SECOND CYCLE, 30 CREDITS STOCKHOLM SWEDEN 2020,
Automated Triage in Digital Primary Care
Assessing the Potential of Using Multi-Criteria Decision-Making Models
ALBIN GRANELL
CHRISTOFER BORÉN
KTH ROYAL INSTITUTE OF TECHNOLOGY
SCHOOL OF INDUSTRIAL ENGINEERING AND MANAGEMENT
Automatiserat Triage i Digital Primärvård
Utvärdering av potentialen att använda Multi-Criteria Decision-Making-modeller
av
Albin Granell Christofer Borén
Examensarbete TRITA-ITM-EX 2020:285 KTH Industriell teknik och management
Industriell ekonomi och organisation
SE-100 44 STOCKHOLM
Automated Triage in Digital Primary Care
Assessing the potential of using Multi-Criteria Decision-Making models
by
Albin Granell Christofer Borén
Master of Science Thesis TRITA-ITM-EX 2020:285 KTH Industrial Engineering and Management
Industrial Management
SE-100 44 STOCKHOLM
Examensarbete TRITA-ITM-EX 2020:285
Automatiserat Triage i Digital Primärvård
Albin Granell Christofer Borén
Godkänt
2020-06-05
Examinator
Jannis Angelis
Handledare
Anna Svarts
Uppdragsgivare Kontaktperson
Sammanfattning
Det ökande underskottet av sjukvårdsresurser gör effektivitetsförbättringar i sjukvårdsbranschen nödvändigt för att säkerställa säker och tillgänglig sjukvård för alla. Digitalisering förväntas fylla en fundamental roll i denna transformation och digitala vårdgivare i primärvården har under de senaste åren växt till en betydande del av den svenska primärvårdssektorn. Flertalet av dessa har byggt lösningar för automatiserat triage, där triagefunktionärens roll ersätts av en automatiserad process där en triagealgoritm direkt hänvisar patienten till den lämpliga vårdnivån.
Trots tillväxten av digitala vårdgivare i primärvården och deras automatiserade triagesystem i
primärvården är forskning kring effekterna av att automatisera triageprocessen i primärvården begränsad.
Denna studie strävar efter att utvärdera potentialen i att använda MCDM-modeller för automatiserat triage i den digitala primärvården genom en casestudie på en av de ledande digitala vårdgivarna i primärvården.
Studien är uppdelad i två delar. I del ett genomförs intervjuer för att kvalitativt fastställa vilka faktorer som bör inkluderas i en automatiserad MCDM-modell för triage. I del två simuleras den resulterande MCDM-modellen för att utvärdera dess resultat jämfört med den traditionella triagemodellen i vilken alla patienter har ett inledande möte med en sjuksköterska.
Studien visar att en automatiserad MCDM-modell för triage kan förbättra kostnadseffektiviteten i termer av lönekostnader och produktivitet i termer av färre konsultationer per patient, jämfört med den
traditionella triagemodellen. Däremot visar den traditionella triagemodellen högre effektivitet i termer av att enbart utnyttja läkarresurser för patienter i absolut behov av läkarvård.
Nyckelord
Triage, Sjukvård, Primärvård, Digital sjukvård, Multi-Criteria Decision-Making
Automated Triage in Digital Primary Care
Albin Granell Christofer Borén
Approved
2020-06-05
Examiner
Jannis Angelis
Supervisor
Anna Svarts
Commissioner Contact person
Abstract
The increasing global deficit of healthcare resources makes efficiency improvements in the healthcare industry a complete necessity to assure safe and available healthcare for everyone. Digitalization is expected to play a fundamental role in this transition and digital primary healthcare providers have in recent years developed into a substantial part of the Swedishprimary care sector. Several of those have built solutions for automated triage, where the role of a triage officer in traditional primary care is replaced by an automated process, in which an triage algorithm directly refers the patient to the appropriate level of care.
Despite the rise of digital healthcare providers and automated primary care triage systems in particular, research on the implications of automating the triage process in primary healthcare is scarce. This study aims to assess the potential of using MCDM models for automated triage in digital primary care, by conducting a single case study at one of the leading digital healthcare providers. The study is separated into two phases. In phase one, interviews are conducted to qualitatively determine what set of factors to include in an automated MCDM triage model.In phase two, the resulting model is simulated to evaluate the performance compared to the traditional triage model in which all patient journeys start with an initial nurse meeting.
