The effect of reimbursement on the intensity of hospital services.

1. Introduction

We examine how an exogenous change in the average reimbursement of a hospital admission affects within-hospital treatment intensity. Treatment intensity has been shown to be correlated with the quality of a patient's outcome (Picone et al. 2003) and may also be correlated with the non-health-related utility of the patient stay. Both health-related and non-health-related intensity will affect overall patient utility and thus a patient's choice of a hospital. However, we focus on treatment intensity that is specific to the patient and disease, such as number of days in the hospital, the number of procedures or tests, or even whether aspirin was given during the patient's hospital stay. Variation in hospital-level attributes that affect all patient's utility, such as nursing staffing ratios or even waterfalls in the lobby, are not examined. Rather, we analyze changes in clinical inputs that are likely to be correlated with clinical quality.

The exogenous change in reimbursement we study is due to the implementation of the Balanced Budget Act (BBA) of 1997. The BBA lowered hospital reimbursement for Medicare inpatient stays by slowing the update factor for the standard dollar payment rate used in the prospective payment system (PPS). Under PPS, hospitals are paid a set amount per Medicare admission based on a patient's diagnosis rather than being paid on the basis of the intensity of services that a patient actually receives. For more severe cases within a diagnosis group, hospitals do not receive additional reimbursement for marginal services provided during the stay. It is well known that the more severely ill patients will incur costs greater than the reimbursement, whereas less severely ill patients will be profitable. The Medicare system includes outlier payments that are designed to lessen the high-powered incentive to reduce intensity for severely ill patients during the hospital stay. However, the outlier payments related to long length of stay were phased out by 2000 under the BBA, but those based on high charges remain. The U.S. Congress sets the level of Medicare reimbursement each year primarily through the annual inflation update factor. One of the primary provisions of BBA was to hold these updates to values less than actual hospital cost inflation. The effect of the BBA on hospital care is important to study because it is likely indicative of future Medicare policy changes. The hospital industry is unlikely to experience another dramatic change in both marginal and average incentives as it did when Medicare shifted from a cost-based payment system to PPS in 1984. Rather, it is likely that policy changes will result in incremental changes in average reimbursement that were similar to what occurred through the BBA.

2. Background

There is a growing literature of the effect of the BBA on hospitals and patients. Bazzoli et al. (2004) looked at operational decisions pre- and post-BBA, including length of stay. They found that hospitals most susceptible to the provisions reacted to the BBA by cutting full-time equivalent employees per bed and costs per admission. Lindrooth et al. (2005), using an identical measure, found that hospitals most affected by the provisions of the BBA cut nurse staffing relative to less affected hospitals. Volpp et al. (2005) found that the BBA had little effect on the process of care for acute myocardial infarction (AMI) patients.

Although there are only a few very recent studies of BBA effects on hospital care, there have been numerous studies that examined the effects of the shift from retrospective, cost-based Medicare reimbursement to PPS. Frank and Lave (1989) looked at the effect of the shift from cost-based to prospective payment on length of stay of psychiatric patients. They found that length of stay increased for less severe patients but decreased for more severe patients. Such a reaction is consistent with theoretical models of the effect of prospective payment (see, e.g., Hodgkin and McGuire 1994; Ellis and McGuire 1996; Ellis 1998). In this case, quality and amenities are provided to less severe (and more profitable patients) and withheld from the more severe, unprofitable patients. Ellis and McGuire (1996) also looked at a shift from cost-based to prospective payments and decomposed the effect into moral hazard, selection, and practice style effects. The moral hazard effect is a change on the intensive margin, whereas the selection effect reflects changes on the extensive margin. The practice style effect was in the spirit of Dranove (1987), who showed that specialization would occur in response to changes in incentives when there are gains to specialization. Meltzer, Chung, and Basu (2002) examined how competition affected the reaction of hospitals to prospective payment. Their model implied that more severe patients would be differentially affected by the shift. They found that competition led to increased costs before prospective payment and was associated with decreased costs after prospective payment.

