Sie sind auf Seite 1von 14

Simulation Modelling Practice and Theory 61 (2016) 1427

Contents lists available at ScienceDirect

Simulation Modelling Practice and Theory


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

Patient referral mechanisms by using simulation optimization


Ping-Shun Chen a, Kang-Hung Yang a,, Rex Aurelius C. Robielos b, Rozel Aireen C. Cancino b,
Lea Angelica M. Dizon b
a
b

Department of Industrial and Systems Engineering, Chung Yuan Christian University, Chung Li District, Taoyuan City 320, Taiwan, ROC
School of Industrial Engineering and Engineering Management, Mapa Institute of Technology, Intramuros, Manila 1002, Philippines

a r t i c l e

i n f o

Article history:
Received 17 August 2015
Received in revised form 13 November 2015
Accepted 24 November 2015

Keywords:
Magnetic resonance imaging
Collaboration
Patient-referring mechanism
Simulation optimization

a b s t r a c t
Building patient referral mechanisms between two hospitals is the focus of this research,
which considers different kinds of magnetic resonance imaging (MRI) services offered by
the case hospitals. The simulation optimization approach is the main tool for analysis,
along with the formulation of a mathematical model and a simulation framework to conceptualize a collaborative MRI patient-referring mechanism. The objective of the study is to
obtain the best feasible number of referral outpatients to minimize patients average
waiting time or to maximize the revenues of both case hospitals. The results can serve
as guidelines for hospital collaboration.
2015 Elsevier B.V. All rights reserved.

1. Introduction
According to [1], collaborations are essential among researchers and practitioners, policymakers, business leaders, and
community advocates. It can also be perceived as significant in the medical field, whose professionals are concerned with
the health of patients who require reliable and fast service [2]. Hospital collaboration has been studied for decades and
has become a critical topic for hospital management. Collaborative activities among allied hospitals achieve objectives that
improve healthcare quality and enhance patient satisfaction [35]. For example, sharing medical resources (physicians, beds,
or machines) or reengineering the hospital processes might reduce patients cycle time and waiting time [6].
The National Health Insurance Administration (NHIA) is concerned with the health insurance of Taiwans citizens. The
major factors affecting the organizations annual losses and earnings come from the premiums, insurance payments, bad
debts, and other incomes, such as those coming from the public welfare. In 1995, NHIA implemented the National Health
Insurance System (NHIS), whereby payments are transferred to the global budget. The global budget system controls the
total NHIS payments in order to remain on budget. Each hospital needs to negotiate annually with NHIA for the annual total
budget and insurance payment for each kind of treatment. Therefore, hospitals can calculate their annual allowable quota. If
hospitals treat more patients than their annual allowable quota, the hospital still obtains the fixed annual total budget from
NHIA for its insurance payment. Yet the global budget system created potential inefficiencies for hospital operations and
alleviated hospitals willingness (such as unwillingness to purchase new medical equipment or to increase the staffs work
hours) to promote quick services for patients. Consequently, patients in overcrowded hospitals wait longer time for their
treatment. However, some hospitals with fewer patients have abundant medical resources, leading to the medical resources
standing idle or being unused and to inefficiencies for hospital operations. From the viewpoint of hospital management, a

Corresponding author. Tel.: +886 3265 4428; fax: +886 3265 4499.
E-mail address: kanghungyang@cycu.edu.tw (K.-H. Yang).
http://dx.doi.org/10.1016/j.simpat.2015.11.004
1569-190X/ 2015 Elsevier B.V. All rights reserved.

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

15

hospital must promote its service to survive and utilize its resources in an efficient way. In order to increase competitive
advantages under the global budget system, collaboration can be used to give hospitals competitive capacities and service
quality.
The annual budget of the NHIS has been in a deficit since 1995. In order to make up the deficit, NHIA increased the insurance premiums. The original insurance premiums of the NHIS remained at 4.25% for the insured people from 1995/03 to
2002/08. NHIA adjusted the insurance premiums rate to 4.55% in 2002/09, to 5.17% in 2010/04, and to 4.91% in 2013/01.
Finally, the accumulated financial report of the NHIS was changed from budget deficits to budget surpluses in 2012/02.
Although the global budget increased about 2% to 6% annually between 2009 and 2015, hospitals are still under strong financial pressure to obtain more insurance payments from the NHIS. Hospital managers therefore needed to improve their operational effectiveness and efficiency, such as collaboration with adjacent hospitals, to prevent hospitals resources from not
being used.
Taiwan has seen increasing demand for healthcare treatment [6]. However, a systematic and efficient collaboration
framework of hospitals is now being established. The referral system has been successful in hospital collaborations in
Taiwan, but needs improvement. The primary objective of the study is to improve Taiwans healthcare system using the simulation optimization method for the mutual benefit of patients and hospitals.
In general, patients are quite concerned with access to and the reputation of their hospital. With this kind of setup,
patients will visit only the best-known hospitals in their vicinity. This mentality creates long waits at prominent hospitals.
Conversely, short waiting lines are found in suburban hospitals. Hence, promoting collaboration among hospitals through
referrals will reduce the long queues and waiting times for patients. At the same time, machines located at each hospital will
not stand idle. This study determines the most feasible numbers of fixed and unfixed referrals of magnetic resonance imaging (MRI) patients per day between the two case hospitals. The objectives are to minimize eligible patients average waiting
time or maximize hospitals revenues.
The remainder of this article is organized as follows. Section 2 reviews the literature on simulation and simulation optimization in healthcare applications. Section 3 introduces the background of the study hospital. Section 4 describes the
details of the simulation model and related parameters. In Section 5, two optimization mathematical models are constructed.
The simulation optimization procedures are then developed to solve the proposed model. Section 6 describes two scenarios
to compare the fixed and unfixed patient referral mechanisms. Two sensitivity analyses are also performed, and the results
and discussions of the patient referral mechanisms are summarized. Finally, Section 7 concludes the paper and suggests
future directions for research.

