus-trate that share CC is uncorrelated with changesin average hospital prices in the pre-reformyears column 2 presents the results from aregression of ln(price) on share CC yeardummies

Coefficient estimates from the reduced-formequation


yeart 2shareCCh postt ht

are presented in Table 7 followed by IV andOLS estimates of equation (7) The IV estimatesfor the elasticity of hospital intensity with re-spect to average hospital price are positive for

38 An alternative instrument for hospital price isln(Laspeyres price)h88-87 However because the actualDRGs sampled for each hospital vary substantially overtime and DRG controls cannot be included in this specifi-cation share CC is a much more accurate measure of themechanical component at this level of aggregation

TABLE 6mdashEFFECTS OF POLICY CHANGE ON AVERAGE

HOSPITAL PRICES

(N 36651)

Dependent variable is ln(price)mean(price) 127

Share CC post 0233(0021)

Share CC year dummies1986 0022

(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)

Adj R-squared 0890 0890

Notes The table reports results from two specifications ofthe first-stage regression relating ln(price) to share CC the1987 share of a hospitalrsquos Medicare patients who are under70 and assigned to the top code of a DRG pair In column1 share CC is multiplied by an indicator for the post-1987period (ldquopostrdquo) In column 2 share CC is interacted withindividual year dummies Both regressions include hospitalfixed effects The unit of observation is the hospital-yearAll observations are weighted by the number of admissionsin the 20-percent MedPAR sample The sum of the weightsis 137 million Robust standard errors are reported in pa-rentheses Signifies p 005 signifies p 001 signifies p 0001


four of the five intensity measures and sta-tistically significant for three The exception isthe in-hospital death rate for which estimatedelasticity is negative but insignificant (a posi-tive coefficient on death rate implies a nega-tive intensity response) The elasticity resultsreveal that an additional dollar of reimburse-ment goes wholly toward patient care Extrareimbursement is associated with longerstays more surgeries more ICU days andpossibly worse outcomes39 The intensity in-creases are consistent with previous studiesthat have examined hospital-wide responsesto changes in the update factor (eg JackHadley et al 1989)40

Hospitals benefiting disproportionately fromthe policy change also enjoyed increases in mar-ket share Given the time resolution of the datahowever it is difficult to determine whetherintensity increases are the signal that elicitedthese gains The intensity results are thereforeconsistent with (at least) two distinct models ofhospital behavior competition in overall inten-

39 Due to computing constraints it is not possible toestimate equations (8) and (9) with individual hospital-yeartrends If share CC is correlated with pre-existing trends inthe dependent variables the estimates in Table 7 will beupward-biased

40 The results can also be reconciled with Duggan(2000) who finds that private hospitals invested funds from

the expansion of Californiarsquos Disproportionate Share Pro-gram (DSH) in financial assets rather than patient careThere are at least two plausible reasons for this differenceFirst the DSH payments were substantially larger repre-senting up to 30 percent of hospital revenues Hospitals mayreact differently to such large budget shocks particularlybecause the marginal benefit of dollars directed to patientcare likely declines with the amount spent Second the DSHpayments include a behavioral response Duggan shows thatprivate hospitals obtained their DSH funds by cream-skim-ming newly profitable patients away from public hospitalsFunds obtained in this manner may be spent differently thanpayments that are ldquomechanicallyrdquo increased Dugganrsquos re-sults suggest that the upcoding-induced payments I examinein Section IV may not have been allocated to patient care

TABLE 7mdashREAL RESPONSES TO CHANGES IN AVERAGE HOSPITAL PRICE

Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778

Reduced formShare CC post 0234 0069 0067 0684 0122 0403

(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10

OLS estimateln(price) 0769 0350 0867 1483 0601 0022

(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651

Notes The top panel of the table reports results from reduced-form regressions of ln(intensity) or ln(volume) on shareCC multiplied by an indicator for the post-1987 period (ldquopostrdquo) Share CC is the 1987 share of a hospitalrsquos Medicarepatients who are under 70 and assigned to the top code of a DRG pair Intensity is measured by average costs lengthof stay number of surgeries and the in-hospital death rate The second and third panels report IV and OLS estimatesrespectively of the elasticity of hospital intensity and hospital volume with respect to hospital price All regressionsinclude year and hospital fixed effects The unlogged weighted mean of the dependent variable is reported at the topof each column The unit of observation is the hospital-year All observations are weighted by the number of admissionsin the 20-percent MedPAR sample The sum of the weights is 137 million Robust standard errors are reported inparentheses



sity and maximization of overall intensity sub-ject to a budget constraint The preponderanceof the evidence does not however support thecommonly assumed model of intensity com-petition at the diagnosis level The lack ofdiagnosis-specific intensity responses contrastswith earlier research and helps to explain whydiagnosis specialization is very limited in inpa-tient care

