1 Department of Operations, Faculty of Economics & Business, University of Groningen 2 Department of Neurology, University of Groningen, University Medical Center Groningen
2013
University of Groningen Author: S. Stegerman Student number: 1840606 e-mail: s.stegerman@gmail.com Supervisors: dr. ir. D.J. van der Zee1M.M.H. Lahr, MSc2 Co-assessor: dr. H. Balsters1
EXPLORING THE
POTENTIAL OF MOBILE
STROKE UNITS: A
COMBINED
ORGANIZATIONAL MODEL
Master thesis, MSc, Technology and Operations Management University of Groningen, Faculty of Economics and Business
ii | P a g e
Abstract
Stroke is a main cause of death worldwide. In developed countries, it even is one of the most common causes of disability.Treatment of acute ischemic stroke (AIS) patients is shown to be effective if started within 4.5 hours after stroke onset.Due to delays in the patient’s path from onset to treatment, in the Netherlands only 11% of all AIS patients are treated with rt-PA while as many as 31% could be treated in optimized settings. It is shown that a large portion of the delay arises pre-hospital.
This observation inspired two research groups to suggest a completely new way of organizing stroke care: pre-hospital diagnosis and treatment with a mobile stroke unit (MSU). A MSU is an ambulance equipped with a computed tomography (CT)-scanner, a point-of-care laboratory, and specialized staff.
The objectives are (1) to develop alternative organizational models for acute stroke care relying on the use of MSUs, and (2) to test and evaluate the developed models for their treatment rates by means of a simulation study.
Different networks of demand coverage, the number of MSUs, and their base locations are designed. This resulted in 13 alternative organizational models for acute stroke care. The accuracy level of emergency medical services dispatchers is of great influence on the effectiveness of MSUs, because they are the ones determining if an emergency call is coming from a possible AIS patient and if a MSU is send. The sensitivity of the alternative organizational models to this factor is determined by testing different accuracy levels of 30%, 78%, and 83%.
All alternative models achieved treatment rates significantly better than the current situation in Groningen (t(78)=2.59, p=0.011). No significant differences were found
between using 1 or 2 MSUs (t(78)=1.60, p=0.113). Locating the MSU (1) at the optimal location(s), (2) at the optimal location(s) next to hospital(s), or (3) at the UMCG resulted in a maximum treatment rate difference of only 0.14%. The test results show that the treatment rate can be improved from currently 21% to 27.6% with only one MSU covering all demand and located next to the central hospital.
iii | P a g e
v | P a g e
Preface
Writing this master thesis and doing the research on which it is build were the last steps toward graduating the master Technology and Operations Management. Four years ago, I did not expect to finish my studentship with the title Master of Science. However, at that time, two of my closest friends pushed me in the right direction and I want to thank them for it. Because after graduating the bachelor Industrial Engineering, I now have completed my master thesis.
Because of all the media attention on the increasing healthcare costs I got really interested in this field of operations management. This motivated me to contribute to slow down the rapidly increasing costs in this sector.
With this thesis I hope to give other students, researchers in the area of
organizational models for stroke care, and neurologists an idea on the potential of the mobile stroke unit in a combined organizational model.
As mentioned before this thesis is written as part of the master Technology and Operations Management. It represents one third of the total points needed to graduate this master.
People I want to thank for contributing to my thesis are the following. I would like to thank my supervisor dr. ir. Durk Jouke van der Zee for making available a lot of his time, the discussions about the content and the feedback on the progress were very motivating and contributed a lot to this thesis. Maarten Lahr for sharing his data set, giving me access to a working paper, and his quick response on my questions. Mr. C.A.J. Vroomen and Mr. R. de Vos for their expert opinions enabling me to validate parts of the simulation model build. dr. H. Balsters for his feedback on the intermediate versions of this thesis.
vii | P a g e
Table of Contents
Abstract ... ii
Preface ... v
Table of Contents ... vii
1. Introduction ... 1
2. Research Objectives and Design ... 2
2.1 Background ... 2
2.2 Research Objectives ... 3
2.3 Research Design ... 4
2.3.1 Context – Organizational Models for Stroke Care ... 4
2.3.2 Diagnosing Current Organizational Model ... 5
2.3.3 Design Alternative Organizational Models ... 5
2.3.4 Evaluate Alternative Organizational Models ... 5
3. Organizing Acute Stroke Care and Designing a MSU Network ... 7
3.1 Acute Stroke Care ... 7
3.2 Organizational Models ... 8
3.3 MSU Network Design ... 9
4. Current System Description ... 11
4.1 Pre-hospital pathway ... 11
4.2 In-hospital Pathway ... 12
5. Analysis ... 13
5.1 Current Organizational Model for Acute Stroke Care Performance ... 13
5.2 Potential Benefit of MSUs ... 13
5.3 Potential Barriers in Adopting MSUs ... 14
6. Design ... 15
6.1 Demand coverage MSUs ... 15
6.2 Number of MSUs ... 15
viii | P a g e
6.3.1 Maximum Coverage Location Problem ... 17
6.3.2 Locations to be Tested ... 18
7. Design and Results of the Simulation Study ... 20
7.1 Experimental Design ... 20 7.2 Simulation Modeling ... 20 7.3 Simulation Setup ... 20 7.4 Results ... 21 7.4.1 Accuracy of Dispatchers ... 21 7.4.2 Demand Coverage ... 22 7.4.3 Number of MSUs ... 23 7.4.4 Location of MSUs ... 23 8. Discussion ... 25
1 | P a g e
1. Introduction
Stroke is a main cause of death worldwide, and in developed countries it is one of the most common causes of disability (Rothwell et al., 2005). Cerebral ischemia causes about 90% of all strokes, with cerebral hemorrhage causing the rest (Caplan, Hier, &D’Cruz, 1983). The most effective therapy for acute ischemic stroke (AIS) is to treat patients as soon as possible with intravenous tissue-type plasminogen activator (rt-PA). Treatment with rt-PA is shown to be effective if started within 4.5 hours after stroke onset (Hacke et al., 2008; The National Institute of Neurological Disorders and Stroke rt-PA Stroke Study Group, 1995). In the Netherlands only 11% of all stroke patients are treated with rt-PA, while as many as 31% could be treated in optimized settings (Bauer, Limburg, &Visser, 2013; Meretoja et al., 2012).
Because the treatment rate of stroke patients is very time dependent a lot of
research has been done to reduce patient delays by improving the organizational model of acute stroke care (Alberts et al., 2011), and as a result major improvements have been made (Evenson, Foraker, Morris, & Rosamond, 2009). Despite the reduction in delays in acute stroke care the onset-to-treatment times (OTT) remain unsatisfactory, mostly because pre-hospital delays have been unaffected (Köhrmann et al., 2011). Moreover, Evenson et al. (2009) show that most time is being lost within this pre-hospital phase.
