Hospital Bailouts and Economic Efficiency



Hospital Closure and Economic Efficiency

Cory Capps*

U. S. Department of Justice

600 E St. NW, Suite 10000

Washington, DC 20530

David Dranove

Department of Management and Strategy

Kellogg School of Management

Northwestern University

2001 Sheridan Rd.

Evanston, IL 60208

Richard C. Lindrooth

Department of Health Administration and Policy

Center for Health Economic and Policy Studies

Medical University of South Carolina

19 Hagood Rd

Charleston, SC 29425

(lindrorc@musc.edu)

843-792-2192

JanuaryApril, 2005

Abstract: We present a framework for assessing the effects of hospital closures measure the effect of five hospital closures on patient and total welfare. While patient welfare necessarily declines because some patients lose access to a hospital, closures also affect costs. Recent research suggests that less efficient institutions are more likely to close and that surrounding hospitals are able to increase efficiency as result of scale economies. Thus, the net effect of closures, and the wisdom of hospital bailouts, is an empirical question. We study five hospital closures in two states and find that hospital bailouts are usually unwise: on balance the cost savings from closures more than offset the reduction in patient welfare. However, in at least one case, closure led to a decline in welfare and an argument for a small bailout could have been warranted.

Preliminary: Please do not cite or quote without checking with the authors for a revised manuscript.

* The views expressed herein are the author’s and not necessarilydo not purport to represent those of the U.S. Dept. of Justice.

Hospital Closure and Economic Efficiency

1. Introduction

The 1990s witnessed the closure of roughly 500 community hospitals across the United States. Communities often responded more strongly to the closure of a local hospital than they would to the loss of other local businesses. In addition to a loss of jobs, hospital closures also meant mean that local residents might have to turn to less attractive alternatives for health care. Accordingly, outcry among community groups and hospital administrators often has led to government intervention in order to keep ailing hospitals afloat. For example, officials in Quincy, Massachusetts, successfully lobbied for a $12.1 million bailout of Quincy Hospital to facilitate its acquisition by a nonprofit enterprise in 1999, and officials in Tampa, Florida, authorized $3.5 million from local tax revenue to bailout Tampa General Hospital in 2000.

Critics of these bailouts argue that the survival of hospitals should be left to market forces. They argue that if a hospital is not financially viable, then it must be either (a) inefficient, (b) in low demand, or (c) both. Why then bail it out? Lindrooth, LoSasso, and Bazzoli (2003) found that in the early to mid-1990s the occupancy rate at future closures was about 48% versus a rate of over 64% at their competitors. This suggests both that local residents did not place much value on these hospitals and that they would be able to find alternative sources for care. However, Lindrooth, LoSasso, and Bazzoli (2003) did not look directly at the effect of closure on patients, focusing instead on hospital costs. Moreover, they tried to study closures that had minimal effects on patient access by limiting the study to closures of hospitals with competitors within 5 miles.

In this study, we measure the effect of closure on hospital costs and on patient access. Measuring both effects is important because it allows us to gauge whether urban hospital bailouts are an efficient use of taxpayer money. This analysis has two parts. First, we implement a methodology designed to measure the value that a hospital brings to its local communitya set of patients. This methodology, elaborated in Capps, Dranove, and Satterthwaite (“CDS”, 2003), uses detailed patient level data to determine the factors that attract patients to different hospitals. A hospital’s value is a function not just of its own characteristics but also of the proximity and degree of service overlap of competing hospitals. For example, CDS show that a suburban hospital that has the only maternity ward in its area may bring more value to its community than a teaching hospital that has several neighboring teaching hospitals.

We use a structural demand model and information about actual patient choices to compute the value of each hospital in our sample, as well as the reduction in utility should a hospital close. For reasons explained below, we denominate the change in utility due to closure in hours of driving time. Specifically, we compute the number of additional hours that would need to be driven pre-closure in order to reduce the patient utility by the same amount as the closure did. If we were working in pricescomputing income changes rather than time changes, this would be the “equivalent variation” of the hospital closure. Our measure accounts for differences in quality and other idiosyncratic elements of the closing hospital beyond simply its location. Then, by drawing on estimates of the value of driving time, we are able to dollar-denominate the utility losses from a closure.

Next, we measure changes in market cost efficiency using the simulation methods of Lindrooth, LoSasso, and Bazzoli (2003). We estimate the aggregate cost changes in the market by combining estimates of where the patients who were treated at the closed hospital would be admitted post-closure with parameter estimates from a multi-product trans-log cost function. Thus, we simulate the costs of treating the closed-hospital’s patients at other hospitals and compare those to the costs at the closed hospital.

Our measure of total welfare changes due to closure is simply the sum of the access and cost effects. If a hospital is very inefficient or if it has very close competitors either in location or product space, the cost savings will likely overshadow the effect of reduced access. Such hospitals should be allowed to close without intervention from the government. If a hospital is relatively efficient and/or it has a unique location or services that patients value highly, the welfare loss associated with reduced access may be larger than the cost savings, and government intervention may be appropriate.

We use hospital discharge data from Arizona and Florida to measure the value of hospitals in metropolitan areas. For analytic convenience, we focus on the Phoenix, AZ, Tucson, AZ, and Tampa, FL, which have all markets that have experienced closure, and compute the social value of these closures.

