a hybrid modeling approach using discrete event simulation
TRANSCRIPT
The Pennsylvania State University
The Graduate School
A HYBRID MODELING APPROACH USING DISCRETE EVENT SIMULATION AND
LAYOUT OPTIMIZATION FOR HEALTHCARE LAYOUT PLANNING PROBLEMS
A Dissertation in
Architectural Engineering
by
Jennifer I. Lather
© 2019 Jennifer I. Lather
Submitted in Partial Fulfillment
of the Requirements
for the Degree of
Doctor of Philosophy
August 2019
ii
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
Chapter 1. Introduction .............................................................................................................................. 1
1.1 Goal and Objectives ................................................................................................................. 6 1.2 Research Scope ........................................................................................................................ 7 1.3 General Methodology Overview ............................................................................................. 7
Chapter 2. Literature Review ..................................................................................................................... 10
2.1 Healthcare Performance Metrics ............................................................................................. 11 2.2 Healthcare Design Process....................................................................................................... 17
2.2.1 Discrete Event Simulation in Healthcare ..................................................................... 18 2.2.2 Optimization in Healthcare Design .............................................................................. 22 2.2.3 Visualization in Healthcare Design .............................................................................. 22
2.3 Discrete Event Simulation ....................................................................................................... 23 2.3.1 Problem Formulation.................................................................................................... 27 2.3.2 Setting Objectives and Simulation Plan ........................................................................ 27 2.3.3 Model Conceptualization .............................................................................................. 28 2.3.4 Data Collection ............................................................................................................. 28 2.3.5 Model Translation ......................................................................................................... 30 2.3.6 Verification ................................................................................................................... 30 2.3.7 Validation ...................................................................................................................... 31 2.3.8 Experimental Design ..................................................................................................... 31 2.3.9 Production Runs and Analysis ...................................................................................... 32 2.3.10 Optimization within Simulation .................................................................................... 32 2.3.11 Documentation and Reporting ...................................................................................... 33
2.4 Facility Planning and Layout Optimization ............................................................................. 34 2.5 Virtual Prototyping and Visualization ..................................................................................... 35 2.6 Integrating Simulation, Optimization, and Visualization ........................................................ 39
2.6.1 Crane Mobilization ....................................................................................................... 39 2.6.2 Stroboscope and Vitascope ........................................................................................... 39 2.6.3 Traffic Simulations ........................................................................................................ 40 2.6.4 Manufacturing Applications (VR Factory) ................................................................... 41 2.6.5 Integration in Healthcare ............................................................................................. 41
2.7 How Has Research Suggested Integration of These Methods? ............................................... 43
Chapter 3. Layout Implication for an Emergency Department: Scenario Tests in a Discrete Event
Simulation .................................................................................................................................................. 45
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
4.2.1 Layout Generation Methods ......................................................................................... 120 4.2.2 Layout Evaluation Metrics ............................................................................................ 128 4.2.3 Expert Evaluation Methods........................................................................................... 131
4.3 Layout Scoring Results ............................................................................................................ 137 4.3.1 Graph Results................................................................................................................ 137 4.3.2 Layout Generation Results ............................................................................................ 138
4.4 Evaluation of Layout Results ................................................................................................... 139 4.4.1 Comparisons of Subjective and Objective Optimal Layout .......................................... 140 4.4.2 Results of Best Scoring Layout Choice ......................................................................... 140 4.4.3 Results of Worst Scoring Layout Choice ...................................................................... 142 4.4.4 Results of Perceived Usefulness.................................................................................... 143 4.4.5 Results of Perceived Need for Decision Details ........................................................... 143 4.4.6 Results of Demographic Variables ............................................................................... 143 4.4.7 General Perceptions of Generative Layouts ................................................................. 144
4.5 Discussion ................................................................................................................................ 145 4.6 Conclusions .............................................................................................................................. 148
Chapter 5. Framework for a Hybrid Simulation Approach for an Integrated Decision Support System in
Healthcare Facilities .................................................................................................................................. 149
vii
5.1 Introduction .............................................................................................................................. 149 5.2 Background Theory ................................................................................................................. 151
5.2.1 Building Lifecycle Process ............................................................................................ 151 5.2.2 Integrated Simulation.................................................................................................... 153 5.2.3 Patient Flow Process .................................................................................................... 153
5.3 Related Simulation Work......................................................................................................... 156 5.3.1 Optimization and Facility Layout Design ..................................................................... 156 5.3.2 Healthcare Layout and Design Studies......................................................................... 157 5.3.3 Discrete Event Simulation in Healthcare ..................................................................... 157 5.3.4 Virtual Reality in Discrete Event Simulation ................................................................ 158 5.3.5 Virtual Reality for Facility Review ............................................................................... 158 5.3.6 Summary of Related Work ............................................................................................ 159
5.4 Development Methodology ..................................................................................................... 160 5.4.1 Healthcare Design Review Process .............................................................................. 161 5.4.2 Conceptualization ......................................................................................................... 161 5.4.3 Hybrid Simulation Objectives ....................................................................................... 163
5.5 Facility Lifecycle Implementation ........................................................................................... 164 5.5.1 Implementation during Operations ............................................................................... 164 5.5.2 Implementation during Planning .................................................................................. 165 5.5.3 Implementation during Design Conceptualization ....................................................... 165 5.5.4 Implementation during Schematic Design .................................................................... 166 5.5.5 Implementation during Design Development ............................................................... 166 5.5.6 Implementation during Construction ............................................................................ 166
5.6 Conclusions .............................................................................................................................. 167
Chapter 6. Conclusions .............................................................................................................................. 169
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
6.5 Concluding Remarks................................................................................................................ 182
References.................................................................................................................................................. 183
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
Survey Procedure ................................................................................................................................ 197 Survey Apparatus ............................................................................................................................... 197 IRB Documentation ............................................................................................................................ 202
viii
LIST OF FIGURES
Figure 1-1: Aspects of design (expanded in Bate and Robert 2007, p. 5, originally from
Berkun 2004). .................................................................................................................. 4
Figure 1-2: Generic model process diagram. Model development in simulation,
optimization, and visualization follow these general steps.............................................. 8
Figure 1-3: Methodology overview diagram with research phases and activities................... 9
Figure 2-1: Emergency Severity Index conceptual algorithm (Gilboy et al. 2011) ................ 17
Figure 2-2: Continuum of approaches to simulation modeling (Robinson 2002, p. 3) ........... 20
Figure 2-3: Typical steps and flow of a simulation process (Banks et al. 2010, p. 15) ........... 26
Figure 2-4: Experienced-based virtual prototyping steps (Kumar 2013, p. 103) .................... 38
Figure 3-1: Existing conditions and expansion diagram (Huddy et al. 2016, p.12). ............... 54
Figure 3-2: Conceptual configuration of Phase 1 (Huddy et al. 2016, p.15). .......................... 55
Figure 3-3. Typical emergency department overview workflow ............................................ 57
Figure 3-4. Overview of typical acuity routing for ED patients .............................................. 60
Figure 3-5. Current room configuration with zones ................................................................ 62
Figure 3-6. Future room configuration with zones .................................................................. 64
Figure 3-7. Conceptual model for patient flow in the current layout ...................................... 67
Figure 3-8. Conceptual model for patient flow in the future layout ........................................ 68
Figure 3-9. Box plots for average LOS of all patients across runs, Current and S1................ 86
Figure 3-10. Box plots for average LOS of all patients across runs, S1-S16 .......................... 86
Figure 3-11. Box plots for average LOS of discharged patients across runs, Current and
S1 ..................................................................................................................................... 87
Figure 3-12. Box plots for average LOS of discharged patients across runs, S1-S16 ............. 87
Figure 3-13. Box plots for average LOS of admitted patients across runs, Current and S1.... 89
Figure 3-14. Box plots for LOS of admitted patients across runs, S1-S16 ............................. 89
ix
Figure 3-15. Box plots for average percent of LOS longer than 3 hours across runs for all
patients, Current and S1 ................................................................................................... 90
Figure 3-16. Box plots for average percent of LOS longer than 3 hours across runs for all
patients, S1-S16 ............................................................................................................... 90
Figure 3-17. Box plot of average time in WR (minutes) across runs, Current and S1 ............ 92
Figure 3-18. Box plots for average waiting time in WR (minutes) across runs, S1-S16 ........ 93
Figure 3-19. Box plots for average number in WR across runs, Current and S1 .................... 94
Figure 3-20. Box plots for average number in WR across runs, S1-S16................................. 94
Figure 3-21. Box plots for average maximum number in WR across runs, Current and S1 ... 95
Figure 3-22. Box plots for average maximum number in WR across runs, S1-S16 ............... 95
Figure 3-23. Box plots of average number in RWR across runs ............................................. 97
Figure 3-24. Box plots for maximum number in RWR across runs ........................................ 98
Figure 3-25. Box plots for the average number in Admits zone across runs........................... 98
Figure 3-26. Box plots for the average length of stay across runs for demand increase
scenarios........................................................................................................................... 112
Figure 3-27. The percentage of length of stay greater than 3hrs across runs for demand
increase scenarios ............................................................................................................ 112
Figure 3-28. Box plot for average number in RWR across for demand increase scenarios .... 113
Figure 3-29. Box plots for average maximum number in RWR across runs for demand
increase scenarios ............................................................................................................ 113
Figure 3-30. Box plots for average number in Admits zone across runs for demand
increase scenarios ............................................................................................................ 114
Figure 4-1. Generative layout methodology ............................................................................ 121
Figure 4-2. (a) Initial tetrahedron formulation of graph theoretical approach (b) Final
maximally planar subgraph .............................................................................................. 123
Figure 4-3. (a) Dual (red) of the adjacency graph (the exterior boundary node is not
shown) (b) A possible block layout formulation meeting all adjacency relationship
requirements..................................................................................................................... 123
Figure 4-4. Serpentine placement pattern, placement path with specified bay size ................ 126
Figure 4-5. Shape grammars for serpentine shape translation ................................................. 127
x
Figure 4-6. BIM objects generated in a parametric BIM authoring tool ................................. 127
Figure 4-7. Block plans for six layout conditions.................................................................... 135
Figure 4-8. Sample layout, Option 4 ....................................................................................... 139
Figure 4-9. Frequency of respondent’s choice of ‘best’ and ‘worst’ layouts with total and
first choices, and the horizontal line for random choice (5.17, n=31) ............................. 141
Figure 5-1. Overview of elements of providing a facility. ...................................................... 152
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). ................................... 154
Figure 5-3. Typical emergency room patient processes. Containers indicate parts in the
process where a patient is roomed. These change based on the condition of the
patient and healthcare processes. ..................................................................................... 155
Figure 5-4. Diagram of hybrid simulation hierarchy proposed for healthcare. ....................... 159
Figure 5-5. Integration technique proposed by Acar et al. (2009) and revised by Arnolds
and Nickel (2015). ........................................................................................................... 161
Figure 5-6. Conceptual diagram for integration of optimization, simulation, and
visualization for healthcare planning. .............................................................................. 163
Figure 6-1. Final conceptual diagram for the optimization-simulation-visualization
framework. ....................................................................................................................... 172
Figure 6-2. Process diagram for facility layout problem to discrete event simulation ............ 174
Figure 6-3. Process diagram for discrete event simulation to facility layout problem ............ 174
Figure 6-4. Taxonomy of aspects of a hybrid simulation approach in healthcare ................... 175
Figure 6-5. Physical mockup of a typical new patient room ................................................... 181
xi
LIST OF TABLES
Table 2-1:Timely & effective care, emergency department throughput (CMS 2017) ............ 12
Table 2-2: Summary relationships between design factors and healthcare outcomes
(Ulrich et al. 2008) ........................................................................................................... 15
Table 2-3: National averages for emergency department healthcare outcomes (CMS
2017), gray cells indicate average is across all emergency department volumes ............ 16
Table 3-1. Performance metrics of interest for ED case study including selection of the
best goals.......................................................................................................................... 49
Table 3-2. Room totals by zone, current.................................................................................. 60
Table 3-3. Redesign room totals by zone, future plan ............................................................. 63
Table 3-4. Summary of room and seat changes from concept to construction documents ..... 66
Table 3-5. Service times, resources, and location summary.................................................... 70
Table 3-6. Summary verification statistics .............................................................................. 76
Table 3-7. Scenarios and current condition control. Latin square experimental design.
Current system scenario based on 2017 Fiscal Year (July ‘16 – June ‘17) ..................... 79
Table 3-8. Demand scenario comparisons ............................................................................... 79
Table 3-9. Simulated patient population .................................................................................. 83
Table 3-10. Summary data for overall length of stay metrics ................................................. 85
Table 3-11. Summary data for length of stay by ESI across runs for all scenarios ................. 91
Table 3-12. Summary data for average WR response variables across runs ........................... 93
Table 3-13. Summary data for Admits zone and RWR response variables across runs ......... 97
Table 3-14. Overall LOS Summary of Current Scenario to Scenario and within Scenario
Differences ....................................................................................................................... 102
Table 3-15. Discharged LOS Summary of Current Scenario to Scenario and within
Scenario Differences ........................................................................................................ 103
Table 3-16. Admitted LOS Summary of Current Scenario to Scenario and within
Scenario Differences ........................................................................................................ 104
xii
Table 3-17. Percent with LOS Greater than 3 Hours Summary of Current Scenario to
Scenario and within Scenario Differences ....................................................................... 105
Table 3-18. Average Number in WR Summary of Current Scenario to Scenario and
within Scenario Differences............................................................................................. 106
Table 3-19. Summary of selection of the best results by response variable ............................ 108
Table 3-20. Simulated patient population for all scenarios, including demand increase
scenarios........................................................................................................................... 110
Table 3-21. Summary of performance metrics across runs for demand increase scenarios .... 111
Table 4-1. Code sketch for graph theoretical approach ........................................................... 124
Table 4-2. Code sketch for placement strategy ....................................................................... 126
Table 4-3. Numeric graph scores ............................................................................................. 138
Table 4-4. Numeric layout scores ............................................................................................ 138
Table 4-5. Hypothesis test results for hypothesis 1 ................................................................. 142
Table 4-6. Hypothesis test results for hypothesis 2 ................................................................. 143
Table 4-7. Pearson correlations and p-values for age, gender, years of experience,
usefulness, and need for decision details ......................................................................... 144
xiii
LIST OF EQUATIONS
Equation 3-1. Critical T-value for screening threshold ........................................................... 80
Equation 3-2. First stage sample mean .................................................................................... 80
Equation 3-3. First stage sample variance ............................................................................... 80
Equation 3-4. Screening threshold ........................................................................................... 80
Equation 3-5. Screening for maximized value ........................................................................ 81
Equation 3-6. Screening for minimized value ......................................................................... 81
Equation 3-7. Rinott’s Constant .............................................................................................. 81
Equation 3-8. Second stage sample sizes ................................................................................ 81
Equation 4-1. Adjacency score ................................................................................................ 128
Equation 4-2. Distance score ................................................................................................... 129
Equation 4-3. Adjacency weighted distance score .................................................................. 129
Equation 4-4. Distance weighted adjacency score .................................................................. 130
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
Castronovo, Yifan Liu, Sreelatha Chunduri, Amanda Webb, Joana Melo, FuJu Wu, and Jesse
Bukenberger for their time listening to me work through those moments of confusion, for their advice
on how to manage the Ph.D. process, for the lunches and coffees that made the long days that much
better, and for the occasional quick turnaround internal reviews of drafts for publications. You all are
amazing.
I would like to thank my family: my mom, dad and sister for their continued support through
life and through making the decision to move across the country to pursue a Ph.D. I would like to
provide a special thank you to Brett Bissinger, who showed me I could find a balance between my
work and life and without whom I wouldn’t have been able to work through many breakthroughs to
complete my dissertation.
Lastly, I would like to acknowledge the sponsors that helped fund this body of work. These
include the Computer Integrated Construction (CIC) Research Group, the Partnership for Achieving
Construction Excellence (PACE), and the US Department of Education. This material is based upon
work partially supported by the US Department of Education Graduate Assistance in Areas of
National Need (GAANN) Contract # P200A180031, project entitled “Integrated Delivery of Ultra-
High-Performance Buildings.” Any opinions, findings, and conclusions or recommendations
expressed in this material are those of the author(s) and do not necessarily reflect the views of the US
Department of Education, the Computer Integrated Construction Research Group, or the Partnership
for Achieving Construction Excellence.
1
Chapter 1.
Introduction
Over half of the inpatient admissions in the US begin with the patient entering the
emergency department (AHQR 2017). Additionally, the number of emergency department (ED)
visits has grown from 2006 to 2014, outpacing the 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 both the first point of care for critical patients and the main source of care for those
who can’t afford or access primary care services (National Academies of Sciences, Engineering,
and Medicine 2007). The typical ED is a highly variable system which serves as the gatekeeper
for a hospital; and with a growing demand, many hospitals are getting bigger.
When investigating changes to a hospital’s operations or redesigning the facility,
simulation tools are useful in planning workflow impact on performance metrics like length of
stay and time to provider; and aid analysis of capacity and room requirements. These tools are not
commonly used in planning a new facility, and only recently have started incorporating static
layout (Taylor et al. 2013). Planners still typically use manual, rule of thumb, and/or experience
methods to arrange and recommend layout choices to solve facility owner problems (Arnolds and
Nickel 2015). Simulation tools fall short of being timely to planners and healthcare administrators
(McGuire 1998). Proposed frameworks in this area have to date been theorized but have been
untested (for examples see Acar et al. (2009); Arnolds and Nickel (2015)). The integration of the
planning tasks, facility layout problems (FLPs), and discrete event simulation (DES) is proposed
to leverage these tools in conjunction, to test different healthcare protocol processes within the
context of layout scenarios to understand layout and design impacts on performance metrics of
2
interest in a healthcare department. These tools when integrated together can potentially provide
critical decision information in context to the healthcare planners. This dissertation investigates
the integration of these techniques and builds a new framework for leveraging data analytics in
the planning of the layout of a healthcare department.
This research is focused on the development and use of a hybrid simulation approach for
a single department in a hospital environment, an emergency department (ED). A case study of a
large emergency department renovation project is used for the context and execution of the hybrid
simulation approach. An ED was identified which was going through an expansion and
renovation, an implementation of a new intake and treatment workflow, and an implementation of
a new electronic medical records (EMR) system. The main problems of the ED were identified as
congestion in the waiting room and lack of connection for staff between the front of the ED and
the patient treatment areas. Early in the conceptual design, the decision makers used a discrete
event simulation (DES) to inform the decisions to expand the ED and to change the front end
intake processes. However, as is typical, when they moved into the design phase, they didn’t
continue to leverage the DES to inform their detailed decisions within the schematic design of the
expansion. This research starts with this base case, and begins with an investigation of the layout
and process pair in a DES to understand how sensitive the DES would be to the layout changes,
secondly an investigation of layout optimization strategies and how end users perceive these
methods is presented, and finally a hybrid modeling framework is developed and presented for
using a data-driven approach to aid decision making throughout a healthcare lifecycle.
Both facility layout problems and discrete event simulation have been investigated for
use in the healthcare application area for the last 35-40 years (Arnolds and Nickel 2015; Jun et al.
1999). Computational methods, communication media, and visualization techniques have
advanced incredibly in the last 15 years, making 3D modeling and 3D visualization easier and
more frequently expected in the fields of design, engineering, and construction. Likewise,
3
complex models such as DES and computationally complex models like facility layout problems
(a class of optimization problems) are expected to be accessible and implementable by healthcare
planners and designers. The rapid industry adoption of building information modeling (BIM) has
allowed for the intelligent modeling of a building’s systems and facilitated a change in the
approach from linear and siloed processes to a greater degree of integration and collaboration in
design and construction as designers, engineers, and contractors coordinate their efforts.
Technology has allowed for rapid development of virtual models to be used in new ways
throughout the design process, such as in engaging end users to confirm that a design meets the
expected use of a facility, design checking, and other options (Castronovo et al. 2013; Liu et al.
2014). In addition, simulation packages, such as discrete event simulation, are increasing their
capabilities by adding animation to simulations to aid model development and validation. Many
of these software applications have relatively simple user interfaces, providing a variety of
professionals with the ability to create compelling visual models with little-to-no design,
animation, or coding experience (Kelton et al. 2014). These technological advances have changed
how the industry performs business and the clients’ expectations from engineers and architects.
Yet there is not enough research into how people use operational research model types in the
design process (O’Keefe 2016). There has been a considerable amount of research in FLPs and
DES but the integration is in its infancy in the literature.
Healthcare design has been a rich application area for evidence based design (EvBD) and
experienced based design (ExBD) practices (Bate and Robert 2007; Ulrich et al. 2008). Hospitals,
labs, medical clinics, etc. have a complex set of design problems and operational considerations.
In the US, these facilities are also subject to market conditions which in effect means patient
satisfaction plays a large role in operational considerations (Brailsford and Vissers 2011). The use
of EvBD aims to address these concerns by utilizing data-driven research into healthcare
practices, such as how patients receive care, how a design impacts time to recovery, how a
4
department gives care (evaluation of a process), and how a design can make a care program more
efficient. The use of ExBD aims to address the problems of healthcare by engaging the users of
the facility within the design process to make their experiences better (e.g., more efficient, higher
quality, better usability, etc.). Both EvBD and ExBD engage the healthcare practitioners and the
patients to gain various expertise and tacit information; both aim to provide better information to
design and engineering teams designing these facilities; and each uses a different strategy to elicit
information to inform the design process.
When designing healthcare facilities, it is important to understand what makes a design
better. Three major aspects of good design are performance, engineering, and usability (Bate and
Robert 2007). Figure 1-1 outlines these three parts of design as the functionality of the design, the
safety and engineering of the design, and the usability and experience of the design expanding the
concept presented by Berkun (2004). EvBD practices fall into the performance and functionality
of healthcare design and ExBD practices fall into the usability and experience of the healthcare
design. The technical, how, is addressed by analytical and knowledge from the engineers to
provide a reasonable solution for human safety. All three in concert create good design. The
decisions made early in the planning and design of facilities impact the functionality, safety, and
usability of these critical infrastructure facilities.
Figure 1-1: Aspects of design (expanded in Bate and Robert 2007, p. 5, originally from
Berkun 2004).
