Developing a simulation model for decision making in a further digitized Swedish healthcare system

(1)

INOM

EXAMENSARBETE TEKNIK OCH HÄLSA, AVANCERAD NIVÅ, 30 HP

STOCKHOLM SVERIGE 2018,

Developing a simulation model for decision making in a further

digitized Swedish healthcare system

SVEN DIZDAREVIC

ANTON HÄMÄLÄINEN

(2)

(3)

Developing a simulation model for decision making in a further digitized Swedish healthcare system

Framst¨ allning av en simuleringsmodell f¨ or beslutsfattande i ett vidare digitaliserat svenskt sjukv˚ ardssystem

Sven Dizdarevic Anton Hämäläinen

Degree Project in Technology and Health Advanced level (second cycle), 30 credits Supervisor: Jayanth Raghothama Examiner: Sebastiaan Meijer TRITA-CBH-GRU-2018:110 School of Engineering Sciences in Chemistry, Biotechnology & Health KTH Royal Institute of Technology Dept Biomedical Engineering & Health Systems H¨alsov¨agen 11C, 141 57 Huddinge, Stockholm www.kth.se/mth

(4)

(5)

Abstract

Owing to computer simulations, healthcare managers-and decision makers are more capable than ever to understand and evaluate the consequences of their decisions. In 2016, the Swedish government put forward ”Vision eHealth 2025”, emphasizing the importance of digitization within the healthcare system. This thesis aimed at studying the current demands regarding eHealth in Sweden, and what simulation architecture is capable of accurately simulating a digitized Swedish healthcare system. An extensive literature study was conducted, followed by an implementation phase, and finally a validation procedure.

It was first concluded that the following three areas of eHealth applications would greatly benefit Swedish healthcare: a fully integrated journal system, systems for care consultations over the internet, and systems for tele-monitoring of chronics and the elderly. The fidelity of a provided first version of a simulation architecture was then examined, and potential areas for improvement were identified. The implementation phase subsequently included changes to the following aspects of the provided simulation platform: level of generality, the patient agent class (pHome), healthcare resources, illness dynamics & levels of care, workflow chart logic, optimization criteria, user-interface, and output variables. The validation procedure consisted of four interviews with professionals knowledgeable about the Swedish healthcare system, for which the developed simulation architecture was demonstrated. It was concluded that, while the level of detail required for a simulation platform to accurately model the consequences of decision making in a digitized Swedish healthcare system is not known, the developed simulation platform is currently not satisfactory. Above all, it lacks specificity in the output variables.

(6)

(7)

Sammanfattning

Tack vare datorsimuleringar kan beslutsfattare inom sjukv˚arden bättre än n˚agonsin först˚a och utvärdera konsekvenserna av sina beslut. ˚Ar 2016 publicerade den svenska regeringen ”Vision eHälsa 2025”. Dokumentet betonade vikten av dig- italisering inom sjukv˚arden. Det här examensarbetet syftar p˚a att studera de aktuella krav som finns inom eHälsa i Sverige, och även simuleringsarkitekturen nödvändig för att sanningsenligt simulera ett digitaliserat svenskt sjukv˚ardssystem.

En omfattande litteraturstudie genomf¨ordes, f¨oljt av en implementationsfas, och till sist en valideringsprocess.

Först drogs slutsatsen att följande tre omr˚aden av eHälsa skulle bidra med stor nytta till svensk sjukv˚ard: ett fullständigt integrerat journalsystem, system för v˚ardkonsultationer över internet, och system för teleövervakning av kroniker och äldre patienter. Sedan undersöktes trovärdigheten av en försedd första version av en simuleringsarkitektur, och dess potentiella omr˚aden för förbättringar identifierades. Den p˚aföljande implementationsfasen inkluder- ade modifieringar till följande aspekter av den försedda simuleringsplattformen:

generalitetsniv˚a, patientagentklassen (pHome), sjukv˚ardsresurser, sjukdomsdy- namik & olika niv˚aer av v˚ardleverans, logiken bakom arbetsflödesdiagrammen, optimiseringskriterier, användargränssnittet, och outputvariabler. Valideringspro- cessen bestod av fyra intervjuer med professionela yrkesmän kunniga om det svenska sjukv˚ardssystemet. Simuleringsplattformen demonstrerades under alla fyra intervjuer. Slutsatsen drogs att detaljniv˚an hos den utvecklade simuleringsplattformen inte är tillräcklig i nuläget för att utföra korrekta simuleringar.

Framför allt saknar outputvariablerna specificitet. Det är fortfarande inte känt vilken detaljniv˚a som skulle vara tillräcklig i det här avseendet.

(8)

(9)

List of Figures & Tables

All figures and tables used in the report are listed here in the order they appear in the text.

Figures

1. Hierarchy of healthcare facilities. p.19.

2. The state chart for the patient class (pHome). p.20.

3. The old state chart for the patient class (pHome). p.21.

4. The state chart for the nurse resource unit. p.23.

5. The tele-monitoring alarm system workflow chart. p.25.

6. The telephone consultation workflow chart. p.26.

7. The video consultation workflow chart. p.27.

8. The nurse home visit workflow chart. p.28.

9. The HC visit workflow chart. p.28.

10. The hospital visit workflow chart. p.29.

11. The ambulance dispatch workflow chart. p.30.

12. The patient emergency alarm workflow chart. p.30.

13. A typical sequence-chain for a patient undergoing a series of interventions p.33.

14. User-interface. Selection of healthcare resource quantities. p.34.

15. User interface. Selection of the distribution of nurses. p.35.

16. User-interface. Selection of the amount of patients in each municipality.

p.36.

