This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The Pennsylvania State University
The Graduate School
A HYBRID MODELING APPROACH USING DISCRETE EVENT SIMULATION AND
LAYOUT OPTIMIZATION FOR HEALTHCARE LAYOUT PLANNING PROBLEMS
The dissertation of Jennifer I. Lather was reviewed and approved* by the following:
John I. Messner
Charles and Elinor Matts Professor of Architectural Engineering
Dissertation Advisor
Chair of Committee
Robert M. Leicht
Associate Professor of Architectural Engineering
S. Shyam Sundar
James P. Jimirro Professor of Media Effects
Catherine Harmonosky
Associate Professor of Industrial and Manufacturing Engineering
Eleanor Dunham
Medical Director of the Department of Emergency Medicine at Penn State Health Milton S.
Hershey Medical Center
Special Member
Sez Atamturktur
Harry and Arlene Schell Professor of Architectural Engineering
Head of the Department of Architectural Engineering
*Signatures are on file in the Graduate School
iii
ABSTRACT
The US is experiencing a growing population of older adults, increasing the demand on
the healthcare system, and the Emergency Department (ED) serves as the main gateway for
inpatient admissions. With this growing demand, EDs and hospitals are expanding and building
new facilities at a growing rate. ED expansion and redesign is a complex design task which takes
into account many operational processes (current and proposed) as well a projected changes in the
system, e.g., patient volume. The effective layout of these critical departments in addition to the
workflow processes that are hospitals influence the efficiency and effectiveness of delivering
healthcare services. Yet, currently workflow processes and layout are not studied together.
Workflow processes are studied via discrete event simulation in a static layout. Layout
optimization finds an optimal layout given a static set of flow or adjacency data. The data from
both methods need to be accessible and timely in delivery for effective use in the rapid pace of
facility design.
Given the lack of integration of computational facility planning techniques in the design
and layout of healthcare facilities, new methods are needed to leverage data in the analysis
planning and design decisions in timely ways. Computational models can be used to evaluate
minimal distances or cost functions. Discrete event simulations can be used to model the
stochastic nature of operations to check the impact on specific performance measures.
Visualization can be used to immerse decision makers in the future environment to aid model
validity, communication, and understanding. In this dissertation, the three techniques are
investigated: discrete event simulation, mathematical layout optimization, and virtual
visualization. First, layout implications in a discrete event simulation of an ED are studied so as
to understand how the healthcare processes are impacted by layout decisions. Second, a layout
optimization methodology leveraging the graph theoretical approach and a placement strategy is
developed and connected to common parametric building information modeling (BIM) authoring
tools for generating layouts with distance weighted adjacency step-wise optimality. Next, the use
of generative layouts is studied with healthcare planning and design professionals. Finally, a
framework for using these techniques in an integrated hybrid simulation modeling approach in
the healthcare planning process is presented.
The results for the study of layout in discrete event simulation show that not all layout
consideration are additive. Two of five layout conditions contributed to the most amount of
improvement over the baseline condition: results waiting (15.1% improvement on all patient
length of stay - LOS) and admits zone (15.7%). A combined improvement was estimated to be
1.19 hours (23.9%) for overall LOS. The addition of fast track bays reduced the improvement by
an estimated 10 minutes. The best scenario included care initiation, results waiting, and admits
zone, and reduced overall LOS by 1.21 hours (24.3%). Study of space allocation and space
utilization found additional fast track bays were not helpful and the results waiting was
underutilized (max utilization = 7.40 people, a fifth of the seats available). Modeling the
stochastic system of an ED in the context of the layout changes can help identify what changes
contribute the most benefit, which changes are additive or compete, and help determine the space
requirements and allocations through the analysis of projected operations in that facility, but
needs operational process inputs and estimated new workflows.
A new method for generating layouts was developed based on the graph theoretical
approach for optimizing adjacency. The method uses an adjacency weighted distance score and a
generative approach to create multiple layouts for review by designers and planners by translating
space content into common parametric BIM tools. The results from the study of layout
optimization and healthcare planners and designers is that the scoring metric aligns relatively well
iv
with expert opinions, but that more advances are needed to make generative layout methods more
accepted by professionals. On average, respondents selected the ‘best’ layout marginally higher
than random chance (proportion = 29.0%, expected = 16.7%). Respondents tended to choose the
higher and lower scoring layouts, respectively: 65% of respondents selected either of the higher
two options; 48% selected either of the lower two options, out of 6 options. Respondents found
generative layouts promising for helping overcome design bias, however the current state of the
technology would need additional development. Across all respondents experience, gender, and
view on generative layouts, respondents wanted to understand the generative layout decision
details. These layouts are based on adjacency ratings, which in an automated methodology could
be updated through simulation.
A hybrid modeling framework is presented which integrates simulation, optimization,
and visualization modeling methods for healthcare facility layout planning activities for
optimizing both process and layout. Objectives are presented to create a systems approach to the
management, planning, design, construction, and operations of healthcare facilities. The main
implications of this body of work are that layout and processes are paired, are in need of greater
investigation, and an integrated approach is presented as a framework for healthcare professionals
and researchers to guide the development of an automated decision support system for healthcare
facility operations, planning, and design. These techniques, while described in a healthcare
context, have implications for other domains where uncertain and latent processes are
components of the layout decision making process.
v
TABLE OF CONTENTS
LIST OF FIGURES ................................................................................................................................... viii
LIST OF TABLES ..................................................................................................................................... xi
LIST OF EQUATIONS ............................................................................................................................. xiii
PREFACE .................................................................................................................................................. xiv
ACKNOWLEDGEMENTS ....................................................................................................................... xv
3.1 Introduction .............................................................................................................................. 45 3.2 Background Theory ................................................................................................................. 45 3.3 Research Questions .................................................................................................................. 48 3.4 Methodology ............................................................................................................................ 50
vi
3.4.1 Discrete Event Simulation Methodology ...................................................................... 51 3.4.2 Emergency Department Test Case ................................................................................ 52
3.5 Model Development ................................................................................................................ 55 3.5.1 Conceptual Model ......................................................................................................... 56 3.5.2 Emergency Department Description of Patient Flow ................................................... 56 3.5.3 Zones in the Emergency Department ............................................................................ 61 3.5.4 Changes to the Floor Plan ............................................................................................ 62 3.5.5 Changes from Conceptual Design Scheme to Final Bid Documents ............................ 64 3.5.6 Input Analysis Methodology ......................................................................................... 65 3.5.7 Model Verification Methodology .................................................................................. 74 3.5.8 Model Validation Methodology .................................................................................... 76 3.5.9 Layout Scenarios ........................................................................................................... 77 3.5.10 Output Analysis Methodology ....................................................................................... 79
3.6 Results ...................................................................................................................................... 82 3.6.1 Population Results ........................................................................................................ 82 3.6.2 Length of Stay for all Patients ...................................................................................... 83 3.6.3 Length of Stay for Discharged Patients ........................................................................ 84 3.6.4 Length of Stay for Admitted Patients ............................................................................ 88 3.6.5 Percent of Patients with LOS greater than 3 hours ...................................................... 88 3.6.6 Length of Stay by Acuity ............................................................................................... 88 3.6.7 WR Waiting Time and Number Waiting........................................................................ 92 3.6.8 Results Waiting Room Analysis .................................................................................... 96 3.6.9 Number in Admits Zone ................................................................................................ 96 3.6.10 Summary of How Layout Impacts Performance Measures........................................... 99 3.6.11 Comparison of the Best in System ................................................................................. 107 3.6.12 Opportunities for Space Allocation .............................................................................. 108 3.6.13 Future Demand Projections .......................................................................................... 109
3.7 Discussion and Conclusions .................................................................................................... 114
Chapter 4. Implementation and Evaluation of Generative Layout Options using the Graph Theoretical
Approach for a Hospital Layout Problem .................................................................................................. 117
4.1 Introduction .............................................................................................................................. 117 4.2 Research Methodology ............................................................................................................ 119
6.1 Summary .................................................................................................................................. 170 6.1.1 Integration of DES and Layout Optimization ............................................................... 172 6.1.2 Visualization of Near Best Options ............................................................................... 173 6.1.3 Implications for Industry............................................................................................... 175
6.2 Contributions to Research ........................................................................................................ 176 6.3 Limitations ............................................................................................................................... 178 6.4 Future Work ............................................................................................................................. 179
6.4.1 Implementation and Validation of the OSV Framework............................................... 180 6.4.2 Development of Software and Methodologies .............................................................. 180 6.4.3 Automation of the OSV Framework .............................................................................. 181
Appendix A. Additional Response Variables Summary Statistics from Discrete Event Simulation ........ 191
Box Plots for Length of Stay for ESI 5 Patients ................................................................................. 191 Box Plots for Length of Stay for ESI 4 Patients ................................................................................. 193 Box Plots for Length of Stay for ESI 3 Patients ................................................................................. 194 Box Plots for Length of Stay for ESI 2 Patients ................................................................................. 195 Box Plots for Length of Stay for ESI 1 Patients ................................................................................. 196
Appendix B. Survey and IRB Materials .................................................................................................... 197
Equation 4-5. Expected value for selection proportion ........................................................... 132
Equation 4-6. Expected value standard deviation from a proportion ...................................... 132
xiv
PREFACE
This dissertation has been organized around 3 main scopes of work, each corresponding
to a distinct chapter: Chapter 3, Chapter 4, and Chapter 5. These bodies of work were developed
in a cohesive manner with an overarching goal and set of objectives guiding the research
methodology. The introduction, literature review, and conclusion are presented to cover the
overarching goal and objectives in Chapter 1, Chapter 2, and Chapter 6, respectively. The
following summary of the contents is provided as a guide for readers to understand the format of
this dissertation:
Chapter 1. Introduction. The introductory chapter includes the overall goal, objectives,
scope, and general methodology for the dissertation.
Chapter 2. Literature Review. The literature review covers relevant general literature
associated with the topics covered throughout the contents of work. A targeted literature review is
summarized in each subsequent chapter.
Chapter 3. Layout Implication for an Emergency Department: Scenario Tests in a Discrete Event Simulation. This chapter pertains to the first major scope of work of the
dissertation, an investigation of the use of layout parameters within a discrete event simulation for
a redesign of an emergency department as a base case.
Chapter 4. Implementation and Evaluation of Generative Layout Options using the
Graph Theoretical Approach for a Hospital Layout Problem. This chapter, the second major
scope of work, presents the development and evaluation of a generative layout procedure
leveraging the graph theoretical approach for providing optimal arrangements of departments
based on adjacency.
Chapter 5. Framework for a Hybrid Simulation Approach for an Integrated Decision
Support System in Healthcare Facilities. This chapter presents the third and last major scope of
work in this dissertation, the development of an optimization-simulation-visualization framework
for use throughout a healthcare facilities lifecycle from planning through operations and redesign.
Chapter 6. Conclusions. This chapter summarizes the major findings and implications of
the previous chapters, describes the general limitations, provides the next steps for future work,
and ends with concluding thoughts.
References. All references cited throughout the dissertation are available in one reference
section.
Appendix A. Additional Discrete Event Simulation Summary Statistics on Response
Variables. Additional data associated with the discrete event simulation output analysis from
Chapter 3 is available in this appendix.
Appendix B. Survey and IRB Materials. The survey procedure, survey apparatus, and IRB
materials used for conducting research on evaluation of the generated layouts associated with
Chapter 5 are documented in this appendix.
xv
ACKNOWLEDGEMENTS
There are many people who have helped me throughout my work on my doctoral degree. First
off, I would like to thank my dissertation advisor, John Messner, without whom I would not have
started this work, and especially his guidance, willingness to meet and discuss research, and the
freedom he gave me to pursue an interesting and complex topic. Likewise, I would like to thank all of
my committee members, Dr. Rob Leicht, Dr. S. Shyam Sundar, Dr. Catherine Harmonosky, and Dr.
Eleanor Dunham, for all their time, their various perspectives, and their invaluable feedback which
helped me to pursue a rigorous interdisciplinary study. I would like express my extreme gratitude to
all participants who volunteered their time in any parts of this study. A huge thank you to both the
Pennsylvania State University Office of Physical Plant and the Department of Emergency Medicine at
the Hershey Medical Center at Penn State Health for their support in the development of this research,
and to HKS Inc. for their interest and support. Without these collaborations, I wouldn’t have been able
to complete this work.
I would like to give an immeasurable thank you to the professionals who have helped me
informally and formally throughout this dissertation research process including and not limited to: Dr.
Susan Promes, Kain Robbins, Deb Medley, Katie Deitrick, Michael Baron, Catherine Brower, David
Barto, Todd Alwine, Paul Seale, Josh Adams, Dr. Katie Kasmire, Jon Huddy, Virginia Minolli,
Michael Klinepeter, Davide Rodney, Tim Shuey, Kate Renner, Tim Logan, Frank Kittredge, Heath
May, Monish Sarkar, and Shannon Kraus. I would like to extend special appreciation to Robert Amor,
Tom Boothby, Bill Bahnfleth, Cindy Reed, Gretchen Macht, and Jason DeGraw, for their help with
research, idea iterations, additional research advice and guidance, and general support throughout the
years at the Pennsylvania State University.
I couldn’t have completed my work with the support of all the AE grad students, past and
present, especially the CIC research group for their openness in sharing their research, experience, and
for planning AE happy hours. Namely, I would like to thank my fellow and past colleagues Fadi
There are many types of computational simulations used in the study of how a system
operates. They can include discrete event simulation, continuous simulation, systems dynamics,
Monte Carlo simulation, agent based simulation, and 3D/virtual reality simulations (Kuljis et al.
2007). Simulations are meant to represent a simplified version of the real world, and in design, a
version of the future or proposed world. For this study, the focus is on discrete event simulation
as a quantitative method to test scenarios in a future facility (Gibson 2007). Deterministic models
such as facility layout problems are a common type of quantitative method in facility planning
(Tompkins et al. 2010). An introduction to the literature in discrete event simulation and facility
layout optimization is reviewed. Additionally, since healthcare problems are focused on people
and need high stakeholder involvement (Robinson 2002), literature on experienced based design
using virtual prototyping is presented to support these data-driven methods. The literature
presented first covers the general and emergency department (ED) performance measures
important to facility owners and administrators. Second, the healthcare design process is
introduced and the application areas of discrete event simulation (DES), facility layout
optimization, and visualization in healthcare design delivery reviewed. Then, details on the
development of facility layout optimization, DES, and visualization for facility design is
presented. Finally, a discussion of the related nature of these methodologies is discussed as well
as why researchers and practitioners might want to integrate them, followed by a discussion of
what is known and needs to be investigated in order to integrate these techniques in design
practices.
11
2.1 Healthcare Performance Metrics
In healthcare, there are many outcome and performance metrics used to understand
healthcare delivery quality. The Agency for Healthcare Research and Quality develops and
establishes quality metrics for healthcare, they have documented 2006 clinical quality measures
and 138 related healthcare delivery measures (AHQR 2017). The Centers for Medicare and
Medicaid Services (CMS) makes data publicly available on all hospitals that provide Medicare
services across the US. Creating standard metrics for understanding healthcare delivery quality
and patient care outcomes is a large initiative which helps managers assess performance by
providing data to benchmark their performance across the country.
