Xenos
et al. Cost Eﬀ Resour Alloc (2017) 15:6
DOI 10.1186/s1296201700685
RESEARCH
Eﬃciency and productivity assessment of public hospitals in Greece during the crisis period 2009–2012
P. Xenos
1
, J. Yfantopoulos
2*
, M. Nektarios
1
, N. Polyzos
3
, P. Tinios
1
and A. Constantopoulos
2
Abstract
Background:
This study is an initial eﬀort to examine the dynamics of eﬃciency and productivity in Greek public hospitals during the ﬁrst phase of the crisis 2009–2012. Data were collected by the Ministry of Health after several quality controls ensuring comparability and validity of hospital inputs and outputs. Productivity is estimated using the Malmquist Indicator, decomposing the estimated values into eﬃciency and technological change.
Methods:
Hospital eﬃciency and productivity growth are calculated by bootstrapping the nonparametric Malmquist analysis. The advantage of this method is the estimation eﬃciency and productivity through the corresponding conﬁdence intervals. Additionally, a Randomeﬀects Tobit model is explored to investigate the impact of contextual factors on the magnitude of eﬃciency.
Results:
Findings reveal substantial variations in hospital productivity over the period from 2009 to 2012. The economic crisis of 2009 had a negative impact in productivity. The average Malmquist Productivity Indicator (MPI) score is 0.72 with unity signifying stable production. Approximately 91% of the hospitals score lower than unity. Substantial increase is observed between 2010 and 2011, as indicated by the average MPI score which ﬂuctuates to 1.52. Moreover, technology change scored more than unity in more than 75% of hospitals. The last period (2011–2012) has shown stabilization in the expansionary process of productivity. The main factors contributing to overall productivity gains are increases in occupancy rates, type and size of the hospital.
Conclusions:
This paper attempts to oﬀer insights in eﬃciency and productivity growth for public hospitals in Greece. The results suggest that the average hospital experienced substantial productivity growth between 2009 and 2012 as indicated by variations in MPI. Almost all of the productivity increase was due to technology change which could be explained by the concurrent managerial and ﬁnancing healthcare reforms. Hospitals operating under decreasing returns to scale could achieve higher eﬃciency rates by reducing their capacity. However, certain social objectives should also be considered. Emphasis perhaps should be placed in utilizing and advancing managerial and organizational reforms, so that the beneﬁts of technological improvements will have a continuing positive impact in the future.
Keywords:
Eﬃciency, Hospitals, Productivity, Malmquist, Tobit, Greece, Crisis
JEL classiﬁcation:
C14, D24, H51
© The Author(s) 2017. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/ ), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the srcinal author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/ publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated.
Background
Analysis of eﬃciency and productivity of the hospital sector has become a considerable concern in Europe. It presents a challenging area of relevant studies because hospitals absorb a large amount of public healthcare spending. In OECD countries and in the European Union of the 27 member states, hospital expenditure represents on the average around 30 and 37% of the total health expenditures in 2012 respectively [1]. e correspond
ing share in Greece is 47%, indicating a hospital based
Open Access
Cost Eﬀectiveness and Resource Allocation
*Correspondence: yfantopoulos@gmail.com
2
School of Economics and Political Science, University of Athens, 6 Themistokleous Str., 10678 Athens, GreeceFull list of author information is available at the end of the article
Page 2 of 12Xenos
et al. Cost Eﬀ Resour Alloc (2017) 15:6
healthcare system and the need to give greater emphases on operational eﬃciency and cost containment in order to balance healthcare expenditure at a feasible level [2, 3].