17. Example of an output graph generated by the simulation platform. p.37.

18. The Likert scale used in the validation inquiry sheet. p.38.

19. Inquiry sheet used for validation of the simulation architecture. p.58.

Tables

1. An example of disease categories and their corresponding doctor specialties.

p.24.

2. 14 key areas where digitization could bring great benefits to Swedish healthcare. p.52.

(13)

(14)

List of Acronyms

All acronyms used in the report are listed here, alphabetically.

• ABS - Agent-Based Simulation

• DES - Discrete-Event Simulation

• GIS - Geographic Information System

• HC - Healthcare Center

• ICT - Information- and Communication Technology

• NDC - Nurse Dispatch Center

• OSP - Original Simulation Platform

• SD - System Dynamics

• SKL - Sveriges Kommuner och Landsting (Swedish Association of Local Authorities and Regions)

• SLL - Stockholm L¨ans Landsting (Stockholm County Council)

• STH - School of Technology and Health

(15)

(16)

1 Introduction

Ever since the 1950s, computer simulation modeling has been used for scientific studies of both natural and sociological phenomena in order to expand human knowledge. During the 1980s, the field of computer simulation modeling un- derwent a radical transition due to rapid technological advances in computer hardware and software. It is also evident that the profound changes were ig- nited by the increasing interest of computer simulation modeling in the world of academia and the elevated expectations of simulation modelers and end-users [1].

Today, simulation modeling has developed into a third way of conducting scientific research, beyond that of theorizing and experimenting. Simulations create representations of systems we want to investigate. More importantly, they allow us to manipulate the systems’ critical variables and parameters in virtual worlds without having to invest the resources or face the potential dangers of doing it in reality [2]. Hence, simulations facilitate learning about reality.

Simulations have raised significant potential to better comprehend complex systems [3]. Complex systems are characterized by consisting of a large amount of components that interact to produce results that cannot be predicted by analyzing each of the components individually. Such systems can be represented by complex networks where nodes constitute the systems’ components, and links constitute their interactions [4].

The healthcare system can be seen as a complex network of a wide range of components: patients, clinicians, hospitals, care centers, the patients’ homes, rehabilitation units, laboratories and so on. All of these components interact with each other non-linearly, meaning that small changes in one part of the system can result in large changes in another part of the system, or vice versa. As a result, the components’ interactions tend to produce unintended consequences such as re-hospitalization of patients, adverse drug reactions and bottlenecks in the patient flow [5].

1.1 Simulations of the Healthcare System

The healthcare system has ever been a popular area for the application of computer simulations, even in the very early days of the discipline. No other modeling approach have proven to be equally capable of capturing the complexity, variability, and uncertainty that is so unique for healthcare [6]. Simulations have been used for modeling healthcare facilities, for achieving resource optimization, and for modeling of patient flows to establish efficient patient throughput in hospitals and subsequently produce low waiting times. It is evident that simulations show great potential for supporting operational decision making in healthcare [7]. Owing to computer simulations, in conjunction with today’s rapid technological advances in computer hardware and software, healthcare managers and decision makers are more capable than ever to understand and evaluate the consequences of their decisions. Most importantly, they can do so without using the healthcare system’s valuable resources or causing unnecessary risk to patients [8].

(17)

1.2 Digitization of Swedish Healthcare

During the last decade, the notion of a completely systematized and coordi- nated digitization of the Swedish healthcare system has become increasingly widespread. There are several reasons for this. Not only is the Swedish society already permeated with a solid and functioning digital infrastructure, but the inhabitants are greatly accustomed to digital technology and show great innovative and entrepreneurial spirit [9].

In 2016, the Swedish government released an official plan detailing its strategy to make Sweden a global pioneer regarding the use of digitization and eHealth for facilitating healthcare services as well as welfare services by the year of 2025 [10]. The document lists three key areas where efforts have to be made in order to achieve this goal: regulatory frameworks, coherent usage of notions and concepts across all actors within the healthcare system, and standardization procedures. What is not mentioned, however, is a prioritization plan for the implementation of certain digital technologies into the Swedish healthcare that would benefit the system and further establish digitization.

1.3 Purpose and goals of the thesis

The complex nature of the healthcare system makes it hard for healthcare managers and decision makers to predict how extensive implementation of digital technologies would affect the efficiency, quality and safety of healthcare. The potential of using computer simulations for obtaining such knowledge has already been stated. The purpose of the study outlined in this report is to design a computer simulation architecture with the potential to facilitate decision- making within the Swedish healthcare organizations. This will be achieved by simulating flows of resources and services in a further digitized healthcare system. An already existing foundation for a simulation architecture will be used for this project. The foundation was developed by two students and serves as a simulation platform capable of performing simple simulations in the context of healthcare delivery in a digitized healthcare system [11]. This project aims to improve the usability of the platform in the purpose of answering the research questions stated below.