Recently, patient experience has become a more important metric for assessing overall
patient care quality. Patient experience is reported from a 32-item questionnaire (development
managed by Hospital Consumer Assessment of Healthcare Providers and Systems - HCAHPS).
HCAHPS standardized the healthcare assessment process for patient experience and has been
implemented throughout hospitals in the US since 2008. The patient experience survey does not
mean patient satisfaction, as the construct for patient experience contains information about
communication with doctors and nurses, cleanliness, perceived satisfaction, as well overall
hospital rating. The assessment methods for patient satisfaction are complex. The US government
makes the HCAHPS and additional data, over 100 total measures, available to the general public
for all hospitals accepting Medicare in the US. Approximately 60 of their measures are used to
generate an overall hospital rating for prospective patients to compare service and quality. Of
those 100+ measures, eight focus on timely and effective care in an ED. These eight are listed in
Table 2-1.
12
Table 2-1:Timely & effective care, emergency department throughput (CMS 2017) Measure
Identifier Technical Measure Title Measure (From on Hospital Compare) Units
Update
Frequency
EDV Emergency department volume Emergency department volume
Patients annually,
categorical: <20K,
20K-40K, 40K-60K,
60K
Annually
December
ED-1b
Median time from emergency
department arrival to emergency
department departure for admitted
emergency department patients
Average (median) time patients spent in the
emergency department, before they were
admitted to the hospital as an inpatient
minutes
Quarterly (April,
July, October,
December)
ED-2b
Admit decision time to emergency
department departure time for admitted
patient
Average (median) time patients spent in the
emergency department, after the doctor decided
to admit them as an inpatient before leaving the
emergency department for their inpatient room
minutes
Quarterly (April,
July, October,
December)
OP-18b
Median time from emergency
department arrival to emergency
department departure for discharged
emergency department patients
Average (median) time patients spent in the
emergency department before leaving from the
visit
minutes
Quarterly (April,
July, October,
December)
OP-20 Door to diagnostic evaluation by a
qualified medical professional
Average (median) time patients spent in the
emergency department before they were seen by a
healthcare professional
minutes
Quarterly (April,
July, October,
December)
OP-21 Median time to pain medication for long
bone fractures
Average (median) time patients who came to the
emergency department with broken bones had to
wait before getting pain medication
minutes
Quarterly (April,
July, October,
December)
OP-22 Patient left without being seen Percentage of patients who left the emergency
department before being seen %
Annually
December
OP-23
Head CT scan results for acute ischemic
stroke or hemorrhagic stroke who
received head CT scan interpretation
within 45 minutes of arrival
Percentage of patients who came to the
emergency department with stroke symptoms
who received brain scan results within 45 minutes
of arrival
%
Quarterly (April,
July, October,
December)
13
Of these general healthcare outcomes, several are related to environment and space.
Organized by three major categories: patient safety, other patient outcomes, and staff outcomes,
Ulrich et al. (2008) summarized the literature results indications of the impacts of 11 environment
and spatial factors on these three major categories of outcomes, separated into 16 specific metrics
(Table 2-2). Some of these outcomes relate to the general statistics gathered by CMS and publicly
reported statistics such as length of stay, communication with patients and family members, and
hospital-acquired infections and patient satisfaction. Design layout strategies such as single-bed
rooms, nursing floor layout, acuity-adaptable rooms, and decentralized materials have been linked
to several patient and staff outcomes. Single-bed rooms has been shown to impact the most
number of healthcare outcomes. All four of these layout strategies have been shown to impact
staff effectiveness. Other design strategies which are not layout specific, such as appropriate
lighting, views of nature, ceiling lifts, and others have also been shown to positively impact
various healthcare outcomes both for patient and staff focused metrics. Equipment can be added,
processes can be changed, lighting fixtures can be replaced, but many of these design strategies
cannot be changed after a hospital is designed and built without considerable added costs and
delays in service. There are competing theories for how design practices impact patient care
outcomes. Most of the design strategies presented in Table 2-2 are associated with inpatient
healthcare outcomes, and not timeliness of care patient outcomes. In an ED, timeliness of care is
important, but all patients need to be treated safely and effectively. There is a need for more
research to understand how design strategies specific to ED care impact key healthcare outcomes.
In ED healthcare outcomes, the timeliness of care is the most common outcome metric
used. Some of these time dependent measures are used as indicators for general performance
metrics, such as how busy staff are (e.g., staff utilization) for staff effectiveness and staff stress;
and a patient leaving before being seen for patient satisfaction. While there may be multiple
reasons why a staff member is stressed or why a patient decided to leave before seeing a doctor,
14
these time-dependent measures are relatively easy to quantify and provide an indication that there
is something not right in the system. If a staff member is too busy, such that they are subjected to
an overly stressful work environment, they cannot provide the best care, leading to healthcare
mistakes, leaving to find a new hospital, or even leaving the industry. If an ED finds that the
percentage of patients who leave an ED before being seen is increasing, they might want
investigate some of the potential causes of that, such as how long people wait before being seen.
These all lead to an analysis of patient flow and time dependent measures to evaluate the
effectiveness of the hospital to address the volume of patients they expect. The current ED
averages by annual patient volume are shown in Table 2-3. In general, as a facility treats more
patients, they have longer averages across the timeliness of care metrics. Other factors that are
important to consider in benchmarking hospital and ED timeliness of care are demographics of
patient population, capacity of beds in the hospital, time of year, insurance rates in patient
population, and trauma level of the hospital for both adult and pediatric care.
Typical healthcare outcome measures used in discrete event simulation of EDs are
percentage of patients leaving without being seen, length of stay for patients, staff utilization,
resource utilization (especially for critical equipment such as MRI, CT, X-Ray), and waiting time.
Waiting times, leaving without being seen, and length of stay are all patient metrics and can be
organized by level of acuity. A common acuity index is the Emergency Severity Index (ESI)
between 1 and 5, indicating the amount of resources needed from nurses and doctors, see Figure
2-1. Acuity 1 patients are the most critical patients, indicating high acuity, and acuity 5 patients
are the least, low acuity, and need the least amount of resources (e.g., tests or procedures).
15
Table 2-2: Summary relationships between design factors and healthcare outcomes (Ulrich
et al. 2008)
Notes: * indicated that a relationship between the specific design factor and healthcare outcome
was indicated, directly or indirectly, by empirical studies reviewed; ** indicated that there was especially strong evidence (converging findings from multiple rigorous studies) indicating that a
design intervention improves a healthcare outcome.
Sin
gle
-bed
ro
om
s
Access
to d
ayli
gh
t
Ap
prop
ria
te l
igh
tin
g
Vie
ws
of
natu
re
Fa
mil
y z
on
e i
n p
ati
en
t roo
ms
Ca
rp
eti
ng
No
ise
-red
ucin
g f
inis
hes
Ceil
ing
lif
ts
Nu
rsi
ng
flo
or l
ayo
ut
Decen
tra
lized
su
pp
lies
Acu
ity
-ad
ap
tab
le r
oom
s
Reduced hospital-acquired infections **
Reduced medical errors * * * *
Reduced patient falls * * * * * *
Reduced pain * * ** *
Improved patient sleep ** * * *
Reduced patient stress * * * ** * **
Reduced depression ** ** * *
Reduced length of stay * * * *
Improved patient privacy and confidentiality ** * *
Improved communication with patients & family members ** * *
Improved social support * * *
Increased patient satisfaction ** * * * * * *
Decreased staff injuries ** *
Decreased staff stress * * * * *
Increased staff effectiveness * * * * * *
increased staff satisfaction * * * * *
Design Strategies or
Environmental Interventions
Healthcare Outcomes
16
Table 2-3: National averages for emergency department healthcare outcomes (CMS 2017), gray cells indicate average is across all
emergency department volumes
Technical Measure Title
Data Collection
Period National Average (2015) Units
Emergency department volume 1/1/15 12/31/15 Low
(<20K)
Med
(20K-
40K))
High
(40K-
60K)
Very High
(60K+)
Volume
Category
Median time from emergency department arrival to
emergency department departure for admitted emergency
department patients
4/1/15 3/31/16 210 258 295 338 minutes
Admit decision time to emergency department departure time
for admitted patient 4/1/15 3/31/16 58 88 115 134 minutes
Median time from emergency department arrival to
emergency department departure for discharged emergency
department patients
4/1/15 3/31/16 113 141 160 172 minutes
Door to diagnostic evaluation by a qualified medical
professional 4/1/15 3/31/16 18 23 27 30 minutes
Median time to pain medication for long bone fractures 4/1/15 3/31/16 52 minutes
Patient left without being seen 1/1/15 12/31/15 2% %
Head CT scan results for acute ischemic stroke or
hemorrhagic stroke who received head CT scan
interpretation within 45 minutes of arrival
4/1/15 3/31/16 69% %
17
Figure 2-1: Emergency Severity Index conceptual algorithm (Gilboy et al. 2011)
2.2 Healthcare Design Process
In general the design process is broken into five typical stages: planning,
conceptualization, design, construction, and operations (Gould 2012). Design is typically
separated into three stages: schematic design, design development, and construction documents.
During planning, the project requirements are developed and feasibility studies are completed.
During conceptual design, the program requirements are further developed. In schematic design,
the layout and space requirements are developed and several options are typically compared.
During design development, one design is developed in more detail. During construction
documents, the final design is detailed for construction. After the design phase, construction of
the design takes place. During operations, the facility is occupied and operated until renovations
or new facilities are needed, and the design cycle begins again. In more integrated projects, where
Patient Dying?
Shouldn’t Wait?
How Many Resources?
Vital Signs
1
2
45
3
yes
yes
consider
no
no
no
A
B
C
D
18
the design team includes all the design and delivery practitioners, these phases usually are not as
distinct. Especially in healthcare renovation projects, where it is important to keep facilities
operational during construction, integrated design and delivery approaches are typically used,
such as design-build. Design-build is characterized by a sole design and construction team with
one contract with the owner (Gould 2012). The single contract gives the design team more power
to complete the design and construction of the project. This can streamline changes during the
design and construction communication process. Under this project delivery structure it is
typically necessary for an owner to have a specific understanding of their design scope early on to
provide adequate guidance to the design-build team.
2.2.1 Discrete Event Simulation in Healthcare
Simulation has been used within the healthcare application domain for the last 35 years
(Brailsford and Vissers 2011; Günal and Pidd 2010). Several reviews of how simulations and
other operations research methods have been used in healthcare domain have been performed
over the years (For early reviews see Wilson 1981, for more modern reviews see Brailsford et al.
2009; Brailsford and Vissers 2011; Fone et al. 2003; Günal and Pidd 2010; Jun et al. 1999; Rais
and Viana 2011). Most of the focus of simulation in healthcare design has been in the operations
of hospital units or departments (Brailsford and Vissers 2011; Günal and Pidd 2010). In
categorizing simulation research in healthcare by implementation areas, Brailsford and Vissers
(2011) identified Region/National, Unit and Hospital Operations, and Patient and Provider
Operations as three separate simulation areas. They found in their survey of research that 43% of
articles focused on the unit and hospital operations, 33% on region and national level, and 25%
on the patient/provider level. The most popular states for operations research in healthcare setting
have been for managing the performance of delivery (39%), for developing programs and plans
(24%), and in evaluating the performance for delivery (18%) (Brailsford and Vissers 2011). The
19
use of simulation has increased in healthcare over the course of the last 35 years (Günal and Pidd
2010).
There are many simulation techniques, and several researchers have talked about the
process for identifying the most appropriate simulation model methods for healthcare settings
(Bhattacharjee and Ray 2014). It is important to identify the goals and objectives of a simulation
technique before determining the method (McGuire 1998). Discrete event simulation has been
used when the focus is on performance of operations and when the performance measures are
quantifiable and measurable. Performance measures include waiting time of patients, congestion
measures, utilization of resources (e.g. equipment, nurses, doctors, beds), length of stay, and cost
assigned measure (Bhattacharjee and Ray 2014). When using discrete event simulation for
healthcare systems, it is important to understand the process and problem being addressed. When
working with a design for a new facility, using baseline data of an existing facility can be helpful
for comparison in decision support for the facility (McGuire 1998).
The role of simulation in healthcare design fits into Robinson's (2002) continuum of
simulation modeling approaches (Figure 2-2) on the side of “simulation as a process of social
change”. In this case, simulation is a tool used developed by a sole modeler for a project to
understand the performance of a system and the input parameters which change the way a
hospital operates (whether that be to increase patient satisfaction or to operate more efficiently).
In these types of simulations, a high degree of stakeholder involvement is needed not only to
understand the system and develop a correct model of its operations, but also to increase
awareness and educate the stakeholders.
20
Figure 2-2: Continuum of approaches to simulation modeling (Robinson 2002, p. 3)
Oh et al. (2016) presented research on the development, validation, and scenario building
of a discrete event simulation for improvement of a large hospital emergency department. The
goal of the study was to investigate impact on length of stay (LOS) for patients, by reducing LOS
from 44% of patients staying less than 3 hours to 80% of patients staying less than 3 hours. They
included the following entities: patients, blood test samples, radiology tests, and patient
registration paperwork. The model included ED staffing, department layout, and patient flow
logic. The key performance indicators were average LOS in each station, waiting times in each
station, number of concurrent patients, and leaving without being seen (LWBS). When evaluating
the model, several target areas were identified as potential areas of improvement without large
capital investment or disruption to service. The authors identified eight target areas: main pod
configuration, main pod bed allocation, radiology turnaround, lab sample re-collection, main pod
nurse staffing, pediatric MD staffing, physician availability, and inpatient bed turnaround time.
From those suggested target areas, 5 different implementation scenarios were built for simulation
experimentation, combining various implementation changes. Some changes included balancing
pod bed allocation, adjustments in staffing, and decreasing time to discharge. The fifth scenario
included the most changes and was found through experimentation to reduce LOS to the desired
levels of 80% of patients stay below 3 hours. The LRH ED implemented the fifth scenario for 5
months. They found their LOS met their goal and the simulation results within a 95% confidence
Simulation as
Software Engineering
Accurate representationLengthy projectsTeam of modelersLow user-involvement
Simulation as
Process of Social Change
Problem understanding & solvingShort projectsLone modeler
High customer-involvement
21
level. They also found a reduction in LWBS, from 2.8% to 0.3%. When compared to similar
sized ED national averages, their LOS and LWBS metrics were found to be below those
averages. Oh et al. (2016) presented a complex model, with a relatively simple experimental
design, which was later implemented and tested to see if the changes met the specified
performance metric goal. Although the example focuses on process changes to an existing
emergency department, the development strategy is useful for understanding how various design
scenarios can be modeled, tested and implemented in an ED context.
Batarseh et al. (2013) presented a method for using system modeling language (SysML)
for incorporating knowledge transfer from stakeholders and aid in validation and verification of
highly granular discrete event simulation. They present a methodology for process integration in a
real-world emergency department by comparing hourly census information with model
simulation results before and after a process intervention. The authors developed 22 activity
diagrams of the processes used in providing care in the emergency department (ED) at Anderson
Emergency Center. They used these SysML activity diagrams to document, develop, validate, and
verify the model logic. The authors compared the hourly patient census within the different
locations of the ED and the daily turnaround times in triage from the simulation to actual data.