2. Background

Research on Hospital Costs

If a hospital closure leads to a decline in social welfare, it raises the question of why an efficient hospital in a valued location would close. Mobley and Frech (1994) found that expected future growth and size were significant determinants of closure. Deily, McKay, and Dorner (2000) found that hospital inefficiency explained a meaningful portion of the probability of closure. Similarly, Ciliberto and Lindrooth (2004) found that inefficiency was a significant predictor of closure. They also discovered that third party payment generosity predicted closures, which suggests that some efficient hospitals could close. Payment generosity among private insurers would likely be positively correlated with the value that a hospital brings to the market, as hospitals with attributes demanded by patients would be better able to negotiate payment rates. However, it is not clear whether government payers, especially Medicaid, would reward valued hospitals in the same way.

Lindrooth, LoSasso, and Bazzoli (2003) found that closure led to an evolutionary improvement in the efficiency of the hospital market. This increase was due more to filling beds at neighboring hospitals and the resulting scale economies than to the baseline inefficiency of the closed hospital. In a different context, several papers have measured the cost of an empty hospital bed and reached conclusions consistent with Lindrooth, LoSasso, and Bazzoli (Gaynor and Andersen, 1995; Keeler and Ying, 1996; and Pauly and Wilson; 1986).

The notion of bailing out a failing firm would not normally arise in the context of traditional business ventures, except as a political matter. In general, a failed firm has costs that exceed the value it brings to consumers. But the hospital industry, comprised of large numbers of non-profit and local government-owned general hospitals, may be characterized by a number of distortions such that “ongoing enterprise” and “value exceed costs” are not synonymous.

We might expect for-profit hospitals to exit markets when their ability to translate value creation into revenues no longer exceeds the cost of remaining in business (Wedig, et. al 1989). We have no such expectations of nonprofit and government-owned hospitals, whose missions may cause them to remain in operation even when the long-term financial outlook is bleak. Thus, an efficient for-profit hospital may depart a market when a nonprofit would persist.[1] Lending support to this notion, Bazzoli and Andes (1995) and Duffy and Friedman (1993) both showed that distressed non-profit hospitals linger in the market, despite financial difficulties,. In addition, many studies have shown that for-profit status is a significant predictor of exit (Ciliberto and Lindrooth, 2004, Succi, et al; Wedig et al 1989; Williams et al., 1992). This all suggests that nonprofit closures could improve the net balance of patient welfare and costs; if anything, the closures may happen less often than is optimal.

On the other hand, other distortions give credence to bailouts. Hospital markets are imperfectly competitive, and hospitals cannot perfectly price discriminate. Thus, the total social surplus generated by a hospital may substantially exceed its profits. Hospitals that rely on Medicaid payments may be especially hard pressed to generate revenues commensurate with value created. Medicaid payments often exceed variable costs (to encourage hospitals to admit Medicaid patients) but do may not cover total average total costs of care. Legal restrictions on balance billing Medicaid enrollees may prevent these hospitals from recouping revenues commensurate with the value they deliver to their communities.

3. Methods

Patient Welfare

To measure the value a hospital brings to a networkmarket, we follow the framework in CDS. That paper focused on estimating the value of a hospital in the context of studying market power and negotiations with managed care organizations, but its framework isis the techniques are also well suited for analyzing hospital closures. In the former context, a hospital commands higher prices in proportion to the reduction in consumers’ utility from its withdrawal from the network. In the current setting, we use this methodology forapply a somewhat different interpretation. We compute , computing the reduction in consumers’ utility when a hospital withdraws completely from a market. To do so, we measure the aggregate difference between patient welfare when a hospital is in operation and when it is not, and we interpret this difference to be the willingness to pay (WTP) to keep a hospital open.

The starting point is a logit model of the utility that patient i derives from being admitted for care at hospital j:[pic]

[pic] (1)

[pic] is a column vector of hospital j’s characteristics, where the vector of variables in Rj includes features that are common across all patient conditions, such as teaching status, and the vector of variables in Sj includes indicators of service offerings whose value depends on the patient’s specific condition, such as delivery rooms. The column vector[pic] is patient i’s type and includes both his socioeconomic characteristics Yi and his clinical attributes Zi that affect what services he may need. [pic] is the out-of-pocket price that patient i with clinical characteristics Zi pays at hospital j. The variable[pic] is the geographical location of his home, and [pic]is the approximate travel time from his residence zip code to hospital j. The function [pic]converts money to utils; it is the utility value of $1 to patient i with characteristics[pic],.[2] As in CDS, we assume ((.) is constant. Finally, the error term εij represents the personal and idiosyncratic component of patient i’s evaluation of hospital j.

Because we intend to value hospitals in terms of travel time, we make the obvious but essential point that patients do not make single one-way trips. Instead each patient makes, or hopes to make, at least one round trip; therefore, at a minimum we should measure Tij as twice the one-way travel time. If the patient has a spouse or family, there may be additional associated trips.[3] We are unaware of any studies that indicate the expected number of round trips associated with a hospital stay. Thus, we estimate the model using an alternative measures of Tij: (a) two times the one-way driving time; and (b) six times the one-way driving timethe effective driving time, computed as the actual driving time times the expected length of stay for the patient’s DRG.[4]. In using our methods to prospectively evaluate a bailout, it would be possible through survey research to more precisely measure the travel patterns of affected patients the number of trips per day in a hospital stay. Note that a linear transformation of travel time do not affect the prediction of which hospital a hospital will choose, it will however affect the utility cost of losing access to a preferred alternative.