The study shows that an automated MCDM triage model can improve cost efficiency in terms of clinician salary costs and productivity in terms of fewer consultations per patient, compared to the traditional triage model. However, the traditional triage model is shown to be more efficient in terms of only utilizing doctor resources for patients in absolute need of doctor care.
Key-words:
Triage, Healthcare, Primary care, Digital healthcare, Multi-Criteria Decision-Making
1 Introduction 1
1.1 Background to Digital Triage . . . 1
1.2 Objective & Research Question . . . 2
1.2.1 Evaluating Efficiency, Productivity and Patient Experience . . . 3
1.3 Delimitations . . . 4
1.4 Expected Contribution . . . 5
1.5 Report Outline . . . 6
2 Triage Literature 7 2.1 Formal Definition of Triage . . . 7
2.2 Previous Research on Triage Models . . . 8
2.2.1 Triage in Emergency Care . . . 8
2.2.2 Triage Judgment Frameworks in Swedish Primary Care . . . 10
3 Empirical Context 11 3.1 The Swedish Primary Care System . . . 11
3.1.1 The Governance System . . . 11
3.1.2 The Act on System of Choice in the Public Sector . . . 12
3.1.3 Public Reimbursement . . . 12
3.1.4 Public Reimbursement of Non-Resident Patients . . . 14
3.1.5 Triage As a Requirement For Reimbursable Digital Healthcare . . . 15
3.2 Guiding Principles in Swedish Healthcare . . . 15
3.2.1 Prioritization and the Ethical Platform . . . 15
3.2.2 Principles of Cost Efficiency . . . 16
3.3 Digital Primary Care Practice . . . 17
3.3.1 Major Players and Triage . . . 18
3.3.2 Criticism . . . 19
4 Theoretical Background 21 4.1 Multi-Criteria Decision-Making . . . 21
4.1.1 MCDM Definition . . . 22
4.1.2 Evaluating and Comparing MCDM Methods . . . 24
4.2 Simulation as a Tool . . . 24
4.3 Performance Evaluation for Predictive Modeling . . . 25
5 Method 27 5.1 Research Design . . . 27
5.2 Phase 1: Interviews . . . 28
5.2.1 Selection of Interviewees . . . 28
5.2.2 Interview Design . . . 29
5.2.3 Data Analysis . . . 31
5.3 Phase 2: Simulation . . . 31
5.3.1 Simulation Strategy . . . 31
5.3.2 Simulation Modules . . . 32
5.3.5 Triage Models . . . 39
5.3.6 Entering the Decision Matrix Values of the MCDM Model . . . 42
5.3.7 Determining the Relative Weights of the MCDM Model . . . 44
5.3.8 Effect of Systematic Errors in the Digital Triage Assessment . . . 45
5.4 Evaluation of Research Method Quality . . . 46
5.4.1 Validity, Reliability and Generalizability . . . 47
5.4.2 7 Criteria of Mixed Method Research Undertaking Quality . . . 48
6 Empirical Findings 50 6.1 Phase 1 - Qualitative Interviews (RQ1) . . . 50
6.1.1 Input Factors Included in Phase 2 Simulations . . . 51
6.1.2 Input Factors Not Included in Phase 2 Simulations . . . 52
6.2 Phase 2 - Simulations Comparing Triage Models (RQ2) . . . 53
6.2.1 Overview of Simulation Results . . . 53
6.2.2 Cost Efficiency - M1 Clinician Salary Costs . . . 54
6.2.3 Cost Efficiency - M2 LEON Enactment . . . 57
6.2.4 Healthcare Productivity - M3 Consultations per Patient . . . 61
6.2.5 Healthcare Productivity - M4 Clinician Idle Time . . . 64
6.2.6 Patient Experience - M5 Patient Waiting Times . . . 65
6.3 Phase 2 - Simulations Evaluating Systematic Assessment Errors (RQ3) . . . 66
6.3.1 Cost Efficiency - M1 Clinician Salary Costs . . . 67
6.3.2 Cost Efficiency - M2 LEON Enactment . . . 69
6.3.3 Healthcare Productivity - M3 Consultations per Patient . . . 72
6.3.4 Healthcare Productivity - M4 Clinician Idle Time . . . 75
6.3.5 Patient Experience - M5 Patient Waiting Times . . . 75
7 Discussion 78 7.1 RQ1 . . . 78
7.2 RQ2 . . . 79
7.3 RQ3 . . . 80
7.4 Reflection on Sustainability Aspects . . . 81
8 Conclusion 82 8.1 Summary and Findings . . . 82
8.2 Contribution . . . 83
8.3 Limitations and Further Research . . . 83
References 89
A Empirical Nursability Distribution 90
B Empirical Meeting Length Distributions 90
1 MTS flow chart for traumatic injuries [15] . . . 9
2 2x2 contingency table . . . 26
3 Overview of the applied mixed method research process . . . 28
