Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
This PDF is a selection from a published volumefrom the National Bureau of Economic Research
Volume Title: Frontiers in Health Policy Research,Volume 5
Volume Author/Editor: Alan M. Garber, editor
Volume Publisher: MIT Press
Volume ISBN: 0-262-07234-3
Volume URL: http://www.nber.org/books/garb02-1
Conference Date: June 7, 2001
Publication Date: January 2002
Title: Hospital Ownership Conversions: Definingthe Appropriate Public Oversight Role
Author: Frank A. Sloan
URL: http://www.nber.org/chapters/c9860
5
Hospital Ownership Conversions: Defining theAppropriate Public Oversight Role
Frank A. Sloan, Duke University and NBER
Executive Summary
This paper reviews recent empirical evidence on the effects of hospital owner-ship conversions on quality of care and provision of public goods, such as Un-compensated care, and presents new results on these topics based on hospitaldischarge data from the Healthcare Cost and Utilization Project's (HCUP) Na-tionwide Inpatient Sample. My analysis of these data reveals that conversionfrom government or private nonprofit to for-profit ownership has no effect onin-hospital mortality but rates of pneumonia complications increased follow-ing conversion to for-profit status. Other research, discussed in the paper,found increased mortality rates following discharge from the hospital for pa-tients admitted to hospitals that had converted to for-profit ownership. Therewas no effect of such conversions on the propensity to admit uninsured orMedicaid patients. Clearly, there is considerable heterogeneity in outcomes at-tributable to conversions. Overall, the evidence suggests a role for public scm-tiny of hospital ownership conversions.
I. Introduction and Policy Context
The hospital industry attracts much public scrutiny, given its importantrole in providing personal health services, its sizeabout 3 percent ofgross domestic product, the importance of hospitals as employers inthe communities they serve, and the high share of hospital revenuefrom public sources. During the last two decades, the industry has ex-perienced a dramatic downsizing at the same time, and for that reason,the methods of paying for hospital care have changed. The market forhospital care in the year 2000 was much more competitive than it wasin 1980.
Downsizing has taken various forms. First, there has been a reduc-tion in the number of hospitals. In 1980, there were 5,830 "community"hospitals (nonfederal short-term general hospitals located outside
124 Sloan
institutions). By 1999, the latest year for which data are publicly avail-able, the number had fallen to 4,956 (figure 5.1). The number of com-munity hospitals peaked during the 1970s. Second, existing hospitalshave reduced bed capacity from a peak of slightly over 1.0 million bedsin 1983 to 830,000 beds in 1999 (figure 5.2). Third, hospitals havediversified, integrating both vertically and horizontally, and they havesought new ownership and management. As an extreme measure,many hospitals have closed.
For-profit (F) hospitals run counter to the national trend in numberand bed capacity (figures 5.1 and 5.2). The number of such hospitalshas remained relatively constant in terms of numbers of hospitals andhas risen in terms of number of beds under such hospital ownership.Although the number of private nonprofit hospitals (N) peaked in1984, as did the number of beds, the subsequent decline in both hasbeen small relative to the decline in number of public (G) hospitals andbeds in such hospitals. In fact, most of the decline in community hospi-tal capacity has occurred in state and local government communityhospitals. Thus, the trend has been to a privatized hospital system withsome relative increase in the share of private hospitals under for-profitownership. These changes reflect hospital closings, mergers, as well asownership changes among existing hospitals.
This study focuses on hospital ownership changes and the effects ofsuch changes. Although government or private nonprofit ownership tofor-profit ownership receives the most publicity, in fact, changes haveoccurred in all directions (Needleman et al. 1997). Relatively more at-tention has been paid to conversions from G and N to F ownership forseveral reasons. There is a concern that hospitals with a profit-seekingmission as an explicit motive are less likely to accept unprofitable casesfor treatment, such as persons who lack health insurance or who areunderinsured. Given the goal of maximizing profit, such hospitals maybe more likely to exploit loopholes in the reimbursement rules; theymay be more willing to reduce quality of care, especially quality attrib-utes that are difficult for patients, and perhaps even their physicians,to monitor ("noncontractable quality"; see, for example, Hart et al.1997).
Allegations of adverse effects associated with ownership conversionare easily made. But obtaining rigorous empirical support for such alle-gations is a much more difficult matter. Any in-depth study of the ef-fects of ownership change on access to and quality of care and onbusiness practices should account for the following factors.
Year
Figure 5.1Number of hospitals in the United States
Total community hospitals
Private not-for-profit hospitals
For-profit hospitals
o Government hospitals
Source: American Hospital Association (1976, 1979, 2001). The data before 1979 are fornonfederal, short-term general and other special hospitals, which include communityhospitals plus hospitals in institutions.
First, hospitals do not change ownership in a vacuum. Hospitals thatconvert may have specific attributes that distinguish them from hospi-tals that do not convert. These attributes may be a characteristic of thehospital and/or of the market in which the hospital operates. If a hos-pital did not change ownership in the particular way proposed, whatwas the alternative course of action? The hospital industry is a matureindustry, in a sense, more like steel than e-commerce. In an industrythat is downsizing, there are rarely many attractive alternatives. Thus,even if the outcomes are worse than before, such outcomes could havebeen even worse if the choice to change ownership had not been made.The alternative to conversion may have been closure. Under such cir-cumstances, all persons in the locality may have experienced a de-crease in access to hospital care, and the loss of jobs to the communitymay have been far greater than occurred as a consequence of"efficiency measures" implemented by the acquirer. It is essential toask the "what if" question, both in policy and in empirical analysis ofeffects of ownership changes.
Second, some of the observed changes reflect change in ownershipper Se, rather than the effects of change in type of ownership. This is
o000
00
AA
00000000000000000000000 00000000000000000
0 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
Hospital Ownership Conversions 125
7,000
6,000
5,000
4,000
0
3,000
2,000
1,000
0
126 Sloan
Year
Figure 5.2Number of hospital beds in the United States
Total ooiranunity hoepitalsDPdvate no44or-pm0t hoapUls
For-ptc4it hoaØas
OGøveaiment hoepilal
Source: American Hospital Association (1976, 1979, 2001). The data before 1979 are fornonfederal, short-term general and other special hospitals, which include community
particularly true during the first few years following a conversion, ashospitals adjust to new management and strategies.
A third point pertains to policy adoption based on empirical evi-dence of undesirable outcomes following specific types of hospitalownership conversions. If changes in mission and in behavior are ob-served, are there more direct and efficient policy instruments for ensur-ing desirable behavior than simply blocking a certain type ofconversion? For example, if fraud and abuse in hospital reporting ofpatient information for purposes of reimbursement is a widespreadpractice, a more direct approach would involve direct public oversightand enforcement rather than indirectly affecting such behavior byinfluencing the mix of hospitals according to their propensities to en-gage in undesirable behaviors. As discussed below, some researchershave found that profit-seeking behavior is contagious. That is, whenfor-profit hospitals engage in certain kinds of behavior, their hospitalcompetitors with different ownership forms may emulate it.
There is a lot of empirical literature on the relationship between hos-pital ownership and various performance measures (see, for example,Sloan 2000). In general, the literature reveals that private hospitals,
..............
.
o
A
0
A
0
A AAAAAAAAAAAAAA
0000000000000
1,200
1,000
8000
a.0
0
26z
600
400
200
0
Hospital Ownership Conversions 127
whether they are for-profit or nonprofit, are more alike than different.The vast number of studies, however, have assessed the effects of own-ership type on hospital behavior rather than ownership conversions.The latter may be particularly insightful because they discuss locationand characteristics that potentially influence behavior associated withlocation constant and examine changes that occurred post- versuspreconversion. Norton and Staiger (1994) found differences in hospitalbehavior by ownership status, but the differences were attributable towhere for-profit versus other types of hospitals decided to locate. Forexample, a profit-seeking hospital may decide to locate in a relativelyaffluent suburb, where well-insured persons are located, rather thanin an inner city, where people are much more likely to be uninsuredor to be enrolled in Medicaid (one of the less generous third-partypayers).
The dearth of studies on hospital ownership conversions reflects thedifficulty of obtaining accurate data on conversion dates and conver-sion types. This problem has been remedied in the present study and insome studies reviewed below. After comparing ownership codes fromtwo independent sources, the Annual Survey of Hospitals (conductedby the American Hospital Association) and Medicare Cost Reports, mystudents called hospitals for which the two sources differed to deter-mine whether the conversion occurred, when it occurred, and the own-ership types before and after conversions. In total, about 300 telephonecalls were made.
Compared to the number of quantitative studies, there have beenvery few qualitative studies of ownership and ownership conversions.Qualitative studies may be called "soft" but at the same time, they canreveal differences among and changes in decision-making processesthat otherwise can only be inferred very indirectly from outcomechanges. By peering inside hospital decision making, especially deci-sion making in the presence of major stress and organizational change,we can enhance our understanding of these changes. Such analysis alsoprovides an important cross-check on findings from the quantitativeanalysis.
Qualitative analysis is not done mostly because it is so difficult to do.Decision makers undergoing change and/or associated with choicesthat did not lead to a successful outcome are not likely to want to havetheir decisions and the consequences scrutinized. Also, there is no ex-plicit hypothesis testing, as in quantitative research. In the end, onegains an impression without the sharply defined results of a rejectednull hypothesis.
128 Sloan
In this paper, I review and evaluate very recent research (mostlypublished in 2000 or later) on the relationship between hospital owner-ship and behavior, and I present new empirical results on ownershipconversions based on empirical analysis of hospital discharge data forthe years 1988 through 1996, the Healthcare Cost and Utilization Pro-ject's (HCUP) Nationwide Inpatient Sample (NIS). HCUP is an intra-mural program of the U.S. Agency for Healthcare Research andQuality.
Using qualitative as well as quantitative evidence, the public policyquestions addressed in this paper are those posed above. Does owner-ship conversion, especially to the for-profit ownership form, lead to in-creased care barriers, diminished quality of care, and profit-seekingbilling practices designed to maximize reimbursement? Do acquirers ofhospitals pay prices for the facilities that are commensurate with earn-ing a fair rate of return on their investments? Or do acquirers pay toolittle for community assets in the form of hospitals? If there is evidenceof market failure in this context, what are the appropriate publicremedies?
Section II of this paper reviews empirical evidence on the effects ofownership and ownership conversions on behavior. Section III de-scribes a new investigation of the effects of ownership conversionsfrom G and N to F ownership and the reverse, from F to G or N. To theextent that hospital decision making becomes more profit-orientedwhen hospitals convert from G or N status, one should observe the op-posite when hospitals convert from F to G or N status. Thus, by exam-ining the effects of conversions in both directionstoward and awayfrom for-profit statusthe empirical analysis makes it possible to dis-tinguish the effects of change of ownership type from the effects of con-version per Se. Section IV presents results from a sample of Medicarebeneficiaries and Section V results from a sample of patients not yetage-eligible for Medicare. Section VI further evaluates the findings andexplores the implications of the results for public policies as they relateto hospital ownership conversions.
II. New Empirical Evidence on Ownership Effects and on HospitalConversions
Why Hospitals Convert
Using qualitative as well as quantitative methods, several recent stud-ies have analyzed why hospitals change ownership status. The major
Hospital Ownership Conversions 129
determinants of ownership change are financial distress prior to thechange, mostly due to a reduction in aggregate demand for hospital in-patient care and an increase in competition among hospitals, which ne-cessitates implementation of new competitive strategies that weremore difficult to introduce under the prior regime. A study of five pub-lic hospitals that converted from public to nonprofit ownership gavetwo reasons for the changes: (1) a reluctance of public facifities to ex-pand beyond their political jurisdiction and (2) governance and man-agement restrictions that made it more difficult to compete with otherhospitals (Legnini et al. 2000).
Burns et al. (2001) performed case studies of sixteen hospitals thatchanged ownership between 1993 and 1996. Among the sixteen hospi-tals were conversions from private nonprofit to for-profit, for-profit tononprofit, public to for-profit, and public to nonprofit ownership. Thecase studies revealed two primary motivations for conversion regard-less of ownership type, even though the two both come down tomoney. First, with one exception, the hospitals reported current or an-ticipated financial stress as a motivation. By joining a chain or reducingmanagerial constraints in the case of public hospitals, the hospitalshoped to improve financial performance. A motivator for the hospitalsconverting from nonprofit to for-profit governance was access to equitycapital. In one case, the hospital was not considered to be "strategicenough" to justify further investment by its nonprofit owner, but thehospital fit the strategy of a for-profit purchaser. In the case of for-profitto nonprofit conversions, the hospital had not met the former owners'required financial return. Reasons for the public to private conversionsincluded poor hospital financial management under the previous own-ership. In the extreme, the alternative for the hospital was closure.