Our paper differs from previous research in several ways. First, we examine the effect of a cut in reimbursement across a broadly defined set of diagnostic related groups (DRGs). We take into account that some services are unprofitable, and thus the provision of treatment intensity both pre- and post-BBA will depend on the profitability of the services. This approach is in the spirit of Newhouse (1989), who examined the number of admissions and dumping across DRGs using a PPS profitability index. Second, most of the studies using patient-level data were limited to California. In contrast, we use data from 11 states. Third, the BBA reflected a reduction in the average payment for a stay and affected the marginal payment for an additional day only through the phaseout of length-of-stay outliers. However, reimbursement for charge-based outliers remained intact. Dranove and White (1998) also examined the effect of a change in average reimbursement, but they looked at neither the effect across the distribution intensity nor the effect by disease category.

Our identification strategy is based on the notion that quality, in part, is a public good. Spence (1975) introduced a model where quality is a public good and a firm is not able to offer different levels of quality for different consumers. Thus, the firm is unable to differentiate its product to offer higher quality to high-valuation consumers and lower quality to low-valuation consumers. Rather, the firm would offer a quality that is optimal given the weighted average of each consumer's price, or valuation. Dranove and White (1998) tested this model against an alternative "private good" model of quality in the context of the hospital industry and found that treatment intensity is a public good. Haile and Stein (2002) also showed that quality is a public good using, like Dranove and White, the California inpatient discharge data. They found that heterogeneity in outcomes is due to variations in care across hospitals rather than variations within a hospital. Subsequent researchers have assumed a public good aspect of quality (see, e.g., Gowrisankaran and Town 2001; Tay 2003).

In our application, we identify the estimates by comparing the effect of the BBA on high-Medicare share hospitals to low-Medicare share hospitals. A differential effect of the BBA would exist only if, at least to some extent, treatment intensity is a public good. At high-Medicare share hospitals, a greater portion of total reimbursement emanates from the Medicare program, and thus average reimbursement, when calculated across all patients, falls more sharply for a high-Medicare share hospital than would occur at a low-Medicare share hospital. If treatment intensity was a private good instead of a public good, then a hospital would adapt treatment intensity for Medicare patients regardless of how many Medicare patients they treat. In reality, there may be some aspects of treatment intensity that are public and some that are private. However, it is not possible to econometrically identify private good effects because they are equivalent across hospitals regardless of Medicare share.

There are unobserved trends during our study period of 1996-2000 that might disproportionately affect treatment intensity at high-Medicare versus low-Medicare share hospitals that we do not identify in our analysis. One example is upcoding. Upcoding within a disease category, which was found by Silverman and Skinner (2004) and Dafny (2005), would lead to a lessening effect of the BBA because it would be manifested in higher charges. In our analysis, we examine changes in broadly defined disease categories. Upcoding would lead to higher measured treatment intensity and thus work against our hypothesis. The level of private reimbursement also changed during this period. A report by MedPAC (2002) shows that private payment-to-cost ratios were declining during this period. However, the declines in the MedPAC report were confounded by the inclusion of Medicaid and Medicare health maintenance organizations (HMOs) in the calculation. Regardless of the direction of private reimbursement changes, our findings represent the net effect of payment changes in Medicare on treatment intensity.

3. Conceptual Model

Our model follows Meltzer, Chung, and Basu (2002) and Hodgkin and McGuire (1994). For simplicity, we present the model for a profit-maximizing firm, but the extensions to other objectives are straightforward and covered elsewhere (see, e.g., Hodgkin and McGuire 1994):

[pi] = D([I.sub.s])[p - c(s) - C([I.sub.s])], (1)

where [pi] is profit, D([I.sub.s]) is the demand for care by patients of severity s, [I.sub.s] is the discretionary treatment intensity given to patient with severity s, p is the reimbursement rate, c(s) is the cost of providing minimally acceptable care to patients with severity s, and C([I.sub.s]) is the cost of providing discretionary treatment intensity. The standard properties of cost and demand utility functions hold: D' > 0, D" < 0, c' > 0, and c" > 0. Profits are maximized by choosing a level of discretionary treatment intensity, for each severity level, subject to a profit constraint:

p - c(s) - c([I.sub.s]) [greater than or equal to] [pi], (2)

where [[pi].bar] is the minimum profit level.