Under the logit model of demand, the formula for the utility value of a particular choice set has beenis well established by previous literature. The expected value of the utility-maximizing option to patient i, who can choose from a set of hospitals, G={1, 2, …, G}, equals

[pic] (2)

Thus, i's utility loss should hospital k close is then [pic]. The community’s utility loss is derived by integrating this utility difference over the population distribution of patients and their characteristics, [pic]. The final step required to convert this difference into a market level measure of the utility loss from a hospital closure is to compute the expectation over patients' potential clinical diagnoses, Zi, because bailout decisions are clearly made prospectively, before patients’ conditions are realized. As shown in CDS, the formula for this value integral is

[pic] (3)

In equation (3), N is the total population in the market, [pic] is hospital k's market share of patients with characteristics [pic], [pic]is the joint density of the demographics, clinical indications, and locations of all consumers who will be sufficiently ill during the next year to cause them to require hospitalization, and 1/ γ is the util-to-dollar conversion factor. Because the constant γ is not identified, (3) gives a util-denominated estimate of the utility loss from the closure of hospital k, up to γ.

Putting Assigning a Dollar Value on to a Util

Equation (3) allows us to derive the total patient utility loss from a closure, measured in utils. To convert this value from utils to dollars, we observe that the travel time coefficients in a patient’s underlying utility function, (1), provide the utility loss from traveling, in utils per hour. We combine (1) and (3) to compute “travel time equivalent of closure,” which is the total time that would have to be added to all hospital commutes so as to generate an equivalent reduction in utility. We then use plausible estimates of the dollar value of driving time to convert this figure into dollars.

This indirect approach would not be necessary if we could directly place a dollar value on utils, but this would only be possible if we included out-of-pocket price in the choice model. This is not practical for two reasons. First, there is little if any variation across hospitals in the prices paid by manyost patients. Patients with fee-for-service insurance, Medicare patients, and Medicaid patients face essentially no variation across hospitals in their out-of-pocket expense for receiving care. Patients in a preferred provider organization (PPO) and managed care organizations (MCOs) face prices that vary depending on whether the provider is in or out of the insurer’s network, which we can not observe. While the inclusiveness of PPO networks is not known with certainty, based on conversations with industry participants we believe that most PPO networks include most hospitals, thus eliminating virtually all inter-hospital price differences.[5] Second, information about out-of-pocket prices is not available.

Due to this lack of price-variation, patients’ marginal utility of income is not identified and cannot be estimated. As a result, we cannot directly convert utils into dollars. Instead, we use our model to convert utils into the “travel time equivalent of closure.” Then, with the time-denominated equivalent in hand, we compute approximate dollar-denominated utility losses for each closure by drawing on estimates of the value of travel time from the transportation economics literature.

While the transportation literature gives no single number for the dollar value of an hour of time, the range of estimates is consistently between $7.00 and $20.00 per hour. For instance, Small, Winston, and Yan (2002) studied commuters’ willingness to pay for the use of congestion-free express lanes in Los Angeles. Using a revealed preference model (preferences are revealed by whether commuters choose to pay for express travel), they found that “the median value of time based on commuters' revealed preferences is $20.36/hour; at 88% of the average wage, it is toward the top of the range expected from previous work." Additionally, theThey also report that the value of time computed from surveys of commuters is $9.22 per hour. A similar study (Brownstone et al., 2003) of travel patterns in San Diego found the median value of one hour to be $15.00.[6] In our study, a disproportionate number of patients affected by hospital closure have lower incomes, as evidenced by the large numbers of Medicaid enrollments and uninsured patients. Thus, we present results using time-valuations of $7.00 and $12.00 we present a range of possible values of time, loosely reflecting the minimum, and average, and maximum estimates of the value of one hour.

To summarize, for a given hospital closure, we derive the market-wide change in driving time that would have the same effect on aggregate utility in the market as the closure does. Note that this is not a measure of the change in actual drive times due to closure; that factor is captured directly by the reduction of ex post utility of all patients who would have chosen the closing hospital.. Instead, we are computing compute how much farther away each patient would have to be from all hospitals, assuming the closure did not occur, in order to be precisely as worse-off as he is following the closure.

While being farther from all hospitals in a market is physically impossible for most residents, we believe this approach makes intuitive sense. For patients who strongly prefer the closing hospital, we may find that the closure is “just as bad” as having to drive an additional hour to receive inpatient care. If we then arbitrarily assume that the value of time is $1216.00/hour, we can infer that such patients would pay $12 16 at the start of the year in order to retain the option of receiving care at the closing hospitals should the need arise.[7] Other patients may have many options that are nearly as attractive as the closing hospital. For such patients, the closure might be “just as bad” as having to drive an extra 4 minutes to receive care, and they would only be willing to pay $1.8007 to retain the option of receiving care at the closing hospital. By adding up these values, we can compute the aggregate of patients’ willingness to pay in a market to keep a given hospital open.

Specifically we seek to find, for each patient i, the[pic] such that, for a the closure of hospital k:

[pic] (4)

Equation (4) must be integrated over patient diagnoses, demographics, and locations in order to compute a market level estimate. Note, however, that both sides of Equation (4) are denominated in utils; the right hand side is the utility loss resulting from hospital k closing – i.e., the WTP for hospital k derived in Equation (3) – and the left hand side is the time increase with the equivalent effect on utility. The [pic] that solves Equation (4) has an established translation to into dollar values, which we will use to translate the utility loss due to closure from utils into dollars. Thus, using Equation (3) and integrating over diagnosis, demographics, and location yields:

[pic] (5)

or equivalently,

[pic] (6)

where [pic], the WTP for hospital j up to the unknown constant γ. We solve for [pic]* Equation (6) to discover the drive-time equivalent, in utils, of the value a hospital brings to its community. Rearranging terms, and expressing the integral in discrete terms yields the final formula we use for constructing the equivalent time change for the closure of hospital k: [8]

[pic].

