Competition and the impact of online hospital report cards.

Journal of Health Economics 34 (2014) 42–58

Contents lists available at ScienceDirect

Journal of Health Economics journal homepage: www.elsevier.com/locate/econbase

Competition and the impact of online hospital report cards Shin-Yi Chou a,∗ , Mary E. Deily b , Suhui Li c , Yi Lu d a

Department of Economics, Lehigh University and National Bureau of Economic Research, United States Department of Economics, Lehigh University, United States c Department of Health Policy, School of Public Health and Health Services, The George Washington University, United States d Health Services Administration, Graduate Program, College of Health Sciences, Barry University, United States b

a r t i c l e

i n f o

Article history: Received 4 December 2012 Received in revised form 4 December 2013 Accepted 4 December 2013 Available online 8 January 2014 JEL classification: I11 L1

a b s t r a c t Information on the quality of healthcare gives providers an incentive to improve care, and this incentive should be stronger in more competitive markets. We examine this hypothesis by studying Pennsylvanian hospitals during the years 1995–2004 to see whether those hospitals located in more competitive markets increased the quality of the care provided to Medicare patients after report cards rating the quality of their Coronary Artery Bypass Graft programs went online in 1998. We find that after the report cards went online, hospitals in more competitive markets used more resources per patient, and achieved lower mortality among more severely ill patients. © 2014 Elsevier B.V. All rights reserved.

Keywords: Quality information Hospital competition CABG report cards Health outcomes Predicted HHIs

“Public Information + Competition = High Quality, Cost Effective Health Care” – Pennsylvania Health Care Cost Containment Council, 1999

1. Introduction The internet has given healthcare consumers unprecedented access to information about the quality of health care providers. Over the past 15 years, the percentage of American adults with internet access has increased dramatically, rising from 10% in 1995, to 50% in 2000, to 75% in 2005; among those users, 61% have looked for health or medical information on the Internet (Fox and Jones, 2009). Given the willingness of internet users to search online for information about healthcare, the increased access that the net provides to credible ratings of the clinical quality of different healthcare providers may significantly affect the nature and degree

∗ Corresponding author at: Department of Economics, Lehigh University, 621 Taylor Street, Bethlehem, PA 18015, United States. Tel.: +1 610 758 3444; fax: +1 610 758 4677. E-mail addresses: [email protected] (S.-Y. Chou), [email protected] (M.E. Deily), [email protected] (S. Li), [email protected] (Y. Lu). 0167-6296/$ – see front matter © 2014 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jhealeco.2013.12.004

of competition among those providers, and thus the quality of the health care they deliver. Early in the debate on the effects of competition in health care markets commentators pointed out that competition would more effectively improve outcomes if information on the clinical quality of healthcare were available to consumers (Brook and Kosecoff, 1988; Ginsburg and Hammons, 1988). Information about clinical quality seems very likely to affect the quality provided by hospitals because “even a small amount of information imperfection” can lead to market failure (Stiglitz, 2000), and asymmetric information in healthcare markets is profound (Arrow, 1963). Publically available data on the quality of care should therefore give competitors an incentive to improve care, and this incentive should be stronger in more competitive markets, because consumers have more choices. Thus, we expect higher quality in those markets where firms face more competition and where consumers are well-informed about quality. As discussed further below, economists have now done a number of studies examining whether quality reports rating hospitals affect patients’ choices or health outcomes, as well as on whether competition improves the quality of healthcare in hospital markets. However, none have examined whether providing quality information to consumers makes competition more effective in improving the clinical quality of hospital services. In this paper, we study

S.-Y. Chou et al. / Journal of Health Economics 34 (2014) 42–58

the relationship between online performance grades, competition, and the quality of health services by examining how the online publication of hospital report cards affected health outcomes for Medicare patients living in Pennsylvania hospital markets with different degrees of competition. Pennsylvania, through the Pennsylvania Health Care Cost Containment Council (PHC4), has been a pioneer in developing performance grades on Coronary Artery Bypass Craft Surgery for health care providers and posting them online (called CABG report cards hereafter). Of course, Pennsylvanians have information about the likely quality of different hospitals other than the CABG report cards. For example, they may form an opinion based on a hospital’s size, reputation, or its teaching status, from their individual experience with it, or from the experiences of others. However, when the PHC4 began publishing CABG report cards for hospitals online in May 1998, it made easily available much more precise data on the aspect of quality most difficult for individual patients to identify: risk-adjusted health outcomes data for a specific serious procedure. We use these report cards to examine whether those hospitals located in more competitive markets increased the quality of their care after report cards rating their Coronary Artery Bypass Graft (CABG) surgery programs went online in 1998. The PHC4 has been publishing Pennsylvania’s Guide to Coronary Artery Bypass Graft Surgery since the early 1990s. The initial two CABG reports (the first was published in 1994, and the second in the fourth quarter of 1995) were printed documents; while they were distributed to hospitals, surgeons, public libraries, business groups, legislature, the media, and any individual who requested them (Schneider and Epstein, 1998), they were nevertheless relatively difficult to access. With the publication of its May 1998 CABG report card on the agency’s website (www.phc4.org), the PHC4 has made the information easily accessible to patients, physicians, hospitals, and health insurance companies. The year 1998 was a landmark year for the PHC4 for several reasons. In addition to posting the CABG report online, the agency upgraded its computer system, moving from a mainframe to a client–server network system, which improved the timeliness of the data used in its reports. In 2000 and 2001 the agency introduced interactive reports on a redesigned website, making the report card information easier to find and use. The result of these changes has been to make the data more easily available and more relevant; the number of hits on the PHC4 website grew rapidly from an average of 1800 per month in 1998 to an average of over 30,000 per month in 2001 (PHC4, Annual Report, various issues). While it is impossible to determine who is accessing the site, the PHC4 states that “Many of PHC4’s Web inquiries are from consumers who have an immediate need for the reports. The Web site presents the public with a quick, simple means of obtaining a copy of our public reports – information that can be downloaded with the click of a mouse” (PHC4, Annual Report for 1999). In 2002, 20,000 copies of the CABG reports were downloaded (PHC4, Annual Report for 2002). We study the impact of newly available quality information on the outcomes for CABG patients covered by Medicare, because reimbursement for these patients is fixed. Economic theory predicts that the relationship between quality and competition is ambiguous when firms may alter both price and quality, but that firms competing in markets where prices are fixed, as is the case for Medicare patients, will use quality in place of price to attract customers.1 We therefore expect higher quality in more

1 The analysis of quality competition among price-regulated firms was developed by economists studying competition between firms such as airlines: a series of papers established that firms unable to compete on price would instead compete on the basis of quality, and that the incentive to compete would be greater as the

43

competitive markets,2 but investigate whether that relationship is magnified by improved availability of credible information about quality in markets with higher levels of competition. Concentration in hospital markets has increased substantially since the mid 1990s (Gaynor, 2006), making it increasingly important to understand the potential impact of competition on health care quality, and the role that quality information plays in that relationship. However, in studies of the effects of market concentration on health outcomes, potential endogeneity poses a major challenge because unobserved heterogeneity may determine both health outcomes and the extent of market competition. We address this problem by estimating conditional logits of hospital choice to generate predicted market shares, and using predicted rather than actual market shares to measure market competitiveness (Kessler and McClellan, 2000). We find a very robust shift in CABG outcomes at the time of online publication in more competitive markets, suggesting that improved quality information caused hospitals in more competitive markets to use more resources to provide better health outcomes for Medicare patients. Specifically, in more competitive hospital markets the online publication of CABG report cards resulted in a roughly 5–10% reduction in mortality at an additional cost of approximately 2000 dollars per case. Our results suggest that better outcomes may be achieved in competitive hospital markets where patients have easy access to quality information, and thus that publically provided quality information made available on the web may play an important role in improving health care quality in these markets. To the best of our knowledge, this is the first study of the role of quality information in determining the impact of market competition on quality in hospital markets. In the next section we review theoretical and empirical literature on relationships between quality information, quality, and market concentration. We describe the Pennsylvania CABG report card program in Section 3, and our data and sample in Section 4. In Section 5 we describe our basic specifications, the variables, and the calculation of our measure of hospital market concentration. We present our results in Section 6, examine the possibility of creamskimming in Section 7, and end with a discussion of our results in Section 8. 2. Literature review Analyses of the relationship between competition and quality assume that consumers are aware of the quality of firms’ goods or services, so that competing in terms of quality makes sense for firms that cannot compete in price.3 But if consumers have difficulty determining the quality of the good or service, then improving that information may give firms a greater incentive to increase quality.4

number of firms increased. For examples, see White (1972) and Douglas and Miller (1974). 2 A number of empirical studies of hospital markets do find that quality is higher in more competitive markets (Kessler and McClellan, 2000; Kessler and Geppert, 2005; Shen, 2003; Tay, 2003; Cooper et al., 2011; Gaynor et al., 2013), although some find mixed relationships, a negative relationship, or no relationship at all (Mutter et al., 2008; Gowrisankaran and Town, 2003; Mukamel et al., 2002; Shortell and Hughes, 1988). See Gaynor (2006) for a critical review of this literature. 3 Cooper et al. (2011) and Gaynor et al. (2013), both of which examine the effect of increased hospital competition in the U.K. on hospital quality, do describe steps the British government has taken to disseminate information about the quality of hospitals’ health outcomes. However, these studies focus on the impact on health outcomes of the change in competitive conditions rather than the change in availability of information. 4 Mutter et al. (2008) hypothesize that their mixed results on the relationship between competition and hospital quality may be explained in part by differences in patients’ ability to assess the quality of hospital services.

44


Dranove and Satterthwaite (1992) model imperfectly competitive firms that choose both price and quality and find that, holding price information constant, better quality information raises the equilibrium level of quality. Gravelle and Sivey (2010) study the effect of better information on quality with a model of duopoly hospitals that are reimbursed at fixed rates. They find that if the costs of producing quality are not too dissimilar across hospitals, then better-informed consumers cause each hospital’s equilibrium quality to increase. Some empirical work indicates that firms do increase quality in response to changes in the availability of information about it: for example, Chassin (2002) and Ettinger et al. (2008) describe steps taken by hospitals receiving poor grades to improve services, and Hannan et al. (2003) show CABG mortality rates are lower in areas with report cards. But other authors have pointed out the possibility that those being graded may try to “game” the system. Lu (2012), for example, shows that report cards cause nursing homes to reallocate effort toward improving those dimensions of quality that are graded, although Mukamel et al. (2002) suggest that this type of reallocation may improve outcomes for hospital patients if it causes hospitals to reallocate effort from improving “hotel amenities,” e.g., room size or food quality, which are easier for patients to assess, toward improved health outcomes, once credible evidence regarding those outcomes becomes available. However, an influential study of hospitals in New York and Pennsylvania in the early 1990s found evidence of “creamskimming,” that is, evidence that hospitals responded to report cards by eschewing the more severely ill and doing more CABG surgery on healthier patients so as to improve their grades (Dranove et al., 2003). More recent work suggests that report cards may give providers an incentive to select patients, but that the incentive depends on the distribution of patient types facing the hospital, on a hospital’s quality, and on the information revealed by report cards, and finds no evidence of selection in Pennsylvania during these early years of the report card program (Chen, 2011; Chen and Meinecke, 2012). This latter work presumes that just as hospitals may respond to report cards by selecting patients, patients may respond to report cards by selecting hospitals. Researchers have found that hospital quality does matter to consumers (see, for example, Jung et al., 2011; Tay, 2003), and a large literature tries to measure consumer response to quality information about healthcare providers.5 Dranove and Jin (2010) summarize findings for a variety of different markets, including work on hospital markets, as showing that consumers are more likely to respond to information that is understandable, easy to access, and “new,” in the sense of contradicting existing beliefs. In sum, these various literatures suggest that giving patients better information about quality gives hospitals an incentive to increase quality, but that under some circumstances such information may also give hospitals an incentive to select healthier patients. However, these authors do not investigate whether the impact of improved information about quality differs across markets with different degrees of competition.

