1"
"
SIMULATION IN ARENA TO DETERMINE POTENTIAL BOTTLENECKS
FOR DERMATOLOGY CLINIC REDESIGN
Tiffany N. Adams
H. Milton Stewart School of Industrial and Systems Engineering
Georgia Institute of Technology Atlanta, Georgia
Abstract
As demand increases for timely healthcare, the
expected queues in the registration process must be
anticipated and ideally reduced. Anne Arundel
Dermatology Clinic in Annapolis, Maryland is preparing
for an increase from currently thirty patients per day to
ultimately one hundred per day. Using the simulation
software Arena, the impact of growth can be evaluated and
bottlenecks in reception area then studied. Assuming an
increase in population, this simulation predicts that patient
wait time is negligible, which allows for a layout design
with the administration suite to be adjacent to the
registration area for staff convenience.
Introduction
Anne Arundel Dermatology Clinic (AADC) in
Annapolis, Maryland has a current throughput over 7,500
patients per year (30 per day). The clinic was able to
provide some specific data, but other data used in this
simulation had to be based on management estimates.
AADC plans to expand their facility (Figure 1), and with
that, their patient population, hoping that change will not
potentially crowd the registration area.
The ideal future design for AADC currently places the
administration area with convenient access, but only has
one passage from the waiting area to the exam rooms, with
the check-in and check-out windows facing this same
hallway. The optimal clinic configuration, from the
perspective of a clinic owner or administrator, should
simultaneously maximize clinic profit, patient satisfaction,
and staff satisfaction. However, determining an optimal
staffing and facility size is complicated by the (often
conflicting) nature of these objectives (Swisher, Jacobson,
Jun, & Balci, 2001). For example, in this clinic, a
configuration that maximizes staff satisfaction by placing
the administration suite near the registration desk may
create a potential bottleneck, thereby decreasing patient
satisfaction. Bottlenecks are areas that increase longer wait
time for the patient, and more concerning for the clinic, a
crowded area.
Arena software was chosen to model this simulation
because of its ease of use. It is compatible with flow
charts, which were easily understood and approved by the
clinic. The receptionists were also easily modeled as
Figure 1: Proposed Future Clinic Floor Layout Design
Single throat
"
Office Area
2"
"
servers in Arena, with the respective queues representing
wait lines for check-in or check-out, which was the main
focus of this simulation.
This simulation runs for an 8-hour day, for a
replication of 10 days. The clinic schedules appointments
in fifteen-minute intervals, with the first appointment at
8:00am and the last at a time to allow the clinic to close at
5:00pm. The data recording system for the clinic only
tracks the appointment times and approximate length, not
the exact length of time for the check-in or check-out
process, so these times were estimated by the clinic. These
estimated times are very important, and a concern for error
here is great, given that the data supplied is only a
conservative estimate and not recorded by a computer
system.
Problem
The current design supports the current capacity,
but with an increase of more than 200% of its current
population, the clinic needs to know that three servers for
the registration process are adequate. The most ideal
design (Figure 1) unfortunately has a single throat between
the waiting area and the exam rooms, which also serves as
the registration window area; however, this design is the
most ideal for the administrative offices, with the office
manager, medical records, and reception area all easily
accessible. A second layout option is shown in Figure 2,
which does have a double throat connecting the waiting
area to the exam rooms, but unfortunately separates the
administrative office area from the reception area. The
clinic would like to use the layout pictured in Figure 1, as
long as wait times and queue lengths are reasonable, and
revert to the plan in Figure 2 as an alternative option.
Project Objectives
The purpose of this project is to prepare the clinic for
likely results as the number of patient appointments per
day increases. The goals of this project are as follows:
• Confirm that the single throat, in Figure 1, is
adequate for a predicted increase in volume
• Collect and analyze data to predict wait times in
check-in and check-out processes
• Evaluate adequacy of three receptionists, as
opposed to two or four as possibilities
• Advise practice on predictions and probability of
wait time, given current data provided by clinic
Methodology
Literature Search
After searching literary databases, few articles appear
to have been written on the specific matter at hand. Most
references for discrete-event simulation are focused on
scheduling policies and admissions, concluding that
admission decision rules for scheduling appointments may
Double throat
"
Separates
administrative
area from
registration area
Figure 2: Alternate Future Clinic Floor Layout Design
Office Area
3"
"
increase utilization of staff and patient throughput (Rising,
Baron, & Averill, 1973); however, the option for altering
schedule rules was out of the scope of this specific project.