Second, nearly all the hospitals said that they were faced with an in-ability to compete for managed-care contracts. By joining a larger, moredominant hospital in the local market, the hospital was in a strongerposition when dealing with managed-care organizations. A desire tochange its mission was a motive only rarely for an ownership conver-sion. Transitions to for-profit ownership were motivated by both "pull"and "push" factors. The former included an attractive financial offer;the latter included siphoning off hospital cash flow by the formerowner. Transitions from for-profit ownership to the other forms tendedto be motivated by push factors, including failure to realize a requiredreturn.
In a review of conversion financing, Robinson (2000) reported thatmost conversions from private nonprofit to for-profit status are
130 Sloan
initiated by the nonprofit trustees and thus resemble friendly lever-aged buyouts. At the same time, many nonprofit hospitals have cre-ated for-profit subsidiaries and continue to function as nonprofitorganizations.
Conover et al. (2001) provide a recent study of the determinants ofhospital closings, mergers, and ownership changes. Hospitals that con-verted experienced on average a major decline in financial perfor-mance in the years immediately preceding conversion. In this sense,converting hospitals were not typically successful businesses prior toconversion. In the absence of a change in ownership, some other out-come, including closing, was often inevitable.
What Is Being Bought and Sold?
The financial transaction involves a purchase or lease price, and muchmore. The transaction may be parsed into two elements, the price andprovisions of the transaction other than price.1 On price, the issue iswhether or not acquirers pay a fair price for the facility, given dis-counted cash flows that are likely to accrue from the transaction. In thevast majority of transactions in the general economy, the underlying as-sumption is that buyers and sellers are sufficiently informed to allowthe market to set the transaction price. In this context, there is a concernthat sellers may not be sufficiently well informed and may obtain toolittle in return for relinquishing a major community asset. On thenonprice dimensions, there is a concern that acquirers will be driven byprofit considerations, and provision of public goods by hospitals, suchas provision of care to uninsured persons, public health programs, andeducation and research other than that sponsored by an outside publicor private organization, will be reduced as a consequence.
Much of the evidence on these two questions comes from case stud-ies of ownership conversions. To date, the literature has provided noclear answers to the price question. As far as the nonprice question isconcerned, the evidence is somewhat reassuring but preliminary.
The price question has been addressed by two recent empirical stud-ies. Conover and Sloan (2001) assessed rates of return on a sample ofhospital purchases starting in the early 1980s. Wide variations in ratesof returns were observed, with conversions to for-profit status exhibit-ing higher rates of return than other types of changes in hospital own-ership. On average, rates of return were well above the cost of capitalwhen hospitals converted to for-profit status, but somewhat closer tothe cost of capital otherwise.
Hospital Ownership Conversions 131
In an earlier study, rates of return tended to be closer to the cost ofcapital, but that study was limited to ownership conversions that oc-curred in North and South Carolina and had a shorter postconversionfollow-up period, meaning that more of the measured return wasbased on projected rather than actual cash flows (Sloan et al. 2000).Methodologies used in the two studies were similar. A weakness ofboth studies is that it was possible to make only crude adjustments fornonprice concessions granted by buyers to sellers.
On the second question, it is necessary to glean bits of evidence fromqualitative studies. Blumenthal and Weissman (2000) provided casestudies of the sales of three teaching hospitals to investor-owned hospi-tal chains, focusing in particular on the effects of the sales on the orga-nizations' medical education missions. In all three, there were noadverse effects. The authors attributed the lack of an effect to three fac-tors. First, the for-profit purchasers considered preservation of someunprofitable activities a cost of doing business at these institutions.The contracts of sale stipulated that specific resources be devoted toteaching, resource, and charity care. Second, private subsidizationof medical education may not be that burdensome to the new owners,given external subsidies such as for graduate medical education. Third,at the time the study was written, all three institutions were doingwell financially. If faced with financial stress, their missions couldchange.
Cutler and Horwitz (2000) studied conversions of the Wesley Medi-cal Center in Wichita, Kansas, and HealthOne in Denver, Colorado.Both were large teaching facilities. The first was purchased by HospitalCorporation of America in 1985. Two foundations were funded withproceeds from the sale. In the HealthOne transaction, the hospital en-tered into a joint venture with Columbia/HCA in 1995. After the trans-action, the new HealthOne organization concentrated on graduatemedical education, paying faculty and residents and administeringmedical education at its facilities. Cutler and Horwitz found improve-ments in financial performance after the conversion and the missionprevious to ownership conversion was maintained. Improved financialperformance came in part from the skill of the for-profits at increasingpublic sector reimbursements, not solely from efficiency gains.
Outcomes following conversion are not uniformly favorable. Burnset al. (2000) described "the Allegheny system debacle." This is a casestudy involving the Allegheny Health, Education and Research Foun-dation, which consisted of a major teaching facility in western Pennsyl-vania and several afffliate hospitals throughout the state. The authors
132 Sloan
attributed failure in part to failure of external oversight mechanisms,including lack of performance of the organization's board, accountants,and auditors; bond-rating agencies; and state government. There wasambiguity regarding the powers of the state attorney general, state poi-itics, and jurisdictional issues with federal bankruptcy court. Pennsyl-vania law is ambiguous regarding the attorney general's power overtransactions with nonprofits.
The sixteen case studies reported in Burns et al. (2001) revealed that,in all but one conversion, the financial status of the hospital improvedafter the conversion. There were funds from the sale or lease of the fa-cility New hospital owners invested in hospital plant and equipment,particularly in hospitals converting from private nonprofit to for-profitstatus, although in some cases, the investment after the conversion wasnot as great as hospital management had anticipated. Improvements inmargins were achieved by cutting staff,2 improving purchasing prac-tices, and consolidating services lines in a network approach.
Changes in decision-making style depended mostly on whether thenew organization was a multihospital system or a freestanding facilityIn fact, the transition to being part of a larger system was on balancemore important than was the specific change in ownership form. Thosehospitals that became part of a chain lost some local autonomy in deci-sion making. But there were differences in the strategic decision-making process among chains and even in treatment of individual fa-cilities within a particular chain. Overall, the organization's generalmission remained intact, as specified, for example, in the sales contract,while the methods for achieving its objectives changed.
Anderson et al. (2001) studied changes in internal decision makingafter conversion in ownership status at the same set of convertinghospitals as in Burns et al. (2001), but they also included a comparisonwith twenty-two nonconverting hospitals matched on location, owner-ship (prior to conversion), and bed size. They found that, relativeto nonconverted hospitals, converted hospitals had greater levels ofphysician and nurse participation in hospital decision making; inthe converted group, these health professionals had greater influenceover final choices made. These agents' interconnections and interac-tions may have intensified as the organizations tried to cope with thechanges brought about by the conversion. Alternatively, grantingmore influence to professionals may have been a compromise struckto overcome professionals' resistance to change. The study assessedparticipation at three to six years following ownership conversion.
Hospital Ownership Conversions 133
Thus, the change in participation was not only transitory However, in-creased participation by health professionals is more closely relatedlogically to provision of care to individual patients than to policies af-fecting the community as a whole, such as provision of care to theuninsured.
Effects of Hospital Ownership Changes on Quality of Care
Two recent studies have assessed differences in quality of care by hos-pital ownership status. One is a comparison of differences in quality byownership. The other explicitly examined the effect of change in own-ership on quality of care. Both studies relied on Medicare administra-tive records to gauge outcomes of care. At least to some extent, bothstudies account for methodological complexities of discerning effectsof conversions discussed in the previous section. Both studies implythat conversion to for-profit status may lead to some reduction in qual-ity of care as measured by mortality rates at various times after dis-charge from the hospital.
McClellan and Staiger (2000) examined all Medicare hospital dis-charges for acute myocardial infarction for 1984-1994 and for ischemicheart disease for 1984-1991. The outcome measures were death withinninety days of admission and cardiac complications leading to re-admission. Many of the details are beyond the scope of this review.One purpose of that study was to develop a hospital-specific measureof quality with a high signal-to-noise ratio. However, the findings onhospital quality, measured in terms of patient survival, are highly perti-nent here.
First, when county-level fixed effects were included, the estimatedmortality difference between N and F hospitals fell by roughly half, im-plying that for-profit hospitals tend to locate in geographic areas wherehospital quality is not as high in general. This may be partly becausefor-profits often acquire facilities that are not doing well financially.This still left some amount of lower quality attributable to being afor-profit hospital.
Second, in the three markets they considered in detail, for-profit hos-pitals did not have higher mortality rates. In one of these markets, afor-profit firm acquired a low-quality hospital, gauged in terms of mor-tality rates, and quality at that hospital subsequently improved.
Overall, these results lend support to the conclusion that some of theresults on quality differences reflect differential patterns by hospital
134 Sloan
ownership in the location of their facilities rather than in fundamentaldifferences in hospital behavior. The authors concluded that:
[O]n average, the performance of not-for-profit hospitals in treating elderly pa-tients with heart disease appears to be slightly better than that of for-profit hos-pitals, even after accounting for systematic differences in hospital size, teach-ing status, urbanization, and patient demographic characteristics. This averagedifference appears to be increasing over time. However, this small average dif-ference masks an enormous amount of variation in hospital quality within thefor-profit and the private nonprofit categories. Our case-study results also sug-gest that for-profits may provide the impetus for quality improvements where,for various reasons, relatively poor quality of care is the norm. (p. 111)
In a comment on the McClellan-Staiger study, Wolfram (2000)stressed that survival to ninety days after the admission date repre-sents only one dimension of quality. While no informed person wouldseek admission at a hospital with a markedly higher mortality rate,other attributes such as the time that providers spend with patients, arealso plausibly important. She also noted that McClellan and Staigermay not have adequately controlled for patient selection. For example,if more severely ifi patients (in ways that the researcher cannot meas-ure) seek care at private nonprofit hospitals, the true differential inquality between these institutions and for-profits will be higher thanthe difference the researchers measured.
Several strategies deal with this issue, none of which is perfect. Oneis to include more explanatory variables, such as Wolfram's suggestionto include a binary variable for the presence of a trauma center.3 If pri-vate nonprofits are more likely to have trauma centers, they may attractthe most vulnerable heart attack victims.4 Other approaches involve in-strumental variables and difference-in-difference. In the end, moreprogress is likely to be made by supplementing statistical approacheswith case studies that identify changes in processes of care that occur inhospitals with different ownership form.
Picone et al. (forthcoming) also assessed health outcomes usingMedicare data for the years 1984-1995. They studied death at thirtydays, six months, and one year. The underlying hypothesis was thathospitals converting to for-profit ownership raise profitability after ac-quiring the facilities in part by reducing dimensions of quality notreadily observed by patients or their physician agents and by raisingprices. Hospital-specific mortality rates after discharge from the hospi-tal are one important dimension of such hard-to-observe quality. Theauthors found that one to two years subsequent to conversion to
Hospital Ownership Conversions 135
for-profit status, the mortality rate of patients at one year following ad-mission rose markedly. At the same time, hospital staffing decreased. Asimilar decline in quality was not observed after hospitals switchedfrom for-profit status to government or to private nonprofit ownershipstatus. At three or more years following conversion to for-profit owner-ship, the decline in quality was much lower (relative to the quality thatprevailed at the hospital at five years before the conversion occurred).A plausible interpretation consistent with the data is that the newfor-profit owners (or their managers) experienced adverse effects afterthe staffing cuts and reversed course at three years or so following thedate of acquisition. By adding staff, some improvements in qualitywere then realized.
Effects of Hospital Ownership Changes on Provision of Public Goods
A major policy concern is that ownership conversions, especially to thefor-profit form, will result in decreased provision of public goods. Mostfrequently mentioned among such public goods is provision of care tothe uninsured, sometimes cast in terms of provision of uncompensatedcare. The empirical evidence on this score is mixed. Comparing provi-sion of uncompensated care across ownership types, one is struck bythe similarity between shares of dollar amounts of uncompensated carerelative to hospital revenue provided by N and F hospitals. Govern-ment hospital uncompensated shares, not surprisingly, tend to behigher than for N and F hospitals (Sloan 2000).