5 See Fung et al. (2008) for a meta-study of research through 2006 and Romano et al. (2011) for a recent study. Most report card studies focus on hospitals and surgeons providing CABG. However, Bundorf et al. (2009) studied fertility clinics and found that report cards affected consumers’ choices, and Howard and Kaplan (2006) studied kidney transplant providers and found that report cards affected the choices particularly of younger and more educated consumers. There is also a growing literature examining the impact of publicly provided information on the quality of care provided by nursing homes. For three recent examples, see Clement et al. (2012), Park et al. (2011), and Werner et al. (2011).

One theoretical study, an extension of the Gravelle and Sivey (2010) model to include horizontal (i.e., geographic) as well as vertical differentiation, suggests that in monopolized markets firms will not change their quality at all in response to consumers having better information (Gravelle and Sivey, 2009).6 In an empirical study of nursing homes, Grabowski and Town (2011) examine whether the impact of online, publicly provided data on quality has a greater impact on homes located in more competitive areas. Using difference in difference methods, they find evidence that the quality of services provided by nursing homes in more competitive markets increased relative to those located in more concentrated markets after quality data became available. However, their quality measures are at the facility level rather than for individual patients, and they do not correct for endogeneity in creating their market share variables. In this paper we investigate whether the effect of better quality information differs across markets with different degrees of competition: our hypothesis is that CABG hospitals in more competitive areas were more likely to provide higher quality to Medicare patients after the release of online quality data than were hospitals in less competitive areas. We test this hypothesis using outcomes for individual patients to study whether the online release of report cards caused those outcomes to improve more in more competitive hospital markets. In doing so we must control for endogeneity in our measure of hospital market concentration, and check for evidence of hospital selection in response to report cards, since creamskimming would result in improvements in observed health outcomes. 3. Quality information and CABG hospitals in Pennsylvania CABG is sophisticated surgery and not all hospitals offer it: in Pennsylvania those that do are mostly large, private not-for-profit, teaching hospitals. In 1995 there were 42 CABG providers, which number grew to 60 in the years following the termination of the Certificate of Need (CON) regulations that had, until then, restricted entry.7 Subsequent entry rates around the state varied, but many of the new programs were started by hospitals located in the more densely populated areas of southeastern Pennsylvania (see Fig. 1). These hospitals collectively performed an average of 18,750 CABG surgeries per year in Pennsylvania during our sample period. As in other states, the number of CABG procedures fell during the period, in Pennsylvania by about 30%, because improved access to preventive care (such as beta-blockers and cholesterol tests) has reduced the incidence of coronary artery disease (Arciero et al., 2004) and because of the development of percutaneous coronary intervention (PCI) and stenting as alternative procedures for those patients who do develop coronary artery disease (Ulrich et al., 2003). Consequently, during the years 1995–2004 new CABG programs entered the market while at the same time demand for CABG procedures began to stagnate, increasing spatial competition both across markets and within markets over time, particularly in nonrural areas.

6 The latter analysis is done in a Hotelling framework. If hospitals are located far from each other, and travel is costly, then hospitals may not respond to better information about their quality because nearby patients will still choose to use the hospital, given the costs of traveling to an alternative. 7 Hospitals have an incentive to enter the CABG market because the procedure is highly profitable under the Medicare reimbursement schedule, and because offering CABG improves a hospital’s technology profile and thus attracts more patients who might come to it for other services in the future. The rapid growth of HMO market share during the 1990s may have reinforced this latter incentive because selectivecontracting managed care plans seem to value the presence of technology in making decisions about the network providers (Baker and Phipps, 2002).


45

Fig. 1. CABG hospital locations in Pennsylvania, 1995 (circles) and 2004 (squares).

The PHC4 report cards rate the performance of individual hospitals with respect to several health outcomes achieved by their CABG patients. A hospital receives a grade of “Same as expected” if its patients’ health outcomes fall within a 95% confidence interval for the state. Hospitals with adverse outcome rates outside the 95% confidence interval receive either a “Lower than expected” grade, if their probability of having an adverse outcome falls below the 95% confidence interval (a grade referred to in this paper as superior), or a “Higher than expected” grade, if mortality or readmission rates are above (a grade referred to here as poor).8 Most hospitals achieve grades of “Same as expected,” but a significant percentage, as many as 17% in one report card, receive a poor grade; superior grades are rarer, ranging from 2 to 10% in any report. Some hospitals performing CABG are not graded, either because they performed too few surgeries,9 or because their CABG programs were new: the first online report card used data from 1994 and 1995, so facilities entering the market after the CON program ended in December 1996 do not appear in the May 1998 report card. In such cases, we identify the hospital as having “no grade reported.” With the publication of the fourth report card in the second quarter of 2002 (based on inpatient data from 2000), the number of ungraded CABG programs fell from 19 to 8; by 2005 there was only one. In sum, since May 1998 any potential consumer of CABG surgery has been able to look online to see if a hospital was graded, and if so what that grade was. Since we expect this type of information about health outcomes to be important to many patients, we expect the improvement in the availability of quality data in May 1998 to have increased the incentives of hospitals in competitive markets to invest in improving their CABG outcomes.

8 Outcomes are risk-adjusted to correct for the possibility that a hospital may on average treat sicker patients. The methods currently used by the PHC4 reflect both administrative data on the patients (e.g., age, sex, gender, co-morbidities) and some clinical data, and are considered to be among the best (Snowbeck, 2000). Methods have evolved over time: details are contained in the Technical Notes to each report card report and are posted on the PHC4 website. 9 A hospital must perform at least 30 isolated CABG procedures on adults in a year to be graded. CABG patients undergoing additional procedures such as mitral valve repair or atrial septal defect repair are excluded from the report card assessment (Shahian et al., 2001).

4. Data and sample Data on individual CABG patients are from the PHC4’s inpatient database, which consists of records for all admissions by Pennsylvania hospitals. These records contain information that includes a patient’s age, gender, ethnicity, admission source/type, diagnostic codes, procedure codes, zip code of residence, and insurance type, the total charge listed by the hospital, and whether the patient died while in the hospital. The records also include individual patient identifiers that allow us to link records for patients over time, and hospital identifiers that allow us to identify the hospital where the procedure took place. We use the procedure codes to identify patients undergoing CABG surgery during the years 1994 through 2005. Because our risk adjustments require information for the 12 months preceding a patient’s admission, and one of our outcome variables requires information from the 12 months following a patient’s CABG surgery, the sample we analyze is the surgeries that occurred during the years 1995 through 2004. Of those surgeries, we excluded the records of patients with invalid patient IDs, patients who underwent CABG in the prior year, patients admitted to hospitals with fewer than five claims for CABG within a year, and patients who were residents of other states. We used the hospital identifiers in the PHC4 data to link information about the admitting hospital from the American Hospital Association’s (AHA) Annual Survey of Hospitals to each patient’s record, and excluded patients with missing AHA data, or whose zip code of residence was missing, because this information is necessary for construction of our measure of market competitiveness (described in Section 5). We next excluded observations with total charges greater than $1,000,000 or with length of stay longer than 30 days (values greater than or equal to three standard deviations from the mean, respectively). We attached to each patient record the Medicare cost-to-charge ratio for the admitting hospital,10 and dropped those with missing ratios since these are necessary to calculate our

10 The Medicare cost-to-charge ratio is computed by dividing a hospital’s total costs on Medicare patients by its total Medicare charges. Cost and charge information are from the Centers for Medicare & Medicaid Services (CMS) cost reports, available at http://www.cms.gov/Research-Statistics-Data-and-Systems/ Files-for-Order/CostReports/index.html. The data are merged with our study sample using the hospital Medicare Provider identification number, which is available in the AHA data.

46


Table 1 Development of samples for main analysis, for conditional logit estimations, and for construction of predicted HHI.

Table 2 Sample statistics for patient outcomes.a Whole sample

Sample size Panel A Patients who underwent an isolated CABG procedure between 1995 and 2004 Exclude patients with an invalid patient ID and patients who underwent CABG in the prior year Exclude patients admitted to hospitals with less than 5 CABG cases Exclude patients who are not Pennsylvania residents Patients with valid individual information, AHA data and distance measures Exclude patients with invalid inpatient charge (>$1,000,000) or length of stay (>30 days) Exclude patients with missing cost-to-charge ratio or invalid adjusted total cost (>$500,000) Exclude patients from zip codes with no HHI for 1997 Exclude all but Medicare patients Final sample used for main analysis Panel B Sample used to predict HHI Patients with valid individual information, AHA data, and distance measures Sample used to fit conditional logit model Exclude all patients except Medicare FFS patients Medicare FFS patients traveling within 50 miles Sample used to fit conditional logit model in robustness check Medicare FFS patients traveling within 75 miles

Panel A: patient outcomes Total costs

187,504 186,628

In-hospital mortality Readmission Nb

185,328 166,657 153,340

$24,149 (14,985) 0.030 0.189 76,862

Online publication of report card Before

After

Change

$23,997 (14,026) 0.034 0.206 28,722

$24,240 (15,529) 0.027 0.180 48,140

$243

$20,947 (11,527) 0.030 0.257 9560

−$2476

$22,541 (12,343) 0.023 0.154 17,732

$1520

$27,179 (18,701) 0.030 0.167 20,902

$728

Panel B: patient outcomes by predicted HHIc Least Competitive [HHI mean = 0.748, SD = 0.099] Total costs $ 21,866 $23,423 (11,748) (11,953) In-hospital mortality 0.031 0.032 Readmission 0.245 0.225 15,121 5615 Nb Competitive [HHI mean = 0.472, SD = 0.071] $21,998 $21,021 Total costs (11,550) (9890) 0.027 0.034 In-hospital mortality Readmission 0.165 0.186 b 27,581 9849 N Most Competitive [HHI mean = 0.177, SD = 0.033] $26,896 $26,450 Total costs (16,743) (17,969) In-hospital mortality 0.032 0.036 0.185 0.212 Readmission b 34,160 13,258 N