First consider the case where the constraint is not binding for the optimal I*. Total differentiation of the first-order condition reveals

[d[I.sub.s]/dp] = [-D([I.sub.s])/D"([I.sub.s])[p - c(s) - c([I.sub.s])] - 2D'([I.sub.s]) c'([I.sub.s]) - D([I.sub.s])c"([I.sub.s])] > 0, (3)

which follows from the shape of the demand and cost functions and profit maximization. Severity, s, enters only into the denominator through c(s), and an increase in severity will lower the denominator, leading to a larger reaction in treatment intensity at higher levels of severity.

When the constraint is binding, the Kuhn-Tucker conditions imply that I* = 0. If there is a further reduction in price, then d[I.sub.s]/dp = 0 because it is neither possible nor ethical (let alone legal) to provide I* < 0. The results hold if providers maximize both profit and patient benefit rather than being pure profit maximizers (Hodgkin and McGuire 1994).

We test two predictions that come out of this simple model. First, a cut in reimbursement will not affect treatment intensity of a service that is already unprofitable. This comes directly from the Kuhn-Tucker conditions where a hospital will set discretionary treatment intensity to zero. Thus, because discretionary treatment intensity is already zero, hospitals will have no cushion to cut treatment intensity further. Second, if a service is profitable, a cut in reimbursement will lead to a quantitatively larger reduction in treatment intensity for the more severe cases than the less severe cases. The effect of price on treatment intensity increases as severity increases. This is due mainly to the fact that payment is prospective and thus price is invariant to change in intensity.

4. Data

The primary data set is the Healthcare Cost and Utilization Project State Inpatient Database (HCUP-SID) for 1996 and 2000. We use the following 11 states in the analysis: Arizona, California, Colorado, Florida, Iowa, Maryland, Massachusetts, New Jersey, New York, Washington, and Wisconsin. We chose these states primarily because the HCUP-SID data were available in both years and reporting was mandatory. We limit our analysis to nonfederal urban short-term general hospitals. The patient demographic variables were drawn from the HCUP-SID data. However, Washington State does not include a race variable, and race was seemingly randomly missing in other states. Thus, we constructed a dummy variable that equals one if race was missing and zero otherwise to control for missing race. We matched the HCUP-SID to hospital-level data from the American Hospital Association's Annual Survey of Hospitals and market-level demographic information from the Area Resource Files.

The primary outcome measure in this analysis is treatment intensity. To derive our measure, we start with gross patient charges. Variation in patient charges reflects the number and types of services given to a patient and also hospital charge policies in each year. Following Lynk (1995) and Keeler, Melnick, and Zwanziger (1999), we control for changes in the charge policy over time by normalizing this variable. The first step is to calculate the weighted average hospital charge for the following 10 DRGs by hospital, payer, and year: 14, 89, 96, 127, 174, 182, 183, 243, 296, and 320. We use the 1996 weights for both years to keep the market basket constant. These DRGs are chosen because they are common and, more important, because there is little variation in severity and treatment intensity within the DRGs. We then divide the gross charge of each patient by the weighted average of these 10 DRGs to obtain the measure of treatment intensity we use in the analysis.

The analysis of treatment intensity is limited to 16 disease categories that encompass over 50 DRGs. These disease categories are consistent with those studied by McClellan (1997), who looked at both the generosity and the amount of risk sharing embedded in the Medicare reimbursement system. We use disease groups rather than specific DRGs so that changes in treatment intensity that may occur if a patient is upcoded from one DRG to another are captured. While such a change in intensity would be associated with a higher reimbursement and, therefore, higher charges, if the change were due strictly to upcoding, we would expect to see an increase in intensity in the postperiod. This works against our underlying hypothesis and thus renders our results conservative.