Recently, this problem led two research groups to suggest a completely new way of organizing acute stroke care: pre-hospital treatment (Kostopoulos et al., 2012; Walter et al., 2010, 2012; Weber et al., 2013). Both groups tested the use of a mobile stroke unit (MSU) to bring the treatment to the patient rather than the other way around, as is
conventional. A MSU is an ambulance equipped with a computed tomography (CT)-scanner, a point-of-care laboratory, and specialized staff. There were minor differences in terms of the staffing and equipment of the MSU used by Kostopoulos et al. (2012) and Walter et al. (2010, 2012), and Weber et al. (2013). The difference in the population setting is more important and suggests that the new approach can be effective in an urban area as well as in a more rural area. These recent studies show that the MSU could reduce the OTT
2 | P a g e
2. Research Objectives and Design
The aim of this chapter is to clarify the objectives of this study, and to show how the study is designed to achieve them. First, it provides some background information on current acute stroke care, the mobile stroke unit, and combining both models. Next, the research objectives are explained, followed by the questions arising from them. The last section of this chapter describes the research design.
2.1 Background
In the dominant traditional organization of the acute stroke care pathway, patients suspected of suffering from acute stroke are transported to the hospital by ambulance. Within this model, the diagnosing and treatment of AIS patients entirely takes place at the emergency department (ED) in the hospital.
In the MSU organizational model a MSU is sent to patients suspected of suffering from acute stroke. Diagnosing and treatment of AIS patients takes place in the MSU, at the patient’s initial location. After diagnosing and treatment the patient is transported to the hospital.
When patients suffer a stroke, it is well established that the sooner they are treated the better they recover (Saver, 2006). For this reason, the acute stroke care pathway is called time dependent. The basic idea behind using MSUs in acute stroke care is to save time by replacing a part of the traditional organizational model, see Figure 1. Sending a MSU eliminates both the time taken for the ambulance staff to diagnose the patient, and the time needed to transport the patient to the hospital.
Figure 1: Simplified Combined Organizational Model for Acute Stroke Care
Study Boundary
Ambulance Emergency Department Call Mobile Stroke UnitDispatching After Care
3 | P a g e
When combining the MSU organizational model with the traditional organizational model, the success of the combined organizational model is very dependent on dispatcher accuracy. A recent study shows that emergency medical service (EMS) dispatchers
correctly diagnose stroke 30% to 83% of the time (Fassbender et al., 2013), in other words a MSU will not be sent to 70% to 17% of the stroke patients. The group of stroke patients not recognized by EMS dispatchers will end up following the traditional organizational model for acute stroke care, possibly with some extra delay.
Until now studies only considered the MSU organizational model isolated from the traditional organizational model for acute stroke care (Kostopoulos et al., 2012; Walter et al., 2010, 2012; Weber et al., 2013). As a result, the potential benefit of MSUs on the treatment rates of the complete system is not yet clear.
The network design of an EMS system determines where to locate how many resources, and is shown to have big impact on the system’s performance (Brotcorne, Laporte, &Semet, 2003; Li, Zhao, Zhu, & Wyatt, 2011). Despite this, the previous studies on the MSU organizational model did not consider the effect of different MSU network designs. They focused instead on one MSU that was located at a convenient location (Kostopoulos et al., 2012; Walter et al., 2010, 2012; Weber et al., 2013).
2.2 Research Objectives
The MSU model clearly has the potential to increase the number of patients treated (Walter et al., 2012; Weber et al., 2013). However, the fact that the success of the MSU organizational model is very dependent on the diagnosis of stroke patients by EMS dispatchers means the organizational model for acute stroke care cannot solely rely on MSUs. As a result, the next step in research should be to examine a combined
organizational model for acute stroke care. Furthermore, this study will investigate the effect of different MSU network designs on the combined organizational models
performance.
Therefore, the research objectives are (1) to develop alternative organizational models for acute stroke care relying on the use of MSUs, and (2) to test and evaluate the developed models for their treatment rates by means of a simulation study.
4 | P a g e
organizational model for acute stroke care, and if so what are the main determinants of such an increase?
Research Questions:
1. What are relevant steps in stroke care and design parameters, when designing a combined organizational model for acute stroke care? (Problem Context)
2. How is acute stroke care organized in the demographic area under investigation? (Problem Situation)
3. How is the model under investigation performing and how can it be improved using MSUs? (Problem Diagnosis)
4. How should the design parameters of the organizational stroke care model with MSUs be set? (Design)
5. Is there a difference in performance between the designed models and the current organizational model for stroke care? (Testing)
6. How can the differences between the organizational models for stroke care be explained? (Design Evaluation)
2.3 Research Design
This study will focus on the situation in the Dutch province of Groningen. The organization of acute stroke care for Groningen is very similar to the dominant traditional organizational model for stroke care (Lahr, Luijckx, Vroomen, Van Der Zee, &Buskens, 2012). The set-up of the study is similar to the regulative cycle developed by Van Strien (1997). However, this study will skip the implementation phase of the regulative cycle.
The remainder of this chapter will give an overview of the methods used during different phases of the methodology, these phases are: (1) context, (2) describing and diagnosing current situation, (3) design alternative organizational stroke care models, and (4) test and evaluate the design.
2.3.1 Context – Organizational Models for Stroke Care
5 | P a g e
2.3.2 Diagnosing Current Organizational Model
Part of this study is a case study of the organizational model for stroke care in Groningen. Because the designed organizational stroke care model is a combination
between the dominant traditional model and MSUs, a detailed explanation of the set-up of the current stroke care model in the province of Groningen will be provided in chapter 4 (research question 2). In chapter 5 the current organizational model for acute stroke care in the province of Groningen is diagnosed, answering the third research question.
Both questions are answered by means of interviews, and an analysis of empirical data from a prior study on the Groningen organizational stroke care model (Lahr et al., 2012). The field experts interviewed are knowledgeable in different areas important to this study. These fields are organizational models for acute stroke care, simulation studies, EMS policies, and neurology.
2.3.3 Design Alternative Organizational Models
Because up to now there has been no research on how to combine MSUs with an existing organizational stroke care model, the design phase in chapter 6 is based on the important design parameters found in chapter 3. A range of settings is developed for each design parameter and by combining these different settings; a number of alternative organizational stroke care models are developed.
To generate and validate creative ideas and to make sure no obvious solutions are missed the field experts mentioned earlier were also consulted during this stage. Finally, the most feasible solutions are listed and explained in such a way they are ready to be tested (research question 5).
2.3.4 Evaluate Alternative Organizational Models
6 | P a g e
7 | P a g e
3. Organizing Acute Stroke Care and Designing a MSU
Network
This chapter provides the context of the design problem. First, it explains important steps in stroke care, followed by the way these steps are currently organized to reduce time delays. Next, the different mobile stroke unit solutions are described. The last section will address the problem of locating the MSUs.
3.1 Acute Stroke Care
Time is an essential factor in acute stroke care. Patients can only be treated within 4.5 hours of stroke onset and the effectiveness of treatment has been shown to decline over time (Hacke et al., 2008; Lees et al., 2010; The National Institute of Neurological Disorders and Stroke rt-PA Stroke Study Group, 1995). Meretoja et al. (2012) show that up to 31% of the patients could be treated in an optimized organizational stroke care model, while currently in the Netherlands only 11% of patients receive treatment (Bauer et al., 2013).