151,028 140,421 137,239 76,862 76,862

153,340

69,754 65,953 68,668

−0.007 −0.026

−0.001 0.031

−0.011 −0.032

−0.006 −0.045

a

Standard deviations are in parentheses. Sample sizes for readmissions are slightly smaller because they exclude patients that died during their original admission to the hospital for CABG surgery. c Most Competitive: HHI in 4th quartile of the HHI distribution; competitive: HHI in 2nd and 3rd quartile of the HHI distribution; Least Competitive: HHI in 1st quartile of the HHI distribution. b

total cost variable. Finally, we excluded patients whose calculated total cost exceeded $500,000 (a value greater than three standard deviations from the mean).11 As will be clarified below, we use the predicted HHI for each zip code for the year 1997 in the analysis, so we excluded patients from those zip codes with no CABG patients in 1997, since calculation of an HHI for 1997 was not possible. Finally, we excluded all but Medicare patients. Panel A of Table 1 outlines our sample selection criteria and the resulting sample size after each exclusion. The final sample size for the main analysis is 76,862. Information about hospital characteristics and location is from the American Hospital Association’s Annual Survey. We obtain hospital grades from issues of Pennsylvania’s Guide to Coronary Artery Bypass Graft, published by PHC4. 5. Specification Our basic specification is: Outcomeikht = ˛ + (HHI97k · Postt ) + ˇ1 Mkt + ˇ2 Pikht + h + k + t + h · Yt + εikht ,

(1)

where Outcomeikht is the outcome realized by patient i from 5-digit zip code area k, admitted to hospital h in year t, HHI97k is a measure of market concentration in 1997, and HHI97k × Postt , our key variable, captures the effect of the publication of online quality information in hospital markets with different levels of concentration (where t for the variable Post refers to quarter). Control variables include Mkt , a vector of other market characteristics, Pikht , a set of patient characteristics, fixed effects h , k , and t , for the patient’s admitting hospital, the patient’s zip code, and the year of the patient’s surgery, and h Yt , a hospital-specific linear time

11

Imposing these various exclusions does not affect our results, but we nevertheless omit the observations because these extreme values may be the result of measurement errors.

trend. Finally, εikht is a mean-zero independently distributed error term so that E(εikht |. . .) = 0. 5.1. Outcome variables We focus on three separate outcomes to capture different aspects of the quality of the services a patient receives. The first is the log of the total cost for each patient, which we use as a measure of the amount of resources used on the patient. The “total charge” reported in the PHC4 data is the sum of the list prices of all goods and services the hospital provided to the patient during their stay, not including physicians’ fees. However, the actual cost of a patient’s care differs from the total charge and is unknown. As a more accurate measure of resource use, we calculate the “total cost” for each patient’s treatment by multiplying the patient’s total charge by the Medicare cost-to-charge ratio for the admitting hospital (Picone et al., 2003; Dafny, 2005). Total costs are inflationadjusted to 1995 dollars using the urban consumer price index (CPI). The second outcome variable is in-hospital mortality, a binary variable equal to one if the patient died in the hospital after being admitted as a CABG patient, and zero otherwise. Mortality is a commonly used measure of the overall quality of a patient’s experience when death in not so rare an occurrence as to make it unusable, as is the case here. However, this measure is limited to capturing only the most extreme outcome, and only during the length of a patient’s stay.12 Therefore, for our third outcome variable we use

12 We know whether a patient died in the hospital from their discharge status as listed in the PHC4 inpatient data record. However, information in the record ends when the patient was discharged, so we are unable to determine if they died after leaving the hospital.


readmission rates after a patient has had CABG surgery for any of a set of conditions that are frequently associated with this procedure.13 That is, the readmission variable equals one if a CABG patient was readmitted (to any hospital) for problems related to ischemic heart diseases within 12 months of their CABG, and zero if they were not. Summary statistics for the patient outcomes variables for the whole sample, as well as for surgeries occurring before and after online publication of the report card and the change in the outcomes, are shown in Panel A of Table 2. (Sample sizes for readmissions are slightly smaller than those given on the table because they exclude patients that died during their original admission to the hospital for CABG surgery.) Overall, total costs increased and mortality and readmission rates decreased after the online publication of the report card. 5.2. Hospital market concentration We do not use HHIs generated by the conventional variableradius method, whereby a hospital’s geographic market is determined by the area from which the hospital draws the majority (e.g., 75%) of its patients, because endogenous patient flows may bias our results. Patient flows may be endogenously correlated with unobservable hospital quality because hospitals attracting patients from further away have larger market sizes: if patients travel further to reach high-quality hospitals, then these hospitals markets will appear more competitive, creating a reverse causality between market competitiveness and quality. In addition, measuring competitiveness based on a patient’s actual admitting hospital can introduce another endogeneity problem: sicker patients are more likely to seek care from high-quality hospitals, which are likely to be located in more densely populated areas where markets tend to be more competitive. Thus, the estimated coefficient of competition measured using patient flows to the admitting hospital could suffer from significant bias due to unobservable determinants of patient’s hospital choice that affect both market competitiveness and health outcomes. We therefore calculate HHIs using predicted market shares for residential zip codes rather than the actual market shares of admitting hospitals (Kessler and McClellan, 2000). Predicted HHIs function like traditional HHIs, in that they are bound below by zero and above by one, and increase in concentration, but are based on exogenous factors of patient demand (the hospital and patient characteristics listed in Table A1 and various distance measures) rather than potentially endogenous actual patient flows. We first estimate the probability that a patient will choose each of the different CABG hospitals located within a 50-mile radius of their residential zip code each year.14 We use only fee-for-service (FFS) Medicare patients for this estimation because these patients are not restricted in their choice of hospital and can choose on the basis of their own tastes for hospital quality and convenience.15 (The FFS-Medicare patients used to estimate the probabilities are a

13 The conditions are those covered by ICD-9 codes 410.XX through 414.XX and consist of acute myocardial infarction, other acute and sub-acute forms of ischemic heart disease, old myocardial infarction, angina pectoris, and other forms of chronic ischemic heart disease. 14 The probabilities are generated from patient-level conditional logit models where hospital choice is a function of hospital characteristics, of travel distances to a hospital compared to distances to hospitals closest to the patient’s residential zip code, and of the characteristics of the individual patient making the choice (where the patient’s characteristics appear interacted with hospital variables). See the Appendix for a more detailed description of these and the following calculations. 15 Ideally we would estimate the conditional logits using all Medicare patients. However, patients with HMO insurance used different HMOs, each of which has its own network of providers, so that we cannot assume an individual with HMO

47

subset of our initial sample before we eliminate observations that are missing information necessary for estimating our main specifications – we call this larger sample the Predicted HHI sample. See Panel B of Table 1 for the relationship of the FFS-Medicare subsample to the Predicted HHI sample.) We then use the parameter estimates from the conditional logits to calculate the probability of each patient in the Predicted HHI sample, both Medicare and non-Medicare, going to each CABG hospital in Pennsylvania.16 (We use the larger Predicted HHI sample to calculate the HHIs so as to get a more complete idea of the degree of competition in different market s, but we test the sensitivity of our results to this decision in the robustness checks.) Descriptive statistics for the patient and hospital characteristics of the Medicare-FFS sample and for the predicted HHI sample for the entire study period are reported in columns (1) and (2) of Appendix Table A1. Once these probabilities are found for each CABG patient in the predicted HHI sample each year, the predicted HHIs are calculated using a series of aggregations, first to the zip code level, and then to the hospital level, where information about the share of patients in each zip code going to each hospital is incorporated. The final step is to translate the hospital-based HHIs back to the zip code level, so that the competitiveness of the hospital market for a patient is not measured by the HHI of the patient’s admitting hospital. We use the predicted HHIs for just one year, 1997, so that the variable HHI97k captures the variation in the level of market competition for zip code area k before the release of online report cards. Fixing the competitiveness of markets at the year before the report cards went online allows us to avoid the confounding effects of changes in quality due to changes over time in market competition. (Note that use of zip code fixed effects means that the variable HHI97k does not itself appear in the specification.) Finally, our initial estimations suggested that the effect of competition was nonlinear. To get a clearer idea of the marginal impact of competition, we follow Kessler and McClellan (2000) and Kessler and Geppert (2005) by grouping zip codes into different categories of market competitiveness, with the categories based on the distribution of zip-code level HHIs over the entire study period. We then replace the continuous variable HHI97k with two dummy variables: the variable Most Competitive equals one if the patient lives in a zip code that was located in one of the most competitive hospital markets in 1997, where most competitive means in the lowest quartile of the predicted HHI measure, and the variable Competitive equals one if the HHI in 1997 for the patient’s zip code lies in the second or third quartile of the HHI distribution. The reference group is the most concentrated quartile of predicted HHIs in 1997. (Appendix Fig. A1 is a histogram of HHI97 and includes the numerical values of the cutoff points identifying the different market classifications.) Consequently, the key variable, HHI97k ·Postt , becomes two variables: Most Competitive·Postt and Competitive·Postt , where the indicator variable Postt equals one for all quarters after quality data went online, that is, after the second quarter of 1998. The coefficients of these variables thus measure the average change in the outcome variable, after grades went online, in most competitive

insurance could freely choose among all hospitals within 50 miles – the resulting estimation would be biased. 16 We assume that FFS Medicare patients’ preferences over hospitals are similar to the preferences of other Medicare and non-Medicare patients. Town and Vistnes (2001) present evidence suggesting that conditional logits estimated using Medicare patients work well to predict choices made by other groups of patients. This same assumption is imposed in Capps et al. (2003) and Ho (2006).

