A modeling framework for hyperacute stroke care simulation with endovascular treatment

(1)

A modeling framework for hyperacute stroke care simulation

with endovascular treatment

By Friso Postema University of Groningen

Faculty of Economics & Business Administration MSc Supply Chain Management

June 2019

Burgemeester van Walsemlaan 10 8162 GG

Epe 0615150992

F.H.Postema@student.rug.nl

(2)

2

ABSTRACT

Purpose: This research aims to accelerate hyperacute stroke pathway simulation by exploring the role of domain specific modeling frameworks in conceptually modeling hyperacute stroke pathways including endovascular therapy, delivering an applicable framework by updating an outdated framework excluding endovascular therapy, and thereby learning lessons on framework reuse and updating in the process.

Methodology: The framework was developed by reviewing literature on hyperacute stroke pathway simulation/organization and was further supplemented by semi-structured interviews with relevant domain experts. Additionally, the framework to be updated served as a source of inspiration and provided framework steps and components. Framework evaluation was achieved by applying the updated framework to a case and evaluating it according to predefined criteria. Lessons learned were derived from observations during these processes. Findings: Domain specific modeling frameworks have a role to play in simulating hyperacute stroke pathways, as this research proves that speed and therefor cost advantages were achieved in applying the developed framework. Furthermore, framework updating appears to be influenced by issues related to trust, familiarization, newness of changes, complexity of changes and the framework components these issues focus on.

Contributions: An applicable framework for conceptually modeling hyperacute stroke pathways with endovascular therapy was delivered. Insights in challenges and pitfalls related to framework updating were illustrated. The beneficial role of domain specific modeling frameworks for hyperacute stroke pathway simulation with endovascular therapy was confirmed.

Limitations/Future research: The framework was applied to only one case with potential developer bias, so more evaluation is needed. Future research should aim to update a more diverse range of frameworks, in multiple domains and in varying updating circumstances.

(3)

3

PREFACE

(4)

4

1. INTRODUCTION

In 2016 some 56.9 million deaths were recorded worldwide of which nearly 6 million were stroke related, making it the second most prevalent cause of death in the world (WHO, 2018). Besides death, disabilities caused by strokes are a major issue and together they account for about 116 million years of healthy life lost worldwide each year (WSO, 2018). The economic impact of strokes is also large. For example, just the USA spends an estimate of $34 billion on stroke related services, materials and missed work days each year (CDC, 2018). Acute stroke treatment can be organized into so called hyperacute stroke pathways (HASPs), detailing care processes from a pre-hospital stroke onset towards an intra hospital admission and initiation of treatment.

Multiple discrete event studies (DES) have made an attempt to improve HASPs by putting various interventions in a pathway set-up and operation to the test, thereby focusing on the whole process or elements of the process (Churilov & Donnan, 2012; Stahl, Furie, Gleason, & Gazelle, 2003). Unfortunately, most DES studies simulating HASPs are one-time exercises and as such were developed from ‘scratch’ each time, i.e., each new simulation study requires the modeler and stakeholders to familiarize themselves with stroke system set-up and operation, possible interventions, and relevant performance criteria. Hence, simulation efficiency and effectiveness may be compromised.

The many efforts in simulating HASPs already undertaken, together with market potential, i.e., the large number of stroke care systems present worldwide, makes conceptual model reuse a relevant issue, especially since is the first step of all DES studies. Where code reuse faces significant hurdles due to its limited applicability given code specifics, possibilities for re-using elements of simulation conceptual models (CMs) are promising due to their general applicability. In recognition of this fact various so-called modeling frameworks have been presented. They provide a step by step guide to modelers that includes the models purpose, which components to include, the variables and parameters associated with the components and the relationships between these components. (Robinson, Brooks, Kotiadis, & van der Zee, 2010).

(7)

7 enhance conceptual modeling efficiency and speed at a lower cost. These frameworks achieve this by exploiting conceptual modeling similarities and the reuse of model elements which are detailed in the domain specific modeling framework. A framework for modeling HASPs was proposed by Monks et al. (2017). However due to recent advances in the treatment of ischemic stroke by means of endovascular thrombectomy (EVT) and accompanying new organizational paradigms coming in the form of new alternatives in HASP set-ups (Ciccone, Berge, & Fischer, 2019; Détraz, Ernst, & Bourcier, 2018), this framework is in need of an extension.

The aim of this research is to develop a DSMF that can be used by researchers and practitioners studying HASPs with EVT by means of DES studies. This framework will extend the framework of Monks et al. (2017) by increasing its coverage to include a new stroke treatment, namely endovascular treatment. The creation of this extended framework will allow for a more speedily and less costly development of new simulation models for HASPs with EVT. Developing this extension will allow researchers that want to initiate DES studies for HASPs to simulate specific hyperacute stroke systems without the need to reinvent the wheel. Which is enabled by providing an updated set predefined set of model components, covering all recent developments in stroke treatment.

This study aims to make contributions to research in several ways. First of all, by exploring how DSMFs and DES studies might have a role to play in modeling HASPs with EVT. Secondly, by providing a domain specific modeling framework for the simulation of hyperacute stroke systems with endovascular treatment. Third, it will provide insights into the challenges and pitfalls of DSMF updating. Currently, there is little experience in this area due to the small amount of domain specific frameworks that have been developed, and those that have been developed are rarely updated. This research will furthermore validate the extended framework by reviewing literature and by applying it to a real case.

(8)

8

2. LITERATURE REVIEW

In this literature review several topics are discussed which are relevant for the development of a DSMF for HASPs including EVT treatment. First of all, the organization of stroke care and factors influencing the way it is organized are discussed in section 2.1. Next, in section 2.2, several models for organizing regional stroke care are considered. In section 2.3, the role of the simulation and the modeling of HASPs will be highlighted. Hereafter, section 2.4 discusses model reuse in general. Finally, in section 2.5 a summary of findings is presented.

2.1 Organizing Stroke Care

2.1.1 Stroke types.

In order to successfully treat strokes a few distinctions must be made in advance that have implications for the organization of stroke treatment. The first of these distinctions relates to the type of stroke that respective patients are suffering from. Strokes can be categorized into two groups. According to Mackay & Mensah (2004) an ischemic stroke occurs when a blood vessel is blocked, which causes the brain tissue beyond the blockage to die off due to a lack of nutrients and oxygen reaching it, in addition to insufficient drainage of carbon dioxide and waste products produced by the cells. The other type of stroke is called an hemorrhagic stroke which is caused by the rupture of a blood vessel within the brain, which then starts to leak blood, thereby increasing the intercranial pressure to such an extent that brain tissues starts to die (Mackay & Mensah, 2004). The occurrence of both types of strokes is 85% for ischemic strokes against a 15% occurrence of hemorrhagic strokes. Since both types of strokes have different underlying causes and challenges, their treatment and the way they are organized also differ. Since this research will focus on ischemic stroke, the following sections will discuss in more detail how ischemic stroke care is organized.

