Mathematical and Computer Modelling 52 (2010) 1095–1102

Contents lists available at ScienceDirect

Mathematical and Computer Modelling

journal homepage: www.elsevier.com/locate/mcm

A model to measure the efficiency of hospital performance

María Caballer-Tarazona ∗ , Ismael Moya-Clemente, David Vivas-Consuelo,
Isabel Barrachina-Martínez
Universidad Politécnica de Valencia, CIEGS. Research Centre for Health Economics & Management, ADE, Edificio 7I, Camino de Vera S/N, C.p: 46022, Valencia,
Spain

article info abstract

Article history: The establishment of specific and thoroughly researched criteria to evaluate hospitals’
Received 22 September 2009 activity is very important because there is a huge and increasing amount of public resources
Received in revised form 27 February 2010 dedicated to healthcare. Therefore, it is necessary to design a system to evaluate healthcare
Accepted 8 March 2010
performance in order to discover and improve potential inefficiencies. With this goal in
mind, the purpose of this paper is to analyse efficiency in three healthcare service units in
Keywords:
Valencian hospitals to establish appropriate guidelines for efficiency performance.
Efficiency
Health services
Three healthcare service units of 22 hospitals in the Valencian Community (East Spain)
DEA Model with a waiting list higher than average were selected (general surgery, ophthalmology,
Efficiency indexes traumatology–orthopaedic surgery). In this study, a non-parametric methodology, known
as the DEA model (Data Envelopment Analysis), was used along with two efficiency indexes
which were specially designed. Discriminant analyses proved the effectiveness of these two
indexes, therefore the indicators are an easier and effective methodology for measuring
efficiency than the DEA Model. In addition, the discriminant analysis offers a function which
classifies new health services as efficient or inefficient.
In conclusion, this paper offers tools for evaluating the performance of hospital activity
which are useful for both hospital management and health administration controlling
hospital performance.
© 2010 Elsevier Ltd. All rights reserved.

1. Introduction

One of the basic objectives pursued by most countries is to improve their health system both in terms of quality services
and efficiency and the extent to which its resources are put to good use.
So one fundamental reason to promote research into the efficiency of publicly financed hospitals in the Valencian
Community is the need to establish the bases for the best distribution and use of healthcare resources (optimum planning)
and to detect the set of problems of various kinds which affect their efficiency and capacity to offer top-quality services to
the population [1].
The current Spanish health system came about as a result of the General Health Law in 1986 and the old Healthcare Social
Security model gave way to the National Health System model. Since then, this new system has been financed by general
taxes and offers practically universal coverage.
Since the General Health Law came out, the Spanish Healthcare System has undergone profound changes, of which the
most outstanding has been the transfer process to the Spanish Autonomous Communities (AC), which concluded in 2002 [2].
In the Valencian Community, the change introduced was the private management of certain health districts, known as
‘‘Administrative Allowance’’. That is, public financing is maintained but a private insurance company manages the healthcare
district (Department). The management agreement (Administrative Allowance) is allocated by public bidding.

∗ Corresponding author. Tel.: +34 3207012009.

E-mail addresses: maria.caballer@gmail.com (M. Caballer-Tarazona), imoya@esp.upv.es (I. Moya-Clemente), dvivas@upvnet.upv.es
(D. Vivas-Consuelo), ibarrach@ade.upv.es (I. Barrachina-Martínez).