After validating and verifying the base model, a change was implemented in the simulation prior
to recommendation to the Anderson Emergency Center for an ED process change. The changes
included the addition of a new pod and staffing changes for room coverage. They used SysML to
document the changes and aid simulation model development. Even though the work was not
implemented (or implementation results were not published), the work presented using SysML to
streamline the logic transfer and validation process potentially can allow for automatic logic
transfer into a discrete event simulation of an ED.
22
2.2.2 Optimization in Healthcare Design
In facility planning, there are location optimization and layout optimization problems.
The first formulation of a hospital layout optimization problem was as a quadradic assignment
problem (QAP) (Elshafei 1977). Current applications of facility layout problems (FLPs) for
optimization have included mixed integer programming (MIP) combining continuous and discrete
variables, as well as meta-heuristic optimization strategies including: simulated annealing, tabu
search, and particle swarm optimization (PSO) (Vahdat et al. 2019). (Arnolds and Nickel 2015)
review layout planning problems in healthcare application setting. These problems are typically
NP-hard (Anjos and Vieira 2017), making them computationally expensive thus the use of
heuristics methods are common to reach approximate optimization in relatively short periods of
time. Recent advances in computing power have made these methods more available as
computational expense is reduced, yet are not commonly used in practice despite research and
development of different formulations of layout problems over the last 40 years.
2.2.3 Visualization in Healthcare Design
Visualization of virtual content associated with planning and design has been a identified
as a method to engage users and aid communication among disparate project team members
(Bassanino et al. 2014; van den Berg et al. 2017; Garcia et al. 2015), thus aiding an experienced
based design methodology. There have been several studies in virtual prototyping of healthcare
designs, for a selection of examples: a cancer ward scenario walkthrough (Kumar 2013), hospital
patient rooms/patient rooms (Dunston et al. 2007; Wahlström et al. 2010), and community
pharmacies (Leicht et al. 2010; Mobach 2008), healthcare facilities (Dunston et al. 2010; Kumar
et al. 2011), and healthcare environments (Zhang et al. 2011). These studies found virtual
23
prototyping a useful tool for communicating and gaining design feedback from end users
throughout the design and delivery process.
Kumar (2013) developed an experienced based design virtual prototyping framework and
tested it with healthcare professionals in two settings: a flexible walkthrough scenario and a
structured tasked based scenario. The findings from this research were that the structured task
based scenario provided more in-depth design feedback from healthcare professionals.
Bullinger et al. (2010) described the use of immersive virtual prototypes for use in a user-
centered approach to architectural design and planning. They found the use of prototypes with
end users throughout the design process was able to increase the quality and performance of
building design and construction process.
2.3 Discrete Event Simulation
There has been extensive research on the topic of discrete event simulation in operations
research and in healthcare applications. Most of the research has focused on simulation of
operation processes, without explicitly describing the design or layout implications with in the
DES. The main goal of using discrete event simulation in healthcare is for process improvement
(McGuire 1998; Robinson 2002). It is used to simulate behavior of a system over time with a
defined process. In a healthcare setting, this usually involves a combination of observation,
review of available data, and interviews to determine service time assumptions for various tasks.
Not all tasks are known, and are subject to simulation error (unknown/inappropriate distribution)
or reporting bias.
A discrete event simulation is a simulation of random events constrained by expected
distributions and means or probabilities of occurrence. A DES is defined as a dynamic, stochastic,
and discrete model of a real-world system (Banks et al. 2010). Dynamic simulation models model
24
a system as the system changes over time, as opposed to static systems which evaluate a system
as a specific point in time. Stochastic, as opposed to deterministic, simulation models contain
random variables such as random arrival rates and random service times. A stochastic simulation
model has one or more random inputs, which in turn leads to random outputs. These random
outputs are response variables defined in the objectives of a simulation study, e.g., performance
measures, are estimates of the actual, real-world, system. Discrete events simulation models
model a system through discrete sets of events, e.g., arrival to system, move to triage, leave
triage, etc., by creating a schedule of events in the simulation environment and then updating the
model based on updates to entity states.
Some aspects of using discrete event simulation in healthcare (or service) industries,
which is different than other non-service related industries (e.g., part manufacturing, assembly
line, warehouse, etc.), is the balance between the efficiency outcomes (e.g., startup costs and
return on investment), and the softer patient satisfaction outcomes (e.g., quality of care metrics)
(McGuire 1998). Thus, it is important to investigate several simulation outcomes before analysts
can make appropriate recommendations. Additionally, success of discrete event simulation, such
as in implementation of recommended changes, in healthcare is highly impacted by stakeholder
expectations and the development timeline. Yet even when the simulation fails to be
implemented, some benefits are found, for instance in using the simulation as a communication
tool throughout the development process (for an example see Bowers et al. 2009).
Simulations studies have a clear step by step development to implementation process.
Figure 2-3 presents a common workflow from Banks et al. (2010, p. 15). Work starts with
problem formulation (usually with simulation modeler and stakeholders) and development of
objectives and overall project plan. It’s best to keep each simulation study small, by having up to
three objectives and breaking up a large simulation into smaller units to keep these objectives
manageable (McGuire 1998). Next analysts collect data and create a model conceptualization.
25
Data collection can involve observations, interviews, review of available data. Model
conceptualization typically starts with flowcharts for each entity type (e.g., patients). Once the
logic of how the system works (in a new facility design, how the new components of the system
function) and the data needed to model that system is obtained the logic can be translated into the
model by the simulation modeler. The next step for a modeler is to verify that the model is
representing the system correctly by checking for errors in the logic and performance. If not, the
model is modified. There are several methods for this step, including comparing with real life
data and changing parameters to see if the model responds appropriately. The next step is to
validate the model. This step ensures that the flowcharts and data used in the model actually
represent the real-world system and is usually done with similar steps to the verification, but can
include stakeholders and experts with the system to ensure the correct real-world logic is in the
model and the system represents the real world. If a change is made to the model, the verification
and validation processes start again until both are satisfied. Once the model is verified and
validated, a design of experiment is developed for each scenario to be simulated, with specified
length of simulation, number of runs, and the need for initialization periods (for steady state
systems). After some production runs and analysis, the simulation modeler determines if there is a
need for more runs depending on the analytical techniques used and the initial analysis. If no
more runs are needed, documentation and reporting the results to stakeholders is the next step.
And finally, implementation of the recommendations from the simulation study in the real world.
The following sections describe the problem formulation, objectives and simulation plan,
model conceptualization, data collection, model translation, verification, validation, experimental
design, production runs and analysis, and finally documentation and reporting.
26
Figure 2-3: Typical steps and flow of a simulation process (Banks et al. 2010, p. 15)
27
2.3.1 Problem Formulation
The goal of the simulation is to evaluate whether design options meet the operational
performance measures of an emergency department. Problem formulation is defined in the scope
portion of this proposal. The problem formulation for a discrete event simulation is more specific
to evaluate design options based on the performance metrics defined. For more information on
key performance metrics of healthcare systems and emergency departments see Section 2.1.
2.3.2 Setting Objectives and Simulation Plan
The simulation objectives are the questions which a simulation analysis can answer.
These are separate from the research objectives. They are typically to select the best alternative
from a set for implementation in the real world. Common rules for defining objectives of a
simulation study are: (1) they should be quantifiable, (2) the number should be limited to three or
less, so as to focus and limit the scope of the simulation modeling efforts, (3) the scope should be
narrow, so as to address the objectives, (4) the results should be useful to the end users and routed
in changes that are implementable, e.g., the modeler shouldn’t suggest a change that is
unreasonable for the budget of the project, (5) the modeling project needs to meet the time
requirements of the owner, and (6) the size of the modeling project is inversely correlated to its
successful implementation, e.g., simple models typically are easier to produce meaningful results
in a timely manner to help owners make decisions (McGuire 1998).
28
2.3.3 Model Conceptualization
Conceptualization is the process for developing the model logic and work flows needed
in the model. In this step it is critical to develop the abstract essential features of the simulation
model in order to find “useful approximation results” (Banks et al. 2010). The start of a model
conceptualization begins with a simple model of the emergency department and more complexity
is added as needed to find meaningful results in the specified performance measures. This step is
an iterative step involving development of patient workflows, defining resources, and processes
which entities (e.g., patients) move through the system. This step was done in tandem with data
collection to ensure that the correct data was used in the model.
2.3.4 Data Collection
There are several types of data needed for an emergency department simulation: arrival
data, patient categories, activities and flowcharts (also used for model conceptualization),
resources (e.g., staff, beds, equipment, depending on model detail), service times for activities,
and distributions. Data must be from a representative point in time. For new designs, data from
similar activities can be used in lieu of historical data (McGuire 1998). The data collected is not
only used to build the model, but also to verify the model, so data associated with historical
performance measures is needed as well for model verification.
Arrival data is needed to understand how often patients arrive. Arrival data can be
stratified by patient type. Depending on the arrival data, patient arrivals can be modeled as a
single entry with probabilities of different conditions representing changing the patient type or
several patient arrivals can be used for different entity flow paths (arrival by ambulance vs. walk-
in arrivals). Initial data analysis is be used to understand arrival patterns and to understand how to
translate the data into a stochastic model of arrival patterns through time of day, patient acuity,
29
arrival mode analysis of distributions and probabilities. Patient categories can be type of patient,
pediatric or adult, acuity of patient (by Emergency Severity Index - ESI), and by mode of arrival
(ambulance or walk-in). The categories used in the simulation model should match as closely a
possible to the categories used by healthcare practitioners and represent the real-life situation.
Activities and patient flowcharts are used for each patient type (or entity) to be modeled.
Information is needed for how many patients receive different types of interventions or follow
various branches in the flowchart. Information about tests and conditions which must be
performed or must be found before moving through the next step in the flowchart need to be
defined. The resources needed for each task in the flowchart and the service time for performing
the task are also needed, depending on the level of detail of the simulation. Using existing
flowcharts is the best, however some translation into simulation modeling flowcharts is
necessary. SysLM has been used to translate patient flowcharts into emergency department
simulation models (Batarseh et al. 2013). Flowcharts can be modeled in various manual ways and
increasingly there are examples of formats to create automatic translation into model
environments, however those have limitations still in standardization and ease of use.
Resource data needed for the models include staff schedules, bed/pod staffing plans,
conditional routing of resources depending on time of day, and time to complete tasks. Staff
schedules show how many nurses or doctors are planned to be available at a given time. Staffing
beds or pods of beds might change throughout the day and during certain times of day routing of
different conditions to different pods may occur. This type of process information is needed to
represent the system logic. The time it takes to perform a task usually follows a lognormal,
Weibull, or gamma distribution (Law and Kelton 1991). These distributions include delays that
may occur when completing a task and an average time to complete a task with an average
amount of training. Uniform distribution can be used to model complete uncertainty between a
range of values. When limited information is known about the actual processing time, triangular
30
distribution, defined by three parameters (mean, minimum, and maximum service time), is
typically a better estimation than using a uniform distribution, defined by minimum and
maximum parameters (Banks et al. 2010).
2.3.5 Model Translation
In this step, the model conceptualization and the data are translated into a simulation
model. There are many software packages available for model translation including Arena, Simio,
ProModel (MedModel), Simul8, Witness, FlexSim (HC), and AnyLogic. Some packages are
medical system specific, such as MedModel, and some are general simulation software (e.g.,
Arena, Simio, AnyLogic, Simulink). Simio software was used for model translation based on it
meeting all the needs of the research: generic model translation, robust pseudo-random number
generator, easy process/flowchart translation, and researcher’s knowledge.
2.3.6 Verification
After the model has been translated, it must be verified that the model logic is represented
accurately, based on the model conceptualization and the data used to generate the model. The
goal of this step is to ensure that the simulation model is performing as expected. This step is
closely related to debugging the model by the model creator. Typically for complex systems, a
simplified version of each part of the model is created and debugged before being integrated into
a more complex model. After integration occurs, the model is again verified to ensure correct
model translation. If the model is found to have errors, corrections are made in the model.
Techniques used include replacing random times with constant values, processing one entity
through the system, making one replication and investigate the reasonableness of the output, and
changing some parameters to see if the model responds correctly.
31
2.3.7 Validation
Similar to the verification process, validation checks to see if the model accurately
represents the real-world. The techniques used are similar to the verification process. It is
common for validation to occur with comparing the model output to the real-world output (e.g.,
compare length of stay in the ED actual vs. simulation model). During this step, it is helpful to
engage people familiar with the system to check the validity of the model. This can be done by
reviewing model flow logic, reviewing output analysis, and by using an animation of the model.
If the model is found to not work as the real-world system, changes are made to either the model
conceptualization or the input parameters, which may include more data or new data collection
activities.
2.3.8 Experimental Design
After the model is verified and validated, then system alternatives can be modeled. In this
step, the alternatives include the various design layout options. In addition, a sensitivity analysis
approach can be used to investigate a range of rates used for either processes or arrival to
investigate the impact of design changes under various conditions. The design layout options
include changes to room configuration (number of beds, private rooms vs. open bed spaces) and
distances between stations for patients and healthcare practitioners. Arrival rates can be varied
based on projected population mix and demand changes. Patient flow processes can be changed
to test different processing configurations using different scenarios, (for example, one with a
patient centered flow, one with a nurse centered flow, and one with a mix between the two). It is
important in the development of the experimental design to model the system such that the
decision variables of interested can be manipulated in the model environment. If the number of
decision variables are reasonable, a fully crossed experimental design can be made. If there are an
32
unreasonable number of combinations, such as for large numbers of decision variables, a Latin
hypercube sampling strategy can be used to develop the experimental design (Duan et al. 2017).
2.3.9 Production Runs and Analysis
When there are K system designs, one can use a selection of the best methodology to
select the best system on a specific performance measure. The procedure used is from Banks et al.
(2010). The steps involved in selection of the best procedure is to first specify the desired
probability of correct selection, set a practical significance difference, and specify the initial
number of runs for each system design. Next, an initial number of simulations are performed and
an initial screening of the scenarios is completed based on the performance measure of interest.
The scenarios that are significantly different are eliminated. Then, additional replications of each
near best scenario are run until either a stopping condition for number of replications is met or a
best scenario is found based on the practical difference and significance levels. For more detail
associated with the selection of the best methodology, see Section 3.5.10.
2.3.10 Optimization within Simulation
One goal of a simulation might be optimization. Often in analysis of simulation or set of
simulations, people are interested in what is the best outcome in an experimental design.
Simulation on its own is not an optimization method. For a simulation to be used for
optimization, several design or procedural inputs are modeled in separate simulation scenarios,
and a performance metric is identified and analyzed as the output of these models. The expected
value of the performance metric is analyzed from the various sets of runs of a simulation
scenarios. It is difficult to analyze multiple performance metrics in conjunction with each other
and methods exist for taking one at a time and then investigating the impact of the optimized
33
value of the first metric and then seeing the impact on the second metric. Another approach is to
combine several metrics into one indicator. A third approach is to use thresholds for secondary
metrics and optimize the main metric. All approaches involve picking the key metrics for
evaluation purposes (Banks et al. 2010).
In the situation where the process is known, and a new design is desired, a simulation
might focus on optimizing the combination of space, layout, and resources around the specific
process. Typically, the process is not necessarily perfect, and simulation of the process reveals
areas where operations and service can be improved or be made more efficient.