A recent study, which focused on ownership conversions, calls theconclusion that the main distinction in provision of uncompensatedcare is between private and public hospitals into question. Using un-published and confidential data on revenue from the American Hospi-tal Association, Thorpe et al. (2000) studied the effects of conversionfrom N to F status on hospital provision of uncompensated care. Theymeasured uncompensated care as bad debt and charity care chargesdeflated by each hospital's cost-to-charge ratio for 1991-1997. To ac-count for variations in hospital capacity and inflation, they divided thisamount by total expenses for the hospital. They included several ex-planatory variables in their analysis, most notably hospital fixed ef-fects. They found little effect of conversion to for-profit status onMedicaid patient loads. However, uncompensated care fell after con-version from N to F status, falling from 5.3 percent to 4.7 percent of hos-pital revenue on average. The 0.6 percentage-point reduction in
136 Sloan
uncompensated care amounted to about $400,000 less being spent perconverted hospital on uncompensated care. For public hospitals con-verting to for-profit status, the decrease was greater, from 5.2 to 2.7 per-cent, or $800,000, per hospital. The authors expressed the amounts interms of admissions lost even though they provided no indicationwhether the decrease occurred on the inpatient or outpatient sides ofthe hospital or in some combination thereof. Thorpe et al. concludedthat:
Of concern, however, is our findings that the provision of uncompensated careis reduced when hospitals convert to for-profit status. Of particular concern isthe large reduction in uncompensated care observed among public to for-profithospital conversions. Because the bulk of these conversions occurred amongsmaller, rural public hospitals, such conversions could limit access to hospitalcare among the uninsured. (p. 192)
The implication is that conversions cause reductions in uncompen-sated care and that these reductions could not be accounted for by in-ward shifts in demand largely exogenous to the hospitals involved inthe conversions. The authors noted that their result for nonprofit tofor-profit conversions had not been found in previous studies (see, forexample, Desai etal. 2000). I shall return to this issue in the next sectionwhen I discuss my own empirical analysis.
Several recent studies have focused not on the effect of ownershipchange on the provision of public goods and more generally on prod-uct mix, but more specifically on the effects of competition fromfor-profit hospitals on the behavior of nonprofit and public hospitals.5The Disproportionate Share (DSH) Program was implemented nation-ally to provide a greater financial incentive for hospitals to deliver careto the poor. Subject to this federal law, each state could design its ownDSH program.
Duggan (2000a, 2000b) studied the change in financial incentivescreated by California's variant of the DSH program on the propensityof hospitals to treat Medicaid recipients. This DSH program providedan explicit financial incentive for hospitals to admit Medicaid patients.It increased revenues to hospitals for which low-income patientsconstituted more than 25 percent of their patients. Hospitals abovethis threshold experienced a revenue increase, and those below thisthreshold had an incentive to increase their low-patient shares to thislevel.
Duggan (2000b) found an appreciable difference in response be-tween public and private hospitals, regardless of whether they were N
Hospital Ownership Conversions 137
or F. The public hospitals faced a "soft budget" constraint; that is, astheir revenues from DSH increased, their sponsors lowered their subsi-dies accordingly. By contrast, both N- and F-owned facilities could ac-cumulate wealth from this new revenue source. He concluded that thegreatest difference in response to the DSH incentive was between pub-lic and private rather than between N and F hospitals.
In the second study, Duggan (2000a) found that DSH resulted in ashift of Medicaid patients from public to private hospitals. The magni-tude of the shift was directly related to the market share of for-profithospitals in the county In particular, the response of N hospitals to theDSH incentive was greater when they faced more competition from thefor-profit sector. The implication is that N hospitals behave more likeprofit-maximizers when faced with the market discipline of thefor-profits. Various tests that Duggan performed rejected alternativeexplanations for his finding (for example, quality of care in the publicfacifities). An examination of the effects of competition from F hospitalson the board composition of N hospitals revealed that N hospitalboards had larger shares of physicians when they faced competitionfrom the F hospitals. Duggan reported that F boards contained largenumbers of physician members.6 In this sense, N boards inF-influenced areas may put physicians on the boards as their missionbecomes more profit oriented. However, other interpretations seem atleast as likely. For one, placing physicians on the board may be a com-petitive response by N hospitals to retain their medical staffs.
Silverman and Skinner (2001) assessed the effects of hospital compe-tition on mission, but from a different perspective. Since implementa-tion of the Medicare Prospective Payment System (PPS), hospitals haveknown that the pattern of reporting of diagnoses and procedures canaffect the diagnosis-related group (DRG) assigned to the case andhence the amount of revenue received from Medicare. Upcoding in-volves rearranging reports of diagnoses and/or procedures, with theresult that the patient falls within a higher-priced DRG. Some upcod-ing may be perfectly legitimate. In some cases, it may improve accu-racy of reporting. But given the financial incentive to upcode, there is alarge gray area. More profit-oriented hospitals may be more willing totake advantage of the incentive to increase revenue from Medicare.Even for those hospitals that are not fully profit oriented, pressuresfrom competition may force them to act in this way. The authors lim-ited their analysis to hospital admissions involving pneumonia and re-spiratory infections. Between 1989 and 1996, the number of the most
138 Sloan
expensive DRG in this diagnosis family rose by 10 percentage pointsamong stable N hospitals, 23 percent among stable F hospitals, but 37percentage points among N hospitals that converted to F status.
Silverman and Skinner obtained two major findings. First, holdingmany other factors constant, for-profit hospitals were more likely toupcode in this diagnostic category than most nonprofit and govern-ment hospitals. Second, the authors found evidence that upcodingamong N hospitals was much more likely when they faced greatercompetition from F hospitals. By contrast, the upcoding behavior of thefor-profits was not affected by the presence of nonprofits.7 During thelatter part of the 1990s, Medicare became more aggressive in monitor-ing hospital billing practices, with the consequence that some hospitalsmade large payments to compensate for shortcomings in past billingpractices.8
III. Evidence from a National Sample of Hospital Discharges
Objectives of the Analysis
To assess further the effect of hospital ownership conversions on qual-ity, patient mix, especially willingness to treat publicly insured and un-insured persons, and upcoding, I assessed hospital discharge data fromthe Healthcare Cost and Utilization Project (HCUP) Nationwide Inpa-tient Sample (NIS). Results from this analysis are reported in this studyfor the first time. This data set offers many advantages. There is a verylarge number of observations, and the data come from many states andare available for several years. The hospital, but of course, not a patientidentifier, is provided. The main disadvantage of the data is that no in-formation is available on patients after they were discharged from thehospital. Also, there is no information on hospital outpatient care.
Data and Methods
The NIS is a compilation of data from state hospital discharge data-collection systems made available through the U.S. Agency for Health-care Research and Quality I limited the analysis to hospital admissionsoccurring during the years 1988-1996. Data from the following stateswere included for some or all of the observational period: Arizona, Cal-ifornia, Colorado, Connecticut, Florida, Hawaii, Illinois, Iowa, Kansas,
Hospital Ownership Conversions 139
Maryland, Massachusetts, Missouri, New Jersey, New York, Oregon,Pennsylvania, South Carolina, Tennessee, Utah, Washington, and Wis-consin. Each year, the NIS provides discharge abstract data from about5 to 7.1 hospital stays from over 900 hospitals per year. The NIS is de-signed to approximate a 20-percent sample of U.S. community hospi-tals. The NIS is not designed to be representative of communityhospitals in terms of ownership. Of the states included in the NIS, Cali-fornia and Florida had the most admissions, by far, to for-profit hospi-tals. Tennessee has a high for-profit market share but is a much smallerstate and was included in the NIS only for part of the observational pe-riod. Texas, another state with a high for-profit market share, was notincluded in the NIS.
Separate analyses were conducted on admissions of (1) persons overthe age of 65 at admissionhereafter called the "Medicare sample"and (2) births and admissions of persons who were between the ages of1 and 64 on the admission datethe non-Medicare sample (although aminority of persons in the under-age-65 group also had Medicare asthe primary payer). For the Medicare analysis, the sample was limitedto persons who were admitted for one of five primary diagnoses:stroke, hip fracture, coronary heart disease, congestive heart failure,and pneumonia. To facifitate analysis, the NIS 10 percent sample wasused with one exception. In the analysis of discharges of persons whowere between the ages of I and 64 on the admission date, I used a 25percent sample of NIS's 10 percent patient sample.
The Medicare and non-Medicare samples were each divided intotwo subsamples. The first consisted of admissions to hospitals thatwere under government or private nonprofit ownership prior to con-version and changed to for-profit ownership or remained under gov-ernment or private nonprofit ownership during the observationalperiod, 1988-1996 (the GN sample). The second included admissions tohospitals that were under for-profit ownership but converted to GNstatus during the observational period, as well as those that remainedfor-profit (the F sample).
Six hypotheses about effects of ownership conversions from GN to Fand F to GN status were tested. Conversions from F to GN ownershipwere hypothesized to have the opposite effect of conversions from GNto F. By employing a two-sided test, I was able to distinguish betweenthe effects of conversion and those attributable to a change in owner-ship. In combining the G and N categories, I focused on changes in be-
140 Sloan
havior to and from the for-profit ownership form rather than on theprivate versus public ownership distinction. The six hypotheses arelisted below:
Quality of care falls following conversion from GN to F ownership.This may occur because profit-seeking hospitals seek to achieve higherprofitability by cutting hospital inputs, such as personnel.
Quality may be increased, but only when it is profitable to do so,that is, when higher quality results in an additional payment sufficientto cover the additional marginal cost of the higher quality level.
Following conversion from GN to F status (when subject to a fixedpayment per case, as under the Medicare Prospective Payment Sys-tem), the hospital becomes more aggressive in reducing length of stay.Again, the motive for reducing stays is to increase profitability
Increasing transfers to nursing homes postdischarge is one manifes-tation of a strategy for reducing the length of stay.
To increase payment from Medicare, the hospital upcodes diagnosesmore frequently following a conversion from GN to F status. This is anextension of Silverman and Skinner's (2001) research, but for a differ-ent set of diagnoses.
Following conversion from GN to F ownership, the hospital be-comes less likely to admit publicly insured and uninsured patients.This analysis used only the non-Medicare sample and sought to repli-cate the finding by Thorpe et al. (2000) that conversion to for-profitownership led to reductions in hospital provision of uncompensatedcare.
For the analysis of the admissions of persons aged 65 and over, thedependent variables were inpatient mortality; extended length of stay;length of stay; pneumonia complication; destination at dischargehome (omitted reference group), nursing home, other hospital, ordeath; and upcoding of diagnoses. Extended length of stay was definedas a stay two standard deviations above the mean stay for the primarydiagnosis and the year in which the admission occurred. The pneumo-nia complication was taken from the list of secondary diagnoses pro-vided on the hospital discharge abstract. In this analysis, elderlypersons admitted with a primary diagnosis of pneumonia wereexcluded.
Upcoding was tested by two analyses. In the first, the dependentvariable was 1 if the primary diagnosis was listed as a transient
Hospital Ownership Conversions 141
ischemic attack (TIA) and 0 if the primary diagnosis was listed as astroke. Medicare pays more for strokes than TIAs. There is some discre-tion in classifying patients between these two diagnoses. The secondwas the DRG weight assigned to the case, limiting the sample to thefive primary diagnoses listed above.
A sample of births was analyzed to test the sixth and second hypoth-eses. The dependent variables were patient did not have private insur-ance, patient stayed in the hospital for less than two days, and patienthad a vaginal birth as opposed to a cesarean sectionthe underlyingpresumption being that C-sections are more profitable on average thanare vaginal births. With the sample of births, the dependent variablewas payment source other than private insurance versus private insur-ance. The other category included Medicaid, Medicare or other govern-ment insurance (such as Veterans Administration, Champus), self-payor no charge (suggesting no health insurance coverage), and private in-surance (the omitted reference group). With the sample of persons withany diagnosis or principal procedure who were between the ages of 1and 64 at the admission date, the dependent variables were each of thepayer categories for public payers and self-pay/no charge with pri-vately insured individuals, the omitted reference group.
With the exceptions noted below, four alternative specifications wereemployed. The methodology for the Medicare analysis is explained indetail here. Specifications for the non-Medicare were similar. Each suc-cessive specification added a set of explanatory variables, retaining theexplanatory variables from the previous specification.
