Multi-expert performance evaluation of healthcare institutions by Ioan Degău

Knowledge-Based Systems 133 (2017) 90–106

Contents lists available at ScienceDirect

Knowledge-Based Systems journal homepage: www.elsevier.com/locate/knosys

Multi-expert performance evaluation of healthcare institutions using an integrated intuitionistic fuzzy AHP&DEA methodology ˙ Otay a,∗, Basar Oztaysi b, Sezi Cevik Onar b, Cengiz Kahraman b Irem a b

Okan University, Faculty of Engineering, Dept. of Industrial Engineering, Akﬁrat-Tuzla 34959, Istanbul, Turkey Istanbul Technical University, Management Faculty, Dept. of Industrial Engineering, Macka, Besiktas 34367, Istanbul, Turkey

a r t i c l e

i n f o

Article history: Received 25 September 2016 Revised 13 June 2017 Accepted 20 June 2017 Available online 21 June 2017 Keywords: Healthcare Data Envelopment Analysis (DEA) Analytic Hierarchy Process (AHP) Fuzzy Triangular Intuitionistic Fuzzy Set (TIFS)

a b s t r a c t Healthcare management and healthcare industry have been one of the popular and complex topics that many researchers and professionals have focused on. This paper proposes a new multi-expert fuzzy approach integrating intuitionistic fuzzy Data Envelopment Analysis (DEA) and intuitionistic fuzzy Analytic Hierarchy Process (IF-AHP) for solving the performance evaluation problem of healthcare institutions. In this paper, intuitionistic fuzzy sets (IFS) have been preferred since they simultaneously provide information on the membership, non-membership, and hesitancy functions. A real life problem is demonstrated to validate the proposed methodology. A total number of 16 hospitals operating in Istanbul have been analyzed based on a broad set of inputs and outputs. Then, a comparison with crisp DEA has been performed. © 2017 Elsevier B.V. All rights reserved.

1. Introduction Performance evaluation has a great importance for decision makers in healthcare resources allocation and service quality improvement [1]. The researchers highlighted that hospital performance evaluation has been gaining importance for a variety of reasons such as increasing health expenditures and competition in the health sector all around the world [2,3]. The World Health Report 20 0 0 by The World Health Organization (WHO) has stated that there has been a widespread concern with regard to measuring health system performances among researchers, policy makers and managers of hospitals [4]. Hospitals are an important component of healthcare organizations and have a signiﬁcant impact on the success of health care system of a country [5]. For eﬃcient use of resources, measuring performances of the hospitals and improvements needed to be done are sine qua non [6]. Today, the health sector is one of the sectors getting the largest share from public spendings. As stated in the Universal Declaration of Human Rights (Article 25th), many governments invest large amounts on healthcare for sustainable and continuous improvement [7]. The performance of hospitals is measured by various methods in the literature [8]. Data Envelopment Analysis (DEA) is one of the frequently used and most popular approaches that measures the performance of similar units under a multi-output and multi-input ∗

Corresponding author. E-mail addresses: irem.otay@okan.edu.tr (I.˙ Otay), oztaysib@itu.edu.tr (B. Oztaysi), cevikse@itu.edu.tr (S. Cevik Onar), kahramanc@itu.edu.tr (C. Kahraman). http://dx.doi.org/10.1016/j.knosys.2017.06.028 0950-7051/© 2017 Elsevier B.V. All rights reserved.

environment. Traditional DEA is a deterministic method where all inputs and outputs are quantitative and identiﬁed with numerical values. However, there is an inherent complex and uncertain nature of real world applications. In case of imprecise and vague data, people are more likely to prefer making linguistic evaluations rather than using exact numerical values. The fuzzy set theory was initiated for dealing with uncertainty and ambiguity [9]. In the literature, there are different extensions of ordinary fuzzy sets such as type 2 fuzzy sets, fuzzy multi-sets, intuitionistic fuzzy sets, hesitant fuzzy sets, and Pythagorean fuzzy sets. All these extensions intend better representing the vague and imprecise information. Intuitionistic fuzzy sets whose components are membership, nonmembership and hesitancy functions are one of the most used extensions for handling vagueness and impreciseness [10]. The superiority of intuitionistic fuzzy sets comes from their ability to deal with hesitancy of decision makers in their decisions. These type of sets successfully model the fuzziness and hesitancy based on well established mathematical theorems. In this study, we utilize intuitionistic fuzzy sets for handling vagueness, imprecision and hesitancy in the performance evaluation of hospitals. Analytic Hierarchy Analysis (AHP), developed by Saaty [11], is a multi-criteria decision making method, which is based on pairwise comparisons among criteria and alternatives. AHP divides a huge problem into small and easy to handle problems. AHP is one of the most reliable methods in weighting criteria. Classical AHP uses a linguistic scale whose numerical values are between 1 and 9. However, according to Buckley [12], the representation of a linguistic term by an exact number may not reﬂect the de-

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

cision maker’s judgments in his/her mind. For instance, a linguistic evaluation such as “Very Strong Importance” is represented by 7 in the scale of classical AHP. However, the decision maker’s judgment “Very Strong Importance” may not be clear enough to assign ‘7’. By “Very Strong Importance”, the decision maker might mean around seven which corresponds to a fuzzy number such as (6.5, 7, 7.5). It provides a better representation of the decision maker’s evaluation. Fuzzy sets are excellent tools for dealing with such type of uncertainties. On the other hand, in some of the fuzzy AHP methods, the evaluations are defuzzified at the initial stages of the method [13]. In those cases, loss of information is caused at the beginning and the advantage of better representation of the uncertainties are not achieved. Fuzzy AHP in our proposed method applies the defuzzification in the latest stage, which reserves the fuzzy information throughout the process. Besides, the intuitionistic fuzzy sets can reflect the hesitancy of decision makers whereas classical sets cannot deal with it. This study proposes an integrated triangular intuitionistic fuzzy (TIF) AHP and TIF-DEA approach [14] for analyzing and solving hospital performance evaluation problem. Since the inputs and outputs have different impacts on the performances of hospitals, we used AHP method to weigh them. TIF-DEA can deal with both tangible and intangible inputs and outputs. Especially, it can simultaneously evaluate the efficiency of many alternatives. Therefore, we integrated DEA method with AHP method. The employed DEA method is based on Constant Returns to Scale (CRS) model since the scale of economies does not change as the sizes of hospitals increase. In the study, inputs and outputs are evaluated by TIF AHP and the weights obtained from TIF AHP are used in TIFDEA analysis. To the best of the authors’ knowledge, this may be the first study conducting a multi-expert decision making analysis by integrating fuzzy AHP and fuzzy DEA methods employing Triangular Intuitionistic Fuzzy Numbers and applying the proposed approach for a real life hospital performance evaluation problem. In the application section, 16 public and private healthcare institutions operating in a particular city are analyzed based on seven inputs and six outputs. In the study, all inputs and outputs are defined by using intuitionistic fuzzy sets. The findings are then compared with the ones derived from traditional DEA approach based on CRS model. The rest of the paper is organized as follows: In Section 2, we provide a detailed literature review on performance evaluation of healthcare institutions particularly hospitals. In Section 3, we describe preliminaries on intuitionistic fuzzy sets. In Section 4, we present our intuitionistic fuzzy DEA model. In Section 5, we propose a new integrated multi-expert intuitionistic fuzzy AHP&DEA methodology. In Section 6, we apply the proposed fuzzy integrated methodology to a real life hospital performance evaluation problem and demonstrate the findings of the proposed approach. In this section, we also perform a comparative analysis, illustrate and discuss the results of the analysis with the findings of the proposed method. In the last section we conclude the paper and present future remarks. 2. Literature review on performance evaluation of healthcare institutions There are a lot of papers on the performance evaluation of healthcare institutions in the literature. For instance, O’Neill et al. [15] made the first taxonomy on hospital efficiency by reviewing approximately 79 studies published between 1984 and 2004, and presented studies using Data Envelopment Analysis and other related techniques. We first classified the papers which analyze healthcare efficiency evaluation problems into 2 groups. The primer is the studies employing classical DEA method and the next is the ones using the

fuzzy set theory. The literature review highlights that the classical DEA and fuzzy DEA papers can be also divided into 3 sub-groups: • •

•

The papers employing only classical or fuzzy DEA, The papers integrating classical or fuzzy DEA with another approach (stochastic, mathematical models, or simulation) and The papers proposing new and modiﬁed DEA models for the healthcare performance evaluation problem.

There are many studies applying classical DEA methods. Even though they can cope with large number of inputs and outputs, they are unable to handle the vagueness and impreciseness in the values of inputs and outputs. For instance, Kazley and Ozcan [16] employed Data Envelopment Analysis and windows analysis to analyze the relationship between electronic medical record use and efficiency for small, medium and large hospitals. The papers integrating DEA models with other approaches can provide the necessary data for the parameters of DEA models. For example, the studies integrating Analytical Hierarchy Process (AHP) with DEA model obtain the weights of inputs & outputs from AHP method and use the weights in the DEA model. Lai et al. [17] suggested integrated knowledge-based system as an benchmarking tool using CCR (Charnes, Cooper and Rhodes) and BCC (Banker, Charnes and Cooper) DEA models for medical centers, followed by regional hospitals and district hospitals in Taiwan. Mitropoulos et al. [21] developed a new approach - a chance constrained DEA (CCDEA) model–integrated it with a stochastic mechanism from Bayesian techniques, and implemented the proposed approach for obtaining efficiency scores of 117 Greek public hospitals. Leleu et al. [19] aimed to assess hospital patient payer mix and technical efficiency and check variations in profits. The researchers applied a weighted DEA and used the data of 138 hospitals locating in Florida for the year 2005. Al-Refaie et al. [5] applied simulation and DEA for improving the performance of emergency department of a hospital in Jordan. For this aim, they focused on reducing the average waiting time of the patients in the emergency department, improving the utilization of nurses, and at the same time increasing the number of served patients. Fang and Li [20] evaluated facility location - allocation problems. The researchers utilized DEA for evaluating the relative efficiency of potential locations and used the solutions obtained from DEA as a goal in a multi objective decision making model. Mitropoulos et al. [18] presented a methodology which combined DEA and integer programming location allocation models as medium and long term decision tools. The authors evaluated efficiencies of health service providers, and used these efficiencies to determine locations of health providers and service allocations. Chowdhury and Zelenyuk [22] analyzed production performance of hospital services in Canada through DEA and used truncated regression estimation with double-bootstrap the significance of determinants. The analysis displayed that some factors such as occupancy rate, rate of unit-producing personnel, outpatient–inpatient ratio, geographic locations, size and teaching status were significant determinants of efficiency. Bahadori et al. [23] integrated DEA and PROMETHEE methods to evaluate performance evaluation of selective wards in a military hospital. Khushalani and Ozcan [24] focused on efficiency of producing quality and hospital sub-divisions from 2009 to 2013 by means of Dynamic Network DEA. In addition to the benefits of employing integrated approaches, this group of papers does not take into account the uncertainties both in inputs and outputs. Studies modifying classical DEA method generally handle limited number of inputs and outputs related to healthcare problems. We can give Ouellette and Vierstraete [25] as an example of the studies proposing modified classical DEA. The researchers modified classical DEA approach by introducing quasi-fixed inputs and used the method for evaluating efficiency of hospital emergency services in Montreal. Wei et al. [26] applied Super-DEA-R-I

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

(DEA-R-based input oriented high-efficiency model) for analyzing the medical sectors in Taiwan. On the other hand, there are also some studies implementing fuzzy methodology in performance evaluation problem in the healthcare literature. These studies aim to capture the vagueness in the evaluations of inputs and outputs and to incorporate the vagueness into the model. For instance, Ebrahimnejad [27] applied a fuzzy DEA approach where the input prices were defined with trapezoidal fuzzy numbers, for evaluating insurance organizations and hospitals. The authors employed the linear ranking functions when comparing the fuzzy numbers and considered four inputs namely “the number of personnel, the total number of computers, the area of the branch and administrative expenses” and four outputs which were “the total number of insured persons, the number of insured person’s agreements, the total number of life-pension receivers and the receipt total sum”. Costantino et al. [28] suggested using a cross-efficiency fuzzy DEA for solving healthcare systems performance evaluation problems in the southern part of Italy. Chang [29] concentrated on hospital service evaluation in Taiwan. The researcher analyzed two public and three private medical centers based on 33 evaluation criteria and proposed fuzzy VIKOR (Vise Kriterijumska Optimizacija I Kompromisno Resenje) approach with triangular fuzzy numbers. In the literature, some other studies integrate fuzzy methods with other approaches. Khodaparasti and Maleki [30] presented a new combined fuzzy dynamic model for locating emergency vehicles and ambulance stations in an emergency medical service. In the study, the input and output parameters were defined as fuzzy numbers. Akdag et al. [31] evaluated the hospital service quality by integrating fuzzy AHP, fuzzy Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) and Yager’s min-max approach. Karadayi and Karsak [3] proposed an imprecise DEA for determining performances of state hospitals operating in 26 districts of Istanbul and obtained Optimistic and Pessimistic Scenario Efficiency Scores. In their study, three inputs which were “number of beds, number of overall staff and operating expenses” and five outputs such as “number of outpatients, number of discharged patients, tangibility and responsiveness” were evaluated. Muriana et al. [32] proposed an expert system for evaluating financial performances of health care structures employing the fuzzy set theory and key performance indicators. Arya and Yadav [33] proposed dual Slack Based Measure (SBM) model with fuzzy DEA and applied the developed model to the health sector. They evaluated the number of doctors, staff nurses and pharmacists as input variables and number of inpatients and outpatients as output variables which were identified with triangular fuzzy numbers. Ameryoun et al. [34] proposed a novel analysis based on DEA and SERVQUAL (Service Quality) for evaluating influence of service quality on hospitals. Table 1 summarizes the studies

3. Preliminaries: intuitionistic fuzzy sets In the fuzzy set theory, the membership of an element to a fuzzy set is a single value between zero and one. However, the degree of non-membership of an element in a fuzzy set may not be equal to 1 minus the membership degree since there may be some hesitation degree. Therefore, a generalization of fuzzy sets was proposed by Atanassov [38] as intuitionistic fuzzy sets (IFS) which incorporate the degree of hesitation, which is deﬁned as “1 ’ minus the sum of membership and non-membership degrees. Deﬁnition 1. Let X = ∅ be a given set. An intuitionistic fuzzy set in X is an object A given by

A˜ =

• •

Implementing only classical or fuzzy DEA approaches, Presenting integrated approaches and Proposing new and modiﬁed classical or fuzzy DEA approaches for measuring the eﬃciency of the healthcare actors such as hospitals and healthcare insurance companies.