Consequently, there are two main goals the study aims for: 1) to know what needs to be simulated by finding out which digital technologies there currently are large demands for in Swedish healthcare, and 2) conclude what level of detail is required for a simulation architecture to be capable of producing such simulations. The following two research questions were formulated prior to this study and are answered in this report:

• RQ1: What are the current implementation demands regarding digitization in Swedish healthcare?

In order to make this study useful for Swedish healthcare, it is necessary to know what demands and desires there currently are regarding the implementation of digital technologies. This will ensure that the simulations will be useful for decision- and decision makers responsible for the Swedish healthcare system.

• RQ2: What level of detail is required of a simulation architecture to accurately simulate these demands?

(18)

It is necessary to know what level of detail is sufficient for accurately simulating the demands in Sweden regarding further digitization of the healthcare system. Obtaining this knowledge is crucial for achieving simulations that provide trustworthy information as to how various decisions will affect the logistical flows of the healthcare system.

1.3.1 Structure of thesis

The structure of this master thesis is as follows. The thesis began with a literature study for the purpose of identifying the requirements for a simulation model to address the demands of a digitized Swedish healthcare system. The thesis then proceeded with examining the already developed foundation architecture developed by two students during the Autumn semester of 2017. The required improvements necessary for conducting the desired research of the thesis were identified, and an extensive implementation process of these changes was subsequently initialized. Finally, the new version of the simulation architecture needed to be validated by professionals with expertise relevant for the research area. This report covers each part in successive order.

1.3.2 Limitations

The simulations will make use of simplifying assumptions regarding patient flows, some structuring of certain parts of the healthcare system, financing issues, failure rates of personnel e.t.c. This is due to limitations both in the amount of data that can be aquired and the amount of parameters that can be addressed in the short span time span of a master thesis.

(19)

2 Method

This section outlines the steps taken in order to answer the research questions in a successful manner. Both methods and methodology will be discussed.

2.1 Literature Study

An extensive literature study was conducted for several purposes:

• To obtain knowledge about the Swedish healthcare system; the way it is organized and financed, the way healthcare is delivered and distributed across the country, the way the different actors within the healthcare system interact with each other, e.t.c.

• To obtain knowledge about the digitization of healthcare systems: how is the healthcare system digitized and in what way does it benefit clinicians and patients?

• To know more about what strategies Sweden has established to implement eHealth applications into its healthcare system. What are the goals for Swedish healthcare regarding eHealth the upcoming years?

• To clarify what particular eHealth applications should be prioritized in Swedish healthcare as the situation is today, and hence know what type of implementations are relevant for this project.

It was deemed early on during the project that status reports and reviews of the Swedish healthcare system from SLL (Stockholm County Council) and SKL (Swedish Association of Local Authorities and Regions) were desired. If such sources proved hard to find, second-hand sources should be sought instead. Ap- propriate sources for the research were found by using Google’s search engine as well as the IEEE Xplore Digital Library, NCBI, and Google Scholar databases.

The research method can be compared to that of inductive coding. For every report that was deemed relevant enough for the goals of the project, each para- graph was read through and summarized with keywords. Once enough keywords were gathered, more specifically 41, they were used to construct a framework for determining what sections of found reports, articles, reviews, e.t.c were relevant and perhaps even important for the project. The keywords were as follows:

Digitization, Kin, Responsibility, Goals, Common initiative, Equality, Adapta- tion, Monitoring, Efficiency, Sustainable, Aging, Integrity, Availability, Par- ticipation, Service, Contact, Knowledge, Coordination, Home care, Personal health journal, Digital infrastructure, Appstore, Support, Process, Workflow, Communication, Distance care, Fewer visits, Sensors, Prevention, Regulation.

Conceptual use. Standardization, Integrity, Flows, Expansion, Specialization, Welfare. Chronics, Availability, Obsolescence.

In section 3.2.2, five fundamental areas were presented as candidates where improvements can be made in order to achieve successful digitization of healthcare in Sweden. Each of the keywords were categorized under one or more of the five areas in the list. Since the majority of the keywords were categorized as number 2: Digital infrastructure and data security, it was decided on that this area should be the focus of the simulations. It was also believed that the effects of

(20)

digital infrastructure and data security implementations and investments were the most suitable for studying with simulations. In order to clarify what is included in digital infrastructure and data security, 14 areas were used to define the concept (see section 3.2.2 and Appendix A).

2.2 Implementing usability and functionality changes to the simulation architecture

All usability updates and functionality changes listed in this report were implemented with the AnyLogic 7 University simulation software tool.

Once it became clear which demands for eHealth applications exist in Swe- den’s current eHealth agenda, it was examined to which extent the provided simulation architecture was already capable of simulating their effects on the healthcare system’s resource distribution and patient flows. This was mainly done by studying the design and workability/applicability of the platform. The functionality was subsequently documented and then suggestions were brought forward as to where improvements should be made. The suggestions were discussed back and forth with the supervisor, and an agreement was reached on which aspects of the simulation architecture should be focused on for this thesis.

