Primary research
A robust analysis of hospital efficiency and factors affecting variability
Health Services Management Research 0(0) 1–10 ! The Author(s) 2017 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0951484817730537 journals.sagepub.com/home/hsm
Herve Leleu1, Mona Al-Amin2, Michael Rosko3 and Vivian G Valdmanis4
Abstract The objectives of this paper are to use data envelopment analysis to measure hospital inefficiency in a way that accounts for patient outcomes and to study the association between organizational factors, such as hospital–physicians integration level and teaching status, and market competition with hospital inefficiency. We apply the robust data envelopment analysis approach to a sample of private (both not-for-profit and for-profit) hospitals operating in the United States. Our data envelopment analysis model includes mortality and readmission rates as bad outputs and admissions, surgeries, emergency room, and other visits as good outputs. Therefore, our measurement of hospital inefficiency accounts for quality. We then use a subsampling regression analysis to determine the predictors of hospital inefficiency. For-profit, fully integrated and teaching hospitals were more efficient than their counterparts. Also hospitals located in more competitive markets were more efficient than those located in less competitive markets. Incorporating quality in the measurement of hospital efficiency is key for producing valid efficiency scores. Hospitals in less competitive markets need to improve their efficiency levels. Moreover, high levels of hospital physician integration might be instrumental in ensuring that hospitals achieve their efficiency goals.
Keywords efficiency, health care organizations, hospitals, integration
Introduction Health care spending is on the rise worldwide and as expensive new technologies emerge and populations age, there is a strong incentive to gain better understanding of health care costs.1 Interest in hospital costs specifically mounted since the early 1970s given the economic burdens hospital costs place on healthcare systems in the United States and elsewhere. In recent years, the interest in hospital cost and quality has intensified in the US and was finally formalized in the Patient Protection and Affordable Care Act, 2010 (ACA). Value Based Purchasing (VBP) places hospitals under financial pressure to deliver care of the highest quality and with optimal efficiency.2 Other health care systems, such as the English NHS, are also struggling to improve their efficiency and contain cost.3 Understanding the drivers of hospital costs and efficiency is therefore paramount to policymakers and healthcare managers alike. However, improving efficiency by either lowering cost or increasing output is of no value in healthcare unless quality is maintained. As Zinn and Flood4 argue, limited studies on hospital efficiency accounted for quality but the need to control or lower cost while maintaining, if not improving, quality ‘‘has never been more urgent or complex’’. Over the years,
data on hospital outcomes have become publicly available through the Center for Medicare and Medicaid Services (CMS), which provides an ample opportunity to improve the validity and reliability of modeling hospital efficiency while accounting for patient outcomes and to determine the predictors of hospital efficiency. Previous research indicates that certain hospital characteristics, such as for-profit (FP) ownership, size, and teaching status, are associated with hospital efficiency.5,6 However, other organizational factors, such as hospital– physician relationships, have been less studied. Hospitals have a variety of integration arrangements they can adopt with their physicians; these arrangements range from very 1
CNRS-LEM (UMR 9221), IESEG School of Management, Lille, France Suffolk University, Boston, MA, USA 3 Widener University, Chester, PA, USA 4 Western Michigan University, Program in Public Health, Grand Rapids, MI, USA 2
Corresponding author: Mona Al-Amin, Healthcare Administration Department, Sawyer Business School, Suffolk University, 120 Tremont Street, Room 5603, Boston, MA 02108, USA. Email: [email protected]
2 tight integration such as the salaried or employment model to loose arrangements such as group practice without walls. Integration arrangements are important because they influence hospital’s ability to control cost, improve quality, increase market share, and achieve other organizational goals.7 In recent years, hospitals have moved towards tight arrangements, such as employment, that ensure the highest levels of integration.8 While earlier research did not find support for the relationship between physician integration and lower hospital cost,9 it is important to note that this era, specifically post the ACA, is different from the 1980s and 1990s since VBP, public reporting, and bundled payments were not common then. Previous research indicates that physician integration mechanisms influence hospital efficiency and operating margins,10,11 however, evidence on the effect of integration levels on hospital efficiency and patient outcomes is insufficient and inconsistent.12 In a more recent study, Baker et al.8 relied on data from a private feefor-service insurance plan to investigate the impact of physician integration strategies on healthcare spending. They found that markets with hospitals with the highest level of physician integration have higher average price per hospital admission and spending per enrollee. Baker et al.8 focused on market level spending, however, more research is needed to understand this relationship at the hospital level. Further research on the relationship between physician integration levels and hospital efficiency is needed. The aim of this paper is two-fold. First, we measure efficiency and adjust for more precise quality measures than included in previous work. This quality adjustment is an improvement over earlier data envelopment analysis (DEA) studies, which focused on the intermediate good of hospital production, i.e. inpatient admissions, number of surgeries et cetera. By including readmission rates and mortality, we can directly measure the quality of hospital services not just the quantity. We are able to include these patient outcome measures because access is now available to publicly reported quality measures through CMS; these measures were not previously available. Second, we investigate the association between market factors and organizational characteristics such as market competition, ownership, level of physician integration, and teaching status with this improved efficiency measure. To derive our efficiency measures, we apply the robust bootstrapping DEA approach to a sample of private (both not-for-profit (NFP) and FP) hospitals operating in the United States. The robust approach allows us to avoid two main shortcomings of traditional DEA approaches. Primarily, it controls for outliers in the derivation of efficiency scores and can be viewed as an intermediate between deterministic and stochastic methods. Secondly, it allows for the utilization of a relevant bootstrap inference in the second stage where determinants of
Health Services Management Research 0(0) inefficiency are analyzed through a regression analysis. This approach had been used in the assessment of home health care services.13
Conceptual framework One of the major theoretical reasons for using DEA in the study of hospitals is that it is a good method for measuring and assessing the underlying causes of hospital X-inefficiency.14 The basic concept of the X-inefficiency theory is that under the proper motivation, managers and employees could behave to operate closer to optimality and when these motivations are absent the distance from optimality increases. There are several reasons for X-inefficiency including: incomplete labor contracts, not all factors of production are marketable given alternative pricing and reimbursement mechanisms, and the production function may change given technology and payment scheme variations. X-inefficiency has also been a popular theoretical framework when describing the NFP hospitals in the US health care system. In the healthcare sector, property rights arrangements that attenuate owner (i.e. stock-holder) wealth decreases the opportunity costs of non-wealth maximizing behavior decreases. Given the market failures in producing hospital care, including asymmetric information, stochastic demand, non-price competition, and the contracting between hospitals and payers, X-inefficiency increases results. A summary of our model is presented in Figure 1 and discussed in the proceeding sections.
DEA background The DEA approach, which uses a best practice frontier to gauge the relative efficiency of hospitals to a ‘‘best practice’’ frontier, is practical for several reasons. First, the economic regularity conditions required for a production function are met including that there is no ‘‘free lunch’’ (inputs are required to produce outputs) and that there is strong disposability of inputs and outputs (outputs are non-decreasing in inputs). A second reason that the DEA is appropriate is its ability to employ multiple inputs and outputs in the specification of the linear programming model from which results are derived. This is important because of the multiple outputs hospitals produce using multiple inputs. A third benefit is the DEA’s ability to use inputs and outputs in their natural units rather than converting them into prices which are often missing. And, finally, there is no requirement to specify a traditional cost minimizing or profit maximizing assumption on the data, an assumption that may not be valid since hospitals do not operate by competitive market norms which includes many buyers and sellers, all information is conveyed by price, products are identical, and free entry and exit. Even though there are FP hospitals in
Leleu et al.
3 Organizaonal Factors
Hospital Efficiency
Physician Integraon Level For-profit Teaching Status Technology Level Medicaid Admissions Medicare Admissions
Good Outputs Surgery Volume Visit Volume
Market Factors Compeon
Bad Outputs 30-day Readmission Rates 30-day Mortality Rates Inputs Case Mix Index Staffing Beds
Figure 1. Predictors of hospital efficiency.
the industry, they cannot conform to the traditional definition of a competitive market. DEA analysis of hospital productivity has been a staple in the health care productivity/efficiency literature since the 1980s.a Advances in DEA included more detailed specifications to separate inefficiency from factors that either endogenously or exogenously affect performance. There was concern that hospitals treating a more severe patient load would be deemed inefficient simply because their patients, by nature, require more hospital services, At first, differentiation by age of the patient (Medicare versus non-Medicare), types of inpatient services (acute versus intensive care), emergency room visits versus outpatient visits, and number of surgeries were used since more processes were needed in terms of nursing care, ancillary services, and beds as patients require these services to become healthier. This estimation of patient disease differentiation was supplanted by the addition of case mix indices to adjust for the types of patients treated more precisely. Another issue missing from earlier research using the DEA approach is the notion of quality. However, more recent studies have begun to expand on the use of quality care indicators. By relaxing the strong disposability assumption, ‘‘congestion’’ can arise wherein economic ‘‘bads’’ increase with the production of economic ‘‘goods’’. Treatment indicators including antibiotic timing, percent of patients given an oxygenation, and the percent of patients given a pneumococcal vaccine were used by Nayar and Ozcan17 on a sample of hospitals operating in Virginia. Using a national sample, Valdmanis et al.5 used failure to rescue, infection due to medical care, postoperative respiratory failure, and postoperative sepsis as ‘‘bad’’ outcomes that were measured as economic congestion. In other words, more hospital care is associated with an increase in the number of bad outcomes. For example, longer length of stays are often considered an economic good since more care often leads
to better outcomes but may also result in bad outcomes such as hospital-acquired infections. Clement et al.18 also assessed congestion noting that more care led to more bad outcomes, which is not surprising given the high levels of iatrogenic diseases reported in hospital. Readmission rates are currently being used as a key measure for assessing and comparing quality of care between hospitals.19 It has been estimated that the 20% of readmission rates for Medicare patients cost an additional $17 billion dollars (Jenks et al., 2009).20 This has induced CMS to impose financial penalties on hospitals with high 30-day readmission rates.21,22 The most commonly used readmission rates are for heart failure, acute myocardial infarction, and pneumonia that occur within 30 days of a patient’s discharge (Davis et al., 2013).23 Since readmissions are costly, it has been recommended that hospital studies should include the 30-day readmission rates as these are considered the most preventable.24 Because of the combination of costs and prevention, readmissions were included in the Hospital Readmission Reduction Program (HRRP) § 3025 of the ACA. Therefore, if inputs are minimized to produce a certain level of output as with the objective function of the DEA, ignoring quality would inevitably lead to incorrect or at least questionable conclusions regarding hospital care.
