Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Essays on organizational forms and performance in California hospitals
(USC Thesis Other)
Essays on organizational forms and performance in California hospitals
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
ESSAYS ON ORGANIZATIONAL FORMS AND PERFORMANCE IN CALIFORNIA HOSPITALS by Mehdi Farsi A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment o f the Requirements o f the Degree DOCTOR OF PHILOSOPHY (ECONOMICS) December 2002 Copyright 2002 Mehdi Farsi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3093759 UMI UMI Microform 3093759 Copyright 2003 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UNIVERSITY OF SOUTHERN CALIFORNIA THE GRADUATE SCHOOL UNIVERSITY PARK LOS ANGELES, CALIFORNIA 90089-1695 This dissertation, written by rlekdli V~o:rSi under the direction o f h j_ s_ dissertation committee, and approved by all its members, has been presented to and accepted by the Director of Graduate and Professional Programs, in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY _____________ f Director 12- 18-2002 D ate_ _____ Dissertation Committee Chair Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ii ACKNOWLEDGEMENTS I would like to thank Janet Currie, Bentley MacLeod and Geert Ridder for their incredible support throughout this research. This dissertation would have been impossible to complete without Janet and Bentley’s insights and their continuous guidance. Chapter 2 is the result of a joint research with Geert to whom I am particularly grateful for his statistical model. I am also grateful to Elizabeth Graddy, Isabelle Perrigne and Quang Vuong for their encouragement, suggestions and helpful comments. I have also benefited from the comments of participants of graduate student workshops at the University of Southern California and University of California at Los Angeles. I gratefully acknowledge the financial support of the USC Graduate School as well as a disseration fellowship from the National Bureau of Economic Research. I also thank the California Office of Health Planning and Development and the UCLA Institute for Social Science Research for providing me with the data, and the UCLA Academic Technology Services for providing some of the computing facilities. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. TABLE OF CONTENTS Acknowledgements ii List of Tables vi List of Figures viii Abstract ix 1. Changes in hospital quality after conversion in ownership status 1 1.1. Introduction 1 1.2. Background 6 1.3. A model of patient selection 7 1.4. Data 11 1.4.1. Conversions 12 1.4.2. Patient-level data 16 1.4.3. Measures of quality 21 1.5. Methods 24 1.5.1. Patient selection 27 1.5.2. Treatment practices 30 1.6. Results 30 1.6.1. Mortality 30 1.6.2. Robustness of the results 36 1.6.3. Operational features 41 1.6.4. Other measures of quality 42 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.7. Conclusions 44 2. Estimating the out-of-hospital mortality rate using patient discharge data 46 2.1. Introduction 46 2.2. Model 49 2.2.1. In-hospital mortality and discharge rates 50 2.2.2. Out-of-hospital mortality and hospitalization rates 51 2.2.3. Measures used in the literature 54 2.3. The patient discharge data 58 2.3.1. Description of the data 5 8 2.3.2. Implementation of the model 59 2.4. Estimation results 61 2.5. Conclusions 69 3. Consequences of mergers and acquisitions in California’s hospital market 70 3.1. Introduction 70 3.2. Background 75 3.3. Data 80 3.3.1. Consolidation of California’s hospitals 80 3.3.2. Major Hospital Chains in California 82 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. V 3.3.3. Patient-level Data 87 3.4. Methods 88 3.5. Results 93 3.5.1. Which hospitals are selected? 93 3.5.2. Consequences of affiliation to the chains 96 3.5.3. Scale economies 100 3.5.4. Market power 103 3.6. Conclusions 104 4. Concluding remarks 107 References 111 Appendix 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vi LIST OF TABLES 1.1. Share of California acute-care hospitals by sector in 1990 and 1998 14 1.2. Number of California acute-care hospitals and hospital beds involved in a conversion in ownership status from 1990 to 1998 15 1.3. A descriptive summary of hospitalizations 18 1.3. Mortality regression without controlling for hospital fixed effects 31 1.5. Mortality regressions accounting for hospital fixed effects 34 1.6. Ownership effects on hospital operational characteristics 37 1.7. Cardiac Catheterization for heart patients 43 1.8. Readmission of AMI patients 44 2.1. Sample statistics for hospital spells 62 2.2. Sample statistics for out-of-hospital spells 62 2.3. Mortality and discharge rates for hospital spells 64 2.4. Mortality and re-hospitalization rates for out-of-hospital spells 66 2.5. Mortality and discharge rates for hospital spells (on converted hospitals) 67 2.6. Out-of-hospital mortality and re-hospitalization rates for out-of-hospital spells (on converted hospitals) 68 3.1. Number of California acute-care hospitals that joined a network 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.2. Number of California acute-care hospital beds that joined a network 86 3.3. Effects of hospital characteristics on the probability of joining a network (including hospital fixed effects) 94 3.4. Effects of hospital characteristics on the probability of j oining a network 95 3.5. Effects of hospital characteristics in the base year (1990) on the probability of joining a network 96 3.6. Effects of joining a network on hospital characteristics 97 3.7. Effect of hospital acquisition on chain’s other hospitals in county 102 3.8. Effect of hospital acquisition on prices and mortality outcome of elderly AMI patients treated in the chain’s other hospitals in the county 105 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF FIGURES 3.1. Number of acute-care hospitals in California 3.2. Average number of hospitals operated by multi-hospital firms in California 3.3. Share (%) of California hospital beds operated by multi-hospital systems 3.4. Share (%) of California hospital beds operated by major multi-hospital systems Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ix ABSTRACT In this dissertation the effect of the recent ownership changes on the quality of medical service and financial performance in California hospitals is analyzed. Chapter 1 examines the effects of hospital conversions on quality of care. The sample includes all Medicare patients treated in California hospitals from 1990 to 1998 for one of four conditions: Acute Myocardial Infarction (AMI), Congestive Heart Failure (CHF), malignant lung cancer, and injuries due to traffic accidents. Measures of quality examined include: in-hospital mortality, the probability of readmission and congestive heart-failure complication for AMI patients, and the rate of cardiac catheterization for both AMI and CHF patients. I find that in-hospital mortality for AMI and CHF patients fell in hospitals that converted from non-profit (NP) to for-profit (FP) status. This finding is stronger among patients admitted from emergency rooms (ER), and does not appear to be driven by changes in transfer/discharge policies. Per day charges are also higher in hospitals that convert to FP status, while waiting times for principal procedures are lower. I develop a model which predicts that if FP hospitals are more cost-efficient than NP hospitals, they will supply more intensive treatments and attract sicker patients. This model is found to be consistent with the data. However, conversion to FP status is also associated with a higher probability of CHF complication among Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. X AMI patients, and a lower rate of admission from the ER, suggesting that quality does not improve uniformly among converting hospitals. In chapter 2 using a transition model it is shown that the out-of-hospital mortality rates can be estimated using the discharge data without post discharge death records. The results indicate a considerable variation in the discharge rate of AMI patients among different hospital types. The common measures of hospital quality used in the literature are studied. Most of these measures confound the mortality rate with the hospital discharge rate. Given the significant variation of discharge rates across hospitals, such measures of quality may be misleading. Regarding the effect of conversion in ownership status on the mortality of AMI patients, the results are consistent with those reported in chapter 1 in that the FP status is associated with lower mortality rates. In addition, using the proposed model it is shown that the discharge rates are higher in hospitals with FP status. However, the results provide no evidence that faster discharge of these patients may lead to higher risks of post- discharge death. Chapter 3 examines the effects of hospital acquisitions on the financial performance of hospitals owned by or affiliated to the four major hospital networks in California. These networks include two FP chains Tenet Healthcare Corporation and Columbia/HCA and two NP systems Catholic Healthcare West and Sutter Health. The sample includes all non-federal acute- Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xi care California hospitals from 1990 to 1999. The results suggest that all these chains have mainly targeted relatively large hospitals. In particular the FP chains have focused on well-equipped hospitals with relatively more costly patient case-mixes. Acquisition by these chains led to a general downsizing of the hospitals especially in medical staff and equipment. On the other hand, the NP systems seem to considerably increase the medical staff of their new hospitals. These networks can guarantee a larger access to patients through a higher number of affiliated physicians. The results provide some evidence of scale economies used by the FP chains. Approximately for every thousand additional hospital beds in any given county, these chains achieve savings of about 2.6 percent in total operating costs and 6.5 percent in administrative expenses. These savings were not accompanied with any decline in quality as measured by the in-hospital mortality of elderly patients hospitalized for AMI. A patient-level analysis of the elderly AMI patients provides suggestive evidence that the studied NP chains exercise market power by charging higher prices as they treat more AMI patients in the county through hospital acquisition. Chapter 4 provides general conclusions and suggestions for further research. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 1. CHANGES IN HOSPITAL QUALITY AFTER CONVERSION IN OWNERSHIP STATUS 1.1. Introduction Recent conversions of public and non-profit (NP) hospitals to for-profit (FP) status have raised public concerns about possible detrimental effects on quality of care (c.f. Goddeeris and Weisbrod (1998), Kuttner (1996a-b) and Ho and Hamilton (2000)). A common perception is that NP institutions are committed to providing quality care regardless of costs. In fact, following Arrow (1963) theoretical models often assume that providers choose the NP form of organization in order to signal this high commitment to quality (c.f. Frank and Sulkever (1994) and Glaeser and Schleifer (1998)). Sloan (2000, 2001) and Baker et al. (2000) provide extensive surveys of the growing literature on the effects of NP and FP status on the quality of care. However, few studies actually examine conversions of hospitals from one form to the other. Sloan (2001) is an important exception. He examines the effects of conversions in a national sample of hospitals on elderly patients admitted for stroke, hip fracture, coronary heart disease, congestive heart failure and pneumonia. He finds that conversions have no effect on in-hospital mortality or on the proportion of uninsured patients. However, he finds that Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2 pneumonia patients in hospitals that converted to FP status experienced an increased rate of complications. He argues that the failure to find a significant effect on in-hospital mortality may reflect shorter hospitalizations after conversion to FP status. This study examines the effects of conversions in the California hospital market over the period 1990 to 1998 using models with hospital fixed effects. It builds on Sloan’s research in several ways. One of the important difficulties in studying the effects of conversions is that patients may be selected differently before and after the conversion. In particular, an institution that switches from NP to FP status may step up efforts to discourage the admission of unprofitable patients. In order to assess the importance of this type of selection, we exploit the fact that different types of patients are selected in different ways. For example, heart attack patients are generally taken to the nearest hospital. Moreover, hospitals are required to treat patients in such an emergency situation regardless of their insurance coverage.1 On the other hand, victims of serious trauma (such as car accident victims) are taken to designated trauma centers. Hospitals can avoid these patients by closing down or scaling-back trauma units. Finally, patients with serious chronic conditions, such as lung 1 Although many hospitals violate these requirements, there is no significant difference in propensity to violate between FP and NP hospitals in California (Blalock and Wolfe, 2001). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3 cancer generally have time to plan their hospitalizations and their decisions may be influenced by many factors within the control of hospitals. These considerations suggest that the measured effects of conversion should be least masked by selection in patients with heart attacks (Acute Myocardial Infarction or AMI) and most subject to selection bias in patients with chronic conditions and trauma victims. A second question is how hospitals that convert are selected from the pool of hospitals. According to my primary analysis using the data on financial characteristics of the hospitals there is little evidence of such a selection. In general financial variables such as costs and revenues have very low explanatory power in explaining the conversions between NP and FP forms.2 However, the available evidence in the literature suggests that conversions are usually preceded by financial difficulties (Sloan (2001) and Mark (1999)). In this case, care may improve post-conversion simply because the hospital’s financial distress is alleviated. The California hospital market was characterized by almost equal numbers of conversion from FP to NP and vice- versa. To the extent that financial distress motivates both types of conversions, the fact that conversions occur in both directions will help us to identify the real effects of conversions. Similar to Sloan (2001) I assume that the 2 The only consistently significant pattern in my results is that among the NP hospitals relatively large ones are more likely to convert to FP status. The insignificant results may be due to the measurement errors in financial variables. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4 conversion from NP to FP form has the opposite effect of conversion from FP to NP status. This symmetry assumption is tested using a model that allows for different effects of opposite conversions. Unlike Sloan, I also control for constant characteristics of hospitals being converted by including hospital fixed effects in the regression models. Finally, I present a simple theoretical framework for understanding the effects of conversion. The model assumes that FP hospitals are more cost- efficient than NP hospitals in the sense that they have lower marginal costs of treatment. Patients are assumed to value both aspects of hospital care that affect their health, and aspects of care (such as location), which do not. In equilibrium, FP hospitals end up supplying more intensive treatments to sicker patients so that on balance outcomes may be better or worse than in NP hospitals. The results suggest that FP status is associated with lower in-hospital mortality for AMI and for Congestive Heart Failure (CHF) patients. I find no significant effects on the mortality of Malignant Lung Cancer (LC) or Traffic Accident (TA) patients. These results are consistent with the hypothesis that effects on quality are easier to detect where patients are less selected. Similarly, I find stronger effects when I control for the reported severity of the patient’s condition, and when I restrict my analysis to patients who were admitted from Emergency Room (ER). Moreover, the in-hospital mortality Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5 results hold up when I account for censoring of hospitalization spells in a hazard-model framework. I also find that patients in FP hospitals experience shorter stays, especially lower waiting times before their procedures, and higher per-day charges. These results are consistent with the model in that they suggest that FP hospitals supply care more intensively and to sicker patients than NP hospitals. Conversions do not have any significant effect in the probability that a cardiac patient receives Cardiac Catheterization (CC), a diagnostic procedure that is used to study the possibility of re-vascularization procedures such as angioplasty or coronary bypass surgery. On the other hand, I find that the probability that AMI patients are re admitted with CHF (the most common complication) is relatively high in hospitals with FP status. Also, while I find few actual closures of ERs following conversion to FP status, I find that the probability that patients are admitted from the ER falls. These results suggest that some of the public concern over these conversions is warranted. The rest of this chapter proceeds as follows: Section 1.2 reviews some of the previous literature. Section 1.3 provides a model of patient selection across providers of medical care. A description of the data and our measures of quality is given in section 1.4. Section 1.5 explains the econometric Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6 methodology and discusses potential sources of bias. Section 1.6 provides the results. Conclusions follow. 1.2. Background Between 1970 and 1995, 330 of 5,000 NP hospitals (about 7%) converted to FP type (Cutler and Horwitz, 2000). These conversions accelerated in the mid-90s. For example, 58 conversions occurred in 1995, up from 34 in 1994 (Kuttner, 1996a). These developments have spurred a large literature on the effects of FP and NP status on quality of care, but the results are far from conclusive largely because of the difficulty of controlling adequately for patient selection. As Kessler and McClellan (2001) suggest, more productive hospitals may attract sicker patients. Geweke et al. (2001) provide some evidence that more severely ill patients are in fact more likely to be hospitalized in high quality institutions. Studies such as Gowrisankaran and Town (1999), Ettner and Hermann (2001) and McClellan et al. (1994) suggest that many patients choose the closest hospital, but this does not mean that FP status can be treated as exogenous determinant of mortality because FP hospitals are more likely to locate in areas with better insured patients, for example in areas with high proportion of Medicare patients (Norton and Staiger, 1994). McClellan and Staiger (2000) find that NP hospitals have slightly lower mortality rates in a sample of elderly AMI patients, but reported that the estimated effects fell by Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7 almost half when county fixed effects were included in the model. Sloan (2001) cites these results and concludes that FP hospitals tend to be located in areas with higher mortality rates. Finally, Kessler and McClellan (2001) find that although in-hospital mortality and the probability of readmission is lower in NP hospitals, the presence of FP hospitals in the same area reduces expenditures at NP hospitals without affecting health outcomes in these hospitals. These spillovers are suggestive of a market in which FP hospitals are more efficient than NP hospitals and also take on the sickest patients. 1.3. A model of patient selection The literature summarized above suggests that it is important to take account of the selection of patients into hospitals. This section presents a simple model of patient selection in which FP hospitals are more cost-efficient. Since they have lower marginal costs of treatment, they supply more intensive treatments. Patients value both intensive treatments and other aspects of hospital care. Sicker patients will place more weight on intensive treatment and hence will be more likely to choose FP hospitals. Consider a continuum of diseases with a normalized index Ae[0,l], representing the nature of the disease regarding patient’s choice of medical care. This index is the weight the patient attaches to the efficacy of treatment. For instance in emergency conditions like heart attacks for which, patients go Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8 to the closest hospital, A =0, whereas in cases like trauma where patients are taken to the best equipped facility, A =1. Each individual patient has a disease with severity level se[0,l]. The individual patient’s ideal choice of hospital is represented by her type /e[0 ,l]. This variable represents parameters like hospital location and certain quality aspects that do not affect the treatment, but are important as a matter of subjective preferences. The probability of a patient’s survival (recovery) is defined by differentiable function P(s,t), where te[0,l], is the level of treatment used. It is assumed that Pt>0, Ps<0, and Pst>0, which implies that the survival probability decreases with severity and increases with treatment level, and also marginal benefit of treatment increases with severity.3 Moreover, it is assumed for simplicity, that second-order differential Pu is equal to zero. The level of treatment is chosen by hospitals. It is assumed that there are only two types of hospital. Hospital characteristics are represented by two variables (Fm d '/i, with H e {1,2}. represents the hospital cost-efficiency and ' f represents hospital type regarding patient preferences. For simplicity suppose that /= 0 , f= l, and (¥>$. Hospitals have a cost of treatment defined by a differentiable function C(t, 6^), that satisfies the following conditions: Ct>0, Ctt>0, C(t, 0*)<C(t, 01 ), Ctit, 0)<Ct(t, e1 ), and C„(t, 01 ) for all 3 In other words treatment matters more for sicker patients, and less severely ill patients survive even with an insufficient treatment. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9 t. This means that cost function is increasing and convex in medical services, and both total and marginal costs are lower for more efficient hospitals (higher value of 0). A simple example of such a function is c(t)/0, where c(t) is a convex function. Hospitals’ revenue consists of expected payments by insurers or other third parties, defined by a differentiable function R(s,t). These payments are non-decreasing in delivered medical services t, that is: Rt>0. It is assumed that the probability of complete reimbursement of a service increases with severity, and decreases with the intensity of medical services.4 Therefore, the expected payment (both in total and marginal value) increases with severity, and the marginal revenue of medical services is decreasing, which implies Rs>0, Rst>0, and Ru< 0. At stage 1, patients observe their disease and type: X, y, and 5 . At stage 2 they choose their hospital H e{ 1,2} to maximize their utility defined as follows: m s, if, ( 1.1) At stage 3, hospitals choose the level of treatment t, to maximize their net revenue, R(s,t)-C(t, 0^), and finally the outcome is observed. 