The time between stroke onset and the actual treatment with rt-PA (onset-to-needle time) is in most cases the sum of the pre-hospital time between stroke onset and arrival at the ED (onset-to-door time), and the in-hospital time between arrival at the ED and the treatment with rt-PA (door-to-needle time). The onset-to-door time itself can be broken down into three distinct time periods: (1) time between symptoms onset and deciding to seek medical attention (onset-to-call time); (2) time between decision to seek medical attention and first medical contact (call-to-arrival time);and (3) time between first medical contact and arrival at the hospital (arrival-to-door time) (Evenson et al., 2009).
8 | P a g e Figure 2: The difference between the two types of stroke(Summit Medical Group, 2013)
3.2 Organizational Models
To organize the steps mentioned in the previous section, existing literature
recognizes three major organizational acute stoke care models: (1) comprehensive stroke center (CSC), (2) primary stroke center (PSC), and (3) telemedicine. A CSC is an
9 | P a g e
To address this problem, some recent studies have focused on a completely different approach: bringing the treatment to the patient. At least one of the solutions to this approach, the mobile stroke unit (MSU) (Walter et al., 2012) showed promising results in terms of to-needle times. Liman et al. (2012) incorporated telemedicine in an ambulance, in an attempt to start with the treatment of AIS patients while they are en route to the hospital. They concluded that in their trial the evaluation of stroke symptoms was not at an acceptable level, mainly because of connection problems with the telecommunication. Both Walter et al. (2012) and Weber et al. (2013)equipped an ambulance with CT, a point-of-care laboratory, and medical staff capable of diagnosing and treating AIS. The staffing and as a result the procedures in both mobile units differed slightly. The stroke emergency mobile unit (STEMO) has a neurologist on board, while the MSU has a neuro-radiologist on board, meaning that during the STEMO experiment consultation with a radiologist is needed to finish the diagnoses. An interesting difference between the two studies is the population setting from which they drew their sample, the STEMO being used in a urban setting whilst the MSU operated in a more rural area (Walter et al., 2012; Weber et al., 2013). This is a possible explanation for the fact that the call-to-needle time was longer for the STEMO (median: 58 [50-65] minutes) than for the MSU (median: 38 [34-42] minutes).
3.3 MSU Network Design
Previous research on MSUs analyzed the use of only one unit, and their most important criterion for its location was that it had access to a large group of potential patients. No further information is available from literature. Therefore, the MSU network design is based on a comparable field of research EMS network design. According to literature from this field, important parameters when deploying an EMS network are the base locations, and how many units to allocate to those bases (Brotcorne et al., 2003; Li et al., 2011).
Reviews by Brotcorne et al. (2003) and Li et al. (2011) point out the location set covering model (LSCM) (Toregas, Swain, Re Velle, & Bergman, 1971) and the maximum coverage location problem (MCLP)(Church & Re Velle, 1974) as two of the earliest
10 | P a g e
demand point being covered by the closest ambulance. If an ambulance is away on a call its coverage area is not covered, this reduces the real-life robustness of the solutions found with these tools. According to Brotcorne et al. (2003) and Li et al. (2011) research came up with two solution directions to overcome this drawback: (1) deterministic tools recognizing additional coverage, and (2) probabilistic tools recognizing the utilization of facilities.
11 | P a g e
4. Current System Description
This chapter will explain the current organizational model for acute stroke care in the province of Groningen shown in Figure 3 (see Appendix A for a more in detail
overview).The organization of acute stroke care in Groningen is according to the
comprehensive stroke center model. First, the pre-hospital pathway of the current acute stroke care model will be explained, followed by the current in-hospital pathway.
Figure 3: Stroke Care Pathway Groningen (Lahr, Luijckx, Vroomen, Van Der Zee, &Buskens, 2013).
4.1 Pre-hospital pathway
The pre-hospital pathway starts with the patient suffering a stroke, the onset. After which most of the patients choose one of three options: (1) call EMS, (2) call general practitioner (GP), and (3) self referral.
When choosing option 1, the EMS dispatcher receives the call and will try to find out which service the patient needs as quickly as possible. A dispatcher has three options: (1) send an ambulance with priority A1, (2) send an ambulance with priority A2, and (3) send an ambulance with priority B1. Priority A1 requires an ambulance to be at the patient within 15 minutes, priority A2 requires the ambulance to be there within 30 minutes. The third priority level (B1) means that the patient’s transport to the hospital is scheduled. If a dispatcher recognizes stroke symptoms and the patient’s time from onset is still within 4 hours they will always send an ambulance with priority A1. A call from a GP with stroke indication will always result in dispatching an ambulance with priority level A1.
12 | P a g e
patient’s transport to the hospital will always have priority level A1. Furthermore, they will perform additional action on the way to the hospital, taking a blood sample and pre-notifying the hospital. This pre-notification is a call from the ambulance staff to the emergency department informing the ED staff that a potential stroke patient is on transport to the ED.
If a patient chooses option 2, the GP has to decide whether to call the EMS immediately. If the GP does not call EMS, they will need to decide whether to see the patient in person. GPs recognizing stroke symptoms by phone or in person almost always call EMS if the time from onset is still within the 4.5-hourtime frame.
4.2 In-hospital Pathway
Patients can be arriving at the ED of the hospital in a number of ways. The five main arrival types are: (1) by ambulance with A1 priority level and pre-notification and a blood sample taken, (2) by ambulance with A1 priority level, (3) by ambulance with priority level A2, (4) by ambulance with priority level B, and (5) by their own transport.
13 | P a g e
5. Analysis
After describing the current organizational model for acute stroke care in the previous chapter, this chapter will first give an analysis of the performance of the current organizational model for acute stroke care. Thereafter, it will discuss the potential benefit and barriers in adopting MSUs.
5.1 Current Organizational Model for Acute Stroke Care Performance
The performance of the organizational model for acute stroke care in the province of Groningen is analyzed by Lahr et al. (2012), they found a rt-PA treatment rate of 21.9%. Other results from the same study are: 43.8% of all patients arrived at the hospital in time (within 4.5 hours), 50% of the patients arriving in time received treatment with rt-PA, the median onset-to-door time for the treated patients is 84 minutes, the median door-to-needle time is 35 minutes, and the median onset-to-door-to-needle time is 124 minutes.5.2 Potential Benefit of MSUs
14 | P a g e Figure 4: Basic difference between EMS and MSU
Call-to-Arrival Departure-to-Door Arrival-to-Departure
Call-to-Arrival EMS
MSU
5.3 Potential Barriers in Adopting MSUs
In the current set-up of the MSU one of the staff members is a neurologist. During office hours this will probably not be a problem, as long as the MSU is located next to the hospital. However, outside office hours there is only one neurologist present at the hospital and he or she cannot leave the hospital. Meaning, using the MSU is only possible during office hours or with an extra neurologist shift outside office hours. To avoid the problem the neurologist should be able to stay in the hospital, a solution could be to use telemedicine. Although Liman et al. (2012) showed this would not be feasible, an expert on this topic suggested it will not take long before the issues encountered during this research can be resolved. For this reason, the assumption is made that a neurologist is not needed in the MSU.