4 Overview of the simulated patient flow . . . 32
5 Illustration of patient flow in simulation model . . . 37
6 RQ2: Staffing distribution per model ID . . . 54
7 RQ2: Boxplot of hourly staffing cost in SEK . . . 55
8 RQ2: Boxplot of P3 Final Doctor Meeting End Time . . . 55
9 RQ2: Boxplot of P4 Final Nurse Meeting End Time . . . 56
10 RQ2: Boxplot of P5 Clinician Salary per Patient Journey (excl. cost for idle) . . 57
11 RQ2: Boxplot of P6 True Positive Rate . . . 58
12 RQ2: Boxplot of P7 True Negative Rate . . . 59
13 RQ2: Boxplot of P8 Positive Predictive Value . . . 59
14 RQ2: Boxplot of P9 Negative Predictive Value for MCDM models . . . 59
15 RQ2: Boxplot of P9 Negative Predictive Value for traditional models . . . 60
16 RQ2: Boxplot of P10 Accuracy for MCDM models . . . 60
17 RQ2: Boxplot of P10 Accuracy for traditional models . . . 60
18 RQ2: Boxplot of P11 Nurse Triage Rate . . . 61
19 RQ2: Boxplot of P12 Doctor Resource Management Rate . . . 61
20 RQ2: Boxplot of P13 Number of Patients . . . 62
21 RQ2: Boxplot of P14/P13 Doctor Meetings per Patient . . . 62
22 RQ2: Boxplot of P15/P13 Nurse Meetings per Patient . . . 63
23 RQ2: Boxplot of (P14+P15)/P13 Meetings per Patient for MCDM models . . . 63
24 RQ2: Boxplot of (P14+P15)/P13 Meetings per Patient for traditional models . . 63
25 RQ2: Boxplot of P18 50th Percentile of Waiting Time . . . 65
26 RQ2: Boxplot of P19 80th Percentile of Waiting Time . . . 66
27 RQ2: Boxplot of P20 100th Percentile of Waiting Time . . . 66
28 RQ3: Boxplot of hourly staffing cost in SEK . . . 67
29 RQ3: Boxplot of P3 Final Doctor Meeting End Time . . . 68
30 RQ3: Boxplot of P4 Final Nurse Meeting End Time . . . 68
31 RQ3: Boxplot of P5 Clinician Salary per Patient Journey (excl. cost for idle) . . 69
32 RQ3: Boxplot of P6 True Positive Rate . . . 70
33 RQ3: Boxplot of P7 True Negative Rate . . . 70
34 RQ3: Boxplot of P8 Positive Predictive Value . . . 70
35 RQ3: Boxplot of P9 Negative Predictive Value . . . 71
36 RQ3: Boxplot of P10 Accuracy . . . 71
37 RQ3: Boxplot of P11 Nurse Triage Rate . . . 71
38 RQ3: Boxplot of P12 Doctor Resource Management Rate . . . 72
39 RQ3: Boxplot of P13 Number of Patients . . . 73
40 RQ3: Boxplot of P14/P13 Doctor Meetings per Patient . . . 73
41 RQ3: Boxplot of P15/P13 Nurse Meetings per Patient . . . 74
42 RQ3: Boxplot of (P14+P15)/P13 Meetings per Patient . . . 74
i
45 RQ3: Boxplot of P20 100th Percentile of Waiting Time . . . 77 46 Empirical mass distribution of nursability scores . . . 90 47 Empirical meeting length (minutes) distribution for nurse meetings with outcome
patient helped . . . 90 48 Empirical meeting length (minutes) distribution for nurse meetings with outcome
referral to doctor . . . 91 49 Empirical meeting length (minutes) distribution for doctor meetings . . . 91
ii
1 Mapping of Principles of Cost Efficiency in Primary Care . . . 17
2 Overview of interview sample groups . . . 29
3 List of interviewees . . . 29
4 Interview themes and topics . . . 30
5 Expected meeting lengths . . . 34
6 Collected parameters for each simulated day . . . 39
7 The ideal triage models simulated . . . 40
8 The traditional triage models simulated . . . 41
9 Parameters used for expected cost calculations . . . 44
10 The MCDM models simulated with their respective relative weight combinations 45 11 The systematic errors simulated . . . 46
12 Table of factors mentioned in interviews . . . 50
13 RQ2: Summary of average daily results for the evaluated triage models . . . 53
14 RQ2: Average values of M2 LEON enactment parameters . . . 58
15 RQ2: Average values of M3 Consultations per Patient parameters . . . 61
16 RQ2: Average values of M4 Clinician Idle Time . . . 64
17 RQ3: Average values of M2 LEON enactment parameters . . . 69
18 RQ3: Average values of M3 Consultations per Patient parameters . . . 72
19 RQ3: Average values of M4 Clinician Idle Time . . . 75
iii