2. Literature review
Kobayashi et al. [7] built an electronic geographical information system (GIS) in order to improve the ability of clinics and
hospitals to collaborate. They concluded that some information from paper maps used as a guide to search for physicians
might be outdated. Consequently, the researchers used GIS with the use of Google Map API to refer patients to other clinics
in a fast and informative manner.
With the same goal, Ramis et al. [8] explored Medical Imaging Centers (MIC). The researchers used a simulator implemented in Flexsim GP to shorten waiting times. With the aid of four hospitals in Chile, the researchers identified all possible
flows, resources, schedules, and exams. After comparing seven setups, they were able to decrease patients total waiting time
by 35% without changing personnel, but allocating common functions. Consequently, the productivity of a center can be
increased by 54%, supposing an infinite demand.
Ahmed and Alkhamis [9] used the same approach, but with the incorporation of optimization. The researchers integrated
simulation with optimization to produce a decision support system that would help decision makers in the operation of an
emergency department unit located in Kuwait. In this paper, budget constraints were observed in the determination of the
optimal number of staff members to maximize patients throughput while reducing patients waiting time.
Rub et al. [10] addressed the importance and usefulness of modeling tools for analyzing complex processes in the healthcare sector. Modeling tools identify process bottlenecks, invalidated data, and cross-functional boundary problems. Model
outputs can be provided on the basis of future scenarios. The study applied Role Activity Diagramming modeling technique
to analyze the collaboration of cancer centers and cancer registries for improving administrative quality of cancer treatment
and enriching collection of information about cancer cases.
Griffiths et al. [11] studied the optimization of the number of beds to minimize the cancelation of elective surgery with
the use of discrete event simulation, which maintained an acceptable level occupied hospital beds. Furthermore, what-if
scenarios were used in the increasing number of beds and ring-fencing beds, reducing the length of stay due to the delayed
discharge and schedule changes for elective surgery. The results showed that, if the hospitals original 24 beds were used, it
would not adhere to the bed-occupancy guidelines. Therefore, additional five beds were required for the hospital. With these,
there would be no elective surgeries would be canceled. Moreover, different configurations were tested to solve the bed
blocking and ring-fencing bed problems for elective patients and unplanned admissions.
Korst et al. [12] supported the claim that it is significant to evaluate the hospitals capability to collaborate first because
data sharing is a challenging step to take. Factor analysis was conducted to determine the most important factor in foreseeing success: 1. the presence of a supportive hospital executive; 2. the extent to which a hospital values data; 3. the presence