Performance
How well it does the job/is fit for the
purpose
Functionality
Engineering
How safe, well engineered and
reliable it is
Safety
Experience
How the whole interaction with the
product/service ‘feels’/is experienced
Usability
5
EvBD and ExBD are terms for a wide array of methods in using data and end user
feedback, respectively, to inform the design process. One way in which analysts use EvBD is
through analysis and research of past facilities, and additionally operations research methods. In
operations research, there are several ways to evaluate performance of a facility. A popular
method within healthcare is to use discrete event simulation (DES), which simulates how entities
(e.g., patients) move through a set of activities to receive care, typically within a department or
unit (e.g., outpatient care unit, inpatient, clinic, and emergency department). Another area, for
layout specifically, is facility layout problems (FLPs), a class of optimization problems, which are
deterministic models with an equation to optimize, given a static set of conditions, typically
optimizing either flow, distance, or adjacency, in a layout. Flow in a healthcare setting is a very
dynamic element, which is sensitive to temporal changes, e.g., yearly fluctuations in health;
weekly and hourly fluctuations in patients seeking care; and demand, population, and market
changes. These techniques grew out of a focus on using data to drive design decisions, especially
using quantitative analysis for justifying large capital and operational changes.
Simulations of operations in the design phase of a project can provide many
mathematically suitable solutions for specific performance metrics, e.g., length of stay, patient
satisfaction, etc., and can aid redesign, planning, and layout efforts. The suitability of these
solutions is usually filtered down to several options which can be implemented, leaving the key
decisions to the stakeholders and facility owners. During the final decision, stakeholders balance
several economic and patient care options to determine the most suitable design plan.
Both simulation of facility operations and layout optimization can be useful tools for
providing analysis and data to help make design decision, additionally experience of the design
can be used to help designers and stakeholders make key decisions and offer introspection along
the planning, design, and implementation of hospital designs. Yet, these techniques, DES and
FLP, are not readily used in practice nor in conjunction with one another. Even though
6
researchers have advocated for integrated approaches to improve design, implementation within
the design and delivery process is still a needed area for researchers to explore. This research is
focused on exploring how these methods can be combined to improve design decisions which are
data-driven and end user validated.
1.1 Goal and Objectives
The goal of this research is to investigate and develop methods to combine facility layout
and workflow processes to provide a data-driven methodology for healthcare facility design. In
order to achieve the goal of this research, three objectives were developed. Initially the objective
of this research is to understand the impact of the combined effect of layout and workflow
processes on performance metrics in the context of a test case example facility, by investigating
the use of layout considerations in a discrete event simulation. The second objective is to
understand stakeholder understanding and engagement in the use of quantitative layout planning
methodologies, by both developing generative layout approaches and evaluating user perception
of those generated layouts. Finally, the last objective is to develop a methodology using a hybrid
modeling approach including layout optimization, discrete event simulation, and visualization to
implement in healthcare facility layout problems. The hybrid modeling system is meant to
leverage temporal and spatial analysis of design options to improve healthcare facility design
decision making for emergency departments, and more generally healthcare facilities, for use
throughout planning, designing, and operations of a facility.
O1. Understand how facility layout impacts the processes used in the healthcare facility,
and thus the respective operational performance metrics of the individuals using the facility.
O2. Develop and evaluate the use of generative layout methodologies in the context of
healthcare planning and design.
7
O3. Develop an integrated framework for using data-driven and experienced based
methods to generate healthcare layouts for use throughout the facility lifecycle.
1.2 Research Scope
The scope of this research is for a hospital; a single hospital department, an emergency
department, and a set of departments in a hospital, the diagnostic and treatment subset of
departments, were used for the context of the objectives. Within this application area, the scope is
of a simulation of the operations of the facility through discrete event simulation, where the
current operational process are known, the future operations are planned to change, and a design
is meant to support that changing process. In general, the design can be new construction or a
renovation of an existing facility, but is best if there is a clear layout and operational data to be
used in the temporal and spatial simulations. For this study, two examples are used, (1) a case
study based on an emergency department renovation project at Penn State Hershey Medical
Center (PSHMC) in Hershey, PA, which recently went through a master plan and is currently on
Phase 1 of a multi-phase expansion, and (2) a case study based on a new hospital project for a 110
bed hospital planned to be built in Pennsylvania, with a subset of 16 departments associated with
diagnostic and treatment, as well as public an-d service departments. These example cases
provide the context for the study and development of the framework.
1.3 General Methodology Overview
The methodology of this study is focused on various models for use by healthcare
administrators, planners, and designers. A model can be defined as a simplified representation of
the real world, and specifically for computer simulation models “a system of postulates, data, and
8
inferences presented as a mathematical description of an entity or state of affairs” (“Model”
2019). The methodology used was developed to investigate computer simulation models used in
quantitative healthcare design and planning; to investigate their connection between the end use
of healthcare professionals and layout considerations; and to develop a system for integration
throughout the lifecycle of the facility.
The following methodology is used for model creation. Model development follows a 5-
step process: (1) develop goals and objectives of the model and analysis, (2) develop a model
concept, (3) translate the model into a virtual simulation of the real-world system, (4) use the
model for analysis and recommendations, and (5) implement changes from recommendations in
the real world (Figure 1-2). The focus of this research is on the integration of these techniques in
model creation and in model usage steps.
Figure 1-2: Generic model process diagram. Model development in simulation,
optimization, and visualization follow these general steps
For this research, there are five phases: Initial Framework Development, Development of
the DES, Layout of the DES, Development of the FLP (implemented as a Graph Theoretical
Approach), Generative Layouts Evaluation, and Finalization of Framework. Figure 1-3 depicts
each phase of the methodology and the associated research activities. Additionally, how these
research phases fit into each of the research objectives is shown. An initial literature review was
performed on operational research, layout planning and visualization for use in healthcare
planning and design. Next a discrete event simulation was developed and analyzed, as part of
Objective 1. Following that, layout optimization was studied through the development of a graph
theoretical approach with layout placement procedure to generate layouts. These layouts were
Goal and Objective
Model Conceptualization
Model Creation Model Usage Implementation
9
evaluated through an objective and subjective scoring method, including survey with experts in
healthcare planning and design, as part of Objective 2. Lastly, the initial framework was revised
and developed further to incorporate the analysis, results, and implications from the previous
studies, as part of Objective 3. For methodology details, please see the associated sections for
each chapter (O1: Section 3.4, O2: Section 4.2, and O3: Section 5.4).
Figure 1-3: Methodology overview diagram with research phases and activities
Objective 1
Objective 2
Objective 3
Development of Initial
Framework
•Literature Review
•System Requirements
• IRB Protocols
•Data Collection
Development of DES
• Input Data Collection & Analysis
•Scenario Development•Conceptual Model
Layout in DES
•Develop Experimental Design
•Conduct Experiment
•Data Analysis
Development of GTA
• Input Data Collection
•Programming GTA & Layout Procedure•Generate Layouts
Generative Layouts
Evaluation
•Development of Survey Tools
•Participant Recruitment
•Conduct Survey
•Data Analysis
Finalization of Framework
10
Chapter 2.
Literature Review
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)
Complete checkout 5 8 Uniform in no. avail? Registration (Swenson 2008)
Lab draw 3 6 Uniform in zone & sch. dep. Nurse (Swenson 2008)
Lab results (complex) 20 60 Uniform in 10 Lab Dept.
Change (was ~u(40,80), applied
to 2s and 3s a proportion of time
Lab results (limited) 10 20 Uniform in 10 Lab Dept. (Swenson 2008)
Assess lab results 3 6 Uniform (remote) schedule dep. Doctor
(Swenson 2008), pneumatic tube
system
Administer X-Ray 7 15 Uniform out 2 ED Radiology (Swenson 2008)
Assess X-Ray 3 6 Uniform out schedule dep. Doctor (Swenson 2008)
Administer EKG 1 3 Uniform in zone & sch. dep. Nurse (Swenson 2008)
Develop EKG 2 10 Uniform (remote) zone & sch. dep. Nurse (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
Cath Lab 30 50 Uniform leave 2 Cath Lab Estimate (Kern 2008)
Consult 5 10 Uniform in schedule dep. Doctor
Estimate, same as doctor visit,
additional doctor resources not
modeled
CT Scan 15 30 Uniform out 2 ED Radiology
Estimate (St. Michael’s Hospital
2019)
CT Results 2 10 Uniform out 1 ED Radiology
Estimate (St. Michael’s Hospital
2019)
ECHO 20 40 Uniform leave 2
Stress/ECHO
Dept. Estimate (Almed 2017)
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)
Current 4.981 (4.77, 5.19) 3.711 (3.48, 3.94) 8.585 (8.42, 8.75) 63.14% (61.55%, 64.74%)
S1 4.929 (4.747, 5.111) 3.660 (3.465, 3.856) 8.518 (8.374, 8.662) 63.26% (61.89%, 64.62%)
S2 4.228 (4.183, 4.273) 2.932 (2.887, 2.977) 7.906 (7.86, 7.951) 53.42% (52.77%, 54.07%)
S3 4.198 (4.112, 4.284) 3.299 (3.208, 3.391) 6.729 (6.657, 6.801) 56.81% (55.80%, 57.81%)
S4 3.792 (3.749, 3.835) 2.885 (2.841, 2.93) 6.372 (6.331, 6.413) 49.49% (48.73%, 50.26%)
S5 4.577 (4.514, 4.641) 3.276 (3.211, 3.341) 8.245 (8.184, 8.306) 59.51% (58.69%, 60.33%)
S6 4.324 (4.272, 4.377) 3.025 (2.972, 3.078) 8.004 (7.954, 8.053) 55.24% (54.46%, 56.01%)
S7 4.079 (4.023, 4.135) 3.181 (3.124, 3.238) 6.618 (6.566, 6.669) 55.23% (54.37%, 56.09%)
S8 3.959 (3.909, 4.008) 3.055 (3.004, 3.106) 6.507 (6.46, 6.553) 52.64% (51.83%, 53.46%)
S9 4.935 (4.8, 5.071) 3.666 (3.521, 3.811) 8.518 (8.404, 8.633) 63.47% (62.31%, 64.63%)
S10 4.252 (4.2, 4.303) 2.954 (2.902, 3.006) 7.929 (7.877, 7.981) 53.78% (53.04%, 54.51%)
S11 4.203 (4.107, 4.299) 3.308 (3.207, 3.41) 6.726 (6.646, 6.807) 56.55% (55.37%, 57.73%)
S12 3.773 (3.727, 3.818) 2.862 (2.816, 2.907) 6.362 (6.321, 6.404) 48.99% (48.17%, 49.81%)
S13 4.560 (4.493, 4.628) 3.261 (3.192, 3.33) 8.218 (8.154, 8.282) 59.36% (58.46%, 60.26%)
S14 4.330 (4.274, 4.385) 3.031 (2.973, 3.088) 8.010 (7.956, 8.064) 55.37% (54.53%, 56.21%)
S15 4.124 (4.058, 4.189) 3.224 (3.157, 3.292) 6.658 (6.598, 6.718) 55.53% (54.61%, 56.44%)
S16 3.937 (3.898, 3.975) 3.034 (2.995, 3.072) 6.486 (6.446, 6.526) 52.32% (51.62%, 53.02%)
S17 5.367 (5.201, 5.533) 4.520 (4.344, 4.697) 7.753 (7.614, 7.891) 72.09% (70.79%, 73.40%)
S18 26.089 (23.084, 29.095) 26.806 (23.563, 30.049) 24.077 (21.74, 26.413) 98.28% (97.93%, 98.63%)
86
Figure 3-9. Box plots for average LOS of all patients across runs, Current and S1
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and quantiles (blue).
Figure 3-10. Box plots for average LOS of all patients across runs, S1-S16
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and quantiles (blue).
87
Figure 3-11. Box plots for average LOS of discharged patients across runs, Current and S1
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and quantiles (blue).
Figure 3-12. Box plots for average LOS of discharged patients across runs, S1-S16
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and quantiles (blue).
88
3.6.4 Length of Stay for Admitted Patients
Admitted patients LOS is summarized in Table 3-10. Admitted patients are mostly ESI 1s
and 2s (82% of 1s and 45% of 2s), and about 21% of ESI 3s. The average LOS for the S1 was not
significantly different than Current, see box plots in Figure 3-13. The box plots for S1-S16 are in
Figure 3-14. For admitted patients, the addition of an Admits zone contributed to the largest
reduction in LOS. Adding the RWR with Admits contributed to a greater reduction in LOS. The
use of care initiation and addition of FT bays had limited impact on LOS. The best scenario in for
admitted LOS was S12, with an average (95% CI) LOS of 6.362 hrs (6.321, 6.404). The top 4
scenarios in increasing LOS were S12, S4, S16, and S8. All of the best scenarios had both Admits
zone and RWR. The FT bays were added in S8 and S16. The care initiation was in S12 and S16.
3.6.5 Percent of Patients with LOS greater than 3 hours
The percent of patients with a length of stay longer than 3 hours is a measurement of risk.
It follows a similar pattern as the overall LOS, with some differences. A summary of the statistics
for each scenario is presented in Table 3-10. The box plots for Current and S1 are available in
Figure 3-15. The box plots for S1-S26 are in Figure 3-16. The best in system was S12.
3.6.6 Length of Stay by Acuity
A summary of the LOS by ESI Acuity levels is presented in Table 3-11. Additional box
plots for each scenario is available in the Appendix A. No significant differences were found
between Current and S1.
89
Figure 3-13. Box plots for average LOS of admitted patients across runs, Current and S1
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and quantiles (blue).
Figure 3-14. Box plots for LOS of admitted patients across runs, S1-S16
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and quantiles (blue).
90
Figure 3-15. Box plots for average percent of LOS longer than 3 hours across runs for all
patients, Current and S1
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and quantiles (blue).
Figure 3-16. Box plots for average percent of LOS longer than 3 hours across runs for all
patients, S1-S16
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and quantiles (blue).
91
Table 3-11. Summary data for length of stay by ESI across runs for all scenarios
Scenario Acuity1 LOS (hr) Acuity2 LOS (hr) Acuity3 LOS (hr) Acuity4 LOS (hr) Acuity5 LOS (hr)
Mean (CIl,Ciu) Mean (CIl,Ciu) Mean (CIl,Ciu) Mean (CIl,Ciu) Mean (CIl,Ciu)
Current 4.877 (4.76, 4.99) 6.521 (6.41, 6.63) 4.785 (4.55, 5.02) 3.923 (3.64, 4.2) 3.576 (3.27, 3.88)
S1 4.910 (4.811, 5.01) 6.485 (6.388, 6.581) 4.731 (4.529, 4.933) 3.853 (3.6, 4.105) 3.438 (3.189, 3.686)
S2 4.462 (4.405, 4.519) 6.032 (5.979, 6.085) 4.079 (4.029, 4.129) 2.741 (2.691, 2.792) 2.604 (2.545, 2.663)
S3 4.738 (4.676, 4.8) 5.378 (5.32, 5.436) 4.045 (3.955, 4.135) 3.354 (3.243, 3.465) 2.938 (2.826, 3.05)
S4 4.449 (4.401, 4.498) 5.114 (5.075, 5.153) 3.683 (3.639, 3.727) 2.663 (2.614, 2.711) 2.563 (2.502, 2.623)
S5 4.741 (4.664, 4.818) 6.315 (6.242, 6.387) 4.388 (4.32, 4.456) 3.284 (3.209, 3.359) 2.925 (2.85, 3.001)
S6 4.563 (4.5, 4.626) 6.106 (6.045, 6.168) 4.160 (4.104, 4.216) 2.896 (2.836, 2.956) 2.667 (2.599, 2.734)
S7 4.677 (4.623, 4.73) 5.323 (5.279, 5.367) 3.932 (3.873, 3.992) 3.152 (3.085, 3.218) 2.735 (2.663, 2.807)
S8 4.582 (4.532, 4.632) 5.230 (5.19, 5.269) 3.830 (3.775, 3.884) 2.943 (2.886, 2.999) 2.657 (2.597, 2.717)
S9 4.917 (4.824, 5.01) 6.488 (6.398, 6.579) 4.745 (4.598, 4.892) 3.844 (3.66, 4.028) 3.455 (3.283, 3.626)
S10 4.482 (4.416, 4.549) 6.061 (5.999, 6.124) 4.099 (4.048, 4.15) 2.762 (2.702, 2.822) 2.656 (2.59, 2.722)
S11 4.716 (4.639, 4.793) 5.370 (5.301, 5.44) 4.058 (3.959, 4.158) 3.356 (3.235, 3.476) 2.950 (2.823, 3.077)
S12 4.423 (4.379, 4.468) 5.089 (5.047, 5.131) 3.666 (3.619, 3.714) 2.639 (2.59, 2.688) 2.549 (2.491, 2.607)
S13 4.722 (4.648, 4.796) 6.284 (6.208, 6.36) 4.372 (4.301, 4.443) 3.279 (3.199, 3.359) 2.871 (2.791, 2.951)
S14 4.554 (4.487, 4.622) 6.132 (6.068, 6.197) 4.162 (4.104, 4.221) 2.889 (2.831, 2.947) 2.696 (2.627, 2.766)
S15 4.733 (4.672, 4.794) 5.355 (5.301, 5.409) 3.978 (3.91, 4.047) 3.205 (3.13, 3.28) 2.780 (2.69, 2.869)
S16 4.555 (4.509, 4.6) 5.211 (5.176, 5.246) 3.808 (3.767, 3.848) 2.919 (2.876, 2.961) 2.649 (2.592, 2.705)
S17 5.699 (5.595, 5.802) 6.266 (6.168, 6.363) 5.358 (5.172, 5.544) 4.455 (4.256, 4.654) 4.232 (4.011, 4.454)
S18 14.393
(13.192,
15.593) 14.868
(13.731,
16.005) 29.215
(25.723,
32.706) 31.078 (27.152, 35.003) 31.010
(27.041,
34.978)
92
3.6.7 WR Waiting Time and Number Waiting
Acuity ESI 1 and 2 patients were routed directly to rooms when they arrived, thus waiting
time in the WR represents the ESI 3-5 patients, from both ambulance and walk-in arrivals. There
was no significant difference found between Current and S1 for waiting time, number in the WR,
and maximum in WR. The box plots for Current and S1 for the waiting time is in Figure 3-17, for
the number in the WR is in Figure 3-19, and for the maximum in the WR in Figure 3-21. The
average (CI) maximum in the WR for Current was 68.04 people (59.74, 76.34), with an average
(CI) of 4.6 people (3.22, 6.07) in the WR. The best scenario for all three performance metrics was
S4, with an average (CI) wait time of 4.1 minutes (3.413, 6.462), 0.85 people (.711, .995), and a
maximum of 41.4 people (38.164, 44.636) in the WR. The top 4 scenarios in increasing order of
waiting time are S4, S12, S16, and S8, all using Admits zone and RWR, S8 and S16 with
additional FT bays, and S12 and S16 using care initiation.
Figure 3-17. Box plot of average time in WR (minutes) across runs, Current and S1
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(blue).
93
Table 3-12. Summary data for average WR response variables across runs
Scenario Waiting Time (min) Max in WR Number in WR
Mean (CIl,Ciu) Mean (CIl,Ciu) Mean (CIl,Ciu)
Current 21.927 (15.31, 28.54) 68.04 (59.74, 76.34) 4.646 (3.22, 6.07)
S1 19.534 (13.892, 39.282) 64.62 (57.954, 71.286) 4.147 (2.932, 5.362)
S2 6.815 (5.83, 10.266) 46.98 (43.055, 50.905) 1.430 (1.223, 1.638)
S3 8.910 (6.705, 16.628) 52.22 (47.031, 57.409) 1.870 (1.404, 2.335)
S4 4.091 (3.413, 6.462) 41.4 (38.164, 44.636) 0.853 (.711, .995)
S5 10.408 (8.978, 15.415) 53.94 (50.086, 57.794) 2.197 (1.892, 2.502)
S6 7.202 (6.027, 11.312) 47.8 (43.792, 51.808) 1.513 (1.266, 1.761)
S7 6.287 (5.195, 10.108) 46.54 (43.34, 49.74) 1.317 (1.086, 1.547)
S8 5.618 (4.613, 9.135) 48.68 (44.571, 52.789) 1.173 (.961, 1.384)
S9 19.378 (15.258, 33.799) 67.16 (60.445, 73.875) 4.111 (3.231, 4.99)
S10 6.815 (5.83, 10.266) 49.44 (45.743, 53.137) 1.536 (1.324, 1.749)
S11 9.520 (7.242, 17.498) 52.48 (47.236, 57.724) 1.997 (1.515, 2.48)
S12 4.191 (3.537, 6.482) 42.8 (39.797, 45.803) 0.874 (.736, 1.012)
S13 9.950 (8.425, 15.288) 51.7 (47.684, 55.716) 2.099 (1.776, 2.421)
S14 7.171 (6.069, 11.03) 47.42 (43.741, 51.099) 1.508 (1.274, 1.742)
S15 7.332 (5.901, 12.34) 51.46 (46.668, 56.252) 1.536 (1.235, 1.838)
S16 5.134 (4.488, 7.397) 45.16 (42.163, 48.157) 1.070 (.935, 1.205)
S17 33.128 (28.029, 50.975) 83.96 (77.449, 90.471) 7.364 (6.224, 8.505)
S18 861.87 (734.017, 1309.405) 422.06 (371.248, 472.872) 200.59 (170.89, 230.29)
Figure 3-18. Box plots for average waiting time in WR (minutes) across runs, S1-S16
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(blue).
94
Figure 3-19. Box plots for average number in WR across runs, Current and S1
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and the quantiles (blue).
Figure 3-20. Box plots for average number in WR across runs, S1-S16
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and the quantiles (blue).
95
Figure 3-21. Box plots for average maximum number in WR across runs, Current and S1
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and the quantiles (blue).
Figure 3-22. Box plots for average maximum number in WR across runs, S1-S16
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and the quantiles (blue).
96
3.6.8 Results Waiting Room Analysis
The Results Waiting Room (RWR) was modeled in S2, S4, S6, S8, S10, S12, S14, and
S16. Patients in the FT were routed to RWR if they needed only either one in room service and or
one out of room service. Only patients with ESI 4 and 5 were routed to the FT area. A summary
of the average number in the RWR and the maximum number is summarized in Table 3-12. Box
plots for average number in RWR is in Figure 3-23, and for maximum number in RWR is in
Figure 3-24. The addition of the FT bays reduced the number in the RWR considerably. The
addition of Admits zone with RWR reduced the average number in the RWR. There were no
changes between the maximum number in the RWR across the different scenarios. The maximum
number in the RWR was between 7.1 (6.874, 7.246) and 9.3 (8.915, 9.645) across scenarios,
which is lower than the seating space allocated.