We would like to express our utmost gratefulness to our supervisor Anna Svarts at the Institution of Industrial Economics and Management, who despite the COVID-19 situation has found ways to continuously give us valuable in-depth feedback and quickly respond to questions whenever needed.
We would also like to thank our partner company and all colleagues and interview participants who allocated time to contribute to this thesis. The supporting environment that you have provided has been a pleasure to work in. An extra expression of appreciation goes out to our supervisor at the partner company.
iv
Section 1: Introduction
In this chapter, the reader is introduced to the subject and objective of the study. The research questions are presented, together with the groups of performance indicators used to evaluate RQ2 and RQ3. Further, the delimitation and expected contribution of the study is presented.
1.1 Background to Digital Triage
The World Health Organization estimates that by 2035, there will be a global deficit of about 12.9 million healthcare professionals [1]. When there exists a scarcity of healthcare resources, decisions must be made about how to allocate these resources. The ability to use available healthcare resources cost efficiently is, and will continue to be of utmost importance. Digitalization of healthcare services is projected to play a fundamental role in improving healthcare cost efficiency.
Known as triage from the French word trier ’to sort’, the practice of allocating resources to patients originates from the battlefields of war in the beginning of the 18th century. At the time, Napoleon’s army recognized a need to categorize wounded soldiers in order to prioritize treatment for those who needed the most urgent medical attention. Today, triage is most prominently found in emergency care units where the need of prioritizing the most critical patients reemerges.
When it comes to triage within the non-emergent primary care sector, research is limited and official guidelines are few compared to those applicable for emergency care. Still, each Swedish citizen visits primary care on average 1.33 times every year, indicating a high volume of healthcare consultations that must be conducted efficiently. The role of triage is to enable such cost efficient operations. In a systematic literature review, it was found that the available evidence indicates that gate-keeping access to specialized levels of care at the primary care level was associated with lower utilization of health services and lower expenditure [2].
Nonetheless, in the context of modest urgency, rather than to determine the urgency of treatment, the objective of primary care triage in Sweden is to guide the patients to the appropriate level of care. In traditional healthcare, patients calling or visiting their healthcare center are given an initial assessment by the triage officer, usually a nurse, who guides the patients to the appropriate level of care.
As Swedish healthcare is regulated by the Health and Medical Services Act on a nation-wide level, healthcare must be performed so that priority is given to those with the greatest needs of medical services. Furthermore, the law stipulates that publicly reimbursed healthcare must be organized such that it encourages cost efficiency. One commonly cited triage policy is the LEON principle (Lowest Efficient Level of Care) which guides triage officers to guide patients to the lowest (i.e. least costly) level of care at which they can be efficiently helped. However exactly how cost efficiency is supposed to be measured remains a debated topic.
The practice of digital primary care has grown rapidly in Sweden since 2015. This enables pa- tients to meet healthcare professionals for assessments and treatments through digital channels such as text or video. As of 2020, the major digital primary care providers in Sweden employ doctors, nurses and psychologists for these services. Having multiple professions available, just like in traditional primary care, requires that some triage system guides patients to the appro- priate level of care.