0895-7177/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mcm.2010.03.006
1096 M. Caballer-Tarazona et al. / Mathematical and Computer Modelling 52 (2010) 1095–1102

Administrative Allowance is a management contract of public services which, in this case, are healthcare services
addressing an assigned population.
Ownership, financing and control are all publicly run, while services are private and respect the principles of efficacy,
efficiency, equity and the cost-free status of services.
The model offers the Regional Government the advantage of achieving quality public services at a lower planned cost
without initial investments [3].
Given the co-existence of both hospital management programmes (public management and Administrative Allowance),
it is necessary to stress the need for studies into hospital-related operative efficiency for the purpose of comparatively
assessing both systems and establishing efficient performance patterns [4].
Stimulating and diffusing comparisons and bench marching based on the works done with already existing data are
initiatives that must be taken into account for the system to work well and with the willingness to promote new information
transparency [5].
Despite the many studies done to measure efficiency in hospitals [6–9] analysing overall efficiency, studying efficiency
through services offers more accurate information about the problem, mainly because when efficiency in hospitals and
medical services is measured, a wide variability of the results among the healthcare units is observed [10–14]. Therefore,
measuring the efficiency of specialized healthcare services to discriminate the inefficiency effect of other causes is of genuine
interest.
By taking these remarks into account, the objective of this article is, on one hand, studying the efficiency of three of the
health services in the Valencian Community with the highest demand through means of the DEA model, on the other hand,
offering an alternative model to measure efficiency in a more simple way. Both methodologies involve a novel approach to
analyse efficiency in the context of hospitals in the Valencian Community.
This article is structured as follows: Section 2 describes data sources and methodology. (This section includes a brief
description of the DEA Model along with the inputs and outputs used in the study. Additionally, a more operative
methodology to measure efficiency is presented.) Section 3 presents the results obtained. Sections 4 and 5 are the Discussion
and Conclusions, respectively.

2. Data sources and methodology

2.1. Data sources

To fulfill the main objective of this article, a comparative study was conducted in 22 hospitals of the Valencian
Community. More specifically, the study into hospital efficiency was conducted in three main hospital units: general surgery,
ophthalmology and traumatology–orthopaedic surgery. These three hospital units were selected in terms of waiting lists,
that is they were chosen for this study because they presented the longest waiting lists and had, therefore, the strongest
need to detect any possible inefficiencies.
It is noteworthy that the hospitals studied included both public hospitals and those run by the Administrative Allowance
System. All the data were provided by the Regional Ministry of Health for the year 2005.

2.2. Data envelopment analysis

The DEA Model (Data Envelopment Analysis) was the first methodology used to analyse the efficiency of the services
considered.
The DEA Model, developed by Charmes, Cooper and Rhodes [15], is a very useful methodology for our particular case as
it measures the productivity and efficiency related to organisational units, like hospitals, which use numerous resources to
produce multiple products [16].
The main objective of the DEA Model is to find a limit for efficiency formed by those combinations of resources which
optimise the amount of products made by minimising production costs. Then, with this limit, it assesses the relative
efficiency of the combinations of resources that do not belong to it [17].
In short, DEA is a model that measures efficiency and may be described as an extension of the simple input/output
analysis ratio which is rigorously generalised to work with multiple inputs and outputs. DEA uses mathematical models
(lineal programming) to calculate an efficiency limit. This limit provides a reference for judging comparatively the results
of the remaining units that do not belong to the limit [18].
Therefore, the model evaluates the efficiency solving in problem (1):
P
us ∗ ys0
max e1 = P
vm ∗ xm0
P
us ∗ ysi (1)
s.t.: P ≤ 1 i = 1, . . . , I
vm ∗ xmi
us vm ≥ 0 m = 1, . . . , M s = 1, . . . , S
 M. Caballer-Tarazona et al. / Mathematical and Computer Modelling 52 (2010) 1095–1102 1097

where:
ys0 = quantity of output s per DMU.
us = weight regarding output s.
xs0 = quantity of input m per DMU.
vm = weight regarding input m.
And the lineal CCR primal is problem (2):

Max = Σ u0 ∗ y0
s.t.: Σ v 0 ∗ xi = 1
(2)
Σ u ∗ yi − Σ v 0 ∗ xi ≤ 0
0

u, v ≥ 0.