The first two columns of Table 1 list the disease categories and the associated DRGs. The third column contains an index of the generosity of payment as derived by McClellan (1997). A higher value reflects more generous reimbursement on average than a lower value. Thus, colorectal cancer is the most generously reimbursed disease category, and AMI is the least generously reimbursed. McClellan's analysis was based on 1990 data, and we assume that the relative generosity persisted until 2000. This assumption is valid only if relative reimbursement was stable (with the exception of the BBA) and there were no hospital-level significant technological innovations that would affect costs for one diagnosis but not the others. Since DRG weights are updated periodically to reflect changes in treatment, we examine these weights in 1990, 1996, and 2000 to see if there were any large changes. The last three columns in Table 1 show the DRG weights in 1990, 1996, and 2000. The DRG weights reflect the relative reimbursement rate (i.e., price) for an admission in each disease category. The columns reveal that the DRG weights of each diagnosis have been reasonably constant over time and thus that relative reimbursement was stable. However, there was a meaningful decline in the weights of aortic aneurism-no rupture and chronic obstructive pulmonary disease. An increase in the weight works against our hypothesis in that an increase reflects a relative price increase. A decline in the weight works with our hypothesis in that there was a relative decrease in price over and above the cuts in reimbursement associated with the BBA. Our interpretation of results will account for changes in the relative prices.

Table 2 contains a summary of patient characteristics and the average share of Medicare patients versus other payers by diagnosis. With the exception of prostate cancer, the majority of the patients are female. Most patients are white and are older than 75 years. The Medicare share is generally above 50% for these diagnoses. A few such as breast cancer, rheumatoid arthritis, and stroke are more common among the younger, non-Medicare population. The standard deviation of Medicare share is in the last column. Note that overall the fifth percentile of Medicare share is about 0.30 and that the 95th percentile is about 0.65.

Table 3 presents the means by year of the hospital- and market-level variables in our data. The most striking difference is that the Herfindahl-Hirschman Index increased while beds per capita decreased, which would imply lessened competition over time. In addition, while admissions per bed day increased over the time period, the percent of inpatient and outpatient surgeries decreased over this time period. Finally, major teaching hospitals had a larger share of Medicare admissions in 2000 than they did in 1996.

5. Econometric Methods

We use a quantile regression at the 25th, 50th, 75th, and 95th quantiles to measure the effect of the cut in reimbursement on treatment intensity. The primary specification is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (4)

where i denotes patient, j denotes hospital, and t denotes time; X is a vector of patient, hospital, and market characteristics (as described in Tables 1 and 2); MCR is the share of Medicare patients at hospital j for each disease; and O is a vector of dummy variables reflecting ownership or mission: investor, public, and major teaching. The excluded category is a nonprofit, nonteaching community hospital. The results are identified using a difference-indifference approach. The difference between treatment intensity at the low- and high-Medicare share hospitals in the preperiod is compared to the difference in treatment at low- and high-Medicare share hospitals in the postperiod:

[DELTA]Intensity = ([Intensity.sup.high.sub.2000] - [Intensity.sup.low.sub.2000]) - ([Intensity.sup.high.sub.1996] - [Intensity.sup.low.sub.1996). (5)

If treatment intensity dropped at high- relative to low-Medicare share hospitals, then [DELTA]Intensity would be negative. We estimate Equations 4 and 5 separately, for each diagnosis category; thus, the set of equations is estimated 64 times: once for each disease category for each quantile. As discussed previously, we expect a larger decline at the right-hand side of the distribution (i.e., more severe patients) for generously reimbursed diagnoses. We define high-Medicare share hospitals as hospitals in the 95th percentile of the continuous Medicare share distribution. Similarly, low-Medicare share hospitals are those in the fifth percentile. On average, moving from a low- to high-Medicare share hospital represents a change of about 35 percentage points. This difference is large enough to exclude hospitals that move from one point to another in response to the BBA.

We compute the standard errors of our estimates using a bootstrap with clustering by hospital as suggested by Bertrand, Duflo, and Mullainathan (2004). We draw the entire time series of hospital observations rather than each individual observation within each iteration of the bootstrap. In doing so, we take into account the fact that the hospital has repeated observations over time, and thus we do not underestimate the standard errors. Furthermore, we reestimate the quantile regression (Equation 4) and the difference-in-difference simulations (Equation 5) within each repetition. The number of repetitions used in the bootstrap is 50.

6. Results

The results based on Equation 5 are summarized in Figures 1-3. The underlying 64 sets of parameter estimates based on Equation 4 are available from the corresponding author.