At several of the major digital healthcare providers, triage has become an automated process without synchronous contact between the patients and the triage officers. Instead of describing their symptoms to nurses, patients fill out a questionnaire and are directly guided to the appropriate level of care based on their answers. Consequently, an accurate triage system could potentially save salary costs, should it be able to perform the role of the triage officer. Nurses currently employed as triage officers could also use their time to help more patients if such an automated triage system was to be employed. However, if the automated triage system would be less accurate in determining the appropriate level of care than nurses as triage officers, it could lead to an overflow of patients being guided to a higher level of care than required which is costly and a misuse of healthcare resources. Limited research has been done regarding these potential savings and what requirements such an automated triage system must fulfill in order to be beneficial.
Nevertheless, this first-iteration of automated triage employed today is still simple in its essence and fails to account for many important factors such as patient experience, policies and societal economics. For example, imagine the scenario where a patient who is deemed to be suited for nurse care is placed in line for a video meeting with the next available nurse while there are doctors ready to have immediate consultations with the patient. The patient experience of a longer waiting time is not optimal and neither is the fact that a doctor could have helped the patient, making room for more consultations for patients in need.
1.2 Objective & Research Question
Designing a digital triage system for primary care presents two main challenges. Firstly, the system must be able to assess the patients’ medical needs and secondly, it must be able to make a judgment based on certain criteria to guide patients to the appropriate level of care. In its most simple form, one such system may simply ask patients what level of care they believe to be in need of and consequently guide patients to that very level of care. In a more complex form, such a system could generate an estimation of the chance that a certain level of care would be able to fill the patient’s healthcare needs. In turn, the more complex system would take that assumption into consideration alongside several other quantitative data points to produce its judgment.
This study aims to investigate a triage system in the more complex setting, where the digital service generates an estimation of the chances that the different potential levels of care could fulfill the patient’s healthcare needs. As briefly previously discussed, the currently employed models of
digital triage at some of the major digital healthcare providers today fail to account for many of the factors that a triage officer would in the traditional sense of triage. Taking multiple factors into account to produce a decision could be difficult if two different factors favors two different outcomes. In operations research, the topic of Multi-Criteria Decision-Making (MCDM) covers evaluation of multiple criteria to produce a decision where there does not exist a unique optimal solution as some criteria might be conflicting. Employing MCDM in a digital triage setting is one way of enabling an expansion of the information upon which to base a triage decision.
Based on this objective and previous background, this study aims to investigate whether digital primary care enables usage of more complex MCDM-based triage policies with comparative advantage to the traditional triage policy. This is performed by answering the three questions below.
At first, a qualitative investigation is needed to outline the set of factors to construct the MCDM algorithm upon.
RQ1 - What set of factors should be considered in an automated digital triage system?
Secondly, the performance of a complex MCDM-based triage model must be investigated and evaluated through a set of relevant criteria.
RQ2 - Is it possible, using Multi-Criteria Decision-Making models to improve cost efficiency, healthcare productivity and patient experience?
Finally, to understand the limitations of an automated triage system, the investigation seeks to explain the impact of a triage system based on bad data input.
RQ3 - What is the impact of a systematic error in the digital triage assessment in terms of cost efficiency, healthcare productivity and patient experience?
1.2.1 Evaluating Efficiency, Productivity and Patient Experience
The cost efficiency, healthcare productivity and patient experience of the healthcare system have been evaluated by five groups of performance indicators. These groups have been decided upon on the basis of Sweden’s national indicators and fundamental principles for good-quality health and medical care, as set forth by the National Board of Health and Welfare (NBHW) [3]. We determined which of the national indicators that have relevance within the scope of this study and adopted them into the appropriate measurements. They are, under each respective field of monitoring:
Cost Efficiency
M1 - Clinician salary costs M2 - LEON enactment Healthcare Productivity
M3 - Consultations per patient M4 - Clinician idle time Patient Experience
M5 - Patient waiting times
1.3 Delimitations
At first, it should be noted that the scope of the study has been confined to the Swedish primary care. That does not necessarily mean that the results obtained are not applicable to healthcare systems in other countries, however the study has been designed based on Swedish healthcare policy. Furthermore, the study will assume a primary care setting in which patients are to be initially guided to one of two levels of care. The lower level of care is represented by nurses and the higher level of care is represented by doctors. This is the most impactful delimitation of this study since this greatly differs from a real setting in which patients may potentially see healthcare workers of many different specialties and professions. However, as patients visiting physical primary care generally start their journey by a nurse meeting regardless of where they eventually get helped, we argue that such a delimitation does not limit our ability to answer the research questions.
