The Interactive Effect of Medicare Inpatient and Outpatient Reimbursement

Transcription

1 The Interactive Effect of Medicare Inpatient and Outpatient Reimbursement JOB MARKET PAPER Andrew Elzinga November 12, 2015 Abstract Hospital care is characterized by inpatient and outpatient departments; however, Medicare reimburses hospitals differently for treatment in each department. In this paper, I examine how hospitals respond to Medicare s disjointed payment system. I exploit the conversion of rural hospitals to Medicare s Critical Access Hospital (CAH) program, which implemented a universal payment system. Although the new structure increased marginal reimbursements, hospitals responded by decreasing treatment intensity. I propose that this decline is explained by high initial levels of outpatient procedures, used to offset losses from inpatient care, and the elimination of inpatient losses through the CAH program, removing the need to maintain high outpatient procedure volumes. Additionally, the decline in treatment is not associated with reductions in health outcomes. My findings suggest Medicare s disjointed payment system created an interactive effect across departments where the financial status of one department impacted behavior in the other. In general, these results highlight the importance of designing and analyzing government programs in a unified approach in order to account for interactions across policies. Brown University, Department of Economics. andrew elzinga@brown.edu. Special thanks to my advisors, Emily Oster, Anna Aizer, and Brian Knight for their insight and dedication; and for their helpful comments I would like to thank Joseph Kofi Acquah, Desislava Byanova, Kenneth Chay, Alex Eble, Morgan Frost, Bruno Gasperini, Philipp Ketz, Kanghyock Koh, Michelle Marcus, and seminar participants at Brown University. All remaining errors are my own. 1

2 1 Introduction Hospital care is characterized by inpatient and outpatient departments; both of which comprise an important share of hospital s total revenue. 1 However, Medicare reimburses hospitals differently depending on where services are provided. Payments for inpatient care are fixed for each discharge, determined by the patient s diagnosis, not the hospital s cost. Outpatient services are reimbursed under a cost-based, fee-for-service structure. 2 Based on existing literature on hospital response to incentives (Duggan, 2000; Dafny, 2005), a disjointed payment system may induce strategic responses by hospitals. This paper examines if and how hospitals respond to this disjointed reimbursement structure. To test for hospital response in this context, I study Medicare s Critical Access Hospital (CAH) program. The CAH program allowed qualifying rural hospitals to receive Medicare reimbursements based on incurred costs for inpatient and outpatient services. This change created a universal payment structure, eliminating the separation between inpatient and outpatient care. As a result, conversion to CAH status changed the payment structure of inpatient care, while maintaining that of outpatient services. Inpatient treatments now received marginal reimbursements equal to treatment costs, opposed to the zero marginal reimbursement under the previous system. Considered in isolation, the increasing marginal reimbursement was expected to increase inpatient treatment intensity (Ellis and McGuire, 1996). I exploit the staggered conversion of CAHs to estimate the impact of the changing reimbursement policy using an event study design. The analysis focuses on the effect of Medicare s payment change on treatment intensity, as measured by the number of procedures per discharge. Contrary to expectation, inpatient data suggest the increased marginal reimbursements led to a decline in treatment intensity for Medicare beneficiaries. In aggregate, I find CAH conversion is associated with a 5-6% reduction in procedures per discharge for Medicare beneficiaries. Estimates of the effect of CAH conversion on the non-medicare population also find small declines in procedures, suggesting potential spillover effects (Baicker et al., 2013). Furthermore, the declining treatment intensity does not appear to be driven 1 In 2013, 45% of total revenues came from outpatient services and 55% from inpatient care (AHA, 2015). 2 Historically, Medicare reimbursed all hospitals under this payment system. In August 2000, Medicare introduced a new structure for outpatient payments, however, small rural hospitals were exempted and effectively continued to be paid on the basis of costs. 2

3 by selection on the part of patients or hospitals and is not observed at non-cahs. In order to understand the declining treatment intensity, I document three empirical facts. First, the decline in procedures was driven primarily by inpatients admitted through the emergency department; opposed to routine admissions. Second, hospitals with a larger baseline Medicare patient share experienced larger declines. Finally, the procedures being reduced were those performed on the day of admission or prior and tended to be characterized by high fixed costs. Reconciling these results, I propose that the decline in procedures is the result of an interactive effect between inpatient and outpatient departments created by Medicare s previously disjointed payment system. Consistent with literature on target-earning and hospital costshifting, 3 the financial status of one department impacted behavior in both departments. I argue this interaction between departments caused the decline in procedures per discharge. Prior to CAH conversion, hospitals provided high levels of profitable outpatient treatments to offset losses from inpatient services. Following conversion, Medicare s inpatient losses were eliminated due to the change to cost based reimbursement, as was the need to sustain high procedure volumes; leading to the observed decline. Overall, due to Medicare s multidimensional payment structure, rural hospitals responded to inpatient losses by driving up outpatient revenues. I show the decline in procedures is not associated with changes in health outcomes for Medicare beneficiaries, which suggests the procedures being eliminated were not affecting patient health. Therefore, although I find the CAH program led to higher Medicare expenditures, the universal payment structure appears to have reduced an inefficiency. Through higher payment rates, Medicare effectively paid rural hospitals to eliminate procedures performed for financial gain. In a more general context, this implies that when government programs are designed in isolation, they can create incentives for people to respond to one program by changing their behavior in another. This behavior is not limited to healthcare. For example, Borghans et al. (2014) finds that cuts in disability insurance benefits are met by a substitution towards other social programs. In general, it is important to design and analyze government programs in 3 Target-earning is the idea that healthcare providers may adopt tactics to maintain their income in response to price changes (Rice and Labelle, 1989). Cost-shifting is a situation where hospitals respond to decreases in the price paid by one insurer by raising prices to patients insured under other plans (Frakt, 2011). This is often viewed as hospitals charging privately insured patients more in response to shortfalls in publicly insured patients (e.g., Medicare and Medicaid). 3