Information Sources
At the beginning of this project, data from August and
September 2010 were collected from the practice
managers. This clinic schedules patient appointments
every fifteen minutes, but will check them in early if they
arrive before appointment time; therefore, the exponential
distribution occurs naturally when describing the lengths
of the inter-arrival times in a homogeneous Poisson
process. However, since this clinic schedules specific
appointment times, this does not satisfy the memoryless
property, and could possibly fall under another distribution
(Alexopoulos, Goldsman, Fontanesi, Kopald, & Wilson,
2008). For simplistic reasons and compatibility with Arena
simulation software, all arrivals are estimated to be in a
random Poisson distribution. When scheduling, the clinic
also may schedule two patients with the same appointment
time, so the simulation allows for two patients to arrive
simultaneously throughout the day, to account for this
higher possibility. The simulation automatically ends
when the 60
th
, 80
th
, or 100
th
patient leaves the system,
depending on the model. The estimates of arrival times and
frequencies are shown in Table 1. Additional data times
supplied, estimated, or agreed upon by the clinic are shown
in Table 2, with their respective distributions.
Model Building
The simulation software Arena was used for this
project. The patients were considered entities and the
receptionists were the resources of focus. A simplified
flowchart of a patient’s experience is shown in Figure 3.
The black background boxes are the processes of focus,
where the patient would be in this potential bottleneck
area: check-in, walking from wait area to exam room, and
check-out.
Assumptions are included in the model for a realistic
result; three receptionists, versatile for the check-in or
check-out process, are each allotted fifteen minutes per
hour for other tasks, such as answering phones, scheduling
patients, or taking a break. The clinic mails registration
forms to patients before their appointment, as an
opportunity for a faster check-in; the clinic estimates about
half of their patients complete forms prior to appointment
Time assigned
Normal (1, 0.2
2
)
Normal (10, 2
2
)
Tri (2, 4.5, 6)
Tri (1, 2, 3)
Tri (0, 30, 40)
Normal (0.5, 0.1
2
)
Tri (15, 45, 90)
Tri (10, 30, 45)
Tri (0.5, 3, 4)
Normal (1, 0.2
2
)
Normal (3, 0.2
2
)
Patient
Volume
Arriving 1 or 2 at
a time?
Frequency
100 patients
1
4 min
100 patients
2
25 min
80 patients
1
5 min
80 patients
2
30 min
60 patients
1
8 min
60 patients
2
60 min
For Testing:
30 patients
1
15 min
30 patients
2
180 min
Figure 3: Patient Flow at Anne Arundel Dermatology Clinic
Table 1: Arrival Frequency Based on Patient Volume
Table 2: Process Times (Poisson)
4"
"
time, so the probability used in the simulation is 0.5. In the
clinic layout as shown in Figure 1, nine exam rooms are
available for use. Two months of clinic data, analyzed by
reason for appointment, show the distribution of 18% new
patients and 81% established patients. The clinic also
wanted to take into consideration Accutane patients, which
account for the remaining 1% of the patient population,
who take slightly longer to check-out because they must
schedule the next appointment. These probabilities are
incorporated into the model, and follow the respective
process times.
Arena also allows the creation of time plots (Figure 4),
which were used as a visual for the number of patients in
the check-in or check-out process. The x-axis is time, in
minutes, beginning at 8:00am; the y-axis is number of
patients in the process. Figure 4 is specifically a replication
for a 60 patient volume scenario.
Verification, Validation, and Testing
Several verification, validation, and testing (VV&T)
techniques were applied during this project. Balci
describes verification as building the model right and
validation as building the right model (Balci, 1997). To
gather results as realistic as possible, the simulation was
run for ten eight-hour days. Another common technique
for testing is to run the new model with historical data, and
compare performance measures (Elbeyli & Krishnan,
2000); this model was tested by running a simulation with
the same assumptions, but current volume of
approximately thirty patients per day. This resulted in
arrival times every fifteen minutes, an average utilization
per receptionist of 7.8%, and an almost nonexistent wait
Check-in
Check-out
Figure 4: Time Plots
Number
of
Patients
in
Process
Number
of
Patients
in
Process
8:00am 9 10 11 1 2 3 4 5:00pm
Noon-1pm is Lunch
5"
"
time, as seen in Table 3. These observations are very
similar to the current state of the practice, and therefore the
model performs adequately well and provides results at the
level of accuracy desired for this project.
In addition to the other VV&T techniques mentioned
above, the simulation was reviewed and discussed with the
Chief Executive Officer of Anne Arundel Dermatology
Clinic.