VI Conclusion

As public and private healthcare insurerscontinue to strengthen financial incentives forefficiency in the production of healthcare it iscritical to understand what the implications ofsuch incentives are for health care quality andexpenditures The fixed-price system used bymany insurers makes hospitals the residualclaimants of profits earned on inpatient staysThese profits differ by diagnosis creatingincentives for hospitals to increase the vol-ume of admissions in profitable diagnosesrelative to unprofitable diagnoses Solicitingthe most lucrative type of business may havefew deleterious consequences in other fixed-price settings (eg utilities) but it is poten-tially dangerous in the healthcare industryFor example doctors at Redding MedicalCenter a for-profit hospital operated by TenetHealthcare Corporation in Redding Califor-nia are currently under criminal investigationfor performing lucrative open-heart surgeriesin place of medically managing symptoms ofheart disease (Kurt Eichenwald 2003)

Resolving the question of how hospitalsrespond to changes in DRG prices which aresimply shocks to the profitability of certaindiagnoses or treatments is therefore criticalfrom a policy standpoint In addition theseresponses provide a window into industryconduct In theory quality erosion is kept incheck by competition among hospitals41 Re-sponses to individual price changes can reveal

whether this competition occurs at the diag-nosis level

This study illustrates how a simple changein the DRG classification system in 1988generated large and exogenous price changesfor 40 percent of DRG codes accounting for43 percent of Medicare admissions Hospitalsresponded to these price changes by upcod-ing patients to DRG codes associated withlarge reimbursement increases garnering$330 ndash$425 million in extra reimbursementannually They proved quite sophisticated intheir upcoding strategies upcoding more inthose DRGs where the reward for doing soincreased more Additionally while all sub-samples of hospitals upcoded in response tothe policy change for-profit facilities availedthemselves of this opportunity to the greatestextent

Whereas coding behavior proved very re-sponsive to diagnosis-specific financial incen-tives admissions and treatment policies didnot I do not find convincing evidence thathospitals increased admissions differentiallyfor those diagnoses with the largest price in-creases although this practice may have be-come more prevalent in recent years Theresults also suggest that healthcare insurerscannot effect an increase in the quality of careprovided to patients with a particular diagno-sis simply by increasing reimbursement ratesfor that diagnosis However there is evidencethat hospitals spend extra funds they receiveon patient care in all DRGs This suggeststhat hospitals do not (or cannot) optimizequality choices by product line which mayexplain the relative lack of specialization inthe hospital industry One anticipated benefitof PPS was that hospitals would specialize inadmissions in which they are relatively cost-efficient If however hospitals do not bal-ance costs and benefits within individualproduct lines such specialization is unlikelyto occur

This research exposes the difficulties inher-ent in implementing a fixed-price system inwhich the output is difficult to verify AsMedicare and other insurers continue to ex-tend prospective payment systems to addi-tional areas they would do well to considerthe evidence that providers of all ownershiptypes may behave strategically to garner ad-ditional resources

41 Of course physicians also play an important role inensuring appropriate care for their patients as highlightedby Kenneth J Arrow (1963)


APPENDIX

REFERENCES

Arrow Kenneth J ldquoUncertainty and the WelfareEconomics of Medical Carerdquo American Eco-nomic Review 1963 53(5) pp 941ndash73

Carter Grace M Newhouse Joseph P andRelles Daniel A ldquoHow Much Change in theCase Mix Index Is DRG Creeprdquo Journal ofHealth Economics 1990 9(4) pp 411ndash28

Coulam Robert F and Gaumer Gary L ldquoMedi-carersquos Prospective Payment System A Criti-cal Appraisalrdquo Health Care Financing ReviewAnnual Supplement 1991 12 pp 45ndash77

Cutler David M ldquoEmpirical Evidence on Hos-pital Delivery under Prospective PaymentrdquoUnpublished Paper 1990

Cutler David M ldquoThe Incidence of AdverseMedical Outcomes under Prospective Pay-mentrdquo Econometrica 1995 63(1) pp 29ndash50