3.6.9 Number in Admits Zone
The Admits zone routing for patients waiting for beds was modeled in S3, S4, S7, S8,
S11, S12, S15, and S16. Patients who were admitted but still waiting for a bed (with the
exception of any ESI 1 patients) were routed to the Admits zone to wait for transfer to an in-
patient bed, typically these are ESI 2 and 3 patients. When used, the average number in the
Admits zone was similar across all scenarios. Since most of the routing changes were for front of
house and the FT zone, the results are not surprising. Table 3-13 summarizes the average number
in the Admits zone. The box plots show that there was only slight variation in the mean and the
spread across the scenarios were consistent (Figure 3-25). Across all scenarios the maximum
number in Admits zone was 7, the total number of beds allotted for that zone.
97
Table 3-13. Summary data for Admits zone and RWR response variables across runs
Scenario Number in Admits Number in RWR Max in RWR
Mean (CIl,Ciu) Mean (CIl,Ciu) Mean (CIl,Ciu)
S2 0.754 (.743, .765) 7.9 (7.61, 8.19)
S3 3.590 (3.576, 3.604)
S4 3.605 (3.591, 3.619) 0.728 (.718, .738) 7.06 (6.874, 7.246)
S6 0.552 (.536, .569) 9.2 (8.809, 9.591)
S7 3.603 (3.589, 3.617)
S8 3.620 (3.607, 3.634) 0.384 (.369, .398) 8.66 (8.33, 8.99)
S10 0.754 (.743, .766) 7.9 (7.628, 8.172)
S11 3.585 (3.571, 3.598)
S12 3.623 (3.611, 3.635) 0.723 (.712, .734) 7.4 (7.123, 7.677)
S14 0.563 (.547, .579) 9.28 (8.915, 9.645)
S15 3.608 (3.597, 3.62)
S16 3.624 (3.61, 3.638) 0.377 (.362, .391) 8.56 (8.213, 8.907)
S17 3.579 (3.566, 3.592) 0.691 (.665, .718) 9.76 (9.486, 10.034)
S18 3.525 (3.512, 3.538) 1.161 (1.143, 1.178) 11.22 (10.902, 11.538)
Figure 3-23. Box plots of average number in RWR across runs
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(blue).
98
Figure 3-24. Box plots for maximum number in RWR across runs
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(blue).
Figure 3-25. Box plots for the average number in Admits zone across runs
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(blue).
99
3.6.10 Summary of How Layout Impacts Performance Measures
The decision variables that impacted the response variables were different across
different types of response variables. The main layout contributors to the reduction of LOS for all
patients was the Results Waiting and Admits zones. The main layout contributor to the reduction
of LOS of discharged patients was the use of RWR. The main layout contributor to the reduction
of LOS of admitted patients was the use of the Admits zone. The main layout contributor to the
reduction of the % of patients staying for over 3 hours in the ED was the use of the RWR. In
addition to the main response variables, the main layout contributors to the reduction in the
number in the WR was the RWR and Admits zone. The most surprising result was the impact of
routing patients to RWR on the percent of patients staying longer than 3 hours, which lowered
LOS significantly across all patients.
For each of these response variables, including time spent in the WR, the differences
between first the scenarios and the current scenario were calculated, and then each pair in the
fully crossed scenario testing were analyzed using a Bonferroni adjusted alpha for each
difference. For our 4 layout factors, there were 𝑛 = 16 combinations run, and thus (𝑛(𝑛 −
1)/2) = 120 non-ordered combination pairs. Each test for difference from each Scenario to the
Current Scenario (the baseline scenario) was found significant if the p-value was lower than 𝛼/𝑚,
where 𝛼 is set at 5% and 𝑚 is the number of scenarios being compared with the baseline scenario.
Each test for difference within Scenarios was found significant if the p-value was lower than
𝛼/𝑚, where 𝛼 is set at 5% and 𝑚 is the number of combinations in the number of tests run, thus
the individual difference would be significant if found to have a p-value lower than 0.05/120 =
0.0004167, when comparing all 16 options, and 120 combinations. All tests were performed
with the scenarios run for 100 days, with 50 replications, and common random numbers. The
100
summary of the differences that were found significant for overall LOS is in Table 3-14, for
discharged LOS is in
Table 3-15, for admitted LOS is in
Table 3-16, for % LOS greater than 3 hours is in
Table 3-17, and for the number in the WR is in Table 3-18.
The first test of differences, comparing the Scenarios to the Current Scenario, showed
that the S1 had no significant differences, thus paths in this model did not significantly impact
these 5 performance measures. Two of five layout conditions contributed to the most amount of
improvement over the baseline condition: results waiting (45.17 minutes, a 15.1% improvement
on overall patient LOS, with Bonferroni adjusted CI between 8.40% and 21.83%) and admits
waiting (47.00 minutes, or 15.7%, between 8.54% and 22.91%). A combined improvement was
estimated to be 1.19 hours (23.87%, between 17.11% and 30.63%) for overall LOS. The greatest
improvement over the baseline Current Scenario was found with S12, with an improvement of a
72.51 minute reduction for overall LOS, a 50.96 minute reduction for discharged LOS, a 133.35
minute reduction for admitted LOS, a 14.15% reduction for percent of patients with LOS greater
than 3 hours, and a 3.77 reduction in average number in the WR.
For the differences with Scenarios, analysis reveals which layout factor played the most
role in these differences. For overall LOS, RWR and Admits zone had the most impact on LOS,
42.05 and 43.87 minutes, respectively (S1-S2 and S1-S3), and 68.21 minutes combined (S1-S4).
There was no significant difference between the two (S2-S3). The addition of fast track bays
reduced the overall improvement by approximately 10 minutes (Bonferroni adjusted CI: 8.51
min, 11.5 min, S4-S8). The addition of the CIA was not significant on its own (S1-S9), but once
RWR and Admits zone were in the scenario, the CIA helped reduce LOS an additional 1.17
minutes (S4-S12). The RWR had the greatest impact on the discharged LOS, 43.69 minutes (S1-
S2), whereas the FT bays reduced LOS 23.09 minutes (S1-S5), and the Admits zone reduced LOS
101
21.66 minutes, and CIA was not significant on its own (S1-S9). Including both RWR and FT
reduced LOS 5.56 minutes less than RWR alone (S2-S6). The best scenario was S12, with a
47.91 minute reduction on LOS, 1.40 minutes better than S4. For admitted patients, the addition
of Admits zone made the biggest impact alone, 107.03 minutes reduced LOS (S1-S3), whereas
adding RWR reduced LOS by 36.73 minutes on average (S1-S2), adding FT alone reduced LOS
16.38 minutes on average (S1-S5), and CIA had no initial impact (S1-S9). Combining Admits
zone, RWR, and CIA had the biggest improvement of 129.33 minutes (S1-S12) for admitted
patients, however this was not significantly different than using Admits zone and RWR without
CIA (S4-S12). The largest contributor to reducing percent of patients whose LOS is greater than 3
hours was the RWR, with a reduction of 9.84% (S1-S2), with Admits alone 6.45% (S1-S3), FT
alone 3.75% (S1-S5), and CIA no significant difference in patients staying greater than 3 hours
(S1-S9). Overall the best scenario was S12 with a 14.26% reduction in patients staying greater
than 3 hours (S1-S12), however this was only 0.50% better than S4, with RWR and admits. For
number waiting in the WR, both RWR and Admits zone had the same impact, 2.72 and 2.28
people on average less in the WR (S1-S2 and S1-S3). There was no significant difference
between the two (S2-S3), and combined they reduced the on average number in the WR by 3.29
(S1-S4), which was the greatest reduction for the average number in the WR. There was no
significant difference for average number in the WR between S4 and S12.
102
Table 3-14. Overall 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
within Scenario difference tests, m = 120
Cur S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16
Cur ns* 45.17 47.00 71.34 24.23 39.40 54.12 61.35 ns* 43.76 46.70 72.51 25.24 39.08 51.45 62.66
S1 42.05 43.87 68.21 21.10 36.28 50.99 58.22 ns** 40.63 43.57 69.38 22.12 35.95 48.32 59.53
S2 ns** 26.16 -20.94 -5.77 8.94 16.17 -42.42 ns** ns** 27.34 -19.93 -6.09 6.28 17.49
S3 24.34 -22.77 -7.59 7.12 14.35 -44.24 ns** ns** 25.51 -21.75 -7.92 4.46 15.67
S4 -47.11 -31.93 -17.22 -9.99 -68.58 -27.58 -24.64 1.17 -46.09 -32.26 -19.89 -8.68
S5 15.17 29.89 37.12 -21.48 19.53 22.47 48.28 ns** 14.85 27.22 38.43
S6 14.71 21.94 -36.65 4.36 7.29 33.11 -14.16 ns** 12.05 23.26
S7 7.23 -51.36 -10.35 -7.42 18.39 -28.87 -15.04 -2.66 8.55
S8 -58.59 -17.59 -14.65 11.16 -36.10 -22.27 -9.89 ns**
S9 41.01 43.94 69.76 22.49 36.33 48.70 59.91
S10 ns** 28.75 -18.52 -4.68 7.69 18.90
S11 25.82 -21.45 -7.62 4.76 15.97
S12 -47.27 -33.43 -21.06 -9.85
S13 13.84 26.21 37.42
S14 12.37 23.58
S15 11.21
S16
103
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
within Scenario difference tests, m = 120
Cur S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16
Cur ns* 46.74 24.72 49.56 26.14 41.18 31.81 39.37 ns* 45.45 -22.58 50.96 27.02 40.83 29.20 40.66
S1 43.69 21.66 46.51 23.09 38.13 28.75 36.32 ns** 42.40 21.11 47.91 23.97 37.78 26.15 37.60
S2 -22.02 2.82 -20.60 -5.56 -14.94 -7.37 -44.03 ns** -22.58 4.22 -19.72 -5.91 -17.54 -6.09
S3 24.84 ns** 16.46 7.09 14.66 -22.01 20.73 ns** 26.24 ns** 16.11 4.48 15.94
S4 -23.42 -8.38 -17.76 -10.19 -46.85 -4.11 -25.40 1.40 -22.54 -8.73 -20.36 -8.91
S5 15.04 5.67 13.23 -23.43 19.31 ns** 24.82 ns** 14.69 3.06 14.52
S6 -9.37 -1.80 -38.47 4.27 -17.02 9.78 -14.15 ns** -11.98 ns**
S7 7.57 -29.09 13.64 -7.65 19.15 -4.78 9.03 -2.60 8.85
S8 -36.66 6.07 -15.21 11.59 -12.35 1.46 -10.17 ns**
S9 42.74 21.45 48.25 24.31 38.12 26.49 37.94
S10 -21.29 5.51 -18.42 -4.62 -16.25 -4.79
S11 26.80 ns** 16.67 5.04 16.50
S12 -23.94 -10.13 -21.76 -10.30
S13 13.81 2.18 13.63
S14 -11.63 ns**
S15 11.45
S16
104
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
within Scenario difference tests, m = 120
Cur S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16
Cur ns* 40.75 111.32 132.77 20.40 34.87 118.02 124.68 ns* 39.34 111.51 133.35 22.01 34.48 115.59 125.94
S1 36.73 107.30 128.75 16.38 30.85 114.00 120.66 ns** 35.32 107.48 129.33 17.99 30.46 111.57 121.92
S2 70.57 92.02 -20.35 -5.88 77.27 83.93 -36.76 -1.41 70.75 92.60 -18.75 -6.27 74.84 85.19
S3 21.45 -90.92 -76.45 6.70 13.37 -107.32 -71.98 ns** 22.03 -89.31 -76.84 4.27 14.62
S4 -112.38 -97.90 -14.75 -8.09 -128.78 -93.43 -21.27 ns** -110.77 -98.29 -17.18 -6.83
S5 14.47 97.62 104.29 -16.40 18.94 91.11 112.95 ns** 14.08 95.20 105.54
S6 83.15 89.81 -30.87 4.47 76.64 98.48 -12.86 ns** 80.72 91.07
S7 6.66 -114.03 -78.68 -6.52 15.33 -96.01 -83.54 -2.43 7.92
S8 -120.69 -85.34 -13.18 8.66 -102.68 -90.20 -9.09 ns**
S9 35.34 107.51 129.35 18.01 30.49 111.60 121.95
S10 72.17 94.01 -17.33 -4.86 76.25 86.60
S11 21.84 -89.50 -77.02 4.09 14.44
S12 -111.34 -98.87 -17.76 -7.41
S13 12.47 93.59 103.94
S14 81.11 91.46
S15 10.35
S16
105
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
within Scenario difference tests, m = 120
Cur S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16
Cur ns* 9.72% 6.33% 13.65% 3.63% 7.90% 7.91% 10.50% -0.33% 9.37% 6.59% 14.15% 3.78% 7.77% 7.61% 10.82%
S1 9.84% 6.45% 13.76% 3.75% 8.02% 8.03% 10.61% ns** 9.48% 6.71% 14.26% 3.90% 7.89% 7.73% 10.94%
S2 -3.39% 3.93% -6.09% -1.82% -1.81% 0.78% -10.05% -0.35% -3.13% 4.43% -5.94% -1.95% -2.11% 1.10%
S3 7.31% -2.70% 1.57% 1.58% 4.16% -6.66% 3.03% ns** 7.81% -2.55% 1.44% 1.28% 4.48%
S4 -10.02% -5.74% -5.73% -3.15% -13.98% -4.28% -7.06% 0.50% -9.87% -5.88% -6.03% -2.83%
S5 4.27% 4.28% 6.87% -3.96% 5.73% 2.96% 10.52% ns** 4.14% 3.98% 7.19%
S6 ns** 2.59% -8.23% 1.46% -1.31% 6.24% -4.12% ns** ns** 2.92%
S7 2.58% -8.24% 1.45% -1.32% 6.23% -4.13% ns** -0.30% 2.91%
S8 -10.83% -1.13% -3.91% 3.65% -6.72% -2.73% -2.88% ns**
S9 9.69% 6.92% 14.48% 4.11% 8.10% 7.94% 11.15%
S10 -2.77% 4.78% -5.58% -1.59% -1.75% 1.45%
S11 7.56% -2.81% 1.18% 1.02% 4.23%
S12 -10.37% -6.38% -6.53% -3.33%
S13 3.99% 3.83% 7.04%
S14 ns** 3.05%
S15 3.20%
S16
106
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
Cur S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16
Cur ns* 3.22 2.78 3.79 2.45 3.13 3.33 3.47 ns* 3.11 2.65 3.77 2.55 3.14 3.11 3.58
S1 2.72 2.28 3.29 1.95 2.63 2.83 2.97 ns** 2.61 2.15 3.27 2.05 2.64 2.61 3.08
S2 ns** 0.58 -0.77 ns** 0.11 0.26 -2.68 -0.11 -0.57 0.56 -0.67 ns** ns** 0.36
S3 1.02 ns** ns** 0.55 0.70 -2.24 ns** ns** 1.00 ns** ns** ns** 0.80
S4 -1.34 -0.66 -0.46 -0.32 -3.26 -0.68 -1.14 ns** -1.25 -0.66 -0.68 -0.22
S5 0.68 0.88 1.02 -1.91 0.66 ns** 1.32 0.10 0.69 0.66 1.13
S6 0.20 0.34 -2.60 ns** -0.48 0.64 -0.59 ns** ns** 0.44
S7 0.14 -2.79 -0.22 -0.68 0.44 -0.78 -0.19 -0.22 0.25
S8 -2.94 -0.36 -0.82 0.30 -0.93 -0.34 -0.36 ns**
S9 2.57 2.11 3.24 2.01 2.60 2.57 3.04
S10 ns** 0.66 -0.56 ns** ns** 0.47
S11 1.12 ns** ns** 0.46 0.93
S12 -1.22 -0.63 -0.66 -0.20
S13 0.59 0.56 1.03
S14 ns** 0.44
S15 0.47
S16
107
3.6.11 Comparison of the Best in System
The best scenario was found to be S12 (RQ1b). A summary of the selection of the best
results with the top selection and next three ‘good’ solutions are presented in Table 3-19. The
next best solution was S4, which had a near best metric for all main response variables and a best
in set rating for waiting room response variables. S16 was the third best option, as it was near best
for every metric except one (discharged patient LOS). The results are surprising, as one would
expect the best in set would be the solution offering all the implementation changes (i.e., solution
S16). That solution is a good solution, but S4 which had less changes performed better. S4 didn’t
have care initiation or the additional fast track beds. S12, the best solution, did use care initiation,
but didn’t use the additional fast track beds either. These results indicate relative performance
under the baseline demand conditions. While S12 was found to be the best, S4 was a close
second. However, S16 was not as close. Care initiation was not significantly reducing the
performance measures unless it was in combination with RWR and admits. The use of FT
detracted from the impact of the RWR. These make sense, as care initiation brings people to a
room faster, but additional workflow changes are needed in how that patient is cared for, thus, if
the back of house is inefficient, bringing them back there faster doesn’t overall give the patient a
lower LOS.
108
Table 3-19. Summary of selection of the best results by response variable
Scenario
Main Response Variable Additional Response Variables
Average
LOS
Discharged
LOS
Admitted
LOS
% LOS >
3hr WR Wait Time
WR Max
Number
S1
S2 n
S3
S4 n n n n X X
S5
S6
S7 n
S8 n n n n
S9
S10 n
S11
S12 X X X X n n
S13
S14
S15
S16 n n n n n
Note: X = best in set, n= near best solution.
3.6.12 Opportunities for Space Allocation
To answer RQ2: “Were there opportunities to optimize space allocation?,” the WR,
RWR, Care Initiation, and Admits resources and utilization can be summarized. In the WR, there
was a programmed 41 seats. Across runs, the best performing simulation had an average of 0.874
people in the WR and a maximum across runs of 42.8. Although, it might not be best to design
for the absolute maximum given these are typically extreme cases and with potentially long
maximum tails in the distribution, the WR is close to adequately sized for the maximum case in
the best scenario. In Current the average number in the WR across runs was found to be 4.65 with
a maximum of 68.04. For the RWR, there was a programmed 36 seats. However, the average
number across runs for the RWR in the best scenario (S12) was 0.723 and the maximum number
was 7.4. The space allocated for the RWR was much higher than the need found in the
109
simulation. For all waiting areas, these numbers did not include any family or persons
accompanying patients, which would also be waiting in these areas. Estimates could be made if
that was deemed important to the ED. Care initiation was programmed to have 4 rooms. The
impact of 2 triage vs 4 care initiations did provide a positive impact on the overall goals for
reduction of the LOS, even though the impact was relatively small. The balance between the
magnitude of that impact and the amount of space needed could be considered by administrators.
Although, a more comprehensive analysis should be performed to model the changing resources
before making a judgement on the actual magnitude of the impact in a given emergency
department study. Finally, the Admits zone had a considerable impact on the overall LOS of the
ED, especially for high acuity patients. Analysis for the size of the Admits zone showed that the
maximum was reached in all scenario runs, the space had a high utilization rate, and the average
number in the Admits zone across runs in the best scenario was 3.6. Given that the maximum was
reached consistently and a high utilization rate, it would be an area that would warrant additional
analysis to determine appropriate sizing under the future workflow and configuration conditions.
3.6.13 Future Demand Projections
The future demands were projected in the S17 and S18. A summary of the population
data in each simulation is shown in Table 3-20. The expected population increase was 12 more
patients per day on average. For S17, the simulation population average across runs increased
from 202.5/day to 214.1/day, an increase of 11.6/day. For S18, the population average across runs
increased to 226.5/day, an increase of 23/day above the Current conditions.
110
Table 3-20. Simulated patient population for all scenarios, including demand increase
scenarios
Scenario Patient
Population SD Resulting Yearly Rate (95% CI)
Comparison to
2017 FY
Current,
S1-16 20250.38 148.4453 73914 (74068.67, 73759.10)
significantly lower,
estimate difference
= 2086 patients/yr
S17 21414.78 148.3953 78164 (78318.68, 78009.22)
significantly higher,
estimate difference
= 2164 patients/yr
S18 22650.3 154.2658 82674 (82834.45, 82512.74)
significantly higher,
estimate difference
= 6674 patients/yr
The length of stay summary performance metrics for S16-S18 are presented in Table
3-21. The performance metrics typically followed the same pattern as the average overall LOS.
The box plots of S16, S17, and S18 are shown in Figure 3-26. The LOS metrics grow
exponentially with the increase in demand. Additionally the percentage of patients who stay in
the ED longer than 3 hours starts to reach 1 for the S18 conditions (Figure 3-27). The average
number in the RWR across runs increased gradually from 0.38 to 1.16 (Figure 3-28). The
maximum number in the RWR also gradually increased from 8.56 to 11.22, still far lower than
the space allocated (Figure 3-29). The average number across runs in the Admits zone was
reduced between the demand scenarios (Figure 3-30). This potentially is because of bottlenecks
earlier in the system stopping patients from reaching the Admits zone. As a reminder, demand
increased but the amount of doctors and nurses scheduled did not change in these scenarios,
which would be necessary with a population increase of this size.
111
Table 3-21. Summary of performance metrics across runs for demand increase scenarios
Performance Metric Scenario
S16 S17 S18
Average LOS (hr) Mean 3.937 5.367 26.089
(CIl,Ciu) (3.898, 3.975) (5.201, 5.533) (23.084, 29.095)
Discharged LOS (hr) Mean 3.034 4.520 26.806
(CIl,Ciu) (2.995, 3.072) (4.344, 4.697) (23.563, 30.049)
Admitted LOS (hr) Mean 6.486 7.753 24.077
(CIl,Ciu) (6.446, 6.526) (7.614, 7.891) (21.74, 26.413)
Acuity1 LOS (hr) Mean 4.555 5.699 14.393
(CIl,Ciu) (4.509, 4.6) (5.595, 5.802) (13.192, 15.593)
Acuity2 LOS (hr) Mean 5.211 6.266 14.868
(CIl,Ciu) (5.176, 5.246) (6.168, 6.363) (13.731, 16.005)
Acuity3 LOS (hr) Mean 3.808 5.358 29.215
(CIl,Ciu) (3.767, 3.848) (5.172, 5.544) (25.723, 32.706)
Acuity4 LOS (hr) Mean 2.919 4.455 31.078
(CIl,Ciu) (2.876, 2.961) (4.256, 4.654) (27.152, 35.003)
Acuity5 LOS (hr) Mean 2.649 4.232 31.010
(CIl,Ciu) (2.592, 2.705) (4.011, 4.454) (27.041, 34.978)
Waiting Time (min) Mean 5.134 33.128 861.866
(CIl,Ciu) (4.488, 7.397) (28.029, 50.975) (734.017, 1309.405)
Max in WR Mean 45.16 83.96 422.06
(CIl,Ciu) (42.163, 48.157) (77.449, 90.471) (371.248, 472.872)
Number in WR Mean 1.070 7.364 200.590
(CIl,Ciu) (.935, 1.205) (6.224, 8.505) (170.89, 230.29)
% LOS > 3 hr Mean 52.32% 72.09% 98.28%
(CIl,Ciu) (51.62%, 53.02%) (70.79%, 73.40%) (97.93%, 98.63%)
Number in Admits Mean 3.624 3.579 3.525
(CIl,Ciu) (3.61, 3.638) (3.566, 3.592) (3.512, 3.538)
Number in RWR Mean 0.377 0.691 1.161
(CIl,Ciu) (.362, .391) (.665, .718) (1.143, 1.178)
Max in RWR Mean 8.56 9.76 11.22
(CIl,Ciu) (8.213, 8.907) (9.486, 10.034) (10.902, 11.538)
112
Figure 3-26. Box plots for the average length of stay across runs for demand increase
scenarios
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and the quantiles (blue).