2.3.11 Documentation and Reporting
After analysis is completed and the objectives of developing a simulation are met, they
need to be conveyed and documented to the future simulation analysts and stakeholders. A report
can be made on the program and on the progress. The program report explains the model and how
the model was built and gives instructions on how to use the model in order to allow others to
make changes to the model in the future. The progress report can be done in stages, the most
common being the final report, where the recommendations are passed to stakeholders for review
and final decision making on the proposed system changes. It is becoming more common practice
to include animations of the simulation in the progress stages of reporting for communication
purposes, as more often researchers describe using animation as a validation strategy (for a few
examples see Batarseh et al. 2013; Vahdat et al. 2019). In these reports, all assumptions, model
specifications, prior model stages and deliverables, program documentation, as well final results
should be clearly documented.
34
2.4 Facility Planning and Layout Optimization
In general, facility planning is the process of planning the location and layout of a
facility, or a set of facilities, and is the overarching process concerned with the strategy of the
management, design, and construction, and eventual reconstruction of the facility(ies) until they
are torn down (Tompkins et al. 2010). While typically used by industrial engineers, facility
optimization software exists but is not typically connected design software used by architectural
designers and engineers (Malmborg 1994). These methods include both manual and computerized
methodologies. A major component of facility planning is facility location and layout
optimization (Tompkins et al. 2010).
Facility layout problems (FLPs) are a class of optimization problems similar to facility
location optimization problems. For a review of the application of different algorithm approaches
see Liggett (2000) and for a recent review of mathematical approaches to these type of problems
see Anjos and Vieira (2017). Research into facility layout has traditionally been in the
manufacturing and industrial sectors (Das 1993; Francis et al. 1992), but additional application
areas have been studied, including in airport terminals (Edwards 2004; Manataki and Zografos
2009), train and railway stations (Li 2000), shipyards and ports (Bruzzone and Signorile 1998),
retail stores (Levy et al. 2014), and healthcare facilities (Arnolds and Gartner 2018; Holst 2015;
Vahdat et al. 2019), where human variation plays a major role in operational performance.
Facility layout problems have been commonly studied through deterministic optimization
problem heuristics which take into account flow information (Francis et al. 1992), while
operation research methods, such as discrete event simulation, use stochastic methods to
approximate the random variation that humans and processes bring to the system (Banks et al.
2010). Supplementing deterministic layout optimization techniques with the total flow path of
people can provide designs based on user experience (for a healthcare clinic example see Vahdat
35
et al. 2019) and, ultimately, a layout design that performs well under a robust set of conditions
(Acar et al. 2009).
Methods in layout optimization for healthcare has been researched and developed over
the last 40 years, beginning with the formulation of the problem as a QAP (Elshafei 1977). These
tools are not typically used in planning new facilities outside of the manufacturing setting. The
software available for healthcare planners and designers do not use these methods, as quantitative
methodologies for layout design are still not widely known. Planners typically use manual
practices including “rules of thumb” and personal experience to arrange and recommend layout
choices to solve owner problems (Arnolds and Nickel 2015). With advancements in technology,
data generation, and computing power, data-driven methods are becoming computationally less
expensive, allowing researchers, engineers, and planners to test an increasing number of “what-
if” scenarios to provide an analysis of future facility performance. These are especially useful for
problems with large human impact and capital costs, where analysis can provide data-driven
results to help inform layout decisions.
2.5 Virtual Prototyping and Visualization
Experienced based design and virtual prototyping are relatively new research areas used
in the architectural, engineering, and construction domains, with literature on the subject more
common over the last 15 years. “Experienced based design in healthcare is design that focuses on
end-user and staff experiences in a facility to identify creative design solutions” (Kumar 2013).
Virtual prototyping is a user-centered design approach which borrows from a broader product
development and human-computer interaction discipline. Virtual prototyping, as a process for
developing digital prototypes, has been an effective method for supporting the evaluation of
alternatives in the product development cycle (Rudd et al. 1996). When researchers investigate
36
the building design as a product which provides a service, or many services as the case may be,
virtual prototyping is a key tool for rapidly iterating through and evaluating design alternatives by
receiving targeted feedback from the product customer, end-user, and stakeholders.
While few have described the virtual prototyping process for building design, many have
discussed its importance in the product development and software development domains. There
isn’t an agreed upon procedure, however it is suggested that a systematic and iterative
prototyping procedure be followed by practitioners (Rudd et al. 1996). A procedure was
developed for experienced based virtual prototyping by Kumar (2013) specific to incorporating
scenarios in healthcare building design (Figure 2-4). In this case scenarios are targeted tasks
defined during goal setting. The steps are similar to the operations simulation steps described in
Section 2.3, with some differences noted. The procedure starts with defining the goals and
objectives of using the prototype, the stakeholders who should be engaged in the process, and the
tasks and users needed for the analysis. The next step is to develop the framework for scenarios
including the model content required and the features required (e.g., hospital unit and equipment
needed for specific task). The third step is to develop the design of the visualization system by
storyboarding how users will use the system and the graphical user interface required to support
the scenario (e.g., what will users touch and do within the prototype, how will they move through
the system, how will that fit into the context of the goals of the prototype, etc.). The next step is to
develop design information workflows to incorporate interactivity in the virtual prototypes (e.g.,
navigation, task scenarios, graphical user interface, and interactive objects). Once a prototype is
developed, the process embedded in the prototype needs to be validated with experts of the
process being represented, and finally implementation within the design process occurs.
The key differences between the virtual prototyping procedure and the simulation
procedure are the development activities required. These steps in the virtual prototyping
procedure might be interpreted as the model conceptualization and translation activities in the
37
simulation steps. There is a more rigorous verification and validation process depicted in the
simulation steps process and a more defined data collection step. Data collection might be an
important addition to the experienced-based virtual prototyping steps depending on the goal and
scenario of interest. Both start with clear engagement with stakeholders by defining goals and
objectives. Building Information Modeling (BIM) data and models are typically used for
gathering, generating, analyzing, communication, and realizing information associated with the
design and construction of facilities (Kreider and Messner 2013). Virtual prototyping is a BIM
tool for analysis and communication tasks.
The goal of the visualization system is to aid stakeholder engagement in design decisions
for the emergency department design. In discussing the role of end users of healthcare facilities to
aid design, Bate and Robert (2006) suggested changing the perspective of end user engagement
from a passive role to a more active role in the design process. They describe the design process
for ExBD as a co-design process where: users and design professionals work together over time;
the focus is on user experiences as opposed to views, attitudes, and perceptions; the focus of
designing experiences is on the subjective pathway and not the objective pathway; users and
design professionals use the process to find deeper understanding; and interpretation includes the
interaction of usability, service, safety, and functionality.
38
Figure 2-4: Experienced-based virtual prototyping steps (Kumar 2013, p. 103)
39
2.6 Integrating Simulation, Optimization, and Visualization
The following examples are some of the published literature on projects which have
combined simulation, optimization, and visualization. Many of these studies focus on the
technological integration aspect of the integrations and on real-time integration of data between
these two systems.
2.6.1 Crane Mobilization
ElNimr et al. (2016) described the A* path finding algorithm integration with discrete
event simulation for crane mobilization planning. They used a spatial analysis component for
planning crane utilization on a construction site where cranes were placed in the planning
sequence based on the path finding algorithm. The example shows the use of spatial path finding
in a simulation of event sequences where the next sequence of the construction site layout using a
two-way communication mechanism between the spatial and event simulation components.
2.6.2 Stroboscope and Vitascope
Rekapalli and Martinez (2011) describe a case study in real-time integration of discrete
event simulations with virtual environments in construction sequence planning. They discuss how
interactivity can improve the model validation in simulation studies. The real-time interactivity
was achieved so that construction engineers can study the model’s response to a simulated
earthmoving operation. The use of real-time linkage to virtual environments was posed as a
capability which enhances the model validation process for use of simulation in construction
planning and design. They used STROBOSCOPE (Martinez 1996) for the simulation engine and
VITASCOPE (Kamat and Martinez 2004) for the visualization engine. Validation of the model
focused on a specific portion of the earthmoving operations, specifically haul truck breakdown on
40
a one-way section of the route. The case study highlighted the need for the visualization to be
targeted to certain areas of the model logic (such as routing errors). The visualization system used
was a proof of concept application of animation and virtual prototyping for simulation model
validation, and didn’t include any user testing studies to test impact on the model validation
process.
2.6.3 Traffic Simulations
In a transportation operations simulation, Chen and Huang (2013) proposed a new system
for 3D animation integration with STROBOSCOPE. The model can be built in 3D space instead
of solely schematic diagrams. They used an augmented reality component to place 3D model
components while viewing the real-world site. The study focused on model conceptualization and
translation. They investigated the effectiveness of their system by surveying 32 transportation
simulation graduate level students who used 4 other simulation platforms which are used for
visualization of DES (STROBOSCOPE, EZStrobe, Vita2D, and VITASCOPE) for each
platform’s pre- and post-processing effectiveness. The asked participants to rate the levels of
intuitiveness, interactivity, reality, ease, prediction, and integration. Some of these platforms are
used in pre-processing and some are used in post-processing. In addition, the results showed
higher average scores for the various metrics but the standard deviation and test for statistical
differences are missing from the study. The results indicate that the proposed new system was
high on integration in comparison with EZStrobe, and similar to VITASCOPE on intuitiveness,
reality, and precision.
41
2.6.4 Manufacturing Applications (VR Factory)
In a simulation education context, the VR Factory was developed and presented (Kelsick
and Vance 1998). Later the VR Factory was further developed with the discrete event simulation
software SLAM II within a six sided CAVETM virtual environment for simulation education of a
manufacturing cell (Kelsick et al. 2003). The manufacturing cell could have several simulations
loaded into the virtual environment which students could explore through a simple user interface.
Direct integration with the model creation process was not implemented (e.g., users couldn’t
change the simulation parameters, such as number of stations or routing), but they could move
through time within the simulation selected and navigate through the model freely to view the
manufacturing cell. The researchers suggest using immersive virtual visualization to study
movement of parts to further understand the design implications of various systems. They suggest
this tool as a “computational steering aid” for improved decisions in a simulation analysis.
However, they did not present an evaluation of decision making impacts in the research.
2.6.5 Integration in Healthcare
Simulation methods have been used in business, automobile, manufacturing, and
construction industries (Kuljis et al. 2007). The simulation methods included in the survey were
discrete event simulation, continuous simulation, systems dynamics, Monte Carlo simulation,
agent based simulation, and 3D and virtual reality simulations. The added constraints of human
actors in healthcare simulation make the methods more difficult (Kuljis et al. 2007). Kuljis et al.
(2001) described the combination of visualization and simulation in a clinical practice, using a
visual simulation called CLINSIM. They found both users and analysts benefit from the
integration of the temporal simulation and virtual world simulation. In discussion on these two
simulation techniques, Kuljis et al. (2001) describe some of the fundamental differences of the
42
two approaches where temporal simulation has the inherent goal of constructing the real world
into a controlled experiment whereby experimentation and impact can be studied and
visualization simulation has the inherent goal of gaining insights from patterns found in the visual
representations. They suggest the integration of these approaches can allow users to focus on
salient patterns otherwise unnoticed by simulation analysts which would (1) strengthen the
understanding of the process and contributing factors and (2) extend the scope of simulation
beyond current practices to incorporate latent processes. The move from salient to latent
processes expands the traditional realm of simulation of systems to engage in the tacit knowledge
of the users of the modeled systems. This deeper understanding can aid simulation models by
allowing them to be better designed, understood, and accepted. In the setting of healthcare design,
or any system which is highly impacted by the human participants, the acquisition and usage of
the latent processes and tacit knowledge is key to appropriate modeling and representation.
In hybrid simulation literature, it is common to focus on the integration of various
simulation types, such as incorporating DES and continuous simulation or DES and agent based
simulation (Djanatliev and Meier 2016). A hybrid modeling approach in contrast is broader
concept connecting different sets of data and tools together to create a larger decision making
framework utilizing a systems level thinking (Mustafee and Powell 2018). The connection
between layout optimization approaches and discrete event simulation approaches have been
investigated in a few settings to develop methods in developing robust layouts under uncertain
workflow practices (Acar et al. 2009; Arnolds and Nickel 2015). For more detail on simulation-
optimization approaches in healthcare see Chapter 5. For integration taxonomy examples see
Figueira and Almada-Lobo (2014) for simulation-optimization, and (Shneiderman 1996) for
information visualization.
43
2.7 How Has Research Suggested Integration of These Methods?
Within simulation, validation has been the most common area cited for the integration of
these techniques (Rekapalli and Martinez 2007, 2011). In addition, model validation and model
implementation have been identified by many simulation experts as some of the key problems in
the use of discrete event simulation in healthcare (Brailsford and Vissers 2011; Günal and Pidd
2010; McGuire 1998; Wilson 1981). In a review of facility layout optimization applications,
Liggett (2000) identified the major items missing in facility layout applications: a transparent
access to the rules and procedures in the algorithm and a connection between the optimal layout
presented and common software used in the building disciplines. Some people have suggested
animation (Chwif et al. 2015) and interactivity as a means to aid model acceptance, increase
stakeholder engagement, and improve usability of discrete event simulations in manufacturing
contexts. Waller and Ladbrook (2002) described the purpose of integration of simulation and
visualization as useful during early design for layout decisions as a tool for communication.
These assertions, while currently anecdotal, present the potential for the use of discrete event
simulation, optimization, and visualization to work cohesively together to create a data-driven
healthcare design decision making framework.
While there are limitations on the types of problems that can be solved in a reasonable
amount of time for simulation and optimization problems, given many facility layout problems
are NP-hard, the computational complexity can be reduced through heuristic methodologies.
Applications that develop the connection of layout optimization strategies and common practices
of healthcare planning and design professionals should consider the incorporation of modern
interactive interfaces and links to both building information models and facilities management
databases (Liggett 2000). In addition, adding the use of simulation (especially to simulate
44
expected and projected future expected processes) is a natural extension for usage throughout a
cohesive integrated decision making framework for both operational and design decision making.
The expectations for visually interactive models is increasing, and with that, simulation is
being used more widely by non-simulation experts (Robinson 2005). The increased use of
simulation by non-simulation experts could mean that new computational methods are needed to
analyze and communicate simulation and optimization logic more effectively.
The literature has shown that the integration of these techniques, simulation,
optimization, and visualization, in a few settings, however they have not been applied in practice
frequently to healthcare operations or healthcare design. Moreover, even though the link between
DES and layout optimization is theoretically discussed in some literature, it is not clear how these
two can best be leveraged in the design review process given the amount of time and the data
needed to develop models and run analysis. Both automated methods and visualization for
communication and stakeholder engagement could be leveraged to aid the use of these data-
driven methods in the design and operations of healthcare facilities.
45
Chapter 3.
Layout Implication for an Emergency Department: Scenario Tests in a
Discrete Event Simulation
3.1 Introduction
Over half of the inpatient admissions in the US begin with emergency department (ED)
visits (AHQR 2017). Additionally, the growth in number of ED visits from 2006 to 2014 have
outpaced population growth in the US, with ED visits increasing 14.8% and population increasing
6.9% (Healthcare Cost and Utilization Project 2017). The ED is the main way that inpatients
enter a facility and trends show an increasing ED utilization rate. EDs are the first line of critical
care service and gate keepers of the overall care paths for most patients. However these systems
are not static, variation in human patterns play a key role in ED patient length of stay. Yet, the
layout is, once built, a static resource. Decisions to change the layout are made based on current
trends in design thinking and theory, and not typically based in data-driven analysis methods to
understand the workflow in the context of a new space. The goal of this research is to understand
how layout considerations impact operational performance measures in an emergency
department, where timely care is of concern.