16

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

of leaders vision for how the collaborative approach advances the hospitals strategic goals; 4. the hospitals use of the collaborative data to track quality outcomes; and 5. the staffs recognition of a strong mandate for collaborative participation.
Zhang et al. [13] utilized demographic and survival analysis, discrete event simulation, and optimization for setting longterm care capacity planning in order to achieve certain target wait time service levels. The researchers developed a decision
support system that would be of help to real-world conditions. Based on the simulation results of long-term bed capacity
planning, this approach was found to be preferable to the fixed ratio approach, stationary, independent, period by period
(SIPP), and modified offered load (MOL) as it lengthened the service times. The fixed ratio approach meant that if the ratio
was 0.075, 75 beds would be needed for every 1000 population aged over 75 years old. Within the same year (2012),
Masselink et al. [14] scrutinized the effect of pharmacy policies on patients waiting time in the Chemotherapy Day Unit
of the Netherlands Cancer Institute Antoni van Leeuwenhoek Hospital. Using a case study and simulation model, patients
waiting time and wastage costs were evaluated to define the trade-off. Thus, this study established a new policy at the career
center that is expected to decrease the waiting time by half while increasing the pharmacys costs by only 1% to 2%.
Al-Araidah et al. [15] examined outpatient clinics at a local hospital in Tehran, Iran. The researchers used a discrete-eventsimulation model with the total visit time and service times at stations as data. Statistical comparison was also used to confirm the model with the performance in the actual system. The outcomes showed that many improvement scenarios can be
applied to have a decrease in waiting time up to 29% and decrease in total visit time up to 19% without buying new resources.
Aeenparast et al. [16] also applied a scenario comparison in which the researchers studied a model that would reduce the
outpatients waiting time through system simulation. The simulation models were categorized from Scenarios 1 to 10, with
different waiting times. Patients waiting times were divided into three levels, depending on their physician. The result of the
study showed that the mix of increasing work time of the resident physicians, increasing work time of the senior staff physicians, and changing the start of patients admission was the best alternative for reducing patients waiting time by about
73.09%.
Gupta et al. [17] applied simulation optimization with Arena 13.0 to analyze the drive-through mass vaccination problem.
The distinctive feature of the simulation optimization allows decision makers to investigate interacting control variables by
optimizing criteria and performance measures under uncertain simulation scenarios. This approach has been applied successfully in Louisville, Kentucky (United States) in 2009.
Chen and Juan [18] studied a three-hospital patient-referral problem, which consisted of two large hospitals and a small
hospital. They considered the patients shortest waiting time as the objective function and added a budget constraint on hospital decisions for allowing staff overtime. In order to simplify the proposed problem, they assumed that the service time of
all computed tomography (CT) scans was the same distribution, the upper bound for the number of the daily referring
patients was 10 for all decision variables, and used the policy that referred the fixed daily patients among three hospitals.
The optimal solution was obtained by Arena software and compared to the practical results. Further, they performed two
sensitivity analyses, such as reducing the value of the maximal average patients waiting time and increasing the number
of annual CT patients, in order to explore the managerial implications for hospital managers. The main differences between
Chen and Juan [18] and this research were that this study relaxed the simplification part of Chen and Juans [18] research and
made their model more realistic, considering two kinds of objective functions, assuming that the service time of each MRI
scan has its own distribution, assuming that the upper bound for the number of the daily referring patients was based on the
maximal number of the daily arrival patients in a month, and considering the policies that referred to the fixed and unfixed
daily patients among two hospitals.
Gillespie et al. [19] compared the deterministic model and the stochastic model of the Orthopedic Integrated Clinical
Assessment and Treatment Service (OICATS). The researchers specifically studied the deficiency in the number of staff members allocated if the uncertainty in the planning stage is disregarded. If this happened, then queues will start to build up, and
the patients will experience a decrease in service quality. In the deterministic model, it was assumed that no variation
occurred in the OICATS department, where the results exhibited that 5625 patients could be treated. In contrast, in that
stochastic simulation with some realistic variations, the OICATS staff can only treat up to 3734 patients. The results showed
that variation had an effect when queuing is an essential property of the system.
Most researchers have used simulation models to understand the behavior of a hospital system and to evaluate the various strategies for the operation of the system. Different types of the simulation models, such as discrete event, deterministic,
and stochastic, have been developed and utilized. Optimization was also taken into consideration in some research studies to
determine the best option for minimizing waiting time and the total cost or maximizing revenues. The objective of most
research studies is to attain a certain waiting-time level for patients. Various facets in the medical field have been tackled
in the prior years, including collaboration among hospitals, MIC, elective surgery, long-term capacity care planning, and
chemotherapy units. However, this study will focus on hospital collaboration to develop a patient-referral mechanism to
improve waiting time. Nevertheless, existing research will contribute to the development of the study.

3. System description
The two case hospitals belong to one organization and have a collaborative relationship. One hospital director is in charge
of the operating procedures for both Hospitals A and B. The hospital director wants to shorten patients average wait, which
is a key performance indicator in the global budget system.

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

17

The system considered in the study involves only the MRI departments for the case hospitals (i.e., Hospitals A and B),
which offer MRI services. For both hospitals, these MRI services are classified into six types depending on the body part
to be examined, such as brain, chest, liver, spine, limb, and artery. The case hospitals offer these kinds of MRI services for
outpatients, inpatients, health check patients, and emergency patients. The physician decides whether or not the patient
requires an enhancement injection undergoing the MRI service. Hospitals A and B both have one MRI machine. The related
parameters of MRI services and MRI patients will be described in Section 4.2.
4. Model simulation
4.1. Simulation framework
Based on the descriptions in Section 3, a simulation framework was established. Patient arrival patterns were determined
using probability distribution with real data provided by case hospitals. Outpatient rules are defined for each hospital.
Optimization models were embedded within the simulation model to determine the best number of referral patients.
Fig. 1 shows an overview of the simulation flow occurring between the two collaborative hospitals from the time the patient
enters the hospital, whether or not the patient is transferred, receives MRI service, and records data until the patient leaves
the original hospital.
The simulation models include the following details:
1. Four types of patients receive MRI services: outpatients, inpatients, emergency patients, and health check patients.
2. The priority level of patients from highest to lowest is emergency patients, inpatients, health check patients, and
outpatients.
3. Only outpatients can be referred to other hospitals as the transfer of inpatients and emergency patients can be a critical
condition. In addition, health check patients require only simple check-ups that consume minimal time; thus, referrals
are not needed in such cases.
4. Only brain, spine, and limb MRI services are considered; these three MRI services account for more than 92% of all MRI
patients in both Hospitals A and B.
5. MRI services are offered 8 h per day in both hospitals.
6. The service times for each type of MRI service for both hospitals are assumed to have the same distribution.
7. MRI machines are assumed to be in good working conditions during the operating hours; the maintenance of MRI machines in the case hospitals is performed during off hours.
8. The referral transportation time from Hospital A to Hospital B is approximately 50 min; the same is true from Hospital B
to Hospital A.
4.2. MRI patients inter-arrival time and MRI service time
The inter-arrival time of patients is important for estimating the time between two consecutive patients entering the system, considering all types of patients (i.e., outpatients, inpatients, health check patients, and emergency patients). However,
service time is essential to determine the average service time of each kind of MRI services.
Using the Input Analyzer Tool,1 inter-arrival times were data-fitted to ensure that they passed chi-square and Kolmogorov
Smirnov tests, wherein the p-value should be greater than 0.05. Figs. 2 and 3 show the distributions of MRI patients interarrival times for Hospitals A and B, which were used to create the patient entities entering in the model. Figs. 46 indicate
the distributions of MRI service times for brain, spine, and limb services used for both case hospitals.
4.3. Parameters of simulation models
In order to obtain steady-state data, simulation model parameters were computed, including the warm-up period, replication length, and replication number. The value for a warm-up period was set to 180 days, 60 for replication length, and 80
for replication number. The coefficient of variation (C:V.), which is the standard deviation divided by the mean, should be
small. This research assumed that the value of C:V. should be less than 0.15 based on the empirical rule. For the replication
number (n) in Eq. (1), the half-width value (the half of the confidence interval, h) is also used as the basis in Eq. (2) and
should fall within an acceptable range. In Eq. (1), n = required replication number; n0 = initial replication number;
h = required half width of the confidence interval; and h0 = initial half width of the confidence interval. In Eq. (2), a = level
of the confidence interval and s = standard deviation of the replication data.
2