This literature review shows us that healthcare performance evaluation is applied to various healthcare problems with different scopes and it requires both objective and subjective criteria to be used. Handling the vagueness in the subjective criteria is possible through the fuzzy set theory and ordinary fuzzy sets have been often utilized with this aim. The extensions of fuzzy sets have been recently utilized in performance evaluation [36,37]. Among fuzzy extensions, intuitionistic fuzzy sets are the most used extension.

(1)

where μA˜ : X → [0, 1] and vA˜ : X → [0, 1] satisfy the condition

0 ≤ μA˜ (x ) + vA˜ (x ) ≤ 1,

(2)

for every x X. Hesitancy is equal to “1-(μA˜ (x ) + vA˜ (x ))” Deﬁnition 2. Intuitionistic fuzzy numbers (IFNs): An IFN A˜ is deﬁned as follows: (i) An intuitionistic fuzzy subset of the real number line (ii) Normal, i.e., there is any x0 ∈ R such that μA˜ (x0 ) = 1 (so vA˜ (x0 ) = 0) (iii) A convex set for the membership function μA˜ (x ), i.e.,

μA˜ (λx1 +(1−λ )x2 )≥min μA˜ (x1 ), μA˜ (x2 ) ∀x1 , x2 ∈ R, λ ∈ [0, 1]

(3) (iv) A concave set for the non-membership function vA˜ (x ), i.e.,

vA˜ (λx1 +(1−λ )x2 ) ≤ max vA˜ (x1 ), vA˜ (x2 ) ∀x1 , x2 ∈ R, λ ∈ [0, 1]. (4) Deﬁnition 3. α -cut of an intuitionistic fuzzy set is given by the following equations: For an intuitionistic fuzzy set the α -cut is deﬁned by Atanassov [38] as the set

A˜ α = x ∈ X |μA˜ (x ) ≥ α , vA˜ (x ) ≤ 1 − α .

(5)

Deﬁnition 4. Triangular Intuitionistic Fuzzy Numbers (TIFN): A TIFN A˜ is an intuitionistic fuzzy subset in R with following membership function and non-membership function:

μA˜ (x ) =

⎧ x − aL ⎪ ⎪ , f or aL ≤ x ≤ aM ⎪ ⎨ aM − aL aU − x

(6)

, f or aM ≤ x ≤ aU ⎪ U M ⎪ ⎪ ⎩a − a 0, otherwise

and •

x, μA˜ (x ), vA˜ (x ) ; x X ,

vA˜ (x ) =

⎧ M a −x ⎪ ⎪ , f or a L ≤ x ≤ aM ⎪ ⎨ aM − a L x − aM

(7)

, f or aM ≤ x ≤ a U ⎪ U M ⎪ ⎪ ⎩a − a 1, otherwise

where a ≤ aL ≤ aM ≤ aU ≤ a U , 0 ≤ μA˜ (x ) + vA˜ (x ) ≤ 1 and TIFN

denoted

A T IF N =

(aL , aM , aU ; a L , aM , a U ) (see Fig. 1).

If A˜ T IF N = (aL , aM , aU ; a L , aM , a U ) and B˜T IF N = (bL , bM , bU ; b L , are two TIFNs, then Addition C˜ = A˜ + B˜ is also a TIFN:

bM , b U )

C˜= aL + bL , aM + bM , aU + bU ; a L + b L , aM + bM , a U + b U . (8)

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

Table 1 Studies implementing DEA for the healthcare sector. Authors

Application area

Classical DEA / Fuzzy DEA

Integrated Methods

Ouellette and Vierstraete [25]

Hospital emergency services in Montreal

Modiﬁed Classical DEA with quasi-ﬁxed inputs (CRS and VRS)

Kazley and Ozcan [16]

The relationship between electronic medical record use and eﬃciency for small, medium and large hospitals

Classical DEA (CRS)

DEA with windows analysis

Wei et al. [26]

The medical sectors in Taiwan

Modiﬁed DEA-R model: Super-DEA-R-I (DEA-R-based input oriented high-eﬃciency model)

Mitropoulos et al. [18]

Eﬃciencies of health service providers and location allocation models

Classical DEA (CCR)

DEA and Integer programming

Al-Refaie et al. [5]

Emergency department of a hospital in Jordan

Classical DEA (CCR)

Simulation and DEA

Leleu et al. [19]

Hospital patient payer mix and technical eﬃciency using the data of 138 hospitals in Florida for 2005

Classical weighted DEA

Regression

Lindlbauer et al. [35]

German acute care hospitals

Classical DEA (BCC)

DEA with difference-in-difference estimation and genetic matching

Chowdhury and Zelenyuk [22]

Production performance of hospital services in Canada

Classical DEA (CRS)

-Truncated regression estimation with double-bootstrapdistributional analysis

Mitropoulos et al. [21]

Eﬃciency of 117 Greek public hospitals

Chance constrained DEA (CCDEA) model

Chance constrained DEA (CCDEA) model with a Bayesian technique

Inputs/Outputs Inputs: Hours worked— excluding physicians—and expenditure on furniture and equipment), two quasi-ﬁxed inputs (number of stretchers and full time equivalent number of physicians), Output: Number of cases Inputs: Equipment, doctors, and nurses Outputs: Discharges, inpatient days, and staff training Inputs: Sickbeds and physicians, Outputs: Out-patients, in-patients, and surgeries

Inputs: Number of doctors, Number of nurses, Treatment population Outputs: Medical exams, Laboratory tests, Transfers Inputs: Number of nurses, Average time in system Outputs: Nurses’ utilization, Number of served patients Inputs: -Medical staff (FTE) -Others personnel (FTE) -Beds Outputs: -Case-mix adjusted total inpatient discharges -Outpatient in estimated days Inputs:Physician, Nurse, Supplies, Admin, Nonclinical, clinical, BedsOutputs:The number of treated in patient cases Inputs: Nursinghours Administrativehours Staffedbeds Medical surgical supplies costs (MSSC) Non-medical surgical supplies costs(NMSSC) Equipment expense(EE) MSSC+NMSSC MSSC+NMSSC+EE Outputs: Ambulatory visits Case-mix weighted inpatient days Inputs: Doctors, Other personnel, beds, operating cost Outputs: Inpatient admissions, outpatient visits (continued on next page)

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

94 Table 1 (continued) Authors

Application area

Lai et al. [17]

Medical centers, regional hospitals and district hospitals in Taiwan

Khushalani and Ozcan [24]

The eﬃciency of producing quality and hospital sub-divisions

Ebrahimnejad [27]

Insurance Organizations and Hospitals

Costantino et al. [28]

Healthcare systems performance evaluation problem in the southern of Italy

Khodaparasti and Maleki [30]

Locating emergency vehicles and ambulance stations

Karadayi and Karsak [3]

State hospitals operating in 26 districts of Istanbul

Muriana et al. [32]

Financial performances of health care structures

Arya and Yadav [33]

Application of the proposed model to the health sector

Classical DEA / Fuzzy DEA

Integrated Methods

CCR and BCC DEA models

Integrated knowledge-based systems as an benchmarking tool

Classical DEA

Multiplication: C˜ ∼ = A˜ B˜ is also a TIFN:

L L

C˜ ∼ = aL bL , aM bM , aU bU ; a b , aM bM , a b

Dynamic Network DEA

C˜ ∼ = aL /bU , aM /bM , aU /bL ; a L /b U , aM /bM , a U /b L

Cross-eﬃciency fuzzy DEA

Fuzzy DEA

A new combined fuzzy dynamic DEA model

Fuzzy DEA

Imprecise DEA with Optimistic and Pessimistic Scenario

Fuzzy DEA

DEA with the fuzzy set theory and key performance indicators

Fuzzy DEA

Dual slack based measure (SBM) model with fuzzy DEA

Inputs:Number of physicians, number of nurses, number of expensive equipment, and diversity of expensive equipment.Outputs:Number of inpatients, number of outpatient, and quality assurance Inputs: Number of beds, Total number of non-physician, Total operating expenses per bed. Outputs: Case-mix adjusted discharges, Total number of surgeries, Emergency visits, Outpatient visits. Inputs: The number of personals, The total number of computers, The area of the branch, Administrative expenses, Outputs: The total number of insured persons, The number of insured person’s agreements, The total number of life-pension receivers, The receipt total sum Inputs: Units of doctors, Units of nurses, Units of other staff, Number of beds Outputs: Number of discharges, Days of hospitalization, Number of surgeries Inputs: Number of existing ambulances, Number of stations, Standard distance coverage, Distances between demand points and sites Outputs: Number of treated patients Inputs: Number of beds, Number of overall staff, Operating expenses Outputs: Number of outpatients, Number of discharged patients, Number of adjusted surgeries,Tangibility, responsiveness

Inputs: Sum of number of doctors and staff nurses, Number of pharmacists Output: Number of inpatients Number of outpatients.

Multiplication with a constant:

Division: C˜ ∼ = A˜ B˜ is also a TIFN:

DEA (CCR) with trapezoidal fuzzy numbers

Inputs/Outputs

(9)

k × A˜ T IF N = k × aL , k × aM , k × aU ; k × a L , k × aM , k × a U , k > 0

(10)

(11)

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

4.1. Traditional DEA

Fig. 1. Membership and non-membership functions of TIFN.

3.1. Defuzziﬁcation of TIFSs

Let Ii = (aLi , aM , aUi ; ai L , aM , aiU ) be a triangular intuitionistic i i fuzzy number. Herein, we propose a new defuzziﬁcation function as in Eq. (12).

df =

aLi + aM + aUi a L + aM + aiU i i + i 3 τ

(12)

where τ is a very large number. This equation is based on the definition of accuracy function given in Zhang and Liu [39] and Liu et al. [40]. τ is the non-membership impact factor which determines the effect of non-membership function in the defuzziﬁcation. As it gets larger, the effect of non-membership function gets smaller. The magnitude of τ is determined by decision makers with respect to the type of decision making problem. 3.2. Aggregation operators for TIFNs

The DEA model is used to evaluate the relative efficiency of a given number of homogeneous decision making units (DMUs) such as bank branches and hospitals taking into account multiple input and multiple output, and solve linear programming problems consecutively for each single DMU [43,44]. DEA is defined as a frontier estimator based on a linear programming approach [22]. DEA simultaneously analyzes the efficiency of each unit (here hospital) using inputs to generate outputs by defining the optimal input/output combination. This combination is described as the “best practice frontier” and/or data envelope [45]. In the traditional Data Envelopment Analysis literature, there are various well-known DEA approaches [14,46]. Constant Returns to Scale (CRS) DEA model is preferred when the scale of economies does not change as size of the service facility increases. If inputs and outputs can be selected in such a way that they conform to the proportionality assumption, the discrimination of the CRS model is better when it is compared to other DEA models [47]. In this study, we use input oriented CRS model for the performance comparison of the hospitals. The essence of the CRS model is the maximization of the ratio of weighted multiple outputs to weighted multiple inputs. CRS model is given in the following [48]:

min θ s.t.

θ xi0 ≥ yr0 ≤

N j=1

j=1

(aLi , aM , aUi ; ai L , aM , ai U ) i i

λ j xi j

λ j yr j

(14)

Suppose Ii = is a set of Triangular Intuitionistic Fuzzy Numbers, then the result is a triangular intuitionistic fuzzy number aggregated by employing Eq. (13) [39].

λ j ≥ 0 ∀j = 1 , 2 , . . . , N

fm (I1 , I2 , . . . , In ) = ⎛ n

i = 1, 2, .., m;

1−

1 − aLi

⎜ i=1 ⎜ ⎝ n w n i L i=1

i=1

,1−

n i=1

1 − aM i

n M wi

i=1

U w i

w i

, 1−

n i=1

1 − aUi

w i

⎞ ,

j=1

λ j = 1, θ ≥ 0

where xi0

⎟ ⎟ ⎠

yr0

(13) where w = ( w1 , w2 , . . . , wn )T is the Ii (i = 1, 2, . . . , n), wi ∈ [0, 1], ni=1 wi = 1.

weight

vector

4. Intuitionistic fuzzy Data Envelopment Analysis Traditional DEA is an efficient performance evaluation method where there exists a multi-output and multi-input environment. Traditional DEA has been extended for various decision-making environments. For instance, Chance Constrained DEA and Stochastic Frontier Analysis DEA are some of these extensions. However, all inputs and outputs are quantitative and identified with numerical values in the traditional DEA method. Usually, the real world problems involve complexity, imprecision and uncertainty. The fuzzy set theory can capture this complexity and vagueness. After the fuzzy set theory has emerged, DEA has been extended to its fuzzy versions. Ordinary fuzzy DEA [41] and Type-2 fuzzy DEA [42] are two of these fuzzy extensions. Ordinary fuzzy sets force the decision makers to define a sharp membership function without considering their hesitancy. Intuitionistic fuzzy DEA can consider the hesitancy of decision makers in the evaluation process. In the following, we first briefly present traditional and ordinary fuzzy DEA and then propose our intuitionistic fuzzy DEA method.

xij yrj

μr vi m s N

λj θ

j = 1, 2, .., N;

r − 1, 2, .., s

Amount of input i for the oth observed decision making unit Amount of output r for the oth observed decision making unit Amount of input i for the jth decision making unit Amount of output r for the jth decision making unit Weight for the rth output given by oth decision making unit Weight for the rth input given by oth decision making unit Number of inputs Number of outputs Number of decision making unit (DMU) Shadow prices regarding the constraints limiting the eﬃciency of each DMU Eﬃciency.