In this work, the CCR (Charles, Cooper and Rhodes) Model has been used; therefore, it has been considered that hospital
activity presents increasing performances.
The CCR model is a radial model which, for input orientation, means that inefficient units reduce inputs to reach the
efficiency limit, while the relationship between the reduced level over the initial level is the efficiency index [19].
The main advantages of the DEA methodology in the hospital efficiency measurement are the following:
– It can be used in DMU (Decision Units) which uses multiple inputs to generate multiple outputs.
– It fits to models where prices of resources and products are unknown.
– None required functional specialization between inputs and outputs.

2.3. Defining inputs and outputs

The Output variables considered were as follows:

– Weighted admissions: (income x case-mix) Number of admissions weighted by the case-mix to consider the complexity
of the cases. We weighted the admission by the case-mix in order to create an equivalent measure in terms of resources
consumption.
– Consultations: (first consultations).
– Successive consultations.
– Surgical interventions: number of surgical interventions.
The Input variables used were the following:
– Number of doctors.
– Number of beds. Number of beds used during the term for each unit.

2.4. Efficiency indicators and discriminant analysis

As a more friendly methodology to DEA, a simpler and more operative measure, by means of indicators, is proposed
which may prove more useful for the hospital management domain.
These indicators offer two basic advantages for an efficiency analysis done by the DEA Model:
– Firstly, it is simple methodology converts them into an efficiency measurement instrument that any hospital director
may use.
– Secondly, fewer variables are needed to calculate the indicators than to estimate the DEA Model. So, when faced with
a situation where data are scarce, it may also prove more advantageous to estimate the efficiency of the service with
indicators.
The indicators proposed are the following:
Indicator I1 : Incomes/doctors.
Indicator I2 : Interventions/doctors.
To verify whether these indicators correctly classify services as efficient and inefficient, their efficacy has been checked
by a discriminant analysis. A discriminant analysis is a statistical technique which allows an activity to be assigned to a
group defined a priori (a dependent variable) in terms of a series of the group’s characteristics.
That is, the problem would be formulated as:
The expression (3) is the discriminant Ds function.

Ds = Bs1 X1 + · · · + Bsp Xp + Bs0 , (3)

where (Xi1 , . . . , Xip ), i = 1, . . . , n, is the sample of n observations of the independent quantitative variables X1 , . . . , Xp , in
the K groups of individuals established by the qualitative dependent variable Y values.
1098 M. Caballer-Tarazona et al. / Mathematical and Computer Modelling 52 (2010) 1095–1102

Fig. 1. Results of the efficiency analysis.

The percentage of correctly classified cases will be an effectiveness index of the discriminant functions. If these functions
were effective with regards to the sample observed, it is expected that they would also be effective when classifying
individuals for whom the group they belong to is unknown [20].
In other words, the objective of the discriminant analysis is to explain the belonging to a group in terms of the classifying
variables by quantifying the relative importance of each one, and by predicting to which group an observation that does not
form part of the analysed data belongs, and from which group we actually know the classifying variables value.
Therefore with this particular study, the belonging group will be the score obtained with the DEA Model, that is, 0 if
the hospital unit is inefficient and 1 if it is efficient. Otherwise, the classifying variables will be the previously calculated
indicators.

3. Results: model interpretation

Applying the DEA Model to the data of the three service units studied with the inputs and outputs described in Section 2.3
provides the results presented in Fig. 1. Here those hospitals classified as efficient score 1, while those classified as inefficient
score less than 1.
Eight efficient units and 14 inefficient units were identified for the general surgery hospital unit, while 9 efficient and
12 inefficient units were found for ophthalmology. Finally, only 6 efficient but 16 inefficient units were identified for
traumatology–orthopaedic surgery.
What clearly comes across from these results is that if an hospital runs one of the studied services efficiently, this does
not necessarily mean that the rest of its services are also run efficiently. Therefore, studying efficiency for each hospital
service individually is much more advisable than studying the overall efficiency of the hospital.
Next, Fig. 2 presents the results of the calculation of the indicators proposed which are a simpler approach to measure
efficiency, where:
Indicator I1 : Weighted admissions/doctors.
Indicator I2 : Interventions/doctors.
We can see in Table 2 how those services with the highest scores for the indicators are classified as efficient as they will
be those services which are capable of producing more inputs with less resources or outputs.
Then if we take the score obtained in the DEA analysis as the dependent variable, and by taking indicators I1 and I2
as the classifying variables, the discriminant analysis was done for the purpose of identifying the exact percentage that
these indicators are able to correctly classify the different hospital units as either efficient or inefficient. Therefore, it is an
alternative methodology to measure efficiency.
This analysis will verify whether the considered indicators are capable to classify the various hospital units into the
corresponding group as either efficient or inefficient, and which of these two indicators contributes to this classification to
a greater extent.
 M. Caballer-Tarazona et al. / Mathematical and Computer Modelling 52 (2010) 1095–1102 1099