3.2 Background Theory
One method commonly used to understand the stochastic operational system of EDs is
discrete event simulation (DES), which simulates the events and resources in a system (Fone et
46
al. 2003). However, in practice these methods are not commonly used in planning a new facility.
Recently DES software have started incorporating static layout (Taylor et al. 2013) and birds-eye
view animation (Kelton et al. 2014). Planners still typically use manual, rule of thumb, and/or
expert experience to develop and recommend layout options to solve facility owner problems
(Arnolds and Nickel 2015). Yet, there is little research in the layout-process interaction for
healthcare projects.
Research has shown that DES in healthcare has challenges with model acceptance and
implementation (Günal and Pidd 2010; McGuire 1998). One area researchers have suggested for
utilizing DES in combination with other visualization tools is during the schematic design of a
new or renovated system (Gibson 2007). To build off these assertions, this study focuses on
initial research into using DES during “what-if” scenario testing of layout options during
schematic design of an ED expansion to understand how layout impacts workflow processes.
Departments are the typical scale for simulation in healthcare facilities. Other scales to
consider are the human scale (micro-level) with agent based simulation or internal processes
system dynamics simulation and the Hospital scale (macro), with the input and output of
departments taken into account, the Health System scale, with the demographics and system
dynamics of the population input and outputs taken into account (Djanatliev and Meier 2016),
and possibly a national or global scale, with the whole population taken into account. Arnolds and
Nickel (2015) documented simulation and optimization on a hospital scale and a departmental
scale. In discussing the use of hybrid simulation in hospital processes, Djanatliev and Meier
(2016) described four scales: the human individual, the human biological processes, the
departmental, and the health system. These scales provide different contexts for analysis. At the
departmental level, certain assumptions are made about the extent of the system in the simulation.
As a definable work unit with definable boundaries, the department-level system presents an
easily definable part of the hospital system for analysis and evaluation.
47
Many researchers have explored discrete event simulation in an emergency department
focusing on evaluation and recommendations for a process change in operations (for example see
Batarseh et al. 2013; Oh et al. 2016). An example of a design problem in a family practice clinic
and using cost metrics was presented by Swisher and Jacobson (2002). Their work presents a
DES model for design decision support in an outpatient clinic. The focus was to determine an
optimal number of medical assistants, physician assistants, and nurse practitioners. They
developed several metrics for performance evaluation by developing a cost model including
negative cost impacts of patient satisfaction, negative cost impacts of decreases in staff
satisfaction, and clinic profit. These metrics and how they were deployed still need some
evidence to support them, but is a good first step to combining various output parameters for
optimization purposes. They used a static floor plan option for evaluation in this example.
In developing a plan for using design layout options in a hospital design and planning
process, Gibson (2007) presents the major goals of using simulation in the planning, master
planning, and schematic design phases. During planning brief, the role of simulation is to study
the clinician’s paths and focuses on output analysis of staffing plans and department locations.
During master planning, path distances between department locations are studied and optimized
based on space requirements (for example, shared reception areas for departments). During
schematic design, the simulation provides an avenue for testing “what-if” scenarios in layout and
design. While Gibson doesn’t present specific evidence and solely proposes a system for design-
via-simulation, this represents some of the research community’s perspective on how simulation
and design can be integrated for evidence-based approaches.
In an initial study into the impact of spatial characteristics on nurses’ productivity rates,
Choudhary et al. (2010) developed empirical models which showed that spatial properties
impacted frequency of trips made by nurses’ in a multi-unit hospital. The study used the level of
room assignment as opposed to unit to study the impact of spatial orientations. The model was
48
found to have a predictive power on frequency of trips, indicating that path layout options
potentially should be addressed at a fine enough level of detail to understand the implications of
design and layout plans.
From these examples, a range of the level of detail and analysis are used in discrete event
simulation for healthcare applications. Typically, design and layout decisions are not well
addressed, if at all, in DESs. However, these spatial considerations potentially have a large impact
on how people will operate their facilities. In the design and renovation of emergency
departments, performance measures are typically based in time, such as length of stay of patients,
thus makes it a good example test case for modeling layout considerations and testing the impact
of layout on performance metrics in an DES environment.
3.3 Research Questions
Using an ED expansion test case for development of this study, the layout implications of
a DES were studied in detail. The ED project, described in more detail in Section 3.4.1, is a large
volume Trauma I facility with congestion problems. The main goals for the facility were to
reduce the average length of stay (LOS) for all patients, especially low acuity (Emergency
Severity Index - ESI 4&5) patients who do not need more than 2 resources or services. They were
identified as the patients who could easily be assessed and discharged because they do not need to
wait for an inpatient bed (a current problem in the ED).
Overall LOS is the average time for all patients who arrive in the ED. Average
discharged LOS is an average time in system for all patients who were discharged from the ED.
Average admitted patients LOS is an average time in system for all patients who were admitted as
a patient to the hospital. Another metric, percent of LOS greater than 3 hours, is a risk
measurement for all patients who have a stay in the ED longer than a specified level, in this case
49
3 hours. These performance metrics are the main metrics of interest for hospital administrators,
but additional metrics exist. Time to provider (the time from a patient entering the ED to being
seen by a doctor), time to roomed in the ED (the time from a patient entering the ED to being
roomed in the ED), and percent of time to room less than 30 minutes were all identified as
important. Time to room were relatively good compared to national averages (13.7 minutes and
94.5% in a room before 30 minutes in 2017), and thus were not the main focus of this study.
Additionally time to provider data was not tracked, but identified at something to track in the
future, so no baseline data existed. The LOS performance metrics were selected as the critical
performance metrics of interest for simulation study. The mean performance of the 2017 baseline
system are presented with their relative performance goals in Table 3-1.
Table 3-1. Performance metrics of interest for ED case study including
selection of the best goals.
Metric 2017 FY Data Unit Goal
Average LOS overall 5.33 Hours Lower is better
Average LOS for discharged patients 4.41 Hours Lower is better
Average LOS for admitted patients 8.14 Hours Lower is better
Percent of LOS > 3 hours 67.73% Percent Lower is better
Given these performance goals, the study focused on understanding how layout impacted
performance, analysis of space allocation, and analysis of performance under future demand. For
the study, the following four research questions were developed:
RQ1a: How does layout impact performance measures?
RQ1b: Which layout is the best?
RQ2: Were there opportunities to optimize space allocation?
RQ3: How does the layout perform under different demand scenarios?
50
3.4 Methodology
To answer the research questions, a methodology was developed to use an existing ED as
a test case. A DES was used to model the workflow processes and layout changes and obtain
estimates for performance measures. An experimental design was developed to operationalize the
key layout changes in the test case, so that each layout change could be investigated individually
and in combination to test if significant improvements on performance metrics were found. The
individual and relative contributions of each layout factor on changes to the performance
measures of interest, while keeping all other factors in the facility design constant, combine to
answer RQ1a, how layout impacts performance measures in an example case of an expansion ED
project. Next, a selection of the best strategy was deployed to select the best of these layout
scenarios, to answer RQ1b. Finding a scenario that is best, or a set of scenarios that are near
enough to each other that they are not discernable, would indicate that some layout choices
complete with one another, in other words, implementing all layout changes does not create the
best result, with best being defined as improvements in the performance measures of interest.
Next, analysis of the DES output of resources utilization on specific measures associated with the
space allocation were investigated, to understand if sizing of the key layout conditions were
appropriately utilized. The results from this analysis are expected to explain if these were
appropriately sized, and if there were opportunities to change and optimize the space
programming and space allocation in the design scenarios, RQ2. Finally, the layout conditions
were tested under future demand projections to help answer RQ3. Below the test case scenario,
DES methodology, and analysis procedures are described in greater detail.
The following sections describe the general discrete event simulation modeling
methodology and the test case ED. Additional details on the methodology are described in model
development, Section 3.5.
51
3.4.1 Discrete Event Simulation Methodology
The DES development methodology follows the modeling methodology outlined by
Banks et al. (2010), including goals, conceptual model, input data collection, model development,
model validation, model verification, and what-if scenario testing of changing parameters of
interest, in this case decision variables associated with layout configuration changes. The work
started with problem formulation (with ED expansion stakeholders) and development of
objectives and overall project plan. The main three objectives were identified and used to direct
the model building activities (McGuire 1998). Data collection was performed to collect data and
create a model conceptualization. Data collection involved observations of ED expansion
workflow review sessions, semi-structured interviews with nurses and doctors, and the review of
available deidentified patient data. Model conceptualization started with flowcharts for the
patients in the system based on the input data. Once the logic of how the system works (or will
work as is the case with a new facility design) and the data needed to model the services and
events in that system were obtained, the logic and data was translated into the model. The next
step was to verify that the model represented the system correctly by checking for errors in the
logic and performance and reviewing assumptions in the model. The model went through several
iterations in the verification including comparison with baseline 2017 FY data and changing
parameters to check that the model responds appropriately. The next step is to validate the model
with healthcare practitioners. This step ensures that the flowcharts and data used in the model
actually represent the real-world system. This was done through review of the conceptual model
with those who understand the baseline system to ensure the correct real-world logic is in the
model and the system represents the real world. Once the model was verified and validated to
enough detail to answer the research questions, a design of experiment was developed for each
scenario to be simulated. A specified length of simulation, number of runs, and the need for
52
initialization periods were selected to warm up the model since the ED runs continuously and is a
steady state system. A confidence interval was selected at 95%.
3.4.2 Emergency Department Test Case
The emergency department is going through a multi-phase redesign and expansion to
help alleviate the problems of overcrowding and disconnect between front of house and back of
house operations. Front of house operations includes the entrances, waiting room, and intake
processes for patients. Back of house operations include the treatment areas, separated into zones
within the emergency department, as well as out-of-room services such as CT scanner and X-Ray
Radiology, as well as many others.
The following details the current state of the Hershey Medical Center Emergency
Department (HMCED) and the master plan for renovation and expansion. HMCED is a trauma 1
emergency center. In the 2015 fiscal year (FY, defined as July 2014 - June 2015), the ED served
72,493 patient and, in the 2017 FY, the ED served 76,020 patients. The ED is expected to
increase in patient volume to serve 95,000 annual patients by 2021 and 110,000 annual patients
by 2026. The main operational challenges of the existing ED include the congested front end
configuration, attending physicians are located far away from the ED front of house, considerable
waiting time in treatment rooms especially for low acuity patients (ESI 4 & 5), excessive
movement of nurses back and forth from front to back of ED to care for patients, lack of visibility
between tending nurses and treatment rooms, and the ED being at or over capacity routinely for
extended periods of time. As part of their master plan to achieve the capacity and improve the
functions of the ED in the near future, PSHMC has a 4-phase expansion and renovation plan
developed by Huddy HealthCare Solutions. The goals for the HMCED redesign were to add
additional capacity, improve front entry for patient access, and improve flow and efficiency for
patient treatment. The goals included both operational changes and layout changes. A planning
53
and conceptual design report was developed by Huddy HealthCare Solutions including
computational simulation of proposed operational changes for each of the four phases. The report
was given to PSHMC in September 2016 (Huddy et al. 2016).
The existing conditions and expansion zones are shown in Figure 3-1. The first phase is
mainly an expansion of the emergency department into the current adjacent cancer ward garden
and out to the ED drop-off access area. See the completed phase 1 conceptual design in Figure
3-2. The second through fourth phases are not to be completed until a later date, depending on
funding and future needs. Work in the first phase is planned in two sequences, first is expansion
and second is renovation in existing areas to decrease impact on operations, especially the
number of patient beds available during construction. In schematic design, the design team had
three layout options for the emergency department, specifically layouts for the ambulance
entrance, the patient (walk-in) entrance, a results waiting room for low acuity patients, and the
extended hall treatment are under consideration. During reviews of these options, questions have
been raised on how nurses will move through the spaces, what the implications are of these
specific design layouts on future changes to operational processes, and tradeoff considerations of
adding other types of spaces not previously investigated in the simulation and conceptual design
in 2016 (HMCED design review meeting, May 16th, 2017). During this process there was
disagreement and problems in creating a common vision for the future envisioned operational
processes in the new design. Since the redesign was focused around solving operational problems
of flow and capacity, the questions raised by the nurses and doctors in HMCED indicate a desire
to have a review process that includes new simulation tests of these layout implications and
assumptions.
54
Figure 3-1: Existing conditions and expansion diagram (Huddy et al. 2016, p.12).
Existing
Entrances
55
Figure 3-2: Conceptual configuration of Phase 1 (Huddy et al. 2016, p.15).
Note: Bright yellow indicates current patient rooms. Light yellow indicates expansion areas. Right, expansion and renovation areas with new entrances marked with red arrows. Upper left,
12 bed Clinical Decision Unit (CDU) not in scope, separate project.
3.5 Model Development
The following sections describe the model development details including the conceptual
model of the ED, patient flow in the ED, design changes during schematic design, input data
analysis (e.g., arrival rates, service times, walking speeds, decision and response variables),
model verification and validation, layout scenario development, and output analysis.
Ambulance
Entrance
Patient
Entrance
56
3.5.1 Conceptual Model
A conceptual model was developed through reviewing workflow notes from a nurse and
doctor workshop on patient flow in the ED conducted during the summer of 2016 and updated
through semi-structured interviews with healthcare practitioners who were familiar with the
typical practices of the 2017 FY workflows. A patient flow diagram was created for the current
practices, from the baseline 2017 FY, (Figure 3-7) and the future practices planned after
occupation of the new layout (Figure 3-8).
The main patient flow changes include the additional space for patients waiting in the
walk in entrance (capacity change), the use of results waiting room (RWR) area for low acuity
patients who have been seen and are only waiting for results from labs and/or minor services
(flow change, see yellow highlighted in Figure 3-8), including potentially those who need X-
Rays, but who do not need the privacy of a bay room, the addition of a separate zone for those
waiting for in-patient beds for stable patients (Admits zone, location change), and the change in
in-take process in the triage area with a doctor who can begin orders for labs and other services
(CIA – Care Initiation Area, flow change), instead of the typical nurse triage. These changes can
be summarized as two main flow changes, one main capacity change, and one location change.
All these changes have capacity and location decisions, but those that changed the flow chart
were considered flow changes. Since Admits zone was using a space already existing in the ED, it
was considered a location change. The waiting room (WR) at the walk-in entrance is mainly a
capacity change, but also a location change.
3.5.2 Emergency Department Description of Patient Flow
A patient can arrive by ambulance or by ‘walk-in’ (car, cab, bus, walk-in, etc.). If a
patient arrives by ambulance, they are brought through a separate ambulance entrance. The
57
current state configuration has the ambulance entrance very close to the patient walk-in entrance.