n  n0 

h0
h

s
h t n1;1a2  p
n
1

Input Analyzer Tool is software developed by the Rockwell Company that can determine inter-arrival and service times of real data.

1
2

18

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

Hospital A

Hospital B

A patient arrives at
Hospital A

A patient arrives at
Hospital B

Outpatient refer to
Hospital B?

Outpatient refer to
Hospital A?

Within quota of
Hospital B?

Within quota of
Hospital A?

Y Y

MRI at Hospital A

Track patient data


in Hospital A

The patient leaves


Hospital A

MRI at Hospital B

Optimization
model

Track patient data


in Hospital B

The patient leaves


Hospital B

Fig. 1. Flowchart model between two collaborative hospitals.

5. Optimization model
5.1. Mathematical models
Revised mathematical models from Chen and Juan [18] were used to define the objective functions and constraints. The
following indices, parameters, objectives, and constraints come from the detailed models.

19

Probability Density Function of Hospital A

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

0.15
0.14
0.13
0.12
0.11
0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0

Expression in ARENA of MRI patients inter-arrival time

20

40

60

80

100

120

140

160

MRI patients' inter-arrival time (in minutes)


Histogram

Beta

Probability Density Function of Hospital B

Fig. 2. Formula based on Beta distribution for Hospital A.

0.44
0.4

Expression in ARENA of MRI patients inter-arrival time

0.36
0.32
0.28
0.24
0.2
0.16
0.12
0.08
0.04
0
0

100

200

300

400

500

MRI patients' inter-arrival time (in minutes)

Histogram

Weibull

Fig. 3. Formula based on Weibull distribution for Hospital B.

Indices:
i and j represent the hospital index. In this paper, i 1; 2; j 1; 2; i j.
k represents the patient index, where k 1; 2; . . . ; N i . N i is the original maximum number of patients in Hospital i.
represents the MRI service. 1 represents the Brain MRI service; 2 represents the Spine MRI service; and 3
represents the Limb MRI service.
Parameters:
Oik represents the revenue for patient k for th MRI service in Hospital i.
Aik represents the appointments start time of patient k for th MRI service in Hospital i.
Pik represents the examination time of patient k for th MRI service in Hospital i.
Q i represents the maximum waiting time at Hospital i.
U ij represents the upper bound of the number of MRI patients referred each day from Hospital i to Hospital j.
Lij represents the lower bound of the number of MRI patients referred each day from Hospital i to Hospital j.
C i represents the close time of MRI each day at Hospital i.

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

Probability Density Function of the brain service

20

0.24
0.22
0.2

Expression in ARENA of the brain MRI service time

0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0

50

100

150

200

MRI service time (in minutes)


Histogram

Gamma

Probability Density Function of the spine service

Fig. 4. Formula based on Gamma distribution for the brain MRI service.

0.3
0.28
0.26
0.24
0.22
0.2
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0

Expression in ARENA of the spine MRI service time

20

40

60

80

100

MRI service time (in minutes)


Histogram

Gamma

Fig. 5. Formula based on Gamma distribution for the spine MRI service.

T i represents the maximum overtime at Hospital i.


S represents the predetermined time by Hospitals i and j so that the hospitals can serve as many patients as possible.
Decision variables:
n represents the number of days.
wik represents the waiting time of patient k for th MRI service in Hospital i.
i

f k represents the finish time of patient k for th MRI service in Hospital i.


xij represents the number of th MRI patients referred each day from Hospital i to Hospital j.
X ij represents the number of MRI patients referred each day from Hospital i to Hospital j. This decision variable is deterP
mined before a simulation experiment. X ij 31 xij .
Optimization Model 1:

z E min

N1X
X 21 X
3

w1k

k1

N 2X
X 12 X
3

w2k

k1

!
3

21

Probability Density Function of the limb service

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

0.64
0.56

Expression in ARENA of the limb MRI service time

0.48
0.4
0.32
0.24
0.16
0.08
0
0

50

100

150

200

MRI service time (in minutes)


Histogram

Weibull

Fig. 6. Formula based on Weibull distribution for the limb MRI service.