48


Fig. 2. Map of Pennsylvania zip codes by competitiveness of hospital market in 1997.

and competitive markets, relative to the change in the outcome variable in the least competitive markets, ceteris paribus. Fig. 2 shows the variation in market competition for different competition groups and zip codes in Pennsylvania in 1997. The map reflects the influence of the locations of CABG hospitals shown in Fig. 1: areas with more CABG hospitals, those in the more populated Philadelphia and Pittsburgh areas, have more zip codes in the Most Competitive category, while other areas of the state, with fewer CABG hospitals, have more zip codes in the Least Competitive category. Blank areas on the map represent zip codes for which we could not predict an HHI in 1997 because there were no CABG patients from those zip codes in that year. Panel B of Table 2 shows the patient outcomes by category of market concentration, for the whole sample and for the sample split into surgeries that occurred before and after the online publication of hospital report cards, as well as the change in the outcomes by market concentration and the average predicted HHI for each of the three concentration categories. 5.3. Control variables The vector Mkt represents variables that are included to control for variation across markets in the type of hospital services available (Kessler and McClellan, 2000). These variables are the predicted number of hospital beds and number of teaching hospitals for each zip code, calculated using the same method as was used to create the predicted HHI (see Appendix). Thus, the variables represent the predicted hospital size and teaching status of hospitals used by patients from each zip code, as determined by the patients’ choices.17 The vector Pikht includes the individual patient characteristics age (measured between 45 and 84 in 5-year categories, or as above 84; the base group is below 45), gender (with female as the base group), and ethnicity or race (Hispanic or African-American, with White or other race as the base group). We also include whether the admission was an emergency (as opposed to scheduled), and

17 We use predicted hospital market characteristics to reduce endogeneity, but our results do not change if we instead use the actual teaching status and size of a patient’s admitting hospital.

whether the patient was a Medicare HMO enrollee (as opposed to Medicare FFS), as well as patient admission source (physician referral, transfer from a hospital, or transfer from a skilled nursing facility; the base group is transfer from some other source such as a clinic or an ambulatory surgical center), and the distance to the closest CABG hospital for patients living in zip code k (measured between the centroids of the zip code of the patient and the zip code of the hospital). We use variables based on the Charlson index to measure the severity of a patient’s illness (Charlson et al., 1987). The Charlson index contains 19 categories of comorbidity, each associated with a weighting score (ranging from 1 to 6) that reflects the adjusted risk of one-year mortality associated with the comorbidity.18 We calculate a weighted Charlson index for each patient using their diagnosis codes at the time of admission for CABG as well as their diagnosis codes from any hospital admission during the 12 months preceding their surgery: for our sample, values for the weighted Charlson index range from zero (no comorbidity) to 15 (very severe burden of comorbidity). We created six dummy variables, one for each value of the index from one to five, and a sixth that indicates if the patient’s index is greater than or equal to six, with zero as the reference group. We include fixed effects for the patient’s zip code, for the admitting hospital, and for the year the patient had their surgery. The use of zip code fixed effects captures unobserved, time-invariant geographic variance in consumer tastes, health status, population density, and access to care that may affect the consumption of hospital services and therefore outcomes. Hospital fixed effects absorb all time-invariant differences among admitting hospitals that may affect the patient’s outcome (while avoiding the introduction of endogeneity that including at least the observable characteristics directly in the specification might involve). Year fixed effects

18 Diagnoses associated with a score of 1: Myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, connective tissue disease, ulcer disease, mild liver disease, and diabetes. Diagnoses associated with a score of 2: Hemiplegia, moderate to severe renal disease, diabetes with end organ damage, any tumor, leukemia, and lymphoma. Diagnoses associated with a score of 3: moderate to severe liver disease. Diagnoses associated with a score of 6: metastatic solid tumor and AIDS.


control changes over time that may be affecting outcomes for all patients in Pennsylvania. Finally, a possible confounding effect is changes over time in hospital characteristics that influence patients’ outcomes, such as changes in technology or nurse staffing. We therefore include a hospital-specific linear time trend by interacting each hospital dummy h with a linear time trend variable Yt . Descriptive statistics for the variables used in the analysis are reported in the columns titled “CABG Patients” in Table A2. 5.4. Quality and patient severity Even if releasing hospital quality information makes competition more effective in improving the quality of care on average, it is not clear that all patients would benefit equally. For example, Kessler and Geppert (2005) found that hospital competition affected the resources used to treat different types of patients. We therefore incorporate the illness severity of the patient into our key variables, so that we can examine whether the extent to which quality information enhances the effects of market competition varies for patients with different degrees of illness. To do so, we add to Eq. (1) interactions between Most Competitivek ·Postt and Competitivek ·Postt and a new dummy variable, Severeikht , where Severeikht equals1 if patient i’s weighted Charlson index is 2 or more, and zero otherwise (see Appendix Fig. A2 for a histogram of the weighted Charlson index for our sample). The main specification becomes:

49

months following their CABG surgery for a related problem. Robust standard errors clustered by hospital are reported in brackets. The estimated coefficients for the variables Most Competitivek ·Postt and Competitivek ·Postt , reported on the first two lines of columns (1), (4), and (7), indicate that on average, after report cards went online, patients in the most competitive markets had significantly more resources expended on their cases compared to patients in the least competitive markets; the coefficient for patients in competitive markets is positive but smaller and not significant. There is no evidence of a difference in mortality or in readmission rates in more competitive or in competitive markets compared to the least competitive markets. However, the estimated coefficients on the triple interaction terms Most Competitivek ·Postt ·Severityikht and Competitivek ·Postt ·Severityikht , reported in columns (2), (5), and (8), suggest that mortality rates fell in competitive and most competitive markets among the more severely ill patients, despite the fact that hospitals in these areas used more resources on all patients. There is also some weak evidence that total cost increased more for severely ill patients in the most competitive markets. Estimated coefficients for readmission, although negative, are insignificant. The estimated coefficients of the pre-trend control variables, Most Competitivek ·Pret and Competitivek ·Pret , reported in columns (3), (6) and (9), are insignificant for costs and mortality rates, suggesting that pre-trends in these variables are not driving our results. As for readmission, a positive pre-trend coefficient as opposed to

Outcomeikht = ˛ + 1 (Most Competitivek · Postt ) + 2 (Competitivek · Postt ) + 1 (Most Competitivek · Postt · Severeikht ) + 2 (Competitivek · Postt · Severeikht ) + ˇ1 Mkt + ˇ2 Pikht + h + k + t + h · Yt + εikht , A non-zero estimate of 1 or 2 implies that the increased availability of quality information caused hospitals facing different amounts of competition to treat patients differently if they were more severely ill. 5.5. Pre-existing trends We are particularly concerned that the estimated ’s might reflect pre-existing trends instead of the actual effects of online report cards. We plotted the three outcome variables by category of market competition and found no obvious signs of pre-trends (see Appendix Figs. A3–A5). Nevertheless, we added variables to Eq. (2) to test for pre-trends in markets with different levels of market competition:

(2)

an insignificant post coefficient for Most Competitivek markets provides evidence that these markets experienced a “reverse trend” of readmission rates in the post period.19 We reran specifications (1)–(3) with two alternative measures of severity in case our result for mortality in particular was caused by the severity measure we used. In one case, we replaced the Severe dummy with a variable equal to the weighted Charlson index, which ranged from 0 (no comorbidity) to 15 (very severe burden of comorbidity), and in the other case we replaced Severe with a truncated version of the Charlson index, with values ranging from 0 to 7 or more. We continued to find strong evidence that hospitals in the most competitive regions had higher total costs for all patients and that hospitals in competitive and most competitive regions had lower mortality for the more severely ill, and no

Outcomeikht = ˛ + 1 (Most Competitivek · Postt ) + 2 (Competitivek · Postt ) +1 (Most Competitivek · Postt · Severeikht ) + 2 (Competitivek · Postt · Severeikht )

(3)

+1 (Most Competitivek · Pret ) + 2 (Competitivek · Pret ) + ˇ1 Mkt + ˇ2 Pit + h + k + t + h · Yt + εikht Pret is a dummy variable that equals one for the four quarters preceding the release of online report card, namely the third quarter of 1997 through the second quarter of 1998 (where t for the variable Pre refers to quarter). The interaction of our market concentration measures with Pret separates the effect of pre-existing trends in the outcome variables from the response to the introduction of online information. 6. Results Table 3 shows the results of estimating specifications (1)–(3) for each of the three different dependent variables: total cost of a patient’s CABG surgery, whether the patient died in the hospital, and whether the patient was readmitted to any hospital in the 12

evidence of any change in readmission rates (results are available on request).20

19 We ran two additional checks for pre-trends. First we reran equation (3) on a sample of observations from the “pre” period (1995-second quarter of 1998) only – essentially using the Pre dummy as a “placebo” version of our Post variable. Second, we changed the “placebo” variable so that it equaled one for eight quarters (third quarter of 1996-second quarter of 1998), again using observations only from the “pre” period. We continued to find no evidence of a pre-trend in total cost or mortality, and evidence of a positive pre-trend in readmissions. 20 We also examined the length of stay of severely ill patients before and after report cards went online for evidence that hospitals in competitive or most competitive markets were discharging these patients earlier so as to reduce the incidence of in-hospital mortality. We found no significant change in length of stay for these

50


Table 3 Information, competition, and patient outcomes.a Log of total cost

Most Competitive × Post Competitive × Post

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

0.0690*** [0.025] 0.0380 [0.024]

0.0631** [0.024] 0.0400* [0.024] 0.0111* [0.007] −0.0039 [0.007]

−0.0008 [0.004] −0.0003 [0.004]

0.0031 [0.004] 0.0046 [0.005] −0.0072** [0.003] −0.0095*** [0.003]

−0.0080 [0.014] −0.0091 [0.013] −0.0110 [0.013] −0.0112 [0.010]

0.475 76,862

0.033 76,862

0.033 76,862

0.0003 [0.005] 0.0032 [0.005] −0.0072** [0.003] −0.0095*** [0.003] −0.0044 [0.004] −0.0013 [0.004] 0.033 76,862

−0.0138 [0.013] −0.0149 [0.013]

0.475 76,862

0.0904*** [0.030] 0.0554* [0.032] 0.0111* [0.007] −0.0039 [0.007] 0.0432 [0.033] 0.0154 [0.027] 0.475 76,862

0.199 74,566

0.199 74,566

0.0038 [0.016] −0.0127 [0.017] −0.0109 [0.013] −0.0112 [0.010] 0.0235* [0.013] −0.0143 [0.012] 0.200 74,566

Most Competitive × Post × Severe Competitive × Post × Severe Most Competitive × Pre Competitive × Pre R-square Observations

1 year readmissionb

In-hospital mortality

a Standard errors, reported in brackets, are clustered by hospitals (67 clusters). All estimations include patient and market characteristics listed in Table A2 in columns titled “CABG patients,” year fixed effects (9 dummies), hospital fixed effects (66 dummies), zip code fixed effects and hospital-specific linear trend. b Sample only includes survivors. * Significant at the 10% level (two-tailed test). ** Significant at the 5% level (two-tailed test). *** Significant at the 1% level (two-tailed test).