4 a unified approach in order to account for interactions across policies. This paper adds to the existing literature on how hospitals respond to incentives by analyzing the effects of disjointed payment structures. Previous research has focused on how healthcare providers respond to payment changes in inpatient (Cutler, 1995; Ellis and McGuire, 1996; Kim, 2011) and outpatient settings (He and Mellor, 2012; Clemens and Gottlieb, 2014). However, the main focus of my study is the overall response of hospitals, rather than the effect of payment changes on a single department. Two notable exceptions to this are Friesner and Rosenman (2004) and Carey (1994) which document the potential for cost shifting between inpatient and outpatient departments. 4 To the best of my knowledge, no paper has studied how hospitals respond to Medicare s disjointed payment structure. The remainder of the paper is organized as follows. Section 2 provides a background of the CAH program and a more complete picture of Medicare s existing payment structure. Section 3 describes the data, while section 4 details the empirical strategy used to estimate the impact of CAH conversion. In section 5, I present the aggregate effects of CAH conversion and then interpret the results in section 6. I discuss welfare effects of the CAH program in section 7 and then conclude. 2 Background 2.1 Critical Access Hospitals Rural hospitals face a number of challenges not experienced by hospitals located in urban areas. From differences in the composition of their patients and Medicare reimbursement rates to significantly lower patient volumes, the financial viability of rural hospitals has long been an issue plaguing the availability of rural healthcare (Moscovice and Stensland, 2002). Recognizing that many of the smallest rural hospitals struggled to recover the incurred inpatient costs of treating Medicare patients, policy makers have created a complex mix of special payment classifications. One of the most important of these hospital classifications is Critical Access Hospital (CAH). The CAH designation was created by the Balanced Budget 4 Friesner and Rosenman (2004) studies dynamic cost shifting between inpatient and outpatient departments. In particular, if hospitals respond to lower government reimbursements for inpatient services by raising outpatient prices. Additionally, Carey (1994) documents the potential for a selection effect, where hospitals shift costs between departments. 4

5 Act of 1997 and was designed to help rural hospitals financially by changing the way Medicare reimbursed them. Hospitals can obtain CAH status by satisfying the following criteria: 5 1. Nonprofit or public hospital 2. Located in a rural area 3. Located more than 35 miles from all other hospitals or be designated as a necessary provider of health care services to residents in the area by the state 4. Provide 24-hour emergency care services 5. Fewer than 25 inpatient beds 6. Inpatient care does not exceed 96 hours In order to examine the relevance of this classification, Figure 1 plots the number of hospitals by classification. The figure shows that the number of CAHs increased quickly following its creation, particularly between 2000 and 2006, while the number of general hospitals decreased. Meanwhile, the total number of providers has remained fairly constant, showing that the increase in CAHs is driven by the conversion of general hospitals rather than the opening of new hospitals. It is important to note, new CAHs are not the building of new hospitals, but rather the conversion of existing hospitals. Regarding Medicare reimbursements, prior to CAH conversion, these small, rural hospitals were reimbursed under the same national system as all hospitals. However, once a hospital converted to a CAH, all of their inpatient and outpatient services were changed to cost based reimbursements. In order to understand the effects of this change, it is important to understand the payment structures that were previously in place. 2.2 Medicare s Inpatient Reimbursement Since its implementation in 1983, Medicare s Inpatient Prospective Payment System (IPPS) has reimbursed hospitals on a per-discharge basis using a nationally-fixed formula. 6 Independent of their geographic location, all general service hospitals are subject to this payment system; including CAHs prior to their conversion. 7 5 The Balanced Budget Refinement Act (BBRA) of 1999 allowed for the inclusion of for-profit hospitals and changed the maximum length of care to an average of 96 hours. The initial bed size cap was 15 until the Medicare Modernization Act (MMA) of 2003 increased it to 25 beds. The MMA also removed a state s ability to waive the distance requirement starting in The Centers for Medicare & Medicaid Services uses separate payment schemes for reimbursement to acute inpatient hospitals, home health agencies, hospice, hospital outpatient, inpatient psychiatric facilities, inpatient rehabilitation facilities, long-term care hospitals, and skilled nursing facilities. 7 Within the Inpatient Prospective Payment System (IPPS) system there are three classifications aimed at helping rural hospitals: sole community hospitals, Medicare dependent hospitals, and rural referral cen- 5