The first specification included explanatory variables identifying ad-missions after ownership conversion occurred, patient characteristics,market characteristics, and binary variables for the year of admission.The patient characteristics were age; race; gender; source of admis-sionemergency room, nursing home, other hospital versus home; bi-nary variables for the five primary diagnoses; DRG weight; and theDxCG score. The DxCG is a case mix measure that accounts for the pa-tient's secondary diagnoses (see wwwdxcg.com).
The market characteristics were a Herfindahi index based on bedshares, the fraction of the population enrolled in health maintenanceorganizations, population density (population per square mile), hospi-tal beds per 1,000 population, and per-capita income. For hospitals lo-cated in standard metropolitan statistical areas (SMSAs), the marketarea was assumed to be the SMSA. For hospitals located outsideSMSAs, the market area was the county.
142 Sloan
In the second specification, I added a conversion fixed effect binaryvariable. The conversion fixed effects identified admissions to hospitalsinvolved in a conversion from G or N to F, or F to G or N during the ob-servational period. In the third specification, I added explanatory vari-ables for hospital characteristics: bedsize; the number of residentphysicians per bed, a measure of commitment to medical education;the hospital's operating margin; debt-to-asset ratio; and occupancyrateall defined for the year in which the admission occurred. For hos-pitals with a negative operating margin, the operating margin was setequal to 0 and an additional binary variable, "no profit," was set to 1.Likewise, if the debt-to-asset ratio exceeded 1, the ratio was set to 1 anda binary variable was included to identify such cases. Although thehospital characteristics are possibly endogenous, there is an argumentfor including them. In their absence, the ownership and ownershipchange variables may represent other hospital characteristics, includ-ing financial distress that may be the true causal determinants of thedependent variables. By considering alternative specifications, it ispossible to gauge the sensitivity of findings to inclusion/exclusion ofthese explanatory variables.
Finally, the fourth specification added area fixed effects. These werebinary variables for the SMSA in which the hospital was located andfor non-SMSA hospitals, a binary variable measuring the hospital'sstate. When the dependent variable was more than one mutually exclu-sive alternative, I used multinomial logit analysis and did not includethe fourth specification. With area fixed effects, there were too manyparameters to estimate.
IV. Results: Medicare Analysis
Sample Characteristics
The main GN sample consisted of 419,000 hospital discharges from1,215 hospitals (table 5.1). Of these, over 16,000 hospital dischargeswere from forty-nine hospitals that changed from G or N to F owner-ship status. Slightly over 6,000 discharges were observed from thirty-five hospitals after the ownership conversion occurred. The main Fsample consisted of 56,000 discharges from 165 hospitals. Amongthese, thirty-two hospitals experienced a conversion from F to N or Gstatus during the observational period. Data were available from
Hospital Ownership Conversions 143
Table 5.1Samplesa
Sample screen Patients Hospitals
GN sample 418,831 1,215
GN to F subsample 16,354 49
GN to F after 6,050 35
F sample 56,231 165
F to GN sample 6,500 32
F to GN after 2,331 20
aDoes not include observations drawn for analysis of coding of stroke versus transientischemic attack (TIA).
twenty hospitals converting from F to N or G ownership for the periodafter the conversion occurred.
I constructed a counterfactual sample. This sample of hospital dis-charges was matched to the conversion sample with respect to baseownership (GN or F). An artificial conversion year was randomly as-signed to the admission. The frequency distribution of conversionyears in the counterfactual sample matched the frequency distributionof actual conversions.
Given the large number of hospital discharges, there were many sta-tistically significant differences between hospitals that did not convertand those that did convert, as well as for the pre versus postconversioncomparisons (table 5.2). The differences, however, were often small.For the GN sample, these are the most noteworthy differences.
Hospitals converting from GN to F status experienced increases inthe proportion of nonwhites that they admitted. Admissions throughthe emergency room fell, as did the receipt of patients from other hos-pitals. For converting hospitals, the mean DRG weight fell after conver-sion. A drop of 0.2 in the mean DRG weight for the five primarydiagnoses is substantial. By contrast, in the counterfactual comparisongroup, the DRG weight rose in the after group in contrast to the beforegroup. The mean DxCG score rose from 2.1 to 2.2 before versus afterconversion to F status. For the comparison group, the mean score rosefrom 2.0 to 2.2. Medicare payment is based on the DRG weight but notthe DxCG score. This comparison suggests something less than a mas-sive change in upcoding for purposes of obtaining higher Medicarepayments postconversion to F ownership.
Tab
le 5
.2M
ean
valu
es o
f ex
plan
ator
y va
riab
les:
con
vert
ing
and
nonc
onve
rtin
g ho
spita
ls: M
edic
are
sam
ples
All
All
Non
conv
ertin
g:N
onco
nver
ting:
Con
vert
ing:
Con
vert
ing:
nonc
onve
rtin
g co
nver
ting
pbe
fore
afte
rp
befo
reaf
ter
p
GN
sam
ple
Num
ber
of o
bser
vatio
ns40
2,47
716
,354
216,
277
186,
371
10,3
046,
050
Patie
ntA
ge77
.356
77.8
49<
.000
177
.250
77.4
79<
.001
77.6
1678
.245
<.0
01W
hite
0.88
60.
817
<.0
010.
893
0.88
0<
.001
0.87
90.
746
<.0
01M
ale
0.46
00.
453
0.11
80.
458
0.46
20.
011
0.46
10.
440
0.01
1A
dmitt
ed f
rom
ER
0.57
20.
580
0.04
20.
560
0.58
7<
.001
0.60
50.
539
<.0
01A
dmitt
ed f
rom
nur
sing
hom
e0.
031
0.03
10.
956
0.03
00.
032
<.0
010.
027
0.03
7<
.001
Adm
itted
fro
m o
ther
hos
pita
l0.
053
0.03
2<
.001
0.04
70.
059
<.0
010.
041
0.01
8<
.001
Dia
gnos
is o
f A
lzhe
imer
's0.
005
0.00
80.
000
0.00
70.
003
<.0
010.
009
0.00
50.
001
Dia
gnos
is o
f ot
her
dem
entia
0.01
40.
014
0.37
30.
020
0.00
8<
.001
0.01
60.
010
0.00
2D
iagn
osis
of
hip
frac
ture
0.08
80.
088
0.86
30.
088
0.08
90.
075
0.09
00.
084
0.16
7D
iagn
osis
of s
trok
e0.
103
0.11
20.
001
0.12
20.
082
<.0
010.
120
0.09
7<
.001
Dia
gnos
is o
f co
rona
ry h
eart
dis
ease
0.39
20.
376
<.0
010.
388
0.39
7<
.001
0.41
30.
313
<.0
01D
iagn
osis
of
cong
estiv
e he
art d
isea
se0.
244
0.24
30.
645
0.23
70.
252
<.0
010.
224
0.27
5<
.001
Dia
gnos
is o
f pn
eum
onia
0.17
20.
182
0.00
10.
164
0.18
0<
.001
0.15
30.
232
<.0
01D
RG
wei
ght
1.55
81.
493
<.0
011.
512
1.61
0<
.001
1.56
71.
367
<.0
01D
xCG
sco
re2.
136
2.14
90.
333
2.04
32.
245
<.0
012.
099
2.23
5<
.001
Hos
pita
lIn
com
e in
zip
cod
e ar
ea (
'000
)30
444.
360
2767
2.74
0<
.000
130
191
.710
3073
7.82
0<
.000
129
680.
220
2425
3.72
0<
.000
1B
ed s
ize
337.
657
237.
889
<.0
0133
7.42
533
7.91
90.
524
302.
886
127.
584
<.0
01R
esid
ents
/bed
s0.
061
0.02
0<
.001
0.05
30.
070
<.0
010.
014
0.03
0<
.001
Ope
ratin
g m
argi
n0.
022
0.04
1<
.001
0.02
20.
023
<.0
010.
015
0.08
5<
.001
Deb
t-ca
pita
l rat
io0.
461
0.64
3<
.001
0.52
40.
387
<.0
010.
695
0.55
5<
.001
Occ
upan
cy r
ate
(%)
66.2
7455
.002
<.0
0167
.110
65.3
03<
.001
58.8
7951
.803
<.0
01
Who
le s
ampl
eC
ount
erfa
ctua
l sam
ple
Con
vers
ion
sam
ple
Tab
le 5
.2 (
cont
inue
d)
Who
le s
ampl
eC
ount
erfa
ctua
l sam
ple
Con
vers
ion
sam
ple
All
All
nonc
onve
rtin
g co
nver
ting
pN
onco
nver
ting:
Non
conv
ertin
g:be
fore
afte
rp
Con
vert
ing:
Con
vert
ing:
befo
reaf
ter
p
Num
ber
of o
bser
vatio
ns
Mar
ket
402,
477
16,3
5421
6,27
718
6,37
110
,304
6,05
0
Her
find
ahi i
ndex
0.23
00.
204
<.0
010.
235
0.22
5<
.001
0.17
70.
252
<.0
01H
MO
sha
re0.
187
0.18
30.
005
0.16
80.
209
<.0
010.
162
0.21
8<
.001
Popu
latio
n de
nsity
('0
00/s
q m
)0.
844
0.68
4<
.001
0.82
40.
867
<.0
010.
670
0.70
9<
.001
Hos
pita
l bed
s/('O
OO
) po
p.0.
390
0.37
8<
.001
0.40
00.
377
<.0
010.
387
0.36
3<
.001
Inco
me
in z
ip c
ode
area
('0
00)
30.4
4427
.673
<.0
0130
.192
30.7
38<
.001
29.6
8024
.252
<.0
01
F sa
mpl
e
Num
ber
of o
bser
vatio
ns49
,731
6,50
029
,218
20,5
134,
169
2,33
1
Patie
ntA
ge77
.581
78.0
08<
.001
77.2
9777
.985
<.0
0177
.861
78.2
720.
045
Whi
te0.
880
0.87
20.
184
0.88
40.
877
0.09
00.
863
0.88
40.
048
Mal
e0.
468
0.44
0<
.001
0.47
00.
466
0.33
70.
451
0.42
10.
020
Adm
itted
fro
m E
R0.
654
0.63
50.
003
0.65
40.
654
0.88
10.
614
0.67
3<
.001
Adm
itted
fro
m n
ursi
ng h
ome
0.02
20.
020
0.26
80.
021
0.02
30.
281
0.02
10.
018
0.31
7A
dmitt
ed f
rom
oth
er h
ospi
tal
0.04
40.
036
0.00
10.
036
0.05
5<
.001
0.03
40.
038
0.39
7
Dia
gnos
is o
f A
lzhe
imer
's0.
014
0.00
40.
466
0.00
70.
002
<.0
010.
006
0.00
0<
.001
Dia
gnos
is o
f ot
her
dem
entia
0.03
60.
010
0.42
40.
016
0.00
4<
.001
0.01
30.
004
<.0
01
Dia
gnos
is o
f hi
p fr
actu
re0.
089
0.10
40.
001
0.08
60.
094
0.00
20.
101
0.10
70.
466
Dia
gnos
isof
str
oke
0.10
00.
110
0.01
60.
121
0.07
0<
.001
0.12
00.
092
0.00
1
Dia
gnos
is o
f co
rona
ry h
eart
dis
ease
0.39
30.
331
<.0
010.
398
0.38
50.
004
0.34
20.
310
0.00
7D
iagn
osis
of
cong
estiv
e he
art d
isea
se0.
249
0.25
60.
219
0.23
60.
268
<.0
010.
243
0.27
90.
002
Dia
gnos
is o
f pn
eum
onia
0.16
90.
200
<.0
010.
159
0.18
3<
.001
0.19
30.
211
0.08
2D
RG
wei
ght
1.47
41.
270
<.0
011.
440
1.52
2<
.001
1.28
41.
246
0.11
1
DxC
G s
core
2.32
52.
161
<.0
012.
200
2.50
2<
.001
2.10
62.
260
<.0
01
Tab
le 5
.2 (
cont
inue
d)
Who
le s
ampl
eC
ount
erfa
ctua
l sam
ple
Con
vers
ion
sam
ple
All
All
nonc
onve
rtin
g co
nver
ting
pN
onco
nver
ting:
Non
conv
ertin
g:be
fore
afte
rp
Con
vert
ing:
Con
vert
ing:
befo
reaf
ter
p
Hos
pita
lIn
com
e in
zip
cod
e ar
ea (
'000
)27
741.
850
3284
8.85
0<
.000
127
521.
830
2805
4.77
0<
.000
132
228.
350
3395
8.60
0<
.000
1B
ed s
ize
217.
581
155.
585
<.0
0122
1.99
021
1.46
0<
.001
170.