Another potential barrier in successful adopting MSUs is the accuracy of EMS
dispatchers. If they are unable to recognize stroke they cannot send a MSU to the patient. The dispatcher accuracy in Groningen is 78%, which is an acceptable level considering that levels in literature range between 30%-83% (Fassbender et al., 2013). An additional effect of this accuracy problem is that dispatchers will also recognize stroke symptoms in
patients not suffering from acute stroke. These so-called false positive recognitions are of influence on the occupation level of MSU because they spent time on patients not
15 | P a g e
6. Design
This chapter considers alternative models for organizing acute stroke care. First, the demand coverage of MSUs is determined. Next, it establishes the number of MSUs needed based on their expected occupation rate. In the last part of this chapter, the optimal choice of locations for the MSUs is determined based on the first two design parameters.
6.1 Demand coverage MSUs
MSUs can be used to cover all demand recognized by EMS dispatchers or they can be used to cover specific demand based on patients characteristics. A MSU is deployed only if it is possible to diagnose and treat the patient within 4.25 hours. Another patient
characteristic is the patient’s location - deploying a MSU if a patient’s transport time to the hospital would exceed a certain threshold. Most of the time saved by using MSUs is the time an ambulance needs to transport a patient to the hospital. Adding to this, the trend towards centralization of healthcare in the Netherlands means there are three
experiments concerning demand coverage. One experiment concerned MSUs covering all patients, and two experiments with MSUs covering patients not living in the neighborhood of the hospital In the first case, more than 10 minutes away from the hospital, and in the second more than 20 minutes away (see Figure 5).
6.2 Number of MSUs
The number of MSUs for the experiments is based on their estimated occupation rate. Based on data from Lahr et al. (2012) every 2 days in Groningen,3 people suffer from AIS. To be precise, 283 AIS patients in 181 days, i.e. 1.56 patients per day, i.e. 1.09*10-3
per minute (ra).
16 | P a g e
Hopp & Spearman (2008) calculate the utilization as follows; with m being the number of stations. Resulting in a utilization with m = 1 of 8.3% and a utilization with m = 2 of 4.1%.This suggests two experiments to investigate the effect of MSU occupation: an experiment with one MSU and an experiment with two MSUs.
Figure 5: Demand coverage MSU; Green >10 minutes, Red >20 minutes
6.3 Locating of MSUs
17 | P a g e
Considering this, the use of the MCLP seems most appropriate in solving the locating problem.
6.3.1 Maximum Coverage Location Problem
The MCLP introduced by Church and Re Velle (1974) tries to optimize demand
coverage with a restriction on the number of available resources. The following notation is used to describe the model.
the set of demand points; the index for demand points;
the set of potential resource locations;
the set of the facility sites covering demand point ; the index for the potential resource locations; the total number of available resources; the population size of demand point ;
binary variable, equal to 1 if and only if demand point is covered at least once;
binary variable, equal to 1 if and only if a resource is located at site .
The model is presented as follows:
max (1)
Subject to (2)
(3)
(4)
As mentioned earlier the objective function (1) tries to maximize the demand coverage. The first constraint (2) guarantees that a demand point is only covered if one or more resources are located within the distance standard, and the second constraint (3)
guarantees that the number of resources used equals the number of resources available. Besides the limit on resources, another advantage of the model is the possibility to
18 | P a g e
and the relatively low number of patients compared to normal ambulances, locating them next to a hospital will probably be the most cost-effective solution. A drawback of this model mentioned by literature is the lack of coverage if a resource is called for emergency services (Li et al., 2011), but the relatively low occupancy of MSUs compared to normal ambulances mitigates this disadvantage.
6.3.2 Locations to be Tested
Not only is the MSU itself an expensive resource, the staff needed is far too expensive to be working less than 50% of the time. With this in mind, two possible designs are assessed, locating MSUs at the best possible position or locating MSUs at the best-located hospital. In Table 1 the locations for the different settings found with the MCLP model are given. Figure 6 shows these locations on a map of Groningen.
20 | P a g e
7. Design and Results of the Simulation Study
In the previous chapter, thirteen alternatives are designed to organize stroke care depending on MSUs.This chapter will first present the design of the experiments to test the developed alternatives. Next, this chapter will introduce the simulation software and the setup of the experiments. The last part of this chapter presents the results sorted by experimental factor.
7.1 Experimental Design
Fixed factors: The inter-arrival rate of patients is modeled using a negative
exponential distribution, see Appendix B for the goodness of fit on the empirical distribution.
Experimental factors: The demand covered by MSUs (all demand, and patients more
than 10 minutes and 20 minutes away from the central hospital), the number of MSUs (one or two), and their location (optimal, next to a hospital, next to the central hospital) are the experimental factors, resulting in 13 different network designs.The sensitivity of these set-ups with respect to the dispatcher accuracy is tested.
Output measures: To show the effectiveness of the different designs, they will be
evaluated by comparing their treatment rates.
7.2 Simulation Modeling
To carry out the simulation experiments Plant SimulationTM 10.1 (Siemens PLM
Software, 2011) is used. There is no need for a warm-up period. Because of the relatively low occupation, the interaction effects between patients will be very low: during the experiments, the system will be empty most of the time, the same condition as it starts in. The chosen run length of the experiment is one year (365 days). To ensure the 95% confidence interval lies within 2.5% of the calculated treatment rate, 40 runs per experiment are needed.
7.3 Simulation Setup
The different design options chosen in the previous chapter generate eight different experiments. Varying the demand of the system by changing the accuracy of the
21 | P a g e
(83% true positives, and 30% true positives). Adding two benchmark experiments gave some idea about the possible range of improvement. The first experiment does not use the MSU at all, and the second experiment uses a large number of MSUs without travel time to the patient. These extra experiments make 41 different experiments in total, they are all listed in Appendix F.
7.4 Results
This section presents the results per experimental setting. The section starts with presenting the sensitivity of the combined model to the accuracy of dispatchers. Results of the other experimental settings are most of the time given with the dispatcher accuracy level at 78%, the current level in Groningen (Table 2). The mean treatment rates are compared using a paired sample t-test.
To show if a certain change in design parameter results in a significant change in treatment rate comparisons are made between experiments that only differ with respect to that design parameter. If there are more such experiments available, the experiment with the smallest difference in mean is used to test on significance. As a general result it can be stated that even the smallest increase in treatment rate (M=22.1%, SD=1.94) of all experiments is significantly higher (t(39)=5.50, p<0.001) than the benchmark representing the current situation in Groningen (M=21.0%, SD=1.92). For the complete list of results, see Appendix F.
7.4.1 Accuracy of Dispatchers
22 | P a g e
A side effect of aiming for higher dispatcher accuracy levels could be an increasing level of false positive recognition. However, there is no significant decrease in treatment rate even if the number of false positives is increased by 50%. Yet, the difference between deploying 1 (M=27.2%, SD=1.89) or 2 (M=28.1%, SD=1.77) MSUs located at the central hospital did become significant (t(39)=5.4, p<0.001).