6 Under IPPS, the amount Medicare pays a hospitals for their services is determined by categorizing the patient s diagnosis into one of 751 diagnosis related groups (DRG). Once the patient has been placed into a specific DRG, hospitals receive a fixed amount, independent of actual costs. The payment rates are meant to represent the national average cost of treating a patient in their specified DRG. A simplified version of the payment formula is as follows: P ayment dh = ST D DRG d (1 + Adjustments h ) where d represents the DRG and h the hospital. ST D is the dollar value where all payments begin (approximately $5,000 in 2014), DRG d is the weight for each DRG, 8 and Adjustments h are hospital specific payment increases. 9 Importantly, this payment scheme does not depend on actual costs, implying that the marginal reimbursement for each treatment provided is effectively zero. Within the IPPS payment structure, rural hospitals were susceptible to negative payment margins. This was due to declining patient volumes and the resulting diseconomies of scale, high percentages of subsidized and uninsured patients, disproportionately elderly populations, and physician shortages (Geyman et al., 2001; Eberhardt et al., 2001). Financial struggles of rural hospitals within this payment system were realized through increases in the closure rates of rural hospitals in the early 1990s. The spike in the closure rate of rural hospitals and the resulting decline in the availability of rural hospital care was the driving force for the creation of the CAH program. However, conversion to a CAH removed the hospital from the IPPS system and allowed the hospital to receive payments based on the realized treatment costs of each Medicare inpatient discharge. 10 ters. Each classification has its own designation criteria and upon classification a hospital receives adjusted/increased Medicare payments. However, these classifications only exist within IPPS and are ignored if the hospitals becomes a Critical Access Hospital (CAH). 8 The DRG weights are updated annually and range from roughly 0.2 to 21. The weight represents the average cost of treating a patient in DRG d relative to the average cost among all Medicare patients. 9 These increases are made to qualifying hospitals for indirect medical education costs and/or treating a disproportionate share of indigent patients. Rural hospitals are much less likely to qualify for either of these adjustments. 10 The determination of incurred costs is based on annual cost reports submitted by hospitals. Rather than receiving a single payment for reimbursement of reported annual cost, Medicare makes interim payments throughout the year. These payments are determined using the previous year s cost report. Following a Medicare beneficiary s discharge, the hospital submits a claim detailing the patients care and stay. Medicare then makes an payment to the hospital based on the average Medicare per-diem calculated from the previous year s cost report. At the end of the year, once the current year s cost report has been submitted, Medicare 6

7 Overall, the change from IPPS to cost based reimbursement, dramatically changed the financial status of the hospitals inpatient department. Furthermore, it changed the treatment incentives for the hospitals. Prior to conversion, the hospitals were reimbursed below costs, and had little incentive to perform marginal treatments due to the zero marginal reimbursement of IPPS. Upon obtaining CAH status, the hospitals were compensated at the cost of their services and, had an incentive to increase treatment intensity because all services faced a marginal reimbursement equal to their cost. In isolation, the increase in the marginal reimbursement of inpatient treatment, was expected to increase treatment intensities (Ellis and McGuire, 1996). 2.3 Medicare s Outpatient Reimbursement Historically, Medicare reimbursed outpatient services provided at all hospitals (mostly) based on reasonable costs (He and Mellor, 2012). Reasonable costs are defined as direct and indirect costs associated with providing services to Medicare beneficiaries (42 CFR 413.9, b). 11 Payments are made on a fee-for-service basis, meaning each time a procedure is performed Medicare reimburses the hospital the reasonable cost of the treatment. Importantly, reasonable costs include a vast amount of overhead and are the same amount each time the procedure is performed. As a result, procedures that were performed enough times and had low marginal costs were actually profitable. Under this system, Medicare payments rose sharply. Between 1983 and 1997, the annual growth rate of Medicare s outpatient expenditure was 12% (MedPAC, 1999). In August 2000, Medicare replaced the cost based reimbursement system with the Outpatient Prospective Payment System (OPPS) for outpatient payments. OPPS has a fee-for-service payment structure that reimburses hospitals a fixed, predetermined amount each time a service is and the hospitals settle any difference between the amount already paid and the actual cost for the year. 11 Outpatient interim payments are determined using hospital s cost report. Each hospital calculates a cost-to-charge ratio (CCR), defined as the ratio of total costs to total charges for Medicare beneficiaries. Every time a Medicare beneficiary receives outpatient treatment, the hospital submits a claim to Medicare documenting all treatments and the charges. To calculate the amount Medicare will pay for the claim, Medicare uses the hospital s CCR. The amount the estimated costs exceed the 20% coinsurance paid by the beneficiary, is then reimbursed by Medicare. In addition to calculating a CCR, Medicare also calculates a payment-to-cost ratio (PCR). The key difference being that the PCR takes into account the coinsurance and deductibles paid by beneficiaries. Formally, the payment-to-cost ratio is calculated as P CR = charges CCR (coinsurance+deductibles) charges CCR. 7