4 One may argue that in a diagnosis-based reimbursement scheme, payments are independent of services. However, a concave form allows us to ignore the institutional constraints that restrict the providers from lowering their services below a reasonable minimum. Moreover, this reimbursement system is far from ubiquitous. For instance, some insurers pay the hospital charges with a negotiated discount. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10 Therefore, the hospitals’ choice of treatment is obtained from: t(s,0) = argmax, [R (s,t)-C (t,0)] (1.2) and the patients’ choice of hospital is written as: max# {IP(s,t(s, ^ ) ) - ( l - X (1.3) Based on the above first-order conditions, the following proposition can be proved (see appendix for the proof): Proposition 1.1. If the payment function R(s,t) is non-decreasing and concave in t, increasing in s, and satisfies Rst>0, hospital cost function C(t, 0), is increasing and convex in t, and both Ct and Cu are decreasing in 0, and also patient survival probabihty is decreasing in s, increasing in t, and satisfies Pw >0 , and Ptt- 0, then: 1) more severely ill patients are more likely to choose more efficient hospitals; 2) hospitals deliver a higher level of care to sicker patients, with more efficient hospitals doing so at a higher rate; 3) compared to other hospitals, more efficient hospitals provide more medical services to the patients with the same severity of illness; and 4) among patients with similar severity s, probability of recovery is higher for those who go to more efficient hospitals. Although more efficient hospitals have a higher probability of survival for a given patient, their aggregate (hospital-specific) survival rate may be Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 11 higher or lower, depending on their case mix. Assuming that the distribution of patients’ taste of hospital is symmetric with regard to hospital types, the following proposition can be proved (see appendix for the proof): Proposition 1.2. If patients’ taste for hospitals (y) has a symmetric distribution around 0.5, then for a given level of severity, the aggregate probability of survival is higher in more efficient hospitals. Moreover, in the special case when A=0 (patients choose only based on their subjective preferences), this statement is true even without controlling for severity of illness. A general implication of the model is that a correct estimation of differences across hospitals depends directly on whether the severity variations are taken into account. Without a sufficient control for severity, the effect of cost-efficiency on aggregate health outcomes is ambiguous. Moreover, as we move from random assignment (T=0) to systematic selection of patients (/L=l), patients are more likely to consider hospital efficiency in their decisions, hence, adjusting for severity becomes more crucial. 1.4. Data The data used in this paper consist of two main data sets prepared by the California’s Office of Statewide Health Planning and Development (OSHPD). The first set is the Patient Discharge Data that includes all the discharge abstracts for the elderly (65 years and older) patients discharged Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 12 from a Californian hospital from 1990 to 1998. The variables include patients’ basic characteristics like age, gender and race, length-of-stay, total charges, severity of the disease, the diagnosed conditions and procedures used for treatment. The hospital charges reported in discharge abstracts are obtained by multiplying services by unit list prices. These prices represent costs rather than transaction prices which are typically negotiated between insurers and providers. Severity of illness is defined in four levels (extreme, major, moderate, and minor) according to APR-DRG (All Patient Refined Diagnosis- Related Group) classification. This severity measure and its validity are discussed later in this section. The second data set is the California’s Hospitals Disclosure Data from 1989 to 1998. This data set consists of the information obtained from the hospital financial reports submitted annually to the Department of Health Services. Hospital characteristics like ownership status, size (number of beds) and type of the hospital are extracted from this data set. 1.4.1. Conversions The changes in California’s acute-care hospital market5 share in private FP, private NP, and public sectors between 1990 and 1998 are given in table 1.1. The changes in average hospital size in terms of the number of available beds in a hospital are also given. These numbers suggest that during this period, an increasing number of patients switched from public hospitals to 5 By acute-care hospitals I mean all non-psychiatric hospitals that reported acute inpatient care. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 13 private hospitals. At the same time FP hospitals became larger on average while public hospitals decreased their capacity. Conversions are responsible for part of these asymmetrical changes across different sectors. The data show that 62 conversions in ownership status occurred in California hospitals between 1990 and 1998. Among 591 acute-care hospitals that have operated in California, 58 hospitals had at least one conversion during this period. These hospitals on average, account for 8.8% of hospital beds in California. Three of the four hospitals that have experienced two conversions, returned to their initial status after a first conversion.6 Table 1.2 gives the distribution of conversions over time. The number of hospitals and hospital beds are both given. As suggested by these numbers, the conversions are spread over the nine years and do not show any broad temporal pattern. Conversions occurred in all directions but the most frequent ones were between NP and FP status, and only a few conversions occurred between FP and public status. The variations in the size of the converting hospitals (given in table 1.2) suggest that among FP hospitals, the larger institutions are more likely to convert to NP status. Similarly, the smaller public hospitals are more likely to change to private form. Using the data on hospital financial characteristics I 6 1 checked the validity of the repeated conversions using the information on the name of the hospital and the owner. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 14 studied whether hospital characteristics can predict the probability of conversion between FP and NP forms. I studied the two sub-samples of FP and NP hospitals separately and for each sub-sample I created a binary indicator representing whether the hospital will convert to the other form in the following year. I regressed this conversion indicator on hospital characteristics and year dummies. I used several specifications with and without hospital fixed effects. My analyses (not shown here) suggest a very low explanatory power for financial variables such as costs and revenues. The only consistently significant pattern in my results is that among the NP hospitals relatively large ones are more likely to convert to FP status. Table 1.1. Share of California acute-care hospitals by sector in 1990 and 1998 Year FP NP Public Share of hospitals (%) 1990 28.6 51.3 20.1 1998 27.5 54.4 18.1 Share of hospital beds (%) 1990 18.1 55.7 26.2 1998 19.4 58.7 21.9 Share of admissions (%) 1990 16.0 65.4 18.6 1998 16.7 69.2 14.1 Average size (number of beds) 1990 122 210 250 1998 138 212 239 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 15 The above results suggest that among private FP and NP hospitals the relatively large hospitals in each sector are more likely to convert to the other form. However, given that hospital capacity is an endogenous parameter that can change with conversions, it is not included in the model.7 Table 1.2. Number of California acute-care hospitals and hospital beds involved in a conversion in ownership status from 1990 to 1998 FP to NP to NP to Public FP to Public NP FP Public to NP Public to FP Total: 17 19 8 15 1 2 (2585) (4013) (394) (2708) (50) (286) Average 152 211 49 181 50 143 size: 1 1 1 1990 (116) - (69) (12) - - 1991 2 2 2 (295) (219) (66) - - - 1992 3 2 1 2 (267) (233) (30) (734) - - 1993 3 1 1 2 (391) (157) (40) (536) - - 1994 3 4 2 (639) (291) - (101) 1 (33) - - 1995 — 1 (274) - - - 1996 2 1 (86) 2 1 (392) (83) - (50) - 1997 1 4 1 2 1 (230) (1635) (106) (165) - (233) 1998 2 4 5 1 (255) (1118) - (1127) - (53) Total number of beds is given in brackets. Average size is the average number of beds. 7 My regressions (not shown here) suggest that converted hospitals may change their capacity. However, including the number of beds in the regressions does not change the results of the paper. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 16 Other hospital characteristics that typically do not change with conversion, are captured by hospital fixed effects. For instance, Keeler et al (1992) found that among all hospital characteristics, the involvement in teaching activities has the most significant effect on their quality measures. There is no association between conversions and teaching status in our sample. In fact there are only two FP hospitals that have teaching status, one of which is a non-converting hospital and the other has converted from NP status, but kept its teaching affiliations after conversion. 1.4.2. Patient-level data Hospitalizations of California’s elderly patients for the following four diagnostic categories have been chosen: acute myocardial infarction (AMI), congestive heart failure (CHF), malignant lung cancer (LC), and hospitalizations due to motor vehicle traffic accidents (TA).8 In each case the sample contains all the patients of 65 years of age and older, hospitalized with the corresponding condition as principal diagnosis.9 Elderly patients provide relatively more homogeneous samples not only regarding age-related risk factors, but also because of a single insurance coverage. 8 Two other diagnostic groups, hypertensive heart disease (HHD) and diabetes mellitus (DM) were also studied using a similar methodology. However, these samples did not show any significant ownership effects, and are excluded from the dissertation to avoid unnecessary repetition. 9 The corresponding codes according to the International Classification o f Diseases, 9th version, Clinical Modification (U.S. Department o f Health and Human Services) are as follows: AMI: 410.xx, CHF: 428.0, 402.xl, 398.91, 404.xl, and 404.x3, LC: 162.x, and TA group consists of E-codes E810.X through E819.X. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 17 In addition to considerations of patient selection, the choice of diagnoses is also based on the variety of treatment methods available. One can expect a higher variation in hospital quality for diseases (like heart diseases) whose treatment is chosen from a relatively wide range of procedures. CHF and lung cancer are chosen to represent this effect. Compared to CHF, lung cancer patients do not benefit from a wide variety of treatment methods. There has been much more innovation in the treatment of heart disease in general and CHF in particular (Braunwald and Bristow, 2000). Since the main measure of quality is in-hospital mortality, the diagnoses are chosen from the most important causes of death. According to the California mortality data, three of the above categories (AMI, CHF, and LC) are ranked among the most deadly diseases in California and throughout the US.1 0 Table 1.3 gives the distribution of the patients and a descriptive summary of some of the features of hospitalizations by sector. The size of the samples varies from 35,700 for the TA group to 565,500 for CHF patients. NP hospitals have the largest share (65 to 73 percent) of hospitalizations in all groups. Comparing the share of FP and public hospitals shows that FP hospitals have a higher share in heart diseases and lung cancer, whereas public hospitals have a significantly larger share of traffic accidents. FP hospitals 1 0 US Vital Statistics Mortality: multiple cause-of-death summary (1968-98), National Center for Health Statistics. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18 have the highest mortality rates in AMI sample and the lowest ones in three other groups. Table 1.3. A descriptive summary of hospitalizations Diagnostic group: AMI CHF LC TA Number of admissions: 290,059 565,536 71,141 35,708 Number of hospital-year groups: 2,020 2,082 1,924 1,780 Distribution of For-Profit 15.5 18.3 14.9 13.1 admissions (%): Non-Profit 72.8 69.4 74.9 64.6 Public 11.7 12.3 10.2 22.3 For-Profit 14.4 5.7 16.2 3.8 Average in- Non-Profit 13.1 5.9 17.4 5.7 hospital death rate Public 14.2 5.8 20.2 8.7 (%) by status and 1990 15.9 7.4 21.3 6.2 year: 1998 11.5 4.8 14.5 5.0 Overall 13.4 5.8 17.5 6.2 For-Profit 76.3 78.8 74.1 75.5 Average age Non-Profit 76.0 78.5 73.7 75.5 (years): Public 76.1 78.1 73.7 75.6 Overall 76.1 78.5 73.8 75.6 Percent of patients For-Profit 46.9 37.5 58.2 26.3 reported in Non-Profit 45.3 37.9 52.7 28.7 extreme or major severity Public 46.1 36.3 56.1 33.2 categories: Overall 45.6 37.7 53.8 29.4 For-Profit 68.7 60.0 31.7 73.8 Percent of admissions Non-Profit 70.6 67.9 31.5 80.8 through ER: Public 81.2 74.2 44.8 89.5 Overall 71.6 67.2 32.9 81.8 For-Profit 6.0 5.6 8.3 6.3 Average length- Non-Profit 6.4 5.6 7.8 6.4 of-stay (days): Public 6.3 5.5 8.4 7.2 Overall 6.3 5.6 8.0 6.6 Average stay For-Profit 1.6 1.6 2.2 1.6 before performing Non-Profit 2.0 1.9 1.8 1.3 the principal Public 1.9 1.6 2.2 1.4 procedure (days): Overall 1.9 1.8 1.9 1.4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 19 Table 1.3. Continued Average per- patient per-day expenses ($1000): For-Profit Non-Profit Public Overall 5.3 3.1 3.7 4.1 5.2 2.9 3.4 5.1 4.0 2.4 2.7 3.8 5.1 2.9 3.4 4.7 Average per- patient expenses ($1000): For-Profit Non-Profit Public Overall 29.4 16.7 30.1 25.2 31.3 15.7 25.7 33.2 23.3 12.4 21.2 25.3 29.9 15.5 25.9 30.2 Total admissions in the diagnostic group per 1000 hospital admissions: For-Profit Non-Profit Public Overall 15.1 30.0 3.6 6.1 17.7 25.9 4.5 7.8 14.6 28.8 3.5 12.9 16.4 27.4 4.4 8.3 These numbers also indicate that except for the TA group, FP hospitals attract older patients and patients admitted to these hospitals are on average more likely to belong to extreme/major severity categories (according to the reported APR-DRG classification). Public hospitals have the highest rate of ER admissions and the longest hospitalizations. Another important feature of the data is that except for the TA group, the average per-day expenses are the highest in FP hospitals and lowest in public facilities. The ratio of the total number of admissions in a diagnostic group over the total admissions in a hospital is also given in different sectors. These numbers suggest that while private hospitals seem to “specialize” in AMI, CHF and LC, in public hospitals a much higher share of hospital capacity is allocated to traffic accidents. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 20 Among private hospitals, FP’s specialize more in CHF whereas NP’s have a higher share of admissions in three other groups. The selection patterns observed in table 1.3, suggest that an unbiased estimation of ownership effects requires controlling for severity variations across hospitals. The risk factors considered in this paper include demographic covariates like age, gender, race (black/non-black), and ethnicity (Asian and Hispanic groups), as well as a binary measure of severity that indicates whether the patient’s condition is more severe than the average patient within the same hospital-year-diagnosis group. This index is constructed based on APR-DRG classification.1 1 In addition, in the case of CHF and TA samples where the diagnosis consists of several main categories, these categories are identified according to the first three digits of the principal diagnosis ICD-9- CM code and are taken into account. The APR-DRG measure of severity has been shown to be a powerful i 'y predictor of mortality. However, this measure is not directly used as a risk- adjustment factor. First, since it includes all the relevant diagnoses reported at discharge, regardless of whether they are developed before or after admission, 1 1 APR-DRG is a system of classification of diseases with severity categories, patented by 3M Health Information Systems. This severity measure is not available for most o f the discharges that occurred in 1990 and 1991. APR-DRG system defines the severity as the "extent of physiologic decompensation or organ system loss of function". Using information like principal diagnoses, procedures, multiple comorbidities, and age, it provides four severity-of- illness and risk-of-mortality subclasses within each DRG (Diagnosis-Related Group). See www.3Mhis.com and the 3M’s APR-DRG Software’s brochure. 1 2 See Romano and Chan (2000) for evidence regarding AMI patients. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 21 it may include some “preventable” complications as well as “natural” comorbidities. Secondly, given that the Medicare reimbursement system is based on the patient’s diagnosis group, hospitals have an incentive to over- 1 o report complications. This problem, known as upcoding or DRG creep, may occur differently among hospitals with different ownership status.1 4 In this paper, a severity variable constructed from the APR-DRG classification is used. This variable represents whether the patient’s severity is higher than the average severity category within the same hospital-year. Since this measure only represents the variation within hospital-year, differential upcoding cannot create any bias in the estimation of ownership effects. 1.4.3. Measures o f quality One of the most commonly used outcome measures of quality is risk- adjusted in-hospital mortality. There are several validation studies suggesting that adjusted mortality rates can be used as a measure of hospital quality. Thomas et al. (1993) studied the in-hospital mortality rates for ten diagnostic groups of patients separately. For many but not all of these groups, the results showed a significant relationship between risk-adjusted in-hospital mortality and the hospital's quality as evaluated by peer reviews based on explicit and 1 3 See for instance Psaty et al. (1999), Silverman and Skinner (2001) and also Foundation for Health Care Quality (1997) section 2. 1 4 Studying Medicare inpatient claims between 1989 and 1997 for pneumonia patients, Silverman and Skinner (2001) provide evidence suggesting that upcoding is more common among FP hospitals and also those NP hospitals that converted to FP type. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 22 implicit process criteria. The strongest evidence of validity was obtained for cardiac diseases, which may suggest less selection for this kind of patients. Kahn et al. (1990) found similar results using mortality rates 30 days after admission. Significant relationship of risk-adjusted 30-day mortality and several process measures of quality was found in four out of five examined conditions. Based upon these studies, the risk-adjusted in-hospital mortality probability is adopted as the main measure of quality in this paper. Like most other health outcomes that potentially have some information about hospital quality, mortality is a rare outcome and sometimes takes a long time to manifest, making its measurement difficult. As the numbers in table 3 indicate, the selected diagnoses have relatively high in-hospital death rates. Moreover, in most of these groups, a relatively large part of deaths occur in acute-care hospitals. For instance during 1998 in California, 29.1% of 17,422 deaths caused by AMI and more than a half of deaths caused by CHF occurred in short-term hospitals. This rate is about 14% for lung cancer patients. Rosenthal et al. (2000) report a strong correlation between 30-day (post-admission) mortality rates and in-hospital death rates for a sample of 13,800 CHF patients. They also provide evidence suggesting that the small differences in hospital ranking caused by replacing in-hospital death rates by 30-day rates are not biased by differences in discharge practices. In-hospital mortality rates can Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 23 therefore be representative of death probability. Moreover, the robustness of this measure to potential differences in discharge and transfer practices across hospitals is studied. This issue is discussed in more detail later. Another outcome measure used in this paper is the risk-adjusted probability of early readmission of AMI patients following discharge from a hospital. Usually readmission within a short period (typically one month) after a prior discharge has been considered as an undesired outcome that could be avoided by the original provider (Thomas and Holloway, (1991) and Carey and Burgess (1999)). In some cases readmission within longer periods of time was used as an indicator of poor quality (Cutler, 1995). However, most of preventable readmissions occur within 10 days of a previous discharge (Frankl et al, 1991). Several authors have found that the variations in readmission probability are related to patient’s clinical conditions rather than hospital quality (Thomas and Holloway (1991), Thomas (1996), and Ludke et al. (1993)). However, a readmission for an AMI patient may imply another heart attack, thus a significant increase in patient’s mortality risk. A second related quality measure is based on the probability of future re-hospitalization of AMI patients with a CHF complication, given that the complication did not develop during the first hospitalization. CHF is the most frequently observed complication among AMI patients (Vaccarino et al., 2001). Both of these Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 24 measures are based on readmissions within one, two, three, and six months after a discharge. Process measures of quality usually require relatively detailed clinical data or physicians’ review. However, in some cases certain procedures are generally believed to be effective. For instance, the use of cardiac catheterization (CC) is commonly recommended as a diagnostic procedure for heart patients. Depending on the result of catheterization, a revascularization procedure like coronary bypass surgery or angioplasty can be performed to improve blood circulation and prevent further damage. The adjusted probability of using CC on two groups of heart patients (AMI and CHF) is therefore considered as an indicator of quality. 