Physicians integration and efficiency Pauly and Redisch25 described hospitals as a physician cooperative wherein physicians are allowed admitting privileges as long as these physicians added to total revenue. Harris26 theorized that hospitals were two firms in one—an administrative side and a medical side. The managers often acquiesce to physicians because of asymmetric information and managers cannot effectively argue with what is needed to treat patients. Pauly27 referred to hospitals as doctors’ workshops in which case physicians substituted hospital inputs for their own time. In these
4 theoretical models of physician-hospital relationships, X-inefficiency is bound to arise due to the incomplete contracts and the agency issues between hospitals and physicians. There are several levels of integration hospitals can adopt with admitting physicians. In a fully integrated model where physicians belong to only one hospital, hospitals can invest safely in physician practices without worrying about spillovers to competitors.28 Efficiency is commonly cited by hospitals and providers as one of the main reasons for vertically integrating in addition to gaining a broader patients base and achieving quality goals.29 Hospital–physicians arrangements play a key role in hospital cost-containment efforts. In recent years, hospitals have moved towards arrangements that ensure the highest levels of integration.8 While earlier research did not find support for a positive association between physician integration and lower hospital cost,9 it is important to note that the shift towards VBP might influence the strength of this relationship. Due to cost containment pressures and reimbursement changes, more formalized agreements between hospitals and physicians have been implemented with the purpose of incentivizing physicians to behave more efficiently. This is an important aspect of efficiency since Chilingerian14 argued; physician practice patterns rather than patient illness characteristics led to higher levels of inefficiency and physicians influence of approximately 80% of total health care costs.30 This finding is not surprising since physicians demand services for their patients and choose treatments, medical technology, and labor associated with their patients’ care. In light of the aforementioned policy changes, hospitals formal employment of physicians in a contractual arrangement will lead to greater market share and financial rewards.31 Even as early of 1995, Goes and Zhan11 found that physician financial integration was associated with higher occupancy rates and lower hospital operating costs. In the United States not all physicians who admit and treat patients at the hospital are employees of the hospital. Hospitals in the US have a spectrum of arrangements they can adopt with physicians. These arrangements range from physicians admitting patients and practicing at any hospital they choose, to physicians being employees and thus practicing only at one hospital. Baker et al.8 classify hospital integration relationships with physicians into four levels: fully integrated, closed physician-hospital organization, open physician-hospital organization, and independent practice association. Full integration is achieved when the hospital adopts any of the following arrangements with physicians: integrated salary model, equity model, or foundation model.8 Full integration is the tightest form of integration.8 Explanations of the three models which full under full integration are included in Table 1. In this paper, we argue that X-inefficiency should be reduced as physicians
Health Services Management Research 0(0) Table 1. Full integration explained. Full integration
Explanation
Integrated Salary Model
Physicians are salaried employees of the organization Physicians are shareholders in the organization A foundation, which is affiliated with the hospital, controls the tangible and intangible assets of the physician practice yet the practice remains under a separate ownership
Equity Model Foundation Model
Source: American Hospital Association Survey Database Fiscal Year 2013 Documentation Manual.
and hospitals economic behavior become more closely linked by moving away from the classical ‘‘doctor’s workshop’ model into a fully integrated model.
Competition and efficiency Previous research has examined the influence of hospital competition on X-inefficiency. Leibenstein32 argues that managers in noncompetitive markets, unlike managers in highly competitive markets, are not pressured to maximize their organizations’ efficiency levels. Therefore organizations in more competitive markets are more likely to be more efficient than those located in less competitive markets. Rosko33 used stochastic frontier analysis (SFA) and found a negative relationship between competition and X-inefficiency while Brown,34 also using SFA, found a positive association between hospital rivalry and X-inefficiency. Other studies which relied on DEA found no relationship between hospital competition and X-inefficiency.35 In a more recent study, Lee et al.36 found that hospitals located in more competitive markets had higher technical efficiency scores. This study however was limited to hospitals in Florida. Previous studies investigating the relationship between competition and efficiency relied on data collected before the passage of the ACA in 2010. It is important to reexamine the nature of the relationship between competition and efficiency given that competition between hospitals, before the passage of ACA, was mainly for gaining market share. However, in the post-ACA era, competition is now also for achieving better patient outcomes, patients’ experiences, and efficiency levels. In addition to the role competition plays in influencing hospital efficiency level, competition also influences quality of care. Kessler and McClellan37 reported a direct relationship between competition and lower readmission rates. Others found partial relationships between increased competition and increased quality (Mutter et al., 2008).38 Poorer outcomes were reported by
Leleu et al. Mukamel et al.,39 whereas others found no relationship between competition and quality in terms of patients’ outcomes.40
Other factors related to efficiency Ownership and teaching status have been associated with both quality and efficiency. X-efficiency Theory and Property Rights Theory suggest that internal pressures for profitability might induce FP hospitals to operate more efficiently than not-for profit hospitals.41 Conversely, hospitals in the FP sector may opt to cream-skim patients, who are less costly to treat, leaving individuals with more intense needs to the non-profit sector. Frontier studies (both DEA and SFA) have mixed results for the relationship between ownership status and efficiency.5 In addition to efficiency, previous studies indicate that NFP hospitals provided better quality as compared to FP hospitals.42 Rosko et al.43 found that teaching hospitals tended to be less efficient than non-teaching hospitals speculating that major teaching hospitals have other objectives and characteristics, such as research and teaching, which may not be consistent with higher efficiency. Joynt and Jha44 also reported that major teaching hospitals were more likely to be penalized under the HRRP than non-teaching hospitals. However, Herrin et al.,22 and Nuckols45 also warn that if hospitals are serving a lower income group of patients, they may be unduly punished for serving the needs of a sicker population. This supposition can be supported by Kupersmith46 who found that major teaching hospitals provided better quality patient care than minor or non-teaching hospitals. This suggests that readmission rates arose at all the sample hospitals, only less so at major teaching institutions. Hospital payments can also lead to increased X-inefficiency. Friedman and Basu47 reported that hospitals earned the most patient revenue from Medicare, Medicaid, and private pay patients. Leleu et al.48 corroborated this finding by demonstrating that these payer groups added to hospital revenues. Some in the political sectors have argued that the public payers are adding to inefficiency49 giving hospitals a buffer from costcontainment, a necessary component of X-inefficiency. Lower administration oversight has also been identified as a culprit for higher Medicare and Medicaid costs.49 Therefore in our model we include FP ownership, teaching status, Medicare share of admissions, and Medicaid share of admissions as predictors of hospital efficiency. There are many factors accounting for variations in hospital productive performance. In the past, crucial data were missing (including CMI, quality of care, physician contracting) that without the necessary adjustment, were considered inefficient. With a richer data set and information on inputs and outputs, we can assess both
5 the efficiency and quality as well as the environmental factors affecting both for hospitals operating in the US. Due to the improved data availability, coupled with employing a robust DEA approach, we are able to expand on earlier productivity studies to provide a more complete picture of hospital performance. In the next section, we describe the robust DEA model and model specification.
Methods We obtained data on hospital characteristics from the American Hospital Association Annual Survey Database (2013)50 and downloaded the quality indicators we incorporated in our efficiency measure from Center for Medicare and Medicaid Services Hospital Compare website. We assessed quality-adjusted efficiency using general acute non-rural hospitals that were either NFP or FP. The final sample consisted of 1847 hospitals in 2013. Due to the improved data availability, coupled with employing a robust DEA approach, we are able to expand on earlier productivity studies to provide a more complete picture of hospital performance. Estimation of hospital performance is based on a benchmarking approach which compares each hospital to a best practice frontier. This approach has its roots in the ‘activity analysis’ which was originally developed by Koopmans51 and Baumol.52 They proposed a linear programming-based technique for modeling a production technology with multiple inputs and outputs. Farrell53 defined the measure of productive inefficiency and included it in the activity analysis framework. Finally Charnes et al.53 introduced and popularized a general approach, the so-called DEA, for estimating the best practice frontier and deriving relevant operational measures of productive inefficiency. Following Charnes et al.,53 the literature for measuring inefficiency grew exponentially. The main objective of DEA can be summed up as the following: in a multidimensional space (inputs, desirable and undesirable outputs), the production technology is defined by the observed best practices of a sample of decision making units (DMUs) and results in a piecewise linear frontier from which any deviations can be interpreted as inefficiency, defined as potential improvements in productivity. The latter is simply defined as the maximization of the desirable outputs and the minimization of the undesirable outputs given a level of resources used for the production. DEA models are now relevant alternatives to the traditional statistical approach found in econometric models such as SFA. While SFA is still widely used in the modeling of production technology for firms operating in traditional market goods, DEA is appealing in the industries that do not conform to traditional competitive norms, such as the hospital sector for several reasons. First, the hospital technology is characterized by the production of multiple outputs. It would be difficult, if
6 not impossible, to validly construct a single composite output measure,43 something that would be required if this were to be used as the dependent variable. Second, the traditional alternative to production-oriented SFA is the use of a cost, revenue or profit function to deal with multiple outputs but first it requires prices that are often unavailable in the health sector and second it necessitates accepting strong assumptions on the economic behavior of DMUs to benefit from the duality theory. In particular, any of the following objectives: cost minimization, revenue maximization or profit maximization is problematic to consider as the objective of a hospital. DEA can directly be applied to a multi-output/multi-input framework and only requires data on quantity to estimate the best practice frontier. Third, the minimal assumption on the economic behavior of DMUs is to avoid waste in resources usage. While traditional regularity conditions and flexible forms can in some ways be helpful for the specification of the technology, no clear economic argument is available to use one form or another for the inefficiency term. Unlike SFA, the non-parametric DEA approach does not assume any specification forms. The observed data are enveloped by piecewise linear facets which let the data determine the form of the frontier. Since the inefficiency is measured as the distance to the frontier, no specification form is required for the distribution of inefficiency. This is the main advantage of non-parametric approaches compared to parametric ones. In short, DEA avoids the possibility of confounding the misspecification effects due to an arbitrary choice of functional forms of the technology and the inefficiency components. However, DEA models suffer from one main disadvantage compared to SFA. They are deterministic and therefore they do not allow for deviations from the efficient frontier to be a function of random error. As such, it is assumed that inputs and outputs are measured without errors, in particular there are no outlier DMUs in the data. This assumption may be too strong in empirical works. The effect of outliers may result in a shift in the best practice frontier and a bias in the measurement of inefficiency. Recently, some robust DEA approaches have been proposed to control or mitigate this effect. Inefficiency estimates derived from traditional DEA are based on the whole sample of DMUs. This is the traditional deterministic approach. However, as discussed in the previous subsection, its main drawback is the sensitivity of the frontier to outliers. To circumvent this problem, we use a robust subsampling approach. We refer interested readers to Cazals et al.,54 Kneip et al.,55 and Simar and Wilson56 for all theoretical and methodological developments. For our purposes, we only provide an intuitive presentation of the approach. If outliers exist and shift away the boundary of the estimated output production set, the robust approach mitigates this problem. This technique is implemented
Health Services Management Research 0(0) as indicated by the following steps. First, a large number of sub-samples, of a predetermined size, are selected from the initial sample of observed DMUs, then the estimated directional distance function is computed for each subsample, and the final estimate is simply the average across the sub-samples. Therefore, since the estimated output production set is based on a random selection of DMUs varies over the sub-samples, a DMU is not always compared with potential outliers. We can interpret the final result as a robust measure of the inefficiency. For details on this methodology, see Valdmanis et al.13 In addition to mitigating the potential biases due to outliers, we are also able to simultaneously address the production of economic goods (patient care) and bads (re-admissions). This is accomplished by adding another constraint in the linear programming’s specification. By combining these approaches, we are better able to model the total hospital care process.