7.4.2 Demand Coverage
The different coverage settings tested are (1) covering all patients, (2) only covering patients living more than 10 minutes away from the hospital, and (3) only covering
patients living more than 20 minutes away from the hospital. The treatment rates of these alternatives range respectively between 27.5%-28.0%, 26.1%-26.8%, and 23.3%-23.6%, see Table 2. Comparing the alternative models, covering only patients living more than 10 minutes away (M=26.8%, SD=2.23) with the models covering all patients (M=28.0%, SD=2.45) reveals that the models covering all patients perform significantly better (t(39)=5.87, p<0.001). Alternative models covering only patients living more than 20 minutes away (M=23.3%, SD=2.07) perform significantly worse than the models covering only patients living more than 10 minutes away (M=26.0%, SD=2.08) (t(39)=12.9, p<0.001).
Table 2: Experiment results with dispatcher accuracy of 78%
23 | P a g e
7.4.3 Number of MSUs
In the alternative organizational models designed, the number of MSUs is either 1 or 2. The models with higher dispatcher accuracy showed a slightly bigger difference in treatment rate between using 1 or 2 MSUs. However, even the biggest difference in mean treatment rate between using 1 MSU (M=26.2%, SD=2.52) and 2 MSUs (M=27.0%, SD=2.15) is not significant (t(39)=4.07, p<0.001), Table 3. Further analysis revealed that, if MSUs were located at the central hospital the difference between one (M=27.1%, SD=1.37) or two (M=27.9%, SD=1.79) MSUs becomes almost significant (t(39)=1.93, p=0.065) when the number of patients is increased by approximately 25%.
Table 3: Biggest difference in treatment rate between using 1 or 2 MSUs
Nu m b er Use MSU Di sp at ch A cc u ra cy De m and Co verag e MSU Nu m b er of MSUs Lo cat io n 1 Lo cat io n 2 Perc Treat m en t 36 TRUE 83% 10 1 H 26.2 37 TRUE 83% 10 2 J G 27.0
7.4.4 Location of MSUs
In the choice of location for the MSUs three alternative options were tested. Locate the MSU (1) at the optimal location(s), (2) at the optimal location(s) next to hospital(s), and (3) at the UMCG. The maximum difference in treatment rate between these three alternatives is with 0.14% far from significant, Table 4.
Table 4: Results of different MSU locating decisions
24 | P a g e
25 | P a g e
8. Discussion
The test results, as presented in chapter 7, show that all alternative organizational models with MSUs perform better than the current organizational model for stroke care in Groningen: this is true even if dispatcher accuracy falls as low as 30% and the MSUs only cover a part of the demand. The potential benefit of MSUs is as expected very sensitive to the accuracy of dispatchers and the results indicate that it should be a focus point when adopting MSUs.
The overall performance of the organizational models where MSUs only cover part of the patients is far worse than the models with MSUs covering all demand. However, the main reason for this is the low occupation level of the MSUs. In scenarios with higher MSU occupation, it could still be worthwhile to reconsider this option.
In Groningen, the patient arrival rate is at such a low level that using more than one MSU does not have a significant impact on the performance of the organizational model. The experiments did show some small trends towards better performance with a higher number of MSUs. Therefore, the conclusion is that if occupation levels of the MSUs get higher the number of MSUs is important when designing the network.
The choice of locations for the MSUs does not have a great impact on the performance of the organizational model, at least in the Groningen case. This design parameter also shows a slight trend toward better improvement if MSUs are located at optimal locations. A larger number of patients or a larger service area could amplify this effect.
Figure 7: Onset-to-Treatment Time Patients
26 | P a g e
Next to the increasing treatment rate, an interesting additional effect of the combined organizational model is a shift in the OTT of treated patients. In the current situation, most treated patients have an OTT between 90 and 180 minutes, if MSUs are deployed most of them have an OTT between 0 and 90 minutes (Figure 7). This shift toward shorter OTTs possibly enhances the positive effect of deploying MSUs on the recovery of AIS patients (Hacke et al., 2004; Lees et al., 2010; Saver et al., 2013)
In a combined organizational model the treatment rates of the traditional model decrease. In case of a 78% dispatcher accuracy level and MSUs covering all demand, this treatment rate even decreases to about 10%. This drop in treatment rates of the
27 | P a g e
9. Conclusion and Future Research
Concluding, there is a lot of potential in combining the dominant organizational stroke care model with the MSU model. Locating one MSU next to the central hospital will be the most cost effective way of implementing this. The treatment rate will increase from 21% to 27.6%, and it would allow the MSU staff to be productive in other ways while the MSU is waiting on patients. If the accuracy of EMS dispatchers improves, the treatment rate becomes more sensitive to the number of MSUs and their demand coverage.
This study discovered several areas for further research. Since the overall performance of the organizational model decreased when MSUs only covered patients living further away from the hospital, it would be interesting to know if the performance on the targeted patients does improve significantly.
I | P a g e
References
Alberts, M. J., Hademenos, G., Latchaw, R. E., Jagoda, A., Marler, J. R., Mayberg, M. R., Starke, R. D., et al. (2000). Recommendations for the establishment of primary stroke centers. Brain Attack Coalition. JAMA : the journal of the American Medical Association, 283(23), 3102–9. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10865305
Alberts, M. J., Latchaw, R. E., Jagoda, A., Wechsler, L. R., Crocco, T., George, M. G., Connolly, E. S., et al. (2011). Revised and updated recommendations for the establishment of primary stroke centers: a summary statement from the brain attack coalition. Stroke; a journal of cerebral circulation, 42(9), 2651–65. doi:10.1161/STROKEAHA.111.615336
Alberts, M. J., Latchaw, R. E., Selman, W. R., Shephard, T., Hadley, M. N., Brass, L. M., Koroshetz, W., et al. (2005). Recommendations for comprehensive stroke centers: a consensus statement from the Brain Attack Coalition. Stroke; a journal of cerebral circulation, 36(7), 1597–616. doi:10.1161/01.STR.0000170622.07210.b4
Bauer, A., Limburg, M., & Visser, M. C. (2013). Variation in clinical practice of intravenous thrombolysis in stroke in the Netherlands. Cerebrovascular diseases extra, 3(1), 74–7. doi:10.1159/000350707
Brotcorne, L., Laporte, G., & Semet, F. (2003). Ambulance location and relocation models. European Journal of Operational Research, 147(3), 451–463. doi:10.1016/S0377-2217(02)00364-8
Caplan, L. R., Hier, D. B., & D’Cruz, I. (1983). Cerebral embolism in the Michael Reese Stroke Registry. Stroke; a journal of cerebral circulation, 14(4), 530–536. doi:10.1161/01.STR.14.4.530
Church, R. L., & Re Velle, C. S. (1974). The maximum covering location problem. Papers Reg. Sci. Assoc., 32(1), 101–18.
Demaerschalk, B., Miley, M., Kiernan, T., Bobrow, B., Corday, D., Wellik, K., Aguilar, M., et al. (2009). Stroke Telemedicine. Mayo Clin Proc, 84(January), 53–64.
Evenson, K. R. Ã., Foraker, R. E., Morris, D. L., & Rosamond, W. D. (2009). Topical review A comprehensive review of prehospital and in-hospital delay times in acute stroke care. International journal of stroke : official journal of the International Stroke Society, 4(June), 187–199.