Cutler David M ldquoCost Shifting or Cost CuttingThe Incidence of Reductions in MedicarePaymentsrdquo in James M Poterba ed Taxpolicy and the economy Vol 12 CambridgeMA MIT Press 1998 pp 1ndash27

Dafny Leemore S and Gruber Jonathan ldquoPub-lic Insurance and Child Hospitalizations Ac-cess and Efficiency Effectsrdquo Journal ofPublic Economics 2005 89(1) pp 109ndash29

Dartmouth Medical School Center for the Eval-uative Clinical Sciences The Dartmouth atlasof health care Chicago American HospitalPublishing 1996

Dranove David ldquoRate-Setting by Diagnosis Re-lated Groups and Hospital SpecializationrdquoRAND Journal of Economics 1987 18(3)pp 417ndash27

Dranove David ldquoPricing by Non-Profit Institu-tions The Case of Hospital Cost-ShiftingrdquoJournal of Health Economics 1988 7(1) pp47ndash57

Duggan Mark G ldquoHospital Ownership and Pub-lic Medical Spendingrdquo Quarterly Journal ofEconomics 2000 115(4) pp 1343ndash73

Duggan Mark G ldquoHospital Market Structureand the Behavior of Not-for-Profit Hospi-talsrdquo RAND Journal of Economics 200233(3) pp 433ndash46

Eichenwald Kurt ldquoOperating Profits MiningMedicare How One Hospital Benefited onQuestionable Operationsrdquo The New YorkTimes August 12 2003 A1

Ellis Randall P and McGuire Thomas G ldquoHos-pital Response to Prospective PaymentMoral Hazard Selection and Practice-StyleEffectsrdquo Journal of Health Economics 199615(3) pp 257ndash77

Gilman Boyd H ldquoHospital Response to DRGRefinements The Impact of Multiple Reim-bursement Incentives on Inpatient Length ofStayrdquo Health Economics 2000 9(4) pp277ndash94

Hadley Jack Zukerman Stephen and Feder Ju-dith ldquoProfits and Fiscal Pressure in the Pro-spective Payment System Their Impacts onHospitalsrdquo Inquiry 1989 26(3) pp 354ndash65

Hodgkin Dominic and McGuire Thomas GldquoPayment Levels and Hospital Response toProspective Paymentrdquo Journal of HealthEconomics 1994 13(1) pp 1ndash29

TABLE A1mdashDESCRIPTIVE STATISTICS FOR HOSPITAL

CHARACTERISTICS

Variable Mean SD Min Max

OwnershipFor-profit 014 035 0 1Non-profit 058 049 0 1Government 028 045 0 1

Financial distress measuresDebtasset ratio 052 032 0 217Medicare bite 037 011 0 100Medicaid bite 011 008 0 084

RegionNortheast 014 035 0 1Midwest 029 046 0 1South 038 049 0 1West 018 038 0 1

Size1ndash99 beds 046 050 0 1100ndash299 beds 037 048 0 1300 beds 017 037 0 1

Service offeringsTeaching program 006 023 0 1Open-heart surgery 013 034 0 1Trauma facility 019 039 0 1ICU beds (except neonatal) 1027 1229 0 194

Market concentrationHSA Herfindahl (AHA) 007 005 0 1HSA Herfindahl (Dartmouth

Atlas of Health Care)067 035 0 1

Notes The table reports descriptive statistics for hospitalswith complete data Hospitals with debtasset ratios in the1 tails of the distribution are also excluded N 5352The analyses in Tables 4 6 and 7 utilize data from thissubset of hospitals the analyses in Tables 3 and 5 do notrequire hospital characteristics and therefore include datafrom all hospitals financed under PPS


Lagnado Lucette ldquoColumbiaHCA Grades ItsHospitals on Severity of Their MedicareClaimsrdquo The Wall Street Journal May 301997 A1 A6

Lichtenberg Frank R ldquoThe Effects of Medicareon Health Care Utilization and Outcomesrdquo inAlan M Garber ed Frontiers in health pol-icy research Vol 5 Cambridge MA MITPress 2002 pp 27ndash52

Murray Lauren A and Eppig Franklin J ldquoIn-surance Trends for the Medicare Population1991ndash1999rdquo Health Care Financing Review2002 23(3) pp 9ndash16

Pstay Bruce M Boineau Robin Kuller LewisH and Luepker Russell V ldquoThe PotentialCosts of Upcoding for Heart Failure in theUnited Statesrdquo American Journal of Cardi-ology 1999 84(1) pp 108ndash09