Independent variables We follow Baker et al.8 in our classification of full integration. Baker et al.8 define fully integrated hospitals as those that adopted one of the following models with their physicians: an integrated salary, foundation or equity model. We predict that fully integrated hospitals are more likely to have lower inefficiency scores. Competition was estimated at a county level by the Herfindahl-Hirschman Index (HHI) calculated by summing the squares of market shares of hospital admissions. For ease of interpretation, competition was coded as 1-HHI. Teaching status was incorporated by the use of two binary variables: major teaching (i.e. hospital is a member of Council of Teaching Hospital (COTH) and other teaching (i.e. hospital has a residency training approval by Accreditation Council for Graduate Medical Education but does not belong to COTH). We classify hospitals as high technology if they offer any of the following services: heart or any major organ transplant, computer assisted orthopedic surgery or electron beam computed tomography. Additional control variables include Medicare share of hospital admissions and Medicaid share of hospital admissions. All of these variables were derived or estimated using data from the AHA Annual Survey Database.
Results In Table 2, we provide the descriptive statistics of the inputs, outputs, a control variable (i.e., Medicare CaseMix Index (MCMI)), environmental factors, and efficiency scores. Average 30-day risk standardized mortality and readmission rates for the year 2013 are publicly reported and were derived from CMS. All outcomes measures were reported twice in 2013 and therefore we
Leleu et al.
7
Table 2. Descriptive statistics of the inputs and outputs used in measuring inefficiency. Variable Bad outputs Mortality rate: heart attack Mortality rate: heart failure Mortality rate: pneumonia Readmission rate: heart attack Readmission rate: heart failure Readmission rate: pneumonia readmission rate: knee, hip Readmission rate: all Good outputs Number of admissions (no surgical) Number of inpatient surgeries Number of outpatient surgeries Number of ER visits Number of other visits Inputs Case Mix Index FTE RN þ LPN Other trainees FTE other Beds Post admission days (proxy for resource use)
Standard deviation
Mean
15.19% 11.60% 11.80% 18.96% 23.84% 18.15% 5.40% 15.95%
1.44 1.33 1.75 1.29 1.91 1.54 0.73 1.13
9494.61 3650.69 6349.64 48,324.57 186,834.60
7689.38 3666.82 6020.77 33,752.37 253,863.50
1.59 523.76 523.76 1108.15 279.21 52,528.13
0.22 530.82 148.81 1299.71 216.75 0.03
ER: emergency room; FTE RN þ LPN: full time equivalent registered nurses and licensed practical nurses.
were able to calculate the average rates except for 30-day all-cause readmission rate which was reported only once. The average hospital mortality rates are between 11.60% and 15.19%. There is more variability among the hospitals in our sample for the 30 day readmission rates for the various diagnosis conditions, with an average overall 30-day readmission rate of 15.95%. The descriptive statistics for inputs and outputs demonstrate a wide variability among the hospitals in our sample showing that opting for the robust DEA approach was the correct model choice. On average, the hospitals in our sample have an MCMI of 1.59 but demonstrates less variability than the other outputs in the hospital sample. As in the case of outputs, the descriptive statistics on inputs also show a wide variability. In our DEA results (Table 3), 1158 hospitals (63%) exhibited inefficiency. The mean inefficiency score is 10.5%. This means that on average inefficient hospitals could simultaneously increase the good outputs and decrease the bad outputs by 10.5% by keeping the same level of inputs. Scores ranged from 0.64 to 0.44. It must be remembered that by virtue of the robust DEA model, we allow for some of the hospitals to be above the
Table 3. Robust DEA results.
Number Percent Robust DEA scores
Efficient hospitals
Inefficient hospitals
Total
689 27% 10.20%
1158 63% 10.50%
1847 100% 2.75%
DEA: data envelopment analysis.
frontier. Since we measure inefficiency, negative scores mean that some hospitals outperform other hospitals. For inefficient hospitals, scores are positive. In our results, 63% of the hospitals exhibited inefficiency. In terms of the organizational environmental factors (Table 3), we report that 42% of the hospitals employed a fully integrated physician model. The average competition index (1-HHI) is 0.57. The share of patients covered by Medicare is approximately 48%. In comparison the percent of patients covered by Medicaid is 18%. Nine percent of the hospitals in our sample are considered as major teaching, i.e. academic medical centers (AMC) belonging to COTH whereas 24% were minor teaching. We also report that 20% of our sample are for-profit and 42% provided high medical technological services. In light of the wide range of the robust DEA measures, it is incumbent upon us to regress these findings on an array of independent variables we believe affect the efficiency (including quality) measure (Table 4). From the regression model (reported in Table 5), we find that FP hospitals were statistically significantly (p < 0.05) more efficient than their NFP counterparts. Similarly, hospitals with highly technological medical equipment, a fully integrated physician work force, and a more competitive environment were more efficient while controlling for quality in the DEA model. Interestingly, the percentage of patients who were covered by Medicare and Medicaid were associated with hospitals increased inefficiency. AMCs and other teaching hospitals were also more efficient than non-teaching hospitals.
Discussion and managerial implications Our research extends earlier work by employing a robust DEA model that includes the joint production of economic ‘‘goods’’ and ‘‘bads’’ given the hospital production process. Due to the derivation of the efficiency measures using the robust approach, we are able to use the findings in regression analysis in order to ascertain the organizational and market based factors which explain the variability of quality adjusted efficiency scores between hospitals. By including the joint production, performance is not upward biased as would be the case if ‘‘bads’’ were excluded. For example, a hospital producing more outputs than another hospital may be deemed as
8
Health Services Management Research 0(0)
Table 4. Descriptive statistics of predictor variables. Variable
Mean (SD)
Standard deviation
Fully integrated hospitals
Competition
0.57
Proportion
Description
42%
Hospitals with one of the following physician integration models: integrated salary, foundation or equity model
0.32
Medicare share
47.91
9.43
Medicaid share
18.03
8.34
Major teaching
9%
Other teaching
24%
For-profit High technology
20% 42%
1-HHI; HHI was calculated by summing the squares of market shares of admissions per county Number of Medicare admissions 100 divided by total hospital discharges Number of Medicaid admissions 100 divided by total hospital discharges Hospital is a member of Council of Teaching Hospital of the Association of American Medical Colleges (COTH) Hospital has a residency training approval by Accreditation Council for Graduate Medical Education but does not belong to COTH For-profit ownership Hospital offers any of the following services: heart or any major organ transplant, computer assisted orthopedic surgery or electron beam computed tomography
HHI: Hirschman-Herfindahl Index.
Table 5. Regression Model results for inefficiency scores. Variable
Parameter
Constant For-profit ownership Percent of patients covered by Medicare Percent of patients covered by Medicaid High technological services Fully integrated physicians Competition Academic medical center Other teaching
0.028 0.028*** 0.002*** 0.001*** 0.036*** 0.025*** 0.024*** 0.032*** 0.018**
N ¼ 1847; adjusted R-squared ¼ 0.08; F ¼ 17.27. **p < 0.05. ***p < 0.01.
best practice. However, if this same hospital had more readmissions and higher mortality rates, this would indicate that the care provided was inferior, which is not compatible with a best practice hospital. Our findings indicate that hospitals with fully integrated physicians are more efficient. This is not surprising since as Chilingerian and Sherman30 argue, physicians control 80% of health care costs. While we are constrained in
the way we measure physician integration by the availability of complete data on hospital physician relationships, we believe our study still contributes to our understanding of the relationship between physician integration level and hospital cost and quality. We classified hospitals as fully integrated if they implemented the: integrated salaried model, equity model, or foundation model.8 Integrated salaried model simply indicates that the hospital employs its physicians while the equity model refers to arrangements where physicians own shares in the organization.12 In the foundation model acquired group practices are aggregated under a nonprofit entity whereby the hospital provides professional services and plays a key role in the management of the foundation.12 The integrated salaried model is by far the most common of the three full integration models listed. Hospitals currently employ more than half of the physician workforce.57 Some argue that physician employment could negatively influence hospitals since a hospital loses $150,000 to $250,000 during the first year it employees a physician.57 Based on our findings though, despite the high upfront cost, full integration, including physician employment, strengthens the ties between hospitals and their physicians which allows hospitals to achieve higher levels of efficiency while maintaining quality. This is especially important for healthcare managers
Leleu et al. as they struggle to configure physician contracts and incentives in a manner which maximizes physicians’ productivity and their adherence to evidence based care. As predicted by X-inefficiency theory, hospitals located in more competitive market are more efficient. This is not surprising, as hospitals operate in the era of VBP where they are competing for higher levels of efficiency and better quality. Managers of hospitals located in less competitive markets need to be aware of this difference and need to realize that they are more likely to operate with higher levels of slacks. Waste need to be identified and proactively reduced. Market competitiveness could change at any time and inefficient hospitals face the risk of losing market share when more efficient hospitals enter their market. The findings are also interesting in that teaching hospitals, particularly AMC are efficient vis-a`-vis non-teaching hospitals. This result should allay fears that teaching hospitals will be punished via higher HRRP penalties. Our results suggest that both efficiency and quality should be vigorously applied when attaching HRRP monitoring. Our findings indicate the for-profit hospitals are more efficient. While it is expected that for-profit hospitals are more efficient than their NFP counterparts, it is time for managers of NFP hospitals to create a major shift in organizational culture and operations to achieve the same levels of efficiency as for-profit hospitals. VBP incorporates efficiency as one of the key domains and therefore this new era of healthcare does not discriminate between FP and NFP in terms of efficiency and quality expectations. By combining quality and efficiency in our measurement of productivity performance, we added to both the methodological and data literature on hospitals. This methodology of measuring efficiency allowed us to more reliably examine relevant environmental factors, especially hospital–physicians arrangements, heretofore ignored, but now can be included. To summarize, we find that ownership, payer type, physician integration, competition, and teaching status all impact the quality/ efficiency dynamic. Healthcare managers should be aware of these factors as they make decisions and implement initiatives to lower cost and improve quality. Moreover, policy makers should review these factors as well as provide incentives that ensure higher quality via readmission penalties, but also account for environmental factors and more efficient treatment modalities. Declaration of conflicting interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