Finally, note that Tthe logit assumption implies that [pic] has a closed form solution:

[pic] (7)

Equation (7) measures the reduction in expected utility to a single patient, i, with characteristics [pic] from a marginal increase in travel time to all hospitals. It implies that changing the drive time to a given hospital will have an effect that is proportional to that hospital's share of patients similar to i (which is an index of the attractiveness of a given hospital to patients like i) and also is determined by how important travel time is within the utility function. The derivatives of U(.) with respect to t in Equation (7) all have analytic solutions that are computable from the interaction from the estimated coefficients and the observed levels of the Hs and Xs, as can be seen from Equation (1).

Finally, the [pic]* implied by Equation (6) is converted to hours and multiplied by the dollar value of time to yield our monetary estimates of the value a hospital brings to the community.

Change in Operating Costs

Because patients value choice, every hospital closure is associated with a some loss in patient utility, though it may be nominal. This may be offset by decreases in costs, however, so that the closures may be socially efficient. Should hospital k close, there will be cost savings equal to the cost of care at hospital k; however, there will also be cost increases at hospitals other than k.

Implementing the translog cost function first requires knowing how patients will allocate themselves across hospitals after k is no longer an option. The coefficients from (1) provide just such an estimate. The predicted probability of a patient with characteristics [pic] of choosing a given hospital j, is given by

[pic] (8)

An alternative interpretation of (8) is that as hospital j's expected market share among patients with characteristics[pic].

Our approach to measuring changes in costs follows Lindrooth, Lo Sasso, and Bazzoli (2003). First, we use the predicted shares from Equation (8) to estimate the effect of closure on the admissions at neighboring hospitals. Specifically, we first estimate the expected number of admissions for the last year the hospital was in operation, denoted E(ADMPRE), and the expected number of admissions at each hospital if the hospital was closed, E(ADMPOST). E(ADMPOST) is calculated using the same parameter estimates the were used to estimate E(ADMPRE), but we eliminate the closed hospital as an option and re-normalize the predicted probabilities so that they sum to one, prior to calculating the expected number of admissions. The change in admissions due to closure at hospital j is then calculated as:

[pic] (9)

Given the change in admissions at each hospital j, we simulate the change in market costs that follow the closure using the parameter estimates based on a national sample of hospitals used in Lindrooth, Lo Sasso, and Bazzoli (2003). Specifically, we assume a short-run translog hospital cost function for the hospital:

[pic] (10)

where Cj is total operating cost; Nj is a vector of all hospital outputs; Ij is a vector of quasi-fixed hospital inputs; Patient mixj is a vector of variables reflecting the payer and case mix at the hospital; Wagesj is the average payroll expense per FTE; and Hospitalj includes dummies indicating ownership and teaching status. We estimate the total cost of treating the patients after closure by adding the change in expected admissions:

[pic] (11)

In addition, we allow ED visits to change because a substantial portion of patients is admitted through the ED. Thus, the difference in market costs due to closure is then:

[pic] (12)

where n is the number of hospitals that closed.

Change in Welfare

The total change in welfare is the dollar denominated WTP, measured by Equation (6), plus the change in market costs, measured by Equation (12). The total change will be positive if the change in market costs is negative and larger in magnitude than the WTP reduction.

4. Data

Our inpatient claims discharge data are from the State Inpatient Database of the Healthcare Cost and Utilization Project (HCUP-SID). We use three two years of data from the Tampa market (1995, 1996, and 1998), one year from Tucson (19991998), and one year from Phoenix (19981997). The years were chosen so that we have a full year of data for the lastat least two years prior to the closure. the hospital was in operation. We narrow our focus to Tampa, Tucson, and Phoenix for several reasons. First, the closures in these markets all occurred after 1995, a period for which the requisite data are available from HCUP. Second, these are all small to mid-sized MSAs with relatively well-defined geographic market areas. Finally, each market has few enough hospitals that the calculations were computationally practical.

In the baseline specifications, we estimate the multinomial logit parameters separately for three groups of patients. The first consists of using a sample of patients who have either Medicare, fee-for-service, or PPO insurance coverage, the patients . We limit the sample to these patients because they are most likely to have access to all of the hospitals in the market. This eliminates potential biases caused by omitting out of pocket price from the model or by misspecification of the choice set. However, our simulations are based on all patients in the market.this group of patients accounts for only about 60% of patients. Our second sample consists of patients enrolled in managed care, whether commercial or as part of Medicaid or Medicare (though we dummy for the latter two payers in the logit model). These patients generally have less discretion over where they receive care. Our third group consists primarily of patients sometimes viewed by providers as undesirable, or at least, unprofitable: Medicaid patients, indigent patients, self-pay patients, and other patients. Thus, in defining our three sets of patients, we have tried to group patients with similar choice decisions together.

We run separate models for admissions through the emergency department and non-emergency admissions because we believe the underlying processes for choosing a hospital are fundamentally different for emergency and non-emergency patients. Additionally, admissions of patients with DRG’s indicating mental health, substance abuse care, rehabilitation care, or psychiatric care are eliminated because these services are not typically performed in general acute care hospitals. Newborns are also dropped to avoid counting a delivery as two admissions. (We implicitly assume that newborns place zero value on driving time!)

In the Tampa market we include all of the general hospitals and patients, subject to the criteria described above, in Hillsborough, Pasco, and Pinellas counties in these counties in the analysis. We add additional patients from surrounding zip codes if more than 60 percent of residents of the zip code sought care in the Tampa-St. PetersbergPetersburg market. This results in the inclusion of patients from a few zip codes in mainly western Polk County.