8 performed. For example, in 2012 each time a hospital performed a computed tomography (CT) scan with contrast, Medicare reimbursed the hospital $ However, Medicare was concerned that the transition from cost based payments to OPPS would lead to significant financial shocks for certain hospitals, in particular rural hospitals with fewer than 100 beds. To protect these hospitals, OPPS included transitional adjustments to limit the decline in payments for small, rural hospitals (He and Mellor, 2012). Specifically, Medicare paid them the maximum of the amount determined by the OPPS fee schedule and the amount that would have been paid under reasonable costs. 12 As a result, the overall impact of converting to a CAH, had minimal impact on the payment structure of hospital s outpatient departments. 3 Data Due to the rural location and small size of CAHs, most publicly available datasets do not provide enough observations to conduct any analysis. For example, to the best of my knowledge, all survey data of hospital visits (both inpatient and outpatient) are primarily comprised of individuals living in urban areas. As a result, this analysis required administrative level data in order detect potential changes brought about by the conversion to a universal payment system. My focus was restricted to data that would allow me to construct a panel of hospitals, in order to partial out unobservable, fixed characteristics of hospitals. Finally, due to the timing of CAH conversion, I am limited to datasets that date back to the late 1990s and early 2000s. 3.1 State Inpatient Databases (SID) The State Inpatient Databases (SID) are state-year databases that are part of the family of databases developed by the Healthcare Cost and Utilization Project (HCUP). The SID contain hospital inpatient stays from all community hospitals and include the universe of inpatient discharges in participating states. The unit of observation is the discharge level 12 Technically, the transitional adjustments required hospitals to calculate reasonable costs using the 1996 payment-to-cost ratio for all years. Thus, the reimbursable amount for services provided in 2001 would equal: Reimbursement 2001 = charges 2001 CCR 2001 P CR However, the difference between hospital s 1996 PCR and the future year s PCR shows little variation suggesting this requirement had little impact. 8

9 and includes variables such as diagnoses and treatment, demographics, and insurance status. The benefit of the SID data is it contains detailed information on how the patient was treated throughout their entire hospital stay. In particular, among admissions that originate in the emergency department, the data also contains information about their treatment prior to being formally admitted to the hospital (e.g. in the emergency department). The analysis that follows is based on SIDs from five states with varying years of observation: Arizona ( ), California ( ), Florida ( ), New York ( ), and Washington ( ). The data allows me to link hospitals across years, creating a panel of hospitals over time. A total of 107 hospitals converted to CAH status from these five states, however, to prevent changes in the composition of hospitals I restrict my sample to hospitals observed at least 24 months before and 24 months after conversion. Following this data restriction, there are a remaining 84 hospitals that have converted to CAH status, which constitute as my main sample. To simplify the complexity of the Internal Classification of Diseases (ICD) codes for diagnosis and procedures commonly used in the medical literature, the analysis will utilize the Clinical Classifications Software (CCS) for services and procedures developed by HCUP. This alternative procedure grouping does not change the reported number of procedures, but rather places them into more aggregate categories. Specifically for procedures, the CCS codes constitute a mapping from approximately 3,900 ICD codes into 231 groups (e.g. CABG, cesarean section, hip replacement, blood transfusion, etc.). The primary limitation of the data is that it does not include patients that visit the hospital in an outpatient setting and are never admitted to the hospital (e.g. emergency department patients who are treated and then released). Therefore, analyses of outpatient behavior will be limited to inpatients admitted through the emergency department. Another limitation is that only significant procedures are included in the discharge records. For example, 70% of the Medicare patients in my sample, whose primary diagnosis is pneumonia receive zero procedures. Table 1 presents the summary statistics of my data. Overall, CAHs tend to have fewer inpatient discharges per year relative to non-cahs. Additionally, the patient composition at CAHs tends to be comprised of more Medicare patients and fewer privately insured patients, relative to non-cahs. My data further shows, treatment intensities tend to be lower at 9

10 CAHs, as measured by the average number of procedures and total charges. 3.2 Medicare Cost Reports The Centers for Medicare and Medicaid Services require all hospitals receiving federal reimbursement to complete annual cost reports. These cost reports are a series of forms that collect descriptive, financial, and statistical data on hospitals. Specifically, they contain information including hospital characteristics, discharge data, total patient charges and costs broken down by department, and Medicare reimbursements. Veterans hospitals, Indian Health Services hospitals, and some children s hospitals are exempt from completing said forms. The principal reason for using the Medicare cost reports is that they include data on how much each hospital actually received from Medicare for both inpatient and outpatient services; a variable that is omitted from the SIDs. A limitation of the cost reports is that the reports do not provide payments on a patient level, rather the data are aggregated to the fiscal year and only include annual totals. Table 2 presents a summary of the Medicare cost report data. As anticipated, CAHs are smaller hospitals and receive lower total Medicare payments. Additionally, CAHs are more likely to be publicly owned and are almost never privately owned. Finally, the CAHs observed in the SID data appear to be very similar to the universe of CAHs Empirical Strategy To estimate the effect of the change to universal cost based reimbursement, I employ an event study model (MacKinlay, 1997). This model utilizes the variation in CAH conversion date to improve the identification strategy as it rules out events that occurred at a single point in time. In particular, my estimates are based on the following framework: y iht = α + β CAH ht + γx iht + θ t + η h + δ t + ε ijt (1) 13 The statistics are calculated using years before and after CAH conversion and therefore the average bed size for CAHs is above the maximum 25 because many hospitals reduced their bed count in order to obtain CAH status. 10