9 Note a. See Hollingsworth15 and O’Neill et al.16 for a review of these studies.
References 1. Negrini D, Kettle A, Sheppard L, et al. The cost of a hospital ward in Europe: is there a methodology available to accurately measure the costs? J Health Organ Manag 2004; 18: 195–206. 2. Nayar P, Ozcan YA, Yu F, et al. Benchmarking urban acute care hospitals: efficiency and quality perspectives. Health Care Manag Rev 2013; 38: 137–145. 3. Dixon J. Making progress on efficiency in the NHS in England: options for system reform. Nuffield Trust. 4. Zinn J and Flood A. Commentary: slack resources in health care organizations – fat to be trimmed or muscle to be exercised? Health Serv Res 2009; 44: 812–820. 5. Valdmanis V, Rosko M and Mutter R. Hospital quality, efficiency, and input slack differentials. Health Serv Res 2008; 43: 1830–1848. 6. Ozcan YA and Luke RD. A national study of the efficiency of hospitals in urban markets. Health Serv Res 1993; 27: 719. 7. Madison K. Hospital–physician affiliations and patient treatments, expenditures, and outcomes. Health Serv Res 2004; 39: 257–278. 8. Baker LC, Bundorf MK and Kessler DP. Vertical integration: hospital ownership of physician practices is associated with higher prices and spending. Health Aff 2014; 33: 756–763. 9. Alexander J and Morrisey M. Hospital–physician integration and hospital costs. Inquiry 1988; 25: 388–401. 10. Wan T, Lin B and Ma A. Integration mechanisms and hospital efficiency in integrated health care delivery systems. J Med Sys 2002; 26: 127–143. 11. Goes JB and Zhan C. The effects of hospital–physician integration strategies on hospital financial performance. Health Serv Res 1995; 30: 507. 12. Burns L and Muller R. Hospital-physician collaboration: landscape of economic integration and impact on clinical integration. Milbank Q 2008; 86: 375–434. 13. Valdmanis V, Rosko M, Mukamel D, et al. Assessing overall, technical, and scale efficiency among home health agencies. Health Care Manag Sci 2017; 20: 265–275. 14. Chilingerian J. Evaluating physician efficiency in hospitals: a multivariate analysis of best practices. Eur J Oper Res 1995; 80: 548–574. 15. Hollingsworth B. The measurement of efficiency and productivity of health care delivery. Health Econ 2008; 17: 1107–1128. 16. O’Neill L, Rauner M, Heidenberger K, et al. A crossnational comparison and taxonomy of DEA-based hospital efficiency studies. Socio-Econ Plann Sci 2008; 42: 158–189. 17. Nayar P and Ozcan YA. Data envelopment analysis comparison of hospital efficiency and quality. J Med Sys 2008; 32: 193–199. 18. Clement J, Valdmanis V, Bazzoli G, et al. Is more better? An analysis of hospital outcomes and efficiency with a DEA model of output congestion. Health Care Manag Sci 2008; 11: 67–77.
10 19. McCormack R, Michels R, Ramos N, et al. Thirty-day readmission rates as a measure of quality: causes of readmission after orthopedic surgeries and accuracy of administrative data. J Healthc Manag 2013; 58: 64–77. 20. Jencks SF, Williams MV and Coleman EA. Rehospitalizations among patients in the Medicare feefor-service program. N Engl J Med 2012; 366: 1364–1366. 21. Berenson R, Paulus R and Kalman N. Medicare’s readmissions-reduction program – a positive alternative. NEJM 2012; 366: 1364–1366. 22. Gohil SK, Datta R, Cao C, et al. Impact of hospital population case-mix, including poverty, on hospital all-cause and infection-related 30-day readmission rates. Clin Infect Dis 2015; 61: 1235–1243. 23. Herrin J, St Andre J, Kenward K, et al. Community factors and hospital readmission rates. Health Serv Res 2015; 50: 20–39. 24. Benbassat J and Taragin M. Hospital readmissions as a measure of quality of health care: advantages and limitations. Arch Intern Med 2000; 160: 1074–1081. 25. Pauly M and Redisch M. The not-for-profit hospital as a physicians’ cooperative. Am Econ Rev 1973; 63: 87–99. 26. Harris J. The internal organization of hospitals: some economic implications. Bell J Econ 1977; 467–482. 27. Pauly M. Doctors and their workshops. Chicago: University of Chicago Press, 1980. 28. Robinson J. Physician-hospital integration and the economic theory of the firm. Med Care Res Rev 1997; 54: 3–24. 29. Burns L and Pauly M. Integrated delivery networks: a detour on the road to the road to integrated health care? Health Aff 2002; 21: 128–143. 30. Chilingerian J and Sherman H. Managing physician efficiency and effectiveness in providing hospital services. Health Serv Manag Res 1990; 3: 3–15. 31. O’Malley A, Bond A and Berenson R. Rising hospital employment of physicians: better quality, higher costs. Issue Brief Cent Stud Health Syst Change 2011; 136: 1–4. 32. Leibenstein H. Allocative efficiency vs. X-efficiency. Am Econ Rev 1966; 56: 392–415. 33. Rosko M. Impact of HMO penetration and other environmental factors on hospital X-inefficiency. Med Care Res Rev 2001; 58: 430–454. 34. Brown HS. Managed care and technical efficiency. Health Econ 2003; 12: 149–158. 35. Bates L, Mukherjee K and Santerre R. Market structure and technical efficiency in the hospital services industry: a DEA approach. Med Care Res Rev 2006; 63: 499–524. 36. Lee KH, Park J, Lim S, et al. Has competition increased hospital technical efficiency? Health Care Manager 2015; 34: 106–112. 37. Kessler D and McClellan M. Is hospital competition socially wasteful? Q J Econ 2000; 115: 577–615. 38. Mutter RL, Wong HS and Goldfarb MG. The effects of hospital competition on inpatient quality of care. INQUIRY: The Journal of Health Care Organization, Provision, and Financing 2008; 45(3): 263–279. 39. Mukamel D, Zwanziger J and Bamezai A. Hospital competition, resource allocation and quality of care. BMC Health Serv Res 2002; 2: 2–10.
Health Services Management Research 0(0) 40. Mukamel D, Zwanziger J and Tomaszewski K. HMO penetration, competition, and risk-adjusted hospital mortality. Health Serv Res 2001; 36: 1019–1028. 41. Mutter R and Rosko M. The impact of ownership on hospital cost-efficiency of U.S. hospitals. In: Blank J and Valdmanis V (eds) Measuring hospital productivity and efficiency advances in health economics and health services research. Vol. 18, Amsterdam: Elsevier Press, 2007, pp.113–138. 42. Landon BE, Normand SL, Lessler A, et al. Quality of care for the treatment of acute medical conditions in US hospitals. Arch Intern Med 2006; 166: 2511–2517. 43. Rosko M, Wong H and Mutter R. Characteristics of highand low-efficiency hospitals. Med Care Res Rev. Epub ahead of print 2017. doi/full/10.1177/1077558716689197. 44. Joynt K and Jha A. Characteristics of hospitals receiving penalties under the hospital readmissions reduction program. JAMA 2013; 309: 342–343. 45. Nuckols T. County-level variation in readmission rates: implications for the Hospital Readmission Reduction Program’s potential to succeed. Health Serv Res 2015; 50: 12–19. 46. Kupersmith J. Quality of care in teaching hospitals: a literature review. Acad Med 2005; 80: 458–466. 47. Friedman B and Basu J. The rate and cost of hospital readmissions for preventable conditions. Med Care Res Rev 2004; 61: 225–240. 48. Leleu H, Moises J and Valdmanis VG. How do payer mix and technical inefficiency affect hospital profit? A weighted DEA approach. Oper Res Health Care 2014; 3: 231–237. 49. Saving T and Goodman J. A better way to approach Medicare’s impossible task. Health Aff Blog, Available at: http://healthaffairs.org/blog/2011/11/15/a-better-way-toapproach-medicares-impossible-task/ (accessed 15 November 2011). 50. American Hospital Association (AHA). American Hospital Association (AHA) Annual Survey Database. Chicago, IL: American Hospital Association (AHA), Health Forum. Codebook: HE-112(2013). 51. Koopmans T. In: Koopmans TC (ed.) An analysis of production as an efficient combination of activities, activity analysis of production and allocation. London: Wiley, 1951. 52. Baumol W. Activity analysis in one lesson. Am Econ Rev 1958; 48: 837–873. 53. Farrell M. The measurement of productive efficiency. J R Stat Soc Ser A 1957; 120: 253–290. 54. Charnes A, Cooper W and Rhodes E. Measuring the efficiency of decision making units. Eur J Oper Res 1978; 2: 429–444. 55. Cazals C, Florens J and Simar L. Nonparametric frontier estimation: a robust approach. J Econometrics 2002; 106: 1–25. 56. Kneip A, Simar L and Wilson PW. Asymptotics and consistent bootstraps for DEA estimators in nonparametric frontier models. Econ Theory 2008; 24: 1663–1697. 57. Simar L and Wilson P. Two-stage DEA: caveat emptor. J Prod Anal 2011; 36: 205–218. 58. Kocher R and Sahni N. Hospitals’ race to employ physicians – the logic behind a money-losing proposition. NEJM 2011; 364: 1790–1793.