079
128.
925
<.0
01R
esid
ents
/bed
s0.
002
0.00
20.
004
0.00
10.
003
<.0
010.
002
0.00
4<
.001
Ope
ratin
g m
argi
n0.
080
0.03
9<
.001
0.06
40.
102
<.0
010.
037
0.04
10.
059
Deb
t-ca
pita
l rat
io0.
521
0.65
9<
.001
0.60
70.
399
<.0
010.
721
0.54
9<
.001
Occ
upan
cy r
ate
(%)
53.1
0545
.105
<.0
0153
.678
52.2
90<
.001
44.4
3148
.743
<.0
01
Mar
ket
Her
find
ahl i
ndex
0.18
00.
155
<.0
010.
182
0.17
80.
056
0.12
90.
200
<.0
01H
MO
sha
re0.
155
0.14
1<
.001
0.12
40.
199
<.0
010.
116
0.18
7<
.001
Popu
latio
n de
nsity
('0
00/s
q m
)0.
823
1.19
2<
.001
0.77
60.
890
<.0
010.
890
1.73
3<
.001
Hos
pita
l bed
s/('O
OO
) po
p.0.
414
0.35
1<
.001
0.42
70.
396
<.0
010.
353
0.34
70.
007
Inco
me
in z
ip c
ode
area
('0
00)
27.7
4232
.849
<.0
0127
.522
28.0
55<
.001
32.2
2833
.959
<.0
01
Hospital Ownership Conversions 147
In the counterfactual group, mean bed size remained about constant.Among converting hospitals, however, mean bed size fell by more than50 percent after conversion. The number of residents per bed roseamong converting hospitals, but this change was fully attributable tothe reduction in beds. Operating margins in converting hospitals rosedramatically. Before conversion, the mean margin per discharge in con-verting hospitals was 0.015. After conversion, the mean was 0.085. Bycontrast, in the counterfactual group, operating margins remainedabout constant. The debt-to-assets ratio fell for both converting andnonconverting hospitals, but much more for those hospitals that didnot convert. The mean occupancy rate fell from 57 to 52 percent forconverting hospitals, in spite of the substantial reduction in bed size.For nonconverting hospitals, however, mean occupancy decreasedonly from 67 to 65 percent. Clearly, low occupancy rates were far morecharacteristic of hospitals converting from GN to F than among hospi-tals that did not convert.
For converting hospitals, the Herfindahi index rose, suggesting a de-cline in competition. By contrast, for hospitals that did not convert, theHerfindahi index was unchanged. Personal per-capita income fell onaverage in the zip code areas in which the converting hospitals were lo-cated. But there was no change in income among nonconvertinghospitals.
Many of the patterns were similar in the F sample. For example, themean DRG weight declined slightly after conversion to GN status. Bycontrast, for hospitals in the counterfactual group, the mean DRG roseafter the conversion. The major difference was in the operating margin.Margins increased much more among nonconverting than among con-verting hospitals. As seen above, for GN hospitals, margins increasedappreciably for hospitals converting to F ownership.
The mean in-hospital mortality rate in the GN sample hospitalized innonconverting hospitals was 8.1 percent (table 5.3). For those hospital-ized in converting hospitals, the mean mortality rate was 7.9 percent.Even with this large sample, this difference was not statistically signi-ficant at conventional levels. Although mortality fell from 8.2 percentbefore to 7.6 percent after conversion in converting hospitals, this dif-ference was not statistically significant at conventional levels (p = 0.19).In the counterfactual sample, the decline was apparently larger, from8.7 percent in the counterfactual before group to 7.5 percent in the aftergroup (p < 0.0001). The decrease reflects the secular decrease in inpa-
Tab
le 5
.3M
ean
valu
es o
f de
pend
ent v
aria
bles
: con
vert
ing
and
nonc
onve
rtin
g ho
spita
ls
Who
le s
ampl
eC
ount
erfa
ctua
l sam
ple
Con
vers
ion
sam
ple
Non
conv
ertin
gC
onve
rtin
gp
Bef
ore
Aft
erP
Bef
ore
Aft
erp
GN
sam
ple
Num
ber
of o
bser
vatio
ns40
2,47
716
,354
216,
106
186,
371
10,3
046,
050
Mor
talit
y0.
081
0.07
90.
375
0.08
70.
075
<.0
001
0.08
20.
076
0.18
7E
xten
ded
stay
0.02
00.
015
<.0
001
0.01
90.
020
0.00
10.
017
0.01
20.
009
Len
gth
of s
tay
8.12
47.
982
0.08
98.
650
7.51
1<
.000
18.
538
7.03
6<
.000
1
Aliv
e at
dis
char
geN
umbe
r of
obs
erva
tions
368,
827
15,0
5319
8,80
517
2,02
29,
463
5,59
0E
xten
ded
stay
0.02
40.
021
0.00
40.
025
0.02
3<
.000
10.
025
0.01
4<
.001
Non
-pne
umon
ia p
rim
ary
diag
nosi
spa
tient
sN
umbe
r of
obs
erva
tions
333,
401
13,3
7818
0,53
115
2,87
08,
731
4,64
7Pn
eum
onia
com
plic
atio
ns0.
042
0.04
20.
847
0.04
20.
042
0.08
10.
038
0.05
00.
001
Non
-pne
umon
ia p
rim
ary
diag
nosi
spa
tient
s: a
live
at d
isch
arge
Num
ber
of o
bser
vatio
ns30
7,86
612
,438
165,
727
142,
139
8,08
94,
349
Pneu
mon
ia c
ompl
icat
ions
0.03
40.
034
0.80
30.
034
0.03
50.
075
0.03
00.
043
0.00
1
Tab
le 5
.3 (
cont
inue
d)
Who
le s
ampl
eC
ount
erf
actu
al s
ampl
eC
onve
rsio
n sa
mpl
e
Non
conv
ertin
gC
onve
rtin
gp
Bef
ore
Aft
erp
Bef
ore
Aft
erp
F sa
mpl
e
Num
ber
of o
bser
vatio
ns49
,731
6,50
029
,218
20,5
134,
169
2,33
1
Mor
talit
y0.
075
0.08
30.
026
0.08
00.
068
<.0
001
0.08
70.
076
0.10
2
Ext
ende
d st
ay0.
034
0.03
50.
701
0.03
40.
034
0.99
20.
026
0.05
1<
.001
Len
gth
of s
tay
7.01
87.
372
0.25
67.
489
6.34
6<
.000
16.
982
8.07
10.
198
Aliv
e at
dis
char
geN
umbe
r of
obs
erva
tions
45,9
885,
953
26,8
5119
,137
3,79
92,
154
Ext
ende
d st
ay0.
037
0.03
40.
156
0.03
80.
036
0.13
40.
026
0.04
7<
.001
Non
-pne
umon
ia p
rim
ary
diag
nosi
spa
tient
sN
umbe
r of
obs
erva
tions
41,3
135,
201
24,5
4216
,771
3,36
31,
838
Pneu
mon
ia c
ompl
icat
ions
0.03
70.
044
0.01
10.
0.35
0.04
00.
006
0.04
50.
044
0.82
9
Non
-pne
umon
ia p
rim
ary
diag
nosi
spa
tient
s: a
live
at d
isch
arge
Num
ber
of o
bser
vatio
ns38
,530
4,80
822
,770
15,7
603,
092
1,71
6
Pneu
mon
ia c
ompl
icat
ions
0.03
00.
035
0.06
40.
028
0.03
40.
001
0.03
40.
038
0.48
5
150 Sloan
tient mortality which apparently was not as favorable among hospitalsthat changed from GN to F ownership status.
The rate of pneumonia complications rose in hospitals convertingfrom GN to F after conversion. The increase was from 3.8 to 5.0 percent(p = 0.001). In the counterfactual sample, there was no change. Rates ofextended stays, as did mean length of stay, declined much more in hos-pitals that were involved in ownership conversions than in those notinvolved in a conversion, implying that hospitals converting from GNto F made a substantial effort to reduce the very long lengths of stay.Mean length of stay declined in both converting and nonconvertinghospitals, but more so in the former.
For the F sample, mortality rates declined by the same absoluteamount for those hospitals that converted versus the counterfactualsample, but those that converted had a higher mortality rate in the be-fore period. Pneumonia complication rates rose among those hospitalsthat remained for-profit, but these rates did not rise among those thatconverted from for-profit to either government or private nonprofitstatus. The rate of extended stays rose dramatically among hospitals inthe converting group, but remained constant among those that did notconvert, a very different pattern from hospitals in the GN sample.
Thus, overall, judging on the basis of mean values alone, a coupleof negative features emerge in the GN-to-F case. The in-hospital mor-tality experience was not as good among hospitals that converted fromGN to F status. Furthermore, pneumonia complication rates roseamong those hospitals that converted from GN to F and among thosefor-profit hospitals that did not convert. Length of stay was related toownership, with for-profit hospitals achieving lower lengths of stayoverall.
Effects of Ownership Type Conversions on Inpatient Mortality,Pneumonia Complications, and Extended Length of Stay
The multivariate analysis of ownership type conversion effects beginswith effects on inpatient mortality, and pneumonia complications dur-ing the stay occurring to patients admitted for stroke, hip fracture, cor-onary heart disease, and congestive heart failure (table 5.4). In the firstspecification in the GN sample, inpatient mortality fell after conversionto F ownership. The decline was 0.8 percent on average, or about 10percent relative to mean mortality. (The numbers in the table in brack-ets are marginal effectschanges in the probability of an outcome for a
Tab
le 5
.4E
ffec
ts o
f ow
ners
hip
conv
ersi
ons
on m
orta
lity,
pne
umon
ia c
ompl
icat
ions
, and
ext
ende
d le
ngth
of
stay
a
GN
sam
ple
F sa
mpl
e
Mor
talit
yPn
eum
onia
com
plic
atio
nE
xten
ded
stay
Mor
talit
yPn
eum
onia
com
plic
atio
nE
xten
ded
stay
Full
sam
ple
Aliv
eFu
ll sa
mpl
eA
live
Full
sam
ple
Aliv
eFu
ll sa
mpl
eA
live
Spec
ific
atio
n 1
Aft
er c
onve
rsio
n0.
191c
_052
1b_0
472b
0222
b0.
148
0.14
50.
342c
0244
d
(0.0
54)
(0.0
75)
(0.0
82)
(0.1
41)
(0.1
32)
(0.0
90)
(0.1
38)
(0.1
51)
(0.1
35)
(0.1
40)
[-0.
008]
[0.0
03]
[0.0
05]
[-0.
004]
[-0.
005]
[0.0
12]
[0.0
03]
[0.0
03]
[0.0
07]
[0.0
06]
Spec
ific
atio
n 2
Aft
er c
onve
rsio
n-0
.014
0260
b03
08b
_042
0b_0
434b
0.06
1-0
.081
0.00
50,
3b0.
423c
(0.0
66)
0.09
5)(0
.106
)(0
.163
)(0
.148
)(0
.108
)(0
.164
(0.1
84)
(0.1
68)
(0.1
73)
[0.0
01]
[0.0
07]
[0.0
08]
[-0.
003]
[-0.
005]
[0.0
03]
[-0.
0021
[0.0
011
[0.0
10]
10.0
11]
Fixe
d ef
fect
_o.l,
ii0.
138c
_012
0d-0
.104
-0.0
3901
71b
0.24
4'0.
148
-0.1
0901
91d
(0.0
39)
(0.0
60)
(0.0
69)
(0.0
83)
- (0
.039
)(0
.063
)(0
.097
)(0
.113
)(0
.111
)(0
.113
)
[-0.
007]
[-0.
003]
[-0.
002]
[-0.
001]
[0.0
01]
[0.0
09]
[0.0
05]
[0.0
03]
[-0.
001]
[-0.
004]
Spec
ific
atio
n 3
Aft
er c
onve
rsio
n-0
.018
0.21
1c0.
259c
-0.3
29_0
.347
c0.
064
-0.0
770.
047
00b
0.40
9c
(0.0
40(0
.097
)(0
.108
)(0
.166
)(0
.152
)(0
.110
)(0
.150
)(0
.169
)(0
.167
)(0
.173
)
[-0.
001]
[-0.
006]
[0.0
06]
[-0.
003]
[-0.
004]
[0.0
03]
[-0.
002]
[0.0
002]
[0.0
10]
[0.0
10]
Fixe
d ef
fect
j126
b-0
.188
"0.