11 where y ijt is the outcome variable for patient i at hospital h in year t. CAH is an indicator variable equal to one if hospital h in year t has converted to CAH status. X iht are individual characteristics, including a quadratic in age, a female dummy variable, and controls for race. t is a flexible linear trend that is allowed to change slope at the time of CAH conversion. Finally, η h are hospital fixed effects and δ t are year fixed effects. The event study model aims to estimate changes in treatment intensity following CAH conversion. I proxy for intensity using the number of procedures performed during a given discharge. Existing literature has also used length of stay and total charges to measure treatment intensity (Card et al., 2009). However, CAH criteria limits the maximum lengths of stay prohibiting a large response along that dimension 14 and total charges are highly correlated with number of procedures. In this model, β is the coefficient of interest and identifies the change in treatment intensity after conversion, relative to before. If hospitals respond to the increased marginal reimbursements of cost based reimbursement by increasing treatment intensities, we would expect β > 0. Furthermore, since the payment structure change is primarily impacting inpatient care, the impacts are expected to be larger for patients that do not spend any time in an outpatient setting (e.g. those that are not admitted through the emergency department). The validity of this empirical strategy rests on the exogeneity of hospital s CAH conversion date. However, the decision to convert was not randomly assigned; in fact, Congress provided funds to help hospitals decide if conversion was in their best interest (Reif and Ricketts, 1999). With this in mind, if there was some unobserved characteristic that is correlated with CAH conversion timing and number of procedures, this estimation strategy may lead to spurious results. To address this, I take advantage of a certain component of the CAH program to provide a robustness analysis. Specifically, included in the Balanced Budget Act of 1997, which created the CAH program, was the requirement that states interested in participating in the CAH program must submit a state plan outlining their process for program implementation. This prerequisite is one explanation for the gradual increase in CAHs observed in Figure 1; 14 Average lengths of stay for Medicare beneficiaries does increase slightly following conversion, but not of a significant magnitude. 11

12 in some instances there were hospitals that were interested in converting, but were restrained due to the state they were located in. The principal reason for the slow creation of the state plan was a lack of funding for the development process; others included a lack of leadership or a lack of political will to explore the idea (Reif and Ricketts, 1999; Gale, 2002) In an attempt to investigate the impacts of CAH conversion timing, I include a specification which contains only hospitals that converted within three years of the state obtaining approval of their implementation plan. In this case, the hospital s actual conversion date is more likely to have been a function of the state they operate in, rather than driven by some unobservable. 5 Impact of CAH Conversion 5.1 Aggregate Effects As a first investigation of the effect of CAH conversion, Figure 2 plots the average number of procedures per inpatient discharge by insurance type among CAHs. The x-axis of Figure 2 is the number of months relative to CAH conversion and the y-axis plots the number of procedures per discharge, having controlled for hospital and year fixed effects. When considering the payment change in isolation, the change from IPPS to cost based reimbursement increased the marginal reimbursement of inpatient procedures and therefore was expected to increase treatment intensity among Medicare patients. However, Figure 2 suggests the opposite. The figure shows that the average number of procedures among Medicaid and private insurance patients exhibit very minor changes following the hospital s conversion to CAH status, while the average number of procedures actually declines among the Medicare population. Table 3 presents the regression equivalent of this figure by estimating equation (1). The results in Table 3 are estimated at the discharge level and all standard errors are clustered at the hospital level. Columns (1) and (2) include hospital fixed effects and a flexible linear trend, while column (1) contains year fixed effects and column (2) controls for quarter specific fixed effects; the results are quantitatively the same. Column (3) includes separate flexible linear trends for each cohort of converting CAHs, and finds similar estimates. Finally, in 12