A robust analysis of hospital efficiency and factors affecting variability
Health Services Management Research 0(0) 1–10 ! The Author(s) 2017 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0951484817730537 journals.sagepub.com/home/hsm
Herve Leleu1, Mona Al-Amin2, Michael Rosko3 and Vivian G Valdmanis4
Abstract The objectives of this paper are to use data envelopment analysis to measure hospital inefficiency in a way that accounts for patient outcomes and to study the association between organizational factors, such as hospital–physicians integration level and teaching status, and market competition with hospital inefficiency. We apply the robust data envelopment analysis approach to a sample of private (both not-for-profit and for-profit) hospitals operating in the United States. Our data envelopment analysis model includes mortality and readmission rates as bad outputs and admissions, surgeries, emergency room, and other visits as good outputs. Therefore, our measurement of hospital inefficiency accounts for quality. We then use a subsampling regression analysis to determine the predictors of hospital inefficiency. For-profit, fully integrated and teaching hospitals were more efficient than their counterparts. Also hospitals located in more competitive markets were more efficient than those located in less competitive markets. Incorporating quality in the measurement of hospital efficiency is key for producing valid efficiency scores. Hospitals in less competitive markets need to improve their efficiency levels. Moreover, high levels of hospital physician integration might be instrumental in ensuring that hospitals achieve their efficiency goals.
Keywords efficiency, health care organizations, hospitals, integration
Introduction Health care spending is on the rise worldwide and as expensive new technologies emerge and populations age, there is a strong incentive to gain better understanding of health care costs.1 Interest in hospital costs specifically mounted since the early 1970s given the economic burdens hospital costs place on healthcare systems in the United States and elsewhere. In recent years, the interest in hospital cost and quality has intensified in the US and was finally formalized in the Patient Protection and Affordable Care Act, 2010 (ACA). Value Based Purchasing (VBP) places hospitals under financial pressure to deliver care of the highest quality and with optimal efficiency.2 Other health care systems, such as the English NHS, are also struggling to improve their efficiency and contain cost.3 Understanding the drivers of hospital costs and efficiency is therefore paramount to policymakers and healthcare managers alike. However, improving efficiency by either lowering cost or increasing output is of no value in healthcare unless quality is maintained. As Zinn and Flood4 argue, limited studies on hospital efficiency accounted for quality but the need to control or lower cost while maintaining, if not improving, quality ‘‘has never been more urgent or complex’’. Over the years,
data on hospital outcomes have become publicly available through the Center for Medicare and Medicaid Services (CMS), which provides an ample opportunity to improve the validity and reliability of modeling hospital efficiency while accounting for patient outcomes and to determine the predictors of hospital efficiency. Previous research indicates that certain hospital characteristics, such as for-profit (FP) ownership, size, and teaching status, are associated with hospital efficiency.5,6 However, other organizational factors, such as hospital– physician relationships, have been less studied. Hospitals have a variety of integration arrangements they can adopt with their physicians; these arrangements range from very 1
CNRS-LEM (UMR 9221), IESEG School of Management, Lille, France Suffolk University, Boston, MA, USA 3 Widener University, Chester, PA, USA 4 Western Michigan University, Program in Public Health, Grand Rapids, MI, USA 2
Corresponding author: Mona Al-Amin, Healthcare Administration Department, Sawyer Business School, Suffolk University, 120 Tremont Street, Room 5603, Boston, MA 02108, USA. Email: [email protected]
2 tight integration such as the salaried or employment model to loose arrangements such as group practice without walls. Integration arrangements are important because they influence hospital’s ability to control cost, improve quality, increase market share, and achieve other organizational goals.7 In recent years, hospitals have moved towards tight arrangements, such as employment, that ensure the highest levels of integration.8 While earlier research did not find support for the relationship between physician integration and lower hospital cost,9 it is important to note that this era, specifically post the ACA, is different from the 1980s and 1990s since VBP, public reporting, and bundled payments were not common then. Previous research indicates that physician integration mechanisms influence hospital efficiency and operating margins,10,11 however, evidence on the effect of integration levels on hospital efficiency and patient outcomes is insufficient and inconsistent.12 In a more recent study, Baker et al.8 relied on data from a private feefor-service insurance plan to investigate the impact of physician integration strategies on healthcare spending. They found that markets with hospitals with the highest level of physician integration have higher average price per hospital admission and spending per enrollee. Baker et al.8 focused on market level spending, however, more research is needed to understand this relationship at the hospital level. Further research on the relationship between physician integration levels and hospital efficiency is needed. The aim of this paper is two-fold. First, we measure efficiency and adjust for more precise quality measures than included in previous work. This quality adjustment is an improvement over earlier data envelopment analysis (DEA) studies, which focused on the intermediate good of hospital production, i.e. inpatient admissions, number of surgeries et cetera. By including readmission rates and mortality, we can directly measure the quality of hospital services not just the quantity. We are able to include these patient outcome measures because access is now available to publicly reported quality measures through CMS; these measures were not previously available. Second, we investigate the association between market factors and organizational characteristics such as market competition, ownership, level of physician integration, and teaching status with this improved efficiency measure. To derive our efficiency measures, we apply the robust bootstrapping DEA approach to a sample of private (both not-for-profit (NFP) and FP) hospitals operating in the United States. The robust approach allows us to avoid two main shortcomings of traditional DEA approaches. Primarily, it controls for outliers in the derivation of efficiency scores and can be viewed as an intermediate between deterministic and stochastic methods. Secondly, it allows for the utilization of a relevant bootstrap inference in the second stage where determinants of
Health Services Management Research 0(0) inefficiency are analyzed through a regression analysis. This approach had been used in the assessment of home health care services.13
Conceptual framework One of the major theoretical reasons for using DEA in the study of hospitals is that it is a good method for measuring and assessing the underlying causes of hospital X-inefficiency.14 The basic concept of the X-inefficiency theory is that under the proper motivation, managers and employees could behave to operate closer to optimality and when these motivations are absent the distance from optimality increases. There are several reasons for X-inefficiency including: incomplete labor contracts, not all factors of production are marketable given alternative pricing and reimbursement mechanisms, and the production function may change given technology and payment scheme variations. X-inefficiency has also been a popular theoretical framework when describing the NFP hospitals in the US health care system. In the healthcare sector, property rights arrangements that attenuate owner (i.e. stock-holder) wealth decreases the opportunity costs of non-wealth maximizing behavior decreases. Given the market failures in producing hospital care, including asymmetric information, stochastic demand, non-price competition, and the contracting between hospitals and payers, X-inefficiency increases results. A summary of our model is presented in Figure 1 and discussed in the proceeding sections.
DEA background The DEA approach, which uses a best practice frontier to gauge the relative efficiency of hospitals to a ‘‘best practice’’ frontier, is practical for several reasons. First, the economic regularity conditions required for a production function are met including that there is no ‘‘free lunch’’ (inputs are required to produce outputs) and that there is strong disposability of inputs and outputs (outputs are non-decreasing in inputs). A second reason that the DEA is appropriate is its ability to employ multiple inputs and outputs in the specification of the linear programming model from which results are derived. This is important because of the multiple outputs hospitals produce using multiple inputs. A third benefit is the DEA’s ability to use inputs and outputs in their natural units rather than converting them into prices which are often missing. And, finally, there is no requirement to specify a traditional cost minimizing or profit maximizing assumption on the data, an assumption that may not be valid since hospitals do not operate by competitive market norms which includes many buyers and sellers, all information is conveyed by price, products are identical, and free entry and exit. Even though there are FP hospitals in
Leleu et al.
3 Organizaonal Factors
Hospital Efficiency
Physician Integraon Level For-profit Teaching Status Technology Level Medicaid Admissions Medicare Admissions
Good Outputs Surgery Volume Visit Volume
Market Factors Compeon
Bad Outputs 30-day Readmission Rates 30-day Mortality Rates Inputs Case Mix Index Staffing Beds
Figure 1. Predictors of hospital efficiency.
the industry, they cannot conform to the traditional definition of a competitive market. DEA analysis of hospital productivity has been a staple in the health care productivity/efficiency literature since the 1980s.a Advances in DEA included more detailed specifications to separate inefficiency from factors that either endogenously or exogenously affect performance. There was concern that hospitals treating a more severe patient load would be deemed inefficient simply because their patients, by nature, require more hospital services, At first, differentiation by age of the patient (Medicare versus non-Medicare), types of inpatient services (acute versus intensive care), emergency room visits versus outpatient visits, and number of surgeries were used since more processes were needed in terms of nursing care, ancillary services, and beds as patients require these services to become healthier. This estimation of patient disease differentiation was supplanted by the addition of case mix indices to adjust for the types of patients treated more precisely. Another issue missing from earlier research using the DEA approach is the notion of quality. However, more recent studies have begun to expand on the use of quality care indicators. By relaxing the strong disposability assumption, ‘‘congestion’’ can arise wherein economic ‘‘bads’’ increase with the production of economic ‘‘goods’’. Treatment indicators including antibiotic timing, percent of patients given an oxygenation, and the percent of patients given a pneumococcal vaccine were used by Nayar and Ozcan17 on a sample of hospitals operating in Virginia. Using a national sample, Valdmanis et al.5 used failure to rescue, infection due to medical care, postoperative respiratory failure, and postoperative sepsis as ‘‘bad’’ outcomes that were measured as economic congestion. In other words, more hospital care is associated with an increase in the number of bad outcomes. For example, longer length of stays are often considered an economic good since more care often leads
to better outcomes but may also result in bad outcomes such as hospital-acquired infections. Clement et al.18 also assessed congestion noting that more care led to more bad outcomes, which is not surprising given the high levels of iatrogenic diseases reported in hospital. Readmission rates are currently being used as a key measure for assessing and comparing quality of care between hospitals.19 It has been estimated that the 20% of readmission rates for Medicare patients cost an additional $17 billion dollars (Jenks et al., 2009).20 This has induced CMS to impose financial penalties on hospitals with high 30-day readmission rates.21,22 The most commonly used readmission rates are for heart failure, acute myocardial infarction, and pneumonia that occur within 30 days of a patient’s discharge (Davis et al., 2013).23 Since readmissions are costly, it has been recommended that hospital studies should include the 30-day readmission rates as these are considered the most preventable.24 Because of the combination of costs and prevention, readmissions were included in the Hospital Readmission Reduction Program (HRRP) § 3025 of the ACA. Therefore, if inputs are minimized to produce a certain level of output as with the objective function of the DEA, ignoring quality would inevitably lead to incorrect or at least questionable conclusions regarding hospital care.