Shleifer Andrei ldquoA Theory of Yardstick Con-sumptionrdquo RAND Journal of Economics1985 16(3) pp 319ndash27

Silverman Elaine M and Skinner Jonathan SldquoMedicare Upcoding and Hospital Owner-shiprdquo Journal of Health Economics 200423(2) pp 369ndash89

Silverman Elaine M Skinner Jonathan S andFisher Elliott S ldquoThe Association betweenFor-Profit Hospital Ownership and Increased

Medicare Spendingrdquo New England Journalof Medicine 1999 341(6) pp 420ndash26

Staiger Douglas and Gaumer Gary L ldquoThe Im-pact of Financial Pressure on Quality of Carein Hospitals Post-Admission Mortality underMedicarersquos Prospective Payment SystemrdquoUnpublished Paper 1992

Steinwald Bruce and Dummit Laura A ldquoHospi-tal Case-Mix Change Sicker Patients or DRGCreeprdquo Health Affairs 1989 8(2) pp 35ndash47

US Census Bureau Statistical CompendiaBranch Statistical abstract of the UnitedStates Washington DC US GovernmentPrinting Office 1987 1991 1998 2001

US Department of Health and Human ServicesCenters for Medicare amp Medicaid Services2002 data compendium Washington DCUS Government Printing Office 2002

Office of the Federal Register National Archivesand Records Service Federal register Wash-ington DC US Government Printing Of-fice 1984ndash2000

Weisbrod Burton A ldquoWhy Private Firms Gov-ernmental Agencies and Nonprofit Organiza-tions Behave Both Alike and DifferentlyApplication to the Hospital Industryrdquo North-western University Institute for Policy Re-search Working Papers No WP-04-08 2004


tion in DRG prices which are typically adjustedto reflect changes in hospital costs As a resultpositive associations between changes in priceand changes in spending or intensity of treat-ment likely reflect bilateral causality and do notconstitute a priori evidence that hospitals altertreatment patterns in response to price changesTo obtain unbiased estimates of hospital re-sponses to price changes this study exploits a1988 policy change that generated large pricechanges for 43 percent of Medicare admis-sions2 The policy change was the eliminationof ldquoage over 69rdquo and ldquoage under 70rdquo in thedescriptions of the DRGs to which patients maybe assigned Qualifiers that formerly read ldquowithcomplications or age over 69rdquo and ldquowithoutcomplications and age under 69rdquo now readldquowith complicationsrdquo or ldquowithout complica-tionsrdquo This seemingly innocuous modificationwhich is described in greater detail in Section Iactually led to substantial changes in prices forDRGs with these qualifiers

I consider both ldquonominalrdquo and ldquorealrdquo re-sponses to these price changes where nominalrefers to hospital coding practices and real re-fers to admissions volumes and intensity of careactually provided Because hospitals are respon-sible for coding patients to the appropriateDRGs raising prices for certain DRGs mayentice hospitals to ldquoupcoderdquo or switch patientsfrom lower-paying to higher-paying DRGsWhile upcoding does not affect real elements ofpatient care it inflates hospital reimbursementsThis was the primary response of hospitals tothe 1988 policy change Hospitals also demon-strated a keen awareness of risk-reward trade-offs in their upcoding which can lead to finesand criminal charges if detected Although thepolicy shock created a blanket incentive to in-crease upcoding in dozens of diagnoses hospi-tals upcoded most in those diagnoses where theincentive to do so was greatest The upcodingresponse was also strongest among for-profithospitals a finding that is consistent with priorresearch

The real responses to the policy change wereless finely tuned I find little evidence that hos-

pitals adjusted the intensity or quality of theircare in response to changes in diagnosis-specificprices where intensity is measured by totalcosts length of stay number of surgical proce-dures number of intensive-care-unit (ICU)days and quality by the in-hospital death rateRather hospitals spread the additional fundsacross all admissions I also find no conclusiveevidence that hospitals attracted more patientsin diagnoses subject to larger price increases

The remainder of the paper is organized intosix sections Section I describes PPS and priorrelated research and lays out the hypotheses Itest Section II offers a detailed explanation of the1988 reclassification followed by a discussionof the data in Section III Sections IV and Vexamine the nominal and real responses to pricechanges respectively and Section VI concludes