13 column (4), the flexible linear trends are allowed to differ based on insurance type. This model finds slightly smaller estimates, but statistical tests cannot reject the equality of the estimates in column (4) from the estimates in the first three columns. Overall, each of the 4 specifications estimate a statistically significant decline of approximately procedures per Medicare discharge, which corresponds to decline of 5-6%. Table 4 investigates these results further to see if the decline in procedures is coming from a decline in the probability of receiving any treatment (extensive margin) or a decline in the number of procedures among patients receiving at least one procedure (intensive margin). Columns (1) and (2) estimate a linear probability model of the likelihood of receiving any treatment and finds that the probability of receiving treatment declines slightly following CAH conversion. Additionally, columns (3) and (4) estimate the decline among the intensive margin and find estimates similar to Table 3. Overall, these estimates imply that both the intensive and extensive margins contribute equally towards the decline in procedures. 5.2 Compositional Changes? Before interpreting the decline in procedures as an impact of the changing payment policy, I first investigate to see if compositional changes caused the decline in procedures. First, it may be that certain patients are selecting into or away from CAHs following conversion. For example, the sickest patients may stop going to their local CAH after conversion because they view the CAH designation as a signal of inferior quality. This potential change in the composition of patients may drive observed decline in Medicare procedures. 15 Figure 3 investigates this by looking at the age, gender, and race profiles of patients. The figures show that none of the three measures appear to change discontinuously following CAH conversion. These three characteristics are consistent with the idea that the composition of patients is not changing after CAH conversion and therefore cannot explain the decline in treatment intensity. In contrast, instead of patients selecting a different hospital following conversion, it may be that newly designated CAHs alter the types of patients they treat or admit. Following conversion, CAHs may have altered the type of patients they treated either trough increased 15 Gowrisankaran et al. (2013) finds some evidence that patient s value of the hospital declines slightly following CAH conversion. 13

14 transfers 16 or shifting patients between inpatient and outpatient care. Figure 5 shows there was no change in the likelihood of a patient being transferred following conversion, independent of their insurance type. Additionally, Figure 4 plots the number of inpatient discharges and shows that while there appears to be a slope change following conversion there was not a discontinuous shift in the number of discharges. Although the changing slope cannot explain the decreased procedures, it does suggest that CAHs responded to the new payment incentives by changing the way they admitted patients. Finally, this decline in treatment may not be specific to CAHs, but rather a trend among all hospitals. I investigate this by looking at three alternative sets of hospitals. The first consists of all rural hospitals that did not converted to CAH status. In order to generate the event study figure, I assign them the conversion date of the nearest CAH hospital. The remaining two robustness datasets are created by taking each CAH and finding their nearest (non-cah) neighbor and also their nearest (non-cah) neighbor that is at least 100 miles away. For each of the two nearest neighbor hospitals, I assign them the conversion date of the original CAH. The average distance between hospitals in the rural alternative dataset and the nearest CAH is 35 miles. Among the nearby and distant (those at least 100 miles away) datasets, the average distance between the hospital and a CAH is 41 and 113 miles, respectively. 17 Panel A of Figure 6 plots the trend in number of procedures for among the distant hospitals. The figure shows that the downward shift in procedures is not experienced by all hospitals. Panels B and C plots the trends for the non-cah rural and nearest neighbor hospitals. Rather than robustness hospitals, an alternative interpretation of these hospitals is a substitutes for the CAH. Therefore, changes observed following the assigned conversion date may be evidence of patient movement between hospitals. However, both figures suggest that there is no discontinuous shift or slope change following CAH conversion. 16 CAHs were required to form a network with one or more hospitals for referral, transfer, use of communication systems, and provision of emergency and non-emergency transportation (Reif and Ricketts, 1999). 17 In all three cases, the distance is measured as a linear distance between two hospitals. These measures should be interpreted as an underestimate of the actual distance between hospitals as travel along roads is likely much longer. 14

15 6 Interpretation In an attempt to understand what caused the decline in the number of procedures per discharge, I conducted interviews with a number of people working in rural healthcare. I consulted with individuals working at rural health research groups and also chief financial officers at existing CAHs in order to get a first hand point of view. During these discussions, two ideas were emphasized: (1) inpatient services for Medicare beneficiaries were reimbursed cents on the dollar prior to CAH conversion leading to small or negative profit margins and (2) the key source of revenue was outpatient services. The fact that rural hospitals struggled under IPPS has been documented. However, those involved in rural healthcare also emphasized the fact that patient volumes (due to the hospital s relative geographic isolation) were unpredictable. This unpredictability, combined with the low reimbursement rates, lead hospitals to effectively cap the potential revenue generated through inpatient care. 18 Without the ability to generate additional income from the inpatient department, these hospitals turned towards outpatients. Outpatient services were an important driver of revenues due to Medicare s reimbursement structure for rural hospitals. While the reimbursement for each service was designed to emulate the hospitals average cost, it also included a large of amount of indirect overhead. The cost allocation process for these hospitals effectively includes pushing down all costs from non-patient departments to the departments that provide patient services. In theory, all overhead is considered in the calculation of the Medicare rate. As a result, procedures that had marginal costs below average cost generated a profit for hospitals. Therefore, increases in procedure volumes was an easy way to increase revenue. Consider the simplification that within Medicare s prospective payment, based on previous experience, hospitals realized they were going to lose money on inpatient care. In response, the hospital could maintain profitable, outpatient procedure volumes at a level high enough to offset these inpatient losses through outpatient profits. Consistent with this, one CFO at a rural hospital described the hospital as being well aware of their volume numbers, saying they knew exactly how many times various procedures had been performed. However, the fiscal condition of these hospitals quickly changed following CAH conver- 18 Nurse s wages was also suggested as a barrier to profits, as nursing costs tended to eliminate much of the profit margin. 15