4.2. Ordinary fuzzy DEA Kahraman et al. [41] presented one of the ﬁrst studies proposing fuzzy DEA analysis. In the study, the authors proposed a fuzzy mathematical programming model with multiple attributes using data envelopment analysis. Guo and Tanaka [49] proposed the following fuzzy CCR DEA model to handle the eﬃciency evaluation

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

problem with fuzzy input and output data.

max μt Yo μ, v

≤

s.t. vt Xo ≈ 1˜

(15)

μt Yj <∼ vt Xj (j = 1, 2, . . . , n) μ, v ≥ 0

maxμ,v μt yo − (1 − h )μt do s.t. vt co ≥ go vt xo − (1 − h )vt co = 1 − (1 − h )e vt xo + (1 − h )vt co ≤ 1 + (1 − h )e μt yj − (1 − h )μt dj ≤ vt xj − (1 − h )vt cj μt yj + (1 − h )μt dj ≤ vt xj + (1 − h )vt cj μ, v ≥ 0

(16)

( j = 1, 2, . . . , N )

where go indicates the optimal value of the objective function, 0 ≤ h ≤ 1 is a predeﬁned possibility level by decision makers, and e is equal to maxj=1,2,...,n (maxk=1,2,...,s cjk /xjk ). 4.3. Proposed triangular intuitionistic fuzzy DEA In case of vague information on inputs and outputs, intuitionistic fuzzy sets reﬂecting the hesitancy of decision makers can be utilized. The proposed intuitionistic fuzzy CRS DEA model is presented in Eq. (17).

min θ s.t. N

j=1 N

y˜r0 ≤

j=1

L j yr j ,

j=1

λ j yr j ,

j=1

λ j yr j ,

j=1

The proposed methodology is composed of an integrated TIF AHP and TIF DEA methods. The criteria, inputs and outputs have been determined based on a comprehensive literature review and viewpoints of the managers of hospitals. The ﬂowchart of the proposed methodology is displayed in Fig. 2. The steps of the proposed methodology with an illustrative example (IE) are stated as follows: Step 1: Apply the triangular intuitionistic fuzzy AHP method. Step 1.1: Deﬁne the problem and establish the hierarchy, goal being at the top, criteria and sub-criteria at the intermediate level and alternatives at the lowest level. IE 1.1: Consider that we have two alternatives (m = 2), two inputs (t = 2) and two outputs (u = 2) in a performance evaluation problem. In the illustrative example, we assume that there is a consensus among the experts. Step 1.2: Construct pairwise comparison matrices for inputs and outputs, and collect expert judgments using TIF scale given in Table 2. The pairwise evaluation matrix of each expert for obtaining the weights of inputs is as in Eq. (19).

⎣

⎡

⎢ A˜ To IF N = ⎢

U x˜i0 = xLi0 , xM i0 , xi0 ; xi0 , xi0 , xi0

L M U U yLr j , yM r j , yr j ; yr j , yr j , yr j

⎢

U L M U y˜r0 = yLr0 , yM r0 , yr0 ; yr0 , yr0 , yr0

⎣

( 1, 1, 1; 1, 1, 1 ) 1/a˜T12IF,iN .. .

···

IF N a˜T1t,i

( 1, 1, 1; 1, 1, 1 ) · · ·

IF N a˜T2t,i .. .

a˜T12IF,iN .. .

⎤ ⎥ ⎥ ⎥ ⎦

· · · ( 1, 1, 1; 1, 1, 1 )

IF N 1/a˜T2t,i

( 1, 1, 1; 1, 1, 1 ) 1/a˜T12IF,oN .. .

···

IF N a˜T1u,o

( 1, 1, 1; 1, 1, 1 ) · · ·

IF N a˜T2u,o .. .

a˜T12IF,oN .. .

IF N 1/a˜T1u,o

a˜T12IF,iN =

min θ

· · · ( 1, 1, 1; 1, 1, 1 )

IF N 1/a˜T2u,o

U L M U aL12,i , aM 12,i , a12,i ; a12,i , a12,i , a12,i

and

U θ xLi0 , θ xM i0 , θ xi0 ; θ xi0 , θ xi0 , θ xi0 N j=1

λ j xM ij ,

N j=1

⎥ ⎥ ⎥ ⎦

where u denotes the number of outputs. In Eq. (19), a˜T12IF,iN , for instance, is as follows:

s.t.

⎤

(20)

Eq. (17) can be expressed as in Eq. (18).

j=1

λ j yr j

j=1

where t denotes the number of inputs. The pairwise evaluation matrix of each expert for obtaining the weights of outputs is given in Eq. (20).

λ j xLi j ,

(19)

U L M U x˜i j = xLi j , xM i j , xi j ; xi j , xi j , xi j

5. A new integrated intuitionistic fuzzy AHP & DEA methodology

IF N 1/a˜T1t,i

where

≥

U j yr j ;

(18)

(17)

θ ≥0

y˜ r j =

θ ≥0

⎢

λj = 1

M j yr j ,

λj = 1

λ j ≥ 0 ∀ j = 1, 2, . . . , N N

j=1

⎢ A˜ Ti IF N = ⎢

λ j y˜ r j

j=1

⎡ λ j x˜i j

λ j ≥ 0 ∀j = 1 , 2 , . . . , N N

where Xj = (xj , cj ) is an s-dimensional fuzzy input vector and Yj = (yj , dj ) is an m-dimensional fuzzy output vector for the jth decision making unit. The fuzzy model can be transformed to the following LP problem as in Eq. (16).

θ x˜i0 ≥

U L M U yLr0 , yM r0 , yr0 ; yr0 , yr0 , yr0

λ j xUi j ;

1/a˜T12IF,iN = N j=1

λ j xiLj ,

N j=1

λ j xiMj ,

N j=1

λ j xiUj

aL12,i aU12,i aM 12,i

;

U L aM a12 a12 12,i ,i ,i

IE 1.2: The pairwise comparison matrices produced by three experts (Exp) including both inputs and outputs are exhibited below.

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

Fig. 2. Flowchart of the proposed methodology. Table 2 Scale for pairwise comparisons. Importance level

Corresponding intuitionistic fuzzy sets

Reciprocal importance level

Equal Importance (EI) Weak Importance (WI) Fairly Strong Importance (FSI) Very Strong Importance (VSI) Absolute Importance (AI)

(1, 1, 1; 1, 1, 1) (2, 3, 4; 1, 3, 5) (4, 5, 6; 3, 5, 7) (6, 7, 8; 5, 7, 9) (8, 9, 9; 7, 9, 9)

REI RWI RFSI RVSI RAI

Inputs

Outputs

Exp1

Exp2

Exp3

Exp1

Exp2

Exp3

I1 I2

EI RWI

EI RFSI

EI RWI

WI EI

FSI EI

WI EI

O1 O2

Reciprocal intuitionistic fuzzy sets (1, 1, 1; 1, 1, 1) ( 14 , 13 , 12 ; 15 , 13 , 1) ( 16 , 15 , 14 ; 17 , 15 , 13 ) ( 18 , 17 , 16 ; 19 , 17 , 15 ) ( 19 , 19 , 18 ; 19 , 19 , 17 )

Exp1

Exp2

Exp3

Exp1

Exp2

Exp3

EI RFSI

EI RVSI

FSI EI

VSI EI

g˜r = a˜Tr1IF N . . . a˜TrnIF N Step 1.3: Examine consistency of the fuzzy pairwise comparison matrices. To this end, the comparison matrix is defuzziﬁed and checked for consistency. If any inconsistency is detected, then the matrix is reformed again. IE 1.3: Using Eq. (12), we defuzziﬁed and checked the consistency of the pairwise comparison matrices in IE 1.2, and found that the comparison matrices are consistent. In fact, a 2 × 2 pairwise comparison matrix is always consistent and there is no need to calculate the consistency for this small example. Step 1.4: Calculate geometric mean of each row (r) in A˜ Ti IF N and A˜ To IF N matrices using Eqs. (21) and (22).

where

⎛

! 1n

aL , ⎜ ⎜ j=1 r j !1/n n ⎝ L ar j

g˜Tr IF N = ⎜

j=1

(21)

! 1n aM rj

n j=1

aM rj

!1/n

j=1

⎞

! 1n

⎟ ⎟ !1/n ⎟ n ⎠ U aUr j

j=1

;

(22)

ar j

IE 1.4: Employing Eq. (22), the geometric mean of pairwise comparison matrices for inputs and outputs are obtained as follows:

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

Inputs

I1 I2

(1,1,1;1,1,1) (0.22,0.28,0.4;0.18,0.28,0.69)

(2.52,3.56,4.58;1.44,3.56,5.59) (1,1,1;1,1,1)

Outputs

O1 O2

(1,1,1;1,1,1) (0.15,0.18,0.22;0.13,0.18,0.28)

(4.58,5.59,6.6;3.56,5.59,7.61) (1,1,1;1,1,1)

Step 2.2: Obtain TIF decision matrix for each decision maker as illustrated in Eq. (25). This matrix includes the whole inputs and outputs of the DEA model.

⎡

D˜ Tk IF N =

Step 1.5: Calculate triangular intuitionistic fuzzy weights of ˜ Tr,iIF N ) and the rth output (w ˜ Tr,oIF N ) using Eqs. the rth input (w (23) and (24), respectively.

# T IF N −1

T IF N d˜11 ,k ⎢d˜T IF N ⎢

T IF N d˜12 ,k d˜T IF N

.. .

··· .. .

T IF N d˜m 1,k

T IF N d˜m 2,k

···

⎢ 21. ,k ⎢ . ⎣ .

22,k

⎤

IF N d˜1Tn,k ⎥ d˜T IF N ⎥

d˜1T (IFnN−1),k d˜T IF N

···

2(n−1 ),k

2n,k

.. .

⎥ ⎥ ⎦

T IF N d˜mn,k

T IF N d˜m (n−1),k

for decision maker k

(25)

˜ Tr,iIF N = g˜Tr,iIF N g˜T1,iIF N . . . g˜T2,iIF N . . . g˜t,i w

(23)

where m is the number of alternatives and n is the total number of inputs and outputs.

IF N IF N ˜ Tr,oIF N = g˜Tr,oIF N g˜T1,o w . . . g˜T2,o . . . g˜u,o

(24)

IE 2.2: Using Table 3, linguistic values are converted to the fuzzy values in the DEA input & output matrix.

Alternatives

A1 A2

(60,65,70;15,23,30) (50,55,60;25,33,40)

(55,60,65;20,28,35) (25,33,40;50,55,60)

(15,23,30;60,65,70) (50,55,60;25,33,40)

(55,60,65;20,28,35) (65,70,75;10,18,25)

IE 1.5: Applying Eqs. (23) and (24) we obtain the following TIF weights of inputs and outputs: Inputs

I1 I2

(0.72,0.78,0.82;0.59,0.78,0.85) (0.16,0.22,0.33;0.11,0.22,0.59)

(0.45,0.78,1.30;0.22,0.78,2.29) (0.18,0.22,0.28;0.15,0.22,0.41)

Outputs

O1 O2

(0.82,0.85,0.87;0.78,0.85,0.88) (0.12,0.15,0.19;0.1,0.15,0.25)

(0.60,0.85,1.18;0.41,0.85,1.67) (0.13,0.15,0.18;0.12,0.15,0.22)

Step 1.6: Defuzzify fuzzy weights to determine importance weights of inputs and outputs using Eq. (12). IE 1.6: Then, we calculated the following normalized and defuzziﬁed TIF weights of inputs and outputs: Inputs I1 I2

wr,i T IF N 0.78 0.22

Outputs O1 O2

Step 2.3: Normalize experts’ evaluations in the DEA input & output matrices.

⎡

Norm d˜iTjIF N = ⎣

diLj , diMj , dUi j ; di Lj , diMj , diUj

max diUj

⎤ ⎦

(26)

∀i,j

IE 2.3: The DEA input & output matrix is normalized using Eq. (26) by dividing the values to the maximum value in the linguistic scale. Alternatives

A1 A2

(0.8,0.87,0.93;0.2,0.31,0.4) (0.67,0.73,0.8;0.33,0.44,0.53)

(0.73,0.8,0.87;0.27,0.37,0.47) (0.33,0.44,0.53;0.67,0.73,0.8)

Alternatives

A1 A2

(0.2,0.31,0.4;0.8,0.87,0.93) (0.67,0.73,0.8;0.33,0.44,0.53)

(0.73,0.8,0.87;0.27,0.37,0.47) (0.87,0.93,1;0.13,0.24,0.33)

Step 2.4: Aggregate the normalized input & output matrices considering the weights of experts based on Eq. (13).

wr,o T IF N

⎡ ˜T IF N

0.85 0.15

T IF N d˜12 ˜ dT IF N

.. .

T IF N d˜m 1

T IF N d˜m 2

d11 ⎢d˜T IF N

Step 2: Step 2.1: Use the scale given in Table 3 to assign the triangular intuitionistic fuzzy preferences into inputs and outputs matrices. IE 2.1: Input and output evaluations of alternatives using Table 3 are presented below: Alternatives

A1 A2

VH MH

H ML

VL MH

H AH

⎢ D˜ TagIF N = ⎢ ⎣

Membership & non-membership

Absolutely Low (AL) Very Low (VL) Low (L) Medium Low (ML) Equal (E) Medium High (MH) High (H) Very High (VH) Absolutely High (AH) Exactly Equal (EE)

(10,18,25;65,70,75) (15,23,30;60,65,70) (20,28,35;55,60,65) (25,33,40;50,55,60) (45,50,55;30,38,45) (50,55,60;25,33,40) (55,60,65;20,28,35) (60,65,70;15,23,30) (65,70,75;10,18,25) (50,50,50;50,50,50)

2(n−1 )

.. .

T IF N d˜m (n−1)

T IF N d˜mn

⎥ ⎥ ⎦

(27)

IE 2.4: Since there is a consensus among the experts, aggregation operations are not needed. So, this step is not applied in the illustrative example. Step 2.5: Obtain the weighted aggregated input & output matrix by multiplying the matrix in Step 2.4 by the weights of inputs & outputs derived from TIF AHP.

Table 3 Linguistic scale and its corresponding TIFN. Linguistic terms

··· .. . ···

⎤

d˜1TnIF N d˜T IF N ⎥

d˜1T (IFnN−1) d˜T IF N

···

⎡ IF N D˜ Tag,w

⎤ T IF N T IF N w1 × d˜11 w2 × d˜12 · · · w(n−1) × d˜1T (IFnN−1) wn × d˜1TnIF N

T IF N T IF N ⎢w1 × d˜21 w2 × d˜22 · · · w(n−1) × d˜2T (IFnN−1) wn × d˜2TnIF N ⎥ ⎢ ⎥ =⎢ ⎥ . . . . .. . . . . ⎣ ⎦ . . . . . T IF N T IF N T IF N ˜T IF N w1 × d˜m w2 × d˜m · · · w(n−1) × d˜m 1 2 (n−1) wn × dmn

IF N D˜ Tag,w

⎡ ˜i d11 i ⎢ d˜21 ⎢ =⎢ . ⎣ ..

i d˜m 1

where t + u = n.

··· ··· . . . ···

d˜1i t d˜i 2t

. . . i d˜

o d˜11 o d˜21

o d˜m 1

··· ··· . . . ···

⎤

d˜1ou d˜o ⎥ 2u

⎥

. ⎥ . ⎦ .

o d˜mu

(28)

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

Table 4 Explanations of inputs & outputs.