e Greek health system is a combination of an National Health System (NHS) model in the supplyside (consisting of an extensive hospital sector of 138 public hospitals and an underdeveloped primary healthcare sector) and a social insurance system in the demand side (consisting of many health funds that all merged in one, EOPPY (National Organization for the Provision of Healthcare), in 2010. Grants from the public budget ﬁnance the ‘ﬁxed’ and contractual expenses (mainly salaries) of public hospitals, while the revenues from the social insurance funds (now consolidate in the single fund EOPPY) ﬁnance the variable expenses.During the crisis period 2009–2012, a reduction of 40% took place in hospital budgets. Additionally, shortages in healthcare workforce and medical supplies have been recorded in the Greek hospital sector [4]. e current eﬀorts of public authorities towards a more eﬃcient allocation of ﬁnancial and human resources to public hospitals raised questions about the criteria used to evaluate the performance of the Greek healthcare system in the previous years.Nowadays, Greece has 136 public plus two NonGovernmental Organizations (NGO) hospitals, managed by 85 NHS Trusts, which belong to the Greek National Health System (ESY). During the crisis period 2009–2012, the Greek Ministry of Health attempted to reform public hospitals operations and restrain healthcare expenditure. One such operation was the initiation of department budgets, which oﬀers better expenditure control, more accurate estimation of hospital products and supports productivity enhancement in hospital departments [5]. Furthermore, two major reforms were implemented regarding Greek public hospitals. e ﬁrst reform was the operational redeployment of the 136 NHS hospitals into 85 Trusts and the second was the implementation of a Diagnosis Related Groups (DRG) prospective reimbursement system, which was introduced in 2011, in order to minimize costs. DRGs are developed in order to identify and price hospital services, based on the diagnosis [6]. Apart from the above measures, additional reforms are in progress in order to restrain healthcare expenditure. For example, the joint purchasing of goods and services by using price–volume agreements can lead to signiﬁcant decline of healthcare costs. Additionally, the consolidation of NHS hospitals, the adoption of speciﬁc policies related to pharmaceuticals and the advancement of public hospitals infrastructure and technology can further contribute to expenditure reductions. At this point, it is important to mention that EOPYY, signs contracts with all Greek hospitals under the KENDRG reimbursement system [2, 5].
To the best of our knowledge, the impact of these hospital reforms has not yet been measured. eoretically, budget cuts are expected to cause a positive shift of eﬃciency provided that outputs remain stable. Intuitively, since shortages in workforce and medical equipment vary between hospitals, their impact would most probably aﬀect eﬃciency change rather than technology. e redeployment of hospitals leads to better management of inputs. erefore, by reducing costs it would most probably increase the overall eﬃciency of the redeployed hospitals. Moreover, the DRGbased reimbursement system, combined with the pharmaceutical pricing reforms, is expected to create economies of scale which would greatly improve hospital eﬃciency [7].
Research objectives
e purpose of this paper is to investigate the dynamics of productivity and eﬃciency in the Greek Hospital sector over the years from 2009 to 2012. e study is limited in this period, due to the unavailability of more recent data. Moreover, data prior to 2009 were not collected and validated according to international organization principles and guidelines.We make use of Malmquist Productivity Index (MPI) through data envelopment analysis (DEA) augmented by bootstrapping techniques. e study contributes to the current literature in several possible ways. First, it takes into account all Greek public hospitals (excluding the specialized in psychiatry and pediatrics). Homogeneity is preserved and selection bias is avoided. Second, the data are collected by the Ministry of Health after several quality controls ensuring comparability and validity of the hospital inputs and outputs. ird, our methodology is based on the nonparametric Malmquist productivity analysis developed by Simar and Wilson [8] not previously applied in Greek hospital sector. e great advantage of this method is the estimation of eﬃciency and productivity change followed by the corresponding conﬁdence intervals. Fourth we decompose the estimated values of productivity into eﬃciency and technological change components. e above points would provide valuable information to decision makers for eﬀective policy guidance during the crisis period of 2009–2012.e rest of the paper is arranged in three sections as follows. e ﬁrst section provides eﬃciency and productivity measurement concepts, with a brief literature review on healthcare eﬃciency measurement in Greece and in some other countries. In the following section, the data and the estimated results are presented and discussed. e ﬁnal section provides the conclusion of the study.
Page 3 of 12Xenos
et al. Cost Eﬀ Resour Alloc (2017) 15:6
Hospital eﬃciency and productivity measurement
e measurement of eﬃciency and productivity is crucial for hospitals because it allows them to compare the performance of their own organization with that of other hospitals in the same NHS and establish a reciprocal policy of “best practices” in order to improve their own performance [9–13].