16 sion. The financial status of inpatient services changed dramatically with conversion and the introduction of cost based reimbursement. Under this payment structure, inpatient losses were eliminated; the driver of inpatient revenue was no longer the patients diagnosis, but rather the hospitals incurred cost. I propose the elimination of inpatient losses is the reason we observe the decline in procedures at CAHs. Prior to conversion, hospitals maintained high outpatient procedure volumes for Medicare patients to offset inpatient losses. Following CAH conversion, inpatient losses were eliminated due to cost based payments. In the absence of inpatient losses the need to maintain outpatient procedure volumes disappeared, causing the estimated decline in procedures. The remainder of this section presents empirical facts in support of this explanation for the observed decline in treatment intensity. Although my interpretation revolves around both the inpatient and outpatient departments, my data is limited to only individuals that eventually became inpatients. However, as previously discussed in section 3.1, data on patients admitted through the emergency department includes treatments they received prior to admission. As a result, my analysis is still able to touch on outpatient behavior Patients Receiving Fewer Procedures First, I break the baseline analysis in equation (1) by patient s source of admission to see if there is a differential decline in the number of procedures. In my data, there are two primary sources of admission: the emergency department (ED) or a routine/planned admission. 19 The reason for making this break is that among patients that are admitted through the ED, the treatments they receive prior to admission are included in their discharge record. For example, if a person suffering from pneumonia-like symptoms visits the ED and the physician orders a chest X-ray to aid in diagnosis and then admits the patient based on the results, the X-ray will be included in patient s discharge record even though it technically occurred in an outpatient setting. Table 5 estimates the change in procedures, breaking the effect by patient s source of admission. Column (1) restricts the sample to only Medicare patients, while columns (2) and (3) show the effect for Medicaid and private insurance patients, respectively. The results 19 Among Medicare patients, the most common principal diagnosis for routine admissions are pneumonia, joint diseases, chronic obstructive pulmonary disease (COPD), and congestive heart failure. While admissions originating from the emergency departmnet are much more likely to be cardiac related. 16

17 suggest the decline in the average number of procedures per discharge is driven by patients that are admitted through the ED rather than routine admissions. For Medicaid and privately insured patients, changes in the number of procedures are larger among ED patients, but statistically insignificant and smaller than those experienced by Medicare beneficiaries. Although the estimates are statistically indistinguishable from zero, the fact that procedures are declining for non-medicare patients is consistent with hospital spillovers (Baicker et al., 2013). When hospitals stopped performing numerous outpatient procedures due to changes in Medicare payments, privately insured and Medicaid beneficiaries may have also received fewer treatments Hospitals Performing Fewer Procedures Among the hospitals converting, those that are most dependent on Medicare inpatient revenue should have the largest incentive to provide excess procedures under IPPS and therefore should exhibit the strongest response following CAH conversion. I proxy for Medicare dependency by dividing hospitals by the fraction of their discharges coming from Medicare. 20 Figure 7 recreates the change in expected number of procedures by dividing hospitals into high and low Medicare dependency; where the split is made at the sample median of fraction of discharges from Medicare (51%). The figure only includes Medicare beneficiaries and shows that the decline is driven by high Medicare hospitals. To provide additional robustness, I ve replicated this figure using my alternative datasets (rural, nearby, and distance hospitals); again focusing only on Medicare patients. The results are presented in Figure 8. None of the figures show a decline in the average number of procedures and no evidence of a differential effect between high and low Medicare hospitals. Table 6 presents the regression analog of this figure for CAHs. Instead of dividing hospitals into dichotomous groups, the specification allows each hospital s dependency on Medicare to vary continuously. The estimated equation is as follows: y iht = α + β CAH ht + λ CAH ht (F rac h 0.40) + γx iht + θ t + η h + δ t + ε iht (2) where F rac h is the average fraction of the hospital s discharges that come from Medicare 20 This is consistent with Salkever (2000) and He and Mellor (2012). 17