The development of the simulation architecture, i.e. the implementation procedure, has closely followed the agile framework known as Scrum. Throughout implementation, regular meetings of about 15 minutes were held with the supervisor where demands and requirements for the simulation architecture were stated. Any progression made since the previous meeting was also discussed.

These meetings were then followed by time-boxed iterations lasting anywhere from a couple of days to a couple of weeks. In the Scrum framework, these time-boxed iterations are called Sprints. During the Sprints, the changes agreed upon during the meetings with the supervisor were implemented to the simulation platform iteratively and the results were tested after each iteration. The Sprints spanned over a total of two months.

2.3 The validation procedure

Finally, once all necessary implementations had been performed, it was time for a validation procedure. The process went as follows. A number of candidates (four in total) were selected for a small demonstration of the simulation architecture. The candidates were selected for their expertise knowledge in different aspects of the Swedish healthcare system. They were all found in the main building for the School of Technology and Health (STH) at H¨alsov¨agen 11 in Huddinge, Stockholm.

The demonstration of the simulation platform lasted approximately 60 minutes for each candidate. During this time, the following components of the platform were explained: workflows, state charts, simulation output, as well as the user-interface. Afterwards, they were handed an inquiry sheet were they documented their opinions and general thoughts about the simulation platform by answering a series of questions. The questions were both of a more open nature as well as having more restricted answers since some questions used the Likert scale. The inquiry sheet can be found in Appendix B.

Once all demonstrations had been done, feedback was collected and summarized. The scores on each Likert scale were summed up, and their mean values

(21)

were subsequently calculated.

(22)

3 Findings of the literature study

3.1 The Swedish Healthcare System

Swedish healthcare is highly ranked among the world’s healthcare systems regarding the quality and delivery of care. In 2012, the expected life time of residents at birth was one of the highest in the world [12]. The Swedish healthcare system provides more or less free healthcare across all socioeconomic groups.

The system achieves this by using local-government with an active role for the central government [13]. The central government accumulates financial resources through national income taxes and then allocates the finances to regional principals throughout the country.

The Swedish healthcare system, like any system designed for nation-wide healthcare delivery, is vastly complex. It is non-linear, dynamic and unpre- dictable. Its components are as varied as they are numerous: hospitals, clinics, rehabilitation units, nursing homes, patient homes, families, hospital personnel, and the patients themselves [5]. Hence decision-and decision making related to health care always risk producing unintended consequences.

The following sections briefly outline how the Swedish healthcare system is governed and describe the typical sequence of events that occurs when a resident in Sweden decides that he or she needs to seek medical care.

3.1.1 Organization of Swedish Healthcare

Sweden has a parliamentary form of government. The government is active on three independent levels: the national/central government, the 21 county councils and the 290 municipalities. All three levels of government are involved in regulating and managing Swedish healthcare. Furthermore, healthcare is delivered from three levels of care: regional medical care, county medical care, and primary care [13, 12]. Regional care is delivered by a total of seven university hospitals located in six medical regions. These hospitals deal with difficult and complicated injuries and pathological conditions, as well as educating medical personnel. At the county medical care level, approximately 70 county council operated hospitals and six private hospitals are active. These healthcare facilities deal with more common injuries and also provide various forms of specialized care. Finally the primary care level consists of approximately 1100 public and private primary care facilities, henceforth called healthcare centers (HCs). The healthcare centers serve as the foundation of the healthcare system. Whenever a resident feels the need to seek medical care, the local HC will usually be con- tacted first, and act as a gateway to more sophisticated healthcare if needed [12].

3.1.2 Seeking out medical care

Whenever a resident in Sweden feels the need of seeking medical care, the norm is for that person to first visit his or hers local HC, unless the pathological condition is deemed to be critical. At the HC the patient may receive a referral for a visit to a hospital, a physiotherapist, a psychologist, or any kind of specialized healthcare. However, seeking primary care first is not mandatory in many coun- ties, and the patient may instead visit a specialized hospital directly, either a public or a private one. Usually, a patient visits a private clinical establishment

(23)

if the patient desires to avoid waiting times and has no problem paying for the care directly out of the pocket [12].

3.2 eHealth in Sweden

Digitization offers great opportunities for healthcare- and social services. The term eHealth refers to the use of digital technologies in purpose of improving the healthcare system as a whole [10]. The use of modern information-and communication technology (ICT) in healthcare allows individuals to easily get involved in their own care. ICT:s also facilitate the means of communication between patient and care provider, and provide more efficient support systems for workers in the healthcare sector. The benefits are many, and the Swedish government and Sveriges Kommuner och Landsting (SKL) has decided to support the work of utilizing the potential of digital technologies in healthcare by developing a common vision for the future of eHealth in Sweden.

3.2.1 eHealth Agenda ”Vision eHealth 2025”

In March 2016, the Swedish central government and SKL adopted a vision concerning the development of eHealth in Sweden called ”Vision eH¨alsa 2025”

(”Vision eHealth 2025”) [10]. Subsequently, an official plan detailing its strategy to make Sweden a global pioneer regarding the use of digitization and eHealth for facilitating healthcare services as well as welfare services by the year of 2025 was released. The plan is designed around a set of fundamental principles such as equality, efficiency, as well as availability, usability, and patient integrity and safety.