The general workflow for patients roughly follows a typical patient workflow of arrival,
registration, triage, evaluation, with possible tests, diagnosis, treatment, including possible
medication or procedures, and discharge, which might include admitting the patient for additional
treatment (Figure 3-3). When leaving the ED, a patient is either discharged or admitted. Typically
these activities are performed in sequence. When a high acuity patient arrives (ESI 1 & 2), they
may have diagnosis and treatment done in parallel. Additionally, triage might be performed
before the patient arrives and registration can occur independent of all other activities.
Figure 3-3. Typical emergency department overview workflow
3.5.2.1 Ambulance Entrance
Prior to arrival, the EMTs call ahead to assign a room. Upon arrival, the EMTs park
outside the entrance and bring patient in through the ambulance entrance. If the patient is in
critical condition, they are brought to the area where care can be administered directly. Spaces
include resuscitation room, Cath lab, and obstetrics. If they are not in critical condition, they are
assigned a room and brought to that space upon arrival. If there is no bed/room available, a lower
acuity patient may be moved from their room to accommodate the ambulance arrival
58
3.5.2.2 Walk-in Entrance
Patients who arrive via walk-in first check-in with the registration desk. If the patient is
emergent, they are immediately brought back to resuscitation room or the space they need to be
treated. They are immediately seen by a physician. Emergent patients will never be taken to the
Physician Directed Queuing zone (PDQ, for more detail on zones see Section 3.5.3). Once seen
by nurse and doctor, emergent patients are stabilized, treated, and possibly given lab work. If in
the resuscitation room, they are then roomed in the ED. The PDQ area is reserved for low acuity
patients (ESI 4 & 5) and is open during the day.
If the patient is not emergent, they are quick registered (sometimes full registered). Quick
registration is the collection of the minimum number of patient identifiable data to start an
electronic medical record (EMR) for the visit. The patient waits in the WR until they are called to
triage. During Triage, the nurse assigns an acuity ESI level to the patient.
If the patient complains of chest pain, they are immediately taken to have an EKG. This
can be the case if the patient has shortness of breath (SOB), irregular heartbeat, palpitations,
syncope (fainting), etc. The results of an EKG need to be given to the doctor within 20 minutes of
the test. If the EKG is not normal, the patient is seen by a doctor, stabilized and treated before
getting ‘roomed in the ED’. If the EKG turns out normal, the patient may be returned to the WR
until their turn in the queue comes. After triage, a patient waits to be called back for a room.
Higher priority patients are taken first and jump the queue.
3.5.2.3 Roomed in ED
Once roomed in the ED, a patient is seen by a nurse and a doctor. These can happen
simultaneously or separately. A doctor will order labs, X-Rays, treatment, procedures, etc. For
labs medication EKG and procedures, these typically occur in the room. Radiology is in the ED
for X-Rays and CT. For an MRI the patient will need to leave the ED for testing. If the patient
59
was quick registered, they will be fully registered in the room. This might begin with the doctor
or might coincide with a doctor visit. After any labs or imaging, a patient will wait for results in
their room. Once results are in, the doctor reviews them, diagnoses the patient and then comes
into the room to provide the diagnosis. If any treatment is needed, this may occur before during or
after diagnosis depending.
3.5.2.4 Discharge
If a patient is discharged, they will exit the ED and the hospital. They may visit the
pharmacy at the hospital (although it is not available 24 hours). Some discharged patients will be
discharged to a home or a rehab unit. Since the hospital is a Trauma 1 facility for both adult and
pediatric, a trauma doctor is available 24h/day on site for adult patients. For pediatric trauma
patients, a pediatric trauma doctor is available on-call to arrive within 20 minutes. Some patients
will be kept for observation.
If a patient was emergent or in critical condition, it is very unlikely that they will be
discharged from the hospital. All patients that will be admitted to the hospital will be in their
room until they are transferred to the inpatient hospital unit. If the patient will be admitted to the
hospital, the doctor will page the relevant unit, order consult, bed recruitment, and logistics. Once
the admit unit is ready for them and there is a bed available, the patient will be transferred by
hospital staff. Burn patients are transferred to another specialty center.
3.5.2.5 Room Zone Routing
Different patients are routed to different areas in the ED. In the current state there are six
zones: PDQ, Red, White, Cobalt, Grey, and Pediatrics. The hours of operation for each zone,
number of beds, and primary patient pool are listed in Table 3-2. An example of room routing for
a high acuity ambulance patient (ESI 1 & 2), an ESI 3 patient, or a low acuity patient (ESI 4 & 5)
60
during daytime operations is shown in Figure 3-4. Typically all services for a high acuity patient
will occur at the room location, whereas a medium acuity patient (ESI 3) typically is registered
and triaged near the WR area before they are roomed. A low acuity patient might not need
extensive tests or treatment, and will only need evaluation to occur in room. Most pediatric
patients will be routed to the pediatrics zone, however those with high acuity will typically be
roomed in the same zone as adult patients.
Table 3-2. Room totals by zone, current
Zone Treatment
Beds
Procedure
Rooms Hours Target ESI Routing
White 10 1 24 hr 1,2
Red 10 1 24 hr 1,2
Cobalt 6 1 8am - 12am 3
Grey 7 1 8am - 12am 3
PDQ 4 1 8am - 12am 4,5
Pediatrics 11 1 24 hr 1-5, <18 yr old
Current Totals 48 6
Figure 3-4. Overview of typical acuity routing for ED patients
61
3.5.3 Zones in the Emergency Department
During the day, the front of house has a triage nurse triage patients from the WR. Because
the WR was designed using a different operational model than one currently being used, patients
waiting for triage have to wait in an area away from Triage nurses’ direct view (See cyan areas in
Figure 3-5).
The back of house operations are separated into different zone which have different
acuity routing. The typical operations in 2017 FY used the zones White, Red, Cobalt, Blue,
Pediatrics, and Physician Directed Queuing (PDQ) (Figure 3-5). The White, Red, and Pediatrics
zones are run 24 hours/day. Acute patients (ESI 1 & 2) are routed to the White and Red zones.
Cobalt and Grey receive mid-level acuity patients (ESI 3) and PDQ receives ESI 4 and 5 patients.
These three zones are operational between the hours of 8am and midnight, with shifts typically
ending around 2am given time to finish care in these zones. All pediatric patients are sent to the
Pediatric zone. If they are full and patient needs immediate care, they are sent to acute zones,
White or Red. Pediatric acute patients are routed to the White and Red zones if they arrive after
the dedicated pediatric doctor shift hours (at night between midnight and 8am). Each zone has its
own staffing of nurses. An APC (Certified Assistant Physician) oversees care in the Gray and
Cobalt areas. When able to fully staff the nurses in the emergency department, the PDQ zone is
used for low acuity patients. When the ED is run without full nursing staff, the PDQ rooms are
not utilized.
62
Figure 3-5. Current room configuration with zones
3.5.4 Changes to the Floor Plan
In the new expanded floor plan, a specific area will be dedicated for patients that are
waiting to be admitted to the Hospital. The ED has a boarding problem, where, in times of high
volume, the demand for in-patient beds exceeds the room and bed resources available. An area
for admits has been informally created in the ED to help operations, where patients wait to be
admitted to the hospital. The current Grey area is planned to house the new Admits zone (Figure
3-6). This will free up space mainly in the White and Red zones (new Acute 1 & 2), where more
63
patients are admitted and more complex care is needed. Additionally, the PDQ zone is being
redesigned as an 8 bay Fast-Track (FT) patient bay zone, which will provide 4 additional patient
treatment bays (see purple rooms in Figure 3-6). These bays need less area than patient treatment
rooms, have a cloth opening, and have patient recliner chairs instead of fully reclined treatment
beds. Adjacent to the FT zone is a zone for Mid-Track (MT), a new location for the previous
Grey zone. All MT patients will be in the patient beds shown in blue in Figure 3-6. In addition, a
dedicated SANE (Sexual Assault Nurse Examiner) patient room and consult bay will be added, as
well as a redesigned decontamination bay, with adjacent isolation room. An additional isolation
room has been designed to be located in the new MT patient zone. The new number of beds, zone
names, and their associated previous zone based on patient treatment, are shown in Table 3-3.
Table 3-3. Redesign room totals by zone, future plan
Future Zone
Assoc.
Previous
Zone
Treatment
Beds/Bays
Procedure
Room Hours Target ESI Routing
Acute 1 White 10 1 24 hr 1,2
Acute 2 Red 10 1 24 hr 1,2
Admits n/a 7 1 8am - 12am 2,3
MT 1 Cobalt 6 1 8am - 12am 3
MT 2 Grey 6 1 8am - 12am 3
FT PDQ 8 0 8am - 12am 4,5
Pediatrics Pediatrics 12 1 24 hr 1-5, <18 yr old
Future Totals 59 6
64
Figure 3-6. Future room configuration with zones
3.5.5 Changes from Conceptual Design Scheme to Final Bid Documents
The design scope changed as added detail was added to the design documents. During the
concept phase, little detail was known about the existing conditions or about the user
requirements. Through design review sessions with administrators, nurses, and doctors a revised
plan was created that included additional necessary requirements to sustain the facilities ability to
be ready for a wide range of emergency care services, such as isolation rooms, SANE exam
room, and updated decontamination room. The seat changes in the WR increased from 28 in the
concept to 41 in the final solution. The seat changes in the RWR increased from a total of 31 to
65
36 in the final solution. A summary of room and seat counts are summarized with the early
concept numbers, three design options from schematic design, and the final solution (Table 3-4).
3.5.6 Input Analysis Methodology
The input modeling of the ED involves analysis of random variables which includes
arrival rates, demand projects, service times, routing probabilities, as well as travel speeds for
different entity types. Additionally, the model assumptions, decision variables, and response
variables are described in the input analysis methodology.
3.5.6.1 Arrival Rates
Arrival rates were analyzed by counts per hour using one year of data from 2017 FY.
Some variation could be explained by day of the week and month of the year in terms of different
acuity level arrival rates, however most of the variation was due to time of day. A non-stationary
Poisson distribution based on hourly rates was the best representation of the variation in the data.
The count of arrivals by hour were tested against the Poisson distribution using a Chi-squared test
for goodness of fit and no significant differences were found, supporting the hypothesis that the
arrivals follow a non-stationary Poisson distribution arrival pattern. This arrival pattern was used
to randomly generate patients in the simulation model.
Future arrival rates were assumed to be an increase of 5.8% percent yearly of the past
data arrival rates, so as to meet the expected total increase of approximately 19,000 additional
patients per year, and increase of approximately 4,400 patients per year for the next 4 years. The
increase in demand arrival pattern was calculated by multiplying each hourly rate by the yearly
increase.
66
Table 3-4. Summary of room and seat changes from concept to construction documents
Rooms Concept Option 1 Option 2 Option 3 Final
Solution
Exam Bays 10 10 10 10 8
Exam Rooms 7 7 6 7 7
Ante Room n/a 1 1 1 1
Isolation Room 1 1 2 2 2
SANE Exam Room n/a 1 1 1 1
Decontamination 1 1 1 1 1
Radiology Room 1 1 1 1 n/a
Bereavement n/a 1 1 1 1
Consult 3 3 3 4 1
Care Initiation 3 3 3 3 4
Equipment Supply n/a 1 1 1 1
EMS Holding 3 n/a n/a n/a n/a
Communication
Services n/a 1 1 1 1
Procedure Room 1 1 1 1 1
Patient Toilet 3 3 3 3 3
Public Toilet 2 2 2 1 1
Staff Toilet 1 2 2 2 2
Meds 1 2 2 2 3
Clean n/a 2 3 3 2
Soiled 1 1 1 1 2
Seats: WR 28 26 26 36 41
Seats: RWR (Sub
Waiting) 29 (3) 40 (6) 33 32 36
67
Figure 3-7. Conceptual model for patient flow in the current layout
68
Figure 3-8. Conceptual model for patient flow in the future layout
Note: Yellow highlighted processes are changed flow from current to future
69
3.5.6.2 Service Times
The service times were estimated by using both data from a previous study of the same
healthcare emergency department facility (Swenson 2008) and through review of those services.
Services times included triage time, registration time, medication, etc. These times were reviewed
with nurse practitioners to determine if they were still good descriptions of those activities. The
boarding time for patients was modeled as a service time estimate based on interviews. These
services are summarized in Table 3-5.
Routing for services was estimated by using aggregate data based on patient acuity level.
Services were assigned to patients using the 2017 FY data for the number of patients who
received a list of 15 different common services. Some services were performed in the room and
some were performed out of the room. The room was held while the patient left to receive out of
room care (MRI, X-Ray, CT scan, etc.). Table 3-5 shows the location of services, the human
resources needed (‘administered by’), and the availability of those resources.
70
Table 3-5. Service times, resources, and location summary
Process
Service Time Estimates Service
Location Number Avail Administered by Notes & Reference min (mean) max Distribution
Quick Registration 2 4 Uniform ent 1 Registration (Swenson 2008)
Triage evaluation 3 5 Uniform ent 2 Nurse (Swenson 2008)
CIA evaluation 3 5 Uniform ent 4(2) Doctor
Estimate, based on Triage, 2
triage at night
Nurse Visit 0.5 4 Uniform in zone & sch. dep. Nurse (Swenson 2008)
Doctor Visit 5 10 Uniform in schedule dep. Doctor (Swenson 2008)
Medication 1 2 Uniform in zone & sch. dep. Nurse (Swenson 2008)
Council patients 3 5 Uniform in schedule dep. Doctor (Swenson 2008)
Assess EKG 1 2 Uniform in schedule dep. Doctor (Swenson 2008)
Diagnosis 3 5 Uniform in schedule dep. Doctor (Swenson 2008)
Discharge Instructions 1 3 Uniform in zone & sch. dep. Nurse (Swenson 2008)
Notes: Service locations include ent = entrance, in = in-room service, out = out of room service, (remote)=service connected remotely (e.g., pneumatic tubes, electronic service), Number avail = Number of stations/resources available, zone & sch. dep.= zone specific and schedule
dependent availability of service/resources (e.g., nurse schedule, doctor schedule), Administered by = the operational resources needed (e.g.,
the people who perform the service).
71
Table 3-5 Continued. Service Times, Resources, and Location Summary
Process
Service Time Estimates Service
Location Number Avail Administered by Notes & Reference min (mean) max Distribution
Acute 2 & 3 waiting for
in-patient room 30 (120) 720 Triangular in none Estimate, based on interviews
Acute 1 Waiting for in-
patient room 10 (20) 30 Triangular in none Estimate, based on interviews
Breathing Treatment 2 5 Uniform in zone & sch. dep. Nurse Estimate
EEG 50 120 Uniform in zone & sch. dep. Nurse Estimate (Mayo Clinic 2019)
In room Registration 2 4 Uniform in zone & sch. dep. Nurse same as quick registration
In room Triage 3 5 Uniform in zone & sch. dep. Nurse same as Triage
IV 2 4 Uniform in zone & sch. dep. Nurse Estimate
MRI (NM, IR, MRI) 15 90 Uniform leave 2 Main Radiology Estimate (NHS 2018)
Ultrasound US 3 5 Uniform leave 2 Main Radiology Estimate (RANZCR 2016)
Ultrasound VL 3 5 Uniform in 2 Main Radiology Estimate (RANZCR 2016)
Notes: Service locations include ent = entrance, in = in-room service, out = out of room service, (remote)=service connected remotely (e.g.,
pneumatic tubes, electronic service), Number avail = Number of stations/resources available, zone & sch. dep.= zone specific and schedule dependent availability of service/resources (e.g., nurse schedule, doctor schedule), Administered by = the operational resources needed (e.g.,
the people who perform the service).