The markets in Tucson and Phoenix are well-defined using 3-digit zip codes. In Phoenix, the following 3-digit zip codes are used: 852, 853, 855, 856, 857, 859, 860, and 865. Over 96 percent of admissions to hospitals in Phoenix resided in either 856 or 857. In Tucson, we use the following 3-digit zip codes: 850, 852, 853, 855, 856, 859, 860, 863, 864, and 865. Over 95 percent of admissions to TusconTucson hospitals came from 850, 852, or 853. Maps of each market are in the appendix.

5. Model Specification

The baseline specification of the choice model includes travel time and dummies for the type of control (for profit, non-profit, and government), for whether the hospital offers transplants, and for teaching status. The model includes one continuous hospital variable, nursing intensity, which measures nursing hours per inpatient day. Each hospital variable is included in isolation and is interacted with travel time. This allows, for example, the data to show whether patients prefer teaching hospitals, or whether patients are willing to travel farther for hospitals that offer transplants. Hospital characteristics are also interacted with patient characteristics, allowing the marginal utility of hospital characteristics to vary according to the patient’s demographic and clinical characteristics.

Additionally, we include “service match” dummies for the following conditions: Oncology, HIV, Labor and Delivery, Circulatory, Transplants, and Pediatric. Patients in each of these categories are unlike to elect to receive care at hospitals without corresponding specialized services. For example, nearly all births occur in hospitals with dedicated labor and delivery rooms, a service typically maintained by 40-60% of hospitals in a market. Similarly, no transplant would ever occur in a hospital that does not offer transplants. If we did not include these match variables, we would underestimate travel aversion for these types of patients because they frequently bypass the closest hospitals for a mores distant one with matching services offerings. Table A1 in the appendix summarizes the independent variables used in the baseline choice model.

For the cost side of the model, we use a trans-log specification: [pic]

Total Costj is total operating cost; Nj is a vector of all hospital outputs, such as hospital and skilled nursing facility admissions; Ij is a vector of quasi-fixed hospital inputs, such as hospital and skilled nursing facility beds; Wagesj is the average payroll expense per FTE; Patient mixj includes the percent Medicare and Medicaid discharges as well as the percent of inpatient and outpatient admissions that were surgical admissions; Hospitalj includes dummies indicating non-profit or local government ownership, teaching status, and system membership, plus a hospital fixed effect, θ.

All of these variables are from the 1994-2001 American Hospital Association’s (AHA) Annual Survey . The survey includes all short-term general hospitals located within an MSA. We perform the simulations on the subset of hospitals located in the Tampa, Phoenix, and Tucson markets as defined above. We assume that hospitals are operating off of the same cost function, allowing for shifts accounted for by ownership, patient mix, and hospital fixed effects. Thus, any cost changes resulting from the simulations are principally a result of scale. Prior to running the simulations wWe adjust the parameter estimates ofpredictions from the cost function for heteroskedastic smearing in order to allow for heteroskedasticity across markets (Manning, 1998). The smearing estimates inflate the estimate of change in markets costs by about 50-154%.

6. Results

Table 21 displays the effect of hospital closure on patient welfare from the fixed effects specification. The first column lists the hospital and its location. The second column reports the drive time equivalents, in hours, of the effect of closure on the patient’s utility. The next four columns map these time changes into dollars under various scenarios. On the strong assumption that most patients choose their hospital with a round trip in mind, we believe the numbers in columns (c) and (d) are more credible.[9] we calculate the utility loss under the assumption that one person or two people make a roundtrip each day. The declines in patient surplus are substantially greater in the Arizona markets than in Tampa, which is likely due to the location of neighboring hospitals. Both Tampa hospitals are located on the peninsula west of Tampa, which has a high concentration of both hospitals and patients. Thus after these two hospitals exit, there is still a number of suitable substitutes available locally. This lead to relatively low welfare effects of closure. In Tucson and Phoenix …..

Table 3 evaluates the change in market costs due to a hospital closure. The second column presents in the ex-post increase in costs at neighboring hospitals from the closed hospital’s patients. As expected, the change in market costs is roughly proportional to the number of admissions at the closed hospital. The third column presents the approximate ex-ante cost of treating those patients at the closed hospital.. This reflects the difference in the total costs at closed hospital less the costs if the patients in the simulation were not treated at the closed hospital. Thus we keep constant outpatient visits and wages when performing these simulations. The costs are only approximate because only total hospital costs are available in the AHA data, whereas we are interested in inpatient costs for general acute care. Thus, we first multiply total hospitals costs by the proportion of inpatient revenue reported in the Medicare Cost reports. This proportion varies from 0.85 to 0.50. Next we multiply this product by the proportion of inpatient visits that are for general hospital services. The result is an admittedly rough measure of the cost of treating inpatients for general hospital services at the closed hospital, but it is the best available from the data. The difference between the second and third columns is reported in the fourth column. We find that each closure led to a decline in market costs.

Our estimates of the total welfare effect are reported in Table 4. Based on results to date, three two of the five closures increased total welfare – the market-wide cost savings outweighed patients’ aggregate lost utility. A fourth closure, involving Community Hospital Medical Center, may have also increased surplus, though when round trips are taken into account and travel time is valued at roughly $10.00 or more, the utility loss outweighs the cost savings. Finally, the closure of Tucson General Hospital apparently reduced total welfare. While the closure did lead to significant cost savings, the utility cost of this closure is particularly high. Referring to the map of Tucson in the appendix, this is not particularly surprising: Tucson General does not have any competitors within two miles in any direction, and none to the northeast for more than ten miles. The map of Phoenix shows that Community Hospital Medical Center, whose closure reduced welfare for time-valuations over $10.00, is also relatively isolated.