JenuAppiah et al. [14] and Kirigia and Asbu [15]
used twostage analysis using DEA eﬃciency measurement and Tobit model in order to examine relationships between hospital ineﬃciencies and environmental factors. Both studies used crosssectional data.Zavras et al. [16], by using DEA, assessed the relative eﬃciency of 133 primary healthcare services, between 1998 and 1999; the results indicated that the primary healthcare centers that had the appropriate technological capacity to carry out laboratory or radiological examinations had the highest eﬃciency scores, whereas the mediumsized centers that covered population areas of 10,000–50,000 people performed better than the other primary healthcare units.In another study, Tsekouras et al. [17], by using Bootstrap DEA, measured the productive eﬃciency of 39 intensive care units (ICUs) of the Greek Healthcare system for 2004. e purpose of the study was to reveal if new medical technology investment into ICUs had a positive impact; the ﬁndings demonstrated that technical eﬃciency improved but scale eﬃciency remained unchanged.Certain studies employ the Malmquist Index methodology and then decompose total factor productivity into technical eﬃciency and technology change. In Greece, the application of DEA in eﬃciency and productivity measurement has gained considerable attention by both researchers and policy makers [18, 19]. In a recent study
Karagiannis and Velentzas [20] estimated productivity growth for Greek public hospitals for the period 2002–2007 including quality variables in their analysis. ey create a qualityadjusted Malmquist productivity index. eir ﬁndings indicate reductions both in productiv ity and quality as well as signiﬁcant variations between hospitals.Androutsou et al. [21] measured the performance in seven homogenous specialty clinics across all National Health System hospitals in the Regional Health Authority (RHA) of essaly, over the period 2002–2006 with Malmquist Index. Overall productivity progressed in all clinics. Technical change progressed except the general medicine clinics, and diachronically the size of the clinics inﬂuences the overall eﬀects on hospital performance. Polyzos [22] analyzed the performance of 117 Greek NHS hospitals by means of DEA, for years 2009–2011. All hospitals, especially middlesized hospitals showed performance improvements on technical eﬃciency terms.is study attempts to make an early assessment of the health reforms in the period 2009–2012 by exploiting the Malmquist methodology which provides a dynamic approach to the assessment of eﬃciency and productiv ity of the hospital sector. Additionally, a Randomeﬀects Tobit regression model is explored to investigate the impact of several contextual factors on the magnitude of eﬃciency in public hospitals.
Methods
Data envelopment analysis
Charnes, Cooper and Rhodes (CCR) [23] calculated the eﬃciency frontier basing their estimates on best practices rather than the average performance in a given sample. Based on their research, Banker et al. [24] introduced the “ Banker, Charnes and Cooper (BCC) model” of eﬃciency measurement. is model assumes a production technology of variable returns to scale, implying that any proportional change in inputs usage results in variable proportional change in outputs [25]. Speciﬁcally, we used the inputoriented approach, since inputs are more easily controlled by hospital administrations, compared to outputs.According to Simar and Wilson [26], twostage approach results are inconsistent and biased unless the DEA eﬃciency scores are corrected by a bootstrapping procedure. Bootstrapping estimates a more robust regression model in order to determine the eﬀect of contextual factors on eﬃciency [27].e DEA model can only be applied to multiple DMUs (Decision Making Units: hospitals in our case) on a peryear basis. erefore, DEA cannot estimate the eﬃciency change over time. e Malmquist Productivity Index (MPI), which is presented in the next subsection, overcomes this limitation. In a nonparametric framework the MPI evaluates the eﬃciency change over time [28].
The malmquist productivity index
Assuming a list of p inputs and q outputs, the production set is deﬁned in the Euclidean space
R
p
+
q
+
as follows:We can deﬁne the input requirement set V(y) as the set of all input vectors that can produce the output vector
y
∈
R
+
:Fare et al. [29] determined the input distance function:
(1)
=
x,y

x
∈
R
p
+
,y
∈
R
q
+
,
x,y
is feasible
(2)
y
=
x
∈
R
p
+

x,y
∈
Ξ
(3)
D
ti
x
t
,y
t
=
sup
{
:
(
x
t
/
,y
t
)
∈
S
t
}
Page 4 of 12Xenos
et al. Cost Eﬀ Resour Alloc (2017) 15:6
where
S
t
=
x
t
,y
t
:
x
t
can produce y
t
. Malmquist Total Factor Productivity change index between period t and t
+
1 as:Equation 4 represents the geometric mean of the two Malmquist indices for periods t and t
+
1. e ﬁrst index employs reference technology, which corresponds to period t, while the second index performs the same function, as the ﬁrst one, for period t
+
1.Fare et al. [29] factor the expression (4) into the product
of technical eﬃciency and technological change (frontier shift) as:orwhere “M” symbolizes Total Factor Productivity Growth index between periods t and t
+
1, and “E” and “T” represent the technical eﬃciency change, and the technological change respectively for the same period. Full interpretation of these indices speciﬁed to health sector can be found in Jacobs et al. [30] and AdensoDiaz [31].