72
3.5.6.3 Estimated Walking Speed
Both patient and healthcare professional walking speeds were modelled as random
variables instantiated upon entity creation. There weren’t any studies of nurse and doctor walking
speed times in an emergency department found in the literature, thus the walking speed was
estimated by using a range of comfortable walking speed for people between the ages of 20 and
79 (Bohannon 1997). Walking speed was modeled as variables assigned during model initiation
based on a uniform distributed random variable between 1.27 and 1.46 m/s. Similarly for patients,
there is little data to pull from on emergency department patients walking speed and more
generally non-healthy patient walking speeds. A study on the threshold of walking independence
for elderly patients identified a minimum walking speed of 0.35 m/s before use of walking
assistive devices such as walker or wheelchair (Graham et al. 2010). Patient walking speeds were
estimated to be between the minimum threshold for walking independence and the maximum
comfortable walking speed, modeled as a uniform distributed random variable between 0.35 and
1.46 m/s.
3.5.6.4 Model Assumptions
The development of the ED model included several assumptions and estimations in the
implementation of the conceptual model of care. First off, the model was based on patient flow,
additional model development, task sequencing, and parameter estimation would need to be done
for modeling the full extent of nurse, doctor, and technician model entities. Registration personnel
were not modelled. Outside of the ED scope was the following closely tied departments and
resources: the trauma teams (who are a separate department and utilize some of the same spaces),
the radiology department personnel (who get patients and perform X-Rays in the department, but
also have resources in the basement for MRI, Ultrasounds, etc.). X-Ray, MRI, Cath, ECHO, and
73
Ultrasound resources were modeled, but every part of their workflow process were not modeled
in detail. The WR was not constrained on size. The subsequent nodes in the model and room
locations were constrained to force patient entities to wait in the input of the WR. The RWR was
modeled as a zone similar to the physician directed queuing and the fast track zones and because
of this had a constraint on the capacity in number of patient entities who could occupy the station.
Zones were modeled as stations with a capacity equal to the number of beds minus any procedure
rooms. The workflow of the procedure rooms were not modeled. The resources at the front of
house, such as triage nurses and CIA doctors, were not modeled. These resources were expected
to change in the future, thus additions to staffing were expected. These resources were assumed to
be able to meet the demand. For example, the resources to staff the additional CIA and Admits
zones were assumed to be adequate, thus they were not modeled so as to not constrain the model.
Optimization of the doctor and nurse resource plans could be an additional objective of the
simulation model in later iterations. Zone locations were modeled at the centroid of the rooms
allocated to represent the average distance and flow. Boarding was modeled as a random variable.
Additional data on boarders could make that value more accurate.
3.5.6.5 Decision Variables
The decision variables used in the model were based off of the main layout and process
changes that were developed for the ED expansion. These included the changes: path lengths (and
subsequent relocation of centroid of zones), addition of Results Waiting (RWR), creation of a
zone specifically for admitted patients waiting for an in-patient bed (Admits zone), additional FT
Bays (an increase from 4 to 8), changes to the intake process from triage nurse, capacity = 2, to
care initiation with a doctor (CIA), capacity = 4, with 2 open at night. Independent sampling was
used for each random variable and decision variables. For testing purposes, common random
numbers were used for each decision variable. Approximately 40 separate random number
74
streams were used in the model to create independent sampling within the model. Simio software
was used to run replications, which automatically spaces each replication sampling at a set
distance in the stream sampling to keep independence between replications.
3.5.6.6 Response Variables
The response variables are the performance metrics of interest for the ED healthcare
workers. These include length of stay (LOS) parameters for all patients, for discharged patients,
for admitted patients, and for the percentage of patients who stay in the ED longer than 3 hours.
The LOS for each acuity level was also tracked for model verification purposes. Additional
response variables were used to assess the performance of the new layout and process. These
included the number in the WR, maximum number in the WR. For the scenarios that were
applicable: number in the RWR, maximum number in the RWR, and number in the Admits zone.
3.5.7 Model Verification Methodology
The model was tested against the 2017 FY aggregate data using a model run length of
100 days and 50 replications. An inspection of more detailed statistics was performed on the
acuity level data to test if each ESI level in the simulation was following the pattern seen in the
baseline 2017 FY data. After inspection, the total population, average length of stay, discharged
length of stay, ESI 2 length of stay, ESI 3 length of stay, and percent of patients who stayed
longer than 3 hours were all found to be less than the expected value from the 2017 FY data,
ranging between 2.8%-17.1% lower (Table 3-6). The admitted patient length of stay, ESI 4 length
of stay, ESI 5 length of stay, and WR waiting time were all found to be higher than the baseline
2017 FY data, ranging between 4.7% and 86.6% higher (Table 3-6). The highest differences were
with ESI 4 and 5 patients LOS, with 1.02 hours (35.0%) and 1.66 hours (86.6%), respectively,
75
higher than the baseline data. The confidence intervals across the 50 simulation runs are shown in
Table 3-6 for the mean values at a 95% confidence interval.
While the model’s averages of performance measures do not match the summary statistic
from baseline 2017 FY data, the model follows the expected trends of longer length of stay for
admitted patients, with the ESI 2 patients staying the longest and ESI 5 staying the least amount
of time, and ESI 1 patients taking the median amount of time to be treated. The current layout
model can be used to support research questions. The current layout model was explored to
understand what resources were driving the model. An inspection of the utilization rates for the
in-room services and the out-of-room services. The main bottlenecks in the system were
identified as resources with a schedule utilization higher than 85%, this included the lab (draw
and results), radiology (both X-Ray and CT radiology in the ED department and MRI outside the
department, imaging and reading results), and in-patient bed transfer times (boarding times).
Additionally the utilization of the zones, doctors, nurses, and techs were explored. All doctors and
the nurses in the blue zone (servicing both cobalt/MT and PDQ/FT patients with Grey nurses) had
high scheduled utilization levels.
76
Table 3-6. Summary verification statistics
Metric 2017 FY Current Difference
(% of 2017 FY)
Population (Day) Mean 208.274 202.5038
-2.77% (CIl,Ciu) (202.08, 202.93)
Average LOS (hr) Mean 5.33 4.981
-6.55% (CIl,Ciu) (4.77, 5.19)
Discharged LOS
(hr)
Mean 4.409 3.711 -15.83%
(CIl,Ciu) (3.48, 3.94)
Admitted LOS (hr) Mean 8.139 8.585
5.48% (CIl,Ciu) (8.42, 8.75)
ESI 1 LOS (hr) Mean 4.656891 4.877
4.73% (CIl,Ciu) (4.76, 4.99)
ESI 2 LOS (hr) Mean 7.154222 6.521
-8.85% (CIl,Ciu) (6.41, 6.63)
ESI 3 LOS (hr) Mean 5.770786 4.785
-17.09% (CIl,Ciu) (4.55, 5.02)
ESI 4 LOS (hr) Mean 2.906316 3.923
34.99% (CIl,Ciu) (3.64, 4.2)
ESI 5 LOS (hr) Mean 1.916459 3.576
86.61% (CIl,Ciu) (3.27, 3.88)
Waiting Time (min) Mean 13.654 21.927 60.59%
*note FY value from all patients,
model data from WR statistics (CIl,Ciu) (15.31, 28.54)
% LOS > 3 hr
Mean 67.70% 63.14%
-6.73% (CIl,Ciu)
(61.55%,
64.74%)
3.5.8 Model Validation Methodology
Model validation was performed with healthcare professionals who are experts on the
operating practices of the emergency department and familiar with workflow practices during
2017 FY. The conceptual workflow was developed and reviewed. Once a comprehensive
conceptual workflow was developed, it was reviewed with additional practitioners, including
pediatrics and nursing professionals. For the pediatrics workflow, a pediatrics doctor was asked to
review the workflow processes. The head nurse also reviewed the workflow from the nurse
77
practitioner perspective. The future workflow processes were based on observations and data
from the workflow planning sessions and were reviewed by several healthcare professionals.
The model was presented to a healthcare professional. The model was reviewed by going
through the translation of the conceptual model to the current layout model of the 2017 FY
operational practices as well as the assumptions in the future workflow model. Changes that were
found during this process included model assumptions for doctor routing for the pediatrics area,
the resources dedicated to the intake processes (triage by nurses in current scenario and by
doctors in the future scenario), location of patients waiting for intake process in the current
scenario, and a dedicated boarding area in the current workflow process.
3.5.9 Layout Scenarios
Scenarios were developed based on the schematic design options. Five areas were of
interest to the designers for changes: (1) the Waiting Room (WR), (2) the Results Waiting Room
(RWR), (3) the area including the SANE consult/exam room and the decontamination/isolation
room, (4) the area with Fast-Track (FT) recliner bays for low acuity patients, and (5) the new
Mid-Track (MT) patient bed area adjacent to the Pediatrics zone. The WR was modeled as
location where people waiting until beds were available, thus it was set up as a response variable.
The addition of SANE consult room was not modeled given the modeling goals, level of detail in
the simulation model, and lack of input routing data. A scenario was created in the discrete event
simulation for each combination of layout decision variables. A summary of each scenario
parameter changes are presented in Table 3-7.
A study of the impact of the layout alone was developed by setting up Scenario 1 (S1)
with the only parameter changed being the path routing based on the new floorplan. A
comparison of the current layout model (Current) to S1 was developed to answer the basic
question, how much does layout impact performance measures? In S1, this was setup as solely
78
changing paths alone, without making any routing, service, or resource changes. The rest of the
scenarios had layout and process changes including the routing of low acuity patients to a RWR
during daytime and evening shifts, location of the admits routing (using the Admits zone),
addition of 4 FT bays, and intake change from 2 Triage rooms to 4 CIA rooms.
To answer RQ1a, first the current layout scenario (Current) is compared with the layout
path changes alone (S1), then a fully crossed experimental design was developed with S1-16 and
these were compared to one another and to the baseline control model of the current (2017)
conditions. To answer RQ1b, a selection of the best methodology was used (Kim and Nelson
2007), which is described further in Section 3.5.10. To answer RQ2, statistics on the additional
response variables associated with layout decisions were explored to understand if there were
opportunities to better allocated space throughout the redesign project, such as WR, RWR, and
Admits zone statistics. To answer RQ3, the demand projections were modeled and tested under
the theoretical best solution with all implemented layout changes (for parameter changes see
Table 3-8).
The estimated increase in future demand was expected to be approximately 4,400
patients/year over the next 4 years, which is equivalent to an increase on average of 12
patients/day (5.8%). With the same resources available and all layout parameters, the simulation
was run under the new demand scenario. Using the future demand scenarios for scenario 17 and
18, an even increase of population of on average 5.8% (Year 1) and 11.9% (Year 2) for each
hourly average was used in the simulation of demand scenarios. A summary of the future demand
scenarios is in Table 3-8.
79
Table 3-7. Scenarios and current condition control. Latin square experimental design.
Current system scenario based on 2017 Fiscal Year (July ‘16 – June ‘17)
Scenario Path
Lengths RWR Admits FT Bays CIA
Current
S1 x
S2 x x
S3 x x
S4 x x x
S5 x x
S6 x x x
S7 x x x
S8 x x x x
S9 x x
S10 x x x
S11 x x x
S12 x x x x
S13 x x x
S14 x x x x
S15 x x x x
S16 x x x x x
Note: x indicates that parameter is modeled as expected in the new design in the scenario. All
operationalized layout parameters are modeled as a Boolean state.
Table 3-8. Demand scenario comparisons Demand
Scenario
Path
Lengths RWR Admits FT Bays CIA Demand
S16 x x x x x 2017 input data
S17 x x x x x 5.8% increase
S18 x x x x x 11.9% increase
Note: x indicates that parameter is modeled as expected in the new design in the scenario.
3.5.10 Output Analysis Methodology
For each of the layout scenarios, the response variables were studied to select the best
layout. A selection of the best methodology was used based on the KN methodology (Kim and
Nelson 2007). In the Simio software, the best scenario was selected if it was significantly better
choice based on one response variable at a time. An initial warm up period of three days was
selected after inspecting the average length of stay performance metrics in an initial run. An
initial number of runs was used for the selection of the best screening and selection procedure, set
at 50 runs. The total length for each run was set at 100 days, or just over 3 months.
80
When there are K system designs, the methodology to select the best system on a specific
performance measure was used following procedure described in Banks et al. (2010). The steps
involved in selection of the best procedure is to first specify the desired probability of correct
selection (𝛼), set a practical significance difference (𝜖), and specify the initial number of runs
(𝑅0) for each system design. Next, an initial number of simulation replications are performed
(𝑅0 = 50) and an initial screening of the performance measure of interest is determined based on
the critical T-value (Equation 3-1), the first stage sample mean across replications (Equation 3-2),
sample variance (Equation 3-3), and the screening threshold between the best first stage sample
mean (minimum in this study: min { 𝑌.𝑖} for 𝑖 = 1,2, … 𝐾 ) and each other system is calculated
(𝑊𝑖𝑗, Equation 3-4). The systems that are significantly different from the best are eliminated
(Equation 3-5, Equation 3-6). For each scenario remaining, the additional number of replications
needed (𝑅𝑖 − 𝑅0) to find a significant difference are calculated using the second stage sample size
calculation using Rinott’s constant (ℎ, Equation 3-7), the standard deviation of the scenario (𝑆𝑖),
and the practical difference initially defined. The additional replications of each scenario are run
(if needed) and the overall sample means by system are calculated. Finally, the system with the
best overall sample mean is selected.
𝑡 = 𝑡1−(1−𝛼 2⁄ )
1𝑘−1,𝑅0−1
Equation 3-1. Critical T-value for screening threshold
�̅�.𝑖 =1
𝑅0∑ 𝑌𝑟𝑖
𝑅0𝑟=1 for 𝑖 = 1, 2, … , 𝐾.
Equation 3-2. First stage sample mean
𝑆𝑖2 =
1
𝑛0−1∑ (𝑌𝑟𝑖 − �̅�.𝑖)2𝑛0
𝑟=1 , for 𝑖 = 1, 2, … , 𝐾.
Equation 3-3. First stage sample variance
𝑊𝑖𝑗 = 𝑡 (𝑆𝑖
2+𝑆𝑗2
𝑅0)
1
2
, for all 𝑗 ≠ 𝑖.
Equation 3-4. Screening threshold
81
�̅�.𝑖 ≥ �̅�.𝑗 − max {0, 𝑊𝑖𝑗 − 𝜖} for all 𝑗 ≠ 𝑖
Equation 3-5. Screening for maximized value
�̅�.𝑖 ≤ �̅�.𝑗 + max {0, 𝑊𝑖𝑗 − 𝜖} for all 𝑗 ≠ 𝑖
Equation 3-6. Screening for minimized value
ℎ = ℎ(𝑅0, 𝐾, 1 − 𝛼 2⁄ )
Equation 3-7. Rinott’s Constant
𝑅𝑖 = max {𝑅0, ⌈(ℎ𝑆𝑖 𝜖⁄ )2⌉} where ⌈. ⌉ means round up
Equation 3-8. Second stage sample sizes
The procedure (proven in Nelson et al. 2001) finds either (1) the system with the
largest/smallest performance measure; or (2) the system within 𝜖 of the best performance
measure, at a level of confidence (1 − 𝛼). A stopping criteria can be used for the maximum
number of replications allowed to find a significant difference, e.g., 𝑅𝑚𝑎𝑥 = 200. If multiple
solutions exist at the stopping criteria, the procedure finds a set of systems within 𝜖 of the best
performance measure.