Physicians integration and efficiency Pauly and Redisch25 described hospitals as a physician cooperative wherein physicians are allowed admitting privileges as long as these physicians added to total revenue. Harris26 theorized that hospitals were two firms in one—an administrative side and a medical side. The managers often acquiesce to physicians because of asymmetric information and managers cannot effectively argue with what is needed to treat patients. Pauly27 referred to hospitals as doctors’ workshops in which case physicians substituted hospital inputs for their own time. In these
4 theoretical models of physician-hospital relationships, X-inefficiency is bound to arise due to the incomplete contracts and the agency issues between hospitals and physicians. There are several levels of integration hospitals can adopt with admitting physicians. In a fully integrated model where physicians belong to only one hospital, hospitals can invest safely in physician practices without worrying about spillovers to competitors.28 Efficiency is commonly cited by hospitals and providers as one of the main reasons for vertically integrating in addition to gaining a broader patients base and achieving quality goals.29 Hospital–physicians arrangements play a key role in hospital cost-containment efforts. In recent years, hospitals have moved towards arrangements that ensure the highest levels of integration.8 While earlier research did not find support for a positive association between physician integration and lower hospital cost,9 it is important to note that the shift towards VBP might influence the strength of this relationship. Due to cost containment pressures and reimbursement changes, more formalized agreements between hospitals and physicians have been implemented with the purpose of incentivizing physicians to behave more efficiently. This is an important aspect of efficiency since Chilingerian14 argued; physician practice patterns rather than patient illness characteristics led to higher levels of inefficiency and physicians influence of approximately 80% of total health care costs.30 This finding is not surprising since physicians demand services for their patients and choose treatments, medical technology, and labor associated with their patients’ care. In light of the aforementioned policy changes, hospitals formal employment of physicians in a contractual arrangement will lead to greater market share and financial rewards.31 Even as early of 1995, Goes and Zhan11 found that physician financial integration was associated with higher occupancy rates and lower hospital operating costs. In the United States not all physicians who admit and treat patients at the hospital are employees of the hospital. Hospitals in the US have a spectrum of arrangements they can adopt with physicians. These arrangements range from physicians admitting patients and practicing at any hospital they choose, to physicians being employees and thus practicing only at one hospital. Baker et al.8 classify hospital integration relationships with physicians into four levels: fully integrated, closed physician-hospital organization, open physician-hospital organization, and independent practice association. Full integration is achieved when the hospital adopts any of the following arrangements with physicians: integrated salary model, equity model, or foundation model.8 Full integration is the tightest form of integration.8 Explanations of the three models which full under full integration are included in Table 1. In this paper, we argue that X-inefficiency should be reduced as physicians
Health Services Management Research 0(0) Table 1. Full integration explained. Full integration
Explanation
Integrated Salary Model
Physicians are salaried employees of the organization Physicians are shareholders in the organization A foundation, which is affiliated with the hospital, controls the tangible and intangible assets of the physician practice yet the practice remains under a separate ownership
Equity Model Foundation Model
Source: American Hospital Association Survey Database Fiscal Year 2013 Documentation Manual.
and hospitals economic behavior become more closely linked by moving away from the classical ‘‘doctor’s workshop’ model into a fully integrated model.
Competition and efficiency Previous research has examined the influence of hospital competition on X-inefficiency. Leibenstein32 argues that managers in noncompetitive markets, unlike managers in highly competitive markets, are not pressured to maximize their organizations’ efficiency levels. Therefore organizations in more competitive markets are more likely to be more efficient than those located in less competitive markets. Rosko33 used stochastic frontier analysis (SFA) and found a negative relationship between competition and X-inefficiency while Brown,34 also using SFA, found a positive association between hospital rivalry and X-inefficiency. Other studies which relied on DEA found no relationship between hospital competition and X-inefficiency.35 In a more recent study, Lee et al.36 found that hospitals located in more competitive markets had higher technical efficiency scores. This study however was limited to hospitals in Florida. Previous studies investigating the relationship between competition and efficiency relied on data collected before the passage of the ACA in 2010. It is important to reexamine the nature of the relationship between competition and efficiency given that competition between hospitals, before the passage of ACA, was mainly for gaining market share. However, in the post-ACA era, competition is now also for achieving better patient outcomes, patients’ experiences, and efficiency levels. In addition to the role competition plays in influencing hospital efficiency level, competition also influences quality of care. Kessler and McClellan37 reported a direct relationship between competition and lower readmission rates. Others found partial relationships between increased competition and increased quality (Mutter et al., 2008).38 Poorer outcomes were reported by
Leleu et al. Mukamel et al.,39 whereas others found no relationship between competition and quality in terms of patients’ outcomes.40
Other factors related to efficiency Ownership and teaching status have been associated with both quality and efficiency. X-efficiency Theory and Property Rights Theory suggest that internal pressures for profitability might induce FP hospitals to operate more efficiently than not-for profit hospitals.41 Conversely, hospitals in the FP sector may opt to cream-skim patients, who are less costly to treat, leaving individuals with more intense needs to the non-profit sector. Frontier studies (both DEA and SFA) have mixed results for the relationship between ownership status and efficiency.5 In addition to efficiency, previous studies indicate that NFP hospitals provided better quality as compared to FP hospitals.42 Rosko et al.43 found that teaching hospitals tended to be less efficient than non-teaching hospitals speculating that major teaching hospitals have other objectives and characteristics, such as research and teaching, which may not be consistent with higher efficiency. Joynt and Jha44 also reported that major teaching hospitals were more likely to be penalized under the HRRP than non-teaching hospitals. However, Herrin et al.,22 and Nuckols45 also warn that if hospitals are serving a lower income group of patients, they may be unduly punished for serving the needs of a sicker population. This supposition can be supported by Kupersmith46 who found that major teaching hospitals provided better quality patient care than minor or non-teaching hospitals. This suggests that readmission rates arose at all the sample hospitals, only less so at major teaching institutions. Hospital payments can also lead to increased X-inefficiency. Friedman and Basu47 reported that hospitals earned the most patient revenue from Medicare, Medicaid, and private pay patients. Leleu et al.48 corroborated this finding by demonstrating that these payer groups added to hospital revenues. Some in the political sectors have argued that the public payers are adding to inefficiency49 giving hospitals a buffer from costcontainment, a necessary component of X-inefficiency. Lower administration oversight has also been identified as a culprit for higher Medicare and Medicaid costs.49 Therefore in our model we include FP ownership, teaching status, Medicare share of admissions, and Medicaid share of admissions as predictors of hospital efficiency. There are many factors accounting for variations in hospital productive performance. In the past, crucial data were missing (including CMI, quality of care, physician contracting) that without the necessary adjustment, were considered inefficient. With a richer data set and information on inputs and outputs, we can assess both
5 the efficiency and quality as well as the environmental factors affecting both for hospitals operating in the US. Due to the improved data availability, coupled with employing a robust DEA approach, we are able to expand on earlier productivity studies to provide a more complete picture of hospital performance. In the next section, we describe the robust DEA model and model specification.
Methods We obtained data on hospital characteristics from the American Hospital Association Annual Survey Database (2013)50 and downloaded the quality indicators we incorporated in our efficiency measure from Center for Medicare and Medicaid Services Hospital Compare website. We assessed quality-adjusted efficiency using general acute non-rural hospitals that were either NFP or FP. The final sample consisted of 1847 hospitals in 2013. Due to the improved data availability, coupled with employing a robust DEA approach, we are able to expand on earlier productivity studies to provide a more complete picture of hospital performance. Estimation of hospital performance is based on a benchmarking approach which compares each hospital to a best practice frontier. This approach has its roots in the ‘activity analysis’ which was originally developed by Koopmans51 and Baumol.52 They proposed a linear programming-based technique for modeling a production technology with multiple inputs and outputs. Farrell53 defined the measure of productive inefficiency and included it in the activity analysis framework. Finally Charnes et al.53 introduced and popularized a general approach, the so-called DEA, for estimating the best practice frontier and deriving relevant operational measures of productive inefficiency. Following Charnes et al.,53 the literature for measuring inefficiency grew exponentially. The main objective of DEA can be summed up as the following: in a multidimensional space (inputs, desirable and undesirable outputs), the production technology is defined by the observed best practices of a sample of decision making units (DMUs) and results in a piecewise linear frontier from which any deviations can be interpreted as inefficiency, defined as potential improvements in productivity. The latter is simply defined as the maximization of the desirable outputs and the minimization of the undesirable outputs given a level of resources used for the production. DEA models are now relevant alternatives to the traditional statistical approach found in econometric models such as SFA. While SFA is still widely used in the modeling of production technology for firms operating in traditional market goods, DEA is appealing in the industries that do not conform to traditional competitive norms, such as the hospital sector for several reasons. First, the hospital technology is characterized by the production of multiple outputs. It would be difficult, if
6 not impossible, to validly construct a single composite output measure,43 something that would be required if this were to be used as the dependent variable. Second, the traditional alternative to production-oriented SFA is the use of a cost, revenue or profit function to deal with multiple outputs but first it requires prices that are often unavailable in the health sector and second it necessitates accepting strong assumptions on the economic behavior of DMUs to benefit from the duality theory. In particular, any of the following objectives: cost minimization, revenue maximization or profit maximization is problematic to consider as the objective of a hospital. DEA can directly be applied to a multi-output/multi-input framework and only requires data on quantity to estimate the best practice frontier. Third, the minimal assumption on the economic behavior of DMUs is to avoid waste in resources usage. While traditional regularity conditions and flexible forms can in some ways be helpful for the specification of the technology, no clear economic argument is available to use one form or another for the inefficiency term. Unlike SFA, the non-parametric DEA approach does not assume any specification forms. The observed data are enveloped by piecewise linear facets which let the data determine the form of the frontier. Since the inefficiency is measured as the distance to the frontier, no specification form is required for the distribution of inefficiency. This is the main advantage of non-parametric approaches compared to parametric ones. In short, DEA avoids the possibility of confounding the misspecification effects due to an arbitrary choice of functional forms of the technology and the inefficiency components. However, DEA models suffer from one main disadvantage compared to SFA. They are deterministic and therefore they do not allow for deviations from the efficient frontier to be a function of random error. As such, it is assumed that inputs and outputs are measured without errors, in particular there are no outlier DMUs in the data. This assumption may be too strong in empirical works. The effect of outliers may result in a shift in the best practice frontier and a bias in the measurement of inefficiency. Recently, some robust DEA approaches have been proposed to control or mitigate this effect. Inefficiency estimates derived from traditional DEA are based on the whole sample of DMUs. This is the traditional deterministic approach. However, as discussed in the previous subsection, its main drawback is the sensitivity of the frontier to outliers. To circumvent this problem, we use a robust subsampling approach. We refer interested readers to Cazals et al.,54 Kneip et al.,55 and Simar and Wilson56 for all theoretical and methodological developments. For our purposes, we only provide an intuitive presentation of the approach. If outliers exist and shift away the boundary of the estimated output production set, the robust approach mitigates this problem. This technique is implemented
Health Services Management Research 0(0) as indicated by the following steps. First, a large number of sub-samples, of a predetermined size, are selected from the initial sample of observed DMUs, then the estimated directional distance function is computed for each subsample, and the final estimate is simply the average across the sub-samples. Therefore, since the estimated output production set is based on a random selection of DMUs varies over the sub-samples, a DMU is not always compared with potential outliers. We can interpret the final result as a robust measure of the inefficiency. For details on this methodology, see Valdmanis et al.13 In addition to mitigating the potential biases due to outliers, we are also able to simultaneously address the production of economic goods (patient care) and bads (re-admissions). This is accomplished by adding another constraint in the linear programming’s specification. By combining these approaches, we are better able to model the total hospital care process.