The estimates in column (3) suggest that the costs increased by 9.46% (=exp(0.0904) − 1) in the most competitive markets after the online publication of report cards, which represents an increase of $2502 relative to the mean in the pre-period of $26,450 in these markets (figures are in 1995 dollars). The mortality rate for the more severely ill in the most competitive markets was approximately (−0.0072 + 0.0003) or 0.69 percentage points lower, while for severely ill patients in competitive markets the reduction was (−0.0095 + 0.0032) or 0.63 percentage points. 6.1. Contemporaneous shocks Since our identification strategy relies on comparing changes in outcomes before and after report cards went online, we must be particularly careful to control for contemporaneous events that may also have affected how hospitals responded to their competitive environment, potentially altering their patients’ outcomes, during the sample period. We re-estimate our specification adding variables to control for several different contemporary shocks to insure that our results show the effects of report cards, as opposed to the effects of these other contemporaneous changes, on hospital quality competition. First, HMO penetration increased during our sample period. Our specification already includes, as a patient characteristic, a variable indicating whether the patient is a Medicare HMO enrollee, thus controlling for any direct effects of being enrolled in an HMO on a patient’s outcomes. However, there may be indirect effects. Hospitals in areas where there is higher penetration by HMOs among all patients (whether Medicare or not) may be under greater financial pressure, which may in turn limit their ability to adopt new technologies, hire more skilled nurses, or spend money on other quality-improving resources. To the extent that these resources are shared by all patients, HMO penetration could have a spillover effect on health outcomes of non-HMO patients. On the other hand, greater scrutiny of their performance by HMOs might force hospitals to improve efficiency and quality in ways that again may spill over to all patients. We control these possible spillover effects by including the HMO penetration among all patients each year

patients, which is perhaps not surprising when we consider that the PHC4 also reports grades for 30-day mortality rates and 7- and 30-day readmission rates.

for the county containing the patient’s zip code, data found in the Pennsylvania Department of Health, Managed Care Reports. Second, passage of the Balanced Budget Act (BBA) in 1997 may also have affected patient outcomes. The BBA imposed substantial reductions in Medicare payments to hospitals, and empirical research suggests that these reductions were particularly hard on those hospitals with a greater reliance on Medicare reimbursements (Wu, 2010). We control for possible effects of the cost-containment legislation (while continuing to avoid potential endogeneity) by including a variable that is the number of Medicare patients admitted to the hospitals in a county each year, divided by the total number of inpatients to hospitals in a county in that year. Hospitals located in counties with a greater share of Medicare patients are more likely to have experienced financial pressure due to the passage of the BBA. A third source of confounding contemporaneous change arises from Pennsylvania’s termination of CON regulation for CABG programs in December 1996, after which new CABG programs gradually entered the market. New entrants reduced concentration directly in individual markets and may also have disrupted tacit agreements among existing providers (Scherer and Ross, 1990; Ivaldi et al., 2003), so that hospitals in markets experiencing entry were likely to be under greater competitive pressure. Further, previous research has shown that in health care markets entry may result in a redistribution of consumers (Cutler et al., 2010), also increasing competition in the markets. We isolate the impact of new entry from the effect of the online report cards, and also minimize endogeneity problems, by adding a variable that is the predicted share of CABG procedures performed at hospitals that entered the CABG market after 1996. We calculate this variable by summing the predicted patient demand each year, by zip code, for hospitals that opened after 1997, divided by the total predicted demand from each zip code each year, where predicted demand in each case is calculated for each zip code using the same procedure as is used to find the predicted HHIs. Finally, we add two dummy variables to control for any other effects of the current level, as opposed to the 1997 level, of competition in each market. The first dummy equals 1 if a market is currently among the most competitive markets, and the second dummy equals 1 if a market is currently a competitive market. (The categorization of competitive and most competitive markets is again based on the distribution of zip-code level HHIs over the


51

Table 4 Impact of contemporaneous events.a Variables

(1)

(2)

Log of total cost Most Competitive × Post Competitive × Post

−0.0032 [0.005] −0.0008 [0.005]

0.0445 [0.034] 0.0186 [0.028] −0.2502 [0.256] 0.1481 [0.219] −0.0482 [0.071] 0.0622 [0.039] 0.0272 [0.036] 76,862 0.476

76,862 0.476

Competitive × Post × Severe

Competitive × Pre HMO penetration rate Percentage of medicare patients Predicted entrant share Most Competitive (current) Competitive (current) Observations R-squared

(4)

(5)

(6) b

In-hospital mortality 0.0938*** [0.031] 0.0628* [0.034] 0.0111* [0.007] −0.0039 [0.007] 0.0446 [0.034] 0.0186 [0.028] −0.2502 [0.256] 0.1486 [0.219] −0.0477 [0.071] 0.0622 [0.039] 0.0272 [0.036]

0.0997*** [0.032] 0.0607* [0.033]

Most Competitive × Post × Severe

Most Competitive × Pre

(3)

1 year readmission 0.0001 [0.015] −0.0151 [0.017]

−0.0042 [0.004] −0.0010 [0.004] −0.0367* [0.022] −0.0527 [0.043] 0.0028 [0.015] 0.0046 [0.005] 0.0020 [0.003]

0.0006 [0.005] 0.0042 [0.005] −0.0072** [0.003] −0.0095*** [0.003] −0.0042 [0.004] −0.0010 [0.004] −0.0368* [0.022] −0.0530 [0.042] 0.0029 [0.015] 0.0046 [0.005] 0.0020 [0.003]

0.0245* [0.013] −0.0127 [0.013] −0.1124* [0.067] 0.1797 [0.119] −0.0108 [0.033] 0.0258 [0.019] 0.0115 [0.015]

0.0058 [0.016] −0.0093 [0.017] −0.0109 [0.013] −0.0112 [0.010] 0.0245* [0.013] −0.0127 [0.013] −0.1126* [0.067] 0.1790 [0.119] −0.0107 [0.033] 0.0258 [0.019] 0.0115 [0.015]

76,862 0.033

76,862 0.033

74,566 0.200

74,566 0.200

a Standard errors, reported in brackets, are clustered by hospitals (67 clusters). All estimations include patient and market characteristics listed in Table A2 in columns titled “CABG patients,” year fixed effects (9 dummies), hospital fixed effects (66 dummies), zip code fixed effects and hospital-specific linear trend. b Sample only includes survivors. * Significant at the 10% level (two-tailed test). ** Significant at the 5% level (two-tailed test). *** Significant at the 1% level (two-tailed test).

entire period.) Descriptive statistics for each of these variables are shown in the column titled “CABG Patients” in Table A2. Table 4 presents the results of estimations that include these additional variables. First, our main results are virtually identical to those in Table 3, except for an increase in the magnitude of the coefficients for total cost. Second, the estimated coefficients for the HMO penetration variable indicate that the HMO penetration is associated with negative but insignificant reductions in total cost, and weakly significant improvements in patient outcomes, which suggests that any negative spillover effect from the financial pressure introduced by HMOs is not as strong as the impact HMOs have on hospital quality and efficiency. Otherwise, we find no evidence that the percentage of Medicare patients, the predicted entrant share, or contemporaneous HHI had any independent effect on our outcome variables. Nevertheless, we continue to include these variables in all subsequent estimations. 6.2. Robustness checks We test the robustness of our results in a number of ways. We begin by further investigating whether our estimates might be the result of changes in market competitiveness over time due to entry rather than because quality information became available online in 1998. We test for this possibility by including two new variables: the two dummy variables that measure market competition in 1997 interacted with the predicted entrant market share of the hospital market of zip code k at time t. The results are presented in Table 5. First, our main results remain unchanged by the inclusion of the two new interaction terms. Second, the estimations give some evidence that entry into most competitive and competitive markets caused costs and readmission rates to fall relative to more concentrated markets, though costs were not significantly lower in competitive markets. The

results suggest that, ceteris paribus, competitive pressure from entry is associated with what may be increased efficiency as well as higher quality, particularly in markets that are most competitive before entry. Table 6 shows result from several additional robustness checks: in each case Panel A shows the results for total cost, Panel B for mortality, and Panel C for readmission. The main message of this table is that our principal results, that hospitals in the most competitive regions had higher total costs for all patients and that hospitals in competitive and most competitive regions had lower mortality for the more severely ill, remain unchanged. Also, both the estimated coefficient of total cost in competitive markets and the estimated coefficient on total cost for severely ill patients in the most competitive markets are positive and have fairly stable magnitudes but are not always significant. The first two columns show the effect of clustering by residential zip code and by three-digit residential zip code rather than by hospital to test whether our results depend on the method of statistical inference.21 Column (3) presents the results of testing for possible mean reversion by the outcome variables. Mean reversion could affect our results if, for example, total cost was unusually low in competitive markets in 1995: subsequent changes in its value might merely represent reversion to its mean. The specification in column (3) includes a set of year dummies interacted with the mean value of the dependent variable in each zip code in 1995. In column (4) we present results that test whether our estimated effects may be picking up the variation over time in outcomes in markets with different levels of competition. We re-estimated our

21 We try clustering our standard errors by 3-digit as well as 5-digit zip codes in case there is spatial correlation. There are 46 3-digit zip codes in Pennsylvania, and, depending on the year, about 2014 5-digit zip codes.

52


Table 5 Impact of new entry.a Variables

(1) Log of total cost

(2) In-hospital mortality

(3) 1 year readmissionb

Most Competitive × Post

0.1047*** [0.034] 0.0667* [0.037] 0.0106 [0.007] −0.0043 [0.007] 0.0479 [0.034] 0.0200 [0.029] −0.2340** [0.112] −0.1040 [0.141]

0.0019 [0.005] 0.0040 [0.005] −0.0073** [0.003] −0.0095*** [0.003] −0.0038 [0.004] −0.0010 [0.004] −0.0235 [0.016] 0.0039 [0.027]

0.0102 [0.014] −0.0018 [0.016] −0.0111 [0.013] −0.0111 [0.010] 0.0263** [0.012] −0.0108 [0.012] −0.1313* [0.078] −0.2104** [0.082]

76,862 0.476

76,862 0.033

74,566 0.200

Competitive × Post Most Competitive × Post × Severe Competitive × Post × Severe Most Competitive × Pre Competitive × Pre Most Competitive × Entrant share Competitive × Entrant share Observations R-squared

a Standard errors, reported in brackets, are clustered by hospitals (67 clusters). All estimations include patient and market characteristics and contemporaneous events listed in Table A2 in columns titled “CABG patients,” year fixed effects (9 dummies), hospital fixed effects (66 dummies), zip code fixed effects and hospital-specific linear trend. b Sample only includes survivors. * Significant at the 10% level (two-tailed test). ** Significant at the 5% level (two-tailed test). *** Significant at the 1% level (two-tailed test).

specification adding interactions of each of the two competitiveness categories in 1997 with a linear time trend variable, so that the coefficients of our key variables identify the shift in the trends of costs and quality after the release of report cards within each market category.22 It is worth noting that the market-specific time trends are likely to capture some of the effects of the change in the availability of quality information, so that the coefficient estimates for the key variables will tend to be conservative, but, as noted above, our results remain unchanged. Columns (5) through (9) show the effects of using alternate versions of our measures of market competition. First, in our original conditional logit estimations we estimated the probability of a patient going to any hospital within a 50-mile radius, which, given that median distance the Medicare FFS patients traveled to their admitting hospital was about 10 miles, seemed adequate. However, in column (5) we report estimates where the conditional logits are re-estimated giving each patient the choice of all CABG hospitals within a 75-mile radius. Next we explore the possible impact of system membership. In measuring market competitiveness we treat all hospitals providing CABG procedures as independent competitors, focusing on the importance of geographic distance for patients in their choice of hospitals. However, some hospitals belonged to health systems that included more than one CABG provider. It is not obvious how system membership might affect quality competition among hospitals belonging to the same system: any effect is likely to depend on such things as how centralized a system’s management of its CABG programs is, how strong are surgeons’ preferences for quality in all the hospitals they use, and how far away system members are from each other. We tested for an effect by measuring competition in 1997 using predicted “system” HHIs, treating all hospitals