When using the selection of the best methodology, each performance measure is
evaluated separately. In order to combine several performance measures, there are three strategies
that can be used. First is to combine performance measures into a single metric. Second, optimize
for one performance measure and evaluate the top solutions with respect for a secondary measure.
Thirdly, optimize for one performance measure but only consider alternatives that meet a certain
constraint on other performance measures.
These methods are estimations of relative performance. When estimating relative
performance, the exact increase in performance is unknown, and thus the method returns relative
differences between the averages measured from each scenario. In the ED, the main performance
measure is a combined metric: length of stay for all patients, combining all zones, patient types,
and discharge types into a single measure. The additional performance measures were used as
secondary if there were no significant difference found on the primary metric. A significance
level of 5% (𝛼 = 0.05) and practical significant difference value of 5 minutes (𝜖 = 0.08333 ℎ𝑟)
82
were defined, e.g., a difference of 5 minutes was used as the threshold for overall improvement
between different scenarios tested with a 5% significance level on averages of 100 days of
simulated patients lengths of stay across a set of 50 simulation runs. The first stage replications
were enough to determine a best scenario.
3.6 Results
In this section, the performance metrics of interest are compared from the current
condition to the 16 different scenarios. For each of these performance metrics, the first research
questions is answered, RQ1a: How does layout impact performance measures? Then the
performance metrics are explored together to assess RQ1b: Which layout is the best?, Next,
response variables associated with space allocation analyzed to answer RQ2: Were there
opportunities to optimize space allocation based on this analysis? And finally, future demand
projections were simulated to address the last research question RQ3: How does the layout
perform under different demand scenarios?
3.6.1 Population Results
All scenarios used a control on the random number stream used for all random variables,
thus all scenarios had the same pattern of patient population. The population for each scenario
was on average (SD) 202.5 (14.8) patients per day and 73,914 (541.8) patients per year. The 2017
FY had a total of 76,020 patients on record. The simulation ranged from 199.5 to 205.4 patients
per day for all 50 runs, or approximately 2086 less than the baseline 2017 FY data.
83
Table 3-9. Simulated patient population Scenario Patient Population SD Resulting Yearly
Rate (95% CI)
Comparison to 2017
FY
Current layout,
S1-16
20250.38 148.4453 73914
(74068.67, 73759.10)
significantly lower,
estimate difference =
2086 patients/yr
3.6.2 Length of Stay for all Patients
A summary of the length of stay for all patients is available in Table 3-10. The average
(95% CI) LOS for Current was 4.981 (4.77, 5.19). There was no significant difference between
the current Current and S1, path changes alone. The box plots of the averages across all
simulations runs for each scenario are show in Figure 3-9, which shows that the average LOS
were similar in variance across these two conditions. For each of the 16 scenarios, Figure 3-10
shows a comparison of the variation with box plots. When adding 4 stations for care initiation vs.
the typical triage, comparison of S1 to S9, a slight reduction in variance is found. When
implementing RWR without admits (S2, S6, S10, and S14) a significant drop in overall length of
stay is found compared to no RWR baseline (S1), the addition of FT bays (S5), the use of care
initiation (S9), and the addition of FT bays and use of care initiation (S13). When implementing a
separate Admits zone (S3, S7, S11, and S15), a similar magnitude drop in overall length of stay is
found compared to no Admits zone (S1 and S9). Combining both RWR routing and Admits zone
have a combined effect (S4, S8, S12, and S16) under the conditions of path changes alone (S1),
additional FT bays (S5), care initiation (S9), and care initiation and additional FT bays (S9),
respectively. The effect of adding FT bays reduces the variance of the simulation results as well
as reduces the average length of stay (compare S5 to S1 and S13 to S9). However the additional
FT bays with all other factors in place did not reduce the overall length of stay (compare S8 to S4
and S16 to S12). The best scenario based on the overall length of stay is S12, with an average
LOS of 3.773 (3.727, 3.818). The order of top 4 ranked solutions based on overall length of stay
are (in increasing LOS): S12, S4, S16, S8. All used both RWR and admits, S12 and S4 did not
84
have the additional FT bays, and S12 and S16 both had care initiation. The results indicate
relative performance, thus the use of RWR and a separate admit area were the largest contributor
to the reduction in LOS. Small gains were found by adding care initiation. The use of additional
FT bays didn’t help the overall LOS once these other factors were taken into account.
3.6.3 Length of Stay for Discharged Patients
Discharged patients LOS is summarized in Table 3-10. Discharge patients are mostly ESI
3s (76% of all 3s), a majority of ESI 2s (54% of all 2s) and a predominant proportion of ESI 4s
and 5s (96% and 94%, respectively). The average LOS for the S1 was not significantly different
from Current, for box plots see Figure 3-11. The scenarios with RWR had the most significant
changes in LOS of discharged patients (Figure 3-12). This makes intuitive sense because
discharged patients aren’t waiting to be admitted, they are discharged and released from the ED.
The addition of the Admits zone did provide some benefit to the discharged LOS, (compare S3 to
S1), but not as much as adding the RWR (compare S2 to S3). The addition of the FT bays
initially significantly decreased the discharged patient LOS, (compare S5 to S1 and S13 to S9).
However, once RWR was introduced, adding additional FT bays increased the discharged LOS
(comparing S6 to S2, estimated difference, S6-S2, = 0.0930 hrs). Adding care initiation had little
impact on the discharged LOS. The best scenario in the system was S12 with an average (95%
CI) LOS of 2.862 hrs (2.816, 2.907). The top 4 scenarios were in increasing LOS were S12, S4,
S2, and S10. None of the top scenarios had the additional FT bays. Both S12 and S10 were using
care initiation. Both S12 and S4 had Admits zone and RWR. S2 only had RWR and still was in
the top 4 scenarios.
85
Table 3-10. Summary data for overall length of stay metrics
Scenario Average LOS (hr) Discharged LOS (hr) Admitted LOS (hr) % LOS > 3 hr
Mean (CIl,Ciu) Mean (CIl,Ciu) Mean (CIl,Ciu) Mean (CIl,Ciu)
Table 3-15. Discharged LOS Summary of Current Scenario to Scenario and within Scenario Differences
Note: Units = minutes; Differences calculated as row minus column; red = negative difference; blue = positive difference; ns* = no significance found at 0.003125, for Current Scenario to Scenario difference tests, m = 16; ns** = no significance found at 0.0004167, for
Table 3-16. Admitted LOS Summary of Current Scenario to Scenario and within Scenario Differences
Note: Units = minutes; Differences calculated as row minus column; red = negative difference; blue = positive difference; ns* = no significance found at 0.003125, for Current Scenario to Scenario difference tests, m = 16; ns** = no significance found at 0.0004167, for
Table 3-17. Percent with LOS Greater than 3 Hours Summary of Current Scenario to Scenario and within Scenario Differences
Note: Units = percent; Differences calculated as row minus column; red = negative difference; blue = positive difference; ns* = no significance found at 0.003125, for Current Scenario to Scenario difference tests, m = 16; ns** = no significance found at 0.0004167, for
Table 3-18. Average Number in WR Summary of Current Scenario to Scenario and within Scenario Differences
Note: Units = minutes; Differences calculated as row minus column; the goal is a positive difference; red = negative difference; blue = positive difference; ns* = no significance found at 0.003125, for Current Scenario to Scenario difference tests, m = 16; ns** = no
significance found at 0.0004167, for within Scenario difference tests, m = 120
Simulation Approach for an Integrated Decision Support System in Healthcare Facilities.” 2018 Winter
Simulation Conference (WSC). December 2018. 3 In reference to IEEE copyrighted material which is used with permission in this thesis, the IEEE does not
endorse any of the Pennsylvania State University's products or services. Internal or personal use of this
material is permitted. If interested in reprinting/republishing IEEE copyrighted material for advertising or
promotional purposes or for creating new collective works for resale or redistribution, please go to
http://www.ieee.org/publications_standards/publications/rights/rights_link.html to learn how to obtain a
License from RightsLink.
150
problem of minimizing distance traveled by patients, formulated as a Quadratic Assignment
Problem (QAP). While this is a classic problem for facility layout problems, in context for
designing and constructing new and renovated healthcare facilities, this method doesn't connect
the processes of a future facility to the implementation of the QAP. Location and layout
optimization is typically done in early stages of the design process when little is known about the
new processes to be implemented in the renovated/new facility yet needs data about appropriate
flow weights or costs, depending on the formulation, to accurately find optimal layout
arrangements. Some research (Acar et al. 2009; Arnolds and Nickel 2015) has looked at an
optimization-simulation approach in these healthcare layout planning problems.
Increase in immersive visualization is one of the key features of communication between
model creators and decision makers (O’Keefe 2016). Virtual reality allows healthcare
professionals to experience their space. Discrete event simulation (DES) allows healthcare
practitioners to test their workflow processes. Virtual reality has been used in the design
evaluation process to allow those not familiar with 2D plans and sections to have a greater
understanding of the spatial arrangement and spatial decisions they are making (van der Land et
al. 2013). 3D visualization has been found to be beneficial in the evaluation of DES models
(Akpan and Shanker 2017). Typical software (e.g., Simio, Flexsim) displays have incorporated
advanced visualization features including 2D, 3D visualization, walkthrough and animation
functionality. However, the features alone don't address including visualization criteria into the
hybrid simulation methodology. The integration of an optimization-simulation-visualization
(OSV) framework can allow for a more iterative structure combining the mathematical and
simulation approaches with immersive visualization evaluation of new processes in future
healthcare facilities to allow for a combined human-centered and data-driven approach.
151
5.2 Background Theory
The context for facility planning should be placed in the facility lifecycle and the
objective of the facility (for example, in this application area: patient care). In this section, the
context of both the facility lifecycle and an overview of the patient evaluation process are
discussed.
5.2.1 Building Lifecycle Process
The building lifecycle is made up of 5 distinct processes: manage, plan, design, construct,
and operate, (Sanvido et al. 1990). Manage includes the business side of building a facility. Plan
defines what the owner of a facility needs, such as the idea of a new facility or a redesign and
developing a program of specific functions and space requirements needed in that facility. Design
consists of functions that communicate the owner's needs with the design team and transforms
those into the design, bid documents, and construction plans. Construct comprises all the building
activities from demolition to all assembly activities. Operate includes all the operational activities
of the facility, including turnover, operations, and maintenance. From an overview of these
processes, manage is the activity which lasts consistently through all stages of the building
lifecycle and connects to all the other aspects of the design and operations lifecycle (Figure 5-1).
152
Figure 5-1. Overview of elements of providing a facility.
These processes are interconnected and can be modeled as distinct parts with inputs,
mechanisms, controls, and outputs (Figure 5-2). When investigating the integrated process, it is
common to think of the plan, design, construct, and operate activities as predecessors to one
another. If we add redesign to the scope, we have a full circle process (Figure 5-1). However,
these processes are interdependent in ways that are more complex than any linear or cyclical
depiction. Sanvido et al. (1990) began to investigate the inputs and outputs of these processes. In
the Integrated Building Process Model, outputs from design, construction, and operations of a
facility feedback into the manage, plan, and design processes of a new or renovated facility
(Figure 5-2), typically as best practices (blue to red lines) and knowledge of what worked and
what didn't which becomes how the project team and owners experience the facility (blue to
green lines). Managers of facilities collect “performance information'” for the facility overall and
“optimization information” to evaluate project performance and facility performance. In the
process model, optimization information means the information used to integrate the expertise of
participants, including designability, constructability, operability, and maintainability
information. This information constrains the manage activities. This same information is used
control the plan and the design and, in turn, impacts how facility owners and users experience the
153
facility. The question arises: how can we better include performance evaluation of the design and
operations in the design and construction processes?
5.2.2 Integrated Simulation
An integrated technique is needed if we want to leverage the computational techniques
effectively in the design of healthcare facilities. Gibson (2007) discussed using discrete event
simulation for scenario testing in the schematic design stage of healthcare projects, yet is not
commonly shown in the literature or implemented in practice. Visualization techniques can be
used in future applications to improve the understanding among the disparate team members,
expand the use cases of experiencing new processing before buildings are built, (i.e., access and
identify elements of importance not modeled) and implement continuous improvement cycles
between design and operations. Integrating layout analysis, healthcare processes, and spatial
visualization may provide a framework where each approach builds off one another while
alleviating common implementation and communication problems.
5.2.3 Patient Flow Process
Patients flow is an important area of research for healthcare professionals. Healthcare
simulation research has been a popular application area throughout the history of the Winter
Simulation Conference as highlighted in a review of healthcare simulation (Arisha and Rashwan
2016). Managers are interested in performance measures such as length of stay of patients. Less
research has been on the use of simulation in design and planning of healthcare applications, such
as in layout or bed capacity analysis (Arisha and Rashwan 2016).
154
Figure 5-2. Elements of providing a facility in the Integrated Building Process Model. Red highlights feedback from Design, Construction,
and Operations into Manage, Plan, and Design. Blue indicates knowledge output. Green indicates experience of the facility resulting from
all phases (Sanvido et al. 1990, p.31).
Notes: FCE = Facility Construction Experience, FCK = Facility Construction Knowledge, FDE = Facility Design Experience, FDK = Facility
Question 10. What is the highest level of education you have completed?
Response type: multiple choice
1. Less than High School, 2. High school, 3. Some College, 4. 2-year College Degree, 5. 4-
year College Degree, 6. Masters Degree, 7. Doctoral Degree, 8. Professional Degree (JD,
MD)
Question 11. How long have you been with your current company? (years)
Response type: text entry, restricted to number between 0 and 80, up to 2 decimals
Question 12. How many years of experience do you have with Healthcare projects?
Response type: text entry, restricted to number between 0 and 80, up to 2 decimals Question 13. What type of experience do you have with Healthcare projects? Select all that
• Arup, Interactive Visualization Team, Visualization and BIM Intern, New York, NY 2014
SELECT PUBLICATIONS
• Lather, J. I., Logan, T., Renner, K., and Messner, J. I. (2019). “Evaluating generated layouts in
a healthcare departmental adjacency optimization problem.” International Conference on Computing in Civil Engineering 2019, American Society of Civil Engineers, Atlanta, GA.
• Lather, J. I. and Messner, J. I. (2018). “Framework for a hybrid simulation approach for an
integrated decision support system in healthcare facilities.” Winter Simulation Conference
2018. Institute for Operations Research and the Management Sciences and Association for
Computing Machinery, Gothenburg, Sweden.
• Lather, J. I., Leicht. R. M., and Messner, J. I. (2018). “Engaging with BIM: Interactive
workspaces in facility design and construction.” Construction Research Congress 2018.
American Society of Civil Engineers, New Orleans, LA.
• Lather, J. I., Amor, R., and Messner, J. I. (2017). “A case study in data visualization for linked
building information model and building management system data.” IWCCE 2017. American