INPUTS

OUTPUTS

Inputs/Outputs

Explanations

References

I1: Patient care supplies and other expenses

This input is related to expenses including patient care supplies and other expenses.

Chirikos and Sear [50]; O’Neill et al. [15]

I2: Number of beds

This input pertains to the total number of beds that a hospital possesses.

Karadayi and Karsak [3]; Retzlaff-Roberts et al. [4]; O’Neill et al. [15]

I3: Number of physicians

This input is related to the total number of physicians in each public or private hospital.

Retzlaff-Roberts et al. [4]; O’Neill et al. [15]

I4: Number of nurses

This input is with regard to the total number of nurses working in each public or private hospital.

Lo et al. [51]; O’Neill et al. [15]

I5: Number of other personnel

This input is relevant to the total number of other personnel such as secretary working in each public or private hospital.

Karadayi and Karsak [3]; O’Neill et al. [15]

I6: Technology level

This input indicates capabilities of technology facilities in a hospital.

Jiang et al. [52]; Gok and Sezen [53];

I7: Service mix

This input represents a range of medical offerings of a hospital.

Harris II et al. [54]

O1: Annual revenues

This output includes annual revenues of hospitals.

Ayanog˘ lu et al. [6]

O2: Number of outpatient department visits

This output is about the total number of outpatient department visits in each public or private hospital.

Karadayi and Karsak [3]; O’Neill et al. [15]; Chowdhury and Zelenyuk [22]

O3: Overall patient satisfaction

This output describes overall satisfaction of each customer served by each hospital.

Gok and Sezen [53]; Hod et al. [55] Benneyan and Ceyhan [56]

O4: Number of inpatient department admissions

This output informs the total number of inpatient department visits in each public or private hospital.

O’Neill et al. [15]; Jat and Sebastian [57]

O5: Conformance to quality procedures

This output explains conformance level of each hospital with regard to quality procedures.

Morey et al. [58]

O6: Bed usage rate (%)

This output is related to the bed usage rate of each evaluated hospital.

Papadaki et al. [59]; Papadaki et al. [60]

Table 5 The pairwise comparisons of inputs.

Table 6 The pairwise comparisons of outputs.

Expert 1

I1 I2 I3 I4 I5 I6 I7

EI RWI AI FSI RWI VSI WI

WI EI AI FSI EI VSI FSI

RAI RAI EI RWI FSI RWI RFSI

RFSI RFSI WI EI RWI WI RWI

WI REI RFSI WI EI VSI FSI

RVSI RVSI WI RWI RVSI EI RWI

RWI RFSI FSI WI RFSI WI EI

Expert 1

O1 O2 O3 O4 O5 O6

EI RFSI WI RWI EI RVSI

FSI EI AI WI FSI RWI

RWI RAI EI RVSI RFSI RAI

WI RWI VSI EI WI RFSI

REI RFSI FSI RWI EI RVSI

VSI WI AI FSI VSI EI

Expert 2

I1 I2 I3 I4 I5 I6 I7

EI RVSI WI WI RFSI AI FSI

VSI EI AI FSI RAI VSI FSI

RWI RAI EI RFSI WI RFSI RVSI

RWI RFSI FSI EI RWI WI RFSI

FSI AI RWI WI EI AI FSI

RAI RVSI FSI RWI RAI EI RAI

RFSI RFSI VSI FSI RFSI AI EI

O1 O2 O3 O4 O5 O6

EI RVSI FSI RFSI RWI RFSI

VSI EI AI FSI VSI RWI

RFSI RAI EI RFSI RVSI AI

FSI RFSI FSI EI FSI RVSI

WI RVSI VSI RFSI EI RAI

FSI WI RAI VSI AI EI

Expert 3

I1 I2 I3 I4 I5 I6 I7

EI RFSI WI VSI RVSI FSI AI

FSI EI AI FSI RAI AI VSI

RWI RAI EI RFSI WI RFSI RVSI

RVSI RFSI FSI EI RWI FSI RWI

VSI AI RWI WI EI VSI FSI

RFSI RAI FSI RFSI RVSI EI RFSI

RAI RVSI VSI WI RFSI FSI EI

O1 O2 O3 O4 O5 O6

EI RFSI RWI EI RWI RAI

FSI EI RAI WI FSI RWI

WI AI EI RAI RFSI RVSI

REI RWI AI EI WI RFSI

WI RFSI FSI RWI EI RVSI

AI WI VSI FSI VSI EI

IE 2.5: The weighted input & output matrix is obtained as follows: Alternatives

A1 A2

(0.62,0.68,0.73;0.16,0.24,0.31) (0.52,0.57,0.62;0.26,0.34,0.41)

(0.16,0.18,0.19;0.06,0.08,0.1) (0.07,0.1,0.12;0.15,0.16,0.18)

Alternatives

A1 A2

(0.17,0.26,0.34;0.68,0.74,0.79) (0.57,0.62,0.68;0.28,0.37,0.45)

(0.11,0.12,0.13;0.04,0.06,0.07) (0.13,0.14,0.15;0.02,0.04,0.05)

Table 7 The weights of inputs & outputs. Weights of inputs

Weights of outputs

I1 I2 I3 I4 I5 I6 I7

O1 O2 O3 O4 O5 O6

0.068 0.035 0.283 0.165 0.045 0.292 0.112

0.275 0.065 0.309 0.113 0.199 0.039

100

Table 8 Evaluations of experts for DEA inputs. Hospital

(52,60,68;50,60,70) (53,5,61,5,69,5;51,5,61,5,71,5) (54,5,62,5,70,5;52,5,62,5,72,5)

(1530,1535,1540;1523,1535,1547) (1530,1535,1540;1523,1535,1547) (1525,1530,1535;1518,1530,1542)

(1780,1787,1794;1775,1787,1799) (1777,1784,1791;1772,1784,1796) (1778,1785,1792;1773,1785,1797)

(748,753,758;743,753,763) (753,758,763;748,758,768) (750,755,760;745,755,765)

(942,950,958;940,950,960) (944,952,960;942,952,962) (937,945,953;935,945,955)

(60,65,70;15,23,30) (60,65,70;15,23,30) (60,65,70;15,23,30)

(47,55,63;45,55,65) (48,56,64;46,56,66) (49,8,57,8,65,8;47,8,57,8,67,8)

(1370,1375,1380;1363,1375,1387) (1373,1378,1383;1366,1378,1390) (1375,1380,1385;1368,1380,1392)

(1493,1500,1507;1488,1500,1512) (1488,1495,1502;1483,1495,1507) (1497,1504,1511;1492,1504,1516)

(690,695,700;685,695,705) (695,700,705;690,700,710) (695,700,705;690,700,710)

(877,885,893;875,885,895) (879,887,895;877,887,897) (872,880,888;870,880,890)

(55,60,65;20,28,35) (60,65,70;15,23,30) (60,65,70;15,23,30)

(60,65,70;15,23,30) (65,70,75;10,18,25) (55,60,65;20,28,35)

(21,29,37;19,29,39) (23,5,31,5,39,5;21,5,31,5,41,5) (23,31,39;21,31,41)

(214,219,224;207,219,231) (217,222,227;210,222,234) (307,312,317;300,312,324)

(343,350,357;338,350,362) (346,353,360;341,353,365) (348,355,362;343,355,367)

(160,165,170;155,165,175) (164,169,174;159,169,179) (167,172,177;162,172,182)

(427,435,443;425,435,445) (425,433,441;423,433,443) (432,440,448;430,440,450)

(60,65,70;15,23,30) (60,65,70;15,23,30) (60,65,70;15,23,30)

(50,55,60;25,33,40) (55,60,65;20,28,35) (60,65,70;15,23,30)

(22,30,38;20,30,40) (25,33,41;23,33,43) (24,4,32,4,40,4;22,4,32,4,42,4)

(186,191,196;179,191,203) (210,215,220;203,215,227) (190,195,200;183,195,207)

(328,335,342;323,335,347) (331,338,345;326,338,350) (323,330,337;318,330,342)

(175,180,185;170,180,190) (180,185,190;175,185,195) (180,185,190;175,185,195)

(527,535,543;525,535,545) (522,530,538;520,530,540) (532,540,548;530,540,550)

(65,70,75;10,18,25) (65,70,75;10,18,25) (65,70,75;10,18,25)

(60,65,70;15,23,30) (65,70,75;10,18,25) (65,70,75;10,18,25)

(17,25,33;15,25,35) (19,5,27,5,35,5;17,5,27,5,37,5) (19,27,35;17,27,37)

(185,190,195;178,190,202) (182,187,192;175,187,199) (180,185,190;173,185,197)

(293,30 0,307;288,30 0,312) (291,298,305;286,298,310) (302,309,316;297,309,321)

(215,220,225;210,220,230) (209,214,219;204,214,224) (219,224,229;214,224,234)

(612,620,628;610,620,630) (614,622,630;612,622,632) (617,625,633;615,625,635)

(65,70,75;10,18,25) (60,65,70;15,23,30) (60,65,70;15,23,30)

(60,65,70;15,23,30) (55,60,65;20,28,35) (55,60,65;20,28,35)

(20,5,28,5,36,5;18,5,28,5,38,5) (19,8,27,8,35,8;17,8,27,8,37,8) (16,5,24,5,32,5;14,5,24,5,34,5)

(245,250,255;238,250,262) (242,247,252;235,247,259) (249,254,259;242,254,266)

(118,125,132;113,125,137) (120,127,134;115,127,139) (124,131,138;119,131,143)

(80,85,90;75,85,95) (90,95,100;85,95,105) (90,95,100;85,95,105)

(137,145,153;135,145,155) (141,149,157;139,149,159) (139,147,155;137,147,157)

(20,28,35;55,60,65) (50,55,60;25,33,40) (15,23,30;60,65,70)

(50,55,60;25,33,40) (55,60,65;20,28,35) (20,28,35;55,60,65)

(81,89,97;79,89,99) (87,95,103;85,95,105) (59,67,75;57,67,77)

(375,380,385;368,380,392) (415,420,425;408,420,432) (393,398,403;386,398,410)

(338,345,352;333,345,357) (358,365,372;353,365,377) (380,387,394;375,387,399)

(295,30 0,305;290,30 0,310) (310,315,320;305,315,325) (282,287,292;277,287,297)

(277,285,293;275,285,295) (281,289,297;279,289,299) (297,305,313;295,305,315)

(60,65,70;15,23,30) (65,70,75;10,18,25) (65,70,75;10,18,25)

(65,70,75;10,18,25) (50,55,60;25,33,40) (55,60,65;20,28,35)

(14,5,22,5,30,5;12,5,22,5,32,5) (12,5,20,5,28,5;10,5,20,5,30,5) (16,24,32;14,24,34)

(185,190,195;178,190,202) (180,185,190;173,185,197) (182,187,192;175,187,199)

(273,280,287;268,280,292) (278,285,292;273,285,297) (278,285,292;273,285,297)

(158,163,168;153,163,173) (160,165,170;155,165,175) (153,158,163;148,158,168)

(337,345,353;335,345,355) (334,342,350;332,342,352) (333,341,349;331,341,351)

(55,60,65;20,28,35) (55,60,65;20,28,35) (60,65,70;15,23,30)

(65,70,75;10,18,25) (60,65,70;15,23,30) (55,60,65;20,28,35)

(19,27,35;17,27,37) (17,8,25,8,33,8;15,8,25,8,35,8) (21,29,37;19,29,39)

(181,186,191;174,186,198) (184,189,194;177,189,201) (180,185,190;173,185,197)

(303,310,317;298,310,322) (304,311,318;299,311,323) (300,307,314;295,307,319)

(220,225,230;215,225,235) (205,210,215;200,210,220) (215,220,225;210,220,230)

(412,420,428;410,420,430) (409,417,425;407,417,427) (416,424,432;414,424,434)

(15,23,30;60,65,70) (50,55,60;25,33,40) (20,28,35;55,60,65)

(25,33,40;50,55,60) (55,60,65;20,28,35) (15,23,30;60,65,70)

H10

(17,25,33;15,25,35) (19,4,27,4,35,4;17,4,27,4,37,4) (16,24,32;14,24,34)

(204,209,214;197,209,221) (207,212,217;200,212,224) (216,221,226;209,221,233)

(213,220,227;208,220,232) (214,221,228;209,221,233) (207,214,221;202,214,226)

(133,138,143;128,138,148) (129,134,139;124,134,144) (130,135,140;125,135,145)

(227,235,243;225,235,245) (222,230,238;220,230,240) (223,231,239;221,231,241)

(55,60,65;20,28,35) (65,70,75;10,18,25) (50,55,60;25,33,40)

(50,55,60;25,33,40) (65,70,75;10,18,25) (60,65,70;15,23,30)

H11

(22,30,38;20,30,40) (24,7,32,7,40,7;22,7,32,7,42,7) (24,5,32,5,40,5;22,5,32,5,42,5)

(465,470,475;458,470,482 (462,467,472;455,467,479) (471,476,481;464,476,488)

(293,30 0,307;288,30 0,312) (290,297,304;285,297,309) (298,305,312;293,305,317)

(182,187,192;177,187,197) (175,180,185;170,180,190) (184,189,194;179,189,199)

(237,245,253;235,245,255) (233,241,249;231,241,251) (233,241,249;231,241,251)

(20,28,35;55,60,65) (50,55,60;25,33,40) (20,28,35;55,60,65)

(15,23,30;60,65,70) (60,65,70;15,23,30) (20,28,35;55,60,65)

H12

(11,5,19,5,27,5;9,5,19,5,29,5) (9,5,17,5,25,5;7,5,17,5,27,5) (14,5,22,5,30,5;12,5,22,5,32,5)

(145,150,155;138,150,162) (150,155,160;143,155,167) (148,153,158;141,153,165)

(219,226,233;214,226,238) (224,231,238;219,231,243) (224,231,238;219,231,243)

(130,135,140;125,135,145) (128,133,138;123,133,143) (127,132,137;122,132,142)

(167,175,183;165,175,185) (167,175,183;165,175,185) (163,171,179;161,171,181)

(15,23,30;60,65,70) (25,33,40;50,55,60) (20,28,35;55,60,65)

(25,33,40;50,55,60) (20,28,35;55,60,65) (15,23,30;60,65,70)

H13

(27,35,43;25,35,45) (23,2,31,2,39,2;21,2,31,2,41,2) (25,33,41;23,33,43)