Independent variables We follow Baker et al.8 in our classification of full integration. Baker et al.8 define fully integrated hospitals as those that adopted one of the following models with their physicians: an integrated salary, foundation or equity model. We predict that fully integrated hospitals are more likely to have lower inefficiency scores. Competition was estimated at a county level by the Herfindahl-Hirschman Index (HHI) calculated by summing the squares of market shares of hospital admissions. For ease of interpretation, competition was coded as 1-HHI. Teaching status was incorporated by the use of two binary variables: major teaching (i.e. hospital is a member of Council of Teaching Hospital (COTH) and other teaching (i.e. hospital has a residency training approval by Accreditation Council for Graduate Medical Education but does not belong to COTH). We classify hospitals as high technology if they offer any of the following services: heart or any major organ transplant, computer assisted orthopedic surgery or electron beam computed tomography. Additional control variables include Medicare share of hospital admissions and Medicaid share of hospital admissions. All of these variables were derived or estimated using data from the AHA Annual Survey Database.
Results In Table 2, we provide the descriptive statistics of the inputs, outputs, a control variable (i.e., Medicare CaseMix Index (MCMI)), environmental factors, and efficiency scores. Average 30-day risk standardized mortality and readmission rates for the year 2013 are publicly reported and were derived from CMS. All outcomes measures were reported twice in 2013 and therefore we
Leleu et al.
7
Table 2. Descriptive statistics of the inputs and outputs used in measuring inefficiency. Variable Bad outputs Mortality rate: heart attack Mortality rate: heart failure Mortality rate: pneumonia Readmission rate: heart attack Readmission rate: heart failure Readmission rate: pneumonia readmission rate: knee, hip Readmission rate: all Good outputs Number of admissions (no surgical) Number of inpatient surgeries Number of outpatient surgeries Number of ER visits Number of other visits Inputs Case Mix Index FTE RN þ LPN Other trainees FTE other Beds Post admission days (proxy for resource use)
Standard deviation
Mean
15.19% 11.60% 11.80% 18.96% 23.84% 18.15% 5.40% 15.95%
1.44 1.33 1.75 1.29 1.91 1.54 0.73 1.13
9494.61 3650.69 6349.64 48,324.57 186,834.60
7689.38 3666.82 6020.77 33,752.37 253,863.50
1.59 523.76 523.76 1108.15 279.21 52,528.13
0.22 530.82 148.81 1299.71 216.75 0.03
ER: emergency room; FTE RN þ LPN: full time equivalent registered nurses and licensed practical nurses.
were able to calculate the average rates except for 30-day all-cause readmission rate which was reported only once. The average hospital mortality rates are between 11.60% and 15.19%. There is more variability among the hospitals in our sample for the 30 day readmission rates for the various diagnosis conditions, with an average overall 30-day readmission rate of 15.95%. The descriptive statistics for inputs and outputs demonstrate a wide variability among the hospitals in our sample showing that opting for the robust DEA approach was the correct model choice. On average, the hospitals in our sample have an MCMI of 1.59 but demonstrates less variability than the other outputs in the hospital sample. As in the case of outputs, the descriptive statistics on inputs also show a wide variability. In our DEA results (Table 3), 1158 hospitals (63%) exhibited inefficiency. The mean inefficiency score is 10.5%. This means that on average inefficient hospitals could simultaneously increase the good outputs and decrease the bad outputs by 10.5% by keeping the same level of inputs. Scores ranged from 0.64 to 0.44. It must be remembered that by virtue of the robust DEA model, we allow for some of the hospitals to be above the
Table 3. Robust DEA results.
Number Percent Robust DEA scores
Efficient hospitals
Inefficient hospitals
Total
689 27% 10.20%
1158 63% 10.50%
1847 100% 2.75%
DEA: data envelopment analysis.
frontier. Since we measure inefficiency, negative scores mean that some hospitals outperform other hospitals. For inefficient hospitals, scores are positive. In our results, 63% of the hospitals exhibited inefficiency. In terms of the organizational environmental factors (Table 3), we report that 42% of the hospitals employed a fully integrated physician model. The average competition index (1-HHI) is 0.57. The share of patients covered by Medicare is approximately 48%. In comparison the percent of patients covered by Medicaid is 18%. Nine percent of the hospitals in our sample are considered as major teaching, i.e. academic medical centers (AMC) belonging to COTH whereas 24% were minor teaching. We also report that 20% of our sample are for-profit and 42% provided high medical technological services. In light of the wide range of the robust DEA measures, it is incumbent upon us to regress these findings on an array of independent variables we believe affect the efficiency (including quality) measure (Table 4). From the regression model (reported in Table 5), we find that FP hospitals were statistically significantly (p < 0.05) more efficient than their NFP counterparts. Similarly, hospitals with highly technological medical equipment, a fully integrated physician work force, and a more competitive environment were more efficient while controlling for quality in the DEA model. Interestingly, the percentage of patients who were covered by Medicare and Medicaid were associated with hospitals increased inefficiency. AMCs and other teaching hospitals were also more efficient than non-teaching hospitals.
Discussion and managerial implications Our research extends earlier work by employing a robust DEA model that includes the joint production of economic ‘‘goods’’ and ‘‘bads’’ given the hospital production process. Due to the derivation of the efficiency measures using the robust approach, we are able to use the findings in regression analysis in order to ascertain the organizational and market based factors which explain the variability of quality adjusted efficiency scores between hospitals. By including the joint production, performance is not upward biased as would be the case if ‘‘bads’’ were excluded. For example, a hospital producing more outputs than another hospital may be deemed as
8
Health Services Management Research 0(0)
Table 4. Descriptive statistics of predictor variables. Variable
Mean (SD)
Standard deviation
Fully integrated hospitals
Competition
0.57
Proportion
Description
42%
Hospitals with one of the following physician integration models: integrated salary, foundation or equity model
0.32
Medicare share
47.91
9.43
Medicaid share
18.03
8.34
Major teaching
9%
Other teaching
24%
For-profit High technology
20% 42%
1-HHI; HHI was calculated by summing the squares of market shares of admissions per county Number of Medicare admissions 100 divided by total hospital discharges Number of Medicaid admissions 100 divided by total hospital discharges Hospital is a member of Council of Teaching Hospital of the Association of American Medical Colleges (COTH) Hospital has a residency training approval by Accreditation Council for Graduate Medical Education but does not belong to COTH For-profit ownership Hospital offers any of the following services: heart or any major organ transplant, computer assisted orthopedic surgery or electron beam computed tomography
HHI: Hirschman-Herfindahl Index.
Table 5. Regression Model results for inefficiency scores. Variable
Parameter
Constant For-profit ownership Percent of patients covered by Medicare Percent of patients covered by Medicaid High technological services Fully integrated physicians Competition Academic medical center Other teaching
0.028 0.028*** 0.002*** 0.001*** 0.036*** 0.025*** 0.024*** 0.032*** 0.018**
N ¼ 1847; adjusted R-squared ¼ 0.08; F ¼ 17.27. **p < 0.05. ***p < 0.01.
best practice. However, if this same hospital had more readmissions and higher mortality rates, this would indicate that the care provided was inferior, which is not compatible with a best practice hospital. Our findings indicate that hospitals with fully integrated physicians are more efficient. This is not surprising since as Chilingerian and Sherman30 argue, physicians control 80% of health care costs. While we are constrained in
the way we measure physician integration by the availability of complete data on hospital physician relationships, we believe our study still contributes to our understanding of the relationship between physician integration level and hospital cost and quality. We classified hospitals as fully integrated if they implemented the: integrated salaried model, equity model, or foundation model.8 Integrated salaried model simply indicates that the hospital employs its physicians while the equity model refers to arrangements where physicians own shares in the organization.12 In the foundation model acquired group practices are aggregated under a nonprofit entity whereby the hospital provides professional services and plays a key role in the management of the foundation.12 The integrated salaried model is by far the most common of the three full integration models listed. Hospitals currently employ more than half of the physician workforce.57 Some argue that physician employment could negatively influence hospitals since a hospital loses $150,000 to $250,000 during the first year it employees a physician.57 Based on our findings though, despite the high upfront cost, full integration, including physician employment, strengthens the ties between hospitals and their physicians which allows hospitals to achieve higher levels of efficiency while maintaining quality. This is especially important for healthcare managers
Leleu et al. as they struggle to configure physician contracts and incentives in a manner which maximizes physicians’ productivity and their adherence to evidence based care. As predicted by X-inefficiency theory, hospitals located in more competitive market are more efficient. This is not surprising, as hospitals operate in the era of VBP where they are competing for higher levels of efficiency and better quality. Managers of hospitals located in less competitive markets need to be aware of this difference and need to realize that they are more likely to operate with higher levels of slacks. Waste need to be identified and proactively reduced. Market competitiveness could change at any time and inefficient hospitals face the risk of losing market share when more efficient hospitals enter their market. The findings are also interesting in that teaching hospitals, particularly AMC are efficient vis-a`-vis non-teaching hospitals. This result should allay fears that teaching hospitals will be punished via higher HRRP penalties. Our results suggest that both efficiency and quality should be vigorously applied when attaching HRRP monitoring. Our findings indicate the for-profit hospitals are more efficient. While it is expected that for-profit hospitals are more efficient than their NFP counterparts, it is time for managers of NFP hospitals to create a major shift in organizational culture and operations to achieve the same levels of efficiency as for-profit hospitals. VBP incorporates efficiency as one of the key domains and therefore this new era of healthcare does not discriminate between FP and NFP in terms of efficiency and quality expectations. By combining quality and efficiency in our measurement of productivity performance, we added to both the methodological and data literature on hospitals. This methodology of measuring efficiency allowed us to more reliably examine relevant environmental factors, especially hospital–physicians arrangements, heretofore ignored, but now can be included. To summarize, we find that ownership, payer type, physician integration, competition, and teaching status all impact the quality/ efficiency dynamic. Healthcare managers should be aware of these factors as they make decisions and implement initiatives to lower cost and improve quality. Moreover, policy makers should review these factors as well as provide incentives that ensure higher quality via readmission penalties, but also account for environmental factors and more efficient treatment modalities. Declaration of conflicting interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