2.1.2 Ischemic stroke treatment options.

(9)

9 blood thinning it cannot be given indiscriminately to every patient. Patients that are suffering from hemorrhagic stroke can under no circumstances receive this treatment since it would worsen the bleeding in their brain (Fugate & Rabinstein, 2015). A computed tomography (CT) scan needs to be performed first in order to determine the type of stroke the patient suffers from. This is normally done in a primary stroke center (PSC) or if the patient lives nearer to a comprehensive stroke center (CSC) it is performed there. When a hemorrhagic stroke is excluded timely, IVT treatment commences. If the CT scan detects a large vessel occlusion (LVO), which occurs in about 29.3% of ischemic strokes, the patient is eligible for EVT treatment (Berkhemer et al., 2015; Lakomkin et al., 2019).

The EVT treatment aims to remove or break up the blood clot by means of a catheter inserted intravenously through the groin. EVT treatment needs to be performed within 6 hours of stroke onset and as such, patients who are diagnosed with LVO will be rushed to the nearest CSC for treatment (Berkhemer et al., 2015).

2.1.3 Stroke treatment phases.

A further distinction that needs to be made when discussing stroke treatment relates to the acuteness or phase that the patient suffering from a stroke is in. According to Kwan (2007), three phases can be identified which are the hyperacute phase, the acute phase and the rehabilitation phase. The hyperacute phase is the most critical phase since it represents the first 6 hours in which the most treatment options are available and furthermore, what occurs during this timeframe very much so determines the eventual health outcomes for the patient (Saver, 2006). The importance of this phase is further underlined by data from the research of Saver (2006) who found that during an ischemic stroke every minute 1.9 million neurons die, which equates to the brain aging 3.6 years every hour whilst undergoing stroke.

(10)

10 Hyperacute stroke pathways are set-up for a number of important reasons. First of all, they are used to identify patients that are suffering from ischemic stroke. After a positive identification, they are used to initiate thrombolysis treatment of the stroke by means of recombinant tissue plasminogen activator (tPA) within 4.5 hours. Furthermore, if patients are diagnosed within 6 hours with an LVO they will receive additional treatment in the form of EVT. There are, however, some issues in the provision of both types of treatment due to a lack of adequate facilities and human resources at every treatment location (Ciccone et al., 2019). This has consequences for the way regional stroke care is organized, these consequences are discussed in section 2.2.

2.2 Organizational Models of Hyperacute Stroke Pathways

Several organizational models detailing how HASPs for regional acute stroke care could be organized have been proposed. There is however no model that is appropriate for all potential acute stroke situations. Regional differences in healthcare infrastructure are a main cause for the variety of situations occurring. This has led to the adaption of current and proposed models to fit specific regional needs. Together, they offer the largest number of patients the greatest number of stroke treatment possibilities within the timeframe they need to receive them (Ciccone et al., 2019; Détraz et al., 2018).

(11)

11 The organizational models do not necessarily exclude one another and often function side by side to adequately serve as many patients as possible. The MI and MSU are however relatively new and are not yet as common as MS and DS models. Pre-hospital screening techniques might aid in routing and triage decisions in the future, by determining patient stroke type (LVO vs. non-LVO) prior to transportation.

2.3 Hyperacute Stroke Pathway Simulation

This section discusses the simulation of HASPs and the role of conceptual modeling in DES studies. Additionally, a section discusses domain specific modeling frameworks for HASPs. These domain specific frameworks are used to guide modelers through the conceptual modeling phase of DES studies.

2.3.1 Simulation studies.

Due to the complex and time sensitive nature of HASPs in addition to the multiple ways they can be organized there is a real need to determine which organizational models work best for specific situations. This cannot be achieved by simply experimenting in practice due to the high stakes involved for patients in addition to the high costs that go hand in hand with testing in practice. In order to avoid this issue, simulation models can be used, which according to Robinson, Nance, Paul, Pidd, & Taylor (2004) are devices that allow for dynamic experimentation. Thus, researchers turn towards DES studies which are risk free tools to improve upon the logistics of HASPs.

The most important of these simulation studies that focused on HASPs where analyzed by Monks et al. (2017) and although they exhibit a lot of overlap, differences can be spotted in the input and decision variables. Furthermore, the studies examined varying pre and intra hospital processes and their influence on several performance indicators related to stroke treatment (Monks et al., 2017a). These studies give a general idea about which components of HASPs are commonly modeled. None of these studies however includes EVT since it is a recent development and therefor, no simulation studies including EVT were published so far.

(12)

12 models. Generating a CM and an actual model are time consuming and therefor costly processes.

2.3.2 Conceptual modeling.

One of the more important aspects of DES studies is the generation of the CM which is achieved by conceptualizing the model from a real or imaginary system (Robinson, 2008a). The CM details a blueprint of components, variables, parameters and relationships between them and is as such the first step in any DES study. The definition of the CM generated is “a non-software specific description of the computer simulation model (that will be, is or has been developed), describing the objectives, inputs, outputs, content, assumptions and simplifications of the model.” (Robinson, 2008, p. 283). The main reasons to generate a CM are that it decreases the chance that simulations are incorrect, it builds more credibility, can be used as a guiding tool, allows for verification and validation in addition to documentation (Robinson, 2008).

Since DES studies for HASP are all executed in the same domain (acute stroke care) it makes sense to exploit similarities that are encountered in the conceptual modeling within this domain. By doing this, future studies within the domain might require less effort. However, benefits achieved will not spillover to other domains.

To benefit from previous models, elements of these models will need to be reused. This reuse is what makes the process of developing new models quicker and less costly. But it might not always be directly clear what to reuse and how much to reuse. So, to simplify the reuse process, a DSMF might be developed.

By developing a domain specific modeling framework, researchers are guided through the conceptual modeling phase of DES studies. Eventually, the usage of such a framework should help researchers in identifying re-usable model elements and speed up the eventual conceptual modeling process.