Vision eHealth 2025 states that by the year of 2025, Sweden shall be exem- plary in utilizing the opportunities of digitization and eHealth in order for people to achieve an equal health and well-being as well as developing and strengthen- ing their own resources for increased independence and participation in society.

It has been made clear that the primary focus of the strategy should be on the individual itself and the caregivers. It is pronounced that the work shall include not only the healthcare system, but also social service and certain divisions of dental care as well [10]. An action plan was released in January 2017 as a follow up of the official plan, presenting in detail the necessary prerequisites for a successful implementation of vision eHealth 2025 in Sweden with respect to regulation, coherent usage of concepts and notions across different actors, and standardization [14].

3.2.2 Current eHealth situation in Sweden

There is a high ambition in Sweden to drive, develop and improve healthcare by utilizing eHealth services. Vision eHealth 2025 has a focal point in utilizing ICT for improving patient safety, increase efficiency of heatlhcare and increase patient participation in care. The work towards digitization and eHealth is ongoing over multiple levels in society, and it is critical for county councils and others healthcare providers to be offensive in their introduction and utilization of eHealth services [15].

Currently, the development of eHealth services is taking place at a slow rate [16]. An interview based study conducted by Socialstyrelsen in 2017 involving

(24)

over 20 selected municipalities in Sweden revealed that there are certain issues that are decelerating the development of eHealth services. These included a lack of knowledge regarding eHealth and welfare technology, financial deficiencies, lack of national directives and standardization, and infrastructure problems, all of which become more apparent in smaller municipalities [17].

Despite the fact that the implementation of new eHealth services is not oc- curring at the desired pace of Vision eHealth 2025, several eHealth services exist and are, to some extent, publicly available today. Inera, a Swedish com- pany owned by SKL, is concerned with coordinating and developing common digital solutions for different actors in Sweden, including residents, employees and decision makers [18]. They offer eHealth services for residents such as video consultations of internet, asking questions to nurses, renewing prescriptive med- ication, reading notes from their own medical journal, and get general medical advice from healthcare personnel. Generally, the digital techniques currently available shows great potential to fundamentally change the healthcare system in Sweden.

An extensive analysis published in 2016 identifies five prerequisites (fundamental areas) where improvements can be made in order to achieve successful digitization of the healthcare system in Sweden [9]:

1. National coordination and distribution of responsibilities.

2. Digital infrastructure and data security.

3. Legislation and personal integrity issues.

4. Financing and compensation.

5. Knowledge about digitization and e-health.

Furthermore, 14 areas were presented where digitization could greatly benefit Swedish healthcare with respect to quality- and cost improvements. These areas are presented in a table in Appendix A.

3.2.3 Common denominator for prioritizations in Swedish eHealth There seems to be a common denominator among sources dealing with the digitization of Swedish healthcare and Swedish eHealth that the following three demands should be prioritized by decision- and decision makers in Sweden as the situation is today [9, 13, 16, 19]:

• A fully integrated journal system.

• Systems for care consultations over the internet.

• Tele-monitoring of chronics and the elderly.

3.3 Simulation Modeling in Healthcare

Healthcare has always been a highly relevant and beneficial application area for computer simulations due to its inherent complexity, and recent studies suggests that it will be in the future as well [20]. Due to the capability of computer simulations to effectively model uncertainty and deal with complex relationship between interacting variables, they present a highly convenient approach for modeling healthcare systems [21].

(25)

3.3.1 Importance of Simulation Modeling in Healthcare

In order for any Healthcare system to function and perform effectively, there are certain demands with respect to financing and material usage that has to be met [22]. Today, due to the varying economic capabilities among different countries, there is an evident variation between the quality of care that can be delivered in each nation [23]. Regardless of this, the ever growing demand for medical care places many of today’s healthcare systems in a situation where the offered medical services does not meet all of the associated needs. This raises an important question which addresses the fact that the health of a population is dependent on two factors; 1) the amount of available healthcare resources and 2) the distribution of these. Therefore, in order for decision makers to successfully make critical decisions regarding the population’s access to relevant healthcare services, optimal allocation of available healthcare resources is imperative [24]. As technology progresses, new mobile health technologies will emerge. Utilization of innovative devices and systems at the point-of-care ren- ders the healthcare system even more remote and patient generated [25]. This is an effect of digitization. It alters the way the healthcare system functions, as new demands are put on healthcare delivery. Studying the potential effects of digitization is thus important, and this is an area where simulation modeling can contribute immensely [26].

(26)

4 What needed to be improved?

In order to produce reasonably accurate results when simulating the effects of eHealth applications on the healthcare system, the level of detail in the original simulation platform (OSP) needed to be improved. The following aspects of the architecture were deemed lacking in this regard:

4.1 Generality

It was desirable that the simulation architecture could be applied, in the future, to any geographical area in Sweden of interest to the user. The original platform implemented data and parameters (i.e. locations such as street names, patient homes, as well as regions such as Stockholm county’s municipalities) in such a way that these could not be modified without significantly modifying the simulation program. Hence it was deemed necessary to reconstruct all parts of the architecture that implicated a required change in source code when modifying the inputs. Only parts requiring hard-coding would be accepted and left untouched, with a general thought in mind of utilizing dynamic coding to the greatest extent possible.