22 We omit the hospital-specific linear trend variables from this specification because of multicollinearity.

in a 50-mile radius that belonged to the same system as a single provider:23 these results are shown in column (6). In column (7) we report the results of measuring competition in 1997 with a measure of predicted HHI that is calculated using only Medicare patients instead of all patients. In column (8) we report the results of replacing the dummy variables Most Competitive and Competitive that are based on the HHIs measured in 1997 with two dummy variables that measure the competitiveness of a market using the HHI as measured in 1996, before hospitals could react to the discontinuation of the CON regulations. In column (9) we report the results of replacing the dummy variables Most Competitive and Competitive based on the 1997 HHIs with two dummy variables that measure the competitiveness of a market using contemporaneous HHI, that is, HHI measured in the year of the patient’s surgery. Again, overall the results in Table 6 continue to suggest that costs are higher in the most competitive markets after report cards went online, and perhaps in competitive markets as well, and that mortality rates for severely ill patients are lower in both. Readmission rates appear largely unaffected by the availability of online quality information; in the few cases where they are significant they may be negative (columns 1 and 8), or positive (column 7). 7. Creamskimming The results we present in Section 6 suggest that when quality information went online, hospitals in more competitive markets increased expenditures on all patients, and mortality rates for severely ill patients fell in competitive and most competitive markets relative to the least competitive markets. We infer that hospitals in more competitive markets are increasing the quality

23 As before, the conditional logit provides the probability of a patient choosing each individual hospital within a 50-mile radius. However, when we aggregated those probabilities to get market shares and HHIs for each zip code, we assumed that hospitals (within the 50-mile radius) that belonged to the same system were a single competitor. Thus, we incorporated system membership into the HHIs during the aggregation process.

Table 6 Robustness checks.a Inference

Log of total cost Most Competitive × Post

Most Competitive × Post × Severe Competitive × Post × Severe In-hospital mortality Most Competitive × Post Competitive × Post Most Competitive × Post × Severe Competitive × Post × Severe 1 year readmissionb Most Competitive × Post Competitive × Post Most Competitive × Post × Severe Competitive × Post × Severe

Alternative HHI measure

(1) Cluster by zip

(2) Cluster by zip3

(3) Mean reversion

(4) Market specific trend

(5) 75 miles

(6) System HHI

(7) Medicare HHI

(8) HHI1996

(9) HHIt

0.0938*** [0.012] 0.0628*** [0.012] 0.0111* [0.006] −0.0039 [0.007]

0.0938*** [0.018] 0.0628*** [0.021] 0.0111 [0.008] −0.0039 [0.006]

0.1145*** [0.038] 0.0506 [0.035] 0.0116* [0.006] −0.0036 [0.007]

0.0871*** [0.028] 0.0581** [0.022] 0.0124* [0.007] −0.0034 [0.008]

0.0953*** [0.031] 0.0578 [0.035] 0.0105 [0.006] −0.0027 [0.006]

0.0933*** [0.030] 0.0571 [0.035] 0.0124* [0.006] −0.0045 [0.007]

0.0761** [0.030] 0.0573** [0.026] 0.0107 [0.008] 0.0035 [0.006]

0.0917*** [0.030] 0.0722** [0.032] 0.0117* [0.007] −0.0037 [0.006]

0.0798** [0.032] 0.0507 [0.036] 0.0062 [0.007] −0.0047 [0.007]

0.0006 [0.005] 0.0042 [0.005] −0.0072** [0.003] −0.0095*** [0.003]

0.0006 [0.005] 0.0042 [0.005] −0.0072** [0.003] −0.0095*** [0.003]

0.0017 [0.005] 0.0033 [0.006] −0.0067** [0.003] −0.0092*** [0.003]

0.0024 [0.008] 0.0045 [0.008] −0.0070** [0.003] −0.0094*** [0.003]

0.0007 [0.005] 0.0049 [0.004] −0.0068** [0.003] −0.0087*** [0.003]

0.0001 [0.005] 0.0015 [0.005] −0.0072** [0.003] −0.0099*** [0.003]

−0.0001 [0.005] 0.0014 [0.004] −0.0061* [0.003] −0.0086*** [0.002]

0.0001 [0.004] 0.0037 [0.005] −0.0064** [0.003] −0.0096*** [0.003]

0.0031 [0.005] 0.0034 [0.005] −0.0083** [0.003] −0.0096*** [0.003]

0.0058 [0.013] −0.0093 [0.012] −0.0109 [0.007] −0.0112* [0.007]

0.0058 [0.011] −0.0093 [0.013] −0.0109 [0.009] −0.0112 [0.009]

0.0111 [0.012] −0.0013 [0.015] −0.0096 [0.013] −0.0091 [0.010]

0.0049 [0.018] 0.0124 [0.017] −0.0127 [0.013] −0.0124 [0.010]

0.0159 [0.013] 0.0191 [0.012] −0.0121 [0.013] −0.0093 [0.010]

0.0106 [0.015] −0.0001 [0.015] −0.0112 [0.013] −0.0110 [0.010]

0.0196* [0.011] 0.0073 [0.014] −0.0061 [0.012] −0.0121 [0.010]

0.0069 [0.014] −0.0049 [0.012] −0.0106 [0.013] −0.0172* [0.009]

0.0074 [0.014] −0.0058 [0.013] −0.0063 [0.013] 0.0021 [0.009]


Competitive × Post

Other specification

a Standard errors, reported in brackets, are clustered by hospitals (67 clusters). All estimations include patient and market characteristics and contemporaneous events listed in Table A2 in columns titled “CABG patients,” year fixed effects (9 dummies), hospital fixed effects (66 dummies), zip code fixed effects and hospital-specific linear trend. b Sample only includes survivors. * Significant at the 10% level (two-tailed test). ** Significant at the 5% level (two-tailed test). *** Significant at the 1% level (two-tailed test).

53

54


Table 7 Probability of different surgical procedures for patients with coronary artery diseases.a

Most Competitive × Post Competitive × Post Most Competitive × Post × Severe Competitive × Post × Severe Sample Observations R-squared

Probability of receiving indicated procedure within one quarter after admission for CAD

CABG patients’ severity

(1) CABG

(2) PCI

(3) Severe

(4) Severe

−0.0050 [0.009] 0.0076 [0.008] 0.0145*** [0.005] −0.0050 [0.006] CADb 371,015 0.159

0.0264** [0.012] 0.0199** [0.009] −0.0066 [0.008] −0.0231*** [0.008] CADb 371,015 0.178

−0.0034 [0.014] −0.0038 [0.014]

−0.0032 [0.015] −0.0036 [0.014]

CABGc 76,862 0.066

CABG survivorsc , d 74,566 0.065

a

Standard errors, reported in brackets, are clustered by hospitals (67 clusters). Sample includes patients admitted to any hospital in PA with a new diagnosis of CAD. Diagnosis codes for CAD are ICD-9-CM 410x-414x. Patients hospitalized for the same diagnosis in the previous three years are excluded. Procedure codes for CABG are ICD-9-CM 3610–3619 and for PCI are ICD-9-CM 3601, 3602, 3605, 3606, 3607, and 3609. All estimations include patient and market characteristics and contemporaneous events listed in Table A2 in columns titled “CAD patients,” year fixed effects (9 dummies), hospital fixed effects (66 dummies), zip code fixed effects and hospital-specific linear trend. c All estimations include patient and market characteristics and contemporaneous events listed in Table A2 in columns titled “CABG patients,” year fixed effects (9 dummies), hospital fixed effects (66 dummies), zip code fixed effects and hospital-specific linear trend. d Sample only includes CABG survivors. ** Significant at the 5% level (two-tailed test). *** Significant at the 1% level (two-tailed test). b

of services provided to all patients, although improving health outcomes only for the severely ill. However, as noted in Section 2, improvements in health outcomes could be the result of creamskimming, behavior which could be particularly prevalent where hospitals face more competitive pressure, so that our mortality results reflect the effects of patient selection by hospitals rather than the effects of quality information. We examine this possibility in two steps. We first look for evidence that online report cards were associated with changes in the patterns of treatment for Medicare patients diagnosed with coronary heart diseases (CAD)24 that might be treated surgically either by CABG or by Percutaneous Coronary Intervention (PCI, also known as balloon angioplasty). PCI is a less invasive and less lucrative procedure than CABG and is generally seen as appropriate for patients with less severe heart disease. However, if hospitals cream skim in response to report cards, they may shift healthier patients from PCI to CABG, so that we might expect to see fewer CABG procedures for sicker patients and fewer PCI procedures for healthier patients in more competitive areas. We identified a sample of Medicare patients that were diagnosed with CAD in the course of a hospital admission during the years of our sample (sample statistics are reported in columns titled “CAD Patients” in Table A2), and estimated the probability of their undergoing either CABG or PCI within one quarter of their initial diagnosis. The results in column (1) of Table 7 show that after report cards went online the probability of CABG increased rather than decreasing in the most competitive markets among more severely ill patients, and the results in column (2) show that the probability of PCI increased rather than decreasing in competitive and more competitive markets. Taken together, these results provide no evidence that the CAD patients that received CABG surgery were systematically healthier after report cards went online. We then test this directly by estimating the probability that a patient is severely ill in competitive and most competitive markets among those patients that did receive CABG surgery before and after report cards went online.

24

Diagnosis codes for CAD are ICD-9-CM 410x-414x.

The results in column (3) of Table 7 (estimated on the sample of all CABG patients) and column (4) (estimated on the sample of CABG patients that did not die during their hospital stay) show no significant change in severity after report cards went online in most competitive or competitive markets relative to the least competitive markets.25 8. Discussion This paper investigates the effect of CABG report cards on hospitals’ quality competition by comparing outcomes, before and after the report cards went online, for CABG patients located in hospital markets in Pennsylvania with varying levels of concentration. Our results suggest that hospitals were more likely to compete along dimensions directly related to health outcomes once clinical information assessing the quality of those outcomes became easily available to the public: after report cards went online, hospitals spent more resources on all CABG patients, and achieved lower mortality for more severely ill patients. If much of hospitals’ additional expenditure was for resources that by their nature were available to all patients once purchased (e.g., hiring a director for the CABG program26 ), then resources used for all patients would increase but would have their greatest impact on the health outcomes of sicker patients. We found no evidence of creamskimming. We ran additional estimations in which we replaced the Post dummy with a series of dummies identifying one, two, three, and four or more years after the report cards went online to get some idea of how quickly the report cards had an effect (results available on request). We found that the total cost increased immediately in both the most competitive and in the competitive markets, but that the improvements in the mortality rate took 2–4 years to achieve, suggesting that the increased expenditures hospitals made to increase quality may have taken a while to have an impact.