Figure 3-27. The percentage of length of stay greater than 3hrs across runs for demand
increase scenarios
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and the quantiles (blue).
113
Figure 3-28. Box plot for average number in RWR across for demand increase scenarios
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(blue).
Figure 3-29. Box plots for average maximum number in RWR across runs for demand
increase scenarios
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(blue).
114
Figure 3-30. Box plots for average number in Admits zone across runs for demand increase
scenarios
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the mean
(brown) and the quantiles (blue).
3.7 Discussion and Conclusions
The results showed that layout considerations have a variety of effects on patient
performance metrics. The largest spatial considerations for overall impact on length of stay were
the addition of the Results Waiting and the Admits zones. Allocation of space (and beds) for the
RWR and the Admits zone are highly coupled with the future process implementations expected,
thus careful consideration of the resources allocated in terms of both capital and operating
resources is important to understand the overall impact on LOS. The addition of the FT bays did
115
not have the effect expected. The additional bays and the modeled routing process did not make a
positive impact on LOS once the RWR and the Admits zones were implemented. There could be
many reasons for this result, including changing population of ESI 4 and 5s from the planning
years of the project (2015) to the data set used in the model (2017). Alternatively, the addition of
both the FT and RWR could have a diminishing combined effect, such that under these conditions
they work against each other. As a result the best in system was not the implementation of all
layout considerations in the model. The best were the use of the RWR, Admits zone, and the care
initiation, the S12 model. The space allocation implications of the analysis are that the additional
FT bays should be considered for different routing, potentially for MT (ESI 3 patients) that do not
need a full bed. The RWR was sized to hold 36 people which is 4.8 times the maximum number
routed there in the best simulation model (and 3.2 times the size in the 2019 demand scenario).
In this case, the next steps would be to analyze the use of different routing and workflow
techniques in addition to testing operational resourcing with demand changes to understand if
different operational practices should be used in this space configuration or if a change in the
space allocation is warranted. With assumptions and model conditions presented, the space design
could be reconfigured to optimally address the patient population if the simulation were coupled
with the process improvement and the design development activities.
The results indicate that the performance measures commonly used in healthcare DES
models are sensitive to both layout and operational changes, with a special emphasis on changes
to space allocation. However, the results indicate that additional processes might need to be
evaluated, such as lab turnaround time or in-patient bed availability, to determine if these changes
are robust when assumption on service times inside and outside of the ED change.
Based on this work, future research should investigate additional input parameters,
service times and distributions. Incorporation of time series analysis to arrival patterns and
disposition probabilities to model different weekly and yearly patterns, could be studied to check
116
if adding detail to the model increases accuracy. In terms of modeling, better tools for integration
of layout and paths within DES software or the use of simulation engines within common drafting
software would be a benefit to the modelling efforts. Linking paths and pathfinding algorithms
are a natural next step in modeling for understanding the layout and process connection.
Additionally, easy & fast translation and iteration with additional common factors of interest is
key for implementation success on future projects, as not every project can hire a dedicated team
for developing and running a complex set of analysis. Future work should investigate automated
workflows and standardization of key performance factors for healthcare decision support. This
study only investigated a small number of layout factors, and did not explore the addition of
operational resources such as doctors and nurses. Once a large number of factors are involved,
additional statistical tools will be necessary for filtering and analyzing the impacts on
performance measures.
The DES provides an avenue to gain insight into complex systems such as emergency
departments and can be implemented throughout the facility’s design process to aid healthcare
and design practitioners to understand the implications of space, program, and layout on
performance measures. The use of a data-driven approach as used here can give rise to data to
help judge the complexity of design decision that healthcare practitioners make, such as the
potential for additive and diminished effects of different layout decisions, however translation of
layout considerations into a DES model is currently a non-trivial task. Much of the research in
healthcare DES has focused on process aspect of changes during operations and early design
phases, not layout changes. Scenario testing of layout options can potentially expand the use
cases of DES and, once automated, could be a useful tool to aid critical decision making
processes.
117
Chapter 4.
Implementation and Evaluation of Generative Layout Options using the
Graph Theoretical Approach for a Hospital Layout Problem1
4.1 Introduction
Healthcare planning and design is a complex process requiring careful consideration of
the key departmental adjacencies throughout the facility to ensure patient safety, operational
efficiency, and reduced travel distances for patients, caregivers, and non-clinical staff. There are
often competing priorities influencing the location and adjacency relationships to other key areas,
particularly in the diagnostic and treatment (DT) departments. These priorities can be dictated by
code and guideline requirements, functional needs, and individual preferences (Carr et al. 2017).
Healthcare owners require increasing amounts of evidence to support design decisions and
healthcare designers are focusing on efficiency and optimal design (Burmahl et al. 2017).
Generative software can be used to develop initial layouts based on evidence-based priorities to
create a quantitative starting point for healthcare planning and design experts to review and refine
layouts with healthcare facility stakeholders. This will allow the development of optimized
configurations of the DT departments within a facility and can lead to a common understanding
of healthcare design implications on efficient and safe care practices.
Healthcare efficiency and flow research has been a rich area of research for both design
and operations. Patient centered design methodologies have been becoming more popular over
1 In preparation, from: Lather, J. I., Logan, T., Renner, K., and Messner, J. I. “Implementation and
Evaluation of Generative Layout Options using the Graph Theoretical Approach for a Hospital Layout
Problem.” Journal of Computing in Civil Engineering. Submitted April 2019.
118
the last 30 years. Patient centered design aims to use the experience of the patient to influence the
design process. In this design process, both patients and healthcare practitioners are involved in
the design and delivery of the healthcare facility to help ‘co-create’ an Experience Based Design
(Bate and Robert 2007). Additionally, other research groups have focused on Evidence Based
Design where the past environment is studied to make recommendations on future design
guidelines and practices (Ulrich et al. 2008). In operations research, researchers have focused on
understanding key performance metrics of interest, such as patient length of stay, for over 50
years (Günal and Pidd 2010; Jun et al. 1999). Simulating patient and practitioner workflow
processes is used in this domain to generate operational performance data for decision makers to
analyze operational practices. These disciplines focus on understanding the patient and healthcare
provider experience, albeit from different perspectives and areas of focus, with the goal of
providing data to inform the quality and performance of the care delivery processes.
Optimization methods have been discussed for healthcare facility layout planning
problems in the literature for more than 40 years. One of the earliest formulations of the problem
was from Elshafei (1977), where it was presented as a Quadratic Assignment Problem (QAP)
where cost and flow were optimized. Many researchers have addressed generic QAPs as well as
other formulations of facility layout planning problems over the years (Anjos and Vieira 2017;
Francis et al. 1992). The graph theoretical approach was developed as a heuristic method for
solving common QAPs (Foulds and Robinson 1978). In a healthcare setting, several researchers
have applied this method to healthcare specific problems, in a hospital setting (Arnolds and
Nickel 2015) and in a Surgery Department setting (Assem et al. 2012). Arnolds and Nickel
(2015) discussed the graph theoretical approach as more useful than other solving techniques for
communicating with healthcare experts and architects who may not be as familiar with typical
facility layout planning optimization methods. While many optimization methods are complex
119
and need experts for proper usage, the graph theoretical approach may be one well suited for use
by healthcare planners and designers.
Even with a growing demand for data-driven methods and years of research on healthcare
layout planning problems, there is a lack of research on the evaluation of these techniques from
experts such as healthcare strategists, planners, and designers. How people behave when they use
models is an important emerging area of research in operations research (O’Keefe 2016) since it
provides insight into the translation from research to the practical. Facility layout problems are
similar in that they provide a model of a future facility, which must be communicated to a set of
stakeholders with varied understanding of the modeled problem or the facility design process to
make critical decisions. This study investigates the use of optimization techniques to generate
layouts of a test case hospital and presents expert evaluation of the techniques for future
development efforts.
4.2 Research Methodology
The research methodology focused on developing, implementing, and evaluating a layout
optimization technique on a hospital layout problem. A recently designed new construction
hospital was selected as a test case. The project had gone through a strategizing, planning, and
designing process with considerable changes in program scope and area over the project timeline.
Layout of the DT and support service departments had changed frequently throughout the design.
A layout optimization technique was proposed to test optimizing and streamlining the layout of
these departments based on programming and adjacency data. Using a graph theoretical
optimization technique, an optimal adjacency subgraph of the hospital departments was obtained.
From the subgraph, a placement strategy was generated to locate departments conceptually on a
floor plan using constraints of structural bay size, department area, and strategy. Two score-based
120
evaluation criteria were used, a weighted adjacency score for the subgraph and a distance
weighted adjacency score for the placed layout. Adjacency ratings from project experts were used
as the input data for the generated layouts. These layouts, using a script between the layout
generating program and common building information model (BIM) authoring software, were
translated into massing objects for layout visualization. To evaluate the usefulness of the
implementation of this technique, healthcare strategists, planners, and designers were surveyed on
their perceptions of the output from this technique. Data includes perceptions of the best and
worst layouts in a generated set; opinions on the advantages and disadvantages of using this
method; and demographic data. The expert evaluation data was used to assess accuracy and
viability of using generative layout techniques in a healthcare hospital layout setting. The
following sections describe the layout generation methods, the layout evaluation metrics, and the
expert evaluation methods used in this methodology.
4.2.1 Layout Generation Methods
Any generative layout methodology needs to address how a layout is created as well as
the evaluation criteria used to drive the process and eventually used to help select the best among
a set of layouts. The layout generation methodology thus starts with goals and objective setting
(Figure 4-1, step 1). Next, both gathering input data (2a) and conceptual problem formulation
(2b) done concurrently. Thirdly, the graph theoretical approach is used to generate an adjacency
model for recommendations (3) which is based on the problem formulation and the input data.
Then adjacency information is used in the block layout development (4) to create block layout
plans. Finally, evaluation of those layouts is done to find their optimal layout (5). This
methodology follows typical optimization and simulation methodologies in the literature (Banks
et al. 2010; Hassan and Hogg 1991; Malmborg 1994).
121
Figure 4-1. Generative layout methodology
4.2.1.1 Development of Graph Theoretical Approach
The graph theoretical approach is a heuristic method of generating a maximally planar
and maximally weighted subgraph of nodes (in this case: departments). It is a commonly used
algorithm for layout planning problems (Francis et al. 1992). It provides an adjacency optimal
graph (Foulds and Robinson 1978). However, it is typical to apply space and area requirements
separate from meeting strict adjacency requirements suggested in the graph approach. There can
be many adjacency relationships in the graph theoretical approach which cannot be met once
layout constraints are applied to the problem.
To define the problem, imagine there is a set of departments with 𝑛 number of nodes,
sometimes called vertices (𝑉), and edges (𝐸′). An edge will connect two nodes and have the
weight of the relationship between the two connected nodes. A complete graph, 𝐺’, can be
122
defined as the set of its nodes and edges, 𝐺’ = (𝑉, 𝐸’), thus all relationships will be represented.
A planar subgraph is 𝐺 = (𝑉, 𝐸), and has all nodes (𝑉) and a subset of edges (𝐸) present such
that 𝐸 ⊂ 𝐸′. The graph 𝐺 would be maximally planar when adding another edge 𝑒 (where 𝑒 ∈
𝐸′) would cause 𝐺 to no longer be planar. To maintain the planar property, no edges can cross one
another. When the sum of the weights present in 𝐺 is maximum, the result is a maximally
weighted and maximally planar subgraph of 𝐺’.
One common method for solving for the subgraph G is the R-construction deltahedral
heuristic developed by Foulds and Robinson (1978) starts with a set of departments with a flow or
adjacency rating between each pairwise combination. An initial arrangement of four nodes is
created by selecting the top weighted nodes and placing them in a tetrahedron fashion, where
each line represents the edge weighted by the relationship between those nodes (Figure 4-2a).
Subsequent nodes are added to the graph step by step: first, by selecting the next node with the
greatest overall impact using the department’s total closeness score (TCS – the sum of all
adjacency ratings for department 𝑖), and, second, by evaluating the entering node’s impact on the
overall score of the graph based on its weight with each triangular segment. A final step-wise
optimal graph is obtained based on the edge weights once all nodes are placed in the graph
(Figure 4-2b). A dual of this graph provides the rough arrangement of departments without
specific area and shape characteristics (Figure 4-3a). This provides abstract information to
planners. The exercise of translating the graph to a block plan is the next step, which includes
applying planarity to the graph and developing a possible layout option which meets the
adjacency requirements (Figure 4-3b).
123
Figure 4-2. (a) Initial tetrahedron formulation of graph theoretical approach (b) Final
maximally planar subgraph
Figure 4-3. (a) Dual (red) of the adjacency graph (the exterior boundary node is not shown)
(b) A possible block layout formulation meeting all adjacency relationship requirements
The step-wise program was developed in C# using the R-construction deltahedral
heuristic from Foulds and Robinson (1978). A sketch of the code is presented in Table 4-1. When
deciding how to arrange the first sx4 nodes, one was randomly selected to enter as the center.
When deciding the next node to enter, if there was a tie with the TCS, a node was selected
randomly. When deciding which location to add the entering node, there could be a tie if it would
contribute to the overall graph score the same among many departments, thus if there was a tie,
the location was selected randomly.
(b) (a)
(a) (b)
124
Table 4-1. Code sketch for graph theoretical approach Step Action
1 Input Spreadsheet matrix data.
2 Create initial set of nodes. 3 Create a dictionary for edge value weights.
4 Initialize graph with 4 nodes, each with three neighbors.
Use the top 4 TCS nodes. One is randomly selected to be
the center. Current iteration score is the sum of all
present edges, where an edge is present if nodes are
neighbors. 5 Create triangles for each triangular segment of the
current graph.
6 Evaluate next node to enter graph.
7 Determine where to place it by evaluating the entering
node’s impact on each triangular segment by adding the
edge weight for each node in the segment. Pick the
location the entering node will create the greatest
impact. Update each node’s list of neighbors. Update
current iteration of graph score.
8 Repeat 6 and 7 until all nodes have entered the graph.
9 Output final graph, score, list of nodes, and each of
their neighbors.
4.2.1.2 Input Data
Most block layout methodologies use either a From-To chart or a relationship diagram
(REL) as input data. The From-To chart data provides measured or estimated flow information
and the relationship data provides a qualitative metric for use in determining the layout of a
facility. In this case there was no prior flow information to measure, thus a relationship diagram
(REL) was used for the input data for this methodology. This method provides a qualitative
measure of the importance of adjacency between departments which can take into account many
factors from an expert’s experience (Francis et al. 1992). The AEIOUX rating system was used,
where each department pair was given an adjacency rating of: (A) absolutely necessary, (E)
especially important, (I) important, (O) ordinary importance, (U) unimportant, and (X)
undesirable. The following numeric weights were used for calculation purposes: A-5, E-4, I-3, O-
2, U-1, X-0. This method is sensitive to the adjacency input values (Francis et al. 1992), so to
125
help alleviate this problem, six experts involved with the project were surveyed to generate
different optimal layout sets.
4.2.1.3 Development of Placement Strategy
After a near-optimal subgraph is obtained, a planner can use the graph and dual to plan an
optimal layout with area and shape information (Figure 4-3b), which typically has been a manual
exercise (Assem et al. 2012). Methods have been explored to create a block layout, such as the
spiral technique (Assem et al. 2012; Malmborg 1994), sometimes called the crystal technique,
and serpentine placement (Francis et al. 1992) as well as many others with or without gaps. Area
requirements are used which are typically simplified to unit blocks. One common algorithm used
is the Computerized Relative Allocation of Facilities Technique (CRAFT) algorithm, which is a
layout improvement strategy, meaning it requires an initial layout.
For implementation, the graph theoretical approach was selected given its easily
communicated methodology and a serpentine placement strategy was selected given its ease of
implementation. Common structural bay sizes for healthcare facilities in the US of 9.144m (30
ft.), 18.288m (60 ft.), and 27.432m (90 ft.) were used and total layout area was based on the
square root of the total area requirements. For deciding how to navigate the graph for placement,
a starting node was used to enter the graph, and each subsequent department was placed based on
the set of neighbors, picking the highest weighted relationship. If there was a relationship weight
tie, the department with the highest TCS was selected, and, if there was still a tie, then the
department with the greatest number of neighbors was selected. Each department was placed in
order based on its programmed area and prescribed bay size (Figure 4-4). Since the bay size and
starting node were unknown parameters, 6 options were run, with two different starting nodes (a
prescribed node and the top TCS node) and all three common bay sizes, providing a total of six
126
layouts. These were each given a distance weighted adjacency score. A code sketch is provided in
Table 4-2.
Figure 4-4. Serpentine placement pattern, placement path with specified bay size
Table 4-2. Code sketch for placement strategy Step Action
1 Create layout boundary canvas with x and y dimensions.
2 Create initial set of departments with areas.
3 Evaluate a starting department based on maximally planar
maximally weighted graph.
4 Place initial department at (0,0) with prescribed bay size
and length depending on area. Calculate coordinates and
centroid.
5 Evaluate next department to add to the layout by picking
greatest weighted neighbor. If tie, pick the neighbor that
has the greatest TCS. If still tie, pick the neighbor that
has the greatest number of neighbors. If still tie, pick
between those left randomly.
6 Place the entering department in serpentine fashion based
on prescribed bay size, area of department, x-, and y-
dimensions. Calculate coordinates and centroid.
7 Repeat 5 and 6 until all departments have been placed.
8 Evaluate the distance weighted adjacency score for the
canvas.
9 Output canvas, its score, and each departments with
coordinate and centroid data.
4.2.1.4 Development of Layout Output
At this stage a design authoring software was selected to generate boundary massing BIM
objects for each department. Each department had a set of coordinates outlining the shape. They
were either rectangular, L-shaped, or S-shaped. Depending on the order placed, the coordinates
127
for each vertex of the shape were in different orders. Figure 4-5 shows the different shape and
coordinate options from this placement strategy. Generic massing elements that corresponded
with the 3 shapes of departments were created as BIM objects with the necessary parameters to be
sized according to the spatial requirements for each department. A script was created to take a set
of layouts and identify shape, define origin points, perform transforms, and ultimately place the
department with parameter values such as name and area in a BIM authoring tool (Figure 4-6).
Figure 4-5. Shape grammars for serpentine shape translation
Figure 4-6. BIM objects generated in a parametric BIM authoring tool
128
4.2.2 Layout Evaluation Metrics
The most common numeric approaches for evaluating layouts are based on either
adjacency or distance (Francis et al. 1992). Some have used a combination of adjacency and
distance scoring metrics (e.g., Computational Relationship Layout Program – CORELAP from
Moore (1971)). For this study, a distance weighted adjacency score metric was used for the
layouts to take both adjacency ratings and distances into account.
4.2.2.1 Adjacency Score
Given 𝑛 departments, the pairwise adjacency rating between departments 𝑖-𝑗 (𝑤𝑖𝑗, ∀𝑛),
and the adjacency state (𝑎𝑖𝑗 ∈ [0,1], integer, evaluating to 1 when departments i-j are adjacent),
the adjacency score (𝑆𝐴) can be calculated as a maximization problem (Equation 1). The goal is to
maximize the higher weighted department pairs, which can be computationally expensive when
there are many competing departments (Malmborg 1994). This scoring metric is used in many
computational layout methods including the graph evaluation in the graph theoretical approach.
𝐦𝐚𝐱 𝑺𝑨 = ∑ ∑ 𝒘𝒊𝒋𝒂𝒊𝒋
𝒏
𝒋=𝟏
𝒏
𝒊=𝟏
(Eq. 1)
Equation 4-1. Adjacency score
4.2.2.2 Distance Score
To provide a score based on information beyond the qualitative adjacency ratings, some
computation methods use a distance based scoring metric (Francis et al. 1992). Given the distance
between departments i-j (𝑑𝑖𝑗) and the cost or flow, depending on the formulation, between
departments 𝑖-𝑗 (𝑉𝑖𝑗), the distance score (𝑆𝐷) can be calculated as a minimization problem
129
(Equation 2). Distances are evaluated by the rectilinear distance between the department centroids
or closest edges, but other distances can be calculated relevant to the application problem.
𝐦𝐢𝐧 𝑺𝑫 = ∑ ∑ 𝒅𝒊𝒋𝑽𝒊𝒋
𝒏
𝒋=𝟏
𝒏
𝒊=𝟏
(Eq. 2)
Equation 4-2. Distance score
4.2.2.3 Adjacency Weighted Distance Score
Given the pairwise adjacency rating between departments 𝑖-𝑗 (𝑤𝑖𝑗, ∀𝑛) and the distance
between departments i-j (𝑑𝑖𝑗), the adjacency weighted distance score (𝑆𝐴𝑊𝐷) can be calculated as
a minimization problem (Equation 3). Distance has been described as calculated in two ways: as
the measured distance between departments and the estimated distance. The measured distance
can be the distance between flow-dependent activities in the departments (e.g., the material
storage location in department A and the input location of department B), and is used when more
details of the layout of a department are known. Estimated distances are more common. In
CORELAP, the adjacency weighted distance score is calculated using the distance between the
shortest distance between department edges (Moore 1971). In this equation, if departments are
adjacent, their contribution to the score is zero, since distance would equal zero and this is a
minimization problem.
𝐦𝐢𝐧 𝑺𝑨𝑾𝑫 = ∑ ∑ 𝒘𝒊𝒋𝒅𝒊𝒋
𝒏
𝒋=𝟏
𝒏
𝒊=𝟏
(Eq. 3)
Equation 4-3. Adjacency weighted distance score
130
4.2.2.4 Distance Weighted Adjacency Score
Given the pairwise adjacency rating between departments 𝑖-𝑗 (𝑤𝑖𝑗, ∀𝑛) and the inverse
distance between departments 𝑖-𝑗 (𝐼𝑖𝑗 = 1/𝑑𝑖𝑗), the distance weighted adjacency score (𝑆𝐷𝑊𝐴) can
be calculated as a maximization problem (Equation 4).