I Background

A A PPS Primer

The defining element of PPS is a reimburse-ment amount that is fixed regardless of a hos-pitalrsquos actual expenditures on a patient Thispayment does vary however by the patientrsquosmedical diagnosis Diagnoses are grouped intoapproximately 500 DRGs Each DRG is as-signed a weight (called a ldquoDRG weightrdquo) thatreflects the relative resource intensity of admis-sions within that group Reimbursement to hos-pital h for an admission in DRG d is given by

(1) Phd Ph 1 IMEh

1 DSHh DRG weightd

where Ph is a hospital-specific amount (inflatedannually by a congressionally approved ldquoupdatefactorrdquo) IMEh represents an adjustment for in-direct medical education (teaching) and DSHhadjusts payment levels to compensate hospitalswith a disproportionate share of indigent patients3

Most of the variation in Phd is due to theDRG weights which range between 009 (DRG448 for allergic reactions) to 228 (DRG 480 forliver transplants)4 The Health Care Financing

2 In 2000 Medicare beneficiaries accounted for 37 per-cent of hospital discharges and 31 percent of total revenues(Data Compendium Centers for Medicare and MedicaidServices and authorrsquos tabulations from the 2000 Survey ofHospitals by the American Hospital Association (AHA))

3 This simplified formula appears in David M Cutler(1995)

4 The range for DRG weights is given for 1985ndash1996

























































DRGcode


1987weight

1988weight



sample)



in volume


08446 09804 16 44989 42314 6


07091 07151 1 4611 10512 128


08136 08535 5 45080 35233 22


06514 05912 9 4182 16829 302


08271 09259 12 45903 38805 15


06984 05791 17 2033 12363 508




III Data



























Unit of observation


N Mean SD N Mean SD




















(N 650)



spread88-87 post 0077 0108(0016) (0015)



(0008) (0005)1987 0077 0011

(0008) (0005)1988 0058 0813

(0011) (0014)1989 0097 0780

(0009) (0014)1990 0115 0764

(0009) (0014)1991 0128 0748

(0010) (0014)Adj R-squared 0948 0960




SPREAD CHANGE






B Hospital Analysis






V Real Responses















NDRG 1381987 NDRG 1391987




(N 650)


For-profitNot-for-



spread88-87 post 0071 0080 0058 0082 0074 0083 0062 0082 0079(0024) (0017) (0018) (0109) (0017) (0016) (0018) (0017) (0024)


(0009) (0009) (0008) (0008) (0009) (0008) (0009) (0009) (0012)1987 0081 0077 0078 0099 0073 0123 0054 0078 0052

(0010) (0008) (0008) (0008) (0008) (0007) (0009) (0008) (0012)1988 0080 0055 0065 0083 0054 0104 0036 0063 0024

(0012) (0011) (0012) (0011) (0011) (0010) (0010) (0012) (0014)1989 0140 0094 0104 0127 0094 0131 0075 0111 0067

(0011) (0009) (0009) (0009) (0009) (0009) (0010) (0009) (0012)1990 0147 0114 0120 0144 0112 0148 0092 0132 0080

(0011) (0009) (0010) (0009) (0010) (0008) (0009) (0010) (0013)1991 0179 0123 0136 0161 0124 0159 0103 0148 0091

(0011) (0011) (0010) (0010) (0010) (0010) (0011) (0010) (0014)89 87 0059 0016 0027 0027 0022 0007 0021 0033 0015

(0010) (0009) (0008) (0009) (0009) (0008) (0010) (0008) (0014)Adj R-squared 0914 0946 0927 0933 0947 0939 0940 0940 0915







pairp year pt














post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS

















































































DRGcode


1987weight

1988weight



sample)



in volume


08446 09804 16 44989 42314 6


07091 07151 1 4611 10512 128


08136 08535 5 45080 35233 22


06514 05912 9 4182 16829 302


08271 09259 12 45903 38805 15


06984 05791 17 2033 12363 508




III Data



























Unit of observation


N Mean SD N Mean SD




















(N 650)



spread88-87 post 0077 0108(0016) (0015)



(0008) (0005)1987 0077 0011

(0008) (0005)1988 0058 0813

(0011) (0014)1989 0097 0780

(0009) (0014)1990 0115 0764

(0009) (0014)1991 0128 0748

(0010) (0014)Adj R-squared 0948 0960




SPREAD CHANGE






B Hospital Analysis






V Real Responses















NDRG 1381987 NDRG 1391987




(N 650)


For-profitNot-for-



spread88-87 post 0071 0080 0058 0082 0074 0083 0062 0082 0079(0024) (0017) (0018) (0109) (0017) (0016) (0018) (0017) (0024)