(381,386,391;374,386,398) (387,392,397;380,392,404) (379,384,389;372,384,396)

(386,393,400;381,393,405) (391,398,405;386,398,410) (393,40 0,407;388,40 0,412)

(220,225,230;215,225,235) (213,218,223;208,218,228) (223,228,233;218,228,238)

(357,365,373;355,365,375) (354,362,370;352,362,372) (356,364,372;354,364,374)

(50,55,60;25,33,40) (55,60,65;20,28,35) (60,65,70;15,23,30)

(60,65,70;15,23,30) (60,65,70;15,23,30) (60,65,70;15,23,30)

H14

(37,45,53;35,45,55) (34,42,50;32,42,52) (34,5,42,5,50,5;32,5,42,5,52,5)

(528,533,538;521,533,545) (530,535,540;523,535,547) (530,535,540;523,535,547)

(543,550,557;538,550,562) (541,548,555;536,548,560) (541,548,555;536,548,560)

(410,415,420;405,415,425) (407,412,417;402,412,422) (405,410,415;400,410,420)

(487,495,503;485,495,505) (484,492,500;482,492,502) (482,490,498;480,490,500)

(15,23,30;60,65,70) (20,28,35;55,60,65) (50,55,60;25,33,40)

(25,33,40;50,55,60) (15,23,30;60,65,70) (55,60,65;20,28,35)

H15

(42,50,58;40,50,60) (42,50,58;40,50,60) (43,5,51,5,59,5;41,5,51,5,61,5)

(995,10 0 0,10 05;988,10 0 0,1012) (992,997,1002;985,997,1009) (990,995,10 0 0;983,995,10 07)

(668,675,682;663,675,687) (672,679,686;667,679,691) (673,680,687;668,680,692)

(384,389,394;379,389,399) (386,391,396;381,391,401) (380,385,390;375,385,395)

(512,520,528;510,520,530) (515,523,531;513,523,533) (517,525,533;515,525,535)

(20,28,35;55,60,65) (50,55,60;25,33,40) (20,28,35;55,60,65)

(25,33,40;50,55,60) (55,60,65;20,28,35) (15,23,30;60,65,70)

H16

(34,42,50;32,42,52) (34,7,42,7,50,7;32,7,42,7,52,7) (36,44,52;34,44,54)

(195,20 0,205;188,20 0,212) (197,202,207;190,202,214) (202,207,212;195,207,219)

(243,250,257;238,250,262) (249,256,263;244,256,268) (246,253,260;241,253,265)

(130,135,140;125,135,145) (134,139,144;129,139,149) (135,140,145;130,140,150)

(152,160,168;150,160,170) (154,162,170;152,162,172) (159,167,175;157,167,177)

(15,23,30;60,65,70) (55,60,65;20,28,35) (15,23,30;60,65,70)

(20,28,35;55,60,65) (65,70,75;10,18,25) (20,28,35;55,60,65)

(60,65,70;15,23,30) (60,65,70;15,23,30) (55,60,65;20,28,35)

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

λ j d˜ ji,k

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16

(0.59,0.68,0.78;0.55,0.65,0.76) (0.53,0.62,0.72;0.49,0.6,0.71) (0.24,0.33,0.42;0.22,0.32,0.43) (0.26,0.34,0.43;0.23,0.34,0.45) (0.2,0.29,0.37;0.17,0.28,0.39) (0.2,0.29,0.38;0.18,0.29,0.4) (0.81,0.89,0.98;0.78,0.89,1) (0.16,0.24,0.33;0.13,0.24,0.35) (0.21,0.3,0.39;0.18,0.29,0.4) (0.19,0.27,0.36;0.16,0.27,0.38) (0.26,0.34,0.43;0.23,0.34,0.44) (0.13,0.22,0.31;0.1,0.21,0.32) (0.27,0.36,0.45;0.25,0.36,0.46) (0.38,0.47,0.56;0.36,0.46,0.57) (0.46,0.55,0.65;0.43,0.54,0.65) (0.38,0.47,0.56;0.35,0.46,0.57)

(0.99,0.99,1;0.98,0.99,1) (0.89,0.89,0.89;0.88,0.89,0.9) (0.16,0.16,0.16;0.15,0.16,0.17) (0.13,0.13,0.13;0.12,0.13,0.14) (0.12,0.12,0.12;0.11,0.12,0.13) (0.16,0.16,0.17;0.15,0.16,0.17) (0.25,0.26,0.26;0.25,0.26,0.26) (0.12,0.12,0.12;0.11,0.12,0.13) (0.12,0.12,0.12;0.11,0.12,0.13) (0.13,0.14,0.14;0.13,0.14,0.15) (0.3,0.3,0.31;0.3,0.3,0.31) (0.1,0.1,0.1;0.09,0.1,0.11) (0.25,0.25,0.25;0.24,0.25,0.26) (0.34,0.35,0.35;0.34,0.35,0.35) (0.64,0.65,0.65;0.64,0.65,0.65) (0.13,0.13,0.13;0.12,0.13,0.14)

(0.99,0.99,0.99;0.99,0.99,1) (0.83,0.83,0.84;0.83,0.83,0.84) (0.19,0.20,0.20;0.19,0.20,0.20) (0.18,0.19,0.19;0.18,0.19,0.19) (0.16,0.17,0.17;0.16,0.17,0.17) (0.07,0.07,0.07;0.06,0.07,0.08) (0.20,0.20,0.21;0.20,0.20,0.21) (0.15,0.16,0.16;0.15,0.16,0.16) (0.17,0.17,0.18;0.17,0.17,0.18) (0.12,0.12,0.13;0.11,0.12,0.13) (0.16,0.17,0.17;0.16,0.17,0.17) (0.12,0.13,0.13;0.12,0.13,0.13) (0.22,0.22,0.22;0.21,0.22,0.23) (0.30,0.31,0.31;0.30,0.31,0.31) (0.37,0.38,0.38;0.37,0.37,0.38) (0.14,0.14,0.14;0.13,0.14,0.15)

(0.98,0.99,0.99;0.97,0.99,1) (0.91,0.91,0.92;0.9,0.91,0.93) (0.21,0.22,0.23;0.21,0.22,0.23) (0.23,0.24,0.25;0.23,0.24,0.25) (0.28,0.29,0.29;0.27,0.29,0.3) (0.11,0.12,0.13;0.11,0.12,0.13) (0.39,0.39,0.4;0.38,0.39,0.41) (0.21,0.21,0.22;0.2,0.21,0.22) (0.28,0.29,0.29;0.27,0.29,0.3) (0.17,0.18,0.18;0.16,0.18,0.19) (0.24,0.24,0.25;0.23,0.24,0.26) (0.17,0.17,0.18;0.16,0.17,0.19) (0.29,0.29,0.3;0.28,0.29,0.31) (0.53,0.54,0.55;0.53,0.54,0.55) (0.5,0.51,0.51;0.49,0.51,0.52) (0.17,0.18,0.19;0.17,0.18,0.19)

(0.98,0.99,1;0.98,0.99,1) (0.91,0.92,0.93;0.91,0.92,0.93) (0.45,0.45,0.46;0.44,0.45,0.46) (0.55,0.56,0.57;0.55,0.56,0.57) (0.64,0.65,0.66;0.64,0.65,0.66) (0.14,0.15,0.16;0.14,0.15,0.16) (0.3,0.3,0.31;0.29,0.3,0.31) (0.35,0.36,0.37;0.35,0.36,0.37) (0.43,0.44,0.45;0.43,0.44,0.45) (0.23,0.24,0.25;0.23,0.24,0.25) (0.24,0.25,0.26;0.24,0.25,0.26) (0.17,0.18,0.19;0.17,0.18,0.19) (0.37,0.38,0.39;0.37,0.38,0.39) (0.51,0.51,0.52;0.5,0.51,0.52) (0.54,0.54,0.55;0.53,0.54,0.56) (0.16,0.17,0.18;0.16,0.17,0.18)

(0.8,0.87,0.93;0.2,0.31,0.4) (0.78,0.84,0.91;0.22,0.33,0.43) (0.8,0.87,0.93;0.2,0.31,0.4) (0.87,0.93,1;0.13,0.24,0.33) (0.83,0.9,1;0.17,0.28,0.37) (0.41,0.5,0.59;0.59,0.68,0.77) (0.84,0.91,1;0.16,0.26,0.36) (0.76,0.82,0.89;0.24,0.35,0.45) (0.4,0.5,0.58;0.6,0.69,0.77) (0.77,0.84,1;0.23,0.34,0.44) (0.42,0.52,0.6;0.58,0.67,0.75) (0.26,0.37,0.46;0.74,0.8,0.87) (0.73,0.8,0.87;0.27,0.38,0.47) (0.4,0.5,0.58;0.6,0.69,0.77) (0.42,0.52,0.6;0.58,0.67,0.75) (0.42,0.52,0.62;0.58,0.67,0.76)

(0.78,0.85,0.92;0.22,0.33,0.42) (0.81,0.88,1;0.19,0.3,0.4) (0.73,0.8,0.87;0.27,0.38,0.47) (0.84,0.91,1;0.16,0.26,0.36) (0.76,0.83,0.9;0.24,0.35,0.44) (0.61,0.68,0.76;0.39,0.5,0.59) (0.78,0.86,1;0.22,0.33,0.42) (0.81,0.89,1;0.19,0.29,0.39) (0.47,0.56,0.65;0.53,0.63,0.71) (0.78,0.86,1;0.22,0.33,0.42) (0.49,0.59,0.7;0.51,0.62,0.71) (0.28,0.38,0.48;0.72,0.79,0.86) (0.8,0.87,0.93;0.2,0.31,0.4) (0.47,0.56,0.65;0.53,0.63,0.71) (0.47,0.56,0.65;0.53,0.63,0.71) (0.56,0.68,1;0.44,0.56,0.65)

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

(29)

Hospital

Step 2.6: Apply CRS based DEA model given in Section 4.3.

min θ s.t.

θ d˜ ei ,k ≥

j=1

λj = 1

m d˜ eo,k ≤ λ j d˜ jo,k

λ j ≥ 0 ∀j = 1 , 2 , . . . , m

j=1

θ ≥0

where $ % d˜ ji,k = dji,,Lk , dji,,M , dji,,U ; dji,,kL , dji,,kM , dji,,kU k k

d˜ ei ,k = dei,,Lk , dei,,M , dei,,Uk ; dei,,kL , dei,,kM , dei,,kU k

djo,,kL , djo,,kM , djo,,kU ; djo,,k L , djo,,k M , djo,,k U

d˜ eo,k = deo,,kL , deo,,kM , deo,,kU ; deo,,kL , deo,,kM , deo,,kU

d˜ jo,k =

0.00 0.00

d˜ ei ,k : Amount of input kth for the eth observed decision making unit d˜ eo,k : Amount of output kth for the eth observed decision making unit d˜ ji,k : Amount of input kth for the jth decision making unit d˜ jo,k : Amount of output kth for the jth decision making unit

0.28 0.00

Output Slacks

0.04 0.00

Input Slacks

0.00 0.00

θ : Eﬃciency value μk : Weight for the kth output given by eth decision making unit vk : Weight for the kth input given by eth decision making unit t : Number of inputs u : Number of outputs m : Number of decision making unit (DMU)

Eﬃciency

0.72 1.00

IE 2.6: When the TIF DEA problem in IE 2.5 is solved by the proposed model in Step 2.6, the following eﬃciency results are calculated.

Alternative

A1 A2

Step 2.7: Prioritize the hospitals with respect to their eﬃciencies (θ ) and advise the hospitals considering the values of input & output slacks and input & output targets.

6. An application

In this section, we analyze a real life hospital performance evaluation problem considering uncertainties in the decision making environment. A total number of 16 hospitals operating in Istanbul have been evaluated. These hospitals have been chosen among both private and public hospitals. According to their privacy policies, the names of hospitals are not shared. For each hospital, we gather the judgments of three experts from different departments namely procurement, ﬁnance, and logistics. In the study, seven inputs, namely patient care supplies and other expenses, the number of beds, number of nurses, number of physicians, number of other personnel, technology level and service mix and six outputs, namely annual revenue, number of outpatient department visits, overall patient satisfaction, number of

Table 9 Aggregated evaluations of experts for DEA inputs.

101

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

102 Table 10 Weighted aggregated evaluations for DEA inputs. Hospital

(0.04,0.047,0.054; 0.037,0.045,0.052) (0.036,0.042,0.049; 0.034,0.041,0.048) (0.017,0.023,0.029; 0.015,0.022,0.03) (0.018,0.024,0.03; 0.016,0.023,0.03) (0.014,0.02,0.026; 0.012,0.019,0.027) (0.014,0.02,0.026; 0.012,0.02,0.027) (0.055,0.061,0.067; 0.054,0.061,0.068) (0.011,0.017,0.023; 0.009,0.016,0.024) (0.014,0.02,0.026; 0.013,0.02,0.027) (0.013,0.019,0.025; 0.011,0.019,0.026) (0.018,0.024,0.03; 0.016,0.023,0.03) (0.009,0.015,0.021; 0.007,0.014,0.022) (0.019,0.025,0.031; 0.017,0.024,0.032) (0.026,0.032,0.038; 0.024,0.032,0.039) (0.032,0.038,0.044; 0.03,0.037,0.044) (0.026,0.032,0.038; 0.024,0.031,0.039)

(0.035,0.035,0.035; 0.035,0.035,0.035) (0.031,0.032,0.032; 0.031,0.032,0.032) (0.0 06,0.0 06,0.0 06; 0.0 05,0.0 06,0.0 06) (0.0 04,0.0 05,0.0 05; 0.0 04,0.0 05,0.0 05) (0.0 04,0.0 04,0.0 04; 0.0 04,0.0 04,0.0 05) (0.0 06,0.0 06,0.0 06; 0.0 05,0.0 06,0.0 06) (0.0 09,0.0 09,0.0 09; 0.0 09,0.0 09,0.0 09) (0.0 04,0.0 04,0.0 04; 0.0 04,0.0 04,0.0 05) (0.0 04,0.0 04,0.0 04; 0.0 04,0.0 04,0.0 05) (0.0 05,0.0 05,0.0 05; 0.0 05,0.0 05,0.0 05) (0.011,0.011,0.011; 0.011,0.011,0.011) (0.0 03,0.0 03,0.0 04; 0.0 03,0.0 03,0.0 04) (0.0 09,0.0 09,0.0 09; 0.0 09,0.0 09,0.0 09) (0.012,0.012,0.012; 0.012,0.012,0.012) (0.023,0.023,0.023; 0.023,0.023,0.023) (0.0 05,0.0 05,0.0 05; 0.0 04,0.0 05,0.0 05)