2.3.3 Domain specific modeling framework for HASPs.

(13)

13 is of note when doing DES studies is the realization that CMs can often be placed on a spectrum going from more general to more specific. The so-called DSMFs are only applicable to specific domains whereas more general conceptual modeling frameworks can be applied to a wide range of domains e.g. healthcare systems or supply chain systems (Kotiadis, Tako, & Vasilakis, 2014; van der Zee & van der Vorst, 2005). The benefits of these types of framework include an increased modeling efficiency and reuse potential as opposed to more general frameworks. An example of a more general conceptual modeling framework was proposed by Robinson (2008) and is composed of five key activities which can be seen in figure 2.2.

Figure 2.2: Key activities of conceptual modeling

Using this general conceptual modeling framework as a basis, Monks et al. (2017) created a DSMF that was suitable and appropriate for studying HASPs. This framework however does not include the option to model HASPs that include EVT, the purpose of this research is to extend the framework of Monks et al. (2017), so that the newly generated framework can be used to perform DES studies for HASPs that do include EVT.

(14)

14

2.4 Model Reuse

Model reuse leads to some of the modeling benefits that are achieved through the usage of domain specific frameworks as discussed previously. How much and what to reuse is however never instantly clear. Decisions related to model reuse are influenced by a number of factors which will be discussed here. Researchers wanting to use a domain specific modeling framework for HASP DES studies will need to take heed of these factors.

A slightly adapted definition of model reuse from Robinson et al. (2004) states that model reuse is all about selection, isolation, utilization and maintenance of specific model artifacts with the aim of reusing them in the creation of new models. Model reuse, like conceptual modeling is placed on a spectrum. Model reuse can take place on several levels starting from a more complex full model reuse, to component reuse, to function reuse and finally to just simply code scavenging (Robinson et al., 2004).

Some of the benefits that go hand in hand with model reuse have been discussed in previous sections. These benefits are for example a lower cost for developing models and also, they can be developed, adapted or upgraded somewhat faster. Besides benefits, model reuse is inherently accompanied by some disadvantages as well.

Most researchers that create models are not very inclined to follow model building protocols that encourage reuse, since it increases modeling time and cost for them, whilst benefiting mostly others (Robinson et al., 2004). Additional negative aspects of model reuse are the need of having to trust on the work of others. Re-users cannot thoroughly check every single element of the model they would reuse since this would make model reuse more expensive. Eventually, that would defeat the purpose of reusing a model in the first place by making it more expensive than developing something new (Robinson et al., 2004). Clearly, modelers should carefully weigh the benefits and disadvantages in order to make an informed and appropriate decision when initiating a new DES study.

2.5 Summary of Findings

(15)

(16)

16

3. METHODOLOGY

The aim of this research is to develop a domain specific modeling framework for HASPs that include EVT treatment. The development of this framework allows other researchers and practitioners to rapidly model and simulate specific HASPs by providing them with a tool that allows for the quick generation of CMs. This is an example of design science which has the function to develop knowledge or tools that can be used by other researchers to design solutions for problems in a specific field (Van Aken, 2005). The nature of this study is exploratory, since no comparable studies have as yet, studied how to do DES studies for HASPs with EVT using DSMFs.

The remainder of this section will discuss how this framework is going to be developed in terms of data collection and analysis. Additionally, the validation of the framework is then discussed. Next, the methodology will describe how the developed framework is going to be evaluated. It also discusses how to comparatively analyze the old and the updated framework.

3.1 Framework Development

For the development of the framework itself, the modeling framework of Monks et al. (2017) will serve as a guide, since it has been developed to conceptually model HASPs without EVT. Additional data detailing HASPs with EVT will be collected by acquiring literature through the search engine Google Scholar and online databases (Business Source Premier, Emerald Insight, Elsevier, SmartCat, Web of Science).

The article of Monks et al. (2017) provides a good overview of the simulation studies done for HASPs prior to 2016 and these will all be considered before the development of the new framework. These articles give an overview of the focus, decision variables and inputs commonly used. Besides providing this overview, the article by Monks et al. (2017) which itself is an extension/refinement of Robinson (2008), proposes a DSMF for HASPs without EVT and can therefor probably be reused partially in developing the new framework. More recent articles on the simulation of HASPs after 2016 will be searched for using the proposed databases.

(17)

17 will be acquired and analyzed. Several recent literature reviews that scrutinize these models are available, e.g. Ciccone et al. (2019) and Détraz et al. (2018).

When all the relevant data has been collected the data will be categorized into the steps that can be seen in figure 2.1. These five key steps are essential in the development of every modeling framework and are therefore suitable categories to sort the HASP data into. When all data is sorted into these categories it will become possible to determine which system elements and related model components of HASP organization occur repeatedly.

Analyzing the data collected this way will yield new framework elements, which will be compared to those present within the framework of monks et al. (2017). Elements that overlap will be kept whilst the remainder will be removed or slightly modified. System elements that were not present in the previous framework but that are relevant to HASPs with EVT will form the extensions needed to update it. What remains is a core of system elements that can be found in HASP with and without EVT.

3.2 Framework Evaluation

After developing the framework, an evaluation should be performed to determine if it is indeed suitable for modeling HASPs with EVT. The first step of the evaluation process is to present the developed framework to researchers of the UMCG who are participants in the CONTRAST research project. This project is a collaboration between researchers, private and public partners from the Netherlands, with the aim to research new treatments for acute strokes (CONTRAST, 2019). These researchers form a source of additional data and model validation by means of semi-structured interviews. This leads to the correction errors within the framework and to the addition of missing framework elements. Subsequent revisions will eventually lead to a more definitive framework.

(18)

18 framework will probably be useful to model other cases as well. When the framework is not suitable to model the entire selected case, it is probably necessary to further expand the framework. After which the framework arrives at its final from.

(19)

19

4. FRAMEWORK DEVELOPMENT

4.1 Approach

4.1.1 Point of departure.

The basis for developing the new modeling framework consists of the framework for modeling HASPs first proposed by Monks et al. (2017). This framework is a domain specific continuation of the framework proposed by Robinson (2008). Both frameworks will serve as guides and examples in developing the new framework. The domain specific knowledge and steps detailed by Monks et al. (2017) will serve where possible as building blocks to be added to the new framework. Furthermore, they serve as inspiration for the addition of new building blocks.

The extensions that are needed to update the framework of Monks et al. (2017) stem from the differences between IVT and EVT treatment. Most importantly, EVT treatment can only be administered in a small number of hospitals, causing new routing options and logic. Furthermore, it requires additional diagnostics to subcategorize the ischemic stroke population and to determine eligibility. Since both types of treatment are given in sequence and have small time windows in which they are effective (the sooner the better), routing decisions require more and better information. So, the proposed required extensions originating from these differences are displayed in table 4.1.