25 These results were not changed if we instead measured severity as ranging in value from 0 to 6 or more. 26 See, for example, Chassin (2002).


quality information together produced better outcomes at a reasonably small extra cost. Acknowledgements We are grateful for comments from participants at the Eastern Economic Conference, the ASHEcon Conference, the Federal Trade Commission Bureau of Economics Seminar Series, and the Lehigh University Health Lunch Seminar. We are also grateful for support from the Lehigh Valley Health Network, the Lehigh University College of Business & Economics, the Lehigh University Martindale Center for Private Enterprise, and the Pennsylvania Department of Health (CURE Grant SAP Number 4100054856). Pennsylvania inpatient data are from the Pennsylvania Health Care Cost Containment Council (PHC4). PHC4 is an independent state agency responsible for addressing the problem of escalating health costs, ensuring the quality of healthcare, and increasing access to healthcare for all citizens regardless of ability to pay. PHC4 has provided data to this entity in an effort to further PHC4’s mission of educating the public and containing health care costs in Pennsylvania. PHC4, its agents, and staff, have made no representation, guarantee, or warranty, express or implied, that the data – financial, patient, payer, and physician specific information – provided to this entity, are errorfree, or that the use of the data will avoid differences of opinion or interpretation. This analysis was not prepared by PHC4. PHC4, its agents and staff, bear no responsibility or liability for the results of the analysis, which are solely the opinion of the authors. Appendix. Calculation of HHI We first estimate patient-level conditional logit models of hospital choice separately for each year in the sample, 1995–2004. The conditional logits estimate the probability that a patient will choose each of the different CABG hospitals located within a 50-mile radius of their residential zip code (using a 50-mile radius to identify the set of hospitals among which patients chose seemed reasonable because the median distance traveled to the admitting hospital by the patients we use to estimate the conditional logit was about 10–10.2 miles in 1995 and 9.7 miles in 2004). We estimate the probabilities as a function of hospital characteristics: size, teaching status, and most recent report card grade. Hospital size is measured categorically by the number of staffed beds: fewer than 200, between 200 and 400, and more than 400. A teaching hospital is one that reported at least 20 full-time residents in a given year. Hospital grades are measured categorically

.05

Fraction

.1

.15

Distribution of Predicted HHI in 1997

0

As we noted in the introduction, increasing numbers of people use the web to access health information, but although the PHC4 reports that many inquiries are from consumers with an immediate need for the information, we are cautious about concluding that patients themselves were using the report card grades to choose better hospitals. A quick search of major Pennsylvanian newspapers for 1998 found only two stories in May that mentioned (at the end of the article) that the PHC4 report cards were available online. However, we know that cardiologists and surgeons in Pennsylvania were aware of the report cards (Schneider and Epstein, 1996). Once the grades became more easily available, physicians may have mentioned them to patients, or more likely, used the hospital grades themselves to recommend hospitals, which response may explain the improved health outcomes we observe. Nevertheless, health care consumers’ use of the web to learn about health care and healthcare providers seems very likely to have increased significantly during the period of our study. Our analysis was complicated by Pennsylvania’s decision to end CON regulations during our sample period, because the subsequent entry of new programs increased competition in some markets, which may also have affected quality competition. However, Robinson et al. (2001) found no evidence that the elimination of CON regulations affected the mortality rate of Pennsylvania CABG patients. In contrast, there is evidence that patient flows were affected by quality information in Pennsylvania (Chou et al., accepted for publication; Wang et al., 2011) as well as in New York (Cutler et al., 2004). Our results indicate that the number of hospitals necessary to create competition is not large, consistent with previous empirical research on manufacturing, local service markets, and hospital markets, as well as experimental results, suggesting that three to five firms may be “enough” (Bresnahan and Reiss, 1991; Xiao and Orazem, 2011; Abraham et al., 2007; Huck et al., 2004). The average HHI for the zip codes in “Most Competitive” hospital markets in our sample was 0.205, and the average HHI in the “Competitive” zip codes was 0.443. Since the HHI measure is sensitive to both the number and the size distribution of the firms, it is difficult to relate these numbers to a particular industry structure. However, an HHI of 0.205 lies between the HHIs for markets with four and five equally sized firms, and an HHI of 0.443 lies between the HHI for two and three equally sized firms. The most concentrated hospital markets, where health outcomes were the worst, had an average HHI of 0.680, which lies between the HHI for two equally size hospitals and one hospital. We do not test for effects related to owner-type because virtually all of the CABG hospitals in Pennsylvania are private notfor-profit (NFP). Instead, our results imply that the NFPs responded to competitive pressures in a way similar to what would be expected of profit-maximizing firms, or at least to firms with a profit goal in their objective function, as predicted in Gaynor (2006). However, our results do provide some evidence that new entry and HMO penetration are associated with lower costs and improved quality, suggesting that these developments in hospital markets may be particularly disruptive to incumbents and cause them to improve their performance. Finally, our results suggest that the extra expenditures made by more competitive hospitals had most value for more severely ill patients. Thus, the type of quality competition encouraged by the report cards seems more likely to be desirable, from a taxpayer point of view, than competition on the basis of amenities. Overall, our results suggest that in more competitive hospital markets, the online publication of CABG report cards resulted in a roughly 5–10% reduction in mortality at an additional cost of approximately 2000 dollars per case. Thus, lower concentration plus better

55

.2

.4

.6 Predicted HHI

.8

Least Competitive: HHI>=0.59, Competitive: HHI 0.27-0.59, Most Competitive: HHI = 6 a b

CABG patients

CAD patients

0.0043 0.0086 0.0164 0.0289 0.2429 0.3002 0.2550 0.1148 0.0259 0.6306 0.0080 0.0320 0.3581 0.1788 0.5626 0.1915 0.0452 10.099 (10.378) 0.2962 0.2424 0.1506 0.0739 0.0339 0.0282

0.005 0.010 0.017 0.027 0.174 0.226 0.227 0.171 0.139 0.508 0.010 0.054 0.572 0.156 0.347 0.103 0.035 9.950 (10.440) 0.324 0.259 0.137 0.048 0.016 0.013

CABG patients

CAD patients

396.509 (198.388) 0.649 (0.375)

407.234 (201.649) 0.670 (0.379)

Most Competitive in year t Competitive in year t

0.406 (0.147) 0.427 (0.060) 0.062 (0.142) 0.420 0.423

0.409 (0.146) 0.426 (0.060) 0.067 (0.146) 0.437 0.410

Number of observations

76,862

371,015

Market characteristics Bed sizeb Teaching statusb Contemporaneous events HMO penetration rate (county level) Percentage of medicare patients (county level) Predicted entrant share (zip code level)b

Standard deviations for continuous variables are reported in parentheses. Predicted values are predicted using the estimated patient demand of each hospital in each zip code.

Finally, characteristics of the individual patient making the choice (age, gender, race, source and type of admission, and whether or not live in urban counties) are introduced to the conditional logits by interacting them with the hospital characteristics (size, teaching status, and most recent report card grade). As discussed in the text, we estimate the conditional logits using only the Medicare patients with fee-for-service (FFS) insurance because these patients are not restricted in their choice of hospital by their insurance plan, and so make their choice of provider solely on the basis of their own tastes for hospital quality and convenience. We then assume that the parameter estimates hold for all patients, irrespective of their insurance type, and use the estimates ˆ ih , each patient’s probability of going to each hospital to calculate, in the sample. (We suppress the subscript for Year here and in the following calculations for greater ease of exposition.) Descriptive statistics for this sample of patients are reported in column (2) of Appendix Table A1. As in Kessler and McClellan (2000), calculation of the HHIs based on predicted market shares then proceeds in three steps. In step one, we sum the individual probabilities of patients from zip code k choosing hospital h (h = 1, . . ., H), and divide it by the summed probabilities of patients from zip code k choosing all hospitals in the sample,

˛h,k =

i∈k

ˆ ih

H h=1

i∈k

ˆ ih

which gives us predicted share of patients from zip code k served by hospital h. We use this predicted share to calculate the first step zip-code level HHIs: pat

HHIk

=

H

˛ ˆ 2hk

h=1

This initial HHI measure represents the availability and attractiveness of hospital options from the patient’s point of view,

independent of unobservable patient and hospital characteristics. However, it is not appropriate as a measure of competition for our specification because it reflects the level of hospital competition at each patient’s residential area instead of the level of competition faced by hospitals, which draw patients from numerous zip codes. Therefore, because we are interested in how hospitals differentiate on quality according to the competitiveness of their neighborhoods, our second step is to calculate hospital-level HHIs by averaging the predicted HHIs across zip code-levels for each hospital, with the weight being the predicted share of demand for each hospital from each zip code: hosp

HHIh

=

K

ˆ kh ∗ HHI pat , ˇ k

k=1

where the weight,

ˆ kh = ˇ

i∈k

ˆ ih

i=1

ˆ ih

Nh

,

is those patients of hospital h that live in zip code k, as a fraction of the hospital’s total predicted demand. The last step is to translate this new HHI back to the zipcode level. We calculate the final HHI using a weighted average of hospital-level HHIs, where the weights are the estimated probability of a hospital being chosen by those living in the patient’s residential zip code: pat∗

HHIk

=

H h=1

hosp

˛ ˆ hk ∗ HHIh

We use the same technique to predict hospital size (number of beds) and teaching status, for each zip code. For example, predicted bed size is: pat∗

Bed sizek

=

H h=1

˛ ˆ hk

K k=1

ˆ hk × ˇ

H h=1

˛ ˆ hk ∗ Bed sizeh

.