175c
-0.0
250.
037
0.16
4'-0
.113
-0.0
23-0
.026
- 0.
114
(0.0
40)
(0.0
61)
(0.0
70)
(0.0
86)
(0.0
73)
(0.0
66)
(0.1
67)
(0.1
87)
(0.1
18)
(0.1
19)
- [-
0.00
7][-
0.00
4][-
0.00
3][-
0.00
02]
[0.0
01]
[0.0
09]
[-0.
003]
[-0.
0004
][-
0.00
1][-
0.00
2]
Tab
le 5
.4 (
cont
inue
d)
aSpe
cifi
catio
n 1
incl
udes
pat
ient
's c
hara
cter
istic
s an
d m
arke
t cha
ract
eris
tics
and
year
dum
mie
s. S
peci
fica
tion
2 in
clud
es s
peci
fica
tion
1 pl
us h
ospi
tal
conv
ersi
on f
ixed
eff
ect (
show
n). S
peci
fica
tion
3 in
clud
es s
peci
fica
tion
2 pl
us h
ospi
tal c
hara
cter
istic
s. S
peci
fica
tion
4 in
clud
es s
peci
fica
tion
3 pl
us a
rea
fixe
d ef
fect
s. S
tand
ard
erro
rs in
par
enth
eses
and
mar
gina
l eff
ects
in b
rack
ets.
Sta
ndar
d er
rors
cor
rect
ed f
or h
eter
osce
dast
icity
.bS
igrM
ican
t at 1
% le
vel (
two-
tail
test
).cS
igni
fica
nt a
t 5%
leve
l (tw
o-ta
il te
st).
dSig
pjfi
cant
at 1
0% le
vel (
two-
tail
test
).
GN
sam
ple
F sa
mpl
e
Mor
talit
yPn
eum
onia
com
plic
atio
nE
xten
ded
stay
Mor
talit
yPn
eum
onia
com
plic
atio
nE
xten
ded
stay
Full
sam
ple
Aliv
eFu
ll sa
mpl
eA
live
Full
sam
ple
Aliv
eFu
ll sa
mpl
eA
live
Spec
ific
atio
n 4
Aft
er c
onve
rsio
n0.
029
0196
d0.
264c
_031
0d0.
337'
-0.0
31-0
.115
0.07
905
33b
0367
d
(0.0
70)
(0.1
01)
(0.1
12)
(0.1
72)
(0.1
57)
(0.1
27)
(0.1
95(0
.217
)(0
.199
)(0
.206
)[0
.002
][0
.005
][0
.006
][-
0.00
21[-
0.00
31[-
0.00
2][-
0.00
21[0
.001
][0
.012
][0
.009
1
Fixe
d ef
fect
O.lO
9-0
.096
0.10
70.
119
0151
d0.
170c
0.12
6-0
.070
0.04
1-0
.024
(0.0
45)
(0.0
68)
(0.0
78)
(0.0
98)
(0.0
83)
(0.0
79)
(0.1
23)
(0.1
45)
(0.1
43)
(0.1
44)
[-0.
006]
[-0.
002]
[-0.
002]
[0.0
01]
[0.0
021
[0.0
09]
[0.0
03]
[-0.
001]
[0.0
07]
[-0.
001]
N41
7,85
034
6,76
532
0,29
241
8,77
338
3,86
456
,864
46,5
1443
,338
56,2
1751
,941
Hospital Ownership Conversions 153
change in binary or a unit change in explanatory variables for continu-ous variables.9) In the remaining specifications, however, there was nochange in inpatient mortality after conversion to F ownership. The re-sults for the conversion fixed effect in specifications 2 to 4 imply thathospitals that converted from G or N to F ownership tended to havelower mortality rates. That is, if anything, the new owners selected GNhospitals with relatively good mortality records.
In the F sample, the first specification implies a mortality increase af-ter conversion to G or N status with an associated marginal effect of0.012 (p = 0.01). However, the after conversion parameter becomes sta-tistically insignificant in the other, more complete specifications. Ingeneral, the results on the key parameters of interestthe binary vari-ables for after conversion and the conversion fixed effectare quitesensitive to changes in equation specification. Thus, again, ownershipconversion had no effect on inpatient mortality
The proportion of patients with very long stays declined in the G andN to F case and rose for conversions in the opposite direction, support-ing the descriptive results presented above in table 5.3. Pneumoniacomplications rose after conversion among patients admitted to hospi-tals that converted to F ownership. The reverse occurred among pa-tients admitted to hospitals that converted from F ownership. Thisconclusion holds for the full sample of Medicare patients in the fourprimary diagnosis categories (excluding those admitted with a pri-mary diagnosis of pneumonia) and for a sample that excludes such pa-tients who died during their stays.
To conserve space and permit focus on the key parameter estimatesand associated marginal effects, table 5.4 does not include most of theparameter estimates in the model. Table 5.5 shows complete specifica-tions from the GN sample for four dependent variables shown previ-ously, but the table does not show area fixed effects. Almost all the pa-rameter estimates on the patient variables are statistically significant atconventional levels and have plausible signs.
Results for the hospital and market variables are somewhat moremixed. The measures of financial distress show no consistent effects onoutcomes. For example, hospitals with an operating loss (no profit) ex-perienced higher rates of pneumonia complications and a higher meanlength of stay, but there were no effects on either inpatient mortality orthe rate of extended stays. Death rates were higher in larger hospitalsbut lower in major teaching hospitals. The result for hospital size plau-sibly reflects case mix not otherwise measured. The coefficient on ex-
Tab
le 5
.5G
N s
ampl
e: m
orta
lity,
pne
umon
ia c
ompl
icat
ion,
ext
ende
d st
ay, a
nd le
ngth
of
stay
: spe
cifi
catio
n 4a
Mor
talit
yPn
eum
onia
com
plic
atio
nE
xten
ded
stay
Len
gth
of s
tay
Coe
ffic
ient
Stan
dard
erro
rbC
oeff
icie
ntSt
anda
rder
rorb
Coe
ffic
ient
Stan
dard
erro
r1'
Coe
ffic
ient
Stan
dard
erro
r1'
Con
vers
ion
GN
to F
aft
er-0
.109
(004
5)d
0.09
6(0
.068
)0.
019
(0.0
98)
0.08
8(0
.099
)G
N to
F f
ixed
eff
ect
0.02
9(0
.070
)0.
196
(0.1
01)e
-0.3
10(0
.172
)e-0
.127
(0.1
24)
Patie
ntA
ge0.
047
(O.0
01)c
0.03
6(0
.001
)c0.
016
(0.0
02)c
0.05
7(O
.003
)cW
hite
0.10
6(0
.027
y-0
.015
(0.0
39)
-0.2
05(0
.049
)c-0
.325
(0.1
10)'
Mal
e0.
072
(0.0
13)c
0.30
7(0
.019
)c-0
.170
(0.0
26)c
-0.6
18(0
.043
)cA
dmitt
ed f
rom
ER
0.34
2(0
.016
)'0.
268
(0.0
24)c
0.04
0(0
.032
)0.
267
(0.0
57Y
Adm
itted
fro
m n
ursi
ng h
ome
0.44
3(0
.034
)'0.
535
(0.0
51Y
0.03
9(0
.072
)-0
.064
(0.1
24)
Adm
itted
fro
m o
ther
hos
pita
l0.
285
(0.0
32)c
0.14
7(0
.047
Y0.
542
(0.0
47)c
2.13
2(0
.275
YD
iagn
osis
of
Alz
heim
er's
0.34
8(0
.052
Y0.
182
(Q08
1)d
0.01
7(0
.120
)-0
.202
(0.3
13)
Dia
gnos
is o
f ot
her
dem
entia
0.13
7(0
.034
Y0.
184
(0.0
46Y
0.27
4(0
.066
)c0.
461
(0.1
46y
Dia
gnos
is o
f st
roke
1.93
5(0
.031
)c1.
087
(0.0
34)'
-0.1
45(0
063)
d1.
288
(0.1
44Y
Dia
gnos
is o
f co
rona
ry h
eart
dis
ease
0.82
7(0
.030
Y-0
.428
(0.0
34Y
-0.2
15(0
.049
)c-3
.822
(0.0
81)c
Dia
gnos
is o
f co
nges
tive
hear
t dis
ease
1.30
5(0
.030
)c0.
913
(0.0
32)c
0.05
2(0
.052
)-0
.416
(0.0
98)c
Dia
gnos
is o
f pn
eum
onia
1.40
2(0
.031
Y-
-0.
241
(0.0
51Y
0.11
9(0
.086
)D
RG
wei
ght
0.08
3(0
.004
)c0.
063
(0.0
05)c
0.33
7(0
.005
)c1.
743
(0.0
26)c
DxC
G s
core
0.42
4(0
.003
)c0.
533
(0.0
05)c
0.40
1(0
.006
)c1.
285
(0.0
22)c
Tab
le 5
.5 (
cont
inue
d)
ayea
r, a
rea,
and
indi
cato
rs o
f m
issi
ng v
alue
bin
ary
vari
able
s no
t sho
wn.
bSta
ndar
d er
rors
in p
aren
thes
es.
csig
r,jf
jcan
t at 1
% le
vel (
two-
tail
test
).ds
igrj
fica
nt a
t 5%
leve
l (tw
o-ta
il te
st).
eSiif
ican
t at 1
0% le
vel (
two-
tail
test
).
Mor
talit
yPn
eum
onia
com
plic
atio
nE
xten
ded
stay
Len
gth
of s
tay
Coe
ffic
ient
Stan
dard
erro
r'C
oeff
icie
ntSt
anda
rder
rorb
Coe
ffic
ient
Stan
dard
erro
r'C
oeff
icie
ntSt
anda
rder
ror'
Hos
pita
lB
ed s
ize
('000
)0.
079
(003
8)d
-0.2
77(0
.060
)c0.
228
(0.0
67)c
0.00
3(0
.137
)
Res
iden
ts/b
eds
-0.1
69(0
.056
)c0.
194
(0.0
75)c
-0.0
56(0
.094
)-1
.113
(0.1
52)c
Ope
ratin
g m
argi
n0.
290
(0.1
39Y
'0.
140
(0.2
33)
-0.1
36(0
.312
)0.
162
(0.3
87)
Deb
t-ca
pita
l rat
io0.
004
(0.0
40)
0.07
0(0
.059
)0.
195
(008
9)d
0.20
1(0
.125
)
Occ
upan
cy r
ate
0.11
2(0
.517
)d1
-0.1
03(0
.730
)0.
494
(1.2
31)'
1.36
0(2
.197
)c
No
prof
it0.
012
(0.0
15)
0.11
3(0
.023
Y-0
.017
(0.0
33)
0.15
1(0
.055
Y
Deb
t-ca
pita
l rat
io>
10.
038
(0.0
41)
0.10
6(0
.061
)e0.
058
(0.0
87)
0.10
6(0
.133
)
Mar
ket
Her
find
ahl i
ndex
0.03
5(0
.038
)0.
292
(0.0
54)c
-0.2
20(0
091)
d-0
.108
(0.1
11)
HM
O s
hare
-0.0
25(0
.053
)0.
078
(0.0
71)
0.17
5(0
.108
)0.
020
(0.1
82)
Popu
latio
n de
nsity
('0
00/s
q m
)-0
.014
(0.0
08)
0.00
4(0
.012
)0.
012
(0.0
12)
0.17
7(0
.034
Y
Hos
pita
l bed
s/('0
00)
pop.
-0.1
40(0
064)
d-0
.021
(0.0
87)
0.52
6(0
.160
)c0.
607
(0.1
50Y
Inco
me
in z
ip c
ode
area
(m
u $)
0.03
3(0
.592
)2.
140
(0.8
67)
-4.3
80(1
.170
)-1
6.70
0(3
.110
)
Con
stan
t-8
.743
(0.1
32)c
-8.2
16(0
.184
)c-9
.121
(0.3
42)c
-0.7
35(0
.385
)e
N41
7,83
334
6,68
841
6,42
841
2,46
2
R2
-0.
091
Pseu
doR
20.
140
0.17
10.
248
-
156 Sloan
tended stay is positive and statistically significant at conventionallevels.