18 prior to conversion. Note that F rac h is standardized to be the percentage points above 40%. This standardization means we can interpret β as the predicted change in the average number of procedures for a hospital that has 40% of its inpatient discharges coming from Medicare. 40% is choosen because it is the 25 th percentile of F rac h. λ measures if the estimated change in procedures per discharges varies according to a hospitals dependency on Medicare. I expect λ > 0; hospitals that treat more Medicare patients should experience larger declines in the number of procedures. One potential issue with this measure of Medicare dependency is that it may be endogenously determined by hospitals, e.g. by moving Medicare beneficiary s care between outpatient and inpatient care. To account for this possibility, I instrument for the fraction of discharges from Medicare by the fraction of the hospital s surrounding population that is above 65 in order to isolate variation that is caused factors exogenous to the hospital. Data on the fraction of the population above 65 is collected at the city level from the 2000 U.S. Census. 21 Column (1) of Table 6 first presents the OLS results of this specification among Medicare beneficiaries. Columns (2) through (4) are the IV estimates that instrument the fraction of discharges coming from Medicare by the fraction of the surrounding population that is above 65. Formally, the first stage equation is the regression of CAH jt (F rac h 0.40) on CAH jt Fraction65+. All regressions include a flexible linear trend, a quadratic in age, and a gender dummy variable. The OLS estimates imply no change in the number of procedures for a hospital at the 25 th percentile of F rac and larger declines as the fraction of discharges from Medicare increases. The OLS estimate for Medicare beneficiaries are slightly smaller than the IV estimate in column (2), but a test of their equality cannot be rejected. The IV estimates also predict larger declines in the number of procedures per discharge as the fraction of discharges from Medicare increases. The estimates suggest that if a hospital s discharges from Medicare increase by 10 percentage points, the expected change in the number of procedures would be 5.7% smaller. Column (3) implies no statistical effect among Medicaid patients and column (4) finds larger declines for high Medicare hospitals, however the estimates are half the size of those implied for Medicare beneficiaries. 21 Data was retrieved from factfinder.census.gov. 18

19 Figure 9 combines the results on the types of patients and hospitals driving the reduction in procedures. Examining ED admissions and routine admissions in isolation, it shows that the primary source of the decline is driven by patients admitted through the ED at high Medicare hospitals. Table 7 presents the regressions of this specification. Columns (1) and (2) estimate equation (2) for Medicare beneficiaries, but breaks the sample by the patients source of admission. Column (2) shows that the decline for Medicare patients is almost entirely driven by ED admission. Table 7 also shows that the smaller effects estimated for the non-medicare population are primarily seen through ED admissions Timing of Procedures Declining One benefit of the SID data is that for each procedure performed, it records when the procedure was performed relative to the day of admission. Utilizing this data, I define early procedures as those that are performed either on the day of admission or before (my data does not allow me to create a more detailed measure). The purpose of this variable is to attempt to focus on procedures that are more likely performed in an outpatient setting. Although I cannot say with certainty that a procedure performed on the day of admission was done while the person was an outpatient, the variable is defined identically before and after CAH conversion so mismeasurement will not drive any estimated impact. 22 Figure 10 plots the average number of early procedures and shows that early procedures decline primarily among patients admitted through the ED at high Medicare hospitals. Table 8 estimates equation (2) using early procedures as the outcome. The results show that a hospitals with 40% of its discharges coming from Medicare is not predicted to change the number of early procedures performed following CAH conversion, but higher Medicare hospitals do experience a decline in early procedures. The estimates imply that approximately two-thirds of the overall decline in procedures is explained by declines in early procedures Types of Procedures Declining Finally, if hospitals are using outpatient services to offset losses from inpatient services, there are certain procedures that should be performed excess. Specifically, hospitals should 22 Additionally, this would create mismeasurement error in the outcome variable and would only increase the standard errors if the measurement error is uncorrelated with CAH conversion. 19

20 be over utilizing services that have low marginal costs. In hospitals, these are often machine dependent procedures that require a large fixed cost investment. To investigate this, for each procedure category within HCUP s Clinical Classification Software I calculated the probability that a Medicare patient receives that treatment for both high and low Medicare hospitals. Columns (1) and (2) of Table 9 depicts this information and column (3) presents the percent difference between the two. The table includes the eight procedures with the largest percentage difference between the two categories of hospitals. An important pattern that emerges is that six of the top eight procedures, those in bold face font in Table 9, are characterized by high fixed costs and low marginal costs. Furthermore, Table 10 replicates this analysis among only high Medicare hospitals and compares the probability of receiving a given treatment between before and after CAH conversion. In this case, five of the eight largest declines in procedure usage is among procedures with low marginal costs. The results from Tables 9 and 10 are consistent with the findings of Kim (2011), which finds that following the introduction of Medicare IPPS, hospitals increased their usage of treatments that had average costs greater than marginal costs in order to take advantage of the economies of scale Reconcilation Empirical results presented thus far are consistent with the decline in procedures being caused by an interactive effect between inpatient and outpatient departments that led to initially high outpatient procedure volumes. Declines through the outpatient department are with my findings that ED admissions drive the result (only patients that spend time as an outpatient) and the majority of them are early procedures (those more likely performed in an outpatient setting). Furthermore, high Medicare hospitals experiencing larger declines is consistent with inpatient losses being the driving force for the increased outpatient treatments. As Medicare reimbursed rural hospitals below costs for inpatient services, hospitals with a higher baseline Medicare patient share likely had larger inpatient losses and thus had the largest incentive to maintain high procedure volumes. 23 Finally, if the excess procedures are done primarily to drive up profits, my finding that low marginal cost procedures are driving the decline in procedures is also consistent as they are the most profitable. 23 Section 6.2 empirically shows that high Medicare patient share is correlated with hospital profits. 20