Fig. 2. Calculating the indicators.

Table 1
Results of the classification (a).
Efficiency Predicted group belonged to Total
0.00 1.00

Original Recount 0.00 12 2 14

1.00 1 7 8
(%) 0.00 85.7 14.3 100.0
1.00 12.5 87.5 100.0
A classified 86.4% of the original grouped cases correctly.

Table 2
Statistical parameters.
Contraste de las funciones Lambda de Wilks Chi-cuadrado gl Sig.

1 0.686 7.158 2 0.028

Therefore, 3 discriminating functions have been obtained, one for each hospital service under study. Functions (4)–(6)
and Tables 1, 3 and 5 show the results.
Discriminant analysis for the general surgery service:
The discriminant function obtained for the general surgery service is the following:
Y = −5.919 + 0.024I1 + 0.007I2 . (4)
Table 1 shows the results of classification and the percentage of cases that the estimated function is able to classify
correctly.
The statistical parameters of the discriminant function are shown in Table 2.
For General Surgery service, a discriminant function was obtained which is capable of classifying 86.4% of the hospital
units correctly by using the previously calculated indicators as classifying variables.
On the other hand, as shown in Table 2, the Wilks’ lambda is statistically significant, therefore, it provides enough
information in order to differentiate the groups significantly. In other words, hospitals can be classified in the group of
efficient or inefficient in terms of the result of the variables introduced (I1 , I2 ) in the discriminant analysis.
1100 M. Caballer-Tarazona et al. / Mathematical and Computer Modelling 52 (2010) 1095–1102

Table 3
Results of the classification (a).
Efficiency Predicted group belonged to Total
0.00 1.00

Original Recount 0.00 7 2 9

1.00 3 9 12
(%) 0.00 77.8 22.2 100.0
1.00 25.0 75.0 100.0
A classified 76.2% of the original grouped cases correctly.

Table 4
Statistical parameters.
Contraste de las funciones Lambda de Wilks Chi-cuadrado gl Sig.

1 0.707 6.230 2 0.044

Table 5
Results of the classification (a).
Efficiency Predicted group belonged to Total
0.00 1.00

Original Recount 0.00 15 1 16

1.00 3 3 6
(%) 0.00 93.8 6.3 100.0
1.00 50.0 50.0 100.0
A Classified 81.8% of the original grouped cases correctly.

Table 6
Statistical parameters.
Contraste de las funciones Lambda de Wilks Chi-cuadrado gl Sig.

1 0.666 7.734 2 0.021

Discriminant analysis for the ophthalmology service:

The discriminant function obtained for the ophthalmology service is the following:

Y = −3.657 + 0.090I1 + 0.010I2 . (5)

Table 3 shows the results of classification and the percentage of cases that the estimated function is able to classify
correctly.
The statistical parameters of the discriminant function are shown in Table 4.
With the Ophthalmology Service, the discriminant function was capable of classifying 76.2% of the different hospital units
correctly as efficient and inefficient.
As shown in Table 4, the Wilks’ lambda is statistically significant, therefore, it provides enough information in order to
differentiate the groups significantly.
Discriminant analysis for the traumatology–orthopaedic surgery service:
The discriminant function obtained for the traumatology–orthopaedic surgery service is the following:

Y = −3.129 + 0.015I1 + 0.005I2 . (6)

Table 5 shows the results of classification and the percentage of cases that the estimated function is able to classify
correctly.
The statistical parameters of the discriminant function are shown in Table 6.
Finally, the discriminant function of the traumatology–orthopaedic surgery service correctly classified 81.8% of the
different hospital units as efficient or inefficient.
In addition, as shown in Table 6, the Wilks’ lambda is statistically significant, therefore, is provides enough information
in order to differentiate the groups significantly.
The correction percentages, with which the discriminant functions classifies the hospital units, are considered high
enough to accept the indicators as suitable classifying variables and, in short, may be used as an alternative methodology to
the DEA Model to measure efficiency.
 M. Caballer-Tarazona et al. / Mathematical and Computer Modelling 52 (2010) 1095–1102 1101

4. Discussion

In the Valencian Community there are some specialities in which the waiting list problem is particularly relevant:
internal medicine, gynaecology, traumatology and general surgery. Therefore, studying the efficiency of these services is
fully justified.
For this study, we firstly used the DEA Model. This model is highly suitable to measure the performance of health services
as it includes essential interactions between inputs and outputs which are present in the productive process.
In addition, the fact that several healthcare managing models exist in the same AC calls for thorough comparative studies
to be conducted into operative efficiency and the management structure and efficiency in its hospitals. The idea is to optimise
the available resources and, at the same time, to guarantee a homogeneous quality health and welfare system for all its
citizens.
On the other hand, improving the quality of the data analysed in terms of inputs and outputs has been considered for
future research.
In relation to outputs, it is advisable to include qualitative variables as, for instance, users’ satisfaction. Regarding inputs,
it is advisable to include synthetic variables which summarise the information from the different variables, for the purpose
of ensuring that all the inputs present in the productive process are considered without excessively increasing the number
of inputs as this could entail a loss of model validity.

5. Conclusions

The fundamental objective of this research work is to offer simple tools to measure efficiency in hospitals in the Valencian
Community. The importance of analysing operative efficiency in hospital units has been highlighted in the Introduction.
Nonetheless, this type of efficiency analysis becomes particularly relevant if we apply it in the context of the Valencian
Community because both the DEA Model and the construction of efficiency indicators are still highly novel and relevant
operative performance analysis methods for hospitals efficiency measurement. Efficiency analysis is still an unsolved
problem in Valencian hospitals and one that has been poorly dealt with.
This objective is of special interest as several service management models exist in this particular AC.
Several conclusions have been obtained with the different analyses done to approach the measure of efficiency.
Firstly, the efficiency analysis using the DEA Model is considered more useful when studying the efficiency of each service
separately instead of studying the overall efficiency of a given hospital. This is because one hospital may present efficient
services but inefficiencies may exist in some other services.
However, the DEA Model may present hospital directors with practical difficulties. For this very reason, the design of two
easily constructed, user-friendly indicators has been proposed as an operative tool to measure efficiency in various hospital
services.
In that sense, once the indicators has been obtained; I1 (Weighted admissions/doctors) and I2 (Interventions/doctors);
their effectiveness has been verified by means of a discriminant analysis. This analysis offers a discriminant function for
each service under study whereby high percentages of correct classifications of hospital units as efficient and inefficient
were obtained by simply using the aforementioned indicators as classifying variables.
Therefore, we may conclude that the indicators here proposed are indeed an alternative measure of efficiency to the DEA
Model.
With these results, healthcare administrations are recommended to provide hospitals with the mean and standard
deviation of the efficiency indicators to serve as self-assessment guidelines for each hospital’s activity.
Additionally, hospitals may also be provided with the previously calculated discriminant functions so that each hospital
service could accurately calculate if it is indeed efficient or inefficient.
As regards this type of management model, we finally conclude that the levels of efficiency of the services analysed
were above the mean. In the future, it would be interesting to make further comparisons to study other services and other
hospitals which are run under the Administrative Allowance Model.