(0.28,0.281,0.282; 0.279,0.281,0.283) (0.235,0.236,0.237; 0.234,0.236,0.238) (0.054,0.055,0.057; 0.054,0.055,0.057) (0.052,0.053,0.054; 0.051,0.053,0.055) (0.046,0.048,0.049; 0.046,0.048,0.049) (0.019,0.02,0.021; 0.018,0.02,0.022) (0.056,0.057,0.058; 0.055,0.057,0.059) (0.043,0.045,0.046; 0.043,0.045,0.046) (0.048,0.049,0.05; 0.047,0.049,0.051) (0.033,0.034,0.036; 0.033,0.034,0.036) (0.046,0.047,0.048; 0.045,0.047,0.049) (0.035,0.036,0.037; 0.034,0.036,0.038) (0.061,0.062,0.064; 0.061,0.062,0.064) (0.085,0.086,0.088; 0.085,0.086,0.088) (0.106,0.107,0.108; 0.105,0.107,0.109) (0.039,0.04,0.041; 0.038,0.04,0.042)

(0.162,0.163,0.164; 0.161,0.163,0.165) (0.15,0.151,0.152; 0.149,0.151,0.153) (0.035,0.036,0.037; 0.034,0.036,0.039) (0.038,0.04,0.041; 0.037,0.04,0.042) (0.046,0.047,0.048; 0.045,0.047,0.05) (0.019,0.02,0.021; 0.017,0.02,0.022) (0.064,0.065,0.066; 0.063,0.065,0.067) (0.034,0.035,0.036; 0.033,0.035,0.037) (0.046,0.047,0.048; 0.045,0.047,0.049) (0.028,0.029,0.03; 0.027,0.029,0.032) (0.039,0.04,0.041; 0.038,0.04,0.042) (0.028,0.029,0.03; 0.027,0.029,0.031) (0.047,0.048,0.049; 0.046,0.048,0.051) (0.088,0.089,0.09; 0.087,0.089,0.091) (0.083,0.084,0.085; 0.082,0.084,0.086) (0.029,0.03,0.031; 0.028,0.03,0.032)

(0.044,0.044,0.045; 0.044,0.044,0.045) (0.041,0.041,0.042; 0.041,0.041,0.042) (0.02,0.02,0.021; 0.02,0.02,0.021) (0.025,0.025,0.025; 0.024,0.025,0.025) (0.029,0.029,0.029; 0.028,0.029,0.029) (0.0 06,0.0 07,0.0 07; 0.0 06,0.0 07,0.0 07) (0.013,0.014,0.014; 0.013,0.014,0.014) (0.016,0.016,0.016; 0.015,0.016,0.016) (0.019,0.02,0.02; 0.019,0.02,0.02) (0.01,0.011,0.011; 0.01,0.011,0.011) (0.011,0.011,0.012; 0.011,0.011,0.012) (0.0 08,0.0 08,0.0 08; 0.0 08,0.0 08,0.0 09) (0.017,0.017,0.017; 0.016,0.017,0.017) (0.023,0.023,0.023; 0.022,0.023,0.023) (0.024,0.024,0.025; 0.024,0.024,0.025) (0.0 07,0.0 08,0.0 08; 0.0 07,0.0 08,0.0 08)

(0.233,0.252,0.271; 0.058,0.089,0.116) (0.226,0.245,0.265; 0.065,0.096,0.124) (0.233,0.252,0.271; 0.058,0.089,0.116) (0.252,0.271,0.291; 0.039,0.07,0.097) (0.241,0.261,0.291; 0.049,0.081,0.108) (0.118,0.145,0.171; 0.173,0.199,0.223) (0.245,0.265,0.291; 0.046,0.077,0.104) (0.22,0.239,0.259; 0.071,0.102,0.13) (0.117,0.144,0.17; 0.174,0.201,0.224) (0.223,0.245,0.291; 0.067,0.1,0.128) (0.122,0.15,0.175; 0.168,0.194,0.218) (0.076,0.107,0.135; 0.215,0.234,0.253) (0.213,0.233,0.254; 0.078,0.109,0.137) (0.117,0.144,0.17; 0.174,0.201,0.224) (0.122,0.15,0.175; 0.168,0.194,0.218) (0.123,0.152,0.18; 0.167,0.196,0.22)

(0.088,0.095,0.103; 0.025,0.037,0.047) (0.091,0.099,0.112; 0.022,0.034,0.045) (0.082,0.09,0.098; 0.03,0.042,0.053) (0.095,0.103,0.112; 0.018,0.03,0.04) (0.086,0.093,0.101; 0.027,0.039,0.049) (0.068,0.077,0.086; 0.044,0.056,0.067) (0.088,0.097,0.112; 0.024,0.037,0.048) (0.092,0.1,0.112; 0.021,0.033,0.044) (0.052,0.063,0.074; 0.06,0.071,0.08) (0.088,0.096,0.112; 0.024,0.037,0.048) (0.055,0.066,0.079; 0.058,0.07,0.08) (0.031,0.043,0.054; 0.081,0.089,0.096) (0.09,0.097,0.105; 0.022,0.034,0.045) (0.052,0.063,0.074; 0.06,0.071,0.08) (0.052,0.063,0.074; 0.06,0.071,0.08) (0.063,0.076,0.112; 0.049,0.063,0.073)

H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16

inpatient department admissions, conformance to quality procedures and bed usage rate, are evaluated. All inputs and outputs are expressed by using Triangular Intuitionistic Fuzzy Sets. Brief explanations of both inputs and outputs are stated in Table 4. 6.1. Weighting inputs & outputs using TIF AHP In this section, the judgments for inputs and outputs are collected from three experts as linguistic evaluations. Tables 5 and 6 exhibit these pairwise comparisons of inputs and outputs based on the linguistic scale in Table 2. Table 7 demonstrates the weights of inputs and outputs obtained from the application of TIF AHP method. 6.2. Measuring efficiencies of hospitals The next operation is to measure efficiencies of the hospitals by using CRS based TIF DEA. Table 8 displays evaluations of three experts for DEA inputs in each cell based on the linguistic scale in Table 3. Table 9 illustrates the normalized aggregated values of these input evaluations by employing Eqs. (13) and (26) while Table 10 presents the weighted aggregated evaluations of inputs as in Eq. (28). Similarly, evaluations of outputs are collected from the decision makers as in Table 11. Once the evaluations are normalized and the experts’ judgments are aggregated, the weighted aggregated evaluations of experts for DEA outputs are obtained as shown in Table 12. Since the hospital performance evaluation problem is too complex and comprehensive to solve using the model given in Eq. (29) in Step 2.6, we preferred defuzzifying the fuzzy scores in Tables 10 and 12. In Table 13, the defuzzified scores of inputs and

outputs are calculated where τ is set to 10 0 0 in Eq. (12). The value of τ is set to a large number since the experts want to significantly decrease the effect of non-membership function. Using the defuzzified values in Table 13, the DEA model has been applied. The solution of the proposed approach is presented in Table 14. According to these results, H3, H5, H6, H8, H9, H10, H12, H13, H14, H15, and H16 are the efficient frontiers. The rest of the hospitals are ranked as H11 > H1 > H4 > H2 > H7. The results reveal that 27% of the efficient frontiers are public hospitals and 60% of the inefficient hospitals are private hospitals. Slacks exist only for those hospitals identified as inefficient. According to Table 14, for H1, the levels of number of beds (I2), number of physicians (I3), number of nurses (I4), number of other personnel (I5), and technology level (I6) should be reduced by 9.14, 160.2, 67.6, 16.1, and 36.3, respectively. However, despite the reduction in these inputs, it would not achieve efficiency. H1 should also augment annual revenue (O1), Number of outpatient department visits (O2), overall patient satisfaction (O3), conformance to quality procedures (O5), and bed usage rate (O6) by 48.6, 4.89, 67.75, 30.01, and 9.7 respectively. Similar situations in different magnitudes exist for H2, H4, H7, and H11. 6.3. Comparison & discussion To examine the validity of the obtained results through TIF AHP & DEA, the results of the proposed integrated fuzzy method are compared with the results of traditional DEA. Table 15 gives the finding from both approaches. According to Table 15, the efficiencies of hospitals are found higher in traditional DEA compared to proposed TIF AHP & DEA. It is interesting that H11 is efficient with traditional DEA model whereas it is inefficient with a score of 0.974 using TIF AHP & DEA. It reveals that TIF AHP & DEA presents

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

103

Table 11 Evaluations of experts for DEA outputs. Hospital

(80,88,96;75,88,101) (76,88,100;70,88,106) (77,89,101;71,89,107) (82,90,98;77,90,103) (81,93,105;75,93,111) (82,94,106;76,94,112) (127,135,143;122,135,148) (120,132,144;114,132,150) (126,138,150;120,138,156) (107,115,123;102,115,128) (113,125,137;107,125,143) (113,125,137;107,125,143) (142,150,158;137,150,163) (141,153,165;135,153,171) (142,154,166;136,154,172) (100,108,116;95,108,121) (99,111,123;93,111,129) (106,118,130;100,118,136) (54,62,70;49,62,75) (59,71,83;53,71,89) (43,55,67;37,55,73) (86,94,102;81,94,107) (79,91,103;73,91,109) (78,90,102;72,90,108) (97,105,113;92,105,118) (103,115,127;97,115,133) (98,110,122;92,110,128) (117,125,133;112,125,138) (106,118,130;100,118,136) (117,129,141;111,129,147) (67,75,83;62,75,88) (65,77,89;59,77,95) (65,77,89;59,77,95) (62,70,78;57,70,83) (58,70,82;52,70,88) (58,70,82;52,70,88) (72,80,88;67,80,93) (65,77,89;59,77,95) (72,84,96;66,84,102) (62,70,78;57,70,83) (60,72,84;54,72,90) (63,75,87;57,75,93) (69,77,85;64,77,90) (66,78,90;60,78,96) (69,81,93;63,81,99) (47,55,63;42,55,68) (45,57,69;39,57,75) (50,62,74;44,62,80)

(965,975,985;967,975,983) (968,978,988;963,978,993) (972,980,988;968,980,992) (875,885,895;877,885,893) (877,887,897;872,887,902) (881,889,897;877,889,901) (670,680,690;672,680,688) (664,674,684;659,674,689) (666,674,682;662,674,686) (270,280,290;272,280,288) (275,285,295;270,285,300) (267,275,283;263,275,287) (490,500,510;492,500,508) (495,505,515;490,505,520) (488,496,504;484,496,508) (440,450,460;442,450,458) (442,452,462;437,452,467) (444,452,460;440,452,464) (290,30 0,310;292,30 0,308) (287,297,307;282,297,312) (304,312,320;300,312,324) (500,510,520;502,510,518) (499,509,519;494,509,524) (504,512,520;500,512,524) (630,640,650;632,640,648) (632,642,652;627,642,657) (639,647,655;635,647,659) (354,364,374;356,364,372) (357,367,377;352,367,382) (352,360,368;348,360,372) (400,410,420;402,410,418) (402,412,422;397,412,427) (400,408,416;396,408,420) (355,365,375;357,365,373) (359,369,379;354,369,384) (353,361,369;349,361,373) (675,685,695;677,685,693) (678,688,698;673,688,703) (681,689,697;677,689,701) (810,820,830;812,820,828) (814,824,834;809,824,839) (816,824,832;812,824,836) (890,90 0,910;892,90 0,908) (887,897,907;882,897,912) (897,905,913;893,905,917) (460,470,480;462,470,478) (465,475,485;460,475,490) (460,468,476;456,468,480)

(50,55,60;25,33,40) (55,60,65;20,28,35) (55,60,65;20,28,35) (50,55,60;25,33,40) (55,60,65;20,28,35) (55,60,65;20,28,35) (65,70,75;10,18,25) (65,70,75;10,18,25) (65,70,75;10,18,25) (20,28,35;55,60,65) (15,23,30;60,65,70) (15,23,30;60,65,70) (65,70,75;10,18,25) (60,65,70;15,23,30) (65,70,75;10,18,25) (20,28,35;55,60,65) (55,60,65;20,28,35) (55,60,65;20,28,35) (25,33,40;50,55,60) (25,33,40;50,55,60) (20,28,35;55,60,65) (55,60,65;20,28,35) (60,65,70;15,23,30) (65,70,75;10,18,25) (25,33,40;50,55,60) (55,60,65;20,28,35) (60,65,70;15,23,30) (60,65,70;15,23,30) (55,60,65;20,28,35) (15,23,30;60,65,70) (25,33,40;50,55,60) (60,65,70;15,23,30) (60,65,70;15,23,30) (15,23,30;60,65,70) (10,18,25;65,70,75) (60,65,70;15,23,30) (50,55,60;25,33,40) (55,60,65;20,28,35) (55,60,65;20,28,35) (20,28,35;55,60,65) (50,55,60;25,33,40) (50,55,60;25,33,40) (50,55,60;25,33,40) (50,55,60;25,33,40) (55,60,65;20,28,35) (15,23,30;60,65,70) (50,55,60;25,33,40) (50,55,60;25,33,40)

(57,65,73;55,65,75) (56,68,80;50,68,86) (65,70,75;61,70,79) (50,58,66;48,58,68) (48,60,72;42,60,78) (55,60,65;51,60,69) (32,40,48;30,40,50) (30,42,54;24,42,60) (40,45,50;36,45,54) (30,38,46;28,38,48) (23,35,47;17,35,53) (25,30,35;21,30,39) (27,35,43;25,35,45) (19,31,43;13,31,49) (26,31,36;22,31,40) (42,50,58;40,50,60) (42,54,66;36,54,72) (49,54,59;45,54,63) (17,25,33;15,25,35) (16,28,40;10,28,46) (24,29,34;20,29,38) (37,45,53;35,45,55) (36,48,60;30,48,66) (39,44,49;35,44,53) (22,30,38;20,30,40) (21,33,45;15,33,51) (28,33,38;24,33,42) (17,25,33;15,25,35) (17,29,41;11,29,47) (17,22,27;13,22,31) (14,22,30;12,22,32) (13,25,37;7,25,43) (15,20,25;11,20,29) (10,18,26;8,18,28) (9,21,33;3,21,39) (19,24,29;15,24,33) (47,55,63;45,55,65) (35,47,59;29,47,65) (53,58,63;49,58,67) (40,48,56;38,48,58) (31,43,55;25,43,61) (35,40,45;31,40,49) (44,52,60;42,52,62) (32,44,56;26,44,62) (45,50,55;41,50,59) (27,35,43;25,35,45) (27,39,51;21,39,57) (34,39,44;30,39,48)