By combining each DMUs distance from the eﬃciency frontier (eﬃciency change) and the overall shift of the frontier over time (technology change), the Malmquist Productivity Index oﬀers a dynamic approach on handling panel datasets [32].However, Eq. (4) limits our ability of determining whether changes in productivity, eﬃciency and technology, really exist or they are merely appearing as such because of the fact that we do not know the actual production frontiers, in which case we must estimate them from the ﬁnite sample [26, 33, 34]. For the above reason,
a bootstrap estimation procedure for obtaining conﬁdence intervals and correcting the Malmquist Index and its components was employed. e estimation is implemented through the data generating process procedure (DGP), by using a series of pseudo datasets to create a bootstrap estimate. e problems that occur when bootstrapping DEA models are discussed by Simar and
(4)
M
I
X
t
,Y
t
,X
t
+
1
,Y
t
+
1
=
M
tI
X
t
+
1
,Y
t
+
1
,X
t
,Y
t
×
M
t
+
1I
X
t
+
1
,Y
t
+
1
,X
t
,Y
t
1
/
2
=
D
tI
X
t
+
1
,Y
t
+
1
D
tI
X
t
,Y
t
D
t
+
1I
X
t
+
1
,Y
t
+
1
D
t
+
1I
X
t
,Y
t
1
/
2
(5)
M
I
X
t
,Y
t
,X
t
+
1
,Y
t
+
1
=
D
t
+
1I
X
t
+
1
,Y
t
+
1
D
tI
X
t
,Y
t
D
tI
X
t
+
1
,Y
t
+
1
D
t
+
1I
X
t
+
1
,Y
t
+
1
D
tI
X
t
,Y
t
D
t
+
1I
X
t
,Y
t
1
/
2
M
=
E
×
T
Wilson [35]. e bootstrapping procedure concerning Malmquist indices is described in detail at Simar and Wilson [8]. us, by obtaining a conﬁdence interval for the Malmquist index and its components it becomes possible to validate whether productivity changes are signiﬁcant at the desired level of conﬁdence.However, Simar and Wilson have expressed doubts about the former methodology. ey argue that the usual semiparametric framework is inconsistent in some cases [34]. Using Monte Carlo simulations, they show that since the data generating process cannot be estimated the Tobit regression is inadequate. ey propose a truncated regression model and perform single and double bootstrapping, ﬁnding that the latter produces better results.
Regression analysis between ineﬃciencies and contextual factors
e point of a twostage analysis of hospital eﬃciency, is to shed more light on the impact of contextual factors beyond the control of the hospitals on eﬃciency. Such factors are the operating status of the hospital, the region that is located, etc. In cases where diﬀerences across the panel variable have inﬂuence on the dependent variable, randomeﬀects models are often used in relevant literature [36–43]. erefore, in order to explore the potential
eﬀect of time as the panel variable, which in this case is expressed in years, we used random rather than ﬁxedeﬀects. Besides that, ﬁxedeﬀects models control for all cannot variables constant across years, such as hospital type, size and RHA, and are therefore unable to measure their eﬀect [44].e Tobit model ensures lower tail censoring of the distribution that DEA creates. e use of OLS estimation is not appropriate for determining the desired factors of hospital eﬃciency, because of the nature of the dependent variable (eﬃciency), which is constrained in the 0–1 interval.Greene [43] proposed a censoring point at zero for computation purposes and transformed DEA eﬃciency scores into ineﬃciency scores leftcensored at zero using the equation as follows:where
DEA eﬀ.score
=
1
D
ti
x
t
,y
t
.Consider the linear regression model with paneldata randomeﬀects:
(6)
ineﬀ score
=
1DEA eﬀ.score
−
1
(7)
y
∗
it
=
β
i
z
it
+
v
i
+
ε
it
y
it
=
y
∗
it
if y
∗
it
<
0 y
it
=
0 if y
∗
it
≤
0i
=
1,2,
. . .
,N
Page 5 of 12Xenos
et al. Cost Eﬀ Resour Alloc (2017) 15:6
where i
=
1,…,N is the number of DMU’s and t is time,
β
i
is the vector of unknown parameters, Z
i
is the vector of explanatory variables. e randomeﬀects v
i
are independent and identically distributed (i.i.d.),
N
(
0,
σ
2 v
)
and
ɛ
it
are i.i.d.