Fassbender, K., Balucani, C., Walter, S., Levine, S. R., Haass, A., & Grotta, J. (2013). Streamlining of prehospital stroke management: the golden hour. Lancet neurology, 12(6), 585–596. doi:10.1016/S1474-4422(13)70100-5
Hacke, W., Donnan, G., Fieschi, C., Kaste, M., Von Kummer, R., Broderick, J. P., Brott, T., et al. (2004). Association of outcome with early stroke treatment: pooled analysis of ATLANTIS, ECASS, and NINDS rt-PA stroke trials. Lancet, 363(9411), 768–74.
doi:10.1016/S0140-6736(04)15692-4
Hacke, W., Kaste, M., Bluhmki, E., Brozman, M., Dávalos, A., Guidetti, D., Larrue, V., et al. (2008). Thrombolysis with Alteplase 3 to 4.5 Hours after Acute Ischemic Stroke. N Engl J Med, 359(13), 1317–1329.
Hopp, W. J., & Spearman, M. L. (2008). Factory Physics (Internatio.). New York: McGraw-Hill/Irwin.
Köhrmann, M., Schellinger, P. D., Breuer, L., Dohrn, M., Kuramatsu, J. B., Blinzler, C., Schwab, S., et al. (2011). Avoiding in hospital delays and eliminating the three-hour effect in thrombolysis for stroke. International journal of stroke : official journal of the International Stroke Society, 6(6), 493–7. doi:10.1111/j.1747-4949.2011.00585.x
Kostopoulos, P., Walter, S., Haass, A., Papanagiotou, P., Roth, C., Yilmaz, U., Körner, H., et al. (2012). Mobile stroke unit for diagnosis-based triage of persons with suspected stroke. Neurology, 78(23), 1849–52. doi:10.1212/WNL.0b013e318258f773
Lahr, M. M. H., Luijckx, G. J., Vroomen, P. C. A. J., Van Der Zee, D. J., & Buskens, E. (2012). Proportion of patients treated with thrombolysis in a centralized versus a decentralized acute stroke care setting. Stroke; a journal of cerebral circulation, 43(5), 1336– 40. doi:10.1161/STROKEAHA.111.641795
II | P a g e Law, A. M. (2007). Simulation Modeling and Analysis (Fourth Edi., p. 768). McGraw-Hill.
Lees, K. R., Bluhmki, E., Von Kummer, R., Brott, T. G., Toni, D., Grotta, J. C., Albers, G. W., et al. (2010). Time to treatment with intravenous alteplase and outcome in stroke: an updated pooled analysis of ECASS, ATLANTIS, NINDS, and EPITHET trials. Lancet, 375(9727), 1695–703. doi:10.1016/S0140-6736(10)60491-6
Li, X., Zhao, Z., Zhu, X., & Wyatt, T. (2011). Covering models and optimization techniques for emergency response facility location and planning: a review. Mathematical Methods of Operations Research, 74(3), 281–310. doi:10.1007/s00186-011-0363-4
Liman, T. G., Winter, B., Waldschmidt, C., Zerbe, N., Hufnagl, P., Audebert, H. J., & Endres, M. (2012). Telestroke ambulances in prehospital stroke management: concept and pilot feasibility study. Stroke; a journal of cerebral circulation, 43(8), 2086–90. doi:10.1161/STROKEAHA.112.657270
Meretoja, A., Strbian, D., Mustanoja, S., Tatlisumak, T., Lindsberg, P. J., & Kaste, M. (2012). Reducing in-hospital delay to 20 minutes in stroke thrombolysis. Neurology, 79(4), 306–13. doi:10.1212/WNL.0b013e31825d6011
Rothwell, P. M., Coull, a J., Silver, L. E., Fairhead, J. F., Giles, M. F., Lovelock, C. E., Redgrave, J. N. E., et al. (2005). Population-based study of event-rate, incidence, case fatality, and mortality for all acute vascular events in all arterial territories (Oxford Vascular Study). Lancet, 366(9499), 1773–83. doi:10.1016/S0140-6736(05)67702-1
Saver, J. L. (2006). Time is brain--quantified. Stroke; a journal of cerebral circulation, 37(1), 263–6. doi:10.1161/01.STR.0000196957.55928.ab
Saver, J. L., Fonarow, G. C., Smith, E. E., Reeves, M. J., Grau-sepulveda, M. V, Hernandez, A. F., Peterson, E. D., et al. (2013). Time to Treatment With Intravenous Tissue Plasminogen Activator and Outcome From Acute Ischemic Stroke. JAMA : the journal of the American Medical Association, 23, 2480–2488.
Siemens PLM Software. (2011). Plant SimulationTM 10.1. Stuttgart, Germany.
Summit Medical Group. (2013). No Title. Retrieved June 11, 2013, from http://www.summitmedicalgroup.com/library/adult_care/ac-strokehemorrhagic_dc/
The National Institute of Neurological Disorders and Stroke rt-PA Stroke Study Group. (1995). Tissue plasminogen activator for acute ischemic stroke. N Engl J Med, 333(24), 1581–1588.
Toregas, C., Swain, R., Re Velle, C. S., & Bergman, L. (1971). The location of emergency service facilities. Operations Research, (19), 1363–1373.
Van Strien, P. J. (1997). Towards a Methodology of Psychological Practice: The Regulative Cycle. Theory & Psychology, 7(5), 683–700. doi:10.1177/0959354397075006
Walter, S., Kostopoulos, P., Haass, A., Keller, I., Lesmeister, M., Schlechtriemen, T., Roth, C., et al. (2012). Diagnosis and treatment of patients with stroke in a mobile stroke unit versus in hospital: a randomised controlled trial. Lancet neurol, 11(5), 397–404. doi:10.1016/S1474-4422(12)70057-1
Walter, S., Kostpopoulos, P., Haass, A., Helwig, S., Keller, I., Licina, T., Schlechtriemen, T., et al. (2010). Bringing the hospital to the patient: first treatment of stroke patients at the emergency site. PloS one, 5(10), e13758. doi:10.1371/journal.pone.0013758 Weber, J. E., Ebinger, M., Rozanski, M., Waldschmidt, C., Wendt, M., Winter, B., Kellner, P., et al. (2013). Prehospital thrombolysis in
III | P a g e
IV | P a g e
Appendix B.
Data analysis
All statistical tests are executed with a 95% confidence interval.
Goodness of fit negative exponential distribution on empirical distribution
patient inter-arrival times
Figure B1: Empirical and theoretical Negative Exponential distribution
Table B1:Chi-Square Test Goodness of fit Negative Exponential Distribution
Distribution Chi statistic Chi value Result Chi Negexp 26.2049689440993 28.8708550104965 true
Table B2: Kolgomorov-Smirnov Test Goodness of fit Negative Exponential Distribution
Distribution KS statistic KS value Result KS Negexp 1.32272742381563 1.358 true 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1000 2000 3000 4000 5000 6000 7000 8000 P( X ≤ x)
Inter-Arrival Time (Seconds)
Distribution Interarrival Time Patients
V | P a g e
Google maps travel time as predictor of Ambulance travel times
Figure B2: Regressielijn Google maps times predicting Ambulance times, linear fit
Table B3-5 : SPSS Outputs regression model; Google maps travel time predicting ambulance travel time
ANOVAa
Model Sum of Squares df Mean Square F Sig.