Table 4.1: Expected extensions to the framework of Monks et al. (2017) Activity Extensions

• Understanding the problem situation

• Study population: categorization due to routing implications related to additional stroke patient types (LVO, non-LVO). • Performance measures: for EVT treatment and performance

measures related to new routings.

• An updated process map: to include EVT treatment and direct vs. indirect routings (CSC only (MS), PSC to CSC (DS)). • Decisions variables: additional areas to be defined linking to

alternative stroke pathway set-ups for EVT. • Setting the

modeling objectives

(20)

20

• Examples of model inputs related to the newly defined decision variable areas are added.

• Constraining factors related treatment timeframes are added. • Determining

model content

• Model components related to EVT treatment and new patient routings.

• Model detail is expanded for the new patient class and new routings.

• New EVT and routing related assumptions.

• Simplifications related to transferal logistics between PSC and CSC.

4.1.2 Sources of information.

Several sources of information will be consulted in order to acquire the information necessary to develop the new framework. Multiple simulation studies studying HASPs without EVT are reviewed in the article of Monks et al. (2017). These articles will be used as a starting point to get familiarized with HASP simulation and will provide information related to stroke treatment and its organization by means of IVT. Unfortunately, no simulation studies have been done for HASPs with EVT. So, to distill the required information, two recent literature reviews and the articles included within them, detailing HASP organization with EVT, are reviewed. To fill in potential gaps and to verify if all necessary elements are included in the framework, the 2018 acute ischemic stroke management guidelines from the national institute of neurological disorders and stroke (NINDS) are consulted. Table 4.2 shows a summary of reviewed literature.

Table 4.2: Literature consulted

Authors Papers reviewed Contents Ciccone, Berge, &

Fischer (2019)

27 HASP EVT organization Détraz, Ernst, &

Bourcier (2018)

11 HASP EVT organization

Monks et al. (2017) 10 HASP modeling focusing on IVT Powers et al. (2018) N/A Acute ischemic stroke management

guidelines

(21)

21 information on the organization underlying a specific HASP. Information gathered through interviewing will be used to supplement the information acquired by reviewing literature.

4.1.3 Framework validation.

(22)

22

4.2 Understanding the Problem Situation

In order to develop a model that correctly reflects a real world situation, it is vital to first understand what the problem situation looks like (Robinson, 2008). Monks et al. (2017) list four areas of knowledge related to a domain specific problem situation that can help to achieve this goal, they are: determining the population to be studied, the creation of a process map, performance measurement of the system and the abstraction of decision variables. In determining the population, researchers decide what the main population to be studied is and potential which subgroups belong to the main group. Performance measures are set to be able to assess the quality of a HASP. The process map forms a starting point for researchers, from which to chart the HASP they are researching. The decisions variables will be guiding in determining which model inputs to use. Together these areas of knowledge form the basis for the later steps of the modeling process. Table 4.3 describes the framework steps related to understanding the problem situation.

Table 4.3: Understanding the problem situation (Adapted from Monks et al. 2017) Area of knowledge Details

• Determining the study population

The main study population consists of two categories which are either confirmed or suspected stroke patients. Confirmed stroke patients can be subclassified according to either ischemic or hemorrhagic stroke types. The ischemic stroke category also consists of an additional subtype, which is the LVO type of ischemic stroke. Suspected stroke patients might be subdivided into eventually confirmed stroke patients which then join the confirmed stroke patient’s category. Further categories are stroke mimics and non-stroke patients. All categories might be considered.

• HASP

performance with EVT

During this step the most important determinants of HASP performance, which are important for the quality assessment of the HASP, are established. Figures for both thrombolysis and thrombectomy are required.

• For thrombolysis: the annual thrombolysis rate, onset-to-door time, onset-to-needle time, door-to-needle time, number of patients arriving within 4.5 h and the percentage of patients with a recorded onset time.

• For thrombectomy:

▪ Direct: onset-to-door time, onset-to needle time, onset-to-groin puncture time and door-to-groin puncture time

(23)

23

▪ Both: The annual thrombectomy rate, the number of patients arriving within 6 h and the percentage of patients with a recorded onset time.

• HASP process map with EVT

For the determination of the model content a process map is a solid starting point. This process map will detail what a HASP looks at present. The map might be altered for experimentation and for the inclusion of case specific elements. Figures 4.1 and 4.2 showcase how IVT and EVT treatment can generally be mapped. The two figures can also be placed side by side to showcase IVT treatment in combination with EVT treatment. When using this process map it is possible to remove, add or zoom in on processes to make it fit the given status-quo. MSU and MI models are not displayed, since they are very rarely utilized, but might be added with subtle changes to the process map.

• Indirect route: PSC to CSC, DS model • Direct route: CSC only, MS model • Explore decision

variables

With the creation of the process map and the determination of the most important performance measures it becomes possible to theorize about potential decision variables. Using both sources of knowledge, the strengths and weaknesses of a HASP might become clear. These are the areas in which it is probably interesting to experiment. Monks et al. (2017) state four areas of decision variables:

• Pre-hospital logistics

• Processes for identification and typifying of stroke patients • Communication between hospital departments

• Workforce scheduling

Two new areas of decision variables are added in this research which deal with the communication between hospitals and the arrangement of transportation between them. This new level of complexity is added due to the introduction of EVT treatment.

(24)

(25)

(26)

26

4.3 Determining the Modeling and General Project Objectives

According to Monks et al. (2017, p. 60): “Modeling objectives clarify the way a simulation study is meant to support client decision making through analyzing various system configurations according to some pre-specified measure of performance.” The modeling objectives furthermore determine the nature of the model that will be developed. Clearly, the modeling objectives play a major guiding role in conceptual modeling. However, in order to be able to come up with proper modeling objectives, it is necessary to state the organizational aims that should be achieved by the development of the model. Once the organizational aims are known, modeling objectives can be conceptualized which should contribute to the full or partial achievement of the organizational aims (Robinson, 2008).

Modeling objectives are stated as belonging to three types of components and are model specific. Achievements are the goals that should be accomplished by the model like an increased throughput. Performance based objectives detail which performance measures should be improved and by how much. Finally, constraints are stated to denote which limitations are present, e.g. budgetary constraints.