NOTE: I’ll report on Arizona in the seminar!

7. Alternative Specifications

[NOTE: At this stage, and pending the alternatives described below, we are not yet confident in the robustness of the above result—that most closures are good.] RICH: I THINK MOST OF THE FOLLOWING SHOULD BE DELETED. WE CAN ADD IT BACK IN WHEN WE HAVE MORE RESULTS FROM NON-FIXED EFFECTS SPECIFICATIONS

There is, however, one significant ambiguity in the above results: we consistently over-estimate the demand for each of the closing hospitals. That is, the predicted market patient shares based on the estimates from the choice model are much higher than the closing hospitals’ actual shares. This indicates that we may be overestimating the utility losses caused by these closures.[10] Over-predicting the demand at the closing hospitals biases the welfare calculation in favor of finding that a closure would decrease total surplus. So the qualitative result that at least 3 of the 5 closures increased welfare is probably unaffected; whether further exploration of this issue will reverse the conclusion for Tucson General is an open question.

There are several possible explanations for this over-prediction. First, because we include a fairly sparse set of hospital-level variables, we may not be fully capturing how unattractive the closing hospitals actually are. For example, the baseline model does not include characteristics such as the attractiveness of the hospital’s neighborhood. One natural way to explore the over-prediction problem is to include community-level socio-economic characteristics such as local crime rate, housing vacancy rate, poverty rate, and median income. This will substantially increase the computational burden, but the simulations are in progress. A second approach we are pursuing is to use hospital-level fixed effects, which will almost surely generate predicted shares that are close to the observed shares.[11]

On the other hand, over-predicting demand may be natural in the context of closing hospitals. For example, these hospitals may be closing because they are poorly managed or have suffered some random negative shocks. If so, then our over-estimated demand reveals the surplus that hospitals in the existing locations with the existing services should generate, if the negative shocks or bad management were removed. The distinction between this and the preceding explanation is important. Consider Tucson General Hospital, whose closure was found to reduce surplus – i.e., a bailout of this hospital would be a good use of tax money. If we are over-estimating demand because the undesirable aspects of the closing hospital are not fully captured then a bailout would actually not be justified. On the other hand, if we are accurately estimating what demand would be absent unique negative conditions, then a bail-out would be justified, conditional on identifying and removing the cause of the unique negative circumstances.

While there is no formal test for which of these explanations is true, the results from the alternative model with additional hospital-level variables should shed light on the question. If adding these controls nearly or completely eliminates the over-estimation problem, then the first explanation is likely true. If, however, the over-estimation persists, then that strongly suggests that a hospital with similar characteristics to the closed hospital should achieve the predicted volumes. In such a case, a bailout could increase efficiency.

[Results Pending]

78. Conclusions

We combine cost and demand estimates to evaluate the impact of hospital closures on economic efficiency. Based on results from threethereof two closure in a mid-size urban markets in the late 1990s, we tentatively find that most hospital closures increase total surplus. The implication is that municipal policy-makers should, absent of unique circumstances, resist the pressure for a bailout that often accompanies a closure announcement. This prescription is strengthened by the observation that our current results appear likely to overestimate the utility generated by the closing hospitals.

However, in each of the markets we study occupancy rates varied from 59% to 64%, well below full occupancy which is between 80-90%. Thus there is plenty of capacity at surrounding hospital to absorb the additional patients. The fact that there are low occupancy rates in these markets is not surprising since closure is most likely to occur in markets that have excess capacity. Furthermore, we have not considered changes in hospital prices that might be a result of closure. We do not think this is an issue in the markets and hospitals we study because they are relatively competitive markets and the hospitals that close are small.

NOTE: Conclusions may change when the Arizona results are in!However, this conclusion is only tentative, and it does appear that bailouts may be warranted when the closing hospital has no nearby substitute. More generally, the model’s over-estimation of demand at the closing hospitals is cause for concern. To address this, we are implementing two additional models: one that expands the set of hospital-level variables to include the socio-economic characteristics of the area hospital’s vicinity, and one that instead uses hospital fixed effects.

References

Bazzoli, G. J. and Andes, S., 1995. “Consequences of hospital financial distress.” Hospital and Health Services Administration 40, 472 - 495.

Brownstone, D., Ghosh, A., Golob, T., Kazimi, C., and Van Amelsfort, D., “Drivers’ willingness-to-pay to reduce travel time: evidence from the San Diego I-15 congestion pricing project.” Transportation Research Part A: Policy and Practice, 37(4), May 2003, Pages 373-387.

Capps, C., D. Dranove, and M. Satterthwaite. 2004. “Competition and Market Power in Option Demand Markets.” Rand Journal of Economics.

Ciliberto, F. and Lindrooth, R.C., 2003. “Exit from the hospital industry.” Working Paper.

Deily, M.E., McKay, N.L., and Dorner, F.H., 2000. “Exit and inefficiency.” Journal of Human Resources 35, 734–747.

Cogswell, S., “A Comparison of Managed Care Organizations in Connecticut,” State of Connecticut Insurance Department, 2002.

Duan, N., 1983. “Smearing estimate: A nonparametric transformation method.” Journal of the American Statistical Association 78, 605–610.

Duffy, S. and Friedman, B., 1993. “Hospitals with chronic financial losses: What came next.” Health Affairs 12, 151–63.

Gaynor, M. and Anderson, G. 1995. “Uncertain demand, the structure of hospital costs, and the cost of empty hospital beds.” Journal of Health Economics 14, 291–317.