1
Regression 3073.906 1 3073.906 150.219 .000b Residual 1309.624 64 20.463
Total 4383.530 65 a. Dependent Variable: Empirical
b. Predictors: (Constant), Google_Maps
Coefficientsa
Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) 1.338 1.422 .941 .350 Google_Maps .646 .053 .837 12.256 .000 a. Dependent Variable: Empirical
y = 0.64x + 1.338 R² = 0.705 0 5 10 15 20 25 30 35 40 45 0 10 20 30 40 50 Verband Reistijden Linear (Verband Reistijden) Model Summary
Model R R Square Adjusted R Square
VI | P a g e
Goodness of fit normal distribution on corrected travel times Google maps.
Figure B3: Frequency Table Corrected Travel Times 2 MSU
Table B6: Chi-Square Test Goodness of fit Normal Distribution
Distribution Chi statistic Chi value Result Chi Normal 9.09731010635336 14.0641087935413 true
Table B7: Kolgomorov-Smirnov Test Goodness of fit Normal Distribution
Distribution KS statistic KS value Result KS Normal 0.807696464662401 1.358 true
Figure B4: Frequency Table Corrected Travel Times 2 MSU 0 5 10 15 20 25 0 -2.5 2.5 -5 5 -7.5 7.5 -10 10 -12.5 12.5 -15 15 -17 .5 17.5 -20 20 -22.5 22.5 -25
All demand/Hospital/2 MSU
0 2 4 6 8 10 12 14 0 -2.5 2.5 -5 5 -7.5 7.5 -10 10 -12.5 12.5 -15 15 -17.5 17.5 -20 20 -22.5 22.5 -25 25 -27 .5 27.5 -30 30 -32.5
VII | P a g e Table B8: Chi-Square Test Goodness of fit Normal Distribution
Distribution Chi statistic Chi value Result Chi Normal 13.0326673393021 14.0641087935413 true
Table B9: Kolgomorov-Smirnov Test Goodness of fit Normal Distribution
Distribution KS statistic KS value Result KS Normal 1.01930540186066 1.358 true
Goodness of fit Normal Distribution on travel times Google maps from
patients to hospital MSU
Figure B5: Frequency Table Travel Times Patient to Hospital MSU
Table B10: Chi-Square Test Goodness of fit Normal Distribution
Distribution Chi statistic Chi value Result Chi Normal 13.0382343565135 14.0641087935413 true
Table B11: Kolgomorov-Smirnov Test Goodness of fit Normal Distribution
Distribution KS statistic KS value Result KS Normal 1.13185312641707 1.358 true 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 18.0 0 -5 5 -10 10 -15 15 -20 20 -25 25 -30 30 -35 35 -40 40 -45 45 -50
VIII | P a g e
Appendix C.
Conceptual Simulation Model
Modeling objectives
Estimate the potential improvement in treatment rates for the designed alternative organizational models for stroke care.
The design options that are going to be considered are the dispatcher accuracy, the demand coverage, the number of MSUs, and the location of the MSU.
General objectives
Time-scale is of great importance, the complete study has to be finished before June 22.
The flexibility needed from the model is not very high, the input may be changing during the study but the outputs are very clear and probably won’t change.
The run-speed should not be too long, max 1 hour.
Keep Visual display at a minimum level but the model will be shown to neurologists, so the logic should be almost self-explaining.
The model is probably only going to be used by the modeler and if there are other users, they will have a lot of modeling experience.
The model should be reusable with other input data to be able to simulate other geographic areas.
Model inputs and outputs
The identification of the models inputs and outputs.
Model outputs
Outputs determining achievement of objective:
Patient treatment rate
Outputs determining reasons for failure to meet objectives:
IX | P a g e
Model inputs
Demand coverage of MSUs Number of MSUs
Location of MSUs
Accuracy EMS Dispatcher
Determine model content
This paragraph will explain the scope of the simulation model and the level of detail that it should achieve.
Table 5: AIS care pathway model: model scope
Components Include? Justification
Entities
Patient Include Key influence on treatment rate MSU Include Experimental factor
Ambulance EXCLUDE Assume always available Activities
Onset Stroke EXCLUDE Delay is taken into account Patient call Include Key influence on treatment rate GP receives call EXCLUDE Delay is taken into account GP visits patient EXCLUDE Delay is taken into account Diagnose based on call Include Key influence on treatment rate Dispatch MSU or
Ambulance Include Experimental factor
MSU to Patient Include Key influence on treatment rate Ambulance to Patient Include Key influence on treatment rate Diagnose Patient
Ambulance Include Key influence on treatment rate Diagnose Patient MSU Include Key influence on treatment rate Ambulance to Hospital Include Key influence on treatment rate Point-of-Care MSU Include Key influence on treatment rate CT scan MSU Include Key influence on treatment rate Prepare rt-PA MSU Include Key influence on treatment rate Treat with rt-PA MSU Include Key influence on treatment rate
MSU to Hospital Include Key influence on CT MSU if transport is done by MSU. Move to Post Include Key influence on CT MSU
X | P a g e Queues
Emergency Department EXCLUDE Queues very exceptional. Patient Call EXCLUDE
If no MSU available, assume Ambulance available. Queues in front of EMS call-center very exceptional
Resources
EMS Dispatcher EXCLUDE Required for dispatching MSU or Ambulance,assume always present. EMS staff EXCLUDE Required for driving and nursing patients,assume always present. Neurologist EXCLUDE Assume not needed in MSU and always present in hospital. MSU: Mobile Stroke Unit, ED: Emergency Department, GP: General Practitioner
Table 6: Level of detail AIS care pathway model
Components Detail Include/Exclude Justification
Entities
Patient Quantity Include Average number of stroke patients per day Arrival Pattern Include Inter-arrival time stroke
patients influences use of MSU.