𝐦𝐚𝐱 𝑺𝑫𝑾𝑨 = ∑ ∑ 𝑰𝒊𝒋𝒘𝒊𝒋
𝒏
𝒋=𝟏
𝒏
𝒊=𝟏
(Eq. 4)
Equation 4-4. Distance weighted adjacency score
In this scoring metric, rather than using a Boolean value of either adjacent or not
adjacent, the score is based on how far apart the centroid of a department is from each pairwise
combination. The impact of the adjacency score is diminished by distance at a rate of 1/𝑑𝑖𝑗,
where 𝑑𝑖𝑗 is the rectilinear distance between the centroids of departments i-j. Compared to the
adjacency weighted distance metric, the distance weighted adjacency score differs in two ways. It
is a maximization problem where the increase in distance is detrimental to the score, whereas
previously, there would be a mis-match between adjacency score and distance, where one would
be maximized and the other minimized. Secondly, centroid to centroid distance was used because
this distance represents an average distance between departments, which is more accurate where
complex sets of tasks are performed in each department instead of the closest rectilinear distance
used in manufacturing settings described by Francis et al. (1992). This metric combines both
distance and adjacency rating into a single metric and can easily be expanded from adjacency
ratings to department flow data. This deviates from typical graph scores because once the shapes
and areas are decided, the distance between centroids of departments can increase considerably
and the activities in adjacent departments may have large distances impacting flow. The 𝑆𝐷𝑊𝐴
takes into account the adjacency rating and the distance even for departments which are adjacent,
131
providing a more realistic metric for complex department dependencies. This score can be
evaluated after the graph is translated into a block layout.
4.2.3 Expert Evaluation Methods
A survey was developed to gain expert perception of the layout outputs and to understand
if experts’ evaluation of these layouts matched the scoring metric. Expert may evaluate layouts
with different evaluation interpretations than used in the objective metric approach where only
the factors in the objective function are optimized. This study was an initial implementation of
this approach and additional features could be added to make this methodology viable in
additional application settings, e.g., add factors, features, and limits, such as site constraints,
different placement strategies, and additional rules. In order to support the research rationale and
further development efforts, a survey question was asked to understand if users found initial
generative layout tools to be useful. The goals of the survey were to assess the layout generation
output, compare the results of the distance-weighted adjacency scored layouts with the expert
opinion of best layouts, and to gather additional general perceptions data on the viability of these
approaches.
4.2.3.1 Hypotheses
To understand expert opinions of generative layout options, a hypothesis was developed
to test if respondents selected layouts that scored well. To confirm the trend of respondents, tests
were performed on whether respondents would also select layouts that scored relatively poorly.
The hypotheses are that respondents will select the best/worst scoring layout more often than one
would expect with random selection. The following hypotheses and null hypotheses were
developed:
132
H1. The proportion of participants who select the best layout will be greater than the
expected random value.
H1 Null: The proportion of participants who select the best layout will be no different
than one would expect with random selection.
H2. The proportion of participants who select the worst layout will be greater than the
expected random value.
H2 Null: The proportion of participants who select the best layout will be no different
than one would expect with random selection.
Given random selection of six items and assuming equal weight to all six options, the
random expected value would be 16.7% of the proportion of participants would select the target
option of either best or worst layouts. The expected value, 𝐸(𝑋), is the number of trials, 𝑛, by the
proportion that selected it, 𝑝 (Equation 5), and the standard deviation, 𝑆𝐷(𝑋), of the expected
value can be found with Equation 6.
𝑬(𝑿) = 𝒏𝒑 = 𝟎. 𝟏𝟔𝟕 𝒏 (Equation
5) Equation 4-5. Expected value for selection proportion
𝑺𝑫(𝑿) = √𝒏𝒑(𝟏 − 𝒑) = 𝟎. 𝟑𝟕𝟑√𝒏 (Equation
6) Equation 4-6. Expected value standard deviation from a proportion
4.2.3.2 Research Questions
In addition to the hypotheses, research questions were developed to understand the users’
perception of the usefulness of the generated layouts. The survey asked participants two questions
about usefulness. First, if they generally thought the generative layouts were useful. Secondly,
since people typically are involved in manual layout planning activities and come up with their
133
own specific strategies for layout, a question asked if the users would want more information
about the strategy used in generating these layouts, or the participants ‘need for decision details’.
RQ1. Do participants find these layouts useful?
RQ2. Do participants have a need for decision details?
Along with understanding usefulness, two research questions were developed to
understand if there were any correlation between demographic data and how users perceived the
usefulness of the layout approach:
RQ3. Does age of participant, years of experience with healthcare projects, or gender
significantly impact a participants perception of usefulness?
RQ4. Does age of participant, years of experience with healthcare projects, or gender
significantly impact a participants need for decision details?
4.2.3.3 Qualitative Measures
In addition to the quantitative measures, open-ended questions were developed to gain
information on the advantages and disadvantages that participants saw for generative layouts in
healthcare planning and design.
4.2.3.4 Participants
A survey sample was selected to represent a diverse range of roles and experience levels
for those with healthcare strategy, programming, planning, and design experience. A single
healthcare design firm was surveyed, utilizing an internal database of individuals who work
primarily on healthcare projects around the world. For the purposes of this study, participants
were limited to individuals working in the US with more than one year of healthcare experience
and a focus on planning and design of healthcare facilities. Additionally, participants were
included from the company’s healthcare consulting group, focused on strategy, programming, and
134
operational planning, and the company’s computational design group, focused on computational
design methodologies for a variety of project types. In total, the survey was sent to 262 potential
participants. Potential participants were given information about the study and asked to provide
consent to participate in the study prior to participating in the survey.
4.2.3.5 Survey Apparatus and Scenario
Survey participants were asked to review a set of layout scenarios. To maintain continuity
and comparability, these layouts were generated given one set of input data. A set of 16
departments were used which included diagnostic and treatment departments and support services
departments. They were given a prompt to imagine they are designing an Ambulatory Surgery
Center with the task to arrange the departments so that they have the best flow between
departments. They were given plans, a scale, and color coded and labeled block plans of the
surgery center (Figure 4-7). They were told that each set of layouts were generated with a few
discrete constraints, a different starting department and different bay sizes. They were told to
ignore site constraints for these questions.
135
Figure 4-7. Block plans for six layout conditions
4.2.3.6 Survey Design
The survey contained three sections: layout scenarios for review, general questions on
generative layouts, and demographic data. The initial section asked participants for their opinions
on the best and worst layouts in the scenario. They were given a set of 6 layout options generated
with the graph theoretical approach and placement strategy. They were asked to “select the layout
that would function the best,” and given an option to select either one, two or three layout plans.
Next they were given the same set of layouts and asked to “select the layout that would not
function well or the worst among the set” and told they could select one to three options. They
were asked to explain their choice for both questions. In the second section, participants were
136
asked their general perceptions of using generative layout and optimization techniques. They
were asked if they had previous experience: “Have you used a system to auto-generate layout
options before?” with a yes, no and maybe answer, and for yes and maybe they were given an
opportunity to describe their answer. To understand usefulness of this system, two questions were
asked on a 7-point Likert scale between agreement and disagreement to the following statements:
(1) “In order to find these useful, I would need to know more about what decisions the system
used to generate these layouts;” and (2) “I would find it useful to use a system to generate options
of layouts.”
Demographic data was collected from respondents on age, gender, education level,
amount of time in the healthcare design field, types of experience, location, and amount of time at
their current company. Age was provided in the following groups: 24 and younger, 25-64 in 5
year increments, and 65 and older. Gender was given three choices: male, female, and decline to
answer. Education level was given 8 categories from less than high school to doctoral degree
options. The types of experience in healthcare projects was asked with categories: strategy,
operational planning, medical planning, architecture, research, and a text entry other category.
Respondent’s location was collected by giving respondents a list of office locations for the target
company.
4.2.3.7 Survey Analysis Methods
For H1 and H2, a one-sample proportion test for the null hypothesis, experts will not
select the best scoring layout more than with random chance was performed. Respondents can
give a best or worst categorization to multiple layouts since there might be only marginal
differences between some layouts, so both respondents first choice and their overall total choices
were analyzed in the same manner. The expected random choice is equal for all layout options,
n=6, where each choice has an equal chance of being selected. The random expected outcome is
137
1/6 or 16.7%, for all choices. These were tested at a 95% confidence level using the exact
method, one-sided, one population test.
For RQ1 and RQ2, the mean and standard deviation responses were reported with a 95%
confidence interval around the mean to describe how useful respondents found these techniques.
For RQ3 and RQ4, the Pearson’s correlation were calculated for each variable of interest,
with significance level. Correlations with alpha’s < 0.05 were considered significant, and those
between 0.10 and 0.05 were considered marginal.
4.3 Layout Scoring Results
The layout scoring results are broken up into the initial graph results, the layout
generation results, and the expert survey results.
4.3.1 Graph Results
To obtain input data for the methodology, six people were surveyed, some in a group
setting, resulting in three sets of input data and one set of combined scores. A graph was
generated for each of these four input sources, with a graph score (G) using 𝑆𝐴 (Equation 1). A
theoretical upper bound (U) of the graph adjacency score can be calculated based on the edge
values, or weights of relationships between department pairs. Since there are (3𝑛 − 6) edges in
the final graph, the total sum of the top (3𝑛 − 6) values of the sorted weights of all combinations
of department pairs gives the maximum theoretical value of the optimal graph, with 𝑛
departments. A relative score for each graph is given by 100𝐺/𝑈, which is a measure of how
close the graph is to its theoretical upper bound (Foulds and Robinson 1978). The upper bound is
typically unattainable as Leung (1992) has discussed, yet it is a useful way to provide a reference
point for comparing performance of graphs (Hassan and Hogg 1991). The graph adjacency score
138
and the relative score for each input dataset on the example case are provided in Table 4-3. The
graph associated with the first input dataset was the highest relative scoring optimal subgraph
(100G/U = 82.9). Input 1 was selected to be used for user input because of its highest relative
graph score.
Table 4-3. Numeric graph scores INPUT
(𝒊)
𝑵𝒊 G (𝑺𝑨𝒊) 𝟏𝟎𝟎𝑮
𝑼𝒊
1 16 174 82.9
2 16 119 69.2
3 17 178 79.5
4 16 437 79.2
4.3.2 Layout Generation Results
The graph with the highest relative score was used to generate six layout options
corresponding to the six different combinations of discrete constraints: start node and bay size.
For each of these, the 𝑆𝐷𝑊𝑆 was calculated (Equation 4). Table 4-4 contains the results. The
highest scoring layouts were option 3 and option 6, with scores of 5.009 and 4.821 respectively.
A sample layout is provided in Figure 4-8.
Table 4-4. Numeric layout scores Option (𝒊) 1 2 3 4 5 6
Start Node Lobby Lobby Lobby Top Dept. Top Dept. Top Dept.
Bay Size 9.14m (30’)
18.29m
(60’)
27.43m
(90’) 9.14m (30’)
18.29m
(60’)
27.43m
(90’)
𝑺𝑫𝑾𝑺𝒊 4.502 4.434 5.009 4.787 4.580 4.821
139
Figure 4-8. Sample layout, Option 4
4.4 Evaluation of Layout Results
The survey had a response rate of 11.8%, with 31 respondents completing the survey.
The average age was between 40 and 44, with one respondent younger than 24 and one older than
65. The mode age category was between 40-44 (7 respondents). Respondents averaged 16.2 years
of experience with a minimum of one year, a maximum of 35 years, and a standard deviation of
8.5. Of the respondents, 54.8% were female. All had 4-year degrees or higher, and 71.0% had
Master’s degrees. 94% had experience in architecture, 84% in medical planning, 45% in
operational planning, 35% in research, and 29% in strategy. All but two respondents had never
seen or used generative layout methods before. Respondents were predominately located in South
Central US (52%), 29% from the Mid-Atlantic region, 13% from the Midwest, and 11% from the
Mountain region.
140
4.4.1 Comparisons of Subjective and Objective Optimal Layout
Respondents were asked to select the best, or set of best, options with a total of three
possible choices. Of the total responses, 11 (35%) chose more than one ‘best’ option, only one
person (3%) chose three options, and 20 (65%) selected either the highest scoring or the second
highest scoring layout. Of the respondents, 29% selected the highest scoring layout and 39%
selected the second highest scoring layout as the best option. One selected both as the best
options. The most commonly chosen ‘best’ functioning layout was the second highest scoring
layout. Respondents were asked to select the worst, or set of worst, layout options with a total of
three possible choices. Of the total responses, 15 (48%) chose more than one, four (13%)
respondents chose three options, and 15 (48%) selected either the lowest and second lowest
scoring layout. When asked to select the worst layout, 23% chose the worst scoring layout and
42% chose the second worst scoring layout. Five selected both as the worst options. The most
commonly chosen ‘worst’ functioning layout was the second worst scoring layout, option 1, with
42% of respondents. Respondents tended to select layouts that scored higher and that scored
lower when asked to select the best flowing layouts and worst flowing layouts, respectively
(Figure 4-9). Respondents were not given the scores of the layouts.
4.4.2 Results of Best Scoring Layout Choice
There is not enough evidence to reject the null hypothesis that experts will select the best
scoring layout as often as one would expect with random chance (expected 16.7%, selected
29.0%, 95% lower bound = 16.1%, p=0.061). Respondents selected the ‘best’ layout marginally
more often than random chance (alpha = 10%). A significant difference from the expected
random chance was found with the second highest scoring layout (expected 16.7%, selected
38.7%, 95% lower bound = 24.1%, p=0.003). Since respondents could select more than one
141
layout, a second test was performed on respondents first choice: there was enough evidence to
reject the null hypothesis that experts will select the second best scoring layout on their first
choice as often as one would expect with random chance (expected 16.7%, selected 35.5%, 95%
lower bound = 21.3%, p=0.009). No other layout option was selected more often than expected
with random chance. This indicates that there is variation in respondents choices for best layout
flow. A summary of the one-sided one-population tests with expected random value 16.7% for
each ‘best’ layout choice, all and first choice, are in Table 4-5.
Figure 4-9. Frequency of respondent’s choice of ‘best’ and ‘worst’ layouts with total and
first choices, and the horizontal line for random choice (5.17, n=31)
4.434 4.502 4.580 4.787 4.821 5.009
0
2
4
6
8
10
12
14
Layout Score (SDWA)
Count
Best 1st Choice Best Choice
Worst 1st Choice Worst Choice
Random Choice
142
Table 4-5. Hypothesis test results for hypothesis 1
Score Layout
Option
Select 'Best' Overall Select 'Best' Frist Choice
Proportion
Selected
95% Lower
Bound p-value
Proportion
Selected
95% Lower
Bound p-value
4.434 2 16.1% 6.6% 0.607 12.9% 4.5% 0.784
4.502 1 12.9% 4.5% 0.784 6.5% 1.2% 0.975
4.58 5 19.4% 8.8% 0.416 12.9% 4.5% 0.784
4.787 4 22.6% 11.1% 0.25 12.9% 4.5% 0.784
4.821 6 38.7% 24.1% 0.003 35.5% 21.3% 0.009
5.009 3 29.0% 16.1% 0.061 19.4% 8.8% 0.416
4.4.3 Results of Worst Scoring Layout Choice
There was not enough evidence to reject the null hypothesis that the worst scoring layout
would be selected as often as expected with random chance (expected 16.7%, selected = 22.6%,
95% lower bound = 11.1%, p-value = 0.250). Respondents selected the ‘worst’ layout no more
often than expected with random chance. Respondents selected the second ‘worst’ layout
significantly more often than expected with random chance (expected 16.7%, selected 41.9%,
95% lower bound = 26.9%, p=0.001), however it was only marginally confirmed with
respondents’ first choice (expected 16.7%, selected 29.0%, 95% lower bound = 16.1%, p=0.061).
Additionally, the third best layout was marginally significantly chosen more than expected
(expected 16.7%, selected 29.0%, 95% lower bound = 16.1%, p=0.061). No other layout option
was selected more often than expected with random chance. A summary of the one-sided one-
population tests with expected random value 16.7% for each ‘worst’ layout choice, all and first
choice, are in Table 4-6.
143
Table 4-6. Hypothesis test results for hypothesis 2
Score Layout
Option
Select 'Worst' Overall Select 'Worst' First Choice
Proportion
Selected
95% Lower
Bound p-value
Proportion
Selected
95% Lower
Bound p-value
4.434 2 22.6% 11.1% 0.250 12.9% 4.5% 0.784
4.502 1 41.9% 26.9% 0.001 29.0% 16.1% 0.061
4.58 5 25.8% 13.5% 0.132 19.4% 8.8% 0.416
4.787 4 29.0% 16.1% 0.061 12.9% 4.5% 0.784
4.821 6 19.4% 8.8% 0.416 12.9% 4.5% 0.784
5.009 3 22.6% 11.1% 0.250 12.9% 4.5% 0.784
4.4.4 Results of Perceived Usefulness
When asked on a scale of 1-7 if respondents found generative layouts useful, with 1 being
extremely useless and 7 being extremely useful, on average they found generative layouts slightly
useful (average of 4.87, with a 95% confidence interval of 4.29, 5.45).
4.4.5 Results of Perceived Need for Decision Details
When asked on a scale of 1-7 if additional information about the decisions the system
was making was needed, with 1 being strongly disagree and 7 being strongly agree, respondents
on average agreed (average 6.26 with a 95% confidence interval of 5.77, 6.75).
4.4.6 Results of Demographic Variables
No significant correlations were found directly between the independent variables (age,
gender, years of experience) and the dependent variables (usefulness and need for decision
details). Years of experience and age were significantly correlated. A summary of the results can
be found in Table 4-7. The data did not provide evidence to support age, years of experience, and
gender to be significantly correlated to respondents’ perception of the usefulness of generative
144
layouts or the need for decision details. Additional data would need to be collected to increase the
sample size to test for interaction effects, such as age and gender on perception metrics.
Table 4-7. Pearson correlations and p-values for age, gender, years of experience,
usefulness, and need for decision details
Factors Need for
Decision Details Usefulness
Years of
Experience Gender
Usefulness 0.455
p = 0.01
Years of
Experience
0.153 -0.021
p = 0.412 p = 0.909
Gender -0.03 0.283 -0.164
p = 0.872 p = 0.123 p = 0.377
Age -0.014 -0.037 0.942 -0.134
p = 0.939 p = 0.842 p < 0.001 p = 0.471
4.4.7 General Perceptions of Generative Layouts
It was found that people were interested in using generative layout techniques, especially
for fast iteration through multiple options not traditionally possible. The main items that people
discussed in their reasons for choosing the layout as the best functioning in the set were the shape
and specific key relationships. Six respondents thought there were flaws with all layouts. The
department relationships which were the most commonly described as impacting the best
functioning layouts were the location of Surgery to the Lab, Perioperative and Recovery, and
Sterile Processing departments, and the connection of the Lobby to ED, Perioperative and
Recovery, and Pharmacy. Several people identified the limiting nature of the linear (30’ bay)
layouts and the use of a single floor. A few respondents used assumptions to qualify their
decisions, such as separate outpatient pharmacy, satellite support services, or separate entrances
for ED and main hospital. Additional issues which were brought up were reducing walking
distance within department, patient safety, and expansion capability. When choosing the worst
functioning layouts, the main items people discussed were the restricted access for public and
145
patients from certain departments or areas, the distance between certain departments, and the
shape of departments. There were many items that were in agreement between respondents,
however there were a variety of criteria were used for judgement on non-universally agreed
strategies, such as the separation of entrances for the main hospital and the ED.
When asking respondents generally about using generative layouts, some respondents
thought the addition of these tools would add to layout accuracy. Additionally, respondents
thought these methods would provide evaluation tools for past and future projects. They thought
these methods would be advantageous for getting “the creative planning started,” teaching young
planners, communicating with other groups, and “quickly generating multiple options for review
and evaluation.” Several discussed the use of generative tools to help filter different evaluative
criteria and overcome personal design bias, as one respondent explained: “every plan has an
inherent bias, whether known, intentional, or non, unintentional.”
4.5 Discussion
Evaluation of the graph theoretical approach shows that respondents aligned well with
higher scoring layouts, but respondents did not consistently identify the worst scoring layout.
Respondents provided common reasoning for selecting the best layout, but had different
decisions, only 29% selected the highest scoring layout. There was not a uniform agreement
across respondents. People 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. These differences could be due to differences in layout strategies,
different experience of experts, and different assumptions within the layouts. If the variation in
choice is due to assumptions, those can be understood as options within a generative layout
methodology. If variation is due to opinions of experts, research can help make strategies and
146
opinions explicit and methods can be used to create scoring metrics that align to specific strategic
approaches.. More research should be done to understand what drives layout and design decisions
to help develop computational methodologies. In this study, the participants were healthcare
planners and designers, who may be subject to a professional bias skewed towards design, as
opposed to healthcare practitioners and administrators, with a potentially different set of opinions
and selection criteria.
The survey apparatus, the set of generative layouts, presents a different approach to
typical layout methods. It is common practice for planners and designers to be the creator of
several layouts in an iterative design methodology. Generative layouts presents a different
approach for planners and designers, where the computational program becomes a key part of the
iterative layout planning and design methodology. This change in methods potentially can
challenge practitioners bias and selection criteria, however is a relatively complex task. More
research is needed to understand the variation in understanding these layouts. Many practitioners
expressed that there were flaws in all layouts presented, yet at least one respondent said the
highest scoring layout was almost perfect, and recommended one adjacency change. While
unanimous agreement might be impossible, it is foreseeable that practitioners are naturally critical
and use different strategies for justifying their critique. Since professionals are used to critiquing
each other in a collaborative environment, there is a lack of an expected human interaction in
using a generative layout approach.
Participants described future beneficial uses of generative layout methodologies such as
for teaching younger planners, for alleviating personal bias, and for providing evaluation criteria.
Some disadvantages from respondents were that planners and designers would become dependent
on computational methods, which could hinder a designer’s creativity. While respondents
identified many advantages, the results were mixed. Respondents found these methods slightly
useful, which indicates that more work is needed on these computational methods to support
147
broader practical usage. Respondents agreed that more details about the generative layout system
were needed to find these techniques useful, which may provide a method for helping to alleviate
disadvantages by increasing planner’s and designer’s understanding of these types of tools. These
advantages and disadvantages speak to a common barrier for new technology: education and
training. To support the implementation of new methodologies, people need to learn the new
process to change how they do business. The perception of change can be negative at first, before
the change provides additional benefits.