Subject to
i

wik f k1 Pik  Aik


wik

6 Qi

Lij 6 X ij 6 U ij

f k1 Pik 6 nC i T i

Optimization Model 2:

z E max

N1 X
3
21 X 12 X
X
k1

O1k

N 2 X
3
12 X 21 X
X

O2k

k1

!
8

Subject to
i

wik f k1 Pik  Aik

6 Qi

10

Lij 6 X ij 6 U ij

11

wik
i

f k1 Pik 6 S

12

Two models were established with different objectives: to minimize the MRI patients average waiting time at Hospitals 1
and 2, as shown in Eq. (3), or to maximize the revenues of Hospitals 1 and 2, as shown in Eq. (8). The two models have the
same constraints except for constraints (7) and (12) because of the different scenario settings. In optimization model 2, a
fixed predetermined time is required as the base for comparing the revenues of two case hospitals without considering
maximum overtime. However, predetermined time is not fixed for optimization model 1 because the objective function
of optimization model 1 is in time unit whiles the objective function of optimization model 2 is in money units (New Taiwan
Dollar, NTD).
Constraints (4) and (9) determine patients waiting time. Constraints (5) and (10) represent the patients average waiting
time in different kinds of MRI services at Hospital i, which must be less than or equal to the maximum allowed waiting time,
Q i . Constraints (6) and (11) represent the number (X ij ) of MRI patients referred each day from Hospital i to Hospital j, which
must be between the lower bound (Lij ) and upper bound (U ij ). Based on the interviews of the hospital manager, the upper
bound is set to the maximal number of the daily arrival MRI patients in a month. Constraint (7) ensures that all patients have
MRI services. Constraint (12) provides a time base for estimating total revenues of different scenario settings.
5.2. Simulation optimization
The researchers used Arena OptQuest software and followed the simulation optimization algorithm formulated by
Klassen and Yoogalingam [20]. First, for the initialization step, candidate solutions were generated and the diverse population identified. Second, in the simulation process, the researchers proceeded with 80 replications to generate more stable and

22

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

better solutions in every run of the simulation. Third, the optimization process combined the existing feasible solution (parents) to create new solutions (offspring) and replace the worst parent with the best offspring. To avoid revisiting the preceding inferior solutions, the offspring was modified with tabu memory functions. Finally, the researchers used a user-specified
number of heuristic iterations, such as 80 iterations, as the stopping criterion. Each iteration contains 80 replications.
The feasible solutions of Arena OptQuest software were generated based on iteration searching (the scatter search
method) with 80 replications for each iteration. The objective value of each iteration was the average objective value of
the 80 replications. The optimal solution was the best average objective value of the 80 iterations.
5.3. Verification and validation of simulation models
After building the simulation model, the researchers used verification and validation to scrutinize the logic and process of
the model. The built models were checked to determine if the logical flow of the entities and processes were correct. For
example, when an outpatient arrives at Hospital A, a decision is made whether or not to transfer that outpatient to Hospital
B. Afterwards, the simulation model will check if the outpatient is within the number of allowed referred patients. If the
referral number is below the referring quota, the patient could be referred to Hospital B with an approximate delay of
50 min for the transportation time; otherwise, the patient referral will not take place.
After the MRI service, the patient will be transferred back to Hospital A for records with a delay of approximately 50 min
for the transportation time. Finally, the patient will leave the original hospital. The verification process will be successful
only if all the logical flows are appropriate for the desired outcome of the system.
In order to make sure that the result of the simulation model will execute a system comparable to the actual condition,
the researchers validated the service time of each kind of MRI processes, average cycle time of all types of patients, and the
total number of patients who received MRI services and left Hospitals A and B for two months by running 80 replications.
This validation process was done to test if a significant difference exists between the actual and simulated systems. Given the
data for patients average waiting time, the values of the mean and standard deviation were used with the help of the paired
t-test of statistical comparison test using the 95% confidence interval.
6. Results and discussions
6.1. First objective: minimize patients average waiting time
Hospital A began its collaboration with Hospital B in 2011. Hospital A is located in a larger city with 230,202 residents;
Hospital B is located in a smaller city with 59,657 residents. Due to the geographical factor, patients arrivals at Hospital A
outnumber patients arrivals at Hospital B. Therefore, patients of Hospital A experience long waiting times. Thus, a patient
referral mechanism will be implemented to accelerate the MRI service. Referral system models of the fixed number and the
unfixed number of referral outpatients per day were built. Therefore, the researchers first objective is to minimize patients
average waiting time in Hospitals A and B.
Scenario 1: Fixed number of referral outpatients per day
Using the OptQuest Tool, the researchers defined the maximum number of referral outpatients as a control, and defined
the waiting times of the three considered MRI services and patients average waiting time as responses. Moreover, the waiting times in the brain, spine, and limb MRI services were set to be less than or equal to 21 days due to hospitals criteria.
Finally, the objective function is to minimize patients average waiting time in Hospitals A and B. After simulating the data,
the researchers obtained an optimal solution that referred to 3 referral outpatients per day with a minimized waiting time
for patients equal to 0.53 days.
Scenario 2: Unfixed number of referral outpatients per day
Due to the fluctuating number of patients entering Hospital A, having a fixed number of referral outpatients might not be
sufficient for addressing patients long waiting time problems in the long run. Therefore, the researchers also considered
building a model based on Scenario 1, but with an unfixed number of referral outpatients per day for a span of one month.
The model for Scenario 2 is similar to that of Scenario 1, except for a few modifications, such as adding the maximum
number of referral outpatients per day, which are variables, for the next 29 days. Initial values of these variables were set
to 0 to determine the appropriate number of referral outpatients per day.
Using the OptQuest tool, the responses, constraints, and objective functions for this scenario were similar in Scenario 1.
However, additional responses were included, such as the maximum number of referral outpatients for the second day until
the last day of the month. Afterwards, the researchers obtained an optimal solution with a minimized waiting time for
patients equal to 0.48 days
6.2. Second objective: maximize hospitals revenues
The researchers considered the hospitals financial needs to render better quality services to patients. From this viewpoint, another objective was defined: to maximize hospitals revenues. In this case, the needs of both the patient and the
hospital are satisfied. Therefore, both case hospitals revenues will be studied further.