This study has been conducted in partnership with one of Sweden’s leading digital healthcare providers. As such, the quantitative input data that has been used in the study to model patients’
needs has been generated by patients’ needs when seeking care at a digital healthcare provider.
Therefore, it can be assumed that the results of the study are best applicable for patients with primarily symptoms suitable for digital care. Furthermore, for some patients present in the input data figures, a previous assessment could potentially have taken place with a nurse by phone which would impact the type of patients and symptoms making up the data.
The nature of the quantitative input data also means that the study disregards seasonal patterns of the healthcare needs of Swedish primary care patients. As the conditions themselves may change, it it however believed that the severity of the treated conditions has a small seasonal effect and may be disregarded.
The study also assumes that all patients who seek care are in need of care and in fact can be helped through a digital consultation. In reality, there will unavoidably be patients who visit healthcare centers, physical and digital, that do not need healthcare or that need to be referred to another healthcare institution. From a triage point-of-view however, these patients should be referred to the appropriate institution before determining the appropriate level of care within
the institution. Further, contact with experienced healthcare professionals and the analytics department at our digital healthcare partner has confirmed that this share of patients is small.
As such, we will disregard these patients in the scope of our thesis.
Regarding the role of triage to distribute healthcare resources in order of patient needs, we will assume a setting of digital healthcare in which patients have a comparable need for care. That means that we study a triage system in which no patient is given priority over another patient based on the urgency of the two patients’ healthcare needs. Rather, we assume a situation where the role of the triage system is solely to determine the appropriate level of care. This is a suitable delimitation when studying a digital healthcare setting as the patients seeking digital primary care all tend to have concerns with little or no urgency.
It should be emphasized that this study does not focus on improving the assessment of patients, but rather the judgment such an assessment should render. These stages of the triage process is more formally defined in the next section. Moreover, when comparing digital tools for triage with traditional judgments performed by a nurse, this study will assume that the judgment of the nurse is completely accurate and that a nurse always will guide a patient to the lowest level of care that can treat the patient.
1.4 Expected Contribution
The healthcare resource deficit makes efficiency improvements within the healthcare sector a complete necessity to assure available and safe healthcare for everyone in the future. In a state public report from 2019, it is concluded that tools for automated anamnesis and triage have a great potential of unlocking healthcare resources by facilitating patients being directly referred to the the most appropriate level of care [4]. Further, following a research request from Stockholm region, Lagerros et. al. published a report in December 2019 studying the impact of digitalizing primary healthcare [5]. It was concluded that automated triage systems show great potential, but emphasized that there is limited research on the subject and that evidence based knowledge about their efficiency, patient safety and resource utilization currently is limited and would be of great value.
This study aims to bridge the knowledge gap about the implications of automated triage in digital primary care. Historically, most triage research has focused on emergency healthcare, but given the current state of the healthcare sector with and increasing resource deficit, there is a great need of efficiency improvements in the primary care sector. By studying the potential of using MCDM models as a tool for automated triage, we hope to contribute with evidence based insights on the efficiency of automated triage systems digital primary care.
1.5 Report Outline
The thesis proceeds as follows:
1) Introduction - In this chapter, the reader is introduced to the subject and objective of the study. The research questions are presented, together with the groups of performance indicators used to evaluate RQ2 and RQ3. Further, the delimitation and expected contri- bution of the study is presented.
2) Triage Literature - This chapter presents the findings of previous and relevant research conducted within the field of triage. Given the limited previous research conducted on triage in primary healthcare, it is mostly focused on triage research in an emergency healthcare setting. However, it also presents two triage decision support tools developed for Swedish primary care.
3) Empirical Context - This chapter provides a description of the Swedish primary healthcare system, including governance, laws, guiding triage policies and the emerging sector of digital primary healthcare.