The results add to healthcare facility layout methodologies by providing expert feedback
on adjacency focused generative layouts. Further evaluation of adjacency ratings from a variety
of experts, including those from other architecture firms, consulting groups, and especially from
care providers, would provide a more robust understanding of the variability in adjacency ratings
and impact on layouts. Additionally, input from those different perspectives would be useful in
testing if an expert’s role or level of experience has a significant impact on their opinion of using
these types of methods.
Development of these generative layout methodologies is time consuming and expert
knowledge can help guide those development efforts. Previous research in facility layout has
focused on the optimization heuristics without testing alignment between the metrics,
implementation methodology, and the practitioners who would use the system. This study
provides expert feedback on a generative layout methodology to provide guidance to future
research and development of computational methodologies for generative layouts.
Future work should develop the method for multi-story optimization by considering both
horizontal and vertical adjacencies, to look at intradepartmental adjacencies of rooms and
activities, and focus on a user interface with parameter selections including the ability to adjust
programming data, size of departments, and the use of a variety of placement strategies.
148
4.6 Conclusions
Generative layouts have many opportunities for aiding healthcare planning and design
experts in optimizing the layout of healthcare facilities. These opportunities include developing a
rapid method that analyzes multiple options and takes into account more factors than typical
designers have time to consider and provides evaluation metrics from which teams can compare
past and future projects. The results show promise in generative layout techniques, however more
details about the process the system performs and the consideration of additional factors were
desired by respondents to be useful in their decision making process. These results indicate a need
for transparent approaches to generative layout methodologies. These future methods need to be
sensitive to designers’ input by incorporating more constraints and objectives in order to be useful
by practitioners.
149
Chapter 5.
Framework for a Hybrid Simulation Approach for an Integrated Decision
Support System in Healthcare Facilities23
5.1 Introduction
Healthcare professionals are increasingly asking for evidence of the functional
requirements of facility designs (Burmahl et al. 2017). Design has typically been based on expert
knowledge from both design and healthcare practitioners. In order to enable data-driven
approaches, this expertise needs to be formalized in the optimization-simulation process. Typical
methods used by operations researchers are not frequently used in the design phase of healthcare
facilities. A recent review of the use of simulation in healthcare facilities found research in design
and capacity planning was limited (Arisha and Rashwan 2016). Similarly, scenario planning was
identified as promising but a limited area of research for healthcare (Rais and Viana 2011). Part
of the reason for the lack of simulation in planning and design is a lack of understanding from
both architects and healthcare practitioners (Arnolds and Nickel 2015; Holst 2015).
Facility layout problems are an important class of problems for operations research. In
the classic example of the optimization healthcare layout problem, Elshafei (1977) explains the
2 © 2018 IEEE. Reprinted, with permission, from: Lather, J. I., and Messner, J. I. “Framework for a Hybrid
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
Design Knowledge, FOD = Facility Operations Documents, FPE = Facility Planning Experience, FPK = Facility Planning Knowledge, PCD =
Post-Construction Documents, PDD = Post-Design Documents, and PEP - Project Execution Planning.
155
Patients go through a series of processes when they visit a healthcare facility. In the
context of an emergency department, the main processes include registration, triage, evaluation,
diagnosis, treatment, and discharge (Figure 5-3). A patient may need tests, medications, or
procedures. A patient may be admitted to the main hospital or transferred to a different facility.
These processes may happen in different locations of the emergency department. A patient who
arrives by ambulance in critical care will immediately be brought to the room or bed needed for
care. A patient who isn't critical typically only has the processes after triage conducted in a
room/bed. A patient who doesn't need much care, such as an emergency severity index (ESI) 4 or
5 patient, might not take up space in a bed for longer than evaluation. Bed resource allocation is
dependent on the processes that the healthcare unit deploys, but information on future processes
is typically not known during the capacity planning or layout of a facility. The experience of the
professionals involved in the functioning of the facility is used in the design process to evaluate
the layout and provide guidance on optimal configuration. But if the processes are unknown, the
decisions are not made with adequate information.
Figure 5-3. Typical emergency room patient processes. Containers indicate parts in the
process where a patient is roomed. These change based on the condition of the patient and
healthcare processes.
156
5.3 Related Simulation Work
In this section, an overview of simulation and optimization approaches in the literature
and how they are used for healthcare facility planning are discussed. A considerable amount of
research has been done in this area, and below only a few of the key areas related to this approach
are discussed.
5.3.1 Optimization and Facility Layout Design
Facility layout optimization problems are often NP-hard mathematical problems (Anjos
and Vieira 2017). Many researchers have investigated methods to solve optimization problems
for the layout of healthcare facilities. These layout solutions are usually a combination of
minimizing distance and flow cost between operational units (weighted average by daily trips).
Anjos and Vieira (2017) discuss the current state of mathematical layout optimization in
three classes of problems: row layout, multi-level, and unequal areas. Unequal areas is a
commonly investigated class of facility layout problems, followed by row layout, which has seen
recent advances, and lastly, multi-level problems are the least common and most difficult
computationally. For more work on the row layout (corridor allocation problem) see Amaral
(2012). While the mathematical optimization techniques are useful for understanding how to
obtain globally optimal solutions to layout problems, they do not take into account the facility
operations. In application, a simulation-optimization approach might be more appropriate to find
an optimal solution with uncertain processes. In a review of layout planning problems, Arnolds
and Nickel (2015) presented literature from 55 articles on various methodologies including
quadratic assignment problems, mixed integer programs, and discrete event simulations. They
suggested a framework for integrating deterministic optimization techniques with stochastic
simulation.
157
5.3.2 Healthcare Layout and Design Studies
Not only minimal distance or flow cost is important for healthcare layout optimization.
Views to rooms, room assignments, and environmental factors can be important factors in the
design of a facility layout. There has been a considerable amount of post-occupancy research
investigating healthcare layout and design and how to it impacts important metrics associated
with a department, namely nurse movement and efficiency of delivered care. In a study aimed at
understanding the relationship between different spatial measures and nurse movement,
Choudhary et al. (2010) found different organizational strategies for pods of beds being served by
nurses led to significant changes in the number of trips to patients beds. While many factors
contribute to the number of visits to patients beds including policy and layout, the study did also
find that patients who were seen more frequently also had more time with the nurse. The findings
show that spatial arrangement impacts time spent with patients when processes and procedures
are similar.
5.3.3 Discrete Event Simulation in Healthcare
Discrete Event Simulation is a common technique in operations research to simulate
healthcare processes in departments to understand how operational changes will impact specific
performance measures. Implementation is one of the more difficult areas of DES research
because decision-makers usually need solutions to problems urgently and thus timing, adequate
data collection, and oversimplification can lead to a lack of confidence in the technique (Günal
and Pidd 2010).
There have been several studies investigating both process changes and layout changes in
clinical healthcare settings. Farahmand et al. (2011) investigated changes to the staff allocation
assignments in a healthcare clinic in concert with minimizing distances. While layout wasn't the
158
main aspect of their study, they modeled patient movement as a stochastic time variable in their
discrete event simulation. This method allows some patients to walk faster than others which
might be more realistic when distances are known.
5.3.4 Virtual Reality in Discrete Event Simulation
Researchers have described the integration of simulation and visualization as helpful for
early design decisions (Waller and Ladbrook 2002), increased model understanding (Akpan and
Brooks 2014), improved model error identification (Akpan and Brooks 2014), demonstrating the
model to the client (Akpan and Brooks 2012), and improved decision making (Chau 1995). The
use of virtual reality in the validation of discrete event simulations has been the most common
area cited for the integration of these techniques (Rekapalli and Martinez 2011). Some people
have suggested animation and interactivity as a means to aid model acceptance, increase
stakeholder engagement, and improve the usability of discrete event simulations in manufacturing
contexts (Chwif et al. 2015). Research has discussed the integration in terms of smart factories
and in continuous improvement strategies (Turner et al. 2016). Kuljis et al. (2001) 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) incorporate latent processes not identified prior.
5.3.5 Virtual Reality for Facility Review
Virtual reality environments are capable of displaying realistic, immersive, and
interactive virtual facilities which support both individual and group understanding (van der Land
et al. 2013). Virtual reality has been used as a tool for integrated project team members to
communicate effectively (Liu et al. 2014). Kumar (2013) developed an experienced based design
159
virtual prototyping framework. In this study, a modeled virtual environment of a future facility
was developed to serve as the backdrop to specific important workflow scenarios (e.g., nurse
finds crash cart). Both structured and unstructured tasks were deployed in this system. Structured
tasks were found to provide more in-depth design feedback from healthcare professionals, yet
both were found helpful in engaging staff feedback into the design process. Physical mockups are
more common than virtual mockups. In a survey of healthcare industry members on recent
projects, 60% of projects were reported to have used some evidence-based design practices and
55% of projects engaged facility staff through physical mockup reviews. However, only 33% of
projects used 3D models and 17% used simulation software (Burmahl et al. 2017).
5.3.6 Summary of Related Work
There has been extensive research into various domains of simulation for process
improvement in the healthcare domain. The investigation revealed three operations research
techniques in healthcare: layout optimization, process simulation, and virtual visualization
(Figure 5-4). For each of these areas, there are at least two types of categories. For optimization,
there are exact mathematical methods and heuristic methods. For simulation, there are
optimization via simulation methods, discrete event simulations, and simulation-visualization
methods. For visualization, there are specific scenario visualization or unstructured, explorative
methods.
Figure 5-4. Diagram of hybrid simulation hierarchy proposed for healthcare.
160
5.4 Development Methodology
Several researchers have proposed a simulation-optimization approach in healthcare
settings that take into account distances as well as stochastic input variables, such as processing
times and arrival times (Arnolds and Nickel 2015) and walking times (Vahdatzad and Griffin
2016). Another hybrid simulation technique proposed the integration of simulation among
different departments and scales of a healthcare facility, including interdepartmental connection,
health demographics connection to DES inputs, and human actors (Djanatliev and Meier 2016).
Instead of focusing on scales of the healthcare system, the focus of this study is on the
implementation of a hybrid simulation approach in different facility phases which then could be
connected to various scales during operations.
Acar et al. (2009) presented a framework for integrating uncertainty into a generalized
mixed integer programming (MIP) optimization through simulation scenario testing. This
methodology was later revised by (Arnolds and Nickel 2015) where it was proposed with QAP.
See Figure 5-5 for the steps in the hybrid modeling technique using QAP or MIP optimization
and discrete event simulation. The steps in the modeling framework are: (1) run MIP/QAP
optimization to minimize cost or distance; (2) run discrete event simulation with updated
workflow scenarios and calculate the difference between the simulated objective value and the
deterministic optimization value; and (3) update the MIP/QAP with the function uncertainty
difference.
161
Figure 5-5. Integration technique proposed by Acar et al. (2009) and revised by Arnolds
and Nickel (2015).
5.4.1 Healthcare Design Review Process
Engaging healthcare practitioners in the design review process can be difficult. From
scheduling to having the correct information available, to guiding the discussion toward the
design decisions that are important to make at the time of the meeting, there are many human
factors that come into play. When discussing schematic design options, healthcare practitioners
struggle with understanding 2D plans and question how operations will occur in a new facility. In
fact, 2D representations of spatial problems have been shown to provide insufficient information
for design evaluation (van der Land et al. 2013). Integrating simulations of the healthcare
processes for operations and the spatial configuration can help healthcare practitioners have the
information they need, or at least understand the impact of uncertain processes, at the time of
their decision.
5.4.2 Conceptualization
Conceptualization of this hybrid simulation approach started with reviewing the relevant
literature, next creating an initial framework, and then observing the planning process in a recent
healthcare project. Limitations exist in current hybrid simulation approaches for healthcare, such
as time to evaluate a solution, healthcare provider trust in solutions, and challenges for input data
162
collection. By not taking into account some of the main limitations of adoption, simulation-
optimization approaches may stay inaccessible to key decision makers and the staff adapting to
managerial changes. Integration with virtual visualization is proposed to aid healthcare provider
understanding of the system and aid the simulation-optimization connection to the design review
process Figure 5-6.
Typically healthcare professionals have not made final decisions about their new
processes in early design phases and may be open to major process changes throughout
construction. Thus, these are unknown and take a long time for a department to formalize and
create consensus. The concept behind the hybrid approach is based on how the inputs and outputs
from optimization of layout, simulation of processes, and visualization of the simulated and
optimized environment (i.e., the design) can be used for a meta-model for design and
implementation. The conceptually integrated approach is shown in Figure 5-6 where each DES
iteration goes through a scenario development in a visualization platform. First, the initial
configuration of a healthcare department is optimized given initial flow cost weights, next a
scenario testing is conducted of the proposed processes in that version of optimal layout
arrangement. Next, the system is virtually visualized for validation and review of layout and
process plans. Next, the new simulation is updated with increased knowledge of facility
processes. The new flow cost weights are calculated and input into the optimization and then
reevaluated for the simulated and validated scenario change. Once a stable healthcare layout and
process combination are achieved, the process comes to a stopping condition and the optimal
layout is found for various healthcare processes.
163
Figure 5-6. Conceptual diagram for integration of optimization, simulation, and
visualization for healthcare planning.
5.4.3 Hybrid Simulation Objectives
For healthcare facilities to manage the design and operation in an integrated approach, we
propose various objective scenarios for practical use of a hybrid optimization-simulation-
visualization framework throughout the various stages of facility development: manage, planning,
designing, construction, and operations. We expect this list to develop and grow as the
technology used is tested and becomes easier to deploy. These simulation objectives include:
• System understanding and investigation. During management of a facility, develop an
understanding of the system for stakeholders (owner, manager, or staff) including
bottlenecks and dynamic impacts.
• Scenario planning and testing. During planning, test different patient and healthcare
workflow scenarios.
• Layout option analysis. During planning and design, test different layout options in
conjunction with patient and healthcare workflows.
• System checking. During operations, check implementation success, tune processes, and
check healthcare practitioner implementation.
164
• Sensitivity analysis. During design and construction, test if newly proposed changes of
layout or procedures will impact important healthcare outcomes. During manage and
planning, test different layout options and workflows to see how robust a design is for
future changes.
5.5 Facility Lifecycle Implementation
In this section, an illustrative case study of the use of an integrated OSV hybrid approach
is presented in various typical stages of the operations and design of a hospital department. While
researchers may want to investigate different scales of a health system, such as the macro-level
(i.e., dynamic health demographics) or the micro-level (individual actor level), focus here is on a
meso-level, e.g., the workflow in a single healthcare department. In the example, the application
of the OSV approach is discussed as it applies to specific phases of a project. Since managing a
facility occurs at all phases of a project, management specific implementation would be a
combination of a selection of these. Design is divided into three common phases:
conceptualization, schematic, and development.
5.5.1 Implementation during Operations
During operational phases of a facility, understanding the current status of performance
measures is important. If simulations were performed during planning, a comparison of the
simulation model can be performed with real-life conditions to check if the simulation model
performed as expected and check if the current operations are following the planned workflows.
Tracking current operations can provide more accurate data for finding implementation problems,
proposing new workflow solutions, and as a method to collect more accurate input data for future
simulations. Corrections to both the simulation and the actual operations can occur. Virtual
165
visualization can be used to help train and communicate motivation of reasoning for operational
changes. The model can be used to explore scenario tests of minor functional changes, e.g.,
adding an additional medication dispensary, moving a registration desk location. More extensive
scenario tests can be done in the hybrid simulation environment for nurse and doctor scheduling,
logistics, or treatment planning.
5.5.2 Implementation during Planning
During planning phases of a facility, it is important to understand the future needs of a
facility. In this phase, forecasting healthcare demands can help determine the stationary resources
needed (bed capacity, location, layout) and thus the capital investments needed to meet those
needs. The hybrid simulation approach can be used to test how sensitive a layout configuration is
to different healthcare forecasts in order to determine space and layout requirements.
Additionally, departments can test different patient and healthcare workflow scenarios to test that
objectives of performance measures are being met in the plan before moving forward with a new
facility, a redesign, or an expansion.
5.5.3 Implementation during Design Conceptualization
In the first stage of design, design conceptualization is performed to create configurations
and massing models of a program defined from the planning stage. A hybrid simulation approach
can be used in this phase of a project to perform room assignments problems, un-equal area
optimization problems, and configuration testing in the context of proposed future processes.
During this phase, configuration optimization can be performed with roughly determined area
requirements and configuration constraints can be formulated. Initial processes should be
formulated and tested.
166
5.5.4 Implementation during Schematic Design
Schematic design is the phase of a design where different design options are configured
and compared. Using a simulation-driven approach, the design team can perform “what-if''
scenario testing of a range of design configurations with constraints formulated in design
conceptualization. Sensitivity analysis can be performed when faced with uncertain future
processes to help understand how different healthcare procedures and policies would, with the
layout decisions, impact performance measures of interest.
5.5.5 Implementation during Design Development
After layout configuration and scenario testing is performed and an optimal layout
solution for the various future processes under consideration is selected, detail design
development can occur. During this phase, details of the architectural and engineering design are
developed. Major changes to layout should be avoided. In this phase, confirmation of room
specific requirements and configurations can be developed. Scenario tests can be refined with
additional information on workflow processes to perform design checks. Equipment specification
and structural requirements may impact layout configurations and the hybrid simulation approach
can be used to perform design checks of changes.
5.5.6 Implementation during Construction
Once a facility is under construction, it can be very costly to make design changes.
However, changes can and will occur on the jobsite. The role of this hybrid simulation in the
construction phase of a project can be to check that changes to the approved design do not impact
projected facility processes. In addition, minor changed can be made as healthcare practitioners
finalize their future proposed workflows. Construction has its own set of complex processes,
167
schedules, and layout planning, which potentially can leverage a similar hybrid simulation
approach to the construction layout, processes, and visualization. Additionally, in future
developments of a hybrid simulation approach, it may be possible to dynamically, or close to
dynamically, make design changes with less impact on the construction schedule, cost, and
quality, through automatic updates of design drawings, conflict recognition, and system checking.
5.6 Conclusions
There is a large amount of simulation research in healthcare processes and an increasing
desire from healthcare managers to increase the efficiency of operations. However, these
simulation techniques usually keep the stationary facility elements as static resources. With an
aging healthcare infrastructure, redesign is more common, which makes these stationary elements
less static than expected. Given that, a surprising amount of literature for healthcare layout
optimization problems and department-level discrete event simulation is published, but not
enough discussion on how these techniques can be implemented effectively in the design of
healthcare facilities. One avenue which is promising is through incorporating virtual visualization
in the review process. Further research is needed to develop test-cases and validate this approach.
If implemented effectively, it may provide an approach to help disparate groups communicate and
come to common decisions on effective and efficient healthcare facility planning and use. A
hybrid simulation approach for implementation in the healthcare lifecycle is presented which
aims to integrate the facility elements and the healthcare processes. This approach combines
layout optimization, process simulation, and virtual visualization in an iterative optimization-
simulation-visualization (OSV) framework to be deployed in various states of a healthcare
facility: manage, plan, design, construct, and operate. The goals of this approach are two-fold: (1)
integrate static facility elements into process simulation as changeable features to aid redesign
168
and planning efforts, and (2) integrate dynamic visualization into the optimization and simulation
processes to ensure adequate review of assumptions in models and increase stakeholder buy-in.
Five different objectives for the framework usage were identified and presented: system
understanding and investigation, scenario planning and testing, layout option analysis, system
checking, and sensitivity analysis. The OSV framework provides an initial approach for use in
domains where there are a large number of stakeholders (such as healthcare setting) and
implementation depends on buy-in from disparate groups of individuals. The framework can be
used by researchers to extend simulation-optimization in connecting performance measures in the
design and construction phases of projects.
169
Chapter 6.
Conclusions
The healthcare industry in the US has increasing demands for professionals to be more
efficient and to provide quality, safe, and timely care to patients. The research presented in this
dissertation explored and developed data-driven methods and an integrated systems approach to
healthcare operations and design. The work presented covers the use of dynamic and
deterministic models for healthcare planning, design, and operations by investigating the process
and layout as a pair which influence one another. The goal of this research was to investigate and
develop methods to combine facility layout and workflow processes to provide a data-driven
methodology for healthcare facility design. The detailed objectives were outlined in Chapter 1.
The literature was investigated in the topic areas of healthcare facility design, discrete event
simulation, healthcare layout optimization, and visualization strategies (Chapter 2). First the
layout implications in a discrete event simulation of a department of a healthcare facility was
studied (Chapter 3). Then a layout optimization strategy was developed and the use of the
generated layouts were evaluated by healthcare planners and designers to understand user
perceptions (Chapter 4). Finally a framework for integrating these techniques as a hybrid
modeling approach throughout the lifecycle of a healthcare facility was presented (Chapter 5).
The main implications of this body of work are that layout and processes are paired, that they are
in need of greater investigation, and an integrated hybrid modeling approach was presented for
healthcare professionals and researchers to guide the development of an automated decision
support system for healthcare facility operations, planning, and design.
170
The following sections describe the main results of the three objectives, the core
contributions of the work, the limitations, future areas of research, and final concluding
discussion of the work.
6.1 Summary
The results for the study of layout in discrete event simulation show that not all layout
consideration are additive. The addition of two of five layout conditions provided the most
amount of improvement over the baseline condition: Results Waiting (estimated 15.1%
improvement, with a Bonferroni adjusted CI between 8.4% and 21.8%) and Admits zone (15.7%,
between 8.5% and 22.9%), a combined improvement of 1.19 hours reduction (23.9%, between
15.6% and 32.1%) in length of stay for all patients. The addition of fast track bays reduced the
overall improvement by approximately 10 minutes (8.51 min, 11.5 min). The best scenario
included Care Initiation, Results Waiting, and Admits zone. Study of space allocation and space
utilization found additional fast track bays were not helpful and the Results Waiting was
underutilized (average of 0.723 people, max of 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, which are competing, and help determine the space requirements and
balance space allocations through the analysis of projected operations in that facility, yet this
level of detail is hard to get to without early development and healthcare practitioner evolvement.
A new method for generating layouts was developed based on the graph theoretical
approach, with translation into common parametric BIM tools. The results from the study of
layout optimization and healthcare planners and designers is that scoring metrics align relatively
well with expert opinions, but that more advances are needed to make generative layout methods
more readily accepted by professionals. On average, respondents selected the ‘best’ layout
171
marginally more than random chance (expected = 16.7%, proportion = 29.0%, lower bound =
16.1%, p-value = 0.061). Although, they 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.