Finally, to the extent that hospitals converting from GN to F statusachieved shorter lengths of stays after ownership conversion, was thisachieved by increased rates of transfers to nursing homes or to otherhospitals? To determine the answer to this question, multinomial logitanalysis of the place to which the person was discharged after leavingthe hospital was conducted (table 5.6). The four destinations werehome (reference group), death, nursing home, and other hospital. Theanalysis controlled, among other factors, for source of admission (nurs-ing home, other hospital, emergency room, home) because patientswere more likely to return to the place from which they came to thehospital. But given the number of parameter estimates in a multi-nomial format, area fixed effects were not included. As in all of theother analyses, year binary variables were included. Results shown inthe table 5.6 are based on specification 3.
In the GN sample, 8.1 percent of persons died in the hospital, 15.3percent were transferred to a nursing home, 10.1 percent were trans-ferred to another hospital, and the remaining two-thirds (65.5 percent)returned home. The pattern of discharge destinations was similar forthe F sample, with a slightly lower percentage of patients returninghome and a somewhat higher percentage being transferred to otherhospitalspossibly reflecting their smaller bed size.
Holding other factors constant, the rate of discharges to other hospi-tals increased after conversion for GN to F hospitals (relative to dis-charge to home). The associated marginal effect of 0.039 is substantialrelative to the sample mean transfer rate (about 40 percent of the sam-ple mean). In view of the dramatic drop in bed size among GN-to-Fconverting hospitals noted earlier, an increase in the transfer rate toother hospitals is not surprising. In the F sample, discharges to nursinghome increased after conversion, but the associated marginal effectswere very small.
Upcoding of Diagnoses
For the measures studied, there is no evidence that ownership conver-sion from GN to F ownership status increased the rate of upcoding(table 5.7). In the analysis of mean DRG weight of the five primary di-agnoses included in the Medicare analysis, the DRG weight fell afterconversion from GN to F. In this specification, many hospital character-
Tab
le 5
.6M
ultin
omia
l log
it an
alys
is: d
estin
atio
n of
dis
char
gea
a B
ased
on
spec
ific
atio
n 3.
Om
itted
gro
up is
des
tinat
ion
to h
ome.
Sta
ndar
d er
rors
in p
aren
thes
es a
nd m
argi
nal e
ffec
ts in
bra
cket
s.b
Sign
ific
ant a
t 1%
leve
l (tw
o-ta
il te
st).
GN
sam
ple
Dea
thN
ursi
ng h
ome
Oth
er h
ospi
tal
Freq
uenc
y8.
12%
15.2
7%10
.12%
Aft
er c
onve
rsio
n0.
012
(0.0
68)
[0.0
02]
0.08
6(0
.054
)[-
0.01
0]0.
373
(0.0
61)
[0.0
39]"
Con
vers
ion
fixe
d ef
fect
0.22
8(0
.416
)[_
0010
]b0.
181
(0.0
34)
[_Q
Q11
]b05
05(0
.042
)[_
0,03
5]b
Pseu
doR
20.
191
Num
ber
of o
bser
vatio
ns41
8,77
3
F sa
mpl
e
Dea
thN
ursi
ng h
ome
Oth
er h
ospi
tal
Freq
uenc
y7.
57%
14.7
7%10
.96%
Aft
er c
onve
rsio
n0.
095
(0.1
10)
[0.0
05]
0.15
3(0
.087
)[0
.006
]"0.
051
(0.0
89)
[-0.
003]
Con
vers
ion
fixe
d ef
fect
0.18
4(0
.068
)[0
.0l0
]'0.
070
(0.0
56)
[0.0
13]
0.01
2(0
.061
)[-
0.00
7]Ps
eudo
R2
0.19
1N
umbe
r of
obs
erva
tions
56,2
17
158 Sloan
istics, including bed size, were held constant. Thus, the results for own-ership conversion do not reflect just downsizing and loss of somesophisticated product lines. Of course, a decrease in the mean DRGmay have reflected subtle changes in case mix. To the extent that thisoccurred, it more than obscures any change in upcoding for the fiveprimary diagnoses included in this analysis that may have occurred.For coding of transient ischemic attack (TIA) versus stroke, there wasan increase in the proportion of TIAs (significant at the 10 percent level)following conversion from GN to F status. By contrast, hospitals thatconverted from F to GN did have higher DRG weights after conversion.Similarly, the proportion of cases coded as stroke rather than TIA rose.
V. Results: Patients Under Age 65
Payer Mix after Ownership Conversion
As discussed in the previous section, there is widespread concern thatownership conversions work to the disadvantage of underserved pop-ulations. However, empirical evidence on actual effects has beenmixed. Table 5.8 presents key results of a multinomial analysis of antic-ipated source of payment. Since the payment categories are confined tothe primary source of payment, they are mutually exclusive. For pa-tients who were between the ages of 1 and 64 at the admission date andwere admitted to a facility in the GN sample, the probability of havingMedicare, other government insurance, or Medicaid, or being classifiedas a self-pay/no charge patient increased after conversion to F status. Gor N hospitals acquired by F owners tended to be those that had rela-tively low proportions of Medicaid and self-pay/no charge patients, orMedicare (disabled, end stage kidney disease) patients. However, thetendency to eschew the poor and the disabled was not sustained afterconversion.
Results for hospitals converting from F to GN ownership are mixed.The proportion of self-pay/no charge patients fell after conversion toGN status. For the sample of births, holding other factors constant, hos-pitals converting from GN to F were more, not less, likely to admitnonprivately insured patients after the conversion occurred (table 5.9).This finding is very sensitive to changes in equation specification. Theconversion fixed effect is negative and the associated marginal effect issubstantial. The implication is that for-profit organizations were morelikely to acquire hospitals that were relatively oriented to privately in-
Tab
le 5
.7U
pcod
mg
of d
iagn
osis
a
TIA
ver
sus
stro
keD
RG
wei
ght
GN
sam
ple
F sa
mpl
eG
N s
ampl
eF
sam
ple
Con
vers
ion
GN
to F
aft
er0.
183
(0.0
98Y
'-0
.173
(002
3)b
GN
to F
fix
ed e
ffec
t-0
.163
(Q05
8)b
0.17
6(0
016)
b
FtoG
Naf
ter
--
-0.5
05(0
162)
b0.
102
(003
4)b
F to
GN
fix
ed e
ffec
t-
0.31
6(0
107)
b-0
.125
(0.0
23)"
Patie
ntA
ge-0
.003
(0.0
01Y
-0.0
11(0
003)
b-0
.013
(000
0)b
-0.0
08(0
001)
b
Whi
te0.
142
(0.0
37)1
)-0
.099
(0.0
99)
0.07
2(0
011)
b0.
045
(0.0
21)c
Mal
e-0
.107
(001
9)b
-0.0
53(0
.051
)0.
181
(000
5)b
0.17
3(0
011)
b
Adm
itted
fro
m E
R-0
.136
(002
2)b
-0.2
79(0
.059
)"-0
.391
(000
6)b
-0.3
02(0
.014
)b
Adm
itted
fro
m n
ursi
ng h
ome
-0.5
24(0
.066
)"-0
.802
(0.1
97)"
-0.0
47(0
014)
b-0
.054
(0.0
38)
Adm
itted
fro
m o
ther
hos
pita
l-
1.48
9(0
080)
b-1
.220
(0.1
88)"
0.67
1(0
018)
b0.
501
(004
5)b
Dia
gnos
is o
f A
lzhe
imer
's-
--0
.091
(001
0)b
-0.0
36(0
.032
)
Dia
gnos
is o
f ot
her
dem
entia
-0.1
14(0
007)
b-0
.058
(002
1)b
Dia
gnos
is o
f st
roke
--0
.443
(000
7)b
-0.5
05(0
.016
)"
Dia
gnos
is o
f co
rona
ry h
eart
dis
ease
-0.0
68(0
006)
b-0
.369
(001
4)b
Dia
gnos
is o
f co
nges
tive
hear
t dis
ease
-0.7
14(0
005)
b-0
.741
(001
3)b
Dia
gnos
is o
f pn
eum
onia
-0.5
17(0
.006
)"-0
.404
(001
6)b
Tab
le 5
.7 (
cont
inue
d)
a Y
ear,
are
a, a
nd in
dica
tors
of
mis
sing
val
ue b
inar
yva
riab
les
not s
how
n.b
Stan
dard
erro
rs in
par
enth
eses
and
mar
gina
l eff
ects
in b
rack
ets.
CSi
gnif
ican
t at 1
% le
vel (
two-
tail
test
).C
I Si
gnif
ican
t at 5
% le
vel (
two-
tail
test
).
TIA
ver
sus
stro
keD
RG
wei
ght
GN
sam
ple
F sa
mpl
eG
N s
ampl
eF
sam
ple
Hos
pita
lB
ed s
ize
('000
)-0
.507
(005
5)b
-0.3
81(0
.259
)0.
824
(0.0
16)"
1.00
9(0
065)
bR
esid
ents
/bed
s-0
.410
(008
8)b
-0.6
95(1
.488
)0.
395
(0.0
26)"
1.39
7(0
.586
)'O
pera
ting
mar
gin
-0.2
10(0
.203
)-0
.597
(0.3
93)
0.07
6(0
.049
)0.
677
(Q08
8)b
Deb
t-ca
pita
l rat
io0.
247
(0.0
53)"
-0.1
65(0
.104
)-0
.107
(0.0
14)"
-0.1
75(0
.025
)"O
ccup
ancy
rat
e (%
)0.
005
(0.0
01)"
0.00
1(0
.002
)0.
003
(000
0)b
0.00
3(0
000)
bN
o pr
ofit
0.07
7(0
.021
)"-0
.116
(0.0
71)
-0.0
65(0
.006
)"0.
018
(0.0
15)
Deb
t-ca
pita
l rat
io la
rger
than
10.
005
(0.0
58)
0.15
7(0
.092
)"-0
.034
(0.0
13)"
0.08
5(0
.020
)'
Mar
ket
Her
find
ahl i
ndex
-0.0
92(0
.041
)'-0
.004
(0.1
25)
-0.0
80(0
.012
)"-0
.083
(0.0
33)'
HM
O s
hare
-0.2
26(0
054)
b0.
093
(0.1
44)
0.17
5(0
.019
)"-0
.062
(0.0
45)
Popu
latio
n de
nsity
('0
00/s
q m
)0.
441
(008
4)"
0.84
3(0
239)
b0.
076
(003
9)b
-0.7
19(0
125)
bH
ospi
tal b
eds/
('OO
O)
popu
latio
n0.
563
(0.0
65)"
0.03
7(0
.194
)-0
.072
(0.0
19)"
-0.0
23(0
.041
)In
com
e in
zip
cod
e ar
ea (
mu
$)3.
240
(0.8
02)
-1.3
00(2
.640
)0.
562
(0.2
31)
2.15
0(0
.667
)C
onst
ant
2.94
1(0
133)
b2.
383
(0.0
42)"
0.62
5(0
340)
d-1
.457
(012
5)b
Num
ber
of o
bser
vatio
ns61
,853
8,01
541
8,77
356
,217
R2
0.03
30.
062
0.14
10.
113
Tab
le 5
.8M
ultin
omia
l log
it an
alys
is: p
ayer
type
of
patie
nts
aged
I to
64
a B
ased
on
spec
ific
atio
n 3.
Sam
ple:
25%
of
one
of th
e su
bsam
ples
pro
vide
d by
HC
UP
data
, exc
ludi
ng w
omen
giv
ing
birt
h. O
mitt
ed g
roup
is p
riva
tein
sura
nce.
Sta
ndar
d er
rors
in p
aren
thes
es a
nd m
argi
nal e
ffec
ts in
bra
cket
s.b
Sign
ific
ant a
t 1%
leve
l (tw
o-ta
il te
st).
CSi
gnif
ican
t at 5
% le
vel (
two-
tail
test
).
GN
sam
ple
Med
icar
e or
oth
ergo
vern
men
t ins
uran
ceM
edic
aid
Self
-pay
or
no c
harg
e
Freq
uenc
y7.
94%
17.6
7%19
.27%
Aft
er c
onve
rsio
n0.
244
(0.0
52)
[0.0
08]"
0.58
2(0
.057
)[0
.067
]"0.
579
(0.0
81)
[0.0
33]"
Con
vers
ion
fixe
d ef
fect
0.06
6(0
.027
)[0
.007
]'0.
412
(0.0
34)
[-0.
039]
"0.
601
(0.0
51)
[0.0
301"
Pseu
do R
20.
105
Num
ber
of o
bser
vatio
ns51
2,13
3F
Sam
ple
Med
icar
e or
oth
ergo
vern
men
t ins
uran
ceM
edic
aid
Self
-pay
or
no c
harg
e
Freq
uenc
y6.