4) Theoretical Background - This chapter presents the theoretical background of the MCDM model. It proceeds with the theoretical aspects of using simulation as a tool and presents the performance evaluation framework used to analyze the simulation outcome.
5) Method - This chapter presents the method of the study. It accounts for methodological choices and provides details of the processes associated with the interviews and simulation.
It is concluded with a discussion of the quality of research
6) Empirical Findings - This chapter presents the findings of the study. It is structured in chronological order, starting with a presentation of the interview findings and then proceeding with the simulation results. The simulation results are structured according to the performance measurements presented in section 1.2.1.
7) Discussion - This chapter discusses the empirical results and is structured according to the research questions. It also includes a discussion of the sustainability aspects of the study.
8) Conclusion - This chapter presents the final conclusion to the stated research questions, its contributions to science and suggestions for further research.
Section 2: Triage Literature
This chapter presents the findings of previous and relevant research conducted within the field of triage. Given the limited previous research conducted on triage in primary healthcare, it is mostly focused on triage research in an emergency healthcare setting. However, it also presents two triage decision support tools developed for Swedish primary care.
2.1 Formal Definition of Triage
Throughout this thesis, the term triage in its traditional sense will be used in accordance with the definition published by Iserson & Moskop [6]. This requires three conditions to be satisfied:
1. At least a modest scarcity of healthcare resources exists. In circumstances where resources always are sufficient to meet the needs of patients with immediate attention, no triage is needed. At the other extreme, without any available healthcare resources, triage is irrelevant. In a primary care setting where triage guides patients to either doctors or nurses, we interpret this condition as focusing on scarcity of the higher level of care, i.e.
doctors.
2. A healthcare worker (often called a “triage officer”) assesses each patient’s medical needs, usually based on a brief examination. This emphasizes that triage is distinguished as a process of allocation on a per-unique-individual basis.
3. The triage officer uses an established system or plan, usually based on an algorithm or a set of criteria, to determine a specific treatment or treatment priority for each patient.
Compared to purely ad hoc or arbitrary decisions, this underlines that triage systems by definition contain systematics.
In regards to digital triage, no formal definition has been published. We argue that the first of Iserson & Moskop’s conditions remains unaltered as it acts as a prerequisite rather than a constraint of what triage is. As for the second criterion, the main aspect of individual assessment is retained. However, digital triage substitutes the healthcare worker with a digital service that performs an assessment without involvement from any individual. We will refer to this as digital assessment. At the same time, this digital assessment depends on the patient’s ability to be able to communicate through the employed digital channel and the service’s ability to quantify the patients’ needs. In its most basic form, this assessment could potentially be one single, fixed questionnaire. In contrast, a digital assessment could also mean customized evaluations interpreted through text or speech, based on the patients’ symptoms, demographic factors and previous contacts with health services, as well as digitally monitored vital parameters (e.g. blood pressure and heart rate). From such a condition, the third of Iserson & Moskop’s criterion is easily altered to define digital triage such that the triage system, rather than a triage officer,
determines the features of the future treatment. As such, we will assume the following conditions for usage of the term digital triage throughout this thesis:
1. At least a modest scarcity of healthcare resources exists.
2. A digital service assesses and quantifies each patient’s medical needs, usually based on brief self-assessed statements by the patient.
3. The digital service uses an established system or plan, usually based on an algorithm or a set of criteria, to determine a specific treatment or treatment priority for each patient.
It should be emphasized that our use of the term digital triage incorporates that the assessment and judgment of patients are automated processes, in addition to being digital. This means that no individual assessment or judgment is performed for each patient by a healthcare worker.
2.2 Previous Research on Triage Models
Existing research on triage is largely focused on emergency healthcare systems and how to ensure clinical justice for patients at emergency departments [7] [8] [9]. A search on the Web of Science database on the topic “emergency” AND “triage” AND (”healthcare” OR ”health care”) in the time span 1975-2019 yields 1,270 results, whereas a search on (”automated” OR “digital”) AND
“triage” AND (”healthcare” OR ”health care”) only yields 57 results. Out of these published articles not specifically focused on triage in emergency departments, most research focus on the medical adequacy of digitizing the triage process within specific medical areas, such as dermatology, oncology or other medical specialities [10] [11] [12].