Results and Conclusion
This project was initially simulated in order to gain a
better understanding of the utilization of receptionists,
which in turn identifies the amount of patients in the
bottleneck area. The Process Analyzer program in Arena
was used to compare the utilization of receptionists and
wait times in line from the increase of 30 patients to 100
patients, as seen in Table 3. “Receptionists Busy”
indicates the number of receptionists busy on average; for
example, 0.793 * 480 minutes is 381 minutes of work per
day among all three receptionists checking in and checking
out patients in the registration process. Individual
utilization per receptionist is also predicted at about 26.4%
on a day with patient volumes at 100; the maximum
utilization per receptionist would be 75%, given the
fifteen-minute allotment for other tasks per hour. As seen
from check-in and check-out times, few patients wait to
register, but those that do, rarely wait more than two
minutes. Given the earlier analysis of optimal clinic
configuration, the clinic layout with the office
administrative area adjacent to the registration area will be
the most beneficial to the clinic up to 100 patients in
volume.
Next Steps
After the completion of this initial project, many
additions could be made to strengthen the model and
expand its application. The simulation could determine
optimum utilization of exam rooms or include a possibility
of extended hours for this specific clinic. This simulation
could also be further reviewed for accuracy once the
clinic’s volume has expanded. Since time was a restricting
factor in this project, direct observation was not available
as an option for data collection; in the future, direct
observation could provide better data estimates.
When a simulation expands with more process steps,
the room for error increases; this project along with other
simulations could be studied to determine if adding a
process step, even if only estimated, could prove beneficial
to the model. Also, once these wait times are confirmed
with the given assumptions, the simulation could be used
to report expected wait times based on patient volume in a
clinic, in order to establish benchmark data.
This model could also be used for other clinics to
determine wait lines based on individual clinic daily
volume.
Acknowledgements
The author wishes to acknowledge the constant
support and advice of David Cowan from the Health
Systems Institute of the Georgia Institute of Technology as
well as Anne Arundel Dermatology Clinic for supplying
data, reviewing the simulation, and approving the use of
the information gathered during this research. She would
also like to recognize Professor Christos Alexopoulos for
his help with the Arena simulation program.
References
Alexopoulos, C., Goldsman, D., Fontanesi, J., Kopald, D.,
& Wilson, J. (2008). Modeling patient arrivals in
community clinics. Omega: The International Journal
of Management Science, 36(1), Retrieved from
Patient'
Volume'
Repli/
cations'
Rep'
Length'
(min)'
Receptionists'
Busy''
(out'of'3)'
Receptionist'
Utilization'(per'
receptionist)'
Check/in''
wait'
average'
Check/'
in''wait''
max'
Check/
out''wait'
average'
Check/
out'wait'
max'
100"Patients"
10"
480"
0.793"
0.264"
0.137"
1.898"
0.071"
1.651"
80"Patients"
10"
480"
0.643"
0.214"
0.104"
1.491"
0.055"
1.196"
60"Patients"
10"
480"
0.477"
0.159"
0.033"
1.037"
0.041"
0.892"
30"Patients"
10"
480"
0.233"
0.078"
0.005"
0.138"
0"
0"
Table 3: Comparison among Potential Patient Volumes
6"
"
http://www2.isye.gatech.edu/~sman/publications/OM
E790_final.pdf
Balci, O. (1997). Principles of Simulation Model
Validation, Verification, and Testing. Transactions of
the Society for Computer Simulation, 14(1), 3-12.
Elbeyli, S., & Krishnan, P. (2000). In-patient flow analysis
using promodel™ simulation package. Informally
published manuscript, Department of Food and
Resource Economics, University of Delaware,
Newark, Delaware. Retrieved from
https://ageconsearch.umn.edu/bitstream/15827/1/sp00
0002.pdf
Rising, E. , Baron, R. , & Averill, B. (1973). Systems
analysis of a university-health-service outpatient
clinic. Operations Research, 21(5), 1030-1047.
Swisher, J., Jacobson, S., Jun, J., & Balci, O. (2001).
Modeling and analyzing a physician clinic
environment using discrete-event (visual) simulation.
Computers & Operations Research, 28(2), doi:
10.1016/S0305-0548(99)00093-3
Biographical Sketch
Tiffany Adams is an Industrial and Systems
Engineering student in her senior year at Georgia Institute
of Technology. This project was completed as part of her
undergraduate research with the Health Systems Institute.
Tiffany will graduate in May 2011 with a Bachelors of
Engineering and plans to continue her studies by pursuing
her Masters of Science degree in Health Systems. Tiffany
joined the Institute of Industrial Engineers during her
freshman year and Society of Health Systems her junior
year with hopes to further her understanding and
knowledge of the healthcare system in order to improve
the efficiency of the current system.
E-mail: tadams7@gatech.edu