N
(
0,
σ
ε
)
independently of v
i
. e observed data y
*it
represents possibly censored versions of y
it
.e estimated empirical model is speciﬁed in the following equation:where “ineﬀ” is the ineﬃciency score and
Z
i
are the following contextual factors: (i) average length of stay (ALS), (ii) bed occupancy rate (OCP), (iii) number of diagnostic procedures (DIAG), (iv) number of patients adjusted by the Roemer index (PAT), (v) type of hospital (1
=
Teaching, 0
=
NonTeaching) (TYPE), (vi) three dummy variables concerning hospital size based on the number of beds. Large hospitals are the ones with more than 400 beds (L), medium hospital are the ones containing between 100 and 400 beds (M) and small hospitals are all the rest, having less than 100 beds (S), (vii) seven dummy variables representing each of the seven Regional Health Authorities (RHA) in which Greece is divided (YPE1–YPE7). e RHAs are responsible for planning, coordinating supervising and inspecting all Health Services within the limits of their region. eir aim is to disperse the health sector in order to address problems related to ineﬃciency in the delivery of healthcare. (viii) four dummy variables signifying the year (YEAR09–YEAR12).e average length of stay (ALS) is the number of days that an inpatient occupies a bed in the hospital. Positive ALS coeﬃcient would indicate a negative impact on eﬃciency, since hospital resources remain committed on the same patient. Bed occupancy rate has the opposite impact, because hospitals operate utilizing all available resources. “Diagnostic procedures” include technical and diagnostic procedures, such as blood tests, MRIs, CTs and biochemical exams. If diagnostics are appointed a negative coeﬃcient, it would indicate a positive eﬀect on eﬃciency. Teaching hospitals are expected to have a positive coeﬃcient, contributing negatively to eﬃciency. is occurs because healthcare is not their only aim and therefore some resources are spent on the teaching procedure.
Sampling
On the base of reforms initiated by the memorandum policies, the Ministry of Health has developed a webbased data repository called “ESYnet”. e base includes all Greek hospitals, covering the period 2009–2012 and several variables concerning organizational, medical and ﬁnancial information. e sample consists of 108 general hospitals for four years (4 years
×
108 hospitals
=
432
(8)
Tobit
(
ineﬀ
)
=
α
+
β
i
Z
i
+···+
v
i
+
ε
i
observations). In order to ensure homogeneity of the sample the specialty hospitals (psychiatric, maternity, dermatological and cardiological hospitals) are excluded. ESYnet has been compatible with the international standards of organizations such as World Health Organization, OECD and Eurostat. Grant of access was oﬃcially oﬀered to researchers in 2011.Based on a study by the Centre of Health Economics of the University of York [45], each pair of adjacent years is called “link” throughout the paper. is way, by perceiving consecutive pairs of years as links of a chain, it is easier to explore changes made over time. Links and Fiscal years are shown in Fig. 1.Given the limitations of the data, the outputs used are: (i) the number of patient discharges adjusted for casemix with Roemer Index [46]. Roemer et al. provide an adjusted estimate for the average length of stay taking into account the occupancy rate of the hospitals. (ii) the number of diagnostic procedures. e inputs include: (a) the number of doctors, (b) the number of beds, (c) the number of other personnel employed and (d) nonlabour expenditures (i.e. pharmaceutical and health technology supplies, etc.) (see Table 1).e expenditure variable has been deﬂated by the GDP price deﬂator (2012
=
100). Following Vassiloglou and Giokas [47], the number of DMUs is greater than three times the number of inputs plus outputs.
Model speciﬁcations
e distance functions that are required in order to obtain Malmquist indices were measured using DEA, assuming constant returns to scale. In order to decompose further the eﬃciency change into pure eﬃciency change and scale eﬃciency change, a variable returns to scale technology (VRS) was considered. Because public hospitals are considered to have smaller ability to control their outputs and more opportunities to lower their inputs, we employed an inputoriented DEA. Moreover, a benchmarking approach was used where the most eﬃcient DMUs were estimated regarding their signiﬁcance as benchmarks for the ineﬃcient DMUs in the sample data.
Fiscal Year2009 2010 2011 2012Link 1Link 2Link 3
Fig. 1
Fiscal years and links