(60,65,70;15,23,30) (50,55,60;25,33,40) (55,60,65;20,28,35) (55,60,65;20,28,35) (50,55,60;25,33,40) (55,60,65;20,28,35) (65,70,75;10,18,25) (65,70,75;10,18,25) (65,70,75;10,18,25) (15,23,30;60,65,70) (10,18,25;65,70,75) (10,18,25;65,70,75) (65,70,75;10,18,25) (65,70,75;10,18,25) (65,70,75;10,18,25) (20,28,35;55,60,65) (65,70,75;10,18,25) (60,65,70;15,23,30) (15,23,30;60,65,70) (20,28,35;55,60,65) (15,23,30;60,65,70) (55,60,65;20,28,35) (60,65,70;15,23,30) (65,70,75;10,18,25) (10,18,25;65,70,75) (60,65,70;15,23,30) (55,60,65;20,28,35) (60,65,70;15,23,30) (60,65,70;15,23,30) (15,23,30;60,65,70) (15,23,30;60,65,70) (55,60,65;20,28,35) (60,65,70;15,23,30) (10,18,25;65,70,75) (10,18,25;65,70,75) (60,65,70;15,23,30) (20,28,35;55,60,65) (50,55,60;25,33,40) (60,65,70;15,23,30) (15,23,30;60,65,70) (50,55,60;25,33,40) (55,60,65;20,28,35) (25,33,40;50,55,60) (50,55,60;25,33,40) (55,60,65;20,28,35) (10,18,25;65,70,75) (50,55,60;25,33,40) (55,60,65;20,28,35)

(90,97,104;85,97,109) (90,98,106;83,98,113) (90,98,106;84,98,112) (90,97,104;85,97,109) (90,98,106;83,98,113) (87,95,103;81,95,109) (81,88,95;76,88,100) (79,87,95;72,87,102) (82,90,98;76,90,104) (77,84,91;72,84,96) (74,82,90;67,82,97) (77,85,93;71,85,99) (78,85,92;73,85,97) (75,83,91;68,83,98) (76,84,92;70,84,98) (76,83,90;71,83,95) (73,81,89;66,81,96) (72,80,88;66,80,94) (70,77,84;65,77,89) (72,80,88;65,80,95) (67,75,83;61,75,89) (83,90,97;78,90,102) (83,91,99;76,91,106) (86,94,102;80,94,108) (82,89,96;77,89,101) (84,92,100;77,92,107) (82,90,98;76,90,104) (73,80,87;68,80,92) (76,84,92;69,84,99) (75,83,91;69,83,97) (78,85,92;73,85,97) (78,86,94;71,86,101) (79,87,95;73,87,101) (78,85,92;73,85,97) (78,86,94;71,86,101) (77,85,93;71,85,99) (86,93,100;81,93,105) (83,91,99;76,91,106) (86,94,102;80,94,108) (88,95,102;83,95,107) (88,96,104;81,96,111) (89,97,105;83,97,111) (88,95,102;83,95,107) (89,97,105;82,97,112) (89,97,105;83,97,111) (88,95,102;83,95,107) (89,97,105;82,97,112) (89,97,105;83,97,111)

H10

H11

H12

H13

H14

H15

H16

a slightly more pessimistic approach with respect to traditional DEA model. Traditional DEA needs less computational effort and it uses deterministic information. The deterministic representation of uncertainty and vagueness decreases the reliability of the obtained solutions. Under certainty, traditional DEA and TIF AHP & DEA may produce same results but traditional DEA is preferred to TIF AHP & DEA since it has computational advantages. Under uncertainty, the accuracy of traditional DEA results is low since it cannot deal with the imprecision in the problem and TIF AHP & DEA is more efficient than traditional DEA since it can incorporate this imprecision and vagueness into the model. The healthcare performance evaluation problem we consider involves high level of uncertainty. Hence, we preferred TIF AHP & DEA rather than traditional DEA. The results indicate that the efficiency scores of H1, H2, H4, H7, and H11 are different in traditional DEA and TIF AHP & DEA. The experts confirmed that the accuracy of TIF AHP & DEA results is higher than traditional DEA.

7. Conclusion Healthcare is a very complex and one of the high priority application fields. It involves many research areas with a large number of constraints and uncertainties. Uncertainty is an inevitable component of decision making process. Linguistic evaluations are generally preferred when exact numerical values cannot be assigned in multicriteria evaluations. Linguistic variables can be best handled by the fuzzy set theory. One of the extensions of ordinary fuzzy sets is intuitionistic fuzzy sets, which includes the definition of membership, non-membership, and hesitancy altogether. The proposed triangular intuitionistic fuzzy AHP & DEA model could successfully handle the linguistic evaluations of more than one expert by aggregating and weighting them. It is observed that traditional DEA has an optimistic point of view to the same problem while TIF DEA has relatively a pessimistic point of view better revealing inefficient performances.

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

104

Table 12 Weighted aggregated evaluations of experts for DEA outputs. Hospital

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16

(0.128,0.145,0.161;0.118,0.144,0.171) (0.134,0.151,0.168;0.124,0.151,0.177) (0.205,0.221,0.238;0.195,0.221,0.247) (0.181,0.198,0.215;0.172,0.198,0.224) (0.232,0.249,0.266;0.223,0.249,0.275) (0.166,0.183,0.2;0.157,0.183,0.209) (0.086,0.103,0.12;0.075,0.102,0.128) (0.134,0.151,0.167;0.124,0.15,0.176) (0.162,0.179,0.196;0.153,0.179,0.205) (0.187,0.204,0.22;0.177,0.203,0.229) (0.108,0.125,0.142;0.098,0.125,0.151) (0.098,0.115,0.131;0.088,0.115,0.141) (0.115,0.131,0.148;0.105,0.131,0.157) (0.101,0.118,0.135;0.092,0.118,0.144) (0.112,0.128,0.145;0.102,0.128,0.154) (0.077,0.094,0.111;0.068,0.094,0.12)

(0.064,0.064,0.065;0.064,0.064,0.065) (0.058,0.058,0.059;0.058,0.058,0.059) (0.044,0.045,0.045;0.044,0.045,0.045) (0.018,0.018,0.019;0.018,0.018,0.019) (0.032,0.033,0.034;0.032,0.033,0.034) (0.029,0.03,0.03;0.029,0.03,0.03) (0.019,0.02,0.021;0.019,0.02,0.021) (0.033,0.034,0.034;0.033,0.034,0.034) (0.042,0.042,0.043;0.042,0.042,0.043) (0.023,0.024,0.025;0.023,0.024,0.025) (0.026,0.027,0.028;0.026,0.027,0.028) (0.023,0.024,0.025;0.023,0.024,0.025) (0.045,0.045,0.046;0.044,0.045,0.046) (0.054,0.054,0.055;0.053,0.054,0.055) (0.059,0.059,0.06;0.059,0.059,0.06) (0.03,0.031,0.032;0.03,0.031,0.032)

(0.219,0.24,0.261;0.09,0.123,0.152) (0.219,0.24,0.261;0.09,0.123,0.152) (0.268,0.288,0.309;0.041,0.074,0.103) (0.07,0.103,0.132;0.239,0.259,0.28) (0.262,0.284,0.309;0.047,0.08,0.109) (0.186,0.211,0.237;0.124,0.156,0.185) (0.097,0.13,0.159;0.212,0.233,0.253) (0.248,0.27,0.309;0.061,0.095,0.124) (0.2,0.226,0.254;0.109,0.142,0.171) (0.207,0.233,0.26;0.102,0.137,0.167) (0.209,0.236,0.264;0.1,0.134,0.163) (0.142,0.175,0.21;0.167,0.201,0.228) (0.219,0.24,0.261;0.09,0.123,0.152) (0.168,0.193,0.218;0.141,0.173,0.2) (0.213,0.233,0.254;0.096,0.129,0.158) (0.163,0.188,0.213;0.146,0.178,0.206)

Hospital

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16

(0.085,0.096,0.108;0.078,0.096,0.113) (0.073,0.084,0.096;0.067,0.084,0.101) (0.049,0.06,0.072;0.042,0.06,0.077) (0.038,0.05,0.062;0.031,0.049,0.066) (0.035,0.047,0.058;0.028,0.046,0.063) (0.063,0.075,0.086;0.057,0.074,0.091) (0.027,0.038,0.05;0.021,0.038,0.055) (0.053,0.065,0.077;0.047,0.065,0.082) (0.034,0.045,0.057;0.028,0.045,0.062) (0.024,0.036,0.048;0.019,0.036,0.052) (0.02,0.032,0.043;0.014,0.032,0.048) (0.018,0.029,0.041;0.01,0.029,0.046) (0.066,0.078,0.089;0.057,0.076,0.093) (0.052,0.064,0.076;0.045,0.062,0.08) (0.059,0.071,0.083;0.051,0.069,0.087) (0.042,0.053,0.065;0.036,0.053,0.07)

(0.148,0.162,0.176;0.05,0.072,0.091) (0.142,0.155,0.169;0.057,0.078,0.096) (0.172,0.185,0.199;0.026,0.048,0.066) (0.032,0.053,0.072;0.167,0.18,0.193) (0.172,0.185,0.199;0.026,0.048,0.066) (0.139,0.159,0.199;0.059,0.083,0.102) (0.044,0.065,0.084;0.155,0.168,0.181) (0.159,0.173,0.199;0.039,0.061,0.08) (0.121,0.139,0.158;0.078,0.101,0.12) (0.138,0.155,0.173;0.06,0.083,0.102) (0.123,0.141,0.159;0.075,0.098,0.117) (0.088,0.109,0.132;0.111,0.133,0.151) (0.121,0.138,0.156;0.078,0.1,0.118) (0.111,0.127,0.144;0.088,0.109,0.127) (0.117,0.133,0.149;0.082,0.102,0.12) (0.108,0.125,0.142;0.091,0.112,0.131)

(0.032,0.035,0.038;0.03,0.035,0.04) (0.032,0.035,0.037;0.03,0.035,0.039) (0.029,0.032,0.034;0.027,0.032,0.036) (0.027,0.03,0.033;0.025,0.03,0.035) (0.027,0.03,0.033;0.025,0.03,0.035) (0.026,0.029,0.032;0.024,0.029,0.034) (0.025,0.028,0.03;0.023,0.028,0.032) (0.03,0.033,0.035;0.028,0.033,0.037) (0.029,0.032,0.035;0.027,0.032,0.037) (0.027,0.029,0.032;0.024,0.029,0.034) (0.028,0.031,0.033;0.026,0.031,0.035) (0.028,0.03,0.033;0.026,0.03,0.035) (0.03,0.033,0.036;0.028,0.033,0.038) (0.032,0.034,0.037;0.029,0.034,0.039) (0.032,0.034,0.037;0.03,0.034,0.039) (0.032,0.034,0.037;0.03,0.034,0.039)

Table 13 DEA inputs & outputs. Hospital

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16

INPUTS

OUTPUTS

48 44 23 24 20 20 63 17 21 19 24 15 25 33 39 33

36 32 6 5 4 6 9 4 4 5 11 4 9 13 24 5

290 243 57 54 49 21 59 46 50 35 49 37 64 89 110 41

168 155 37 41 49 20 67 36 49 30 41 30 50 92 86 31

45 42 21 26 30 7 14 16 20 11 12 8 17 24 25 8

255 248 255 273 267 151 269 242 149 256 155 113 237 149 155 158

97 102 91 104 94 79 100 102 65 100 69 45 98 65 65 86

149 155 228 204 257 189 106 155 185 210 128 118 135 121 132 97

66 60 46 19 34 31 21 35 44 25 28 25 47 56 61 32

243 243 291 110 287 216 136 278 231 237 240 182 243 198 237 193

99 87 62 51 48 77 40 67 47 37 33 30 80 65 73 55

164 158 187 58 187 168 69 179 142 158 144 114 141 131 136 128

36 36 33 31 31 30 28 34 33 30 32 31 34 35 35 35

Even DEA is an effective optimization technique to evaluate the performance of hospitals, it has some limitations which need to be underlined. When the number of inputs and outputs are relatively large and the number of alternative hospitals is low, discriminatory power of the DEA is limited. For further research, we suggest other extensions of ordinary fuzzy sets such as hesitant fuzzy sets, fuzzy multisets, or nonstationary fuzzy sets to be used for the solution of the same prob-

lem. Additionally, the recently proposed sets such as neutrosophic sets and Pythagorean fuzzy sets can be used to deal with uncertainty with a different perspective. Besides, Electronic Health Record (EHR) systems, which keep patients’ historical information electronically, can provide more reliable data for the proposed method to be used in the performance evaluation of the hospitals [61,62].