Several general modeling objectives also hold, which can be stated for any model. Robinson (2008) names five general modeling objectives which are flexibility, run speed, visual display, ease-of-use and model/component reuse. All of these should be considered but the importance of each of these five elements is mainly case dependent. Table 4.4 shows the organizational aim and objectives related to HASPs with EVT.

Table 4.4: Determining the modeling and general project objectives (Adapted from Monks et al. 2017)

Activity Details • Determine

organizational aim

(27)

27

• Determine modeling objectives

• Achievement

o Health outcomes (primary): the most important outcomes that can be achieved considering HASPs are health outcomes. So, when modeling it is essential that good results are achieved on the following health outcomes.

▪ Survival rates

▪ Functional independence at 90 days (modified Rankin Scale (mRS) score of 0-2)

▪ Functional dependence at 90 days (mRS score of 3-5)

o Logistical outcomes (secondary): this type of outcomes might have a large impact on health outcomes and as such it is vital to achieve good outcomes.

▪ Thrombolysis treatment rate ▪ Thrombectomy treatment rate ▪ Onset-to-treatment time ▪ Minimized flow time ▪ Optimized throughput rates • Performance

o Treatment rates

o Logistical performances: onset-to-needle, onset-to-groin puncture etc.

• Constraints

(28)

28

4.4 Identifying the Model Outputs

Now that the modeling objectives have been determined, the next step is to determine the model inputs and outputs. The model outputs will be described in this section first, since they are somewhat easier to determine. This is due to the fact that in the early stages of conceptual modeling, what people want to know from a model is conceptualized more easily, in contrast to the experiments they might want to perform with them. Modeling outputs related to HASPs can be split into two categories of outputs which are primary health outputs and secondary logistical outputs (Monks et al., 2017). The logistical outputs can be used as parameters for the determination of the health outcomes achieved for the study population. Proportions of the study population treated within predefined time intervals might be linked to established health outcome measures (Lahr et al., 2013). Table 4.5 shows the most commonly used health and logistical outputs.

Table 4.5: Model outputs (Adapted from Monks et al. 2017) Output category Output Details

• Health outputs

o In-hospital death rate Percentage of patients dying before or during treatment

o Survival rate at 90 days Percentage of patients alive after 90 days

o Modified Rankin scale at 90 days

The mRS ranges from 0 to 6, with higher scores indicating greater degree of disability. o Functional independence at 90 days mRS score of 0-2 • Logistical outputs o Onset-to-EMS response time

Total time from onset to EMS response

o Onset-to-door time Total time from onset to hospital entry o Onset-to-IVT

recanalization

Total time from onset to recanalization by means of thrombolysis

o Onset-to-EVT recanalization

Total time from onset to recanalization by means of thrombolysis and

thrombectomy

(29)

29

o Door-to-neurologist evaluation time

Total time from hospital entry to neurological evaluation

o Door-to-needle time Total time from hospital entry to IVT treatment

o Door-to-groin puncture time

Total time from hospital entry to groin puncture

o Groin puncture-to-recanalization time

Total time from groin puncture to recanalization

o Referral call-to-leaving PSC time

Total time from referral to leaving PSC

o Referral call-to-arrival at CSC time

Total time from referral to arriving at CSC

o Arrival at CSC-to-groin puncture time

Total time from arriving at CSC to groin puncture

o Arrival at CSC-to-revascularization time

Total time from arriving at CSC to revascularization

o Onset-to-treatment time IVT

Total time from onset to the administration of IVT o Onset-to-treatment

time EVT

Total time from onset to the administration of EVT o Thrombolysis

treatment rate

Number of patients receiving IVT treatment divided by the total amount of patients

o Thrombectomy treatment rate

(30)

30

4.5 Identifying the Model Inputs

According to Robinson (2008) inputs are the model data that can be altered with the aim of achieving the modeling objectives. In some cases, the inputs can be determined by simply analyzing the modeling objectives and distilling the inputs from them. When this is not the case the inputs to choose can be determined by looking at one’s decision variables. Analyzing these decision variables and identifying which inputs might change them provides a basis for experimentation. Table 4.6 shows the decision variable areas which provide a starting point for experimentation and additionally gives some examples of specific inputs to potentially experiment with.

Table 4.6: Model inputs (Adapted from Monks et al. 2017)

Input category Inputs

• Pre-hospital logistics o EMS stroke referral protocols o PSC/CSC geographical distribution o EMS point of origin

o EMS delays • Processes for identification and

typifying of stroke patients

o Diagnostic equipment availability o Diagnostic staff availability o Time required for diagnostics o Stroke identification protocols • Communication between hospital

departments

o Information sharing o Diagnostics reporting

o Priority arrangement protocols o Multi-disciplinary communication

protocols • Communication between PSCs and

CSCs

o Information sharing protocols o Diagnostics sharing protocols o Pre-notifying protocols • Transfer logistics o EMS availability

o EMS delays

o PSC/CSC geographical distribution • Workforce scheduling o Availability of staff

(31)

31

4.6 Determining the Model Content

The determination of the model content can be split up into two separate activities. These separate activities are the determination of the model scope and the determination of the model level of detail. This makes sense since the model scope denotes the breadth of the model by establishing its components whilst the level of detail says something about the depth of the model by specifying the attributes belonging to these components (Monks et al., 2017). Decisions with regards to these areas are informed by knowledge flowing from assumptions and simplifications

4.6.1 Model scope.

To determine the model scope, an approach which emulates Monks et al. (2017) is chosen, who took as a basis four classes of component types: entities, activities, queues and resources that were defined by Robinson (2008). The entities that are defined represent the units that moves through the system and to which the processes are applied, in this case only potential stroke or confirmed stroke patients. Following the example of Monks et al. (2017) the components have been categorized according to the location of the patient. The split consists of: pre-hospital, intra-hospital, direct route and indirect route. The pre-hospital components are adopted from Monks et al. (2017), except ‘call for help’ and ‘pre-hospital stroke recognition protocol’. Intra hospital component ‘Hospital/stroke team handover’ is a merged form of two prior components proposed by Monks et al. (2017) whilst ‘IVT eligibility assessment’ and ‘IVT treatment’ are directly emulated. The resources have also been provided by Monks et al. (2017) Additional components were added to account for the newly added processes. In this model no queues are specified, but they could essentially be defined for every activity. The most common components encountered in the literature are stated in table 4.7.