(0009) (0009) (0008) (0008) (0009) (0008) (0009) (0009) (0012)1987 0081 0077 0078 0099 0073 0123 0054 0078 0052

(0010) (0008) (0008) (0008) (0008) (0007) (0009) (0008) (0012)1988 0080 0055 0065 0083 0054 0104 0036 0063 0024

(0012) (0011) (0012) (0011) (0011) (0010) (0010) (0012) (0014)1989 0140 0094 0104 0127 0094 0131 0075 0111 0067

(0011) (0009) (0009) (0009) (0009) (0009) (0010) (0009) (0012)1990 0147 0114 0120 0144 0112 0148 0092 0132 0080

(0011) (0009) (0010) (0009) (0010) (0008) (0009) (0010) (0013)1991 0179 0123 0136 0161 0124 0159 0103 0148 0091

(0011) (0011) (0010) (0010) (0010) (0010) (0011) (0010) (0014)89 87 0059 0016 0027 0027 0022 0007 0021 0033 0015

(0010) (0009) (0008) (0009) (0009) (0008) (0010) (0008) (0014)Adj R-squared 0914 0946 0927 0933 0947 0939 0940 0940 0915







pairp year pt














post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS



























III Data



























Unit of observation


N Mean SD N Mean SD




















(N 650)



spread88-87 post 0077 0108(0016) (0015)



(0008) (0005)1987 0077 0011

(0008) (0005)1988 0058 0813

(0011) (0014)1989 0097 0780

(0009) (0014)1990 0115 0764

(0009) (0014)1991 0128 0748

(0010) (0014)Adj R-squared 0948 0960




SPREAD CHANGE






B Hospital Analysis






V Real Responses















NDRG 1381987 NDRG 1391987




(N 650)


For-profitNot-for-



spread88-87 post 0071 0080 0058 0082 0074 0083 0062 0082 0079(0024) (0017) (0018) (0109) (0017) (0016) (0018) (0017) (0024)


(0009) (0009) (0008) (0008) (0009) (0008) (0009) (0009) (0012)1987 0081 0077 0078 0099 0073 0123 0054 0078 0052

(0010) (0008) (0008) (0008) (0008) (0007) (0009) (0008) (0012)1988 0080 0055 0065 0083 0054 0104 0036 0063 0024

(0012) (0011) (0012) (0011) (0011) (0010) (0010) (0012) (0014)1989 0140 0094 0104 0127 0094 0131 0075 0111 0067

(0011) (0009) (0009) (0009) (0009) (0009) (0010) (0009) (0012)1990 0147 0114 0120 0144 0112 0148 0092 0132 0080

(0011) (0009) (0010) (0009) (0010) (0008) (0009) (0010) (0013)1991 0179 0123 0136 0161 0124 0159 0103 0148 0091

(0011) (0011) (0010) (0010) (0010) (0010) (0011) (0010) (0014)89 87 0059 0016 0027 0027 0022 0007 0021 0033 0015

(0010) (0009) (0008) (0009) (0009) (0008) (0010) (0008) (0014)Adj R-squared 0914 0946 0927 0933 0947 0939 0940 0940 0915







pairp year pt














post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS








































Unit of observation


N Mean SD N Mean SD




















(N 650)



spread88-87 post 0077 0108(0016) (0015)



(0008) (0005)1987 0077 0011

(0008) (0005)1988 0058 0813

(0011) (0014)1989 0097 0780

(0009) (0014)1990 0115 0764

(0009) (0014)1991 0128 0748

(0010) (0014)Adj R-squared 0948 0960




SPREAD CHANGE






B Hospital Analysis






V Real Responses















NDRG 1381987 NDRG 1391987




(N 650)


For-profitNot-for-



spread88-87 post 0071 0080 0058 0082 0074 0083 0062 0082 0079(0024) (0017) (0018) (0109) (0017) (0016) (0018) (0017) (0024)


(0009) (0009) (0008) (0008) (0009) (0008) (0009) (0009) (0012)1987 0081 0077 0078 0099 0073 0123 0054 0078 0052

(0010) (0008) (0008) (0008) (0008) (0007) (0009) (0008) (0012)1988 0080 0055 0065 0083 0054 0104 0036 0063 0024

(0012) (0011) (0012) (0011) (0011) (0010) (0010) (0012) (0014)1989 0140 0094 0104 0127 0094 0131 0075 0111 0067