[FIGURES 1-3 OMITTED]

Figures 1-3 show the effect of the BBA on treatment intensity controlling for patient, hospital, and market characteristics (defined previously) at not-for-profit, for-profit, and public hospitals, respectively. The horizontal axis reflects the diagnosis, and the vertical axis is the change in treatment intensity. The four sets of results reflect the 25th, 50th, 75th, and 95th quantiles of treatment intensity. The diseases are ordered from most generous on the left to least generous on the right. In other words, the left-hand side of the x-axis reflects disease with more generous reimbursement, and the right-hand side contains disease with less generous reimbursement. For example, a value of 1 reflects the first row of Table 1, colorectal cancer. A value of 16 reflects the last row of Table 1, AMI. The point estimates and 90% and 95% confidence intervals are graphed. The darker shading reflects the 90% confidence interval, and the lighter shading indicates the broader range of the 95% confidence interval. The solid line in the middle of the confidence interval is the parameter estimate.

The results imply that there is no effect of the BBA in the 25th quantiles of treatment intensity regardless of hospital ownership. At the 50th, 75th, and 95th quantiles, treatment intensity is lower for the more generously reimbursed disease categories at not-for-profit hospitals (Figure 1). However, there is not a significant difference for any of the diseases regardless of generosity for either for-profit or public hospitals (Figures 2 and 3, respectively). Note also that the estimates have larger confidence intervals at the 95th quantile for all ownership types. The results presented in Figures 1-3 are robust to changes in the variables used as controls. For example, if we drop both the patient and the hospital characteristics from the analysis, the standard errors increase, thereby widening the confidence intervals. The only meaningful change is that the results at the 50th quantile are no longer significant for not-for-profit hospitals if only market-level controls are included. The coefficient estimates from the 64 underlying quantile regression are available from the corresponding author. The magnitude, statistical significance, and sign on the control variables vary by disease. For example, only the coefficients on major teaching hospital, bed size, percent of inpatient admissions that were surgical, and patient race (white vs. nonwhite) are somewhat consistent across diseases in the specification of the 75th quantile. White patients generally received less intensity, whereas patients admitted to teaching or large hospitals received more intensity. Patients admitted to hospitals that did a higher percentage of surgeries received less intensity.

7. Discussion

We have shown that high-Medicare share hospitals cut treatment intensity of patients with more generously reimbursed diagnoses in reaction to changes in reimbursement associated with the BBA. This reaction was concentrated in the 50th, 75th, and 95th quantiles of treatment intensity and is measured relative to low-Medicare share hospitals. This result is consistent with the predictions of the theoretical model. Furthermore, the model predicted that we would not observe a change in treatment intensity for diagnoses that were unprofitable. This aspect of the model is also consistent with the data.

Overall, the results reveal a less dramatic change at public hospitals and for-profit hospitals. Public hospitals likely have a different objective function when compared to the average nonprofit hospital and also have access to funds through tax revenue that are unavailable to other ownership types. Thus, it is not surprising that the reaction to payment changes at public hospitals is small relative to nonprofit hospitals. In contrast, for-profit hospitals are predicted to have lower baseline treatment intensity in the basic model but were expected to react similarly to not-for-profit hospitals when a change in reimbursement level occurred (Hodgkin and McGuire 1994). The relative nonresponsiveness of for-profit hospitals likely reflects that they had already cut treatment intensity to minimally acceptable levels before the BBA and thus were unable to cut it further after BBA implementation. This is consistent with Hoerger (1991), who found that profits at for-profit hospitals were more volatile than at not-for-profits. Hoerger suggests that an inability to cut quality (or uncompensated care) in response to reimbursement reductions is the reason for this volatility at for-profit hospitals. This result is also consistent with Ellis (1998), who found that if providers are pure profit maximizers in a duopoly model, then changes in reimbursement levels will not have an effect on the level of services provided.

On the other hand, it may be that for-profit hospitals are simultaneously reducing treatment intensity and upcoding patients' diagnosis categories to obtain higher-Medicare payments. Our results may reflect the net effect of these two reactions. Silverman and Skinner (2004) and Dafny (2005) showed that for-profit hospitals were more likely to upcode; this is possible in our data, though we cannot identify it.