Attributes: Stroke yes/no, Location, Who to call
Include
Routing EXCLUDE Determined by process not patient
MSU Quantity Include Experimental factor Arrival Pattern EXCLUDE
Attributes EXCLUDE
Routing EXCLUDE Determined by process not MSU
Activities
All activities Quantity Various Nature Various Cycle time Various Breakdown/repair EXCLUDE Set-up/changeover EXCLUDE Resources EXCLUDE Shifts EXCLUDE Routing Include
Receive Patient Call Quantity EXCLUDE Assume infinite Nature Include Who to call changed Cycle time Include Key influence on treatment
rate
Routing Include Patient to Diagnose based on call
Diagnose based on call Quantity EXCLUDE Assume infinite Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include Patient to Dispatch MSU or Ambulance or Patient exits system
Dispatch MSU or Ambulance Quantity EXCLUDE Assume infinite Nature EXCLUDE Attributes of patient are
changed
XI | P a g e Routing Include Patient to MSU to Patient
or to Ambulance to Patient. (Experimental Factor)
Ambulance to Patient Quantity EXCLUDE Assume infinite Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include Patient to Diagnose Patient Ambulance
MSU to Patient Quantity Include Number of MSUs Nature Include Combine Patient and MSU Cycle time Include Key influence on treatment
rate
Routing Include MSU with Patient to Point-of-Care MSU
Diagnose Patient Ambulance Quantity EXCLUDE Assume infinite Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include Patient to Ambulance to Hospital
Ambulance to Hospital Quantity EXCLUDE Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include Patient to Laboratory Tests ED
Point-of-Care MSU Quantity Include Number of MSUs Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include MSU with Patient to CT scan MSU
CT scan MSU Quantity Include Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include MSU with Patient to Diagnose Patient MSU Diagnose Patient MSU Quantity Include Number of MSUs
Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include MSU with Patient to Prepare rt-PA MSU or to MSU to Hospital Prepare rt-PA MSU Quantity Include Number of MSUs
Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include MSU with Patient to Treat with rt-PA MSU
Treat with rt-PA MSU Quantity Include Number of MSUs Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include MSU with Patient to MSU to Hospital
XII | P a g e Cycle time Include Key influence on treatment
rate
Routing Include Patient exit system MSU to Move to Post
Move to Post Quantity Include Number of MSUs Nature EXCLUDE
Cycle time Include MSU needs cleaning Routing Include If Patient arrives move to
MSU to Patient Laboratory Tests ED Quantity EXCLUDE Assume infinite
Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include Patient to Queue CT scan ED
CT scan ED
Quantity Include 1, only one CT scan standby
Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include Patient to Diagnose Patient ED
Diagnose Patient ED Quantity Include 1, Patient still in CT scan Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include Patient to Prepare rt-PA ED or exit system
Prepare rt-PA ED Quantity EXCLUDE Assume infinite Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include Patient to Treat with rt-PA ED
Treat with rt-PA ED Quantity EXCLUDE Assume infinite Nature EXCLUDE
Cycle time Include Key influence on treatment rate
Routing Include Patient exit system MSU: Mobile Stroke Unit, ED: Emergency Department, GP: General Practitioner
Identify possible simplifications
Point of care MSU, CT-scan MSU, and Diagnose MSU are combined in Parallel Activities MSU.
Treat and prepare rt-PA MSU are combined in TPA mixing MSU.
Laboratory tests ED, CT-scan ED, and Diagnose patient ED are combined in Parallel Activities ED.
Assumptions
P
ATIENT_C
ALLXIII | P a g e
MSU_T
O_P
ATIENT_H
OMEAssumed neurologist not needed on board of MSU.
Assumed that the time in minutes MSU staff (nurse, driver) needs to get to MSU: minimal 0, maximal 2, likely 1. Triangle distribution.
Assumed that MSU travels with same speed as an ambulance with highest priority (A1).
Assumed normal distribution with mean and standard deviation changing per experiment based on designed location. Travel times based on average corrected Google maps travel times from chosen MSU locations to covered demand.
Correction based on Lahr et al. (2012) empirical data, see Appendix B.
Assumed that time needed to get patient in MSU is equal to scoop and run delay. Only 50% of the time needed that is used in normal situation as modeled by Lahr et al. (2012).
P
ARALLEL_A
CTIVITIES_MSU
Assumed that delay is equal to in-hospital equivalent, without the lab-results delay (point-of-care).
TPA_M
IXING_MSU
Assumed that same time is needed as in hospital.
MSU_P
ATIENT_T
O_H
OSPITALAssumed that patients are treated in MSU.
Assumed that MSU can be moving during treatment of patient.
Assumed that there is no time pressure. Travel times equal to average Google maps travel times. Normal distribution with mean 23.4 minutes, and standard deviation 8.2 minutes, Appendix B.
P
REPARE_MSU
Assumed triangle distribution with: minimum 15 minutes, maximum 30 minutes, likely 20 minutes.
XIV | P a g e
Appendix D.
Coded Model
XV | P a g e
Figure C1 shows the complete coded model as it has been build in Plant
SimulationTM10.1 (Siemens PLM Software, 2011), the top part (see Figure C2) is the same
as the model by Lahr et al. ( 2013). The bottom part has been added to simulate the MSU organizational model.
Figure C2: Part of model similar to model by Lahr et al. ( 2013)
First, the part of the model were the patients are sent to either the MSU model or the traditional model is explained, Figure C5. In contrast to the model by Lahr et al. ( 2013), the assumption that patients do not interact with each other probably will not hold because of the longer cycle times of the MSU compared to the bottleneck in the
traditional model. Therefore, patients are generated at the source Patient_Call with a negative exponential distributed inter-arrival time. Another change with respect to the model by Lahr et al. ( 2013) is the creation of non-stroke patients, these are needed to simulate the false-positive recognized patients by EMS dispatchers.
XVI | P a g e Figure C3: Method Initialize_Patient_Data Flowchart
Route? Stroke? 1 EntranceHospital 2 EntranceTreatment 3 First_Responder TRUE FALSE 4 Who to Call 1 2 3
Table C1: Method Initialize_Patient_Data Frequency tables
Stroke Frequency Route Frequency
Who to Call Frequency TRUE 67% 1 207/Stroke(Frequency,TRUE) 1 18.57 FALSE 33% 2 7 2 26.79 3 60 3 30 4 6
XVII | P a g e Figure C4: Method First_Responder_Diagnosis Flowchart
Stroke TRUE? MSU TRUE? TRUE True positive or false negative? TRUE MSU available and Patient in Covered area? True Positive TRUE: MSU MSU_to_Patient_Ho me EMS_to_Patient_Ho me FALSE: A1 False Negative A2 B EMS Diagnosis FALSE A1 True negative of false positive? FALSE
False Positive: MSU
EntranceTreatmen
True Negative
Table C2: Method First_Responder_Diagnosis Frequency tables
True Positive
False
Positive Diagnosis Frequency
XVIII | P a g e
Second, some parts of the model simulating the MSU organizational model are explained, Figure C6.
Method Parallel_Activities_MSU:The MSUs with non-stroke patients are directly moved from ParallelActivitiesMSUto MSU_Patient_to_Hospital, MSU with stroke patients go to TPA_Mixing_MSU. Delays from Lahr et al. (2013).
Method MSU_To_Preparing: MSUs are moved to Prepare_MSU, patients are moved to Hospital.
XX | P a g e
Appendix E.
Simulation Model Validation
The simulation model used is an extension to the model developed by Lahr et al. (2013). A big difference between their model and the extended model is the time dependency of the system. Lahr et al. (2013) assumed that there would not be any interaction between different patients because of the very low cycle time compared to the average inter-arrival time. A MSU needs a lot more time per patient, meaning that the assumption of no interaction between patients cannot be made. The arrival rate and distribution are based on data from a previous study on the Groningen model (Lahr et al., 2012). The extended model also incorporates non-stroke patients, these are needed because MSUs will be send to these patients affecting the occupation of the MSUs. No significant difference (t(78)=0.14, p=0.889) is found between outcomes of the extended model (M=21.0%, SD=1.92) and the model by Lahr et al. (2013) (M=21.0%, SD=1.76). Most of the steps are validated by comparing them with similar steps from the original
XXI | P a g e