(0011) (0009) (0009) (0009) (0009) (0009) (0010) (0009) (0012)1990 0147 0114 0120 0144 0112 0148 0092 0132 0080

(0011) (0009) (0010) (0009) (0010) (0008) (0009) (0010) (0013)1991 0179 0123 0136 0161 0124 0159 0103 0148 0091

(0011) (0011) (0010) (0010) (0010) (0010) (0011) (0010) (0014)89 87 0059 0016 0027 0027 0022 0007 0021 0033 0015

(0010) (0009) (0008) (0009) (0009) (0008) (0010) (0008) (0014)Adj R-squared 0914 0946 0927 0933 0947 0939 0940 0940 0915







pairp year pt














post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS







































(N 650)



spread88-87 post 0077 0108(0016) (0015)



(0008) (0005)1987 0077 0011

(0008) (0005)1988 0058 0813

(0011) (0014)1989 0097 0780

(0009) (0014)1990 0115 0764

(0009) (0014)1991 0128 0748

(0010) (0014)Adj R-squared 0948 0960




SPREAD CHANGE






B Hospital Analysis






V Real Responses















NDRG 1381987 NDRG 1391987




(N 650)


For-profitNot-for-



spread88-87 post 0071 0080 0058 0082 0074 0083 0062 0082 0079(0024) (0017) (0018) (0109) (0017) (0016) (0018) (0017) (0024)


(0009) (0009) (0008) (0008) (0009) (0008) (0009) (0009) (0012)1987 0081 0077 0078 0099 0073 0123 0054 0078 0052

(0010) (0008) (0008) (0008) (0008) (0007) (0009) (0008) (0012)1988 0080 0055 0065 0083 0054 0104 0036 0063 0024

(0012) (0011) (0012) (0011) (0011) (0010) (0010) (0012) (0014)1989 0140 0094 0104 0127 0094 0131 0075 0111 0067

(0011) (0009) (0009) (0009) (0009) (0009) (0010) (0009) (0012)1990 0147 0114 0120 0144 0112 0148 0092 0132 0080

(0011) (0009) (0010) (0009) (0010) (0008) (0009) (0010) (0013)1991 0179 0123 0136 0161 0124 0159 0103 0148 0091

(0011) (0011) (0010) (0010) (0010) (0010) (0011) (0010) (0014)89 87 0059 0016 0027 0027 0022 0007 0021 0033 0015

(0010) (0009) (0008) (0009) (0009) (0008) (0010) (0008) (0014)Adj R-squared 0914 0946 0927 0933 0947 0939 0940 0940 0915







pairp year pt














post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS



























SPREAD CHANGE






B Hospital Analysis






V Real Responses















NDRG 1381987 NDRG 1391987




(N 650)


For-profitNot-for-



spread88-87 post 0071 0080 0058 0082 0074 0083 0062 0082 0079(0024) (0017) (0018) (0109) (0017) (0016) (0018) (0017) (0024)


(0009) (0009) (0008) (0008) (0009) (0008) (0009) (0009) (0012)1987 0081 0077 0078 0099 0073 0123 0054 0078 0052

(0010) (0008) (0008) (0008) (0008) (0007) (0009) (0008) (0012)1988 0080 0055 0065 0083 0054 0104 0036 0063 0024

(0012) (0011) (0012) (0011) (0011) (0010) (0010) (0012) (0014)1989 0140 0094 0104 0127 0094 0131 0075 0111 0067

(0011) (0009) (0009) (0009) (0009) (0009) (0010) (0009) (0012)1990 0147 0114 0120 0144 0112 0148 0092 0132 0080

(0011) (0009) (0010) (0009) (0010) (0008) (0009) (0010) (0013)1991 0179 0123 0136 0161 0124 0159 0103 0148 0091

(0011) (0011) (0010) (0010) (0010) (0010) (0011) (0010) (0014)89 87 0059 0016 0027 0027 0022 0007 0021 0033 0015

(0010) (0009) (0008) (0009) (0009) (0008) (0010) (0008) (0014)Adj R-squared 0914 0946 0927 0933 0947 0939 0940 0940 0915







pairp year pt














post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS





























B Hospital Analysis






V Real Responses















NDRG 1381987 NDRG 1391987




(N 650)


For-profitNot-for-



spread88-87 post 0071 0080 0058 0082 0074 0083 0062 0082 0079(0024) (0017) (0018) (0109) (0017) (0016) (0018) (0017) (0024)