9 Note a. See Hollingsworth15 and O’Neill et al.16 for a review of these studies.
References 1. Negrini D, Kettle A, Sheppard L, et al. The cost of a hospital ward in Europe: is there a methodology available to accurately measure the costs? J Health Organ Manag 2004; 18: 195–206. 2. Nayar P, Ozcan YA, Yu F, et al. Benchmarking urban acute care hospitals: efficiency and quality perspectives. Health Care Manag Rev 2013; 38: 137–145. 3. Dixon J. Making progress on efficiency in the NHS in England: options for system reform. Nuffield Trust. 4. Zinn J and Flood A. Commentary: slack resources in health care organizations – fat to be trimmed or muscle to be exercised? Health Serv Res 2009; 44: 812–820. 5. Valdmanis V, Rosko M and Mutter R. Hospital quality, efficiency, and input slack differentials. Health Serv Res 2008; 43: 1830–1848. 6. Ozcan YA and Luke RD. A national study of the efficiency of hospitals in urban markets. Health Serv Res 1993; 27: 719. 7. Madison K. Hospital–physician affiliations and patient treatments, expenditures, and outcomes. Health Serv Res 2004; 39: 257–278. 8. Baker LC, Bundorf MK and Kessler DP. Vertical integration: hospital ownership of physician practices is associated with higher prices and spending. Health Aff 2014; 33: 756–763. 9. Alexander J and Morrisey M. Hospital–physician integration and hospital costs. Inquiry 1988; 25: 388–401. 10. Wan T, Lin B and Ma A. Integration mechanisms and hospital efficiency in integrated health care delivery systems. J Med Sys 2002; 26: 127–143. 11. Goes JB and Zhan C. The effects of hospital–physician integration strategies on hospital financial performance. Health Serv Res 1995; 30: 507. 12. Burns L and Muller R. Hospital-physician collaboration: landscape of economic integration and impact on clinical integration. Milbank Q 2008; 86: 375–434. 13. Valdmanis V, Rosko M, Mukamel D, et al. Assessing overall, technical, and scale efficiency among home health agencies. Health Care Manag Sci 2017; 20: 265–275. 14. Chilingerian J. Evaluating physician efficiency in hospitals: a multivariate analysis of best practices. Eur J Oper Res 1995; 80: 548–574. 15. Hollingsworth B. The measurement of efficiency and productivity of health care delivery. Health Econ 2008; 17: 1107–1128. 16. O’Neill L, Rauner M, Heidenberger K, et al. A crossnational comparison and taxonomy of DEA-based hospital efficiency studies. Socio-Econ Plann Sci 2008; 42: 158–189. 17. Nayar P and Ozcan YA. Data envelopment analysis comparison of hospital efficiency and quality. J Med Sys 2008; 32: 193–199. 18. Clement J, Valdmanis V, Bazzoli G, et al. Is more better? An analysis of hospital outcomes and efficiency with a DEA model of output congestion. Health Care Manag Sci 2008; 11: 67–77.
10 19. McCormack R, Michels R, Ramos N, et al. Thirty-day readmission rates as a measure of quality: causes of readmission after orthopedic surgeries and accuracy of administrative data. J Healthc Manag 2013; 58: 64–77. 20. Jencks SF, Williams MV and Coleman EA. Rehospitalizations among patients in the Medicare feefor-service program. N Engl J Med 2012; 366: 1364–1366. 21. Berenson R, Paulus R and Kalman N. Medicare’s readmissions-reduction program – a positive alternative. NEJM 2012; 366: 1364–1366. 22. Gohil SK, Datta R, Cao C, et al. Impact of hospital population case-mix, including poverty, on hospital all-cause and infection-related 30-day readmission rates. Clin Infect Dis 2015; 61: 1235–1243. 23. Herrin J, St Andre J, Kenward K, et al. Community factors and hospital readmission rates. Health Serv Res 2015; 50: 20–39. 24. Benbassat J and Taragin M. Hospital readmissions as a measure of quality of health care: advantages and limitations. Arch Intern Med 2000; 160: 1074–1081. 25. Pauly M and Redisch M. The not-for-profit hospital as a physicians’ cooperative. Am Econ Rev 1973; 63: 87–99. 26. Harris J. The internal organization of hospitals: some economic implications. Bell J Econ 1977; 467–482. 27. Pauly M. Doctors and their workshops. Chicago: University of Chicago Press, 1980. 28. Robinson J. Physician-hospital integration and the economic theory of the firm. Med Care Res Rev 1997; 54: 3–24. 29. Burns L and Pauly M. Integrated delivery networks: a detour on the road to the road to integrated health care? Health Aff 2002; 21: 128–143. 30. Chilingerian J and Sherman H. Managing physician efficiency and effectiveness in providing hospital services. Health Serv Manag Res 1990; 3: 3–15. 31. O’Malley A, Bond A and Berenson R. Rising hospital employment of physicians: better quality, higher costs. Issue Brief Cent Stud Health Syst Change 2011; 136: 1–4. 32. Leibenstein H. Allocative efficiency vs. X-efficiency. Am Econ Rev 1966; 56: 392–415. 33. Rosko M. Impact of HMO penetration and other environmental factors on hospital X-inefficiency. Med Care Res Rev 2001; 58: 430–454. 34. Brown HS. Managed care and technical efficiency. Health Econ 2003; 12: 149–158. 35. Bates L, Mukherjee K and Santerre R. Market structure and technical efficiency in the hospital services industry: a DEA approach. Med Care Res Rev 2006; 63: 499–524. 36. Lee KH, Park J, Lim S, et al. Has competition increased hospital technical efficiency? Health Care Manager 2015; 34: 106–112. 37. Kessler D and McClellan M. Is hospital competition socially wasteful? Q J Econ 2000; 115: 577–615. 38. Mutter RL, Wong HS and Goldfarb MG. The effects of hospital competition on inpatient quality of care. INQUIRY: The Journal of Health Care Organization, Provision, and Financing 2008; 45(3): 263–279. 39. Mukamel D, Zwanziger J and Bamezai A. Hospital competition, resource allocation and quality of care. BMC Health Serv Res 2002; 2: 2–10.
Health Services Management Research 0(0) 40. Mukamel D, Zwanziger J and Tomaszewski K. HMO penetration, competition, and risk-adjusted hospital mortality. Health Serv Res 2001; 36: 1019–1028. 41. Mutter R and Rosko M. The impact of ownership on hospital cost-efficiency of U.S. hospitals. In: Blank J and Valdmanis V (eds) Measuring hospital productivity and efficiency advances in health economics and health services research. Vol. 18, Amsterdam: Elsevier Press, 2007, pp.113–138. 42. Landon BE, Normand SL, Lessler A, et al. Quality of care for the treatment of acute medical conditions in US hospitals. Arch Intern Med 2006; 166: 2511–2517. 43. Rosko M, Wong H and Mutter R. Characteristics of highand low-efficiency hospitals. Med Care Res Rev. Epub ahead of print 2017. doi/full/10.1177/1077558716689197. 44. Joynt K and Jha A. Characteristics of hospitals receiving penalties under the hospital readmissions reduction program. JAMA 2013; 309: 342–343. 45. Nuckols T. County-level variation in readmission rates: implications for the Hospital Readmission Reduction Program’s potential to succeed. Health Serv Res 2015; 50: 12–19. 46. Kupersmith J. Quality of care in teaching hospitals: a literature review. Acad Med 2005; 80: 458–466. 47. Friedman B and Basu J. The rate and cost of hospital readmissions for preventable conditions. Med Care Res Rev 2004; 61: 225–240. 48. Leleu H, Moises J and Valdmanis VG. How do payer mix and technical inefficiency affect hospital profit? A weighted DEA approach. Oper Res Health Care 2014; 3: 231–237. 49. Saving T and Goodman J. A better way to approach Medicare’s impossible task. Health Aff Blog, Available at: http://healthaffairs.org/blog/2011/11/15/a-better-way-toapproach-medicares-impossible-task/ (accessed 15 November 2011). 50. American Hospital Association (AHA). American Hospital Association (AHA) Annual Survey Database. Chicago, IL: American Hospital Association (AHA), Health Forum. Codebook: HE-112(2013). 51. Koopmans T. In: Koopmans TC (ed.) An analysis of production as an efficient combination of activities, activity analysis of production and allocation. London: Wiley, 1951. 52. Baumol W. Activity analysis in one lesson. Am Econ Rev 1958; 48: 837–873. 53. Farrell M. The measurement of productive efficiency. J R Stat Soc Ser A 1957; 120: 253–290. 54. Charnes A, Cooper W and Rhodes E. Measuring the efficiency of decision making units. Eur J Oper Res 1978; 2: 429–444. 55. Cazals C, Florens J and Simar L. Nonparametric frontier estimation: a robust approach. J Econometrics 2002; 106: 1–25. 56. Kneip A, Simar L and Wilson PW. Asymptotics and consistent bootstraps for DEA estimators in nonparametric frontier models. Econ Theory 2008; 24: 1663–1697. 57. Simar L and Wilson P. Two-stage DEA: caveat emptor. J Prod Anal 2011; 36: 205–218. 58. Kocher R and Sahni N. Hospitals’ race to employ physicians – the logic behind a money-losing proposition. NEJM 2011; 364: 1790–1793.