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

23

Scenario 1: Fixed number of referral outpatients per day


In this model, the same logic of the system was considered, but additional variables for the calculation of revenues were
included, such as the revenues, injection points, non-injection points, and equivalent points. These variables were added
because these can affect the revenues of the hospital. The injection point for one patient is 11,500 points compared to
6500 points for non-injection patients. To convert these points to a monetary value, they were multiplied to the equivalent
point, 0.9568. The equivalent point is predetermined by hospitals and government agencies every year.
With the OptQuest tool, the researchers defined the maximum number of referral outpatients as a control, and defined
the waiting times of the three considered MRI services and total revenues for Hospitals A and B as responses. Furthermore,
the waiting times in the brain, spine, and limb MRI services were set to be less than or equal to 21 days, based on the hospitals criteria. Finally, the objective function was to maximize the revenues of both Hospitals A and B.
In considering the revenues of Hospital A, the researchers obtained the optimal solution with one number of referral outpatients per day with a maximized revenue of 6,071,664 NTD for Hospital A and 3,157,533 NTD for Hospital B. The computed
costs were subsequently subtracted from the revenues of both Hospitals A and B. These are the labor and transportation
costs of transferring an outpatient to another hospital. The distance between the two hospitals is 45.2 km. After the labor
and transportation costs are deducted from the revenues, the remaining revenues for Hospitals A and B are 6,061,849
NTD, and 3,147,718 NTD, respectively. The depreciation cost of the MRI machine and associated cost are not considered
in this research. In general, the profits of Hospitals A and B will be much less than the remaining revenues.
Scenario 2: Unfixed number of referral outpatients per day
Using the OptQuest tool, the constraints and objective function for this scenario are the same as in Scenario 1. In addition,
there were more responses, such as the maximum number of referral outpatients for the second day until the last day of the
month. In considering the revenues of Hospitals A and B, the researchers obtained the optimal solution with the number of
referral outpatients per day with a maximized revenue of 2,450,025 NTD for Hospital A and 2,271,577 NTD for Hospital B.
Successively, the computed costs were subtracted from the revenues of both Hospitals A and B. These are the labor and transportation costs of transferring an outpatient to another hospital. The remaining revenues for Hospitals A and B are 2,445,117
NTD and 2,266,669 NTD, respectively.
Table 1 summarizes the number of daily referral outpatients in the first 10 days with different objective functions under
case settings. For the scenario of fixed number of referral outpatients per day, the number of daily referral outpatients considering minimizing patients average waiting time is larger than that of considering maximizing hospitals revenues. The
reason comes from extra cost of referral outpatients. This implies if Hospital A takes patients position rather themselves,
it will improve its service level comparing to its original operations. The tendency of the second scenario of unfixed number
of referral outpatients per day is the same as the first scenario. In general, the number of referral outpatients per day with
considering the first objective is larger than that with considering the second objective.
6.3. Sensitivity analysis
In order to investigate the effect of two key parameters (revenues and patients average waiting time) by adding a new
constraint on revenues or patients average waiting time, this study conducted two sensitivity analyses. The details are
described in the following subsections.
6.3.1. First objective: minimize patients average waiting time
Scenario 1: Fixed number of referral outpatients per day
In minimizing patients average waiting time, revenues were reduced at most by 5%, 10%, 15%, and 20%. These values are
the revenues that the hospital director is willing to sacrifice in accordance with the hospitals criterion of less than or equal to
21 days for patients average waiting time. Thus, if the hospital director is willing to sacrifice revenues under 10%, the optimal number of outpatients to be referred per day is 1, that is, cases of 5% and 10% revenues reduction make the director do
the same decision, which results in a 6.49-day patients average waiting time. By reducing the revenues by less than or equal
to 15% to 20%, the optimal number of outpatients to be referred per day are 2 and 3, respectively. Moreover, patients average
waiting times are 0.62 days and 0.55 days respectively when reduced to less than or equal to 15% and 20% of the revenues.
Fig. 7 shows results of Scenario 1 of the first objective.
Scenario 2: Unfixed number of referral outpatients per day
Fig. 8 shows results of Scenario 2 of the first objective. If the hospital director considers the unfixed patient referral mechanism and sacrifices the revenues by at most 5%, 10%, 15%, and 20%, patients average waiting times are 1.27, 1.27, 0.52, and
0.49 days, respectively, with corresponding revenues of 2,910,097 NTD, 2,910,097 NTD, 2,628,224 NTD, and 2,421,620 NTD.
6.3.2. Second objective: maximize hospitals revenues
Scenario 1: Fixed number of referral outpatients per day
In the maximized revenues in Hospitals A and B, patients average waiting time was reduced to less than or equal to 18,
15, 12, and 9 days. Fig. 9 shows results that indicate the impact of reducing patients average waiting time for revenues. If the
hospital director is willing to prioritize patients average waiting time by reducing it to less than or equal to 18, 15, or
12 days, the optimal number of outpatients to be referred per day is 1, resulting in a value of 6.49 days with 6,071,664

24

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

Table 1
Number of referral outpatients per day in the first 10 days.
Day

1st

2nd

3rd

4th

5th

6th

7th

8th

9th

10th

1st obj.