(0009) (0009) (0008) (0008) (0009) (0008) (0009) (0009) (0012)1987 0081 0077 0078 0099 0073 0123 0054 0078 0052

(0010) (0008) (0008) (0008) (0008) (0007) (0009) (0008) (0012)1988 0080 0055 0065 0083 0054 0104 0036 0063 0024

(0012) (0011) (0012) (0011) (0011) (0010) (0010) (0012) (0014)1989 0140 0094 0104 0127 0094 0131 0075 0111 0067

(0011) (0009) (0009) (0009) (0009) (0009) (0010) (0009) (0012)1990 0147 0114 0120 0144 0112 0148 0092 0132 0080

(0011) (0009) (0010) (0009) (0010) (0008) (0009) (0010) (0013)1991 0179 0123 0136 0161 0124 0159 0103 0148 0091

(0011) (0011) (0010) (0010) (0010) (0010) (0011) (0010) (0014)89 87 0059 0016 0027 0027 0022 0007 0021 0033 0015

(0010) (0009) (0008) (0009) (0009) (0008) (0010) (0008) (0014)Adj R-squared 0914 0946 0927 0933 0947 0939 0940 0940 0915







pairp year pt














post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS


































NDRG 1381987 NDRG 1391987




(N 650)


For-profitNot-for-



spread88-87 post 0071 0080 0058 0082 0074 0083 0062 0082 0079(0024) (0017) (0018) (0109) (0017) (0016) (0018) (0017) (0024)


(0009) (0009) (0008) (0008) (0009) (0008) (0009) (0009) (0012)1987 0081 0077 0078 0099 0073 0123 0054 0078 0052

(0010) (0008) (0008) (0008) (0008) (0007) (0009) (0008) (0012)1988 0080 0055 0065 0083 0054 0104 0036 0063 0024

(0012) (0011) (0012) (0011) (0011) (0010) (0010) (0012) (0014)1989 0140 0094 0104 0127 0094 0131 0075 0111 0067

(0011) (0009) (0009) (0009) (0009) (0009) (0010) (0009) (0012)1990 0147 0114 0120 0144 0112 0148 0092 0132 0080

(0011) (0009) (0010) (0009) (0010) (0008) (0009) (0010) (0013)1991 0179 0123 0136 0161 0124 0159 0103 0148 0091

(0011) (0011) (0010) (0010) (0010) (0010) (0011) (0010) (0014)89 87 0059 0016 0027 0027 0022 0007 0021 0033 0015

(0010) (0009) (0008) (0009) (0009) (0008) (0010) (0008) (0014)Adj R-squared 0914 0946 0927 0933 0947 0939 0940 0940 0915







pairp year pt














post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS






























pairp year pt














post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS































post pairp year pt















(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS



































(N 650)

Dependent variable

ln(cost)mean $9489

ln(LOS)mean 937






(0133) (0146) (0158) (0380) (0352) (0272)IV estimate

ln(price) 0223 0079 0009 0694 0412 0265(0147) (0157) (0171) (0425) (0383) (0285)


x 05 0 0 0 0 041 021x 1 0 0 0 0 006 001


(0053) (0049) (0068) (0136) (0134)













yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS



































yeart lnpriceht ht










1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS






























1shareCCh postt ht








HOSPITAL PRICES

(N 36651)




(0040)1987 0015

(0038)1988 0229

(0038)1989 0212

(0039)1990 0174

(0040)1991 0270

(0047)Year dummies

1986 0039 0041(0001) (0004)

1987 0057 0058(0001) (0004)

1988 0063 0064(0002) (0004)

1989 0088 0090(0002) (0004)

1990 0094 0099(0002) (0004)

1991 0119 0116(0002) (0004)










Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS
































Dependent variable

ln(cost)mean $9014

ln(LOS)mean 881




ln(volume)mean 778


(0075) (0034) (0104) (0186) (0098) (0052)IV estimate

ln(price) 0998 0296 0291 3457 0536 1728(0312) (0141) (0445) (0950) (0423) (0276)


x 05 096 006 031 10 0 10x 1 050 0 004 10 0 10


(0027) (0011) (0036) (0065) (0031) (0018)N 36169 36651 35897 28226 34992 36651





VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS



























VI Conclusion









APPENDIX

REFERENCES



















CHARACTERISTICS


























APPENDIX

REFERENCES



















CHARACTERISTICS


























how do hospitals respond to price changes? files/16_dafny_how do hos… · 6 bruce steinwald and...

Documents