Ghemawat, P. and Nalebuff, B., 1990. “The devolution of declining industries.” Quarterly Journal of Economics 105, 167–186.

Gómez-Ibáñez, J., Tye, B., and Winston, C., editors, Essays in Transportation Economics and Policy: A Handbook in Honor of John R. Meyer. Brookings Institution Press, Washington D.C., 1999.

Keeler, T.E. and Ying, J.S., 1996. “Hospital costs and excess bed capacity: A statistical analysis.” Review of Economics and Statistics 78, 470–481.

Lindrooth, R., A. Lo Sasso, and G. Bazzoli, 2003. “The Effect of Urban Hospital Closure on Markets.” Journal of Health Economics, 22(5), pp. 691-712.

Manning, W. G., 1998. “The logged dependent variable, heteroscedasticity, and the retransformation problem.” Journal of Health Economics, 17, 283-95.

McNamara, P.E., 1999. “Welfare effects of rural hospital closures: A nested logit analysis of the demand for rural hospital services. American Journal of Agricultural Economics 81, 686–691.

Mobley, L.R. and H.E. Frech. 1994. Firm growth and failure in increasingly competitive markets: Theory and application to hospital markets.” Journal of the Economics of Business 1(1), 77–93.

Pauly, M and Wilson, P. 1986. “Hospital output forecasts and the cost of empty hospital beds.” Health Services Research 21, 403–428.

Small, K., Urban Transportation Economics.Vol. 57 of Fundamentals of Pure and Applied Science. Harwood Press, Switzerland: 1992.

Small, K., Winston, C., and Yan, J., "Uncovering the Distribution of Motorists' Preferences for Travel Time and Reliability: Implications for Road Pricing.” Manuscript, University of California at Irvine, March 2003.

Succi, M.J., Lee, S.D., and Alexander, J.A. 1997. “Effects of market position and competition on rural hospital closures.” Health Services Research 31, 679–699.

Wedig, G. J., Hassan, M. and Sloan, F.A. 1989. “Hospital investment decisions and the cost of capital.” Journal of Business 62, 517–37.

Williams, D.J., Hadley, J., and Pettengill, J., 1992. “Profits, community role, and hospital closure: An urban and rural analysis.” Medical Care 30, 174–187.

|Table 1. Characteristics of Closed Hospitals |

|Hospital Closure |Year |Market Share (AHA Annual Survey) |

|Phoenix Regional Med. Ctr., Phoenix, AZ |1998 |1.9% |

|Community Hospital Med. Ctr., Phoenix, AZ |1998 |0.7% |

|Clearwater Community Hosp., Clearwater, FL |1998 |1.6% |

|University General Hosp., Seminole, FL |1996 |1.3% |

|Tuscon General Hosp., Tuscon, AZ |1999 |3.0% |

|Table 22. The Effect of Closure on Patients |  |  |  |  |  |

|Hospital Closure |Hour-Equivalent of Utility Cost|One way person |RoundtripTwo people |

| | |(a) |(b) |(c) |(d) |

| | |$1220/hour |$716/hour |$1220/hour |$716/hour |

|Phoenix Regional Med. Ctr., Phoenix, AZ |1.00376,306.43 |40.00$4,515,677 |32.00$2,634,145 |$80$9,031,354 |$64$5,268,290 |

|Community Hospital Med. Ctr., Phoenix, AZ |1.00356,702.14 |40.00$4,280,426 |32.00$2,496,915 |$80$8,560,851 |$64$4,993,830 |

|Clearwater Community Hosp., Clearwater, FL |1,776.36128,835.37 |$71,054$1,546,024 |$56,844$901,848 |$142,109$3,092,049 |$113,687$1,803,695 |

|University General Hosp., Seminole, FL |858.14163,755.18 |$34,326$1,965,062 |$27,460$1,146,286 |$68,651$3,930,124 |$54,921$2,292,573 |

|Tuscon General Hosp., Tuscon, AZ |1.00590,101.72 |40.00$7,081,221 |32.00$4,130,712 |$80$14,162,441 |$64$8,261,424 |

|Table 3. Estimated Cost Changes Due to ClosureTable 3. Estimated Cost Changes Due to Closure |

|Hospital ClosureHospital Closure |Change in Expenditures at |Equivalent Inpatient |Change in Market Expenditures Change|

| |other hospitalsChange in |Expenditures at Closed |in Market Expenditures ($1000s) |

| |Expenditures at other |HospitalEquivalent Inpatient | |

| |hospitals ($1000s) |Expenditures at Closed Hospital | |

| | |($1000s) | |

|Phoenix Regional Medical Center, Phoenix, AZPhoenix Regional Med. Ctr., |$7,127 |$19,210 |-$12,083 |

|Phoenix, AZ | | | |

|Community Hospital Medical Center, Phoenix, AZCommunity Hospital Med. Ctr., |$4,562 |$10,774 |-$6,212 |

|Phoenix, AZ | | | |

|Clearwater Community Hospital, Clearwater, FLClearwater Community Hosp., |$8,195,533$10,104 |$13,603,817$18,459 |-$5,408,285-$8,355 |

|Clearwater, FL | | | |

|University General Hospital, Seminole, FLUniversity General Hosp., Seminole, |$4,249,418$11,689 |$5,978,459$15,720 |-$1,729,042-$4,031 |

|FL | | | |

|Tuscon General Hospital, Tuscon, AZTuscon General Hosp., Tuscon, AZ |$4,264,269$4,219 |$7,445,210$9,847 |-$3,180,942-$5,629 |