A long-term response by hospitals (regardless of whether they are for-profit or not-for-profit) to a cut in reimbursement of a service below a profitable level is to drop the service. In all likelihood, a hospital cannot distinguish between an average intensity patient and a low intensity patient, ex ante. Thus, a severity threshold, along the lines of Ellis (1998), used to determine whether a patient is dropped, may be greater than zero. In practice, the entire service line may be dropped if it is unprofitable on average. It remains to be seen whether and to what extent unprofitable services were shed in response to the BBA. Recent research by Horwitz (2005) has shown that for-profit hospitals are more likely to add or drop services in response to changes in reimbursement than not-for-profit and public hospitals. However, one of the most generously reimbursed services (cardiac valve disorders) and the least generously reimbursed service (AMI) in our study are part of a bundle of cardiac services, which, as a whole, are known to be quite profitable (Horwitz 2005). High-quality treatment of AMI patients may be a loss leader to attract patients in need of more profitable cardiac valve disorder.

In summary, we show that the BBA affected treatment intensity for disease categories that were generously reimbursed and for patients who required higher treatment intensity. The normative implications of this analysis are less clear. On the one hand, the BBA may improve welfare if the reductions in intensity increased efficiency or spared patients from minimally beneficial services that might put them at harm. On the other hand, as Ellis (1998) and others have shown, it may be the case that quality was increasingly underprovided to the more severely ill patients and overprovided to the less severely ill patients. If this is the case, we find that the BBA pushed the care further below the first-best level for the most severely ill patients while not affecting care for the less severely ill patients. This would likely lead to worsening outcomes among patients who require high treatment intensity.

Received July 2006; accepted September 2006.

References

Bazzoli, Gloria, Richard C. Lindrooth, Romana Hasnain-Wynia, and Jack Needleman. 2004. The Balanced Budget Act of 1997 and U.S. hospital operations. Inquiry 41:401-17.

Bertrand, M., E. Duflo, and S. Mullainathan. 2004. How much should we trust difference-in-differences estimates. Quarterly Journal of Econometrics 119:249-75.

Dafny, Leemore. 2005. How do hospitals respond to price changes? American Economic Review 95:1525-47.

Dranove, David. 1987. Rate-setting by diagnosis related groups and hospital specialization. RAND Journal of Economics 18:417-27.

Dranove, David, and William D. White. 1998. Medicaid-dependent hospitals and their patients: How have they fared? Health Services Research 33:163-85.

Ellis, Randall P. 1998. Creaming, skimping, and dumping: Provider competition on the intensive and extensive margins. Journal of Health Economics 17:537-55.

Ellis, Randall P., and Thomas G. McGuire. 1996. Hospital response to prospective payment: Moral hazard, selection, and practice-style effects. Journal of Health Economics 15:257-77.

Frank, Richard G., and Judith R. Lave. 1989. A comparison of hospital responses to reimbursement policies for Medicaid psychiatric patients. RAND Journal of Economics 20:588-600.

Gowrisankaran, Gautam, and Robert J. Town. 2001. Does competition kill? Hospital quality and competition. Unpublished paper, Washington University in St. Louis.

Haile, Philip A., and Rebecca M. Stein. 2002. Managed care incentives and inpatient complications. Journal of Economics and Management Strategy 11:37-79.

Hodgkin, Dominic, and Thomas G. McGuire. 1994. Payment levels and hospital response to prospective payment. Journal of Health Economics 13:1-29.

Hoerger, Thomas J. 1991. Profit variability in for-profit and not-for-profit hospitals. Journal of Health Economics 10:259-89.

Horwitz, Jill R. 2005. Making profits and providing care: Comparing nonprofit, for-profit, and government hospitals. Health Affairs 24:790-801.

Keeler, Emmett B., Glenn Melnick, and Jack Zwanziger. 1999. The changing effects of competition on non-profit and for-profit hospital pricing behavior. Journal of Health Economics 18:69-86.

Lindrooth, Richard C., Gloria Bazzoli, Jack Needleman, and Romana Hasnain-Wynia. 2005. The effect of changes in hospital reimbursement on staffing decisions at safety net and non-safety net hospitals. Health Services Research 41:701-20.