A framework was developed and presented which integrates simulation, optimization,
and visualization for healthcare facility layout planning activities for optimizing both process and
layout (Figure 6-1). Objectives are presented to outline the uses of the hybrid modeling approach
as a systems approach throughout the management, planning, design, construction, and operations
of healthcare facilities. The use of this framework with discrete event simulation and layout
optimization problems is presented using an iterative process which engages the decision makers
and the model and analysis content. The main implications of this body of work are that layout
and processes are paired and that they warrant investigation and methodological development by
researchers. 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 layout setting
and in a healthcare context, have implications for other domains where uncertain and latent
processes are components of the layout decision making process.
172
Figure 6-1. Final conceptual diagram for the optimization-simulation-visualization
framework.
When considering optimal layout, certain criteria are expected to be known during early
stages of facility planning, such as traffic between stations, area, and stories. However those are
typically unknown at that early stage of the design pipeline. Additionally, planning new processes
through a department or hospital are also typically worked on for many stages throughout the
design process. It isn't enough to implement optimization in conceptual planning, and scenario
building in schematic design, and visualization in design development. With new models for how
facilities are delivered, a more integrated approach is needed to have on-demand analysis and
planning tools for healthcare delivery and healthcare practitioners.
6.1.1 Integration of DES and Layout Optimization
To integrate discrete event simulation and layout optimization, the inputs and outputs of
each need to be considered. For facility layout problems (FLP) optimization, the data output is a
block configuration; the input data is area and flow data (adjacency ratings or flow data).
Typically early programmatic data is used for these types of problems, thus initial estimates for
flow (e.g., early adjacency ratings) and area estimates. For discrete event simulation, the data for
173
input and output depends on the objectives. Output data is typically performance measures of
interest, additional data can be used such as flow data estimates for layout optimization. The data
input for the DES is layout data, workflow information, and processing times.
To go from FLP to DES, the layout of the workflow processes needs to be identified and
mapped (Figure 6-2). First a set of near best layout options are generated, they are then in parallel
analyzed and input into the DES for further analysis. Coordinate data is needed to automate the
activity location data, however, additional manual processing is common. For the connection
from the DES to the FLP, the flow and space requirement data needs to be gathered and updated
(Figure 6-3). Flow data is not automatically generated from DES. The different activities, zones,
and/or departments (generically called nodes) initially need to be identified. Then trackers need to
measure the number of entities or resources that pass between each pair of nodes. For a set of
nodes, n, where n = 16, there are n(n – 1), or 240, directional pairs, and (n(n – 1))/2, or 120,
bidirectional pairs. This flow information can then be used in the FLP optimization to update the
adjacency ratings with simulated flow data. Additionally, layout considerations need to be
formulated in the DES methodology, in conjunction with the typical DES creation methodology.
6.1.2 Visualization of Near Best Options
An important aspect of DES and FLP is the translation into usable content by decision
makers, planners, and designers. These models are not commonly implemented in practice, yet
information and communication technology in the Architectural, Engineering, and Construction
(AEC) Industry are creating processes for leveraging data throughout the building lifecycle
through building information modeling (BIM). The uses of FLP and DES models are for analyses
and visualization tasks. Visualization of the analytical data is key to engage stakeholders for
validation and eventual implementation of the data generated through these model methodologies.
174
Figure 6-2. Process diagram for facility layout problem to discrete event simulation
Figure 6-3. Process diagram for discrete event simulation to facility layout problem
175
Figure 6-4. Taxonomy of aspects of a hybrid simulation approach in healthcare
In the taxonomy (Figure 6-4), virtual visualization of the healthcare facility can be
categorized into two task types: structured and unstructured tasks. Utilization of virtual
visualization aims to gather illicit and tacit knowledge through the exploration of salient and
latent processes. The 3D model data generated in common BIM modeling practices can be
leveraged to aid communication and visualization of spatial information in conjunction with
analytical information. During scenario analysis, when several near best layout options are being
considered, visualization of the spatial relationships can help decision makers understand the
workflow processes and iterate through different procedural strategies, thus allowing for an
iterative model approach which incorporates risk and uncertainty. Future work should investigate
how, technically and feasibly, to implement virtual visualization of healthcare simulation and
optimization content.
6.1.3 Implications for Industry
There is a growing desire to have evidence to support layout decisions. While common
layout methodologies are not data-driven or quantitative methods, the methods developed in this
work provide a framework for data-driven methods to support decision makers for healthcare
facilities. Evidence can be found to support decision makers from studying past practices,
studying current practices, and (as in this research) developing methods to study future practices.
176
The use of computational methods by industry is in its infancy. Industry needs support from
researchers to develop these tools and methods so that they are confident in their results. Careful
consideration should be taken to practitioner use requirements. However, these methods do
present a change in workflow and practice. Change management is difficult in any industry. In
the AEC Industry, changes in Information Communication Technology and Building Information
Modeling have been changing the way business is done. These quantitative methods, such as the
use of discrete event simulation and layout optimization, change the type of skills needed for
planners and designers, and careful consideration of the limits of inferences are needed for value
to be found in the industry. Training on methods and potentially adding technical experts in
planning and design fields are both needed for successful use of data-driven methods in
healthcare planning and layout design.
6.2 Contributions to Research
Both healthcare layout and process optimization are difficult problems. They require the
use of data to help understand the system and to develop new solutions to problems, but they are
complicated by human factors from both the process and the decision making. An integrated
approach is needed to leverage the output from these tools in effective and timely ways to provide
meaningful information to decision makers in the design and operations of facilities. Layout
optimization typically uses static flow data or adjacency ratings, but doesn’t take into account the
process variation in a dynamic system. Discrete event simulation uses past data to understand
performance, but typically keeps a layout static. The contributions of this research were to (1)
investigate how facility layout impacts operational performance measures; (2) develop and
evaluate a layout optimization problem; and (3) develop a framework for using layout
optimization (deterministic model) with discrete event simulation (stochastic model) and
177
visualization model (validation and decision making support). These contributions are in a cross
section of domains including operational research, architectural engineering, system engineering,
decision making, and data-driven design.
The main contributions of the first study was to model layout considerations to gain
insight into the space allocation program and to understand the current and new healthcare
workflow processes in the context of the layout considerations. The results showed that some
layout considerations competed with each other, indicating that better consideration of layout
changes in a DES methodology could help design teams create more adequate space planning.
Developing easy methods for set up and testing additional factors is an important area of research
for DES, and these should include layout considerations, such as location and path changes,
capacity changes and space allocation.
The main contributions of the second study were to develop a method to develop a set of
step-wise optimal block layouts of departments and import them into a parametric BIM modeling
tool. Additionally, the study provides data on the evaluation of the block layout by healthcare
planning and design professionals, the first study of its kind known to the author, thereby adding
data to researchers about how layout scoring metrics align to experts in the domain. A scoring
metric is variable based on the optimization equation used, and thus far researchers haven’t
questioned how the layout scoring metric changes the optimal layout created. This study provides
a user evaluation of a layout score, which aligned relatively well with users perceptions.
The third study presented a healthcare hybrid simulation taxonomy of layout
optimization, process simulation, and virtual visualization (Figure 6-4). The main contribution
was the development a framework for implementation throughout a healthcare facility lifecycle,
whereby current data is fed into the framework and decisions can be made as to current and future
demand projections (Figure 6-1). Test scenarios for changing space allocation in redesign and
178
expansion of the facility could also be performed rapidly, and scenarios for different stages of
managing, operating, planning, designing, and constructing a healthcare facility were presented.
6.3 Limitations
The limitations of this work include limitations on the accuracy of the discrete event
simulation, on the sample of healthcare experts, and on the validation of the optimization-
simulation-visualization framework. The main limits of the discrete event simulation are that a
series of assumptions were made to develop the simulation such that the model represents a
simplification of the emergency department workflows in an ideal state. Additionally,
assumptions were made on the future workflow process that could change at any time. Some
service times were assumed based on literature and could have a better numerical description
given additional observation and input data. The level of detail is set on the zone level of the
emergency department and at the patient treatment path based on acuity level. These assumptions
do not model all inter-department activities. Some treatment paths might not make sense in the
model since diagnosis and specific treatment plans were not in the model as a result of the acuity
level probability routing. Bottlenecks from outside departments were modeled as unchanging,
which could change if management in those departments were also working on operational
changes. The model could be expanded to include these parameters if the goals of the study were
to include those areas of the hospital. Likewise, the room-level detail and person-level detail was
not modeled, so additional goals and objectives associated with that level of detail cannot be
explored without further model development activities.
The main limitations of the second study on using and evaluating the graph theoretical
approach to a department adjacency hospital layout problem are associated with the input data,
the methodology, and the evaluation survey. The input data could be studied in more detail to
179
understand if different practitioners have different ideal adjacency requirements. In this study,
several practitioners provided their expert opinion on adjacency ratings, however those are known
to be sensitive to input bias. If the validity of the ratings were explored in more detail and given
more emphasis in the study, these ratings could be formalized and adjusted to owner
requirements. That would require more individuals being involved in the study, increasing the
amount of time needed. Another approach would be to use measured flow data, or more
immediately available, simulated flow data.
The third study focus was on the development of a framework, thus its main limitation is
on validity and limits on translation into other facility types. Industry 4.0 is a vision for the use of
real-time data and analytics in manufacturing and industrial settings (Vieira et al. 2018), where
simulation plays an important role. The healthcare industry is an important area to consider for
development of the Industry 4.0 vision and the OSV framework could be a useful tool for
development and implementation in both healthcare and additional complex system settings. The
future work should focus on validity and translation into generic application areas to provide
adequate detail and vision for implementation.
6.4 Future Work
The future work includes implementation and validation of the different parts of the
framework, the development of software and methodologies to integrate the parts of the
framework, and the automation of the framework. The following sections describe the future
work in these three major areas as well as discuss potential additional future work.
180
6.4.1 Implementation and Validation of the OSV Framework
As discussed in the limitations, the implementation and validation of the OSV framework
is a significant next step in the research laid forth in this dissertation. The scope of validation and
implementation needs to be detailed to understand the complexity of the system and dedicate
appropriate detail to guide implementation and develop the framework for additional application
areas in healthcare, health systems, and additional application areas, for example aviation,
manufacturing, construction, and industrial settings. Testing the framework throughout different
stages of a facility lifecycle is also needed. Immediate testing of the framework with generated
flow data from zones of the emergency department can be done with the current scope of work.
The framework could be validated in different stages and then integrated into the overall
framework (e.g., specific processes described in Figure 6-2 and Figure 6-3).
6.4.2 Development of Software and Methodologies
Thus far advances in this dissertation have described the layout optimization
methodology, the discrete event simulation, and the framework for integration. Additional
research is needed into the methods for implementing and the software for development of hybrid
approaches. Both the human-in-the-loop methodologies and software designed to develop
appropriate layout generation rule sets are useful future directions of research. In addition,
development of software to analyze real-time data can provide additional advantages for updated
baseline models. Additional work in methods for describing both the healthcare system and the
flows of interest could be done at different levels of detail, such as at the disposition level (as
opposed to acuity level).
181
6.4.3 Automation of the OSV Framework
To support data-driven decision making, the right information in the right format needs to
be available when decisions are being made. This is particularly hard in a healthcare environment,
where once a need is determined, there is an immediacy of realizing the solution to the need.
Thus, in order to develop the maximum benefits of the OSV framework, automation is necessary
at different points in the framework. Future research should focus on the automation of collecting
input data; analyzing that data to identify bottlenecks, additional needed capacity, resource
demand changes, and potential workflow changes to aid efficiency; and in the automation of the
workflow between the optimization, simulation, and visualization to streamline “what-if”
scenario testing. In the future, it would be helpful to see data in context of the future facility to aid
the visualization of important tacit information often forgotten in the design and realized as
mistakes later on. Physical mockups of patient rooms (such as that shown in Figure 6-5) could
contain the simulation data in an augmented reality framework. Additionally, virtual mockups
could contain additional simulation data superimposed onto a virtual representation of the facility
in an augmented virtuality framework, aimed at running structured and unstructured tasks for
program and design requirement feedback from healthcare professionals.
Figure 6-5. Physical mockup of a typical new patient room
182
6.5 Concluding Remarks
While there has been a considerable amount of research in both operations and evidence
based design practices for healthcare facilities, there is still a lack of research investigating the
operations of a future facility in the context of its future workflow. This provides a unique
opportunity to look at the facility as an non static and adaptable resource. While layout
optimization continues to be a difficult combinatorial problem, most often NP-hard, there are new
heuristics available, computing power previously unavailable, and healthcare practitioner and
designer driven needs for data to support large capital decisions. Simulation provides a
methodology to generate data and test stochastic systems in the future context. The framework
developed and presented here can be a backdrop to research and development of methodologies
which integrate these computational tools with the human decision makers, bridging the gap
between facility planning algorithms, operations research, and healthcare practitioner workflows.
183
References
Acar, Y., Kadipasaoglu, S. N., and Day, J. M. (2009). “Incorporating uncertainty in
optimal decision making: Integrating mixed integer programming and simulation
to solve combinatorial problems.” Computers & Industrial Engineering, 56(1),
106–112.
AHQR. (2017). Guidelines and Measures. Agency for Healthcare Research and Quality,
US Department of Health and Human Services, Rockville, MD.
Akpan, I. J., and Brooks, R. J. (2012). “Users’ perceptions of the relative costs and
benefits of 2D and 3D visual displays in discrete-event simulation.”
SIMULATION, 88(4), 464–480.
Akpan, I. J., and Brooks, R. J. (2014). “Experimental evaluation of user performance on
two-dimensional and three-dimensional perspective displays in discrete-event
simulation.” Decision Support Systems, 64, 14–30.
Akpan, I. J., and Shanker, M. (2017). “The confirmed realities and myths about the
benefits and costs of 3D visualization and virtual reality in discrete event
modeling and simulation: A descriptive meta-analysis of evidence from research
and practice.” Computers & Industrial Engineering, 112, 197–211.
Almed, M. (2017). “How Long Does an Echocardiogram Take?” MyHeart.
Amaral, A. R. S. (2012). “The corridor allocation problem.” Computers & Operations
Research, 39(12), 3325–3330.
Anjos, M. F., and Vieira, M. V. C. (2017). “Mathematical optimization approaches for
facility layout problems: the state-of-the-art and future research directions.”
European Journal of Operational Research, 261(1), 1–16.
Arisha, A., and Rashwan, W. (2016). “Modeling of healthcare systems: Past, current and
future trends.” IEEE, 1523–1534.
Arnolds, I., and Nickel, S. (2015). “Layout planning problems in health care.”
Applications of Location Analysis, International Series in Operations Research &
Management Science, H. A. Eiselt and V. Marianov, eds., Springer International
Publishing, Cham, Switzerland, 109–152.
Arnolds, I. V., and Gartner, D. (2018). “Improving hospital layout planning through
clinical pathway mining.” Annals of Operations Research, 263(1), 453–477.
Assem, M., Ouda, B. K., and Wahed, M. A. (2012). “Improving operating theatre design
using facilities layout planning.” 2012 Cairo International Biomedical
Engineering Conference (CIBEC), 109–113.
Banks, J., Carson, J. S., Nelson, B. L., and Nicol, D. M. (2010). Discrete-event system
simulation. Prentice Hall, Upper Saddle River, NJ, USA.
Bassanino, M., Fernando, T., and Wu, K.-C. (2014). “Can virtual workspaces enhance
team communication and collaboration in design review meetings?” Architectural
Engineering and Design Management, 10(3–4), 200–217.
184
Batarseh, O. G., Goldlust, E. J., and Day, T. E. (2013). “SysML for conceptual modeling
and simulation for analysis: A case example of a highly granular model of an
emergency department.” 2013 Winter Simulations Conference (WSC),
Washington, D.C., USA, 2398–2409.
Bate, P., and Robert, G. (2006). “Experience-based design: from redesigning the system
around the patient to co-designing services with the patient.” Quality and Safety
in Health Care, 15, 307–310.
Bate, P., and Robert, G. (2007). Bringing User Experience to Healthcare Improvement:
The Concepts, Methods and Practices of Experience-Based Design. Radcliffe,
Oxford, U.K.
van den Berg, M., Hartmann, T., and de Graaf, R. (2017). “Supporting design reviews
with pre-meeting virtual reality environments.” Journal of Information
Technology in Construction (ITcon), 22(16), 305–321.
Berkun, S. (2004). “Programmers, designers, and the Brooklyn Bridge.” Essays on
design, engineering and project management.
Bhattacharjee, P., and Ray, P. K. (2014). “Patient flow modelling and performance
analysis of healthcare delivery processes in hospitals: A review and reflections.”
Computers & Industrial Engineering, 78, 299–312.
Bohannon, R. W. (1997). “Comfortable and maximum walking speed of adults aged 20—
79 years: reference values and determinants.” Age and Ageing, 26(1), 15–19.
Bowers, J., Ghattas, M., and Mould, G. (2009). “Success and failure in the simulation of
an Accident and Emergency department.” Journal of Simulation, 3(3), 171–178.
Brailsford, S. C., Harper, P. R., Patel, B., and Pitt, M. (2009). “An analysis of the
academic literature on simulation and modelling in health care.” Journal of
Simulation, 3(3), 130–140.
Brailsford, S., and Vissers, J. (2011). “OR in healthcare: A European perspective.”
European Journal of Operational Research, 212(2), 223–234.
Bruzzone, A., and Signorile, R. (1998). “Simulation and Genetic Algorithms for Ship
Planning and Shipyard Layout.” SIMULATION, 71(2), 74–83.
Bullinger, H.-J., Bauer, W., Wenzel, G., and Blach, R. (2010). “Towards user centered
design (UCD) in architecture based on immersive virtual environments.”
Computers in Industry, Human-Centered Computing Systems in Industry - A
Special Issue in Honor of Professor G. Salvendy, 61(4), 372–379.
Burmahl, B., Hoppszallern, S., and Morgan, J. (2017). “2017 Hospital construction
survey.” Health Facilities Management, 30(2), 18–24.
Carr, R. F., and WBDG Health Care Subcommittee. (2017). “Health care facilities:
Hospitals.” Whole Building Design Guide, National Institute of Building
Sciences, Washington D.C.
Castronovo, F., Nikolic, D., Liu, Y., and Messner, J. (2013). “An evaluation of
immersive virtual reality systems for design reviews.” CONVR 2013, London,
UK.
Chau, P. Y. K. (1995). “Factors Used in the Selection of Packaged Software in Small
Businesses: Views of Owners and Managers.” Information & Management, 29(2),
71–78.
185
Chen, H.-M., and Huang, P.-H. (2013). “3D AR-based modeling for discrete-event
simulation of transport operations in construction.” Automation in Construction,
Augmented Reality in Architecture, Engineering, and Construction, 33, 123–136.
Choudhary, R., Bafna, S., Heo, Y., Hendrich, A., and Chow, M. (2010). “A predictive
model for computing the influence of space layouts on nurses’ movement in
hospital units.” Journal of Building Performance Simulation, 3(3), 171–184.
Chwif, L., Pereira, W. I., and Montevechi, J. A. B. (2015). “Are visually appealing
simulation models preferable?” Proceedings of the 2015 Winter Simulation
Conference, IEEE Press, Huntington Beach, CA, USA, 835–843.
CMS. (2017). Hospital Compare. Centers for Medicare and Medicaid Services,
Baltimore, MD.
Djanatliev, A., and Meier, F. (2016). “Hospital processes within an integrated system
view: A hybrid simulation approach.” Proceedings of the 2016 Winter Simulation
Conference, IEEE, Arlington, Virginia, 1364–1375.
Duan, W., Ankenman, B. E., Sanchez, S. M., and Sanchez, P. J. (2017). “Sliced Full
Factorial-Based Latin Hypercube Designs as a Framework for a Batch Sequential
Design Algorithm.” Technometrics, 59(1), 11–22.
Dunston, P. S., Arns, L. L., and McGlothin, J. D. (2007). “An Immersive Virtual Reality
Mock-Up for Design Review of Hospital Patient Rooms.” Proceedings of the 7th
International Conference on Construction Applications of Virtual Reality, Penn
State University, University Park, PA.
Dunston, P. S., Arns, L. L., and McGlothin, J. D. (2010). “Virtual reality mock-ups for
healthcare facility design and a model for technology hub collaboration.” Journal
of Building Performance Simulation, 1.
Edwards, B. (2004). The Modern Airport Terminal : New Approaches to Airport
Architecture. Taylor & Francis, London, UK.
ElNimr, A., Fagiar, M., and Mohamed, Y. (2016). “Two-way integration of 3D
visualization and discrete event simulation for modeling mobile crane movement
under dynamically changing site layout.” Automation in Construction, 68, 235–
248.
Elshafei, A. N. (1977). “Hospital layout as a quadratic assignment problem.” Journal of
the Operational Research Society, 28(1), 167–179.
Farahmand, K., Karim, R., Srinivasan, R., Sajjadi, S. R., and Fisher, L. (2011). “Clinic
space design using discrete event simulation.” Proceedings of the IIE Annual
Conference, T. Doolen and E. Van Aken, eds., IISE, Norcorss, Georgia, 1–8.
Figueira, G., and Almada-Lobo, B. (2014). “Hybrid simulation–optimization methods: A
taxonomy and discussion.” Simulation Modelling Practice and Theory,
Simulation-Optimization of Complex Systems: Methods and Applications, 46,
118–134.
Fone, D., Hollinghurst, S., Temple, M., Round, A., Lester, N., Weightman, A., Roberts,
K., Coyle, E., Bevan, G., and Palmer, S. (2003). “Systematic review of the use
and value of computer simulation modelling in population health and health care
delivery.” Journal of Public Health, 25(4), 325–335.
Foulds, L. R., and Robinson, D. F. (1978). “Graph theoretic heuristics for the plant layout
problem.” International Journal of Production Research, 16(1), 27–37.
186
Francis, R. L., McGinnis, L. F., and White, J. A. (1992). Facility layout and location: an
analytical approach. Prentice Hall, Englewood Cliffs, NJ.
Garcia, A. S., Roberts, D. J., Fernando, T., Bar, C., Wolff, R., Dodiya, J., Engelke, W.,
and Gerndt, A. (2015). “A collaborative workspace architecture for strengthening
collaboration among space scientists.” IEEE, 1–12.
Gibson, I. W. (2007). “An approach to hospital planning and design using discrete event
simulation.” Proceedings of the 2007 Winter Simulation Conference, S. G.
Henderson, B. Biller, M.-H. Hsieh, J. D. Shortle, J. D. Tew, and R. R. Barton,
eds., IEEE, Piscataway, New Jersey, 1501–1509.