|*Based on 1997 data | | | |

|Table 4. Welfare Effect of Closure |

|Hospital Closure |One way |Roundtrip |

| |$20/hour$12/hour |$16/hour$7/hour |$20/hour$12/hour |$16/hour$7/hour |

|Phoenix Regional Medical Center, Phoenix, AZ |$7,567,396 |$9,448,928 |$3,051,719 |$6,814,783 |

|Community Hospital Medical Center, Phoenix, AZ |$1,931,764 |$3,715,275 |-$2,348,661 |$1,218,360 |

|Clearwater Community Hospital, Clearwater, FL |$5,337,230$6,808,906 |$5,351,441$7,453,082 |$5,266,176$5,262,881 |$5,294,597$6,551,235 |

|University General Hospital, Seminole, FL |$1,694,716$2,065,798 |$1,701,581$2,884,574 |$1,660,390$100,736 |$1,674,121$1,738,287 |

|Tuscon General Hospital, Tuscon, AZ |-$1,452,591 |$1,497,918 |-$8,533,811 |-$2,632,794 |

Appendix

|Table A1: Choice Model Variables |

| | |Hospital Characteristics |

| | |Time |For Profit |Non-Profit |Teaching |Trans-plants |Nurses/ |

| | | | | | | |Patient Day |

| |

| |Income in Zip |X |X |X |X |X |X |

|Patient | | | | | | | |

|Characteristics | | | | | | | |

| |Male |X |X |X |X |X |X |

| |Age: 0-5 |X |X |X |X |X |X |

| |Age: 6-18 |X |X |X |X |X |X |

| |Age: 41-64 |X |X |X |X |X |X |

| |Age: 65_75 |X |X |X |X |X |X |

| |Age: >75 |X |X |X |X |X |X |

| |Black |X |X |X |X |X |X |

| |Hisp. |X |X |X |X |X |X |

| |Other |X |X |X |X |X |X |

| |#Procs |X |X |X |X |X |X |

| |#Diags |X |X |X |X |X |X |

| |%Travel |X |X |X |X |X |X |

| |Time |-- |X |X |X |X |X |

| |1 |X |X |X |X |X |X |

| |

|Service Match |Oncology |All interactions of Patient Characteristics and Hospital Characteristics are included in the Choice Model, |

|Indicators | |as indicated by “X”. |

| | |Hospital Fixed Effects are Interacted with time |

| | |Time and Hospital Characteristics are included as stand-alone independent variables (i.e., interacted with |

| | |“1”). |

| | |Service Match Indicators equal one when a hospital offers the service that “matches” the patient’s clinical |

| | |needs. |

| |HIV | |

| |Delivery | |

| |Circulatory | |

| |Transplant | |

| |Pediatric | |

Phoenix Closures

(Red circles denote closing hospitals)

[pic]

Note: Not all hospitals in the Phoenix market are shown.

Tucson Closures

(Red circles denotes closing hospital)

[pic]

Note: Not all hospitals in the Tucson market are shown.

Tampa-St. Petersberg Closures

(Red circles denote closing hospitals)

[pic]

|Market |Capacity Utilization |Capacity Utilization |

| |Before Closure |After Closure |

|Phoenix |62.2% |65.0% |

|Tucson |63.3% |66.4% |

|Tampa 1996 |59.4% |59.2% |

|Tampa 1998 |61.3% |57.2% |

-----------------------

[1] This might occur because the nonprofit is willing to sacrifice expected profits to sustain patient welfare, or, in a competitive market, because the for-profit does not believe it can outlast a nonprofit in a war of attrition (Ghemawat and Nalebluff, 1990).

[2] As discussed above, ((.) is not identified because relatively few patients face meaningful or observable cross-hospital variation in their out-of-pocket expense.

[3] As we ignore the cost side of outpatient care, we also ignore travel for outpatient care. If a closure forces patients to travel considerable distances for outpatient care, this could be an important consideration.

[4] IN OUR ONGOING REVISIONS, WE ARE EXPERIMENTING WITH OTHER SPECIFICATIONS OF DRIVING TIME.

[5] To our knowledge there is only one direct study of this issue, a study of Connecticut hospital networks (Cogswell, 2002), and it supports our belief. Unlike most other states, Connecticut provides highly detailed data on hospital participation in managed care networks. Cogswell reports that, of the 9 HMOs in the state; 8 contract with at least 87% of the state’s hospitals and 5 contract with at least 93%. Of the 15 PPOs, 13 contracted with 87% of the hospitals, and 5 contracted with 93%. Thus, at least in Connecticut, almost all networks consist of very nearly all hospitals. We have no reason to believe that Arizona and Florida networks are more restrictive.

[6] For a methodological overview of the techniques used in the transportation literature, see Gómez-Ibáñez, Tye, and Winston (1999), particularly Chapter 2; also see Small (1992).

[7] This $126 is what this hypothetical patient, looking forward and evaluating possible health states (including the possibility of not needing hospitalization), would pay to retain the option of receiving care at the closing hospital. That is, it is the ex-ante value of having the hospital available to provide care. Should the patient in fact draw a condition for which he prefers the closing hospital, then his willingness to pay would of course be much higher.

[8] We only integrate ofover[pic]%ABbd†ˆ«­Ö×Øyz{|¢£¤µ¶ÄÅÆÍÒÙäçÿôíçôíôíôíôíÏÇù¯ ¯• † ¯‚ÃoeÃ`ÃVhkD”hIï0Ja[9]?j[pic]hIïU[pic][pic]?jH[pic]hƒÌ”&hIïU[pic]

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download