Scenario 1
Scenario 2

3
7

3
4

3
5

3
6

3
6

3
5

3
2

3
7

3
6

3
6

2nd obj.

Scenario 1
Scenario 2

1
6

1
2

1
6

1
6

1
3

1
5

1
2

1
4

1
6

1
2

Patients' average waiting times


7

Number of outpatients

6
3

5
4

2
3
2

1
0
0%

0
5%

10%

15%

20%

25%

Actural average waiting time under referral policy

Optimal number of outpatients to be referred per day

Percentage of reduced revenues (under the referral policy) comparing to the


original revenues (no referral policy)
Fig. 7. Simulation optimized cases for Hospital A with the first objective for Scenario 1.

Patients' average waiting times


1.4

4.0

1.2

Revenue (10e6)

3.0

1
0.8

2.0
0.6
0.4

1.0

0.2
0.0
0%

5%

10%

15%

20%

0
25%

Actual average waiting time under the referral


policy

Revenue

Percentage of reduced revenues (under the referral policy) comparing to the


original revenues (no referral policy)
Fig. 8. Simulation optimized cases for Hospital A with the first objective for Scenario 2.

NTD in revenue. By reducing patients average waiting time to less than or equal to 9 days, the optimal number of outpatients to be referred per day is 3, resulting in a value of 0.55 days with 4,936,275 NTD in revenue.
Scenario 2: Unfixed number of referral outpatients per day
Fig. 10 indicates the results of Scenario 2 of the second objective. If the hospital director considers the unfixed patient
referral mechanism and prioritizes patients average waiting times by reducing them to less than or equal to 18, 15, 12,

25

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

Optimal number of outpatients to be referred per day


7

Number of outpatients

6
3

5
4

2
3
2

1
0

0
8

10

11

12

13

14

15

16

17

18

19

Actual average waiting time under referral policy

Patients' average waiting times


4

Pre-determined patients' waiting time under the referral policy


Fig. 9. Simulation optimized cases for Hospital A with the second objective for Scenario 1.

Patients' average waiting times

Revenue (10e6)

1
8

10

11

12

13

14

15

16

17

18

19

Actual average waiting time under referral policy

Revenue

Pre-determined patients' waiting time under the referral policy


Fig. 10. Simulation optimized cases for Hospital A with the second objective for Scenario 2.

and 9 days, the corresponding revenues are 3,004,292 NTD, 3,004,292 NTD, 3,004,292 NTD, and 2,993,874 NTD. The reduction of patients average waiting times to less than or equal to 18, 15, and 12 days yielded the same results. The result indicates that if the hospital is willing to sacrifice some revenues to make patients waiting times be reduced by 3 days, in reality,
it might have advantages that the actual reduction time can reach 9 days, that is, the average patients waiting time for the
pre-determined criterion is 12 days.
7. Conclusions and future research
The first academic contributions of this research is that it proposed the more rigid mathematical formulations, which
relaxed the simplification part of Chen and Juans [18] research. The proposed models of this research were more realistic
for the further academic study. Second, for simulation methodology, combining stochastic parameters (random patients
arrival and random patients service time) with constraints, the what-if simulation scenario would be required in either
a trial-and-error way to search for an optimal solution. Take the unfixed referral policy as an example, 30 decision variables,