4.2 Healthcare facilities

The OSP did not differentiate between different types of healthcare facilities and levels of care. The only healthcare facility included in the original model was hospitals. This facility was not represented by an agent and did not provide any interactive feature with the patients. It was merely represented by a delay function that kept patients in a certain state before returning to their default state.

In order to simulate healthcare delivery in a more accurate way, multiple kinds of healthcare facilities needed to be included in the simulations; different types of hospitals, healthcare centers, emergency wards, and other primary care facilities. Interwoven with this differentiation is a hierarchical structure of Swedish healthcare delivery that needs to be recognized in the simulations as well; regional healthcare, county healthcare, and primary healthcare.

4.3 Patient distribution

The patient distribution in the OSP suffered due to hard-coding. Patients were distributed across the municipalities of Stockholm County in a limited manner for all simulation runs in that their home locations were constant, few in number, and shared over large parts of the patient population. This represented an unlikely patient distribution and prevented the study of any other area in Sweden but Stockholm County. It was desirable to be able to locate patient homes in any area of Sweden and adapt the distribution to whatever scenario you want to study. The same applied for the distribution of healthcare facilities.

4.4 Workflow logic

Healthcare delivery is simulated according to workflow charts in the DES part of the platform, describing the logic of the resource flows and patient flows.

(27)

The workflow charts incorporated in the OSP were few in number and overly simplified in certain aspects. For instance, patients only had the opportunity to seize nurses from the closest healthcare facility when acquiring a nurse. The logic did not take into consideration the size of the queue of other healthcare facilities, and nurse availability at other locations did not have any impact on the choice of resource. This had patients placed in unnecessary large queues.

4.5 The relation between patient illness severity and healthcare interventions

The OSP provided a strict linear relationship between the extent of patient illness severity and the type of healthcare intervention provided to the patient.

A high illness severity could only lead to a drastic healthcare intervention, and low values only resulted in smaller interventions. Consequently, the patients pathways through the workflow charts in the DES part of the simulation model were predetermined prior to the patients entering them. Different patients with approximately equal illness severity would always move through the same chain of healthcare intervention in a highly predictable way. Thus, the outcome of each workflow and the underlying path through it would always be decided beforehand, and the overall outcome of the patient would be the same for every single patient with the same illness severity.

4.6 Patient healthcare demands

The OSP did not take into consideration the general fact that patients require different forms of healthcare interventions, depending on what type of illness they possess. In reality, the type of care that a patient requires is dependent on one more factor beyond that of the severity of the disease: the type of disease.

Different patients usually suffer from different diseases and thus require different types of healthcare interventions. Overall, the original model did not allow for a dynamical interaction between the patients and the healthcare providers by taking into consideration the specific needs that each patient might have with respect to interventions. Therefore, functionality that could make the model capable of supporting such scenarios was investigated.

4.7 Resource unit home locations

Multiple improvements could be made to the way the OSP handled resource pools. The architecture did not define home locations for all resources. Even though the architecture, for instance, included doctors as an available resource for the healthcare system, their locations were not defined on the map. This made it impossible for the model to keep track of how many doctors were available at each healthcare facility at any given time during the simulations.

4.8 Optimization criteria

By implementing multiple-criteria decision analysis into the logic responsible for the decision making in scenarios where patients seize different types of resources could greatly benefit the OSP. In its original condition, the decision making in the platform was constituted by simple and often single conditions when

(28)

different types of decisions had to me made. Such decisions included patients selecting a doctor for a video consultation, or patients receiving a nurse for a home visit intervention. By carefully considering and incorporating multiple, suitable conditions for evaluation as a base for decision making, the interaction between the patients and the healthcare delivery can be made to more closely resemble the true nature of the system being modeled. As this would allow for a better representation of reality, it was desired to investigate and implement an optimization criteria for each healthcare intervention.

(29)

5 Usability improvements and implementations

The simulation platform has undergone major changes as a result of a period of extensive implementation. Based on the literature study and a thorough analysis of the OSP, necessary implementations have been performed. These implementations serve the purpose of directing the usability of the simulation platform towards the scope of this thesis - developing a simulation platform capable of simulating the Swedish healthcare system in a further digitized society, with a primary focus on internet-based care consultations and tele-monitoring of elderly and chronically ill patients.

The following sections outline the implemented changes to the simulation architecture.

5.1 Generalization

The locations of patients and healthcare facilities are no longer hard-coded in the model, but are instead accessed via web queries (see section 5.3.2). As a result, the simulation platform is no longer restricted to the Stockholm County area but can be applied to any geographical area in Sweden.

5.2 Healthcare facilities

A number of significant changes have been made to how the simulation architecture handles healthcare facilities and how patients interact with them. The different aspects of these changes are explained below.

5.2.1 Hierarchy of healthcare facilities

A healthcare facility hierarchy has been implemented to the simulation architecture in order for the simulation of healthcare delivery to more adequately represent real Swedish healthcare delivery. The following diagram displays the classes of healthcare facilities used in the simulation platform and the relationship between them.