29%
12.9
0%21
.36%
Aft
er c
onve
rsio
n0.
068
(0.0
86)
[0.0
02]
0.45
4(0
.096
)[0
.048
]"0.
266
(0.1
21)
[0.0
11]c
Con
vers
ion
fixe
d ef
fect
0.34
3(0
.059
)[-
0.04
9]"
0.32
6(0
.077
)[0
022]
b0.
210
(0.0
90)
[0.0
13]'
Pseu
doR
20.
103
Num
ber
of o
bser
vatio
ns40
,405
162 Sloan
Table 5.9Logit analysis of probability of payment other than private insurance for labor delivery,less than two-day stay, and vaginal birtha
a Based on specification 3. Standard errors corrected for heteroscedasticity. Standarderrors in parentheses and marginal effects in brackets.
b Significant at 1% level (two-tail test).Significant at 5% level (two-tail test).
d Significant at 10% level (two-tail test).
sured patients, but after conversion, there was a major increase innonprivately insured patients. The marginal effect is 0.224. Since timefixed effects were included, the marginal effect for the after-conversiondummy does not reflect a secular growth in persons not covered by pri-vate insurance. Patterns for conversions from F to GN status were quitesimilar to those from GN to F.
Changes in Length of Stays: Birth Sample
There is widespread concern that the length of hospital stays for laborand delivery have declined to unsafe levels. Such reductions are said tobe motivated mostly by a desire to increase profit. Some states have im-plemented laws to restrict the ability of health care providers to limitstays for labor and delivery to less than two days. Table 5.9 shows thatthe probability of a less-than-two-day stay declined after ownershipconversion, both for hospitals that converted from GN to F status andthose that converted in the opposite direction.
GN sample F sample
Probability of payment other than private insurance
After conversion 0.916 (0.046) [0224]b 0.884 (0.107) [0.218]"Conversion fixed effect 0.417 (0.026) [0098]b 0.187 (0.083) [_0.044]cNumber of observations 558,268 32,527Pseudo R2
0.188 0.213Probability of less than two-day stayAfter conversion 0.391 (0.056) [0025]" 0.206 (0.117) [0017]dConversion fixed effect 0.347 (0.028) [-0.023]" 0.085 (0.092) [-0.007]Number of observations 566,927 32,635Pseudo R2
0.015 0.024Probability of vaginal birthAfter conversion 0.340 (0.049) [0.025]" 0.123 (0.103) [0.0241Conversion fixed effect 0.252 (0.024) [0025]b 0.299 (0.075) [0061]bNumber of observations 567,605 32,635Pseudo R2 0.293 0.315
Hospital Ownership Conversions 163
Changes in the Probability of Vaginal Births
The proportion of vaginal births declined after hospitals convertedfrom government or not-for-profit to for-profit status (table 5.9). Forhospitals that converted from F to GN status, there was an increase inthe probability of vaginal births, but this increase was not statistically
significant at conventional levels.
VI. Discussion, Conclusions, and Implications
The main question addressed by this paper is whether or not the mar-ket for hospitals is fundamentally broke. In my review of the literatureof cross-sectional evidence on differences in hospital behavior by own-ership type, I conclude that hospitals with various types of ownershipwere more alike than different (Sloan 2000). With longitudinal data onthe same set of hospitals (panel data), the research focus has been onchanges in behavior for the same hospital following a change in owner-ship. With a panel, isolating the effect of the change in ownership canbe done more precisely.
In many respects, the empirical evidence from hospital conversionsis reassuring. In particular, in this study, I did not find that in-hospitalmortality changed as a result of changes in ownership type. Payer mix,including the proportion of persons admitted without health insur-ance, was not affected. On the whole, hospitals' missions appear to bepreserved postconversion. In large part, constancy of mission has beensafeguarded by contract provisions that are made part of the agree-ment between the buying and selling hospitals at the time of the trans-action. The analysis presented in the previous section did not revealsystematic upcoding by hospitals that converted to for-profit owner-ship status. Thus, the upcoding for pneumonia and respiratory infec-tions reported in an earlier study does not appear to generalize to otherdiagnoses.
However, rates of pneumonia complications rose following conver-sions from government or private nonprofit to for-profit ownership. Inanother recent study cited in this paper, mortality rates following dis-charge rose appreciably following conversion at hospitals that con-verted from government or private nonprofit to for-profit ownership,especially during the first two years after the conversion occurred. Onereason that there would be no change in in-hospital mortality, butan increase in mortality observed at thirty days, six months, andone year following admission to the hospital, may be that hospitals
164 Sloan
converting to for-profit ownership reduced lengths of hospital staydisproportionately.
The evidence pointing to some adverse effects of conversions is areason for concern and should affect public policy regarding hospitalownership conversions. There is a clear role for public oversight. Suc-cess is not guaranteedthe "Allegany system debacle," where effec-tive public oversight was lacking, is a case in point. Communities thatplace their hospital assets for sale would do well to exercise due dili-gence. This may take the form of oversight by the state attorney generalor the state certificate of need agency as well as local elected officialsPractices such as upcoding should be monitored by hospital compli-ance programs. Some public oversight of hospital staffing during theyears immediately following conversion seems warranted based on theempirical evidence.
Is there a chance that this study has failed to document importantchanges in quality of care and/or public goods provision associatedwith changes in hospital ownership status? Almost certainly yes. Theissues involved are complex, and there are so many kinds of outcomes.This consideration adds further force to the conclusion that scrutinyshould be case-specific. Rather than conclude that all conversions of agiven type are "bad" or "good," public policy should recognize the het-erogeneity and multiplicity of outcomes.
Notes
I acknowledge the capable research assistance of Jingshu Wang, Department of Econom-ics and Center for Health Policy, Law, and Management, Duke University, and the grant,Hospital Ozvnership Conversions, from the Robert Wood Johnson Foundation and adininis-tered by the Academy for Health Services Research and Health Policy
Of course, the two aspects are related, If there are many concessions on nonprice di-mensions of the transaction, the price should reflect this.
This result is consistent with the finding by Thorpe et al. (2000) that margins increasedafter conversion to for-profit status, mainly because of expense reductions.
In some unpublished work, colleagues and I followed the suggestion to includea bi-nary for trauma center in analysis of survival following inpatient stays. The variable typ-ically did not affect outcomes. One might argue that a trauma center is endogenous toownership.
Their technique includes the equivalent of hospital fixed effects.
A more complete discussion of the effects of hospital competition on quality ofcare isbeyond the scope of this paper. On this subject, see, for example, Kessler and McClellan(2000).
Hospital Ownership Conversions 165
This has been documented by others as well. See, for example, Eldenburg et al. (2001).
This result is consistent with a view advanced by Lakdawalla and Philipson (1999).They argued that nonprofit behavior should not affect for-profit behavior because the lat-ter, by virtue of the lack of tax advantages and charitable endowments, are the marginalfirms and hence influence market outcomes (price, quality, etc.).
Medicare compliance, coding, self-referral, and joint ventures have become major is-sues to hospitals. See, for example, a newsletter for hospital administrators explicitly de-voted to these issues: Eli Research, "Hospital Compliance Alert."
In table 5.4, all explanatory variables shown are binaries.
References
American Hospital Association (1976). Hospital Statistics. Chicago, IL: AHA, 1976.
American Hospital Association (1979). Hospital Statistics. Chicago, IL: AHA, 1979.
American Hospital Association (2001). Hospital Statistics. Chicago, IL: AHA, 2001.
Anderson, Ruth, Charlene Allerd, and Frank A. Sloan (2001). "Effects of Hospital Con-versions on Organizational Decision Making and Service Coordination," unpublished.
Blumenthal, David, and Joel S. Weissman (2000). "Selling Teaching Hospitals to Inves-tor-Owned Hospital Chains: Three Case Studies," Health Affairs 19(2):158-166.
Burns, Lawton R., John Cacciamani, James Clement, and Welman Aquino (2000). "TheFall of the House of AHERF: The Allegany Bankruptcy," Health Affairs 19(l):7-41.
Burns, Lawton R., Amit Shah, and Frank A. Sloan (2001). "The Impact of GovernanceChanges on Strategy: Results from a Study of Hospital Ownership Conversions," Uni-versity of Pennsylvania.
Conover, Christopher J., Michael E. Ezell, and Frank A. Sloan (2001). "UnderstandingHospital Conversions," unpublished. Duke University.
Conover, Christopher J., and Frank A. Sloan (2001). "Rates of Return from Hospital Con-versions," unpublished. Duke University.
Cutler, David M., and Jill R. Horwitz (2000). "Converting Hospitals from Not-for-Profitto For-Profit Status," in David M. Cutler, ed., The Changing Hospital Industry: ComparingNot-for-Profit and For-Profit Institutions. Chicago, IL: University of Chicago Press, 45-79.
Desai, Kamal R., Carol VanDeusen Lukas, and Gary J. Young (2000). "Public Hospitals:Privatization and Uncompensated Care," Health Affairs 19(2):167-172.
Duggan, Mark G. (2000a). "Hospital Market Structure and the Behavior of Not-for-ProfitHospitals: Evidence from Responses to California's Disproportionate Share Program,"NBER Working Paper 7966.
Duggan, Mark G. (2000b). "Hospital Ownership and Public Medical Spending," Quar-terly Journal of Economics, 115(4):1343-1373.
Eldenburg, Leslie, Benjamin E. Hermalin, Michael S. Weisbach, and Marta Wosinska(2001). "Hospital Governance, Performance Objectives, and Organizational Form,"NBER Working Paper 8201.
166 Sloan
Hart, Oliver, Andrei Shleifer, and Robert W. Vishny (1997). "The Proper Scope of Govern-ment: Theory and an Application to Prisons," Quarterly Journal of Economics112(4):1127-1162.
Kessler, Daniel P., and Mark B. McClellan (2000). "Is Competition Socially Wasteful?"Quarterly Journal of Economics, 115(2):577-615.
Lakdawalla, Darius, and Tomas Philipson (1999). "The Nonprofit Sector and IndustryPerformance," mimeo, University of Chicago.
Legnini, M. W., L. E. Rosenberg, M. J. Perry, and N.J. Robertson (2000). "Where Does Per-formance Measurement Go from Here?" Health Affairs 19(3):173-177.
McClellan, Mark, and Douglas Staiger (2000). "Comparing Hospital Quality at For-Profitand Not-for-Profit Hospitals," in David M. Cutler, ed., The Changing Hospital Industry:Comparing Not-for-Profit and For-Profit Institutions. Chicago, IL: University of ChicagoPress, 93-112.
Needleman, J., D. J. Chollet, and J. Lamphere (1997). "Hospital Conversion Trends,"Health Affairs 16(2):187-195.
Norton, Edward C., and Douglas 0. Staiger (1994). "How Hospital Ownership AffectsAccess to Care for the Uninsured," Rand Journal of Economics 25(1):171-185.
Picone, Gabriel, Shin-Yi Chou, and Frank Sloan (forthcoming). "Are For-Profit Conver-sions Harmful to Patients and to Medicare?" Rand Journal of Economics.
Robinson, James G. (2000). "Capital Finance and Ownership Conversions in HealthCare," Health Affairs (19)1:56-71.
Silverman, Elaine, and Jonathan Skinner (2001). "Are For-Profit Hospitals Really Differ-ent? Medicare Upcoding and Market Structure," NBER Working Paper 8133.
Sloan, Frank A. (2000). "Not-for-Profit Ownership and Hospital Behavior," in A. J.Culyer and J. P. Newhouse, eds., Handbook of Health Economics, vol. lB. Amsterdam:Elsevier, 1141-1174.
Sloan, Frank A., Donald H. Taylor, and Christopher J. Conover (2000). "Hospital Conver-sions: Is the Purchase Price Too Low?" in David M. Cutler, ed., The Changing Hospital In-dustry: Comparing Not-for-Profit and For-Profit Institutions. Chicago, IL: University of Chi-cago Press, 13-44.
Thorpe, Kenneth E., Curtis S. Florence, and Eric E. Seiber (2000). "Hospital Conversions,Margins, and the Provision of Uncompensated Care," Health Affairs 19(6):187-195.
Wolfram, Catherine (2000). "Comment on Chapters 3 and 4: McClellan and Staiger," inDavid M. Cutler, ed., The Changing Hospital Industry: Comparing Not-for-Profit andFor-Profit Institutions. Chicago, IL: University of Chicago Press, 155-158.