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

105

Table 14 DEA results, eﬃciencies, input slacks, and output slacks. Hospital

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16

Eﬃciency

0.954 0.847 1.0 0 0 0.946 1.0 0 0 1.0 0 0 0.515 1.0 0 0 1.0 0 0 1.0 0 0 0.974 1.0 0 0 1.0 0 0 1.0 0 0 1.0 0 0 1.0 0 0

INPUTS

OUTPUTS

0.00 0.00 0.00 4.24 0.00 0.00 15.93 0.00 0.00 0.00 3.07 0.00 0.00 0.00 0.00 0.00

9.14 12.6 0.00 0.00 0.00 0.00 0.97 0.00 0.00 0.00 5.93 0.00 0.00 0.00 0.00 0.00

160.2 119.2 0.00 9.98 0.00 0.00 0.05 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

67.6 46.9 0.00 0.00 0.00 0.00 9.56 0.00 0.00 0.00 1.96 0.00 0.00 0.00 0.00 0.00

16.1 12.9 0.00 4.08 0.00 0.00 0.00 0.00 0.00 0.00 0.50 0.00 0.00 0.00 0.00 0.00

36.3 23.4 0.00 44.48 0.00 0.00 24.38 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 15.56 0.00 0.00 0.00 0.00 0.00 0.00 5.22 0.00 0.00 0.00 0.00 0.00

48.6 23.9 0.00 0.00 0.00 0.00 14.84 0.00 0.00 0.00 31.3 0.00 0.00 0.00 0.00 0.00

4.89 0.00 0.00 12.39 0.00 0.00 3.76 0.00 0.00 0.00 5.05 0.00 0.00 0.00 0.00 0.00

67.75 16.12 0.00 142.30 0.00 0.00 37.30 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 10.44 0.00 0.00 0.00 0.00 0.00

30.01 20.21 0.00 109.40 0.00 0.00 46.84 0.00 0.00 0.00 8.78 0.00 0.00 0.00 0.00 0.00

9.7 6.1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 9.41 0.00 0.00 0.00 0.00 0.00

Table 15 Comparison of TIF AHP & DEA and traditional DEA. Hospital

TIF AHP & DEA Results

Traditional DEA results

Difference

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16

0.954 0.848 1 0.946 1 1 0.515 1 1 1 0.974 1 1 1 1 1

0.921 0.902 1 0.971 1 1 0.555 1 1 1 1 1 1 1 1 1

−0.034 0.054 0 0.026 0 0 0.040 0 0 0 0.026 0 0 0 0 0

References [1] D. Tengilimog˘ lu, S¸ .A. Toygar, PATH Method in the measurement of hospital performance, J. Soc. Sc. (2013) 50–78. [2] M. Azadi, R.F. Saen, A combination of QFD and imprecise DEA with enhanced Russell graph measure: a case study in healthcare, Socio. Econ. Plan. Sci. 47 (4) (2013) 281–291. [3] M.A. Karadayi, E.E. Karsak, Imprecise DEA framework for evaluating the efficiency of state hospitals in Istanbul, in: Proceedings of the World Congress on Engineering 2014, London, U.K., II, 2014 WCE 2014, July 2-4. [4] D. Retzlaff-Roberts, C.F. Chang, R.M. Rubin, Technical efficiency in the use of health care resources: a comparison of OECD countries, Health Policy 69 (1) (2004) 55–72. [5] A. Al-Refaie, R.H. Fouad, M.-H. Li, M. Shurrab, Applying simulation and DEA to improve performance of emergency department in a Jordanian hospital, Simul. Model. Pract. Theory 41 (2014) 59–72. [6] Y. Ayanog˘ lu, M. Atan, U. Beylik, Assessments and measurement of financial performance using the methodology of Data Envelopment Analysis (DEA) in hospitals, Sag˘ lıkta Performans ve Kalite Dergisi 2 (2010) 40–62. [7] J.A. Delgado-Osuna, M. Lozano, C. García-Martínez, An alternative artificial bee colony algorithm with destructive–constructive neighbourhood operator for the problem of composing medical crews, Inf. Sci. 326 (2016) 215–226. [8] G. Büyüközkan, O. Feyziog˘ lu, F. Gocer, Evaluation of hospital web services using intuitionistic fuzzy AHP and intuitionistic fuzzy VIKOR, in: IEEE International Conference on Industrial Engineering and Engineering ManagementVolume 2016-December, 2016, pp. 607–611. 27 December 2016, Article number 7797947. [9] L.A. Zadeh, Fuzzy Sets, Inf. Control 8 (3) (1965) 338–353. [10] J. Ye, Multicriteria fuzzy decision-making method using entropy weights-based correlation coefficients of interval-valued intuitionistic fuzzy sets, Appl. Math. Model. 34 (12) (2010) 3864–3870. [11] T.L. Saaty, The Analytic Hierarchy Process, McGraw-Hill, New York, 1980. [12] J.J. Buckley, Fuzzy hierarchical analysis, Fuzzy Sets Syst. 34 (1985) 187–195. [13] S. Kubler, J. Robert, W. Derigent, A. Voisin, Y. Le Traon, A state-of the-art survey; testbed of fuzzy AHP (FAHP) applications, Expert Syst. Appl. 65 (2016) 398–422.

[14] A. Charnes, W.W. Cooper, E. Rhodes, Measuring the efficiency of decision making units, Eur. J. Oper. Res. 2 (1978) 429–444. [15] L. O’Neill, M. Rauner, K. Heidenberger, M. Kraus, A cross-national comparison and taxonomy of DEA-based hospital efficiency studies, Socio. Econ. Plann. Sci. 42 (3) (2008) 158–189. [16] A.S. Kazley, Y.A. Ozcan, Electronic medical record use and efficiency: a DEA and windows analysis of hospitals, Socio. Econ. Plann. Sci. 43 (2009) 209–216. [17] M.-C. Lai, H.-C. Huang, W.-K. Wang, Designing a knowledge-based system for benchmarking: a DEA approach, Knowl. Based Syst. 24 (5) (2011) 662–671. [18] P. Mitropoulos, M.A. Talias, I. Mitropoulos, Combining stochastic DEA with Bayesian analysis to obtain statistical properties of the efficiency scores: an application to Greek public hospitals, Eur. J. Oper. Res. 243 (1) (2015) 302–311. [19] H. Leleu, J. Moises, V.G. Valdmanis, How do payer mix and technical inefficiency affect hospital profit? A weighted DEA approach, Oper. Res. Health Care 3 (2014) 231–237. [20] L. Fang, H. Li, Multi-criteria decision analysis for efficient location- allocation problem combining DEA and goal programming, Rairo–Oper. Res. 49 (4) (2015), doi:10.1051/ro/2015003. [21] P. Mitropoulos, I. Mitropoulos, I. Giannikos, Combining DEA with location analysis for the effective consolidation of services in the health sector, Comput. Oper. Res. 40 (9) (2013) 2241–2250. [22] H. Chowdhury, V. Zelenyuk, Performance of hospital services in Ontario: DEA with truncated regression approach, Omega 63 (2016) 111–122. [23] M.K. Bahadori, K. Abolghasemi, E. Teymourzadeh, Performance evaluation and ranking of selective wards in a military hospital using DEA and promethee method, J. Mil. Med. 18 (4) (2017) 325–334. [24] J. Khushalani, Y.A. Ozcan, Are hospitals producing quality care efficiently? An analysis using Dynamic Network Data Envelopment Analysis (DEA), Socio Econ. Plann. Sci. (2017). [25] P. Ouellette, V. Vierstraete, Technological change and efficiency in the presence of quasi-fixed inputs: a DEA application to the hospital sector, Eur. J. Oper. Res. 154 (3) (2004) 755–763. [26] C.-K. Wei, L.-C. Chen, R.-K. Li, C.-H. Tsai, Exploration of efficiency underestimation of CCR model: based on medical sectors with DEA-R model, Expert Syst. Appl. 38 (4) (2011) 3155–3160. [27] A. Ebrahimnejad, Cost efficiency measures with trapezoidal fuzzy numbers in data envelopment analysis based on ranking functions: application in insurance organization and hospital, Int. J. Fuzzy Syst. Appl. 2 (3) (2012) 51–68. [28] N. Costantino, M. Dotoli, N. Epicoco, M. Falagario, F. Sciancalepore, Using cross– efficiency fuzzy Data Envelopment Analysis for healthcare facilities performance evaluation under uncertainty, in: IEEE International Conference on Systems, Man, and Cybernetics, 2013, pp. 912–917. [29] T.-H. Chang, Fuzzy VIKOR method: a case study of the hospital service evaluation in Taiwan, Inf. Sci. 271 (2014) 196–212. [30] S. Khodaparasti, H.R. Maleki, A new combined dynamic location model for emergency medical services in fuzzy environment, in: 13th Iranian Conference on Fuzzy Systems, 2013, pp. 1–6. [31] H. Akdag, T. Kalaycı, S. Karagöz, H. Zülfikar, D. Giz, The evaluation of hospital service quality by fuzzy MCDM, Appl. Soft Comput. 23 (2014) 239–248. [32] C. Muriana, T. Piazza, G. Vizzini, An expert system for financial performance assessment of health care structures based on fuzzy sets and KPIs, Knowl. Based Syst. 97 (2016) 1–10. [33] A. Arya, S.P. Yadav, et al., A Fuzzy Dual SBM Model with fuzzy weights: an application to the health sector, in: K. Deep, et al. (Eds.), Proceedings of Sixth International Conference on Soft Computing for Problem Solving. Advances in Intelligent Systems and Computing, 546, 2017, p. 66. [34] A. Ameryoun, S. Najafi, B. Nejati-Zarnaqi, S. Khalilifar, M. Ajam, A. Ansarimoghadam, Factor selection for service quality evaluation: a hospital case study, Int. J. Health Care Qual. Assur. 30 (1) (2017) 58–66.

106

I.˙ Otay et al. / Knowledge-Based Systems 133 (2017) 90–106

[35] I. Lindlbauer, J. Schreyögg, V. Winter, Changes in technical efficiency after quality management certification: a DEA approach using difference-in-difference estimation with genetic matching in the hospital industry, Eur. J. Oper. Res. 250 (3) (2016) 1026–1036. [36] G. Büyüközkan, O. Feyziog˘ lu, F. Gocer, Evaluation of hospital web services using intuitionistic fuzzy AHP and intuitionistic fuzzy VIKOR, in: IEEE International Conference on Industrial Engineering and Engineering Management, 2016, pp. 607–611. [37] S.-X. Shi, Performance evaluation of urban ecological environment construction with interval-valued intuitionistic fuzzy information, J. Intell. Fuzzy Syst. 32 (1) (2017) 1119–1127. [38] K. Atanassov, Intuitionistic fuzzy sets, Fuzzy Set Syst. 20 (1986) 87–96. [39] X. Zhang, P. Liu, Method for aggregating triangular fuzzy intuitionistic fuzzy information and its application to decision making, Technol. Econ. Dev. Econ. 16 (2) (2010) 280–290. [40] Y. Liu, J. Wu, C. Liang, Attitudinal ranking and correlated aggregating methods for multiple attribute group decision making with triangular intuitionistic fuzzy Choquet integral, Kybernetes 44 (10) (2015) 1437–1454. [41] C. Kahraman, E. Tolga, Z. Ulukan, Selection among advanced manufacturing technologies using fuzzy data envelopment analysis, Tatra Mountains Mathematical Publications 16 (2) (1999) 311–324. [42] Qin R., Liu Y., Liu Z., Wang G. (2009). Modeling Fuzzy DEA with Type-2 Fuzzy Variable Coefficients. In: Yu W., He H., Zhang N. (eds) Advances in Neural Networks - ISNN 2009. ISNN 2009. Lecture Notes in Computer Science, vol 5552. Springer, Berlin, Heidelberg. [43] H. Yan, Q. Wei, Data envelopment analysis classification machine, Inf. Sci. 181 (22) (2011) 5029–5041. [44] M.Z. Angiz, L.A. Emrouznejad, A. Mustafa, A.S. Al-Eraqi, Aggregating preference ranking with fuzzy Data Envelopment Analysis, Knowl. Based Syst. 23 (6) (2010) 512–519. [45] K. Ersoy, S. Kavuncubasi, Y.A. Ozcan, J.M. Harris II, Technical efficiencies of Turkish hospitals: DEA approach, J. Med. Syst. 21 (2) (1997) 67–74. [46] R.D. Banker, A. Charnes, W.W. Cooper, Models for the estimation of technical and scale inefficiencies in Data Envelopment Analysis, Manage. Sci. 30 (1984) 1078–1092. [47] V.V. Podinovski, Bridging the Gap between the constant and variable returns-to-scale models: selective proportionality in Data Envelopment Analysis, J. Oper. Res. Soc. 55 (3) (2004) 265–276. [48] S.C. Ray, Data Envelopment Analysis Theory and Techniques for Economics and Operations Research, Cambridge University Press, New York, USA, 2004. [49] P. Guo, H. Tanaka, Fuzzy DEA: a perceptual evaluation method, Fuzzy Set Syst. 119 (1) (2001) 149–160.

[50] T.N. Chirikos, A.M. Sear, Measuring hospital efficiency: a comparison of two approaches, Health Serv. Res. 34 (6) (20 0 0) 1389–1408. [51] J.C. Lo, K.S. Shih, K.L. Chen, Technical efficiency of the general hospitals in Taiwan: an application of DEA, Acad. Econ. Pap. 24 (3) (1996) 275–296. [52] S. Jiang, W-M. Wu, P. Fang, Evaluating the effectiveness of public hospital reform from the perspective of efficiency and quality in Guangxi, China, Springerplus 5 (1) (2016), doi:10.1186/s40064- 016- 3598- y. [53] M.S. Gok, B. Sezen, Analyzing the efficiencies of hospitals: an application of Data Envelopment Analysis, J. Global Strateg. Manage. 5 (2) (2011) 137–146. [54] J. Harris II, H. Ozgen, Y. Ozcan, Do mergers enhance the performance of hospital efficiency? J. Oper. Res. Soc. 51 (20 0 0) 801–811. [55] R. Hod, O. Maimon, E. Zimlichman, The relationships between hospital service quality and operational efficiency: an alternative model, Health Econ. Outcome Res. Open Access 2 (2016) 119. [56] J. Benneyan, M.E. Ceyhan, Data Envelopment Analysis of national healthcare systems and their relative efficiencies, in: M.H. Elwany, A.B. Eltawil (Eds.), Proceedings of the 37th International Conference on Computers and Industrial Engineering, Alexandria, Egypt, 2007 October 20-23. [57] T.R. Jat, M.S. Sebastian, Technical efficiency of public district hospitals in Madhya Pradesh, India: a Data Envelopment Analysis, Glob Health Action 6 (2013), doi:10.3402/gha.v6i0.21742. [58] R.C. Morey, D.J. Fine, S.W. Loree, D.L. Retzlaff-Roberts, S. Tsubakitani, The trade-off between hospital cost and quality of care. an exploratory empirical analysis, Med. Care 30 (1992) 677–698. [59] S. Papadaki, P. Stankova, P. Klímek, The effectiveness of horizontally integrated hospitals in the Czech Republic, Int. J. Math. Models Methods Appl. Sci. 10 (2016) 110–119. [60] S. Papadaki, S. Pavla, P. Stankova, Comparison of horizontally integrated hospitals in private and public sectors of Czech Republic, Econ. Sociol. 9 (3) (2016) 180–194. [61] V.C. Kaggal, R.K. Elayavilli, S. Mehrabi, J.J. Pankratz, S. Sohn, Y. Wang, D. Li, M.M. Rastegar, S.P. Murphy, J.L. Ross, R. Chaudhry, J.D. Buntrock, H. Liu, Toward a learning health-care system-knowledge delivery at the point of care empowered by big data and NLP, Biomed. Inform. Insights 8 (Suppl 1) (2016) 13. [62] M.A. Ellsworth, M. Dziadzko, J.C. O’Horo, A.M. Farrell, J. Zhang, V. Herasevich, An appraisal of published usability evaluations of electronic health records via systematic review, J. Am. Med. Inform. Assoc. 24 (1) (2017) 218–226.