(30)

Figure 1: Hierarchy of healthcare facilities.

The superclass Healthcare F acility has two subclasses; Hospital and P rimary Care F acility. Two more classes: County Hospital and Regional Hospital further inherit from the Hospital class. Finally, the P rimary Care F acility class has two subclasses; Healthcare Center and Other P rimary Care. To- gether these classes comprise the three distinct levels of care in Sweden; regional care, county medical care, and primary care. Visits to a hospital will, for instance, have a larger effect on the illness status of a patient than any visit to a primary care facility, since hospitals are able to provide a higher level of care.

Each distinct type of healthcare facility is represented by its own icon on the model’s geographic information system (GIS) map.

5.2.2 Hospital specializations

In addition, all hospitals have been given a Speciality parameter (String). The thought process behind this is discussed in section 4, and the role the parameter plays is explained in more detail in 5.5: Illness dynamics and delivery of care.

Each county hospital is given one specialty and each regional hospital is given two specialties. All possible specialties that hospitals can have are stored in an external text file outside the model. The simulation program accesses the specialties stored there by looping through the text file and extracting the specialties as String objects. The contents of the text file can be modified by the user, and both the number of specialties and the names of the specialties can easily be altered.

(31)

5.3 Patient modifications

A number of significant changes have been made to how the simulation architecture handles patients. These include a completely revamped state chart for the patient class, as well as a complete new approach to how patients are distributed on the model’s GIS map. In addition, the patient class has been given new parameters and variables in order to function together with the changes made in other classes. The following sections explain patients’ new state chart and how they are distributed across the map.

5.3.1 Patients’ new state chart

Figure 2: The state chart for the patient class (pHome).

When a new patient object is created, it is immediately put into the Home- Telemonitoring state. There it stays until its illIntervention parameter increases its value. Then, an alarm is triggered, and the patient is sent to the AlarmTriggered state. AlarmTriggered acts only as a temporary, intermediate, state and the patient is almost instantly transferred to the third and final Inter- vention state. What happens in the Intervention state depends on the patient’s illIntervention value.

(32)

If one observes figure 2 more closely, it is apparent that there is a way for patients to reach the Intervention state directly from HomeTelemonitor- ing. This happens when a patient uses the emergency alarm function of the tele-monitoring system, see section 5.6.8. The U-shaped transition arrow at the bottom of the figure represents the cases when an intervention ensues an intervention; for instance, when a video consultation with a doctor results in a hospital visit for the patient. Once care has been delivered to the patient, it transitions back to its original HomeTelemonitoring state. The arrow that exits the state chart at the top of the figure represents the case when a patient dies.

This occurs when the patient’s illnessSeverity ≥ 7 and illIntervention ≥ 17, after which the patient is removed from the model.

The purpose for changing the state chart of the patients was to simplify it and make it more understandable while at the same time allowing for a more dynamic intervention pattern. For example, it was desired that patients could undergo several interventions before returning to the HomeTelemonitoring state.

This would have the model more accurately representing reality. The inclusion of the hospital visit workflow chart also made it inappropriate for the patients’

state chart to have a hospital state. The following figure displays the patients’

old state chart.

Figure 3: The old state chart for the patient class (pHome).

(33)

5.3.2 Patient distribution

A general way of adding patients to the geographical map used in the simulation architecture has been implemented. The approach takes advantage of the ability of the web server nominatim.openstreetmap to return an address information object when issued a web query containing a set of geographical coordinates. During patient population build-up, each individual patient undergoes the process of receiving a home location on the map. For each patient in the given municipality, a random number generator is used to create a set of random coordinates, latitude and longitude, within that specific municipality.

This set of geographical coordinates is issued automatically to the web server, which returns a text file in XML format containing address information for the nearest street address of that geographical point. The text file is analyzed and the new coordinates are extracted from the text file and given to the patient.

The patient receives this location as its home location. Each newly created patient undergoes this process.

The approach described here provided means for achieving a realistic patient distribution, where the patient locations are no longer restricted to the Stock- holm County area. The fidelity of the model is also improved by the fact that all patients are actually positioned at real roads/streets on the map, and not far out in a forest or in the middle of a lake. The latter was not an issue with the OSP, but the patient locations were, however, restricted to Stockholm County.

5.4 Resource alterations

A number of changes have been made to the resource pools being used in the model. There are still three types of resources available to patients: nurses, ambulances, and doctors. However, nurses and doctors have received new properties and all three resources have been given home locations on the GIS map.

5.4.1 New properties for nurses and doctors

Nurses and doctor agents have each been given a new parameter (boolean):

nurseAvailable and dAvailable respectively. In order for a patient to be able to seize a nurse or a doctor, these new parameters have to be true. nurseAvailable and dAvailable will both be false if the nurse or doctor is already seized by a patient or is in some other way occupied. Whether or not the resource happens to be occupied with something else is determined by a state chart. The transitions between the two states in the chart are triggered either by patients seizing the resource or by an event, which regularly switches the value of the nurseAvailable or dAvailable parameter. The following figure shows the state chart for the nurse resource:

(34)

Figure 4: The state chart for the nurse resource unit.

The state chart for the doctor resource unit is more or less identical.