Lynk, William. 1995. Nonprofit hospital mergers and the exercise of market power. Journal of Law and Economics 38:437-61.

McClellan, Mark. 1997. Hospital reimbursement incentives: An empirical analysis. Journal of Economics and Management Strategy 6:91-128.

MedPAC. 2002. Report to Congress: Medicare payment policy, 2002. Washington, DC: MedPAC.

Meltzer, David, Jeanette Chung, and Anirban Basu. 2002. Does competition under Medicare prospective payment selectively reduce expenditures on high-cost patients? RAND Journal of Economics 33:447-68.

Newhouse, Joeseph P. 1989. Do unprofitable patients face access problems? Health Care Financing Review 11:33-42.

Picone, Gabriel, Frank A. Sloan, Shin-Yi Chou, and Donald H. Taylor. 2003. Does higher hospital cost imply higher quality of care? Review of Economics and Statistics 85:51-62.

Silverman, Elaine, and Jonathan Skinner. 2004. Medicare upcoding and hospital ownership. Journal of Health Economics 23:369-89.

Spence, A. Michael. 1975. Monopoly, quality, and regulation. Bell Journal of Economics 6:417-29.

Tay, Abigail. 2003. Assessing competition in hospital care markets: The importance of accounting for quality differentiation. RAND Journal of Economics 34:786-814.

Volpp, Kevin G., R. Tamara Konetzka, Jingsan Zhu, Lori Parsons, and Eric Peterson. 2005. Effects of cuts in Medicare reimbursement on process and outcomes of care for acute myocardial infarction patients. Circulation 112:2268-75.

Richard C. Lindrooth, * Gloria J. Bazzoli, ([dagger]) and Jan Clement ([double dagger])

* Medical University of South Carolina, 151 Rutledge Avenue, Building B, P.O. Box 250961, Charleston, SC 29425, USA; E-mail [e-mail]; corresponding author.

([dagger]) Virginia Commonwealth University, P.O. Box 9802031008 East Clay Street, Richmond, VA 23298, USA; E-mail [e-mail].

([double dagger]) Virginia Commonwealth University, P.O. Box 9802031008 East Clay Street, Richmond, VA 23298, USA; E-mail [e-mail].

Table 1. Diagnosis Group Characteristics

                                   Diagnosis-Related     Generosity (a)
Disease                              Groups (DRGs)
                                                              1.01
Colorectal cancer (b)            172-173                      0.98
Cardiac valve disorders (b)      135-137                      0.91
Breast cancer (c)                257-262                      0.91
Rheumatoid arthritis (b)         240-241                      0.9
Lung cancer (b)                  82                           0.89
Hip fracture (b)                 236                          0.89
Aortic aneurism-no rupture (b)   121; 130-131                 0.86
Prostate cancer (d)              338; 344; 346-347            0.86
Leukemias/white blood cell (d)   398-399; 400-405; 473
Chronic obstructive pulmonary                                 0.86
  disease (b)                    88                           0.84
Pneumonia (b)                    79-81; 89-91                 0.83
Syncope/presyncope (b)           141-142
Acute upper respiratory                                       0.81
  infection (b)                  96-98                        0.81
Stroke (b)                       14-15; 12                    0.76
Ventricular arythmia (b)         138-139                      0.73
Acute myocardial infarction (b)  115; 121-125; 144-145

                                           DRG      DRG       DRG
                                         Weight    Weight    Weight
Disease                                  (1990)    (1996)    (2000)

Colorectal cancer (b)                     0.94      0.98      1.06
Cardiac valve disorders (b)               0.68      0.75      0.78
Breast cancer (c)                         0.71      0.78      0.83
Rheumatoid arthritis (b)                  0.85      0.91      0.94
Lung cancer (b)                           1.22      1.33      1.38
Hip fracture (b)                          0.74      0.77      0.77
Aortic aneurism-no rupture (b)            0.85      0.76      0.72
Prostate cancer (d)                       0.82      0.88      0.98
Leukemias/white blood cell (d)            1.59      1.77      1.83
Chronic obstructive pulmonary
  disease (b)                             1.00      0.98      0.93
Pneumonia (b)                             1.12      1.10      1.09
Syncope/presyncope (b)                    0.60      0.62      0.64
Acute upper respiratory
  infection (b)                           0.93      0.96      0.95
Stroke (b)                                0.74      0.74      0.75
Ventricular arythmia (b)                  0.68      0.65      0.67
Acute myocardial infarction (b)           1.42      1.49      1.50

(a) From McClellan (1997, table 4).