References

[1] J. Barea, A. Gómez, El problema de la eficiencia del sector público en España, Especial consideración de la sanidad, Publicación del Instituto de Estudios
Económicos, Madrid, 1994, ISBN: 84-88533-09-8.
[2] A. Duran, J.L. Lara, M. Waveren, Snapshots of health systems: Spain, in: European Observatory on Health Systems and Policies, 2006.
www.euro.who.int/observatory.
[3] E. Tarazona, R. Torner, M. Marín, La experiencia del ‘‘Modelo Alzira’’ del Hospital de La Ribera a la Ribera-Área 10 de salud, la consolidación del modelo,
Revista de Administración Sanitaria 3 (1) (2005) 83–98.
[4] J. Magmussen, K. Nyland, Measuring efficiency in clinical departments, Health Policy 87 (2008) 1–7.
[5] R. Belenes, Un balance personal de 25 años de gestión sanitaria moderna en el Sistema Nacional de Salud, Gaceta Sanitaria 17 (2) (2003).
[6] L.A. Aday, C.E. Begley, D.R. Larison, Slater, Evaluating the Medical Care System: Effectiveness, Efficiency, and Equity, Health Administration Press,
Ann Arbor, MI, 1993, p. 30.
[7] J. Barea Tejeiro, Organización hospitalaria y eficiencia, Revista de Gestión y Evaluación de Costes Sanitarios (Marzo) 2 (1) (2001).
[8] J. Puig-Junoy, Eficiencia en la atención primaria de salud: una revisión crítica en las medidas de frontera, Revista Española de salud pública 74 (2000)
483–495.
[9] M. Staat, Efficiency of hospitals in germany: a DEA-boot strap approach, Applied Economics 38 (19) (2006) 2255–2263.
1102 M. Caballer-Tarazona et al. / Mathematical and Computer Modelling 52 (2010) 1095–1102

[10] F. Dexter, Sensitivity of super-efficient data envelopment analysis results to individual decision-making units: an example of surgical workload by
specialty, Health Care Management Science 11 (4) (2008) 307–318.
[11] K. Field, A. Emrouznejad, Measuring the performance of mental care units in Scotland, Journal of Medical System 27 (4) (2003) 315–324.
[12] M. Katharaki, Approaching the management of hospital units with an operation research technique: the case of 32 Greek obstetric and gynaecology
public units, Health Policy (2007).
[13] N. Kontodimopoulos, D. Niakas, Efficiency measurement of haemodialysis units in Greece with data envelopment analysis, Health Policy 71 (2005)
195–204.
[14] E. Sánchez, D. Vivas, Efficiency and quality measurement in hospital acute units: a multi-criteria model, in: International Health Economic Association,
Inaugural Conference, Vancuver, 1996.
[15] A. Charmes, W.W. Cooper, E. Rhodes, Measuring the efficiency of decision making units, European Journal of Operational Research 2 (1978) 429–444.
[16] R. Banker, A. Charmes, W. Cooper, Some models for estimating technical and scale inefficiencies in data envelopment analysis, Management Science
30 (9) (1984) 1078–1092.
[17] R. Jacobs, P. Smith, A. Street, Measuring Efficiency in Health Care. Analytic Techniques and Health Policy, Cambridge University Press, 2006 (Chapter 5).
[18] A. Charnes, W. Cooper, A. Lewin, L. Seiford, Data Envolment Analysis. Theory, Methodology and Applications, Kluwer Academic Publishers, 1994
(Chapters 1–3).
[19] T. Coelli, D. Rao, G. Battese, An Introduction to Efficiency and Productivity Analysis, Kluwer Academic Publishers, London, 1999.
[20] B. González, Análisis Multivariante. Aplicación al Ámbito Sanitario, SG Editores, SA, Barcelona, 1991.