In addition, doctors have been given a Speciality parameter (String). All possible specialties that doctors can have are stored in an external text file outside the model. Doctors are given the same specialty as the hospital they are located in. This means that the simulation model loops through the text file containing all specialties, gives them to hospitals according to the order in which they are listed, and subsequently assigns the same specialties to the doctors located in each respective hospital. As for the doctors located at HCs (from now on refered to as ”general doctors”), they do not have any specialties but instead possess more general knowledge. The general doctors act as a bridge between the patients and the specialized doctors at hospitals, see section 5.6.

5.4.2 Home locations

A home location has been added for each of the available resources in the model:

nurses, ambulances, and doctors. Nurses can be found at the following healthcare facilities: nurse dispatch centers (NDCs), hospitals, and HCs. Specialized doctors can be found in hospitals and general doctors are located at HCs. Am- bulances are positioned in hospitals only. As a result of these changes, the model is able to keep track of all resource quantities at the various healthcare facilities and thus have patients seize the specific resource unit most suited for delivering care, based on the patient’s disease, illness severity, and location (see section 5.6 for a description of the so-called ”optimization criteria”).

5.5 Illness dynamics and delivery of care

A patient’s illness status is still determined by the two parameters: illnessSeverity and illIntervention. However, patients can now be inflicted by five different categories of diseases. These categories are, just like hospital specialties, accessed by the simulation program from an external text file. The number of categories is thus arbitrary and can be modified by the user by changing the text file’s contents. More importantly, a certain type of disease can only be

(35)

treated by doctors with the correct specialty. If such doctors do not exist, the patient can only receive minor care from general doctors or nurses. The following table shows an example of a relationship structure between patient disease categories and doctor specialties, and is included here to further clarify how the model works with respect to the above.

Disease category

Doctor specialty

1 A, B, or E

2 B or D

3 D or F

4 C

5 A or F

Table 1: An example of disease categories and their corresponding doctor specialties.

5.5.1 Different levels of care

Different levels of healthcare is delivered to patients in the simulation model.

Higher levels of care will have a larger impact on the patient’s illnessSeverity and illIntervention values, i.e. reduce them to a larger extent than low level care. The different forms of healthcare interventions in the model are ranked in the following order, starting with the lowest level of care:

1. Telephone consultation (with nurse).

2. Video consultation (with doctor).

3. Nurse home visit.

4. Healthcare center visit.

5. Voluntary hospital visit (by taxi for example).

6. Forced hospital visit (by ambulance).

5.6 DES workflow chart logic

Beyond the implementation of the healthcare facility hierarchy, the manner of which delivery of care is simulated has been altered in another significant way;

each patient is assumed to be monitored through a tele-monitoring alarm system. As a result, new logic has been created for the DES-part of the simulation architecture. All workflows present in the OSP have been modified to some extent, and a couple of new workflows have been added due to additional forms of interventions being included in the model; hospital visits and healthcare center visits. The old workflow charts which have been modified include: nurse home visit, video consultations, telephone consultations, alarm triggering, and signal- ing an emergency situation. The following sections will discuss each workflow chart in more detail.

(36)

5.6.1 The tele-monitoring alarm system workflow

Figure 5: The tele-monitoring alarm system workflow chart.

The alarm system workflow chart represents the tele-monitoring system in use for hospital personnel to track patients’ vital parameters. If sudden changes occur in these values, an alarm will be triggered, notifying nurses and/or doctors at a nearby hospital. The changes in the patients’ vital parameters are in the model represented by changes in the patients’ illnessSeverity and illIntervention parameters. It is important to note that interventions only occur after an alarm has been turned on, meaning that an intervention never occurs without an alarm being triggered. The only exception to this is when an intervention results in an ensuing intervention which, for instance, could be the outcome of a telephone consultation, or a video consultation for that matter.

The selection of the outcome of the alarm workflow is dictated by a set of probabilities. The probability values will depend on the value of the patient’s illIntervention parameter. If the value is relatively low, the probability for the patient to be sent home will be relatively high. In the same manner, if illIntervention is relatively high, the probability for being sent to a hospital will be relatively high as well. The probabilities are thus weighted for each individual patient undergoing the intervention in a way that increases the likelihood for an ensuing low level intervention if illIntervention is low, and vice versa. This approach for selecting the outcome of certain interventions is applied in other workflow charts too.

Observe once more figure 5. Whenever a patient enters the workflow, a decision is made regarding what intervention should ensue. The decision process has two steps; 1) check the patient’s illIntervention value and weight the probability for each intervention based on that value, and 2) select an outcome based on

Developing a simulation model for decision making in a further digitized Swedish healthcare system

Developing a simulation model for decision making in a further

digitized Swedish healthcare system

SVEN DIZDAREVIC

ANTON HÄMÄLÄINEN

Developing a simulation model for decision making in a further digitized Swedish healthcare system

Framst¨ allning av en simuleringsmodell f¨ or beslutsfattande i ett vidare digitaliserat svenskt sjukv˚ ardssystem

Abstract

Sammanfattning

Contents

List of Figures & Tables

List of Acronyms

1 Introduction

2 Method

3 Findings of the literature study

4 What needed to be improved?

5 Usability improvements and implementations