(b) Medical diagnoses.

(c) Surgical diagnoses.

(d) Mix of medical and surgical diagnoses.

Table 2. Patient Characteristics, by Diagnosis Group

                                                        Age     Missing
Disease                              Female   White   (65-75)    Race

Colorectal cancer                     52%      63%      39%       20%
Cardiac valve disorders               66%      72%      19%       15%
Breast cancer                         99%      67%      52%       21%
Rheumatoid arthritis                  76%      60%      33%       20%
Lung cancer                           51%      64%      50%       22%
Hip fracture                          59%      66%      36%       19%
Aortic aneurism-no rupture            80%      69%      16%       25%
Prostate cancer                        0%      63%      51%       18%
Leukemias/white blood cell            54%      65%      44%       23%
Chronic obstructive pulmonary
  disease                             60%      67%      43%       18%
Pneumonia                             54%      65%      27%       21%
Syncope/presyncope                    61%      70%      29%       12%
Acute upper respiratory infection     60%      65%      32%       19%
Stroke                                71%      58%      29%       17%
Ventricular arythmia                  60%      70%      38%       19%
Acute myocardial infarction           53%      64%      38%       21%

                                     Medicare     Medicare Share
Disease                              Share     (Standard Deviation)

Colorectal cancer                     0.66             0.16
Cardiac valve disorders               0.66             0.26
Breast cancer                         0.39             0.15
Rheumatoid arthritis                  0.46             0.23
Lung cancer                           0.63             0.13
Hip fracture                          0.63             0.12
Aortic aneurism-no rupture            0.80             0.15
Prostate cancer                       0.72             0.20
Leukemias/white blood cell            0.52             0.19
Chronic obstructive pulmonary
  disease                             0.74             0.08
Pneumonia                             0.71             0.10
Syncope/presyncope                    0.73             0.10
Acute upper respiratory infection     0.75             0.09
Stroke                                0.25             0.13
Ventricular arythmia                  0.72             0.08
Acute myocardial infarction           0.66             0.10

Table 3. Hospital and Market Characteristics, by Year

                                                          Year

                                                   1996         2000

Hospital characteristics
  Investor owned                                   0.014        0.003
                                                  (0.116)      (0.056)
  Public                                           0.040        0.034
                                                  (0.195)      (0.181)
  Major teaching hospital                          0.118        0.192
                                                  (0.323)      (0.394)
  Medicare share of inpatient discharges           0.692        0.685
                                                  (0.142)      (0.142)
  Total beds                                     314.183      396.056
                                                (158.564)    (395.608)
  Admissions per bed day                           0.127        0.140
                                                  (0.035)      (0.038)
  Registered nurses per adjusted patient day       0.003        0.003
                                                  (0.001)      (0.001)
  Outpatient surgeries/outpatient visits           0.051        0.049
                                                  (0.039)      (0.040)
  Inpatient surgeries/inpatient visits             0.324        0.310
                                                  (0.102)      (0.116)
  Nonemergent outpatient visits/total              8.709        9.066
    outpatient visits                             (4.739)      (4.545)

Market characteristics

  Herfindahl-Hirschman Index                       0.208        0.258
                                                  (0.209)      (0.210)
  Percent of residents age 65 or older             0.130        0.126
                                                  (0.018)      (0.017)
  Per capita income (1000s)                       27.328       31.625
                                                  (4.866)      (6.460)
  Overall hospital occupancy rate of               0.641        0.656
    metropolitan statistical area (MSA)           (0.124)      (0.106)
  Hospital beds per capita in MSA                  0.371        0.339
                                                  (0.528)      (0.449)
  Number of observations                          136,876      165,657