1.5. Methods The empirical model used in this paper can be formulated as follows: m ijt = P X y t + yZjt +rYt +A.j+ sijt (1.4) where myt is the quality indicator of patient i hospitalized in hospital j in year t. The quality indicators are binary variables representing the patient’s mortality outcome, whether the patient was readmitted after discharge, or whether certain procedures were used during the hospitalization. Xyt is the vector of patient’s characteristics including age, gender, race and their pair-wise interactions. This vector also includes a constructed severity index as defined in the previous section, as well as additional dummies Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 25 for CHF and TA samples, which represent the main categories of these conditions. These variables account for 3 and 16 dummies respectively for CHF and TA groups. Zjt is the vector of hospital characteristics, which includes the ownership dummies for FP, NP, and public facilities. Yt is the vector of year dummies and Xj is the hospital-specific fixed effect. Finally £yt is an i.i.d. random error that represents the unobserved heterogeneities among patients, hospitals, and years. Since the effect of risk factors differs across different health conditions, the model is estimated separately in four diagnostic groups. The standard errors are corrected for the correlation of errors within hospital- year groups. It should be noted that due to the presence of hospital fixed effects the model in equation (1.4) captures only the effect of those hospital characteristics that change over time. The profit status effects are therefore driven only by hospitals that converted from one form to another. Moreover, the symmetry assumption that the effect of conversion from one status to another is reversible is implicit in the model. This means that for instance, conversion from NP to FP form has the opposite effect of conversion from FP to NP status. This assumption is tested using a model that allows for different effects of opposite conversions. This model is similar to equation (1.4) with the exception that instead of profit status dummies, two conversion indicators are Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 26 used. These binary variables respectively indicate whether the hospital has converted from FP to NP or NP to FP in the past. The model is estimated over all private hospitals and for each of the four diagnoses separately. The results (not shown here) suggest that the sum of the coefficients of conversion indicators is not significantly different from zero, thus not rejecting the symmetry hypothesis. However, both coefficients are insignificant. The symmetry assumption is also tested using a model without hospital fixed effects. In this model I control for the two conversion indicators as explained above and two other binary indicators respectively representing whether the hospital has been FP and NP throughout the sample period. The results are similar to the previous results in that the symmetry hypothesis cannot be rejected but the coefficients of conversion indicators are generally insignificant. Although the weak statistical significance may be related to the loss of information (using only half of the conversions for each coefficient), the above tests do not give a conclusive evidence of symmetry. However, to the extent that similar reasons (such as financial problems) motivate both conversions, the symmetry assumption will help to identify the real effects of status by pooling information across both directions. Moreover, in the studied sample the number of conversions from NP to FP is almost the same as those in the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 27 opposite directions. Therefore, similar to Sloan (2001) and Carey and Burgess (1999) I assume that conversions have symmetric effects on quality. 1.5.1. Patient selection Patient level data can be used to estimate hospital-specific measures of quality. However, these measures are affected by a variety of confounding factors such as caseload characteristics. Any unobserved systematic selection of particular types of patients to a category of hospitals creates an estimation bias in the results. An unbiased estimation of ownership effects on hospital performance requires sufficient adjustment for the unobserved risk factors that potentially vary across different sectors. Patients with different severity may favor hospitals in one sector over another. Hospitals may also have different incentives in targeting certain groups of patients or avoiding “costly” or more severely ill patients to make more profits. To the extent that patients go to the closest hospital and hospital location does not change with ownership, hospital fixed effects (A .j) can help to avoid selection effects. The emergency nature of diagnoses like AMI can help in this regard. Similarly, patients admitted through ER may be less affected by selection. Comparing the results between such patients and the whole sample and also across different diagnostic groups can help us to understand the extent to which patients’ choice of hospital can bias the estimations. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 28 Another source of selection bias might be the fact that hospitals have different shares of ER admissions. One can expect these patients to be more severely ill. The data used in this paper show that FP hospitals are more likely than other hospitals, to close their ER services or to have practically no ER admissions. However, there is no significant evidence of any association between ER closures and conversion in ownership status. To see if hospital practices regarding ER services can create bias in the results, the relation between share of ER admissions and conversions is studied. Moreover, restricting the samples to ER patients and controlling for whether a patient is admitted through ER can help identify the extent of such biases. There is a possibility that certain types of hospitals (for instance FP ones) get rid of their most severe patients by transferring them to other hospitals (say NP ones). There is also a possibility that for instance FP hospitals discharge their patients earlier than is required. In this case, keeping other factors constant, the length-of-stay of patients would have to be lower in FP hospitals. In addition to studying the variations in length-of-stay across hospitals, the conditional in-hospital mortality risk is analyzed using a proportional hazard model that takes both mortality outcome and the length of hospital stay into account. In this model, longer hospitalizations are considered as an indication of lower hazard rates. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 29 The theoretical model presented in this paper predicts certain patterns of severity variation across hospitals with different levels of cost-efficiency. According to this model, the more severely ill are more likely to choose more efficient hospitals. The available data on clinical severity of illness (as measured by APR-DRG classification) help us to study the variation of severity across different sectors. However, for the reasons stated above, particularly the possibility of differential upcoding, these estimates should be considered with caution. To the extent that upcoding practices are hospital- specific and do not change with conversion, they are captured by hospital fixed effects. Moreover, as stated by Silverman and Skinner (2001), upcoding is most likely limited to a few conditions where there is clinical uncertainty about the appropriate diagnosis. AMI is a prominent example since hospitals can increase their revenues by falsely reporting CHF complication. For instance, Medicare reimbursements increase about 40% if the patient has complications (Psaty et al., 1999). My regressions (not shown in the paper) indicate that after controlling for the usual risk factors, there is no significant evidence of any association between the rate of reported CHF complications and conversions. However, since there may be other unknown types of upcoding, the analyses based on reported severity can only provide suggestive evidence. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 30 1.5.2. Treatment practices According to the theoretical model presented above, more efficient hospitals are likely to use more intensive treatments, suggesting shorter hospital stays and more intensive utilization of resources in these hospitals. The association between conversions and length-of-stay and the intensity of treatment is studied. The volume of services delivered to the patient is proxied by the list prices reported in discharge abstracts. Another interesting question is whether the differences across hospitals of different types result from differences in the volume of resources expended or from a more efficient use of resources. To study this issue, I look at mortality differences in models that control for the total expense of a given hospitalization. 1.6. Results This section presents the main results of this paper. The effects of ownership status on mortality probability, admissions from ER, hospital stays, and other measures of performance are discussed respectively. 1.6.1. Mortality Table 1.4 gives the estimated effects of conversion on in-hospital mortality in models that do not include hospital fixed effects. The results are also given separately for converted and non-converted hospitals. The results for AMI in the first and second columns are consistent with McClellan and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 31 Table 1.4. Mortality regression without controlling for hospital fixed effects Mortality Probability Entire Sample Non-converted hospitals Converted hospitals Acute Myocardial Infarction Non Profit Public R-square Sample size -.0090* (.003) -.0024 (.004) .025 290,059 -.017* (.003) -.0040 (.004) .026 253,777 .025* (.0063) .0045 (.0074) .024 36,282 Congestive Heart Failure Non Profit Public R-square Sample size .0023A (.0013) .0017 (.0019) .0077 565,536 .0020 (.0014) .0028 (.0023) .0079 491,903 .0050A (.0026) -.0005 (.0031) .0069 73,633 Malignant Lung Cancer Non Profit Public R-square Sample size .013A (.0077) .037* (.0095) .0079 71,141 .011 (.0083) .027* (.011) .0073 63,265 .019 (.013) .069* (.018) .016 7,876 Injuries d u e t o Traffic Accidents Non Profit Public R-square Sample size .016* (.0041) .040* (.0062) .068 35,708 .017* (.0047) .047* (.0075) .071 29,979 .017A (.0096) .021* (.0092) .061 5,729 For-Profit status is considered as base line (coefficient zero). Standard errors are given in brackets. * indicates significant at 5% level and A indicates significant at 10% level. Other explanatory variables are not shown. Severity dummy is excluded from explanatory variables. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 32 Staiger (2000) and Shen (2001), in that they suggest a lower probability of mortality in NP hospitals. In contrast, regressions using the same group but restricting the sample to converting hospitals result give the opposite result, suggesting that in these hospitals, FP status was associated with lower mortality rates. These results suggest that converting hospitals may consist of a selected sample of hospitals. These contrasting results are however limited to the AMI group. In other groups, FP hospitals have generally lower mortality rates than NP ones regardless of hospital sample. An interesting result is the remarkably high lung cancer mortality rates of public hospitals—there is a difference of 2 to 7 percent compared to private hospitals. Since the mortality outcome of this group of patients may be beyond the hospitals’ control, these relatively large differences suggest a high degree of selection among these patients. One explanation may be that private hospitals, particularly FP ones, avoid lung cancer patients at their terminal stage. Traffic accident patients also show large differences in mortality rates by hospital type. As discussed earlier, patient selection is relatively strong among these patients, and higher mortality in public and NP hospitals may only reflect the admission of sicker patients to these hospitals. This appears to be consistent with the higher participation of public hospitals in trauma Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 33 networks, suggested by their disproportionately high share of traffic accident admissions.1 5 The results given in the table highlight the importance of selection issues and the crucial importance of controlling for severity. These considerations motivate the use of hospital fixed effects models below. Moreover, these results suggest that due to potential variations among different diagnostic groups, estimates based on one condition should be considered as overall measures of hospital quality. The main results regarding the effect of conversions on mortality are presented in table 1.5. Here, hospital fixed effects are included as explanatory variables, so that the ownership effects are driven only by converting hospitals. The first column gives the regression results without controls for severity. These results suggest that FP status is associated with lower mortality rates for AMI and CHF patients. As it can be seen for the AMI and CHF groups, the results are generally similar to those obtained without hospital fixed effects on converting hospitals (last column of table 1.4), suggesting that there is relatively little selection of patients to hospitals. However, for LC and TA patients the inclusion of fixed effects dramatically changes the results. These findings suggest that selection 1 5 As table 1.4 shows the share of TA hospitalizations in public hospitals (22%) is almost twice as that of the other three diagnostic groups. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 34 Table 1.5. Mortality regressions accounting for hospital fixed effects Model: I II III IV V Non Profit .013* (.006) .016* (.007) .013* (.007) .020* (.007) .016* (.0066) o < + - ( & *3 c 3 o o Public Severity ER .0095 (.0090) .017A (.0093) .21* (.002) .013 (.009) .21* (.002) .027* (.002) .024* (.009) .22* (.002) .015 (.009) .21* (.002) 3 a < c Charges ($106 ) - - - - -.099* (.037) R2 .032 .104 .105 .105 .103 Sample size 290,059 290,059 290,059 208,266 255,715 Non Profit .0044 (.003) .0071* (.003) .0071* (.003) .011* (.003) .0084* (.003) < D u Public -.0005 (.004) .0019 (.004) .0019 (.004) .0087* (.005) .0033 (.004) c 3 u- S < L > X Severity - .088* (.001) .088* (.001) .090* (.001) .071* (.001) .s < L > 6 X ) C 3 ER - - .0004 (.0008) - - O a Charges ($106 ) - - - - 1.2* (.07) R2 .013 .040 .040 .039 .052 Sample size 565,536 565,536 565,536 380,517 518,251 I: Regressions without control for severity dummy. II: Regressions with control for severity dummy. Ill: Regressions controlling for ER admissions. IV: Regressions exclusively on ER admissions. V: Regressions controlling for patient’s hospital expenses. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 35 Table 1.5. Continued Model: I II III IV V Non Profit -.021 (.013) -.020 (.014) -.018 (.013) -.087* (.026) -.019 (.014) * - < L > O Public .035 (.025) .033 (.026) .037 (.027) -.022 (.040) .035 (.026) S 0 t M } 41 1 .§> Severity ER - .12* (.004) .11* (.004) .11* (.004) .11* (.007) .12* (.004) C 3 S Charges ($ 106) - - - - .23* (.064) R2 .049 .070 .085 .050 .073 Sample size 71,141 71,141 71,141 23,372 63,344 Non Profit -.0037 (.0087) -.0011 (.008) -.0012 (.0083) -.0059 (.0097) .0006 (.009) o n 2 o o Public .0083 (.013) .0051 (.014) .0041 (.014) .0067 (.015) .006 (.013) < o ££ C 3 U t H Severity - .067* (.004) .067* (.004) .073* (.004) .059* (.004) 2 o > ■ § o n ER - - .014* (.003) - - © ’C 3 Charges ($ 106) - - - - .41* (.07) R2 .090 .103 .103 .106 .114 Sample size 35,708 35,708 35,708 29,227 33,711 For-Profit status is considered as base line (coefficient zero and hazard ratio 1.0). Standard errors are given in brackets. * indicates significant at 5% level, and A indicates significant at 10% level. Other explanatory variables are not shown. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 36 is a much more important issue for these patients, as I argued above on a priori grounds. The results given in column II are obtained from a model that controls for severity within hospital-year groups. The first observation is that controlling for severity increases the model’s explanatory power (reflected in an increase in the R-square). The ownership effects also become larger when severity is taken into account. Another interesting observation is that in the AMI sample, ownership effects are not sensitive to whether the hospital fixed effects and the severity measures are taken into account. These findings suggest that selection bias is much less important for this group of patients than it is for others. 1.6.2. Robustness o f the results The validity of the lower mortality rates for AMI and CHF patients in FP hospitals is tested against several hypotheses. First, the ER is presumably an admission channel for the sickest patients. After conversion to FP status hospitals might close their ER service or downsize their ER-related services to avoid such patients. The regressions of probability of admission through ER (given in first column of table 1.6) show that conversion to FP status is associated with a significant decrease in the probability of admission of similar AMI and CHF patients through the ER. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 37 Table 1.6. Ownership effects on hospital operational characteristics Dependent variable: ER Length- of-stay (days) Log of expenses ($) Log of Per-day expenses ($) Waiting period before principal procedure (days) APR Severity index (1-4) (ordered logit) NP .085* (.018) .26* (.09) -.029 (.025) -.080* (.02) .20* (.081) -.026 (.06) AMI Public .12* (.03) .78* (.19) -.004 (.046) -.12* (.038) .35* (.14) -,15A (.079) R2 .159 .11 .32 .42 .060 .021 CHF NP .041* (.012) .12 (.11) -,042A (.024) -.051* (.02) .11* (.056) -.12* (.05) Public .074* (.025) .22 (.14) -,091A (.050) -.14* (.04) .074 (.12) -.27* (.10) R2 .093 .051 .27 .42 .064 .020 NP -.013 (.019) .36 (.26) .019 (.042) -.059* (.030) .14 (.099) -.18* (.066) o Public -.038 (.029) .39 (.45) -.13 (.075) -.20* (.064) .095 (.16) -.17 (.10) R2 .11 .091 .19 .41 .076 .024 NP .0096 (.021) -.026 (.41) -.052 (.053) -.072* (.029) .033 (.17) -.29* (.094) < H Public .075* (.031) .092 (1.3) -.11 (.099) -,082A (.045) -.11 (.41) -.12 (.18) R2 .198 .106 .383 .493 .045 .049 For-Profit status is considered as base line (coefficient zero). Standard errors are given in brackets. * indicates significant at 5% level. A indicates significant at 10% level. The sample sizes are approximately equal to the corresponding numbers given in table 1.4. Hospital fixed effects are included as explanatory variables. Severity dummy is controlled for in all the regressions except for the last column (APR severity regressions). Other explanatory variables are not shown. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 38 Column III in table 1.5 summarizes the results of mortality regressions controlling for whether the patient was admitted through the ER. These results show that the estimated ownership effects are generally robust to this change in specification. The next column gives the mortality regressions exclusively for ER patients. These results show that the ownership effects are stronger and more significant among ER patients. This general pattern is observed in all diagnostic groups and can be explained by relatively weaker selection effects for ER patients. The second hypothesis concerns the transfer and discharge policies practiced by different hospitals. The lower in-hospital mortality might be an artifact of transfer or premature discharge of costly patients. In fact, the regressions shown in table 1.6 (second column) suggest that FP status is associated with significantly shorter hospital stays especially for AMI patients. A proportional hazards model of in-hospital mortality was estimated in order to test this hypothesis.1 6 The starting time of being exposed to risk is assumed to be the admission day. The patients who were discharged alive are considered to be censored from the sample. The model can therefore take into account the differences in the length-of-stay across different hospitals. To avoid computational complexity this analysis is conducted on the sample of patients admitted to hospitals that converted at some point in the study period. 1 6 Similar results are also obtained using exponential and Weibull survival models. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 39 The hazard model is assumed to be identical for all patients but shifts across hospitals by hospital fixed effects. The results (not shown here) confirm the previous findings suggesting that FP status is associated with lower AMI and CHF mortality rates even after length-of-stay is accounted for. A more systematic and detailed discussion of the effect of discharge rates on hospital quality measures is postponed to chapter 2. Third, it might be the case that conversion in status resulted in an unobservable change in the severity of the patient caseload. Conversion to NP status might send a signal of “trust” and higher quality to consumers, and thereby attract more severely ill patients. This would result in lower mortality rates for FP hospitals. To answer this question I use the available measure of severity. The last column of table 1.6 gives a summary of estimation results from an ordered multinomial logit model of the APR-DRG severity index. These results show that FP status is generally associated with a more severe case mix in all four diagnostic groups. As was previously discussed, this measure of severity may be subject to reporting errors that are potentially related to hospital ownership status. However, the uniform pattern of these results suggests that they cannot be entirely attributed to upcoding, since the feasibility of upcoding differs across diagnostic groups. Moreover, the smallest estimated changes in reported severity occur among AMI patients, even though Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40 AMI is supposed to be one of the diagnoses most susceptible to upcoding (Psaty et al., 1999). Therefore, the variations in reported severity with conversion suggest that FP status is associated with a higher severity of illness. These results indicate that we are likely to under-estimate the negative effects of conversion to FP status on in-hospital mortality. Moreover, as it can be seen in table 1.6 (last column), FP status is also associated with higher degrees of severity among LC and TA patients, even though there is no change in mortality (as seen in table 1.5). Thus, it appears that conversion to FP status improves outcomes among these patients as well. Finally, the differences in mortality may be due to higher costs of treatment. As shown in table 1.6 (column 3), FP status is generally associated with higher expenses per patient in AMI and CHF groups. One may argue that better mortality outcomes are an indication of higher performance only if they do not cost too much. Column V in table 1.5, gives a summary of the mortality regressions controlling for per-patient expenses. These results show that conversion to FP status improves outcomes even when costs are controlled for. These models are consistent with the model presented in this paper, which predicts that more severely ill patients are likely to choose more efficient hospitals. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 41 1.6.3. Operational features Access to health services through the ER can also be considered to be an indicator of quality. One of the main concerns of critics of hospital conversions to FP form is the possible reduction in access to ER services. To investigate the potential relationship between ER closure and conversion, I studied the discharge data for all patients hospitalized between 1990 and 1998. I extracted the number of ER admissions and the fraction of total patients admitted through the ER for each hospital and year. The ER service is considered to be closed in a given year if there are no ER admissions during that year and if the number of ER admissions is above one percent of all admissions to the hospital during the previous year. Based on this definition, • • 1 7 there were 23 cases of ER closures in California during this period. Only two hospitals closed down their ER after conversion to FP status (during the same year). One of these two cases was a NP acute care hospital that converted to a FP long-term care facility. Long-term care facilities are excluded from all the estimations presented here. The other case is a small hospital with 64 beds that had about 776 admissions and admitted only 8.5% of patients from the ER in the year prior to conversion. In our data, FP hospitals on average have more than 100 beds and about a 25% share of ER admissions. Thus, there is only 1 7 Among these 23 cases of ER closure, nine were FP hospitals, 7 were NP, and 5 were public hospitals. None of these 21 hospitals converted in ownership status. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 42 one case of an outright ER closure in the 23 NP-to-FP conversions in the sample. However, conversions may reduce access to ER services even when ERs remain open. The first column of table 1.6 shows that conversion to FP status is associated with a significant decrease in the probability of admission through the ER for AMI and CHF patients. The ER admission rates in LC and TA groups do not change with conversions. The estimated effects of conversions on length of hospitalizations are given in table 1.6 (column 2). These results indicate that conversion to FP status results in shorter stays. However, as seen in column 4 of the same table, hospitals that convert to FP status deliver a significantly higher volume of services as reflected in higher per-day expenses for a given hospitalization. These results suggest that the shorter stays in FP hospitals are accompanied by more intensive treatment. Moreover, regressions of the length of the waiting period before the patient undergoes the principal procedure (also shown in table 6) suggest generally shorter waits and thus a more efficient use of resources. These results are consistent with the hypothesis that more efficient hospitals are more likely to use more intensive treatments. 1.6.4. Other measures of quality Regression of process-based indicators of quality is done for AMI and CHF groups separately. This analysis shows that hospitals that convert to FP Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 43 status are more likely to perform Cardiac Catheterization. However, the difference is only borderline significant for AMI sample. These results are given in table 1.7. The average rates of using CC are also given. The results are generally consistent with estimated mortality differences, and also suggest quality improvement with conversion to FP status. The probabilities of readmission and complication among AMI patients are analyzed using two different measures. The first measure is based on readmissions with AMI as a principal diagnosis. Readmissions within 1, 2, 3, and 6 months are considered. The estimation results are given in table 1.8. These results indicate no significant differences across hospitals of different forms. However, the coefficients are in the direction of lower readmission rates for hospitals with public and NP forms. Table 1.7. Cardiac Catheterization for heart patients Regression results Average rates (%) AMI CHF AMI CHF For-profit: 23,8 2.7 Non-profit: 34.3 5.1 Public: 25.7 3.9 Non Profit Public R-square -.014 (.0093) -.0085 (.022) .031 -.0019 (.0029) .0054 (.0070) .084 Severity dummy is controlled for in all the regressions. See also notes to table 1.6. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 44 Table 1.8. Readmission of AMI patients Readmission within: 1 month 2 months 3 months 6 months AMI Readmission Non Profit Public R-square -.0025 (.0039) -.011* (.0052) .0097 -.0014 (.004) -,010A (.006) .0095 -.0027 (.0045) -.0098 (.006) .0095 -.0020 (.0047) -.0084 (.0067) .010 CHF Complication Non Profit Public R-square -.012* (.0053) -.014* (.0057) .016 -.0093 (.0057) -.012 (.0078) .020 -.010A (.0061) -.013 (.0092) .023 -,011A (.0063) -.013 (.011) .028 The sample for AMI readxnissions includes 232,218 observations consisting of AMI elderly patients with valid ID who were discharged alive. The sample for CHF complications includes 153,606 observations consisting of AMI elderly patients with valid ID who did not have CHF complication originally and were discharged alive. Severity dummy is controlled for in all the regressions. See also notes to table 1.6. The second measure is based on the probability of CHF complication for AMI patients within 1, 2, 3, and 6 months after a discharge. CHF is the most common complication among AMI patients. All hospitalizations with CHF as principal diagnosis or as one of the four secondary diagnoses are considered as a complication. The results are also given in table 1.8. These results suggest that NP and public types are associated with lower complication rates. 1.7. Conclusions In California in the 1990s, conversions from NP to FP status were associated with significant reductions in in-hospital mortality rates among AMI Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 45 and CHF patients. These estimates are obtained from models that control for possible selection biases by including hospital fixed effects and controlling for the severity of patients’ illnesses. The results are even stronger when only admissions from the ER are considered, and they do not appear to be driven by changes in discharge or transfer policies following conversion. Hospitals that convert are also more likely to have higher per day charges, and lower waiting times before procedures are performed, suggesting that they supply services more intensively. Moreover, hospitals that convert to FP status have lower mortality rates conditional on charges, suggesting that conversion improves the efficiency with which services are delivered as well as the intensity of service provision. These results are consistent with a theoretical model in which more efficient hospitals supply services more intensively to sicker patients. On the down-side, although there did not appear to be a significant relationship between conversion to FP status and ER closures, conversion was associated with a reduced probability of being admitted from the ER. Finally, AMI patients were more likely to be readmitted with CHF complications following a conversion. These results suggest that hospital conversions to FP status have mixed effects, improving the intensity of services and reducing mortality while, at the same time, reducing access to emergency care and increasing the probability of some types of complications. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 46 2. ESTIMATING THE OUT-OF-HOSPITAL MORTALITY RATE USING PATIENT DISCHARGE DATA 2.1. Introduction One of the most commonly used measures of hospital quality is the risk-adjusted in-hospital mortality rate. This measure is a convenient measure in that it is readily applicable to most administrative data like Patient Discharge Data (PDD). Although a few studies suggest that the in-hospital 1 R mortality is a valid measure of quality, this result cannot be generalized . Validation studies need extensive clinical data that are usually expensive if at all available. A major concern is that the in-hospital mortality outcome only reflects the short-term effects of hospital treatment. Particularly, given the competitive pressures on hospitals the hospital stays are increasingly short. Moreover, the differences in discharge/transfer policies across hospitals may distort the estimations. Low-quality hospitals might discharge their most severely ill patients prematurely or transfer them to better hospitals. This can only be assessed if we follow patients over time. However such longitudinal data are expensive and hardly available or limited to certain groups of patients. 1 8 For instance Thomas et al. (1993) studied the in-hospital mortality rates for ten diagnostic groups of patients separately. For many but not all of these groups, the results showed a significant relationship between risk-adjusted in-hospital mortality and the hospital's quality as evaluated by peer reviews based on several process criteria. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 47 In his study of the effect of ownership conversion on the quality of care, Sloan (2001) reports that while the in-hospital mortality is not affected by conversions, the longer-term mortality probability increased as hospitals converted to for-profit status. These results suggest that hospitals with shorter stays may have higher mortality rates after discharge. While using short-term hospital data many studies use the inpatient mortality along with additional measures like the probability of re-hospitalization in the future with or without complication (c.f. Ho and Hamilton (2000), McClellan and Staiger (2000), and Cutler (1995)). However, several studies like Thomas (1996) and Ludke et al. (1993) have found that readmission risks are related to patient’s clinical conditions rather than hospital quality. Moreover, because of negative correlation between mortality and future readmission as suggested by McClellan and Staiger (2000), the estimations based on readmission usually do not provide additional information on hospital quality. Another approach used by Gowrisankaran and Town (1999) is a censored hazard model of in-hospital mortality. Such models (as used in chapter 1 of this dissertation) control for the variations in hospital stays across different hospitals1 9 . However, the information on out-of-hospital spells particularly from the patients who were re-hospitalized is not used. Moreover, these models could not give any information about the post-discharge mortality risk. 1 9 A similar model is used by Chou (2002) to estimate the quality of nursing homes. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 48 A complete measure of hospital-specific mortality risk should therefore include the risk of post-discharge mortality. Lacking long-term mortality data, a measure that can derive the out-of-hospital mortality risk from the data on out-of-hospital spells is useful. In this chapter hospital quality is assessed using both the in-hospital mortality rate and mortality rate after discharge. As a byproduct the discharge rate is also measured. Discharge rate is often confounded with the mortality rate in measures of hospital quality that are used in the literature. Another complication is that like most administrative data, the PDD are available to researchers in public use files. In such files certain variables that could reveal the identity of individuals are omitted. The administrative file from which the public use file is derived, contains all hospital spells for an individual that ended in any year from 1990 through 1998. In the public use file only the year and the month in which the spell started and its length in days are retained. If more than one spell started in that month, the multiple spells are reported in a random order. The plan of this chapter is as follows. Section 2.2 introduces the statistical model. It is shown that the out-of-hospital mortality rate is identified even if deaths after discharge are not recorded. The section also discusses the quality measures used in the literature. Section 2.3 provides more information Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 49 on the patient discharge data and on how the statistical model is implemented. Conclusions end the chapter. 2.2. Model In this section it is assumed that for a member of population the complete hospitalization history during the observation period [0,1] is observed. A complete hospitalization history consists of a sequence of hospital stays and spells outside hospital or equivalently, of a sequence of transitions between two states: hospitalize (H) and discharged (£>). A hospital spell ends if the patient is discharged or if she dies. An out-of-hospital spell ends if the patient is admitted or if she dies. The shortcomings of the available data are discussed in section 2.3. Death can considered as a transition to a third absorbing state. The time of death is observed only if the patient dies in hospital. The problem is to estimate the transition rates and in particular, the out-of-hospital mortality rate from the observed hospitalization records. The hospitalization record has multiple time scales: the observation times (0 is the start of the observation period), the duration of hospital or out- of-hospital spells (0 is the start of the spell), the time since the onset of the disease, calendar time, and age. In the sequel both observation and duration time are used. It is clear from the context which time scale is used. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 50 2.2.1. In-hospital mortality and discharge rates A hospital spell is denoted by tn■ A hospital spell ends with the death of the patient with intensity //#(/) or with the discharge of the patient with intensity A ,D {t). A hospital spell could also end with the transfer of the patient to another hospital. This could be considered by introducing a transfer intensity. Here, only one hospital spell is considered and there is no distinction between discharge and transfer. Hence, Xo(t) is a weighted average (with weights depending on the hospital spell) of the discharge and transfer densities. Let Dh be 1 if the spell ends with discharge and 0 if it ends with the death of the patient. The joint distribution of tn, Dh has the following pdf: = (2.1) t t with = j juH(s)ds and AD(t) = jx D(s)ds . f.in and Xd are assumed to be 0 0 piecewise constant over k intervals 0= t0 < tj < ...< Ai < tm a x where tm a x is the longest hospital stay. j U h and Xd are also functions of independent variables like patient and hospital characteristics. An exponential form is assumed. If X is the vector of independent variables, these rates can be written as: = (2.2) i~ \ ^ ( 0 (2.3) i~ \ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 51 where I(.) is the indicator function, //,■ and /I, are constant ratios corresponding to interval (/,_/, t( ] with ju o =Ao = 1, and y and [3 are the vectors of coefficients corresponding to the independent variables including an intercept. The pdf given in equation (2.1) is the basis for the likelihood function for the in- hospital spells. 2.2.2. Out-of-hospital mortality and hospitalization rates For the identification of the out-of-hospital mortality rate the spell spent outside hospital denoted by tD is considered. This spell starts at the time of discharge from the hospital. It ends if the patient returns to the hospital (not necessarily the same hospital) or if she dies. However, the death is not observed. Let 2# denote the hospitalization rate and Ho the mortality rate outside hospital. These rates may depend on the time since the last hospitalization t. This may be due to dependence on a number of time scales such as age, time since the onset of the disease, etc. For ease of exposition it is assumed that this spell starts at time 0 and that it is censored at time T. the distribution of to is mixed discrete-continuous with a positive probability that to -T. To show this consider for t<T: Pr(tD > 0 = + j’ {iD(s)e-A 'As)-M °(s)ds (2.4) 0 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 52 t i with AH{t)= ^XH{s)ds and MI)(t)= ^/uD(s)ds. The first term on the right- 0 0 hand-side is the probability that by t neither a death nor a hospitalization has occurred. The second term is the probability that during [0,f] the individual has died. In this case since deaths outside hospitals are not observed, the observed spell is still in progress. In fact for all patients who die before re hospitalization the observed spell to is of infinite length. This means that the distribution of to is defective and the probability of observing an infinite spell is the average of the probability of death before hospitalization, where the average is computed over the duration of the latent out-of-hospital spell, that Q U is: Pr(rD = co) = | fj.D (s)e~ hfI(s)-M £ ,(s)(A , If the observation period is finite, to is observed if tD <T. Otherwise the event to>T is observed. Define Du as the indicator of the event tu<T. The probability density of to given Dd =1 is written as: f ( t | D d = 1) = , t<T (2.5) Moreover: i Pt(Dd =0) = Pr(tD >T) = + j ^ Dis)e~A ’ AshM D (s)ds (2.6) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 53 The pdf given in equation (2.5) and the probability given in equation (2.6) are the basis of the likelihood estimation of the out-of-hospital mortality and re hospitalization rates, j u j j and Ah are assumed to be piecewise constant over k intervals 0 = Tg<Tj< ...< 7*. i < 7* = T. These rates can be written as: Mo(0 = el,xf d MMT,-l « ± T ,) (2.7) /= ] (2 .8 ) /= ! where H i and A, ■ are constant ratios corresponding to interval (7)./, T,] with fio =Ao = 1, J5 and y are the vectors of coefficients. To show that both hospitalization and mortality rates are identified consider first the special case where both rates are constant over time. In this case the conditional pdf in equation (2.5) reduces to: (A + u \e~ iX H + ! ' lD )t f ( t \ B D=l) = — ~ ,t<T (2.9) which is the pdf of a truncated (at 7) exponential distribution with parameter k= iid+Ah- Hence, from the distribution of spells that end in hospitalization the sum of mortality and hospitalization rates is identified. Moreover, the probability of re-hospitalization before T is: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 54 Since ic=jun+AH.is identified from (2.9), Ah is identified using the probability given in (2.10). The joint distribution of to,Do has the following pdf: ' AH e ^ T^ D D Ah + fJ .D Ah + flD (2.11) / The above argument can be extended to piecewise constant rates /& > (£ ) and Adf). It suffices to first censor the out-of-hospital spells at 7j (the first interval). The rates are constant over the interval thus identified using the censored spells. The spells that end with a hospitalization in the interval identify the sum of mortality and hospitalization rates and the fraction of spells that are censored identity the rates separately. Next, consider the out-of- hospital spells that end with a hospitalization in the second interval (T),^]. It can be shown that the distribution of these spells is such that fo-Ty has a truncated (at T2- T 1) exponential distribution with a parameter that is the sum of mortality and hospitalization rates on the second interval. Hence this distribution identifies the sum. The hospitalization and mortality rates are identified from the fraction of spells that are censored at I). This argument can be repeated for the remaining intervals. 2.2.3. Measures used in the literature In this section the quality measures used in the literature are discussed using the proposed model. The measures can be divided into three categories: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 55 in-hospital mortality outcome, mortality outcome within a given period after admission, and readmission within a given period after discharge. For ease of exposition it is assumed that all transition rates are constant. A large number of papers used the mortality outcome at discharge as a measure of hospital quality. Examples are Ho and Hamilton (2000), Thomas et al. (1993), and DesHamais et al. (1991). This measure can be written as a function of in-hopsital mortality and discharge rates: Pr(Z)„=0) = T i i t — (2.12) " D t^ H It can be shown that the in-hospital death probability is increasing in hh and decreasing in Xd. To the extent that discharge practices differ across hospitals, this measure cannot be used as a hospital-specific mortality. An alternative used by Geweke et al. (2001) is the inpatient death probability within 10 days after admission. This in-hospital mortality probability within period t after admission can be written as: Pr{Dh = 0,tH <t) = (2.13) XD+/iH In this case depending on the chosen value of t, the death probability can be decreasing or increasing in ////. Therefore, even assuming a constant discharge rate across hospitals, this cannot be used as a measure of hospital-specific mortality. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 56 Perhaps the most commonly used measure of hospital mortality is the mortality probability within a given period after admission. Hartz et al. (1988), Carey and Burgess (1999), Keeler et al. (1992), Jencks et al. (1988), Kahn et al. (1988), and Chen et al. (1999) used 30-day mortality probability, whereas Sloan et al. (2000), McClellan and Staiger (2000), and Kessler and McClellan (2001) used longer-term mortality probabilities up to one year after admission to a hospital. These deaths may occur inside hospitals or after discharge. This measure may seem appealing because it can measure a relatively long-term outcome that is seemingly independent of discharge rates. The probability of death within t days after admission can be written as the sum of the probabilities of the in-hospital and post-discharge death before t, that is: ! \/uH e u'> n i!')sds + 0 0 It is easy to show that such measures are affected by discharge and hospitalization rates. Another frequently used measure of quality is the re-hospitalization probability within a given period after discharge. Ho and Hamilton (2000) used 90-day readmission rates for heart attack patients. DesHamais et al. (1991) used 30-day readmissions and Carey and Burgess (1999) considered the re admission probability within 14 days after discharge. Kessler and McClellan Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 57 (2001) use the readmissions up to one year after a first hospitalization. The readmission probability within t days after discharge can be written as: Pr(tD < 0 = ——— (2.14) The problem with this measure is that for short readmission periods (small t), it is not increasing in XH , and for relatively large periods the correlation between readmission risk and hospital quality is low (see Ludke et al. (1993)). Moreover, as it can be seen this measure depends on the out-of- hospital mortality rate. In fact, for short periods (small t) this measure is a decreasing function of / ud- In cases where the out-of-hospital mortality is not observed, small rates of re-hospitalization may be associated with high mortality rates, hence not necessarily a higher hospital quality. The above problems provide an explanation to why the readmission measures of quality as used in the literature are inconsistent with other measures of hospital quality (see Thomas (1996) and Ludke et al. (1993)). Ettner and Hermann (2001) used the readmission within 30 days after discharge for psychiatric patients. Given that mortality rates are quite low for these patients, the re-admission measure may be appropriate. Assuming that ju d is close to zero, the probability given in (2.14) can be simplified as: (1 - e , which is a non-decreasing function of Xh, and therefore can be used as a proxy for the re-hospitalization rate. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 58 2.3. The patient discharge data 2.3.1. Description of the data The data used in this chapter are extracted from the two data sets (California Hospital Discharge Data and Disclosure Data) used in chapter 1. These data sets are described in section 1.4. The population considered in this chapter are all individuals of 65 years of age or older who were hospitalized during 1990-1998 with Acute Myocardial Infarction (AMI) as their principal 9 A diagnosis and who were in an initial episode of treatment. As it is shown in chapter 1, AMI is an acute condition and these patients are less subject to selection problems. The above age group is chosen because all these patients benefit from Medicare and are less likely to be rejected by hospitals. Moreover, the identification of post-discharge mortality relies on the assumption that an out-of-hospital spell ends in re-hospitalization or death. A third possibility is that the patient leaves the state of California. The migration is less likely for the elderly patients with an acute condition. To protect the privacy of the patients the exact dates of admission and discharge are omitted in the public use file. Only the year and month of admission and the length-of-stay in days are retained. Moreover, if more than one spell start in a given month, the spells are not reported in the order they 2 0 AMI in an initial episode of treatment is coded as 4 10.x 1 according to the International Classification o f Diseases, 9th version, Clinical Modification. Initial episode o f treatment is typically referred to the care given to the newly diagnosed myocardial infarction patients. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 59 occurred, but in a random order. Therefore the data consist of a sequence of hospital spells together with the month in which each of these started. The first month is the month in which the initial hospitalization for AMI occurred. The second and later hospitalizations need not be for AMI and can be for any condition. Hospital spells of a given patient can be in several hospitals. However, the first hospitalization is more likely to represent the effect of medical care on the health outcome. Moreover, the first hospitalization is relatively less affected by selection, because in their later hospitalizations patients may get experienced or have more time to choose their hospital. Therefore to avoid unnecessary complication for the patients with multiple hospitalizations only the first in-hospital and out-of-hospital spells are considered for the estimations. The estimations for the hospital and out-of-hospital spells are done separately. 2.3.2. Implementation o f the model The estimation of hospital spells is straightforward and the joint distribution of tn, Dh with piecewise constant rates can be directly derived from equation (2.1) using the expressions in (2.2) and (2.3). For the out-of- hospital spells, because of the limitations of the data the exact length of spell is not known. Rather, the spell can be specified with its lower and upper bounds. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 60 For the spells that do not end in re-hospitalization the contribution to the likelihood function is given by equation (2.6) using the expressions in (2.7) and (2.8). The end of observation period (T) is a patient-specific variable. T in days is given by: T = 30.5(M + l - m l) - ^ t m (2.15) i= i where M is number of the months in the sample period (1990-1998), m\ is the number of months before the first AMI admission (including the admission month), k\ is the number of hospital spells that started in the same month as the first AMI admission, and tH i are the length (in days) of the hospital spells that started in month mi,with i - 1, 2, ..., k\. The spells that end in re-hospitalization to is specified by the following lower and upper bounds: = 30.5(m2 -m , - l ) - m a (2.16) = 30.5(m2 +1 - m,) - tm - f £ Tm - max {t m; i = 1,..., k2} (2.17) where m2 is the number of months before the month of re-hospitalization , k2 is the number of hospital spells that started in that month, and xm are the length (in days) of the hospital spells that started in month m2,with z'=l, 2 ,..., k2. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6 1 The spells that end in re-hospitalization make the following contribution to the likelihood function: 4 “ p Pr(t”f <tD <O = | XH(s )e h»(s)-U D (s)ds (2.18) .in f lD where the integrals Au(t) and MD (t) are obtained using the expressions in (2.7) and (2.8). 2.4. Estimation results The data on the first reported hospital spell are used to estimate the in- hospital mortality rate and the discharge rate. In the observation period 1990- 1998, 209,362 elderly (65 years or older) individuals were hospitalized for the first time for AMI. The summary statistics are given in table 2.1. The average hospital spell is about 6.5 days and about 14% of the spells end with the death of the patient. For 177,138 patients from this sample the out-of-hospital spell are calculated. Note that the patients who are dead at discharge are excluded. About 66% of these patients were re-admitted after their first hospitalization and before the end on the observation period. For these patients the lower bound of out-of-hospital varies from 0 to about 3,000 days (with an average of 299 days) and the upper bound varies between 2 and 3,100 days (with an average of 355 days). Table 2.2 gives the summary statistics for out-of-hospital spells. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 62 Table 2.1. Sample statistics for hospital spells (N=209,362) Mean Standard deviation Hospital stay (days) 6.49 5.89 Discharged alive .86 .35 Non-Profit hospital .73 .44 Public hospital .12 .32 Number o f beds 280 166 Male .54 .50 Black .046 .21 Age 70-74 .24 .43 Age 75-79 .22 .41 Age 80-84 .17 .38 Age 85 + .16 .36 Moderate severity .40 .49 Major severity .28 .45 Extreme severity .19 .39 Table 2.2. Sample statistics for out-of-hospital spells (N=177,138) Mean Standard deviation Re-hospitalized before the .66 .47 end of observation period Lower bound of out-of- 299 467 hospital stay (days) Upper bound of out-of 355 469 hospital stay (days) Spell until the end of 1529 893 observation period (days) Non-Profit hospital .74 .44 Public hospital .12 .32 Number o f beds 282 166 Male .55 .50 Black .047 .21 Age 70-74 .25 .43 Age 75-79 .22 .41 Age 80-84 .16 .37 Age 85 + .14 .35 Moderate severity .44 .49 Major severity .27 .45 Extreme severity .13 .34 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 63 Several specifications are considered. In the first analysis, hospital fixed effects are excluded from the independent variables. Discharge and in- hospital mortality rates are assumed to be piecewise constant over 6 intervals: 0 to 2 days, 2 to 4, 4 to 6, 6 to 10, 10 to 14, and more than 14 days. Table 2.3 gives a summary of these results. The calendar year effects (not shown in the table) indicate that there is no significant trend in the mortality rate, but there is a strong upward trend in the discharge rate. Both severity and age have a positive effect on mortality and a negative effect on the discharge rate. The significant changes in transition rates over the intervals show that the rates are time-variant. For instance the mortality rate in the first two days of the spell is significantly higher. The crucial importance of the immediate stabilization of AMI patients is a well-known medical fact. Hospital size has a significant effect on both mortality and discharge rates with large hospitals having lower rates. These results suggest that compared to For-Profit (FP) hospitals, Non-Profit (NP) hospitals have a lower mortality and a higher discharge rate. The significant variation of discharge rates across hospitals with different ownership status supports the concern that lower in-hospital mortality rates may be associated to higher discharge rates. However, to the extent that higher discharge rates do not lead to higher chances of post-discharge mortality, the lower discharge rates do not indicate lower quality. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 64 Table 2.3. Mortality and discharge rates for hospital spells (without hospital fixed effects) Model I Model II MLE Standard error MLE Standard error Discharge rate: Non-Profit hospital .026 .007 .008 .008 Public hospital .044 .009 .043 .01 Number of beds -.74 .016 -.70 .017 Male .045 .005 .02 .005 Black -.018 .011 .003 .01 Age 70-74 -.081 .007 -.038 .008 Age 75-79 -.16 .007 -.074 .008 Age 80-84 -.20 .008 -.10 .009 Age 85 + -.22 .008 -.09 .009 Interval 2 to 4 days .85 .009 .88 .009 Interval 4 to 6 days 1.33 .008 1.42 .009 Interval 6 to 10 days 1.41 .008 1.62 .009 Interval 10 to 14 days 2.1 .01 2.37 .01 More than 14 days .30 .01 1.0 .01 Moderate severity - - -.36 .008 Major severity - - -.91 .009 Extreme severity - - -1.7 .01 Mortality rate: Non-Profit hospital -.068 .016 -.023 .018 Public hospital -.0095 .022 .024 .025 Number of beds -.69 .038 -.92 .044 Male -.062 .012 -.046 .013 Black -.057 .029 -.097 .032 Age 70-74 .25 .02 .17 .02 Age 75-79 .47 .02 .31 .02 Age 80-84 .76 .02 .60 .02 Age 85 + 1.0 .02 .78 .02 Interval 2 to 4 days -.93 .02 -.94 .02 Interval 4 to 6 days -1.1 .02 -1.2 .02 Interval 6 to 10 days -1.0 .02 -1.4 .02 Interval 10 to 14 days .093 .03 -.31 .03 More than 14 days -1.1 .02 -1.8 .03 Moderate severity - - .75 .06 Major severity - - 2.05 .06 Extreme severity ” ~ 2.8 .05 Calendar year dummies are included in the model but are not reported in the table. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 65 The estimation results for the out-of-hospital mortality and re hospitalization rates are given in table 2.4. Hospital fixed effects are not controlled for. Transition rates are assumed to be piecewise constant over 2 intervals: 1 day to 2 months and over 2 months21. The year effects (not shown in the table) show that both hospitalization and mortality rates have grown up significantly during the 1990s. The excessive fall of both rates at the cut-off point (two months) suggests that the time-variations of the transition rates are too large to be controlled by only two intervals with constant rates. It should be noted that for many patients the out-of-hospital spell is longer than a few years. In such long periods the mortality risk may increase because of age. Therefore, a higher number of intervals is needed for more accurate results. A surprising result is that both severity and age have a positive effect on the re hospitalization rate and a negative effect on the mortality rate. This result seems consistent with the lower discharge rate for older and more severely ill patients. It may also point to the imperfections of our analysis. Moreover, because the spells in the first month are reported in a random error, the coefficients of hospital characteristics and also severity are biased due to measurement error. However, the results do not provide any evidence that the higher discharge rates result in higher out-of-hospital mortality. 2 1 Other cut-off points (1 month, 3 months and 6 months) are also used in separate regressions. The results show that the estimations are not sensitive to the cut-off point. Since more intervals complicate the integration of the pdf the number of intervals is kept at two. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 66 Table 2.4. Mortality and re-hospitalization rates for out-of-hospital spells (without hospital fixed effects) Model I Moc el II MLE Standard error MLE Standard error Re-hospitalization rate: Non-Profit hospital -.023 .01 -.021 .011 Public hospital -.004 .013 -.007 .015 Number of beds -.0076 .02 -.023 .024 Male -.024 .007 -.02 .008 Black .075 .016 .061 .018 Age 70-74 .039 .01 .023 .011 Age 75-79 .091 .01 .057 .012 Age 80-84 .12 .01 .084 .013 Age 85 + .18 .01 .12 .014 Longer than 2 months -2.8 .01 -2.7 .01 Moderate severity - - .14 .012 Major severity - - .24 .013 Extreme severity - - .36 .015 Mortality rate: Non-Profit hospital .0021 .018 .010 .02 Public hospital -.0089 .024 -.013 .027 Number of beds -.069 .038 -.042 .043 Male .045 .013 .051 .014 Black -.16 .03 -.15 .03 Age 70-74 -.12 .02 -.10 .02 Age 75-79 -.18 .02 -.16 .02 Age 80-84 -.18 .02 -.16 .02 Age 85 + -.015 .02 -.003 .02 Longer than 2 months -5.7 .05 -5.5 .05 Moderate severity - - -.29 .02 Major severity - - -.46 .02 Extreme severity “ “ -.54 .03 Calendar year dummies are included in the model but are not reported in the table. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 67 The next step is to control for hospital fixed effects. To avoid excessive computation only the converted hospitals are considered in this specification. This sample includes all the patients treated in hospitals that converted from one ownership status (FP, NP or public) to another in any year between 1989 through 1998. The results are compared with a model without fixed effects on the same samples. Table 2.5 gives a summary of these estimations for the in- hospital mortality and discharge rates. Since the purpose of this exercise is to assess the effect of conversions on hospital quality, only the coefficients of ownership status are reported in the table. Table 2.5. Mortality and discharge rates for hospital spells (on converted hospitals) Withou Fixec t Hospital Effects With Hospital Fixed Effects MLE Standard error MLE Standard error Discharge rate: Non-Profit hospital -.071 .016 -.054 .022 Public hospital .0004 .02 -.16 .037 Mortality rate: Non-Profit hospital .11 .04 .105 .056 Public hospital -.011 .048 .085 .096 Other independent variables included in the model are similar to those in table 2.3. The severity variables and hospital size (number of beds) are excluded. The sample includes 25,554 patients. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 68 These results are consistent with the results reported in chapter 1 in that FP status is associated with lower in-hospital mortality rates. However, here it is clearly seen that the discharge rate in these hospitals is significantly higher. Another interesting point is that similar to the analysis of chapter 1, this result is not sensitive to whether the hospital fixed effects are included or not. This suggests that the selection effects are less important for AMI patients. Table 2.6. Out-of-hospital mortality and re-hospitalization rates for out-of hospital spells (on converted hospitals) Without Hospital Fixed Effects With Hospital Fixed Effects MLE Standard error MLE Standard error Re-hospitalization rate: Non-Profit hospital -.039 .024 -.070 .039 Public hospital -.018 .027 -.077 .060 Mortality rate: Non-Profit hospital -.036 .042 -.072 .063 Public hospital -.099 .048 -.18 .10 Other independent variables included in the model are similar to those in table 2.3. The severity variables and hospital size (number of beds) are excluded. The sample includes 21,662 patients. Using a similar model the out-of-hospital mortality and re hospitalization rates are estimated. The results are given in table 2.6. According to these results there is no significant differences between FP and NP hospitals regarding the post-discharge mortality. However, the sign of the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 69 coefficients point to higher mortality rates for FP hospitals. An interesting result concerns the public hospitals. While these hospitals have a significantly higher in-hospital mortality they have much lower post-discharge mortality rates. 2.5. Conclusions Using a transition model it is shown that the out-of-hospital mortality rates can be estimated using the discharge data without post-discharge death records. The results indicate a considerable variation in the discharge rate of AMI patients among different hospital types. The common measures of hospital quality used in the literature are studied. Most of these measures confound the mortality rate with the hospital discharge rate. Given the significant variation of discharge rates across hospitals, such measures of quality may be misleading. Regarding the effect of conversion in ownership status on the mortality of AMI patients, the results are consistent with those reported in chapter 1 in that the FP status is associated with lower mortality rates. In addition, using the proposed model it is shown that the discharge rates are higher in hospitals with FP status. However, the results of this chapter provide no evidence that faster discharge of these patients may lead to higher risks of post-discharge death. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 70 3. CONSEQUENCES OF MERGERS AND ACQUISITIONS IN CALIFORNIA’S HOSPITAL MARKET 3.1. Introduction California hospital market has been subject to a dramatic consolidation 99 during the past decade. The number of acute-care hospitals in California has decreased from 497 in 1990 to 441 in 1999 and the total number of hospital beds has decreased from 95,000 to 88,000. In the period between 1990 and 1999, while the number of single hospitals has decreased (from 269 to 190), more hospitals have joined networks increasing the number of affiliated hospitals from 228 to 251. In 1990, 56 percent of California’s acute-care hospital beds were run by firms owning more than one hospital, whereas in 1999 this share has increased to 67 percent. Among non-single hospitals the average number of hospitals owned by a firm has almost doubled from 9.3 to 17.9 hospitals. In the same period, the hospital size as measured by the number of beds has slightly increased from 192 to 200 beds. These trends suggest that the market has been re-structured to have slightly larger but fewer hospitals operated by fewer firms. These changes are 2 2 Acute-care hospitals include all short-term general or surgical hospitals. The psychiatric and rehabilitation hospitals are excluded. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 71 believed to be the consequences of a large decline in demand driven partly by the technological progress and partly by the control mechanisms implemented since the 1980’s. Barro and Cutler (1997) state three objectives for consolidation: closure, efficiency gains through economies of scale, and network creation. A large body of literature has studied the potential market power and efficiency gains in medical care markets. The evidence is however mixed. To some researchers the key to success for hospital chains is their more efficient organization and better use of the economies of scale. Many others consider that hospitals join mainly to create large networks with market power. The work presented in this chapter is different from most of the previous literature in three aspects. First, in order to study the efficiency and market power hypothesis the owning firm as opposed to the hospital, is considered as the unit of production. It is argued that the neighboring hospitals operated by a single firm can share the benefits of synergies and/or networking. Secondly, similar to Krishnan (2001) it is argued that market power may exist within diagnosis-specific markets. But different diagnoses have different degrees of flexibility of choice for the patients and thus may provide different potentials for monopoly. For instance, those patients who can easily switch from one hospital to another are least affected by monopoly prices. In such cases the markets tend to be competitive. However, in other Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 72 cases such as emergency conditions, patients typically go to the closest hospitals regardless of the hospital type and a monopolist can exploit this limitation. Here, the market for AMI (heart attack) patients is considered to study the possibility of monopoly. Finally, it is assumed that only growing chains have fully utilized the potential scale economies and/or monopoly possibilities. The recent history of hospital chains shows that while certain firms have greatly succeeded, some others have hardly survived in the rapidly changing health care market. Including the chains that have failed may result in an underestimation of potential benefits or harms of consolidations. Therefore, the motivations and effects of market consolidation with respect to specific hospital chains that had a relatively successful record. In particular, the acquisitions made by the two largest For-Profit (FP) chains in California: Tenet Healthcare Corporation and Columbia-HCA, between 1990 and 1999 are addressed. Two other Non-Profit (NP) chains (Catholic Healthcare West and Sutter Health System) that had the two highest numbers of acquisitions are also studied. Several hypotheses are considered. First, it has been argued that hospitals that join the multi-hospital FP corporations are those facilities that were operated inefficiently or had a significant financial loss. In this case those hospitals should have relatively high costs and low revenues and/or low demand prior to joining. However, the results of this paper suggest that the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 73 hospitals that joined the studied FP chains have relatively high patient revenue and operating costs per bed. There is no evidence of financial distress or low occupancy in these hospitals. In general large hospitals are more likely to be taken over by both FP and NP chains. The results also indicate that FP chains are particularly interested in well-equipped hospitals (relatively high assets per bed) with relatively costly patient mix. The second hypothesis is that chains gain efficiency through the economies of scale and decrease the costs. An analysis of the changes in hospital financial characteristics before and after joining the chains suggests that the FP chains have cut down the size of the acquired hospitals on average, by 3 percent of total beds and about 5 percent of licensed beds. These chains have also decreased hospital assets substantially (by 26% on average). The number of operating rooms and the active medical staff have also decreased with joining these networks. As a result the hospital operating costs have decreased by 10 percent on average. However, these results are not necessarily indicative of efficiency gains because the results show a similar decrease (11 93 percent on average) in the hospital’s total patient revenue. At the same time the total number of admitted patients and the total patient-days have remained constant. These results suggest that the FP chains have turned the hospitals 2 3 Patient revenue is the total revenue actually paid to the hospitals by patients and insurers for medical care. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 74 with an expensive patient mix to facilities delivering relatively low-cost medical care. In order to further investigate the possibilities of scale economies, it is hypothesized that the acquisition of new hospitals may result in savings for the chain’s other hospitals in the neighborhood. The results suggest that the hospital’s operating costs are negatively correlated with the number of hospital beds operated by the chain in the county. As the number of a FP chain’s hospital beds in the county grows, the hospital’s patient revenue remains unchanged, but the hospitals operate with significantly lower costs particularly lower administrative expenses and labor costs. These results are consistent with the efficiency hypothesis in that hospitals operated by a single firm can share some of the labor and administrative expenses. The third issue is the market power hypothesis. If chains have any market power, it would reflect as a price increase as their market share rises by acquisition of new hospitals. This issue is addressed by studying a sample of Medicare patients hospitalized for Acute Myocardial Infarction (AMI) in any of the hospitals owned by chains. A patient-level analysis shows that the charged prices rise as the number of AMI patients treated in the respective NP chain’s hospitals in the county increases as a result of acquisitions. Interestingly this result is valid only for the NP chains studied in this paper. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 75 Finally, the hospital quality in terms of patients’ health outcome is studied. It is argued that as chains take over hospitals, their aggressive cost- saving methods may result in a lower quality of care. Our analyses show that joining chains has no significant effect on the in-hospital mortality of elderly AMI patients. It should be noted that AMI sample is chosen for two reasons. First, because of the emergency nature of the disease the selection effects are expected to be minimal and secondly, our analyses show that when hospitals join the networks the number of AMI patients and their average severity index do not change significantly. Our analyses suggest some distinctive contrasts between FP and NP chains. Although both types are interested in acquiring relatively large hospitals, there is no evidence of significant change in hospital size after joining a NP network. Another interesting contrast between NP and FP chains is in the number of medical staff. The FP chains significantly decreased (by 20 percent on average) the medical staff in the acquired hospitals whereas the NP systems increased the medical staff by about 9.7 percent on average. 3.2. Background Several studies have dealt with the causes and consequences of market consolidation. Barro (2000) studied the financial performance of hospitals in Massachusetts from 1985 to 1995. He found little evidence of cost reduction or revenue-increase (market power). However, his results on bed and staff Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 76 reduction support consolidation for efficiency motivation. Conner et al. (1998) studied 3500 general hospitals between 1986 and 1994. That paper provides some evidence of cost-reduction after hospital mergers. The reduction is lower in less competitive markets. Other papers like Dranove (1998) suggest that scale economies exist only for small hospitals and no efficiency gains can be obtained from mergers of hospitals as large as 200 beds. Dranove (2000) gives an extensive account of its previous empirical literature. Summarizing the mixed evidence of the economies of scale, he concludes that the savings from mergers are not substantial. In their survey of the related literature, Gaynor and Vogt (2000) point out some problems that have plagued the empirical research on the extent of scale economies. First, patient case-load varies across hospitals of different sizes. Large hospitals tend to treat more severely ill thus more costly patients. Moreover, these hospitals usually deliver a broader variety of services resulting in lower potential for scale economies. Aggregated measures of output and costs can therefore lead to an underestimation of scale economies. Secondly, given that hospitals have an uncertain demand but a relatively fixed capacity of various outputs, some additional scale economies may be realized by consolidating the services of several hospitals. On the other hand some studies suggest that the hospital markets are not perfectly competitive and mergers can result in higher prices through Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 77 monopoly. For instance Keeler et al. (1999) suggest that mergers may potentially increase prices by up to 26%. Simpson and Shin (1998) provide some evidence of relatively high prices in non-profit California hospitals in more concentrated markets. Studying the mergers and acquisitions occurred in Ohio and California during 1994-95, Krishnan (2001) provides some evidence of increase in prices within the Diagnostics Related Group (DRG) level after mergers, especially where the merging hospitals gain significant DRG-specific market share. Krishnan’s results suggest that monopoly can be created at diagnosis-specific markets. Gaynor and Vogt (2000) and Dranove and Satterthwaite (2000) provide their respective reviews of the empirical literature on the relation between price and market concentration. Dranove and Satterthwaite consider that since different providers may offer slightly different services and treat patients with different severities, the price-concentration relationship is difficult to show. Moreover, most of the available data do not provide the actual prices paid to the hospitals. Another problem is that the market’s geographic boundaries are difficult to determine. The commonly used approach based on aggregate inflows and outflows of consumers has been criticized. For instance Capps et al. (2001) provide some evidence that a majority of patients show an aversion to travel for medical care. Moreover, whether a patient chooses to travel depends on the type of their disease. For example patients with emergency Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 78 conditions are unlikely to travel far to get medical care. Hospitals with a large market share can set relatively high prices at least for those patients who are not flexible to choose their hospital. Market power can also lead to lower quality of service. More competitive (less concentrated) markets induce hospitals to attract more patients by delivering higher quality of service. There are quite a few papers that provide evidence of higher levels of utilization and more advanced technology in less concentrated markets.2 4 However, these studies are mainly limited to the fixed price environment of the 1980s and more recent papers provide rather mixed evidence. Hamilton and Ho (2000) find that hospital mergers have no effect on the mortality of heart-attack patients. They however find that the acquisition of independent hospitals raises readmission rates. Studying the treatment of Medicare heart-attack patients between 1985 and 1998, Madison (2001) finds that multi-hospital system membership is associated with more intensive treatments and lower expenditure but no significant change in mortality outcomes. However, using a sample of Medicare patients from 1985 to 1994, Kessler and McClellan (1999) find that the mortality of heart-attack patients is higher in more concentrated markets suggesting that hospitals may use market power to lower quality of service. 2 4 See Gaynor and Vogt (2000) for a brief review o f this literature. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 79 Classical economic theories suggest that mergers can create market power and therefore result in higher prices or lower quality of service. By joining networks, hospitals can have more bargaining power in price negotiations with managed care organizations and other medical insurers. Affiliation to large networks helps hospitals to secure access to managed care contracts. These contracts represent an increasingly large part of the demand for medical care. If mergers were motivated by market power they would reflect as higher revenues (through higher prices and/or demand) or lower quality of service. On the other hand there are theoretical reasons that hospital mergers can result in cost reduction and even quality improvement through scale economies and other efficiency gains. Multi-hospital firms can save costs by creating centralized administration and information systems. They can also secure purchasing discounts from suppliers. System membership may facilitate hospitals’ access to capital as well as medical and managerial expertise. Another important source of efficiency is the additional ability of multi hospital systems to consolidate different service centers. Hospitals in a network can more easily specialize because they can secure their demand through other affiliated hospitals in the neighborhood. Finally, hospital chains can have a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 80 better “inventory” management facing demand fluctuation.2 5 Any reduction in one or more cost components that is not associated with a lower level of quality, can be indicative of economic efficiency gains. 3.3. Data The data used in this chapter consist of two main data sets. The first data set is obtained from the California’s Hospitals Disclosure Data (CHDD) from 1990 to 1999.2 6 The CHDD consists of the information obtained from the hospital financial reports submitted annually to the Department of Health Services. This data set includes hospital characteristics like ownership, profit status, size (number of beds) and other financial variables like costs and revenues, total discharges, etc. The second data set is a sub-sample of the California’s Hospitals Patient Discharge Data, which includes all the discharge abstracts for the elderly (65 years and older) patients hospitalized for Acute Myocardial Infarction in a California hospital from 1990 to 1998. The variables of this data set are already explained in section 1.4. 3.3.1. Consolidation o f California’s hospitals Figure 3.1 shows the number of acute-care hospitals in California from 1990 to 1999. The independent hospitals and hospitals that are affiliated to a 2 5 Lynk (1995) estimates that larger hospitals can reduce expenses by about 5 percent by efficiently managing the peak hour problem. 2 6 The data also include a part of year 2000. But this part is used only in cases where completeness is not necessary. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 81 network are shown separately. Between 1990 and 1999, the total number of acute-care hospitals has decreased by 11.3 percent and the total number of hospital beds has decreased by 7.6 percent. In the same period the independent hospitals have decreased by about 30 percent while the number of affiliated hospitals has increased by about 10 percent. Figure 3.1. Number of Acute-Care Hospitals in California ............. T otal-----------Independent----------Affiliated 500 450 400 - 350 300 250 200 150 100 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Year Figures 3.2 gives the average number of hospitals operated by multi hospital firms separately for private FP and private NP organizations. These trends indicate that hospital chains especially the FP ones have operated an increasing number of hospitals varying from 9.2 to 17.9 hospitals on average. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 82 Figure 3.2. Average Number of Hospitals Operated by Multi-hospital Firms in California Any Private F irm -----------For-Profit F irm ----------Non-Profit Firm 25 20 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Year 3.3.2. Major Hospital Chains in California Figure 3.3 gives the share of California’s hospital beds operated by multi-hospital systems by ownership status. This figure shows that in 1999, 54 percent of California acute-care hospital beds were owned by private multi hospital organizations. This share was about 39 percent in 1990. As it can be seen in the figure the share of NP systems has increased from 25 percent to about 39 percent of hospital beds, while the FP firms’ share has remained around the same 14 to 15 percent. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 83 Figure 3.3. Share (%) of California Hospital Beds Operated by Multi-hospital Systems Total (Private Chains) For-Profit Non-Profit 60 50 40 30 20 10 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Year Figure 3.4. Share (%) of California Hospital Beds Operated by Major Multi-hospital Systems Cath. West — ♦— Sutter T en et Col./HCA 10 9 8 7 6 5 4 3 2 1 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Year Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 84 Spetz et al. (1999) report that six organizations operate over one third 77 of California’s hospitals. The largest FP corporations in California are Tenet Healthcare Corporation and Columbia/HCA. The NP systems with the largest number of hospitals are Catholic Healthcare West (CHW), Kaiser Foundation Hospitals, Sutter Health and Adventist Health. Over the sample period (1990 to 19990) Kaiser Foundation Hospitals and Adventist Health have not much expanded.2 8 These two systems are therefore excluded from our analysis. Tenet, CHW, Columbia/HCA and Sutter had the largest numbers of hospital acquisitions in California between 1990 and 1999. The shares of the four hospital chains studied in this paper are shown in figure 3.4. These chains represent about 23 percent of California acute-care hospital beds in 1999 compared to only 6 percent in 1990. Between 1990 and 1999, Tenet and Sutter have more than doubled their size in terms of acute-care beds and Columbia/HCA and CHW have multiplied by more than six. Tables 3.1 and 3.2 give an account of the hospitals that joined these four networks by year and by ownership status prior to the acquisition. Tenet is the largest FP chain in California. Before 1995 it was known as National Medical Enterprises (NME). After acquiring FP hospital chain 2 7 The University of California and the State o f California owning respectively about 4 and 7 percent of California acute-care hospital beds in 1999, are also important players. 2 8 Between 1990 and 1999, the number of Kaiser’s acute-care hospitals in California remained unchanged at 28 and only 2 acute-care hospitals joined Adventist Health. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 85 American Medical International it was renamed as Tenet. Today, Tenet Corporation owns 116 acute-care hospitals in 17 states.2 9 The second largest FP hospital chain in California is Columbia/HCA. With about 200 hospitals in 24 states this chain is the largest chain in the US. Columbia/HCA was formed in 1994 by the merger of Hospital Corporation of America with another already large hospital chain known as Columbia. At that time Columbia was not present in California and for the purpose of our analysis HCA and Columbia/HCA are considered as the same entity. Table 3.1. Number of California acute-care hospitals that joined a network Network: Tenet Columbia/HCA Catholic West Sutter Health Year: 1992 1 - 2 - 1993 - - - - 1994 - 2 2 3 1995 2 1 - - 1996 6 5 5 1 1997 9 6 3 3 1998 9 - 6 1 1999 4 1 9 - 2000* - 1 2 - Status prior to joining: 25 10 - - For-Profit Non-profit 5 6 27 6 Public: 1 - 2 2 Total: 31 16 29 8 * The numbers of year 2000 are not complete. 2 9 More information about Tenet can be found at their website at www.tenethealth.com. 3 0 See www.hcahealthcare.com for more information about Columbia/HCA. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 86 Table 3.2. Number of California acute-care hospital beds that joined a network Network: Tenet Columbia/HCA Catholic West Sutter Health Year: 1992 80 - 394 - 1993 - - - - 1994 - 384 615 258 1995 384 231 - - 1996 989 379 1,234 169 1997 1,917 1,445 907 695 1998 1,809 - 1,447 115 1999 1,055 48 2,372 - 2000* - 188 345 - Status prior to joining: For-Profit 4,384 1,218 - - Non-profit 1,617 1,457 6,813 1,010 Public: 233 - 501 227 Total: 6,234 2,675 7,314 1,237 * The numbers of year 2000 are not complete. With 29 acute-care hospitals in 1999, Catholic Healthcare West (CHW) is the largest and the most growing NP hospital group in California. CHW represents nine religious orders and operates 42 acute-care hospitals in California, Arizona and Nevada.3 1 Sutter Health is a secular NP organization • T9 that operates 21 acute-care hospitals mainly m Northern California. The recent history of FP hospital companies suggests that the US medical care markets have undergone a great deal of reorganization. Kuttner (1996a) provides an account of the failures and bankruptcies of FP hospital chains in the 1980s. Associating this instability to the falling profits in the 3 1 See www.chw.edu for more details about Catholic Healthcare West. 3 2 For more information about Sutter Health System see www.sutterhealth.org. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 87 1980s Kuttner calls the 1990s a decade of “lucrative consolidation.” However, bankruptcies as recent as 2000 suggest that even large FP chains remain at risk.3 3 It is reasonable to assume that those FP chains that could not survive the competition did not fully use the potential efficiency gains or monopoly power. The two FP chains studied in this chapter are considered as the most profitable hospital chains in the US. An analysis of the strategies of these chains can shed some light on the motivations for hospital acquisitions. NP systems Sutter and CHW are also the fastest growing NP chains in California. As reported by Spetz et al. (1999) both chains are considered as financially healthy organizations recording positive net income in 1997 and 1998. The pattern of acquisitions by the four hospital networks suggest that both FP chains are more or less concentrated in the urban areas in southern California and Bay Area, whereas the hospitals affiliated with the two NP systems are relatively spread across large cities as well as rural areas. 3.3.3. Patient-level Data Hospitalizations of California’s elderly patients for Acute Myocardial Infarction (AMI) are used for the patient-level analyses in this chapter. This sample includes all the patients of 65 years and older, hospitalized with AMI 3 3 For instance Vencor Hospitals and Paracelcus Healthcare Corporation filed for bankruptcy in 1999 and 2000 respectively. Each one of these chains had 9 acute-care hospitals in California one year before bankruptcy. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 88 as their principal diagnosis.3 4 This sample is used for the assessment of quality and prices changes associated with hospital acquisitions. Charged prices and health outcomes both depend on the patient’s severity of illness. If acquisitions led to higher prices or lower quality of service, patients would perhaps react to these changes and switch to hospitals with better quality and lower prices. However, in the case of AMI patients because of the emergency nature of the disease, patients are most likely to choose the closest hospital, thus minimizing the selection effects. Moreover, elderly patients provide a relatively homogeneous sample regarding age-related risk factors as well as insurance coverage.3 5 3.4. Methods Financial problems are often considered as the primary motivation of hospitals to join networks. In order to test this hypothesis three specifications are used. The first model can be written as: Cjt =pXJt+rYt+Xj+sJt (3.1) where Cjt is a dummy variable indicating whether hospital j has joined a chain in year t+1, Xjt is the vector of hospital j ’ s characteristics in year t including occupancy rate and one of the following variables: total revenues, total costs, 3 4 AMI is coded as 410.xx in the International Classification of Diseases. 3 5 Practically all elderly patients are covered by Medicare and therefore less likely to be refused by certain types of hospitals. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 89 total debt and whether the hospital had a non-positive net income in year t. Yt is the vector of year dummies and Xj is the hospital-specific fixed effect. Finally £jt represents an i.i.d. random error that represents the unobserved heterogeneity among hospitals. This model is estimated for FP chains (Tenet and HCA) and NP chains (CHW and Sutter) separately. The samples include the observations (hospital-years) in which the hospital is not a part of the corresponding chains. This model can identify the effect of financial changes over time on hospitals’ decision to join a network. In the second specification hospital fixed effects are excluded but the county fixed effects are controlled for. The county fixed effects can capture some of the geographical patterns of acquisitions. This model is written as: C ^ pX ^ +T Y '+ tit+ ej" (3.2) where subscript c indicates county, t]jC is the county fixed effect and Xjct includes hospital size (number of beds), ownership status (non-profit and public hospital dummies), occupancy rate and one of the following variables: patient revenue per bed, operating costs per bed and total assets per bed. Similar to the first model, this model is estimated separately for FP and NP chains. Since the cross-sectional variation of hospital characteristics even within a given county dominates the changes over time, this model helps identify which type of hospital is more likely to join a network. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 90 Finally, the third model is one in which only the cross-sectional variations are considered. This model can be defined as: Cjc ~ PXjc + rjjc + ejC (3-3) where Cjc is a dummy variable indicating whether hospital j has joined a chain in any year in the study period (between 1991 and 2000), X JC is the vector of hospital j ’s characteristics in 1990 including hospital size (number of beds), ownership status (non-profit and public hospital dummies), occupancy rate and one of the following three variables: patient revenue per bed, operating costs per bed and total assets per bed. This model is estimated for NP and FP chains separately. In each case the sample includes all the hospitals that were not a part of the corresponding chains in 1990. The second hypothesis is that chains gain efficiency through scale economies or by consolidating services. This hypothesis is studied by estimating the effects of joining a network on the hospital’s performance and also on the costs and revenues of the chain’s other hospitals in the same county. Two specifications are considered. The first one is formulated as: H ^ S d t+ r Y '+ X j+ S j, (3.4) where H j t is hospital f s characteristics in year t like size, revenue, costs, etc., djt is a vector of two dummy variables that indicate respectively whether the hospital has joined a FP or NP chains in year t or any year before t within the study period. Other variables are defined before. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 91 In the second specification we estimate the effect of the changes in the number of a chain’s hospital beds in a given county on the operating costs, occupancy and the revenues of that chain’s hospitals in that county. This model is written as: Hjet = cSjet +rYf +Aj+ ejct (3.5) where S jc t is the number of hospital beds in county c that are operated by hospital f s owning chain. The market power hypothesis is studied using a sample of elderly patients hospitalized for AMI in the hospitals owned by the chains. As suggested by previous literature like Krishnan (2001), monopoly may be created within diagnosis-specific markets. It is hypothesized that the market power of a given chain in a specific disease (AMI) depends on that chain’s share of those patients within the county. To the extent that patients cannot switch to other hospitals, chains may gain market power by acquiring more hospitals thus serving more patients in the market. Because of the emergency nature of their disease, AMI patients are most likely to go to the closest hospital. Any positive relationship between prices charges to AMI patients and the chain’s market share would therefore 3 6 In California paramedics are instructed to take heart-attack patients to the closest hospital and hospitals are required to accept these patients regardless o f their insurance coverage. These patients may however be transferred after their situation has been stabilized. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 92 suggest that the chains exercise market power. The econometric model used for this analysis can be summarized as: Py c t = PX ij c t + a S jct +?Yt + Xj + £ijc t (3.6) where Pyc l is the patient i ’s charges in hospital j in county c in year t, Xyct is a vector of patient characteristics including gender, race, age and the interaction terms. Sjc t is the total number of elderly AMI admissions in all the hospitals operated by hospital f s owning firm in county c in year t, £yct represents the error term, and other variables are defined before. This model is estimates separately for FP and NP chains. In each analysis the sample includes hospitals operated by the corresponding chains. A similar model is used to investigate the effect of market share on the in-hospital mortality of AMI patients. The idea is that hospitals may lower the quality of service by using their monopoly power. Finally, another patient-level analysis is used to study whether joining networks has any effect on the quality of service. The quality measure is based on the in-hospital mortality outcome of elderly AMI patients. The econometric model is as follows: m9 t = fiX 9 l + 5dJl+TYt +AJ +61 l t (3.7) where mijl is a binary variable indicating if patient i died before discharge from hospital j, and is a vector of two dummies representing whether hospital j has joined a FP or NP chain in year t or any year before t in the study period. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 93 3.5. Results This section presents the main results of this chapter. First, the results regarding the selection of hospitals by hospital chains are presented. A discussion of consequences of acquisitions with respect to scale economies and market power follows. 3.5.1. Which hospitals are selected? Table 3.3 gives the estimated effects of hospital characteristics on the probability of joining a network. These estimations are based on the model shown in equation (3.1). Hospital fixed effects are included in the model. The very low values of R2 suggest that after fixing the hospitals, the time variations in financial conditions have little explanatory power in predicting the probability of joining networks. The very least that can be said here is that the results do not provide any evidence of financial distress before joining the chains. The estimation results of equation (3.2) are presented in table 3.4. Here, the model controls for county fixed effects. In contrast with the previous results, occupancy rate has no significant effect on the probability of joining. This may indicate that the studied FP chains actually focused on the counties where the demand grew relatively rapidly (or declined less). Another interesting result is that both chain types have shown an interest in relatively Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 94 large hospitals. In particular, FP chains were attracted to large and well- equipped hospitals with relatively high amounts of assets per bed. Table 3.3. Effects of hospital characteristics on the probability of joining a network (including hospital fixed effects) Network: Tenet and Columbia/HCA Catholic West and Sutter Health Specification: I II III IV I II III IV Occupancy .016* .015* ,012A .016* -.0011 -.0022 -.0019 -.0023 rate (.0073) (.007) (.0064) (.0068) (.006) (.006) (.0055) (.007) Log o f total patient revenue -.012 (.0078) — — - -.0022 (.0067) - - - Log o f total operating costs - -.015 (.01) - - - .0013 (.009) - - Non-positive net income - - .0041 (.004) - - - -.0006 (.003) - Log of long term debt - - - .0036* (.0018) - - - .0009 (.0016) Sample Size 4,286 4,313 4,431 3,672 4,438 4,465 4,583 3,688 R-square .0037 .0030 .0082 .0060 .0039 .0061 .0049 .0079 * indicates significant at 5% level. A indicates significant at 10% level. The standard errors are given in brackets. Money values are in nominal US dollars. Patient revenue and operating costs are per day amounts. Occupancy is defined as the total number of patient-days per day divided by the number of hospital beds. Hospital fixed effects and year-specific dummies are controlled for but are not shown. Table 3.5 presents an estimation of the effects of hospital characteristics in 1990 on the probability of joining chains in the ten following years. The model as shown in equation (3.3), includes county fixed effects and ownership status in 1990. These results suggest that hospitals that have joined the FP chains are selected among relatively large hospitals with relatively high Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 95 amounts of per bed patient revenues, costs and assets. This result suggests that hospitals acquired by the FP chains treated a relatively more expensive patient case-mix before being taken over. Summarizing the results of tables 3.3 through 3.5, we can conclude that the four studied chains were interested in relatively large hospitals and there is no direct evidence that any financial distress has driven hospitals to join networks. FP chains Tenet and HCA were mainly interested in the hospitals located in counties with a relatively low decline in demand and also in relatively well-equipped hospitals perhaps with relatively expensive patient case-mix. Table 3.4. Effects of hospital characteristics on the probability of joining a network (without hospital fixed effects) Network: Tenet and Columbia/HCA Catholic West and Sutter Health Specification: I II III I II III Log of total .0040* .0040* ,0040A .0042* .0043* .0051* hospital beds (.0006) (.0023) (.0023) (.0020) (.0020) (.0021) Occupancy rate .0006 .0008 .0008 -.0014 -.0015 -.0013 (.0041) (.004) (.004) (.0035) (.0034) (.0035) Log of patient revenue per bed Log of .0043 (.0028) — - -.00096 (.0035) - - operating costs per bed - .0044 (.0031) - - -.0011 (.0027) Log of total assets per bed - - .0047* (.0021) - - -.0018 (.0018) Sample Size 4,286 4,313 4,260 4,438 4,465 4,412 R-square .034 .034 .035 .014 .014 .014 County fixed effects, year-specific dummies and ownership status (public and non profit hospital dummies) are controlled for but are not shown. The first footnote of table 3.3 applies. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 96 Table 3.5. Effects ofhospital characteristics in the base year (1990) on the probability of joining a network Network: Tenet and Columbia/HCA Catholic West and Sutter Health Specification: I II m I II III Log of total hospital beds in .052* .051* .052* .034* .034* .033A 1990 (.020) (.020) (.022) (.018) (.017) (.019) Occupancy rate -.041 -.033 -.030 -.020 -.018 -.024 in 1990 (.038) (.038) (.038) (.033) (.033) (.033) Log of patient revenue per bed .055* .013 in 1990 (.023) (.020) Log of operating costs .049* .0094 per bed in 1990 (.029) (.025) Log of total assets per bed .038* .019 in 1990 (.020) (.018) Sample Size 449 451 440 457 459 448 R-square .14 .14 .14 .070 .069 .072 County fixed effects and ownership status (public and non-profit hospital dummies) in 1990 are controlled for but are not shown. The first footnote o f table 3.3 applies. 3.5.2. Consequences o f affiliation to the chains Table 3.6 gives a summary of the estimated effects of network membership on hospital financial performance. The econometric specification is given in equation (3.4). The results suggest that as hospitals joined the networks, while the total number of admissions has not changed significantly, the FP chains have significantly decreased the number of hospital beds (about 3 percent in total beds and 5 percent in licensed beds). This result is a suggestive evidence of economic efficiency gain in joining the FP chains. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 97 Table 3.6. Effects of joining a network on hospital characteristics Dependent variable: Total hospital beds Hospital licensed beds Operating costs per day Patient revenue per day Total discharges per day Occupancy rate Joined FP chain -.029A -.049* -.10* -.11* -.017 .0063 (Tenet, HCA) Joined NP (.016) (.013) (.019) (.024) (.028) (.011) chain -.0075 -.024 .00089 .0018 .035 -.021 (CHW, Sutter) (.019) (.016) (.023) (.029) (.034) (.014) Sample Size 4,988 4,988 4,723 4,687 4,972 4,988 R-square .0001 .0001 .019 .022 .0025 .0022 Administ Dependent variable: Labor costs per day Employee benefits per day rative expenses per day Total assets Total medical staff Total operating rooms Joined FP chain -.11* -.078* -.095* -.26* -.20* -.28* (Tenet, HCA) Joined NP (.018) (.028) (.029) (.038) (.044) (.087) chain -.0048 .085* -.0017 -.13* ,097A -.072 (CHW, Sutter) (.022) (.018) (.036) (.047) (.054) (.11) Sample Size 4,658 4,656 4,766 4,670 4,952 4,525 R-square .018 .016 .018 .016 .0079 .84 * indicates significant at 5% level. A indicates significant at 10% level. The standard errors are given in brackets. All dependent variables except occupancy rate are in log terms. Money values are in nominal US dollars. Occupancy rate is defined as the total number of patient-days per day divided by the number o f hospital beds. Labor costs include only the employees’ expenses. Hospital fixed effects and year-specific dummies are controlled for but are not shown. The results also suggest that after joining the FP chains both hospital operating costs and patient revenues have decreased by about 10 percent while the total admissions remained unchanged. This result together with the results Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 98 discussed above, suggest that these hospitals have switched from an expensive patient case-mix to a less expensive one. Another analysis (not shown here) in which the hospital states one year, two years and three years before and after acquisition are included in the model, suggests that hospitals gradually switch to a less costly case mix after joining the studied FP chains. The lower panel of table 3.6 gives the changes in different components of hospital costs. According to these results the FP chains have cut hospital labor costs and administrative expenses respectively by 11 and 9.5 percent on average. Compared to a similar decrease in patient revenues, these numbers do not necessarily reflect any efficiency gains. These results also show that while the FP chains have decreased the total employee benefits in their acquired hospitals (by 7.8% on average), the NP systems have increased those benefits (by 8.5% on average). However, after comparing to a decrease of 11% in total labor costs, it turns out that the FP chains have actually increased the employee benefits relative to total labor costs. Therefore the employee benefits rise after joining both FP and NP chains but the increase is higher in the case of NP chains. Both NP and FP chains have decreased hospital assets, which suggests that chains may consolidate and share services among several hospitals. However, larger decreases in the case of FP chains suggest that these chains are more aggressive in this regard. Particularly, the FP chains have decreased Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 99 the number of operating rooms by 28 percent on average. This large cut may be partly related to the changes in patient case mix towards patients who are less likely to need surgical operations. There is a contrasting difference between FP chains (Tenet and HCA) and NP systems (CHW and Sutter). While the FP chains have cut the medical staff by about 20 percent on average, the NP chains have increased the number of physicians by about 10 percent on average. If we assume that the rate of general downsizing for FP chains was about 10 percent (as shown by total operating costs), these numbers suggest that the FP chains have decreased the medical staff by an additional 10 percent while NP systems have increased it by a similar amount. The interesting question is that the hospitals that joined these FP chains may suffer a shortage in medical staff and equipment, which might lead to a lower level of quality. However, the results of patient-level analyses based on the model in equation (3.7) show that the mortality outcome of the elderly AMI patients is not significantly affected by joining any of the studied chains (including the NP ones).3 7 Moreover, a patient-level ordered multinomial logit regression of the four severity categories using a model similar to equation (3.7) shows that the severity of elderly AMI patients does not change 3 7 These analyses include a linear probability model and a proportional hazard model based on equation (3.7). An additional specification in which the four severity categories are also included is used. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 100 significantly when hospitals join the networks. Finally, to make sure that the AMI patients are not affected by any specialization strategy taken by a hospital chain, using the model in equation (3.4), the effect of joining chains on the number of AMI elderly patients admitted in the hospital is shown to be insignificant. These results are not shown in the paper. Other analyses (not shown here) suggest that affiliation with the NP systems is associated with a lower proportion of MediCal patients while joining FP chains has an opposite effect. Moreover, the proportion of Medicare patients rises with joining the FP chains, but remains unchanged as the hospital joins a NP network. 3.5.3. Scale economies So far, the results do not show any direct evidence of scale economies or any efficiency gains through operation by hospital chains. Although the FP chains studied in this paper have cut the costs considerably, hospital revenues have decreased by an approximately similar rate. Therefore, the savings may be at least partly due to specialization in generally less costly diagnoses. There is some anecdotal evidence that hospitals (especially FP hospital chains) avoid costly services like trauma centers or neonatal intensive care.3 8 3 8 For instance see Allen Myerson’s article “Hospitals specialize to enhance profits” in the New York Times, October 7, 1997. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 101 In order to detect the potential efficiency gains obtained by serving a larger body of patients, we estimate the effect of hospital acquisition on the performance of the chain’s other hospitals in the same county. The econometric model is according to equation (3.5). The model is estimated for hospitals owned by the FP and NP chains separately. Tables 3.7-a and 3.7-b show respectively the results for the FP and NP chains. These results as shown in table 3.7-b, indicate that the number of acquired hospital beds in a given county has no significant effect on the cost elements of the NP chains’ hospitals in that county. However, as seen in table 3.7-a, the FP chains have gained some considerable cost-savings by purchasing more hospital beds in the county. According to these results the FP chains have saved approximately 2.6 percent in their hospitals’ average operating costs by acquiring every thousand hospital beds in the county. These results also suggest that on average each thousand additional hospital beds in the county allowed the FP chains to decrease their employees’ labor costs by about 3.3 percent. A large part of the savings on labor costs is related to the employees’ benefits that have decreased by about 6.2 percent. These savings may be partly due to the fact that having more hospitals these chains can contract more of their labor force to outside contractors.3 9 Note that 3 9 It is a common practice for the hospital chains to contract out some o f their low-skilled services like cleaning, laundry, etc. See for instance the Service Employees International Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 102 the labor costs used in our analysis are limited to employees’ wages and benefits and do not include outside contracts. Table 3.7. Effect of hospital acquisition on chain’s other hospitals in county a- Hospitals owned by Tenet Corporation and Columbia/HCA Dependent variable: Operating costs per day Labor costs per day Employee benefits per day Administ rative expenses per day Patient revenue per day Occupancy rate Number of the corresponding FP chain’s -.026A -.033* -.062* -.065* -.006 < O O q beds in the (.015) (.015) (.019) (.026) (.019) (.009) county (xlOOO) Sample Size 386 386 386 386 386 386 R-square .032 .031 .018 .058 .034 .020 b- Hospitals operated by Catholic Healthcare West and Sutter Health Dependent variable: Operating costs per day Labor costs per day Employee benefits per day Administ rative expenses per day Patient revenue per day Occupancy rate Number of the corresponding NP chain’s .002 .006 .042 .090 .001 -.017 beds in the (.04) (.045) (.076) (.088) (.04) (.025) county (xlOOO) Sample Size 234 233 233 232 234 234 R-square .035 .031 .042 .070 .032 .0011 Hospital fixed effects and year-specific dummies are controlled for but are not shown. The first footnote o f table 3.6 applies. Union’s report “Staffing and labor practices at Tenet Healthcare Corporation hospitals in Los Angeles and Orange Comities” to the California Attorney General, October 2001. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 103 The savings are even higher for administrative expenses amounting to an average of 6.5 percent for a thousand additional beds. Using a centralized administration, hospital chains can save some of the related costs in individual hospitals. Overall, these results suggest that the FP chains have used a considerable amount of scale economies. Moreover, the variations in the hospitals occupancy rates show that the FP chains’ hospitals can fill more beds as the chain takes over other hospital in the county. This increase (about an average of 1.8 percent per 1,000 beds) is not associated with an increase in patient revenues and is therefore not suggestive of monopoly power. A plausible explanation may be that the FP chains redistribute their patients among all their hospitals in the county. When they purchase a new hospital they may close some of the services and parcel out some of those patients to their other hospitals in the neighborhood. 3.5.4. Market power The market power hypothesis is studied with respect to prices and quality. Hospital chains may charge higher prices and/or deliver lower quality of service especially to those patients who cannot switch to other hospitals. As seen in tables 3.6 and 3.7, neither the acquired hospital nor other chain’s hospitals in the county could increase the aggregate hospital revenue after acquisition. However, hospitals may still use monopoly power in a specific diagnosis. A sample of elderly AMI patients is used to study this hypothesis. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 104 These patients are likely to choose the closest hospitals and less likely to be rejected by hospitals, thus reducing the selection effects. The model used for this study is given in equation (3.6). This model is estimated for the hospitalizations in hospitals owned by the FP and NP chains separately. The results for the FP chains are shown in table 3.8-a. These results suggest that hospital acquisition in any given county does not have any significant effect on the prices or mortality of the AMI patients treated in the chain’s other hospitals in the same county. Table 3.8-b gives the corresponding results for the NP chains. These results suggest that the number of AMI elderly patients treated by the NP chains in the county has a positive and significant effect on the prices charged to these patients but no significant effect on the quality of service as measured by their in-hospital mortality. The estimated price increase depends on whether the patient’s severity of illness is controlled for. Approximately, for every thousand additional patients in the county these chains could increase their prices by 3 to 5 percent on average without raising the quality of service. 3.6. Conclusions Our analysis of the strategies of four major hospital chains in California from 1990 to 1999 suggests that all these chains have targeted relatively large hospitals. The two FP chains (Tenet and HCA) were particularly interested in well-equipped hospitals with relatively large assets and perhaps with relatively Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 105 Table 3.8. Effect of hospital acquisition on prices and mortality outcome of elderly AMI patients treated in the chain’s other hospitals in the county a- Hospitals owned by Tenet Corporation and Columbia/HCA Dependent variable: Log o f Charges ($) Mortality outcome (0: discharged alive, 1: discharged dead) Specification I II I II Number of the corresponding FP chain’s .068 .072 -.040A -.032 elderly AMI (.054) (.053) (.024) (.023) patients in the county (xlOOO) Sample Size 18,327 15,949 18,332 15,953 Number of 60 58 60 58 hospitals R-square .053 .146 .025 .177 b- Hospitals operated by Catholic Healthcare West and Sutter Health Dependent variable: Log o f Charges ($) Mortality outcome (0: discharged alive, 1: discharged dead) Specification I II I II Number of the corresponding FP chain’s .054* .029* -.038 -.033 elderly AMI (.0086) (.0098) (.035) (.040) patients in the county (xlOOO) Sample Size 18,303 16,386 18,305 16,388 Number of 37 37 37 37 hospitals R-square .138 .159 .026 .168 In model I the severity categories are not controlled for. In model II the severity categories are included in the explanatory variables. Hospital fixed effects and year-specific dummies are controlled for but are not shown. * indicates significant at 5% level. A indicates significant at 10% level. The standard errors are given in brackets. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 106 costly patient case mixes. No evidence of financial distress was found in the hospitals prior to joining these chains. The main consequence of being taken over by the studied FP chains is a general downsizing especially in medical staff and equipment. In contrast, the two NP chains (CHW and Sutter) have significantly increased the medical staff in their new hospitals. None of these changes were associated with any decline or improvement in the quality as measured by the mortality outcome of elderly AMI patients. There is also some suggestive evidence that the FP chains change the patient case mix of their acquired hospitals to less costly patients. There is suggestive evidence of both economies of scale and market power. The results suggest that the FP chains achieved significant savings in their hospitals’ operating costs especially in administrative expenses. On average, these chains could decrease their operating costs and administrative costs respectively by 2.6 and 6.5 percent by acquiring every thousand additional hospital beds in the county. The studied NP chains on the other hand did not gain any economic efficiency. Our analysis of a sample of elderly AMI patients suggests that as they have treated more patients in the county these NP chains have charged higher prices. The extent of this price increase is estimated to be in the rage of 3 to 5 percent for every thousand additional patients. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 107 4. CONCLUDING REMARKS The re-structuring of the California hospital market in the past decade is characterized by a large number of ownership changes. Some of these changes in ownership are accompanied by a change in profit status of the hospital. Changes in legal ownership status of the hospitals have been the focus of a fair amount of research. The public concerns regarding the possible detrimental effects of conversion to FP status on the quality of medical care have attracted a lot of attention. The evidence is however mixed. In the first two chapters of this dissertation, the effects of changes in ownership status on hospital quality are studied. The issues of patient selection and the choice of quality measures are addressed in detail. The findings suggest that the conversions to FP status had mixed effects on hospital quality. While FP status is found to be associated with a lower in-hospital mortality of AMI patients, it has an increasing effect on the probability of complication among these patients. Moreover, the access to the emergency services is more limited in hospitals with FP status. In terms of economic efficiency, FP status is found to provide some advantages compared to NP form. Hospitals in FP status have shorter hospital stays and particularly shorter inpatient waiting periods before a principal procedure is performed. The treatment in FP hospitals is also more intensive than in hospitals in NP status, and more Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 108 severely ill patients are found to generally prefer hospitals in FP status over public and NP hospitals. It should be noted that all these results are obtained from hospitals that converted from one profit status to another. There is some evidence that hospitals that have a stable ownership status are different from the converted hospitals. This dissertation does not provide any conclusive result regarding the selection of hospitals into conversion. Neither does it control for such selection issues. Therefore, the above results do not imply that conversion to one or the other profit status can improve hospital quality. Rather, the findings are limited to the conversions that actually occurred during the sample period from 1990 to 1998. Moreover, there is another limitation to these results. Implicit in the models is a symmetry assumption that conversion effects are reversible that is, conversions in opposite directions have opposite effects on performance. Although this hypothesis cannot be rejected using the data, due to the relatively low number of conversions in each direction, the effects of conversion without the symmetry assumption are not statistically significant and the results do not provide conclusive evidence of symmetry. Following the previous literature the symmetry of conversion effects is assumed in this dissertation. Further research in this regard requires more data and is recommended. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 109 Another important aspect of this study is the choice of quality measure. The use of hospitalization records is very common in the literature. However, most of the quality measures used in the literature confound mortality and discharge probabilities. This issue is addressed in chapter 2. The results suggest a significant variation in discharge rates among different hospital types. The econometric model developed in this chapter can be used to estimate these variations. Moreover, this model is used to estimate post discharge mortality rates using the hospitalization history. These estimations do not require death records after discharge. The model is implemented on California hospital discharge data. However, due to the shortcomings of the public-use files the estimations of out-of-hospital mortality rates are not sufficiently accurate. This issue can be solved either by using more accurate data on the exact hospitalization dates or by refining the model to include a higher number of piece-wise changes in mortality rates. The effect of organization and profit status cannot be studied without a addressing the firms’ behavior in the market. One of the main driving forces behind market consolidation in California hospitals is the growth of multi hospital Systems. Chapter 3 studies the behavior of four of the largest hospital networks in California. The results suggest that these networks behave differently depending on their profit status. These differences can shed some light on the differences resulted from different organizational forms. The FP Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 110 chains are mainly interested in efficiency gains by acquiring more hospitals in a few clusters mostly in urban areas. These chains are found to do aggressive downsizing and cost-cutting to their new hospitals. On the other hand the NP networks are mainly interested in networking benefits of affiliating more hospitals. Further research in this regard especially on the potential specialization of hospitals in a single network is strongly suggested for future research. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I l l REFERENCES 1. Arrow, Kenneth (1963) “Uncertainty and the welfare economics of medical care” American Economic Review, 53 (5): 941-973. 2. Baker, Constance M., P. L. Messmer, Ch. C. Gyurko, et al. (2000) “Hospital ownership, performance, and outcomes: assessing the state-of- the-science” Journal o f Nursing Administration, 30 (5): 227-240. 3. Barro, Jason R. (2000) “Efficiency and Monopoly as explanations for hospital mergers” NBER conference on Industrial Organization of Medical Care (Summer 2000, Nashville, Tennessee), forthcoming in RAND Journal o f Economics. 4. Barro, J. R., and D. M. Cutler (1997) “Consolidation in the medical care marketplace: a case study from Massachusets” NBER Working Paper 5957, National Bureau of Economic Research, Cambridge, MA. 5. Blalock, K. and S. M. Wolfe (2001) Questionable Hospitals: A detailed look at “ patient dumping", Public Citizen Health Research Group (www.citizen.org) available at www.questionablehospitals.org. 6. Braunwald, E. and M. R. Bristow (2000) “Congestive Heart Failure: Fifty Years of Progress” Circulation, American Heart Association, 2000 (102): IV-4-23 (available at www.circulationaha.org). 7. Capps, Cory S., David Dranove, Shane Greenstein and Mark Satterthwaite (2001) “The silent majority fallacy of the Elzinga-Hogarty criteria: a critique and new approach to analyzing hospital mergers” NBER Working Paper 8216, National Bureau of Economic Research, Cambridge, MA. 8. Carey, K. and J. F. Burgess Jr. (1999) “On measuring the hospital cost/quality trade-off’ Journal o f Health Economics, 8: 509-520. 9. Chou, Shin-Yi (2002) “Asymmetric information, ownership and quality of care: an empirical analysis of nursing homes” Journal o f Health Economics, 21 (2002): 293-311. 10. Connor, Robert A., Roger D. Feldman and Bryan E. Dowd (1998) “The effect of market concentration and horizontal mergers on hospital costs and prices” International Journal o f the Economics o f Business, 5 (2): 159-180. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 1 2 11. Cutler, David M. (1995) "The incidence of adverse medical outcomes under prospective payment" Econometrica, 63 (1): 29-50. 12. Cutler, David M., and J. R. Horwitz (2000) “Converting hospitals from not-for-profit to for-profit status: Why and what effects?” in The Changing Hospital Industry: Comparing Not-for-Profit and For-Profit Institutions, edited by David M. Cutler, The University of Chicago Press, pp. 45-79. 13. DesHamais, S., L. E. McMahon Jr., and R. Wroblewski (1991) "Measuring outcomes of hospital care using multiple risk-adjusted indexes" Health Services Research, 26 (4): 425-445. 14. Dranove, David (2000) The Economic Evolution o f American Health Care, Princeton University Press. 15. Dranove, David (1998) “Economies of scale in non-revenue producing cost centers: implications for hospital mergers” Journal o f Health Economics, 17 (1998): 69-83. 16. Dranove, David and Mark Satterthwaite (2000) “The industrial organization of health care markerts” ” in Handbook o f Health Economics, Vol. I, Edited by A. J. Culyer and J. P. Newhouse, Elsevier Science, pp. 1093-1139. 17. Ettner, Susan L., and R. C. Hermann (2001) “The role of profit status under imperfect information: evidence from the treatment patterns of elderly Medicare beneficiaries hospitalized for psychiatric diagnoses” Journal o f Health Economics, 20: 23-49. 18. Foundation for Health Quality (1997). Assessing Hospital Performance, Quality Measurement Advisory Service, available on line at http://www.qmas.org/tools/guide-assessing/. 19. Frank, Richard G., and D. S. Sulkever (1994) “Nonprofit organizations in the health sector” Journal o f Economic Perspectives, 8 (4): 129-144. 20. Frankl, S. E., J. L. Breeling, and L. Goldman (1991) “Preventability of emergent hospital readmission” American Journal o f Medicine, 90: 667- 674. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 113 21. Gaynor, Martin and William B. Vogt (2000) “Antitrust and competition in health care markets” in Handbook o f Health Economics, Vol. I, Edited by A. J. Culyer and J. P. Newhouse, Elsevier Science, pp. 1405-1487. 22. Geweke, John, Gautam Gowrisankaran, and Robert J. Town (2001) “Bayesian Inference for Elospital Quality in a Selection Model” NBER Working Paper 8497, National Bureau of Economic Research, Cambridge, MA. 23. Glaeser, Edward L. and A. Shleifer (1998) “Not-for-profit Entrepreneurs” NBER Working Paper 6810, National Bureau of Economic Research, Cambridge, MA. 24. Goddeeris, J. H., and B. A. Weisbrod (1998) “Conversion from Nonprofit to For-profit Legal Status: Why does it happen and should anyone care?” in To Profit or Not to Profit: the Commercial Transformation o f the Nonprofit Sector, edited by Burton A. Weisbrod, Cambridge University Press, pp. 129-148. 25. Gowrisankaran, Gautam, and Robert J. Town (1999) “Estimating the quality of care in hospitals using instrumental variables”, University of Califomia-Irvine working paper, Forthcoming in Journal o f Health Economics. 26. Hartz, A. J., H. Krakauer, E. M. Kuhn et al. (1989) “Elospital Characteristics and Mortality Rates”, New England Journal o f Medicine, 321:1720-1725 27. Himmelstein, D. EL, S. Woolhandler, I. Helander, et al. (1999) “Quality of Care in Investor-owned vs Not-for-profit HMOs” Journal o f the American Medical Association, 282 (2): 159-163. 28. EIo, Vivian, and B. H. Hamilton (2000) “Hospital mergers and Acquisitions: Does market consolidation harm patients?” Journal of Health Economics, 19: 767-791. 29. Jencks, S. F., J. Daley, et al. (1988) "Interpreting hospital mortality data: the role of clinical risk adjustment" Journal o f the American Medical Association, 260 (24): 3611-3616. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 114 30. Kahn, K. L., W. H. Rogers, et al. (1990) "Measuring Quality of Care with Explicit Process Criteria before and after Implementation of the DRG- based Prospective Payment System" Journal o f the American Medical Association, 264: 1969-1973. 31. Kessler, Daniel P. and Mark B. McClellan (1999) “Is hospital competition socially wasteful” NBER Working Paper 7266, National Bureau of Economic Research, Cambridge, MA. 32. Keeler, E. B., G. Melnick and J. Zwanziger (1999) “The changing effects of competition on non-profit and for-profit hospital pricing behavior” Journal o f Health Economics, 18 (1999): 69-86. 33. Keeler, E.B., L.V. Rubinstein, K. L. Kahn, D. Draper et al. (1992) “Hospital Characteristics and Quality of Care” Journal o f the American Medical Association, 268 (13): 1709-1714. 34. Kessler, Daniel, and Mark McClellan (2001) “The Effects of Hospital Ownership on Medical productivity” NBER Working Paper 8537, National Bureau of Economic Research, Cambridge, MA. 35. Krishnan, Ranjani (2001) “Market restructuring and pricing in the hospital industry” Journal o f Health Economics, 20 (2001): 213-237. 36. Kuttner, Robert (1996a) “Columbia/HCA and the Resurgence of the For- Profit Hospital Business” (Part I) The American Prospect, 335 (5): 362- 367. 37. Kuttner, Robert (1996b) “Columbia/HCA and the Resurgence of the For- Profit Hospital Business” (Part II) The American Prospect, 335 (6): 446- 451. 38. Lohr, K. N. (1988) "Outcome Measurement: Concepts and Questions" Inquiry, 25: 37-50. 39. Ludke, R. L., B. M. Booth, and J. A. Lewis-Beck (1993) "Relationship between Early Readmission and Hospital Quality of Care Indicators" Inquiry, 30: 95-103. 40. Lynk, W. (1995) “The creation of economic efficiencies in hospital mergers” Journal o f Health Economics, 14 (1995): 507-530. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 115 41. Madison, Kristin (2001) “Essays on Organizational Forms in the Health Care Industry” PhD Dissertation, Stanford University. 42. Mark, Tami L. (1999) “Analysis of the Rationale for, and Consequences of, Nonprofit and For-profit ownership Conversions” Health Services Research, 34 (1) Part I: 83-101. 43. McClellan, Mark, B. J. McNeil, and J. P. Newhouse (1994) “Does More Intensive Treatment of Acute Myocardial Infarction in the Elderly Reduce Mortality?: analysis using instrumental variables” The Journal o f the American Medical Association, 272(1): pp. 859-866. 44. McClellan, Mark, and Douglas O. Staiger (2000) “Comparing Hospital Quality at For-profit and Not-for-profit Hospitals” in The Changing Hospital Industry: Comparing Not-for-Profit and For-Profit Institutions, edited by David M. Cutler, The University of Chicago Press, pp. 93-112. 45. McClellan, Mark, and Douglas O. Staiger (1999) “The Quality of Health care Providers” NBER Working Paper 7327, National Bureau of Economic Research, Cambridge, MA. 46. Norton, Edward C., and Douglas O. Staiger (1994) “How Hospital Ownership Affects Access to Care for the Uninsured” Rand Journal o f Economics, 25 (1): 171-185. 47. Psaty, Bruce M., R. Boineau, L. H. Kuller, and R. Luepker (1999). “The Potential Costs ofUpcoding for Heart Failure in the US” Excerpta Medica, 84 (1): 108-109. 48. Romano, P. S., and B. K. Chan (2000) “Risk-adjusting Acute Myocardial Infarction Mortality: Are APR-DRGs the right tool?” Health Services Research, 34 (7): 1469-1489. (The article is followed by N. Goldfield and R. Averill’s commentary and the authors’ reply, pp: 1491-98). 49. Rosenthal, Gary E., D. W. Baker, D. G. Norris, et al. (2000) “Relationships between In-hospitals and 30-day Standardized Hospital Mortality: implications for profiling hospitals” Health Services Research, 34 (7): 1449-1468. 50. Shen, Yu-Chu (2001) “Identifying the Effect of ownership on the Quality of Care” Unpublished manuscript, Harvard University. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 116 51. Silverman, Elaine, and Jonathan Skinner (2001). “Are For-profit Hospitals Really Different? Medicare upcoding and market structure” NBER Working Paper 8133, National Bureau of Economic Research, Cambridge, MA. 52. Simpson, John and Richard Shin (1998) “Do Nonprofit Hospitals Exercise Market Power?” International Journal o f the Economics o f Business, 5 (2): 141-157. 53. Sloan, Frank A. (2001) “Hospital Ownership Conversions: Defining the Appropriate Public Oversight Role” forthcoming in Frontiers in Health Policy Research, edited by Alan M. Garber, Vol. 5, Book in progress, National Bureau of Economic Research. 54. Sloan, Frank A. (2000) “Not-for-profit ownership and hospital behavior” in Handbook o f Health Economics, edited by A. J. Culyer and J. P. Newhouse, Elsevier Science, Vol. 1, Chapter 21,1141-1174. 55. Spetz, Joanne, Jean Ann Seago and Shannon Mitchell (1999) “Changes in Hospital Ownership in California” PPIC Report, Public Policy Institute of California. 56. Thomas, J. W. (1996) "Does risk-adjusted readmission rate provide valid information on hospital quality?" Inquiry, 28: 258-270. 57. Thomas, J. W., J. J. Holloway, and K. J. Guire (1993) “Validating risk- adjusted mortality as an indicator for quality of care” Inquiry 30: 6-22. 58. Thomas, J. W. and J. J. Holloway (1991) “Investigating early readmission as an indicator for quality of care studies” Medical Care, 29 (4): 377- 394. 59. Vaccarino, Viola, H.M. Krumholz, J. Yarzebski, et al. (2001) “Sex differences in 2-year mortality after hospital discharge for myocardial infarction” Annals o f Internal Medicine, 134 (3): 173-181. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 117 APPENDIX Proof of proposition 1.1: Hospitals’ first-order condition: Rt(s,t)— Ct(t,0), can be differentiated with respect to s to show: = (1) OS OS which combined by the assumptions Rst> 0, Ctt> 0, and Rtt< 0, gives: — >0. 8s Since by assumption Ct(t, $)<Ct(t, $ ), from the hospitals’ FOC it follows that Rt(s,t(s, (?))<Rt(s,t(s, $)), and since Rt is decreasing in t, then: t(s, $)>t{s, 6 * 1 ). Moreover, considering the assumption Ctt(t, $ ), and using equation (jt (1), one can show that — (s,62) > — (s,9l) . These results imply that hospitals ds ds utilize higher levels of treatment for sicker patients, and that more efficient hospitals not only deliver more service to patients of equal severity, but increase treatment with a higher rate. Moreover, since P is increasing in t, a given patient has a higher survival probability if she chooses the more efficient hospital (H -2). In summary, optimal treatment function is characterized by the following expressions: # > o , t{s, &)>t(s, (s, ex ), os ds ds andP(s,t(s,$ ) ,rj)>P(s,t(s,&),rj) with & >$ (2) Patients’ choice of hospital is written as: max# {AP(s,t(s, tf1 ), tj}-{ 1 - A )(y -f)2}. (3) Considering the assumption (/= 0 , /=1), the optimal choice can be obtained: H= 2 if and only if: P(s,t(s, & * ), jf)-P(s,t(s, & * ), rj)>/4l-'2.y), (4) H= 1 if and only if: P(s,t(s, (?\ rf)-P(s,t(s, 01 ), rj)</u(\-2y), (5) and H is randomly chosen if the equality holds, I - A where n= — j — represents the relative weight patient attaches to her subjective taste as opposed to the quality of medical care, that affects her survival probability. 1 -2 / measures how much the patient prefers one hospital type to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 118 another. It can be seen from the optimal choice given in (4) and (5), that the patient chooses the more efficient hospital if the benefit gained through a higher probability of recovery is larger than the disutility of switching from her favorite hospital to the other. It can also be shown that all patients with y>0.5, choose the more efficient hospital, H=2. In order to see how severity affects patient choice, one can show that the LHS of inequality (4) is an increasing function of s: 4 - {P M s, & ) , l)}= Ps(s,t(s, rf) ~Ps(s,t(s, 0), 7j)+ . . . ds +Pt(s,t(s, &), ij)dt{Sf - ^ + Pls,t{s,& \ Ti)dtiS: 0l) >0- os O S Note that since Pst>0, then Ps(s,t(s, &), r/) >Ps(s,t(s,$), rj). Moreover, since Pft=0, Pt(s,t, rf) does not depend on t, implying the equality of Pt(s,t(s, $), rf) and n . , j . . _ ■ 1 1 • * u 8t{s,02) dt{s,9l) Pt(s,t(s, (f),rf). Finally as it was shown, — -------> --------- . ds 8s Therefore, the LHS of inequality (4) is increasing in severity and the inequality is more likely to hold with higher levels of s, indicating that in similar conditions, sicker patients are more likely to choose the more efficient hospital. ♦ Proof of proposition 1.2: Consider first the simple case where 2=0. In this case the optimal choice of patients reduces to: H — 2 if and only if: y> 0.5, (6) H — 1 if and only if: y< 0.5. (7) The aggregate probabilities of survival can be written as: 0.5 + q o P,=P (s,A,0')= j J P isd isJ'ltfd F ^d F y (8) y=Q r/= — co 1 +00 P2=P(.a,02)= j I P (s,t(s,02),9)dF,dFr (9) /= 0.5 7= - o o where Fr and F n are cumulative probability distribution of y and ij. Note that since P(s,t(s, &), rf)>P(s,t(s, $ ), rf), for all values of y and rj, as long as y has a symmetric distribution around 0.5, it can easily be seen that Yy> Pi. Moreover, since severity s, is independent from both y and rj, the equations (8) and (9) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 119 can be integrated over the possible values of s, to show that fP(s ,A,02)dF5 > Jp {s,X,9lyiFs , which implies that the aggregate survival s s rate is higher in more efficient hospitals regardless of severity. Now, consider the case 2>0. Define sets Ei and E2 as the respective events that a patient with given type chooses hospital 1 and 2. These sets can be written as: E\(s, y rj)={(y, rj)\ P(s,t(s, &), rj)-P(s,t(s,tf), rj) < E2(s,y rj)={{% 7 )1 Pis Ms, &), rj)— P(s, t(s, 0X ), rj) >/j(\-2y)} The aggregate survival probabilities are: 1 J \P {s,t{s,eX ),V)dFvdFr (10) 1 P2=P(s,A,02)= J \P {s,t{s,e2),tl)dFndFY (11) r= 0 E 2 Considering that Ej and {(y, rj)\ y>0.5} are disjoint, that {(yrj)\ ;k>0.5} is a subset of E2, and that ^and 77 are independent, one can show: p, = j j p (, ,r1 ^ x )dFr ) dFr (12) y = 0 E x 0.5 P2 = J \p (s ,n ,0 ‘)dFt dF,+ J E ,[P (s,r;,0 !)]rff; y~0 E2 y~0.5 ... = j E,[i>(s,)7,« J)] (13) y=0 E2 where Ex is the mean operator over random variable x. Subtracting the above equations, the following statement is straightforward: Y1-Yi>l -^[P{s,t] ,92 j]+a '\ \P{s,T,,e2 )dF/Fr- \ J P(s,rj,0')dFr ] dF ? + J \P{s,V ,9l)dF n d F r hence: 2 ^ y= 0 E 2 y = 0 £ 2 P 2-P ,> |e n[P(s,r} ,92 )-P(s,T),0')]+ | \[P(.s,Ti,02 ) + P(s,Tj,0l)]dFn dF r 7» o e 2 and since both the right-hand-side terms are positive and non-zero: P2 >Pi ♦ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Essays on consumption behavior, economic growth and public policy
PDF
Essays on contracting in the construction industry
PDF
A methodology to identify high -risk patients with diabetes in the California Medicaid populations (Medi -Cal)
PDF
Intellectual property rights, quality control and incentives
PDF
Essays on regulation of public utilities and the provision of public goods
PDF
Essays on new product preannouncements
PDF
Effects of a formulary expansion of the use of SSRIs and health care services by depressed patients in the California Medicaid program
PDF
A study of employee health plan choice and medical cost: Panel data probit regression and sample selection model
PDF
Infertility and assisted reproductive technology in a pluralistic world: A development and application of a Hindu ethic
PDF
A new paradigm to evaluate quality-adjusted life years (QALY) from secondary database: Transforming health status instrument scores to health preference
PDF
Controlling for biases from measurement errors in health outcomes research: A structural equation modeling approach
PDF
Essays on technological evolution and financial returns to innovation
PDF
Assessing the cost implications of combined pharmacotherapy in the long term management of asthma: Theory and application of methods to control selection bias
PDF
Compliance study of second-generation antipsychotics on patients with schizophrenia
PDF
Cost analysis of three pharmacy counseling programs for diabetics in a health maintenance organization
PDF
Individual heterogeneity and program evaluation
PDF
Assessment of prognostic comorbidity in hospital outcomes research: Is there a role for outpatient pharmacy data?
PDF
Essays on auctions
PDF
Business political influence on agency regulatory decision making: The efficacy of corporate political activity in Taiwan's CATV license awarding
PDF
Health care finance reform and market failure at King/Drew Medical Center, Los Angeles: An analysis of pediatric social outcomes in an urban public safety -net hospital
Asset Metadata
Creator
Farsi, Mehdi Nasser
(author)
Core Title
Essays on organizational forms and performance in California hospitals
School
Graduate School
Degree
Doctor of Philosophy
Degree Program
Economics
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
Economics, General,health sciences, health care management,OAI-PMH Harvest
Language
English
Contributor
Digitized by ProQuest
(provenance)
Advisor
MacLeod, W. Bentley (
committee chair
), Currie, Janet (
committee member
), Graddy, Elizabeth (
committee member
), Ridder, Geert (
committee member
)
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c16-248180
Unique identifier
UC11339344
Identifier
3093759.pdf (filename),usctheses-c16-248180 (legacy record id)
Legacy Identifier
3093759.pdf
Dmrecord
248180
Document Type
Dissertation
Rights
Farsi, Mehdi Nasser
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA
Tags
health sciences, health care management