26

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

such as the number of the daily referring patients, were integer variables, ranging from 0 to the maximal number (i.e., N
patients) of the daily arrival patients in a month. The problem is similar to the multi-stage stochastic problem (cf. Dyer
and Stougie [21]). For each day (stage), Hospital A has to determine the number of referral outpatient (decision) under uncertain environment parameters. This implies the problem of this study is a NP-hard problem. This led to long computation time
for exhaustive searching. However, by simulation optimization, the convergence could be quickly obtained within 80 iterations and 80 replications, which were total 6400 runs. The simulation time was not proportional to the value of N. Therefore, the simulation optimization method indeed shortened a lot of computing time compared to the traditional simulation
method.
From the viewpoint of hospital management, this study determined a collaborative patient referral mechanism between
two case hospitals with different objectives: minimizing patients average waiting time and maximizing the revenues in both
hospitals. The researchers used scenarios for each objective by determining a fixed and unfixed optimal number of outpatients to be referred per day between the two hospitals.
After evaluating the two objectives with different scenarios, it was observed in the first objective that there is a drastic
decrease occurred in patients average waiting time along with the revenues. In maximizing revenues, there will also be a
decrease in patients average waiting time, but only a small decrease in revenues. These factors could help the hospital director make decisions in guiding, leading, and rendering a quality service.
Finally, the researchers validated that simulation optimization could be a helpful tool in the healthcare sector without
changing any processes and entities in a system. With the use of Arena software, the researchers were able to conduct analyses and achieve the purpose of their study.
For further development, the researchers recommend that future researchers examining a similar topic with a deeper discussion of hospital collaboration. Other considerations, such as referral not only of MRI outpatients but also other types of
patients, can be tackled. In this paper, the researchers focused on minimizing patients average waiting time or maximizing
the revenues for Hospitals A and B. On the other side, the objective function can be changed by simultaneously minimizing
patients average waiting time and maximizing revenues.
Due to service time constraints, future researchers might also consider another type of hospital services in the imaging
center, such as X-ray, CT, or ultrasound tests. A collaboration involving more than two hospitals could have a more convenient service. Other service industries can also apply collaboration to fully comprehend its importance.
Acknowledgements
The authors are grateful to anonymous referees for their very valuable comments, which have significantly enhanced this
work. This research is supported by the Ministry of Science and Technology under contract No. MOST 104-2221-E-033-025-.
References
[1] G. Bammer, Enhancing research collaborations: three key management challenges, Res. Policy 37 (5) (2008) 875887.
[2] A. Lomi, D. Mascia, D.Q. Vu, F. Pallotti, G. Conaldi, T.J. Iwashyna, Quality of care and interhospital collaboration: a study of patient transfers in Italy,
Med. Care 52 (5) (2014) 407414.
[3] C.M. Callahan, M.A. Boustani, F.W. Unverzagt, M.G. Austrom, T.M. Damush, A.J. Perkins, B.A. Fultz, S.L. Hui, S.R. Counsell, H.C. Hendrie, Effectiveness of
collaborative care for older adults with Alzheimer disease in primary care: a randomized control trail, J. Am. Med. Assoc. 295 (18) (2006) 21482157.
[4] D.A. Campbell, E.P. Dellinger, Multihospital collaborations for surgical quality improvement, J. Am. Med. Assoc. 302 (14) (2009) 15841585.
[5] R.A. Dudley, K.L. Johansen, R. Brand, D.J. Rennie, A. Milstein, Selective referral to high-volume hospitals: estimating potentially avoidable deaths, J. Am.
Med. Assoc. 283 (9) (2000) 11591166.
[6] P.-S. Chen, C.Y. Yang, H.L. Chao, A simulation analysis of hospital collaboration: a preliminary study, in: Proceedings of the 2012 IIE Asian Conference,
Singapore, 2012.
[7] S. Kobayashi, T. Fujioka, Y. Tanaka, M. Inoue, Y. Niho, A. Miyoshi, A geographical information system using the Google map: API for guidance to referral
hospitals, J. Med. Syst. 34 (6) (2009) 11571160.
[8] F.J. Ramis, F. Baesler, E. Berho, L. Neriz, J.A. Sepulveda, A simulator to improve waiting times at a medical imaging center, in: Proceedings of the Winter
Simulation Conference, 2008, pp. 15721577.
[9] M.A. Ahmed, T.M. Alkhamis, Simulation optimization for an emergency department healthcare unit in Kuwait, Eur. J. Oper. Res. 198 (3) (2009) 936
942.
[10] F. Rub, M. Odeh, I. Beeson, D. Pheby, B. Codling, Modelling healthcare processes using role activity diagramming, Int. J. Modell. Simul. 28 (2) (2008)
147155.
[11] J.D. Griffiths, M. Jones, M.S. Read, J.E. Williams, A simulation model of bed-occupancy in a critical care unit, J. Simul. 4 (1) (2010) 5259.
[12] L.M. Korst, C.E. Aydin, J.M.K. Signer, A. Fink, Hospital readiness for health information exchange: development of metrics associated with successful
collaboration for quality improvement, Int. J. Med. Inform. 80 (8) (2011) 178188.
[13] Y. Zhang, M.L. Puterman, M. Nelson, D. Atkins, A simulation optimization approach to long-term care capacity planning, Oper. Res. 50 (2) (2012) 249
261.
[14] I.H.J. Masselinka, T.L.C. van der Mijden, N. Litvak, P.T. Vanberkel, Preparation of chemotherapy drugs: planning policy for reduced waiting times,
Omega 40 (2012) 181187.
[15] O. Al-Araidah, A. Boran, A. Wahsheh, Reducing delay in healthcare delivery at outpatients clinics using discrete event simulation, Int. J. Simul. Modell.
11 (4) (2012) 185195.
[16] A. Aeenparast, S.J. Tabibi, K. Shahanaghi, M.B. Aryanejhad, Reducing outpatient waiting time: a simulation modeling approach, Iranian Red Crescent
Med. J. 15 (9) (2013) 865869.
[17] A. Gupta, W.E. Gerald, S.H. Sunderesh, Simulation and optimization modeling for drive-through mass vaccination a generalized approach, Simul.
Modell. Pract. Theory 37 (2013) 99106.
[18] P.-S. Chen, K.L. Juan, Applying simulation optimization for solving a collaborative patient-referring mechanism problem, J. Indust. Product. Eng. 3 (6)
(2013) 405413.

P.-S. Chen et al. / Simulation Modelling Practice and Theory 61 (2016) 1427

27

[19] J. Gillespie, S. McClean, F. FitzGibbons, B. Scotney, F. Dobbs, B.J. Meenan, Do we need stochastic models for healthcare? The case of ICATS?, J Simul. 8 (4)
(2014) 293303.
[20] K.J. Klassen, R. Yoogalingam, Improving performance in outpatient appointment services with a simulation optimization approach, Product. Oper.
Management 18 (4) (2009) 447458.
[21] M. Dyer, L. Stougie, Computational complexity of stochastic programming problems, Math. Program. 106 (3) (2006) 423432.

Das könnte Ihnen auch gefallen