Table 4.7: Model scope (Adapted from Monks et al. 2017)

Class Component

Entities Patients

Pre-hospital Call for help Patient reaches out for help Contact GP Contact GP

GP EMS call GP on scene GP life support

(32)

32

EMS contact & dispatch

Contact EMS

EMS ambulance dispatch EMS rapid responder dispatch Self-presentation Transport to hospital

EMS on scene EMS life support EMS diagnostic tests Pre-hospital stroke

recognition protocol

ABC check

Parameter assessment

Cincinnati pre-hospital stroke scale Glasgow coma scale

Glycaemia measurement EMS transport EMS transport to hospital

EMS pre-notification to hospital Intra-hospital Hospital/stroke team

handover

EMS handover to ED ED triage

ED stroke diagnostic test Assignment of ED cubicle Assessment by ED physician Acute stroke team travel to ED

Assessment by member of acute stroke team Laboratory testing Bloodwork

Neurological testing NNIHS score measurement Neurological imaging Booking CT/MRI scanning

CT/MRI scanning In-hospital transportation

Reporting and interpretation of CT/MRI scan Ambulance paramedic takes patient direct to CT/MRI scanner IVT eligibility assessment Laboratory examination Inter-disciplinary consultation Eligibility decision

IVT treatment tPA administration CTA/MRA scan Booking CTA/MRA scan

CTA/MRA scan

Intra hospital transportation Interpretation of CTA/MRA EVT eligibility assessment Diagnostics examination Inter-disciplinary consultation Eligibility decision

Direct route EVT treatment Transportation to operating theater Groin puncture

Stent retrieval Recanalization

Admission Admission to acute stroke unit Indirect route EMS transfer to CSC Contact EMS

EMS ambulance dispatch EMS on scene

EMS transport

Hospital prenotification

CSC handover EMS handover to acute stroke team Transportation to operating theater EVT treatment Groin puncture

Stent retrieval Recanalization

(33)

33

Resources General Practitioners ED Physicians ED Nurses

Acute Stroke Physicians Acute Stroke Team Radiographers Radiologists Ambulances Ambulance paramedic CT scanners MRI scanners CTA scanners Neuro-interventionist 4.6.2 Level of detail.

After the determination of the model scope, which introduces the components that are present within the model, the level of detail of the included components needs to be established. This is achieved by defining attributes that can be linked to specific components, which in turn generates the necessary depth of these components. The level of detail for this framework was largely predefined by Monks et al. (2017) and is already fairly complete. Only a slight adaption was needed for the attribute patient classification, to enable the creation of the LVO ischemic stroke classification. A second somewhat larger adaptation is the inclusion of the hospital type attribute, which allows for the classification of patients into categories according to which type of hospital they arrived at first. The proposed attributes can be found in table 4.8.

Table 4.8: Level of detail (Adapted from Monks et al. 2017) Attribute Example

Patient classification Classification of patients into hemorrhagic stroke, ischemic stroke, LVO ischemic stroke, stroke mimic, and non-stroke classes based on retrospective local or national data.

Patient age band Dichotomy of patients to incorporate variable treatment windows or local hospital policies for different age groups. For example, above or below the age of 80

Contraindication to treatment

(34)

34

Timestamp of stroke onset

Simulated onset time of stroke. Used to assess time remaining. Can also be zero in cases where onset time is not available (e.g., due to wake up strokes).

Stroke severity Dichotomy into mild/moderate or severe strokes based on historical profile of cases. Used to incorporate alternative activity cycle times depending on severity.

Mode of arrival Classification of patients into those that arrived by emergency ambulance services, primary care followed by emergency ambulance, or self-presentation at the ED. Provides routing logic. Hospital type Classification of patients into which type of hospital they arrived at

first. This provides routing logic for the direct CSC only and indirect PSC to CSC routes.

Diagnostic test applied A flag indicating a stroke diagnostic test has been applied. In a model this is used for basic logic. I.e. to test if a new test should be applied. Note there may be multiple diagnostic tests applied and hence require an array of flags.

Diagnostic test result Dichotomy into positive and negative results. Provides routing logic; for example, only patients with positive results would enter a fast track for treatment.

Diagnostic test classification

Classification of diagnostic results into true/false positives/negatives. Each diagnostic test for stroke has a sensitivity (ability to classify a true stroke correctly) and specificity (ability to correctly identify a non-stroke). Important when modeling arrival rates to an acute stroke unit as false positive patients may spend some time on an alternative ward before identification and transfer. Triage classification The level of priority given to a (suspected) stroke patient in the emergency department. Used to either reorder patients in a queue or in the case of a Monte Carlo estimate of queueing time alter the choice of sampling distribution.

4.6.3 Identifying assumptions and simplifications.

(35)

35

Table 4.9: Assumptions & Simplifications (Adapted from Monks et al. 2017) Assumptions

A1. Activity cycle times are independent of the time remaining until the treatment deadline

A2. Patients identified with suspected ischemic stroke arriving within 4.5 h of onset of symptoms have priority on hospital resources and queue jump.

A3. Patients identified with suspected LVO ischemic stroke arriving within 6.0 h of onset of symptoms have priority on CSC hospital resources and queue jump. A4. Independence of patient attributes.

A5. Patients are routed to the hospital type (PSC/CSC) they are located closest to. A6. Patients with known IVT contraindications are always directly routed to CSC.

A7. Patients suffering from a non-LVO ischemic stroke receive IVT treatment when within 4.5 h timeframe.

A8. Patients suffering from LVO ischemic stroke receive both treatments when within 4.5 h timeframe for IVT and 6 h timeframe for EVT.

Simplifications

S1. Non-stroke patients that share resources in the stroke pathway are excluded S2. ED processes are modelled as a series of random variables without queues (ED activities are not capacity constrained, but include queueing time as a delay) S3. Pre-hospital EMS logistics and network management are simplified to a series of random variables without queues (EMS activities are not explicitly capacity constrained and include queue times)

S4. Pre-hospital onset-to-door time is simplified to a single random variable. S5. Severity of stroke is excluded.

S6. Patients who wake up with stroke and hence do not have an onset-to-door time are not modelled in detail.

(36)

36

4.7 Comparative Analysis

This section will compare the framework of Monks et al. (2017) with the framework that was developed in this chapter. The addition of EVT treatment has changed the status quo in several ways. The main differences that were caused by the introduction are:

1. Additional routing options and logic have emerged relating to the new forms of organizing acute stroke care.

2. New diagnostic processes are utilized both pre-hospital and intra-hospital to asses EVT eligibility.

3. New processes related to the actual intra-hospital EVT treatment.

4. New processes related to interhospital transportation and communication. 5. A new ischemic stroke patient category emerged (LVO vs. non-LVO).