58


Thus, the variables represent the predicted hospital size and teaching status of hospitals used by patients from each zip code, as determined by the patients’ choices. References Abraham, J.M., Gaynor, M., Vogt, W.B., 2007. Entry and competition in local hospital markets. Journal of Industrial Economics 55, 265–288. Arciero, T.J., Jacobsen, S.J., Reeder, G.S., Frye, R.L., Weston, S.A., Killian, J.M., Roger, V.L., 2004. Temporal trends in the incidence of coronary disease. American Journal of Medicine 117, 274–276. Arrow, K., 1963. Uncertainty and the welfare economics of medical care. American Economic Review 53, 941–973. Baker, L.C., Phipps, C.S., 2002. Managed care, technology adoption, and health care: the adoption of neonatal intensive care. RAND Journal of Economics 33, 524–548. Bresnahan, T.F., Reiss, P.C., 1991. Entry and competition in concentrated markets. Journal of Political Economy 99, 977–1009. Brook, R.H., Kosecoff, J.B., 1988. Competition and quality. Health Affairs 7, 150–161. Bundorf, M.K., Chun, N., Goda, G.S., Kessler, D., 2009. Do markets respond to quality information? The case of fertility clinics. Journal of Health Economics 28 (3), 718–727. Capps, C., Dranove, D., Satterthwaite, M., 2003. Competition and market power in option demand markets. RAND Journal of Economics 34, 737–763. Charlson, M.E., Pompei, P., Ales, K.L., MacKenzie, C.R., 1987. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. Journal of Chronic Disease 40, 373–383. Chassin, M.R., 2002. Achieving and sustaining improved quality: lessons from New York state and cardiac surgery. Health Affairs 21, 40–51. Chen, Y., 2011. Why are health care report cards so bad (good)? Journal of Health Economics 30, 575–590. Chen, Y., Meinecke, J., 2012. Do healthcare report cards cause providers to select patients and raise quality of care? Health Economics 21 (Supp. 1), 33–35. Chou, S.Y., Deily, M.E., Li, S., 2014. Travel distance and health outcomes for scheduled surgery. Medical Care (accepted for publication). Clement, J.P., Bazzoli, G.J., Zhao, M., 2012. Nursing home price and quality responses to publicly reported quality information. Health Services Research 47, 86–105. Cooper, Z., Gibbons, S., Jones, S., McGuire, A., 2011. Does hospital competition save lives? Evidence from the English NHS Patient Choice Reforms. Economic Journal 121 (554), F228–F260. Cutler, D.M., Huckman, R.S., Landrum, M.B., 2004. The role of information in medical markets: an analysis of publicly reported outcomes in cardiac surgery. American Economic Review 94 (2), 342–346. Cutler, D.M., Huckman, R.S., Kolstad, J.T., 2010. Input constraints and the efficiency of entry: lessons from cardiac surgery. American Economic Journal: Economic Policy 2, 51–76. Dafny, L.S., 2005. How do hospitals respond to price changes? American Economic Review 95 (5), 1525–1547. Douglas, G.W., Miller III, J.C., 1974. Quality competition, industry equilibrium, and efficiency in the price-constrained airline market. American Economic Review 64, 657–669. Dranove, D., Jin, G.Z., 2010. Quality disclosure and certification: theory and practice. Journal of Economic Literature 48, 935–963. Dranove, D., Daniel, K., McClellan, M., Satterthwaite, M., 2003. Is more information better? The effects of ‘report cards’ on health care providers. Journal of Political Economy 111, 555–588. Dranove, D., Satterthwaite, M.A., 1992. Monopolistic competition when price and quality are imperfectly observable. RAND Journal of Economics 23 (4), 518–534. Ettinger, W.H., Hylka, S.M., Phillips, R.A., Harrison, L.H., Cyr Jr., J.A., Sussman, A.J., 2008. When things go wrong: the impact of being a statistical outlier in publicly reported coronary artery bypass graft surgery mortality data. American Journal of Medical Quality 23, 90–95. Fox, S., Jones, S., 2009. The Social Life of Internet Users. Pew Internet & American Life Project, Washington, DC. Fung, C.H., Lim, Y.-W., Mattke, S., Damberg, C., Shekelle, P.G., 2008. Systematic review: the evidence that publishing patient care performance data improves quality of care. Annals of Internal Medicine 148, 111–123. Gaynor, M., 2006. What Do We Know About Competition and Quality in Health Care Markets? NBER Working Paper #12301. Gaynor, M., Moreno-Serra, R., Propper, C., 2013. Death by market power: reform, competition and patient outcomes in the National Health Service. American Economic Journal: Economic Policy 5 (4), 134–166. Ginsburg, P.B., Hammons, G.T., 1988. Competition and the quality of care: the importance of information. Inquiry 25, 108–115. Gowrisankaran, G., Town, R.J., 2003. Competition, payers, and hospital quality. Health Services Research 38, 1403–1421. Grabowski, D.C., Town, R.J., 2011. Does information matter? Competition, quality, and the impact of nursing home report cards. Health Services Research 46, 1698–1719. Gravelle, H., Sivey, P., 2009. Imperfect Quality Information in a Quality-Competitive Hospital Market. University of York Department of Economics and Related Studies Discussion Paper.

Gravelle, H., Sivey, P., 2010. Imperfect information in a quality-competitive hospital market. Journal of Health Economics 29, 524–535. Hannan, E.L., Vaughn Sarrazin, M.S., Doran, D.R., Rosenthal, G.E., 2003. Provider profiling and quality improvement efforts in coronary artery bypass graft surgery. Medical Care 41 (10), 1164–1172. Ho, K., 2006. The welfare effects of restricted hospital choice in the US medical care market. Journal of Applied Econometrics 21 (7), 1039–1079. Howard, D.H., Kaplan, B., 2006. Do report cards influence hospital choice? The case of kidney transplantation. Inquiry 43 (2), 150–159. Huck, S., Normann, H.-T., Oechssler, J., 2004. Two are few and four are many: numbers effects in experimental oligopolies. Journal of Economic Behavior and Organization 53, 435–446. Ivaldi, M., Jullien, B., Rey, P., Seabright, P., Tirole, J., 2003. The Economics of Tacit Collusion, Final Report for DG Competition. European Commission. Jung, K., Feldman, R., Scanlon, D., 2011. Where would you go for your next hospitalization? Journal of Health Economics 30, 832–841. Kessler, D.P., Geppert, J.J., 2005. The effects of competition on variation in the quality and cost of medical care. Journal of Economics and Management Strategy 14, 575–589. Kessler, D.P., McClellan, M.B., 2000. Is hospital competition socially wasteful? Quarterly Journal of Economics 115, 577–615. Lu, S.F., 2012. Multitasking, information disclosure and product quality: evidence from nursing homes. Journal of Economics and Management Strategy 21, 673–705. Mukamel, D.B., Zwanziger, J., Bamezi, A., 2002. Hospital competition, resource allocation and quality of care. BMC Health Services Research 2. Mutter, R.L., Wong, H.S., Goldfarb, M.G., 2008. The effects of hospital competition on inpatient quality of care. Inquiry 45 (Fall), 263–279. Park, J., Konetzka, R.T., Werner, R.M., 2011. Performing well on nursing home report cards: does it pay off? Health Services Research 46 (2), 531–554. Pennsylvania Department of Health, 2010. Managed Care Reports. Pennsylvania Health Care Cost Containment Council, 2014. Annual Report (various issues). Picone, G.A., Sloan, F., Chou, S.-Y., Taylor Jr., D.H., 2003. Is medical care productive? Evidence from medicare. Review of Economics & Statistics 85, 51–62. Robinson, J.L., Nash, D.B., Moxey, E., O’Connor, J.P., 2001. Certificate of need and the quality of cardiac surgery. American Journal of Medical Quality 16 (5), 155–160. Romano, P.S., Marcin, J.P., Dai, J.J., Yang, X.D., Dravitz, R.L., Rocke, D.M., Dharmar, M., Li, Z., 2011. Impact of public reporting of coronary artery bypass graft surgery performance data on market share, mortality, and patient selection. Medical Care 49, 1118–1125. Scherer, F.M., Ross, D.M., 1990. Industrial Market Structure and Economic Performance, 3rd ed. Houghton Mifflin Co, Boston, MA. Schneider, E.C., Epstein, A.M., 1996. Influence of cardiac-surgery performance reports on referral practices and access to care: a survey of cardiovascular specialists. New England Journal of Medicine 335, 251–256. Schneider, E.C., Epstein, A.M., 1998. Use of public performance reports: a survey of patients undergoing cardiac surgery. JAMA 279, 1638–1642. Shahian, D.M., Normand, S.-L., Torciana, D.F., Lewis, S.M., Pastore, J.O., Kuntz, R.E., Dreyer, P.I., 2001. Cardiac surgery report cards: comprehensive review and statistical critique. Annals of Thoracic Surgery 72, 2155–2168. Shen, Y.C., 2003. The effects of financial pressure on the quality of care in hospitals. Journal of Health Economics 22, 243–269. Shortell, S.M., Hughes, E.F., 1988. The effects of regulation, competition, and ownership on mortality rates among hospital inpatients. New England Journal of Medicine 318, 1100–1107. Snowbeck, C., 2000. State Hospital Council Report Finds HMOs Approve Too Many Hysterectomies. Pittsburgh Post-Gazette. Stiglitz, J.E., 2000. The contributions of the economics of information to twentieth century economics. Quarterly Journal of Economics 115, 1441–1478. Tay, A., 2003. Assessing competition in hospital care markets: the importance of accounting for quality differentiation. RAND Journal of Economics 34, 786–814. Town, R., Vistnes, G., 2001. Hospital competition in HMO networks. Journal of Health Economics 20, 3–53. Ulrich, M.R., Brock, D.M., Ziskind, A.A., 2003. Analysis of trends in coronary artery bypass grafting and percutaneous coronary intervention rates in Washington state from 1987–2001. American Journal of Cardiology 92, 836–839. Wang, J., Hockenberry, J., Chou, S.Y., Yang, M., 2011. Do bad report cards have consequences? Impacts of publicly reported provider quality information on the CABG market in Pennsylvania. Journal of Health Economics 30 (2), 392–407. Werner, R.M., Konetzka, R.T., Stuart, E.A., Polsky, D., 2011. Changes in patient sorting to nursing homes under public reporting improved patient matching or provider gaming? Health Services Research 46 (April (2)), 555–571. White, L.J., 1972. Quality variation when prices are regulated. Bell Journal of Economics and Management Science 3, 425–436. Wu, V.Y., 2010. Hospital cost shifting revisited: new evidence from the balanced budget act of 1997. International Journal of Health Care Finance and Economics 10, 61–83. Xiao, M., Orazem, P.F., 2011. Does the fourth entrant make any difference? Entry and competition in the early U.S. broadband market. International Journal of Industrial Organization 29, 547–561.

[Text Comprehensibility of Hospital Report Cards].

Hospital report cards: necessary but not sufficient?

Considering potential benefits and consequences of hospital report cards: what are the next steps?

Report cards and quality: do center report cards predict quality or simply predict the next report card?

Public report cards for in-hospital cardiac arrest: empowering the public with location-specific data.

Performance report cards increase adenoma detection rate.

The Role of Severe Dementia in Nursing Home Report Cards.

Competition between global and local online social networks.

Considerations of reliability and validity of transplant center report cards.

Has competition increased hospital technical efficiency?

Online Actions with Offline Impact: How Online Social Networks Influence Online and Offline User Behavior.

Evaluation of online catalog systems for the hospital library.

Hospital consolidation, competition, and quality: is bigger necessarily better?

Impact of online pharmacies on veterinary practices.

Physical activity report cards: Active Healthy Kids Global Alliance and The Lancet Physical Activity Observatory.

The impact of participating in suicide research online.

The Report of Bethlehem Hospital.

The impact of aeration on the competition between polyphosphate accumulating organisms and glycogen accumulating organisms.

Steroid cards: patient compliance.

Showing your cards.

Physician report cards and rankings yield long-lasting hand hygiene compliance exceeding 90%.

Impact of changing health care economics on Michigan hospital libraries: report of a survey.

Speed of Dog Adoption: Impact of Online Photo Traits.

Development and impact of the Iranian hospital performance measurement program.