Gilboy, N., Tanabe, T., Travers, D., and Rosenau, A. M. (2011). Emergency Severity
Index (ESI): A Triage Tool for Emergency Department Care, Version 4.
Implementation Handbook 2012 Edition. Agency for Healthcare Research and
Quality, Rockville, MD.
Gould, F. E. (2012). Managing the construction process: estimating, scheduling, and
project control. Prentice Hall, Boston.
Graham, J. E., Fisher, S. R., Bergés, I.-M., Kuo, Y.-F., and Ostir, G. V. (2010). “Walking
Speed Threshold for Classifying Walking Independence in Hospitalized Older
Adults.” Physical Therapy, 90(11), 1591–1597.
Günal, M. M., and Pidd, M. (2010). “Discrete event simulation for performance
modelling in health care: a review of the literature.” Journal of Simulation, 4(1),
42–51.
Hassan, M. M. D., and Hogg, G. L. (1991). “On constructing a block layout by graph
theory.” International Journal of Production Research, 29(6), 1263–1278.
Healthcare Cost and Utilization Project. (2017). Trends in Emergency Department Visits,
2006-2014. 20.
Holst, M. K. (2015). “Optimal hospital layout design.” Ph.D. Thesis, Aalborg University,
Aalborg, Denmark.
Huddy, J., Bonalumi, N., Minoli, V., Jepson, J., and Meyer, D. (2016). Recommenations
Report - Planning and concept design services for the emergency department.
Penn State Hershey Medical Center, Hershey, PA, 104.
Jun, J. B., Jacobson, S. H., and Swisher, J. R. (1999). “Application of discrete-event
simulation in health care clinics: A survey.” Journal of the Operational Research
Society, 50(2), 109–123.
Kamat, V. R., and Martinez, J. C. (2004). “General-purpose 3D animation with
VITASCOPE.” Simulation Conference, 2004. Proceedings of the 2004 Winter,
1691–1697 vol.2.
Kelsick, J., and Vance, J. (1998). “The VR Factory: Discrete Event Simulation
Implemented in a Virtual Environment.” Mechanical Engineering Conference
Presentations, Papers, and Proceedings.
Kelsick, J., Vance, J. M., Buhr, L., and Moller, C. (2003). “Discrete Event Simulation
Implemented in a Virtual Environment.” Journal of Mechanical Design, 125(3),
428.
Kelton, W. D., Smith, J. S., and Sturrock, D. T. (2014). Simio and simulation: modeling,
analysis, applications. Simio, Sewickley, Pa.
187
Kern, M. (2008). “How long should it take to do a cardiac catheterization?” Cath Lab
Digest, 16(3).
Kim, S.-H., and Nelson, B. L. (2007). “Recent advances in ranking and selection.” 2007
Winter Simulation Conference, IEEE, 162–172.
Kreider, R., and Messner, J. I. (2013). The Uses of BIM. The Pennsylvania State
University, University Park, PA, Available at: bim.psu.edu.
Kuljis, J., Paul, R. J., and Chen, C. (2001). “Visualization and simulation: Two sides of
the same coin?” Simulation, 77(3–4), 141–152.
Kuljis, J., Paul, R. J., and Stergioulas, L. K. (2007). “Can health care benefit from
modeling and simulation methods in the same way as business and manufacturing
has?” 2007 Winter Simulation Conference, 1449–1453.
Kumar, S. (2013). “Experienced-based design review of healthcare facilities using
interactive virtual prototypes.” Doctoral Dissertation, The Pennsylvania State
University, University Park, PA, USA.
Kumar, S., Hedrick, M., Wiacek, C., and Messner, J. I. (2011). “Developing an
Experienced-based Design Review Application for Healthcare Facilities using a
3D Game Engine.” Information Technology in Construction, 16(Special Issue:
Use of Gaming Technology in Architecture, Engineering and Construction), 84–
103.
van der Land, S., Schouten, A. P., Feldberg, F., van den Hooff, B., and Huysman, M.
(2013). “Lost in space? Cognitive fit and cognitive load in 3D virtual
environments.” Computers in Human Behavior, 29(3), 1054–1064.
Law, A. M., and Kelton, W. D. (1991). Simulation modeling and analysis. McGraw-Hill,
New York, NY, USA.
Leicht, R., Kumar, S., Abdelkarim, M., and Messner, J. (2010). “Gaining End User
Involvement Through Virtual Reality Mock-Ups: A Medical Facility Case
Study.” Proceedings of the 27th International Conference on Applications of IT in
the AEC Industry, Cairo, Egypt.
Leung, J. (1992). “A new graph-theoretic heuristic for facility layout.” Management
Science, 38(4), 594–605.
Li, J. P. (2000). “Train Station Passenger Flow Study.” Proceedings of the 2000 Winter
Simulation Conference, A. J. Joines, R. R. Barton, K. Kang, and P. A. Fishwick,
eds., Society for Computer Simulation International, San Diego, CA, USA, 1173–
1173.
Liggett, R. S. (2000). “Automated facilities layout: past, present and future.” Automation
in Construction, 9(2), 197–215.
Liu, Y., Lather, J., and Messner, J. (2014). “Virtual Reality to Support the Integrated
Design Process: A Retrofit Case Study.” Computing in Civil and Building
Engineering (2014), American Society of Civil Engineers, 801–808.
Malmborg, C. J. (1994). “Facility design: The block layout planning process for
manufacturing systems.” Handbook of Design, Manufacturing and Automation,
John Wiley & Sons, Ltd, 461–479.
Manataki, I. E., and Zografos, K. G. (2009). “A generic system dynamics based tool for
airport terminal performance analysis.” Transportation Research Part C:
Emerging Technologies, 17(4), 428–443.
188
Martinez, J. C. (1996). “Stroboscope: State and resource based simulation of construction
processes.” University of Michigan.
Mayo Clinic. (2019). “EEG (electroencephalogram).” Mayo Clinic,
<https://www.mayoclinic.org/tests-procedures/eeg/about/pac-20393875> (May 6,
2019).
McGuire, F. (1998). “Simulation in healthcare.” Handbook of simulation: Principles,
methodology, advances, applications, and practice, J. Banks, ed., John Wiley &
Sons Inc.; Engineering & Management Press, New York, New York, 605–627.
Mobach, M. P. (2008). “Do virtual worlds create better real worlds?” Virtual Reality,
12(3), 163–179.
“Model.” (2019). Merriam-Webster Dictionary.
Moore, J. M. (1971). “Computer program evaluates plant layout alternatives.” Industrial
Engineering, 3(8), 19–25.
Mustafee, N., and Powell, J. H. (2018). “From Hybrid Simulation to Hybrid Systems
Modelling.” Proceedings of the 2018 Winter Simulation Conference, WSC ’18,
IEEE Press, Piscataway, NJ, USA, 1430–1439.
National Academies of Sciences, Engineering, and Medicine. (2007). Hospital-Based
Emergency Care: At the Breaking Point. The National Acadamies Press,
Washington, D.C., USA.
NHS. (2018). “MRI.” nhs.uk, <https://www.nhs.uk/conditions/mri-scan/what-happens/>
(May 6, 2019).
Oh, C., Novotny, A. M., Carter, P. L., Ready, R. K., Campbell, D. D., and Leckie, M. C.
(2016). “Use of a simulation-based decision support tool to improve emergency
department throughput.” Operations Research for Health Care, 9, 29–39.
O’Keefe, R. M. (2016). “Experimental behavioural research in operational research:
What we know and what we might come to know.” European Journal of
Operational Research, 249(3), 899–907.
Rais, A., and Viana, A. (2011). “Operations Research in Healthcare: a survey.”
International Transactions in Operational Research, 18(1), 1–31.
RANZCR. (2016). “Ultrasound.” InsideRadiology, The Royal Australian and New
Zealand College of Radiologists,
<https://www.insideradiology.com.au/ultrasound/> (May 6, 2019).
Rekapalli, P. V., and Martinez, J. C. (2007). “Gaming Perspective Based Visual
Interactive Simulation for Validation of Simulated Construction Operations.”
International Workshop on Computing in Civil Engineering, ASCE, Pittsburgh,
Pennsylvania, USA.
Rekapalli, P. V., and Martinez, J. C. (2011). “Discrete-Event Simulation-Based Virtual
Reality Environments for Construction Operations: Technology Introduction.”
Journal of Construction Engineering and Management, 137(3), 214–224.
Robinson, S. (2002). “General concepts of quality for discrete-event simulation.”
European Journal of Operational Research, 138(1), 103–117.
Robinson, S. (2005). “Discrete-event simulation: from the pioneers to the present, what
next?” Journal of the Operational Research Society, 56(6), 619–629.
Rudd, J., Stern, K., and Isensee, S. (1996). “Low vs. high fidelity prototyping debate.”
Interactions, 3(1), 76–85.
189
Sanvido, V. E., Khayyal, S. A., Guvenis, M., Norton, K., Hetrick, M., Muallem, M.,
Chung, E. K., and Medeiros, D. J. (1990). An integrated building process model.
Computer Integrated Construction Research Program, Department of
Architectural Engineering, the Pennsylvania State University, University Park,
PA.
Shneiderman, B. (1996). “The eyes have it: A task by data type taxonomy for information
visualizations.” Visual Languages, 1996. Proceedings., IEEE Symposium on,
IEEE, 336–343.
St. Michael’s Hospital. (2019). “Medical Imaging CT (CAT) Scan.” St. Michael’s
Hospital, <http://www.stmichaelshospital.com/programs/imaging/ctscan/faq.php>
(May 6, 2019).
Swenson, E. R. (2008). “Using discrete-event simulation to improve patient flow in an
emergency department.” Master of Science in Industrial Engineering, The
Pennsylvania State University, University Park, Pennsylvania.
Swisher, J. R., and Jacobson, S. H. (2002). “Evaluating the Design of a Family Practice
Healthcare Clinic Using Discrete-Event Simulation.” Health Care Management
Science, 5(2), 75–88.
Taylor, S. J. E., Chick, S. E., Macal, C. M., Brailsford, S., L’Ecuyer, P., and Nelson, B.
L. (2013). “Modeling and simulation grand challenges: An OR/MS perspective.”
Proceedings of the 2013 Winter Simulations Conference, IEEE, 1269–1282.
Tompkins, J. A., White, J. A., Bozer, Y. A., and Tanchoco, J. M. A. (2010). Facilities
Planning. John Wiley & Sons.
Turner, C. J., Hutabarat, W., Oyekan, J., and Tiwari, A. (2016). “Discrete Event
Simulation and Virtual Reality Use in Industry: New Opportunities and Future
Trends.” IEEE Transactions on Human-Machine Systems, 46(6), 882–894.
Ulrich, R. S., Zimring, C., Zhu, X., DuBose, J., Seo, H.-B., Choi, Y.-S., Quan, X., and
Joseph, A. (2008). “A review of the research literature on evidence-based
healthcare design.” HERD: Health Environments Research & Design Journal,
1(3), 61–125.
Vahdat, V., Namin, A., Azghandi, R., and Griffin, J. (2019). “Improving patient
timeliness of care through efficient outpatient clinic layout design using data-
driven simulation and optimisation.” Health Systems, 1–22.
Vahdatzad, V., and Griffin, J. (2016). “Outpatient clinic layout design accounting for
flexible policies.” Proceedings of the 2016 Winter Simulation Conference, WSC
’16, IEEE, Piscataway, New Jersey, 3668–3669.
Vieira, A. a. C., Dias, L. M. S., Santos, M. Y., Pereira, G. a. B., and Oliveira, J. A.
(2018). “Setting an Industry 4.0 Research and Development Agenda for
Simulation - a Literature Review.” International Journal of Simulation Modelling
(IJSIMM), 17(3), 377–390.
Wahlström, M., Aittala, M., Kotilainen, H., Yli-Karhu, T., Porkka, J., and Nykänen, E.
(2010). “CAVE for collaborative patient room design: analysis with end-user
opinion contrasting method.” Virtual Reality, 14(3), 197–211.
Waller, A. P., and Ladbrook, J. (2002). “Virtual Worlds: Experiencing Virtual Factories
of the Future.” Proceedings of the 34th Conference on Winter Simulation:
190
Exploring New Frontiers, WSC ’02, Winter Simulation Conference, San Diego,
California, 513–517.
Wilson, J. C. T. (1981). “Implementation of computer simulation projects in health care.”
Journal of the Operational Research Society, 32(9), 825–832.
Zhang, L., Grossmann, J., Stevenson, C., Chi, M., Cauwenberghs, G., Schulze, J., Otto,
P., Jung, T., Peterson, R., Edelstein, E., and Macagno, E. (2011). “Spatial
cognition and architectural design in 4D Immersive virtual reality: testing
cognition with a novel audiovisual CAVE-CAD tool".” Proceedings of the Spatial
Cognition for Architectural Design Conference, SFB/TR 8 Spatial Cognition,
New York, NY, USA.
191
Appendix A.
Additional Response Variables Summary Statistics from Discrete Event
Simulation
Box Plots for Length of Stay for ESI 5 Patients
A-1. Box plots for length of stay for ESI 5 patients for Current and S1 under the
conditions of 2017.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
192
A-2. Box plots for average LOS of ESI 5 patients across runs, S1-S16.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
193
Box Plots for Length of Stay for ESI 4 Patients
A-3. Box plots for average LOS of ESI 4 patients across runs, Current and S1.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
A-4. Box plots for average LOS of ESI 4 patients across runs, S1-S16.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
194
Box Plots for Length of Stay for ESI 3 Patients
A-5. Box plots for average LOS of ESI 3 patients across runs, Current and S1.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
A-6. Box plots for average LOS of ESI 3 patients across runs, S1-S16.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
195
Box Plots for Length of Stay for ESI 2 Patients
A-7. Box plots for average LOS of ESI 2 patients across runs, Current and S1.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
A-8. Box plots for average LOS of ESI 2 patients across runs, S1-S16.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
196
Box Plots for Length of Stay for ESI 1 Patients
A-9. Box plots for average LOS of ESI 1 patients across runs, Current and S1.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
A-10. Box plots for average LOS of ESI 1 patients across runs, S1-S16.
Note: Box plot set at 75% and 25% quantiles, includes 95% confidence interval about the
mean (brown) and quantiles (blue).
197
Appendix B.
Survey and IRB Materials
Survey Procedure
Potential participants were asked participate in a Qualtrics survey via email. In the email
there was a link to the survey, a description of the project, and information about expected length
of time to complete the survey. Potential participants were given 10 days to enroll in the study
and complete the survey. They were reminded about the survey data collection period twice
throughout the data collection period. The survey consisted of several sections collecting
information about general layout planning of a hospital, information about which layout option
works best in 1 scenarios, their opinions about the generated layouts, and their demographic and
experience level information. They were asked at the end if they would be willing to be contacted
for a follow up interview.
Survey Apparatus
(Initial page)
Thank you for your participation.
This survey is meant to get an initial set of input from healthcare planners and strategists on
layout generation techniques. This survey is part of a research project by a team at The
Pennsylvania State University. We have planned this survey to take on average 9 minutes,
depending on your length of responses.
Your participation is voluntary and you can exit the survey at any time. We ask you to participate
to help us gather accurate information from the industry. Your participation, identity, and
responses will be kept confidential and will not be disclosed to your employer or anyone other
198
than those in the research team at PSU. Any information reported from this study will be reported
in an aggregate and anonymous format.
If you have any questions for the research team, you can contact the Principal Investigator of the
study, Jennifer Lather, at [email protected] or at (814) 865-6394 via the Department of
Architectural Engineering, 104 Engineering Unit A, University Park, PA 16802.
Again thank you for your time in providing valuable input to help advance both research and the
industry,
Jennifer Lather
(next page)
In this section, we would like you to give your opinion on some layouts that were automatically
generated. These were generated to give an optimal layout based on a set of adjacency ratings for
the Diagnostic and Treatment and Service departments of a healthcare facility. Each set of 6 have
a few discrete constraints with different 'starting' department and different bay sizes (30', 60', 90').
Imagine you are designing an Ambulatory Surgery Center. You've planned out the space
requirements for 16 departments in the facility, and you are trying to arrange them to provide the
best flow between departments. For each question below select either the best or the worst among
the set.
Please ignore site constraints for these questions.
Legend:
Question 1a. Please select the layout(s) that would function the best. Select at least 1 and up to
3. To select, click anywhere on the layout(s) of choice.
(image)
Response type: three mouse clicks allowed, location (x,y) and order recorded
199
B-1. Layout Scenarios as presented in the survey
Question 1b. Reason for selection. Please provide a brief reason for why you made this choice.
Response type: text entry
Question 2a. Please select the layout(s) that would NOT function well or the worst among the
set. Select at least 1 and up to 3. To select, click anywhere on the layout(s) of choice.
(same layouts given as image above)
Response type: three mouse clicks allowed, location (x,y) and order recorded
Question 2b. Reason for selection. Please provide a brief reason for why you made this choice.
Response type: text entry
(next page)
Questions 3. Have you used a system to auto-generate layout options before?
Response type: multiple choice, with text entry if response was yes or maybe
1. Yes (Please Describe), 2. Maybe (explain), 3. No
200
Question 4. Please respond with your agreement or disagreement with the following statement:
In order to find these useful, I would need to know more about what decisions the system used to
generate these layouts.
Response type: multiple choice 1. Strongly agree, 2. Agree, 3. Somewhat agree, 4. Neither agree nor disagree, 5. Somewhat
disagree, 6. Disagree, 7. Strongly Agree
Question 5. If you were to make a system to auto-generate layouts, what key feature(s) would
you like to adjust? Select all that apply.
Response type: multiple answer
1. Bay size (dimensions), 2. Department size (area), 3. Cost estimates, 4. Bed/capacity
quantities, 5. Demand changes (# patients/year), 6. Typical department layout
configurations, 7. Circulation configurations, 8. Other (specify): (text box)
Question 6. What advantages do you see for using generative layouts in your work and/or in the
larger industry? Please answer openly and truthfully.
Response type: text entry
Question 7. What disadvantages do you see for using generative layouts in your work and/or in
the larger industry? Please answer openly and truthfully.
Response type: text entry
(next page)
Demographic Data
Question 8. What is your age?
Response type: multiple choice
1. <24, 2. 25-29. 3. 30-34, 4. 35-39, 5. 40-44, 6. 45-49, 7. 50-54, 8. 55-59, 9. 60-64, >65
Question 9. What is your gender?
Response type: multiple choice 1. Male, 2. Female, 3. Decline to answer
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
apply.
Response type: multiple answer
1. Strategy, 2. Operational planning, 3. Medical planning, 4. Architecture, 5. Research, 6.
Other
201
Question 14. What region do you work in?
Response type: multiple choice
1. Mid-Atlantic, 2. Midwest, 3. South Central, 4. Mountain, 5. Southeast, 6. Pacific
(next page)
Thank you for taking the time to complete the survey!
Please leave your name and contact information so that the research team can keep track of who
responded. We value your privacy and confidentiality in the responses you provided. This
information is kept confidential and your personal information will not be shared with anyone at
HKS and those not part of the research team. Data collected will be shared as aggregate and
anonymous information.
Follow up. Would you be willing to participate in a follow-up interview or usability study? Y/N
Name and email.
Response type: text entry, optional
202
IRB Documentation
B-1. Layout Planning Study Survey and Interview with Healthcare Planning and Design
Professionals
We would like to know how the IRB Program can better serve you.
Please fill out our survey; it should take about a minute: https://www.research.psu.edu/irb/feedback.ID27
EXEMPTION DETERMINATION
Date: November 12, 2018
From: Philip Frum, IRB Analyst
To: Jennifer Lather
Type of Submission: Initial Study
Title of Study: Layout Planning Study
Principal Investigator: Jennifer Lather
Study ID: STUDY00010926
Submission ID: STUDY00010926
Funding: Not Applicable
Documents Approved: • HRP-591 - LPS Protocol for Human Subject
Research.pdf (0.03), Category: IRB Protocol
• Semi-structured Interview Questions (0.01),
Category: Data Collection Instrument
• Survey Questions (0.02), Category: Data
Collection Instrument
The Office for Research Protections determined that the proposed activity, as
described in the above-referenced submission, does not require formal IRB review
because the research met the criteria for exempt research according to the policies
of this institution and the provisions of applicable federal regulations.
Continuing Progress Reports are not required for exempt research. Record of this
research determined to be exempt will be maintained for five years from the date of
this notification. If your research will continue beyond five years, please contact the
Office for Research Protections closer to the determination end date.
Changes to exempt research only need to be submitted to the Office for Research
Protections in limited circumstances described in the below-referenced Investigator
Manual. If changes are being considered and there are questions about whether IRB
review is needed, please contact the Office for Research Protections.
Penn State researchers are required to follow the requirements listed in the
Investigator Manual (HRP-103), which can be found by navigating to the IRB
Library within CATS IRB (http://irb.psu.edu).
This correspondence should be maintained with your records.
VITA
Jennifer I. Lather
EDUCATION
• MS in Architectural Engineering, Construction Option, College of Engineering, The
Pennsylvania State University, University Park, PA May 2016
• BS in Conservation of Resource Studies, AOI: Regenerative Design, College of Natural
Resources, University of California, Berkeley, Berkeley, CA Aug 2012
• BA in Architecture, College of Environmental Design, University of California, Berkeley,
Berkeley, CA Aug 2012
RESEARCH AND TEACHING EXPERIENCE
• Computer Integrated Construction Research Group, AE, PSU, University Park, PA 2013-2019
• Integrated Delivery of Ultra-High-Performance Buildings, Fellow, Graduate Assistance in
Areas of National Need (GAANN), AE, PSU 2019
• Partnership for Achieving Construction Excellence, Graduate Research Assist., AE, PSU 2018
• AE 372 Introduction to the Building Industry, Teaching Assistant, AE, PSU 2015-2017
• BIM and BMS Sensor Integration, Visiting Graduate Researcher, Department of Computer
Science, University of Auckland, Auckland, New Zealand 2016
• BIM for Energy Retrofits, Consortium for Building Energy Innovation, AE, PSU 2015
• Collaborative Workspaces and Information Technology, Consortium for Building Energy
Innovation, AE, PSU 2014-2015
• Building Energy Informatics, Consortium for Building Energy Innovation, AE, PSU 2013-2014
PROFESSIONAL EXPERIENCE
• HKS Inc., Consulting and LINE Groups, Simulation Graduate Intern, Washington, D.C. 2018
• Aditazz, Inc., Operational Modeling Group, Interactive Intern, Brisbane, CA 2015
• 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
Society of Civil Engineers, Seattle, WA.