6. New performance measures linking to the new routings, diagnostics, EVT treatment and patient categories.

7. The linkage between onset-to-treatment (EVT) time and the mRS score as primary health outcome.

The changes that are denoted above have had an impact on the framework first detailed by Monks et al. (2017). This has led to the following extensions as displayed in table 4.10.

Table 4.10: Extensions to the framework of Monks et al. (2017) Activity Extensions

• Understanding the problem situation

• Study population: the ischemic stroke patient category is split into regular ischemic stroke and LVO ischemic stroke. Leads to redefined attributes and assumptions. Framework users have to take note of an increased number of patient categories and collect data accordingly.

• Performance measures: related to EVT treatment established. Measures related to the MS and DS organizational models defined. Leads to additional model outputs. Framework users need to define which of the new and old measures are relevant to them and include them if they can extract from the problem situation.

(37)

37

either simplify, adapt or expand this process map according to their needs.

• Decision variables: two new areas are added related to transferal logistics and interhospital communication. They stem from the addition of new organization models with different hospital types. Framework users can explore these and previously defined areas to potentially find new experimental variables.

• Setting the modeling objectives

• Logistical outputs: related to routings and EVT treatment. Additional health outputs proposed. These are used to evaluate the performance of new processes and the new treatment/routings. These logistical outputs can be used by framework users to evaluate the performance of the routings and the old/new treatment.

• Model inputs: are provided for the two new decision variable areas. Potential to experiment with the new decision variable areas is shown this way. Framework users might use these for experimentation or as inspiration for the definition of other experimental inputs.

• Determining

model content

• Model components: added for EVT processes and new routings. Accompanying resources are defined. Allows for the required extra model breath. Framework users can pick and chose these to get the model components fitting to the situation they want to model.

• Attributes: one adapted to allow for additional patient categorization. New attribute defined to classify patients according to hospital type arrived at first. Allows for the extra model depth required. Framework users can employ the attributes to further specify the patient characteristics of all types of patient categories.

• Assumptions: related to new patient classifications, routings and EVT treatment added. Guides framework users by illustrating to them how believes and uncertainties related to the new status quo can be made certainties.

(38)

38

5. FRAMEWORK EVALUATION

In this chapter the framework developed previously will be applied to conceptually model a real case in order to illustrate its workings and indicate its added value for performing simulation studies. After applying the framework, model components for which the framework offered no direct support will be discussed in order to determine why the framework was lacking. Important components that are missing will be added to the framework. Components that can be modeled will not be discussed in detail since they illustrate that the framework sufficiently covers them. Next, an evaluation will take place to determine if the supposed time and cost benefits of DSMFs were achieved in conceptually modeling this case. By doing this, an overall value judgment can be made about the entire framework.

5.1 Selected Case

The case setting for putting the framework to the test is based on a study by Holodinsky et al. (2018). Their study does not rely on simulation, but may be characterized as a theoretical conditional probability modeling study. As such, a theoretical HASP is the focus, although based on existing data from clinical trials. The theoretical HASP was projected on the California region of the USA. More particularly, Holodinsky et al. (2018) researched how to triage suspected LVO stroke patients by using three different pre-hospital screening tools with varying probabilities in predictive accuracy in combination with two transport strategies, which are the MS and DS models. The decision for a transport strategy was furthermore dependent on the speed of treatment at PSC and CSC respectively and the patients travel time from both types of hospitals. The probability of a good health outcome (mRS score 0-1) was determined for both the MS and DS models.

(39)

39 as specific input, and as such detailed within the CM, but do not exclude alternative non-deterministic input scenarios. Five logistical input variables are considered: onset-to-door, door-to-needle (PSC), door-to-leaving PSC, door-to-needle (CSC) and door to groin puncture. All of these inputs are of a non-deterministic nature.

By using non-deterministic inputs, a plethora of situations might be simulated, enhancing the explanatory power, accuracy and coverage of the simulation study that might be performed, compared to the three scenarios modeled by Holodinsky et al. (2018). Three ranges of predictive probabilities relating to the pre-hospital stroke protocols were detailed by Holodinsky et al. (2018), and will also be used as input. A summary of the applied framework can be found in table 5.1.

Table 5.1: Summary of applied framework steps for simulating the case presented by Holodinsky et al. (2018)

Framework step Details

Problem situation

Study population • Included are ischemic stroke patients, LVO ischemic stroke patients, hemorrhagic stroke patients and stroke mimics

Performance • Performance in this case is not yet known, due to this being a theoretical study not working with a pre-existing HASP.

Process map • See figures 4.1 & 4.2. No adaptions required.

Decision variables • Pre-hospital screening protocols • Logistical performance measures

Modeling & general objectives

Organizational aim • The aim is to identify the optimal transport and triage strategies.

Model outputs • Health outcomes

o The probability of a mRS score of 0-1 at 90 days.

• Logistical outcomes

o Onset-to-treatment time IVT o Onset-to-treatment time EVT Model inputs Pre-hospital screening inputs:

(40)

40 Logistical inputs: • Onset-to-door time • Door-to-needle time (PSC) • Door-to-leaving PSC time • Door-to-needle time (CSC) • Door-to-groin puncture time

The three predefined scenarios are as follows: o Scenario A

Logistical input Time in minutes Onset-to-door 60 Door-to-needle (PSC) 60 Door-to-leaving PSC 50 Door-to-needle (CSC) 30 Door-to-groin puncture 60 (direct), 30 (indirect) o Scenario B

Logistical input Time in minutes Onset-to-door 60 Door-to-needle (PSC) 60 Door-to-leaving PSC 120 Door-to-needle (CSC) 30 Door-to-groin puncture 60 (direct), 30 (indirect) o Scenario C

Logistical input Time in minutes Onset-to-door 60 Door-to-needle (PSC) 60 Door-to-leaving PSC 120 Door-to-needle (CSC) 60 Door-to-groin puncture 90 (direct), 60 (indirect) 3 Constraints • N/A Model content

Model components See table 5.1

Model attributes L1, L3, L4, L6, L7, L8, L9, L11, see table 5.2 for arguments Assumptions A1, A2, A3, A4, A6, A7, A8

Simplifications S1-S7

5.2 Understanding the Problem Situation

(41)

41 correct predictive power. They might have some value in predicting what routing decision to make in combination with the speed of treatment at both hospital types and the travel time to and between them.

The goal of this simulation study is to determine how patient routing is influenced by the pre-hospital screening protocols with variable probabilities of predictive accuracy, in tandem with variable logistical performances also influencing these routing decisions. Since Holodinsky et al. (2018) chose to work with a simplified HASP process, the current process map could be utilized. However, intra-hospital processes should be simplified into a single time-based logistical variable, since these processes are not considered in detail in this simulation.

The study population is made up out of four categories of patients, i.e., those facing ischemic stroke, LVO ischemic stroke, hemorrhagic stroke and stroke mimics respectively. Furthermore, the authors predefined three deterministic logistical input scenarios, but with varying probabilities in predictive accuracy of the decision variables (pre-hospital screening protocols). Three different pre-hospital screening protocols and accompanying ranges of probabilities were used.

5.3 Determining the Modeling and General Project Objectives

The organizational aim that is to be achieved in the theoretical study is to identify the optimal triage and transport strategies the yield the highest probability of positive health outcomes. The modeling objectives that are to be achieved are the highest possible probability of a mRS score of 0-1 at 90 days. Further modeling objectives are the secondary logistical outputs, for which is the aim to achieved the lowest total onset-to-treatment times for both IVT and EVT treatments.

Since this model will only be used to test a theoretical situation not all general objectives are important. Time is not an issue since there is rush to improve an actual HASP. Flexibility, model reuse, visual display and ease-of-use are the objectives that are useful to ease experimentation. In this case, no specific constraints as specified in table 4.4 are detailed.

(42)

42

5.4 Identifying Model Outputs

The primary model output in this case the probability of a mRS score of 0-1 at 90 days, computed according to specified equations (Holodinsky et al., 2018). This health outcome was measured for both types of transport strategy (MS, DS), under the influence of the predictive accuracy of the pre-hospital protocols, and was mainly based on the onset-to-treatment times achieved. In this case, onset-to-onset-to-treatment times for both IVT and EVT treatment are determined. Both these logistical outputs might be linked to the primary health outcome, the formulas detailing this can be found in Holodinsky et al. (2018). This allows for determining how much effect the speedily administration of both types of treatment has on the probability of a good mRS score.

5.5 Identifying Model Inputs

Three types of pre hospital screening protocols are used in order to determine if a patient is suffering from an LVO. The three protocols researched by Holodinsky et al. (2018, p.1478): “Cincinnati Prehospital Stroke Severity Scale (C-STAT, 3 item scale), the Rapid Arterial Occlusion Evaluation (RACE, 9-point scale) and the Los Angeles Motor Scale (LAMS, 5 point scale).” The probability to correctly predict if a patient has an LVO varies per protocol and per test scores. These probabilities might be used as input parameters to experiment on how to triage and transport.

Additional model inputs are required to experiment with logistical performances. Therefore, the logistical inputs described within the three scenarios might all be varied. This means that the following logistical inputs present a source of experimentation: onset-to-door time, door-to-needle time (MS), door-to-leaving PSC time, door-to-needle time (DS) and door-to-groin puncture time.

5.6 Determining the Model Content

5.6.1 Scope.

(43)

43 including or excluding model elements are presented alongside the decision for in or exclusion.

Table 5.2: Model scope

Component Include/Exclude Argument Entities

Patients Include The only entity flowing through the system Pre-hospital

Call for help Include Triggers the variable: onset-to-door

Contact GP Exclude Study includes only patients arriving by EMS

GP on scene Exclude Study includes only patients arriving by EMS

EMS contact & dispatch Include Necessary to determine response time/ part of experimental variable: onset-to-door Self-presentation Exclude Study includes only patients arriving by

EMS

EMS on scene Include Necessary to determine on scene time/ part of experimental variable: onset-to-door Pre-hospital stroke

recognition protocol

Include Influences routing decision

EMS transport Include Part of experimental variable: onset-to-door Intra-hospital

Hospital/stroke team handover

Include Endpoint of experimental variable: onset-to-door

Triggers the variables: door-to-needle (MS), door-to-needle (DS), door-to-leaving PSC and door-to-groin puncture time

Laboratory testing Include Combines all pre-treatment diagnostic processes to yield one diagnostic related logistical performance

Neurological testing Exclude Diagnostic processes are combined Neurological imaging Exclude Diagnostic processes are combined IVT eligibility assessment Include Provides routing logic

IVT treatment Include Endpoint for the variable: door-to-needle (MS, DS)

CTA/MRA scan Include Provides routing logic EVT eligibility assessment Include Determines routing Direct route

EVT treatment Include Endpoint for the variable: door-to-groin puncture (MS)

Admission Exclude Study only wants to know onset-to-treatment Indirect route

EMS transfer to CSC Include Endpoint for variable: door-to-leaving PSC CSC handover Include Triggers experimental variable

door-to-groin puncture.

EVT treatment Include Endpoint for the variable: door-to-groin puncture (DS)

(44)

44 5.6.2 Level of detail.

Table 5.3 presents the model attributes to be used in the CM and the arguments for in or excluding them.

Table 5.3: Model level of detail

Attribute Include/Exclude Argument

Patient classification Include Provides necessary patient categorization.

Patient age band Exclude Age is not a factor considered by this study.

Contraindication to treatment

Include Provides routing logic.

Timestamp of stroke onset

Include Study assumes there is a known stroke onset time.

Stroke severity Exclude Stroke severity is not considered by this study.

Mode of arrival Exclude No categorization of patients into mode of arrival is needed, as all patients arrive by EMS transport.

Hospital type Include Provides routing logic.

Diagnostic test applied Include Confirms if patients underwent diagnostic testing or not.

Diagnostic test result Include Is needed to route patients according to the intervention required.

Diagnostic test classification

Exclude Unnecessary complication for this study.

Triage classification Include Gives priority to patients nearing their upper treatment time windows.

5.6.3 Identifying assumptions and simplifications.

(45)

45

5.7 Framework Evaluation

5.7.1 CM development.

The application of the framework was for the most part without obstacles, there are however some things of note. The logistical outcomes that have been chosen as model output did not exist in this form prior to application. The logistical outcome onset-to-treatment was split into one for both EVT and IVT. Furthermore, the specific model inputs used were not detailed by the developed framework. These model inputs were specifically derived from the case that was modelled. Most notably the three types of pre-hospital screening protocols that were used. In addition to this, customized logistical performances that fit the simplified status quo were derived. Other than this, the model was sufficiently complete to model the remaining aspects of the case. So, overall the framework seems to be covering all important aspects of HASPs with EVT, only requiring slight adaptions and customizations in order to be applied to a specific case.

5.7.2 Benefits gained from DSMF application.

A modeling framework for hyperacute stroke care simulation with endovascular treatment