2.2.1 Triage in Emergency Care
As previously mentioned, the purpose of triage in emergency healthcare is to determine the priority of treatment for emergency patients based on the severity of their condition. Triage systems serve as a method for systematic prioritization and are often based on trigger tools for vital signs, usually including systematic questionnaires adapted for different settings and conditions. Research on emergency healthcare often refers to three phases of triage [13];
• Pre-hospital triage - Aiming at allocating and dispatching ambulance and pre-hospital care resources
• Triage at scene - Performed by the first clinician attending the patient, giving a first in-person judgement of the patient’s condition
• Triage on arrival - Performed as patient arrives to the emergency department or receiving hospital and aimed at prioritizing the patient’s treatment according to the urgency in need of care
Research has been performed on all different phases of triage, but most established triage systems are those developed during the 1990s and 2000s for use in the emergency departments, i.e.
in the triage on arrival phase. The Australasian Triage Scale (ATS), Canadian Triage and Acuity Scale (CTAS), Manchester Triage System (MTS), Emergency Severity Index (ESI) and Medical Emergency Triage and Treatment System (METTS) are some of those, which all have disseminated and been implemented at emergency departments all over the world [14]. In Europe, MTS is the most commonly used system in emergency departments. In this system, orthopedic disorders are divided into five groups; traumatic injuries, joint pain, vertebral pain, postoperative disorder, and musculoskeletal infection. For each group, a flow chart has been developed to help the triage officer assign the patient one of five different urgency categories, each with a standardized colour tag and a maximum waiting time ranging from immediate (0 min) to non- urgent (240 min) [15]. Below, an example of a flow chart in the MTS is illustrated:
Figure 1: MTS flow chart for traumatic injuries [15]
MTS is one of many different triage systems, but the concept of urgency categories based on a severity criteria is similar for most systems [13]. There are numerous studies evaluating and comparing the performance of different emergency triage systems, but differences in study design, study populations, reference standards etc. have shown to have great impact on the results.
Hence, up until today there is no established consensus on the most reliable and efficient triage system in emergency care [8].
2.2.2 Triage Judgment Frameworks in Swedish Primary Care
Even though there, in terms of evaluating patients’ need of care to allocate healthcare resources, are some similarities between triage in emergency and primary care, the prerequisites and the triage systems differ a lot. Unlike in emergency departments, primary care patients are usually not in urgent need of care. Therefore, evaluation of the severity of the patients’ condition to prioritize who gets help first plays a vital role in emergency care, while not being as important in primary care triage. Instead, primary care triage aims at utilizing healthcare resources in the most efficient way by sending the patient to the lowest (i.e. least costly) level of care at which they can be efficiently helped [4].
Triage systems in primary care are far less developed than the widely researched emergency triage systems previously presented. However, some of the Swedish regions have developed their own triage handbooks, aimed at supporting triage officers in primary healthcare with guidelines on how to decide needed level of care for primary care patients. Region Sk˚ane has developed
”Triagehandboken”, a decision support tool developed and reviewed by specialist representatives from various medical fields. Just like many of the established emergency triage systems it applies a symptom based approach, where judgment of vital functions is a central part of the triage process. For each symptom, a systematic guide has been developed to help the triage officer gather relevant information, such as pain intensity, duration and other simultaneous factors, e.g.
age or other complicating diagnoses. The answer to those questions helps the triage officer to decide the most appropriate level of care.
In a similar way, N¨arh¨alsan, the public provider of primary care in the V¨astra G¨otaland region, has developed ”En handbok f¨or N¨arh¨alsans sjuksk¨oterskor ”, a triage guide for nurses in V¨astra G¨otaland. It has many similarities with ”Triagehandboken”, containing instructions on how to systematically assess vital functions to increase understanding of the symptoms and consequently, be able to triage the patient to the most appropriate level of care.
Section 3: Empirical Context
This chapter provides a description of the Swedish primary healthcare system, including governance, laws, guiding triage policies and the emerging sector of digital primary healthcare.
3.1 The Swedish Primary Care System
Primary care in Sweden is a part of the open care which serves patients’ day-to-day healthcare needs. At 55.9 billion SEK, the primary care sector accounts for 10.6% of Sweden’s 526.2 billion SEK annual healthcare bill as of 2018 [16] [17]. In most cases, primary care acts as the first point-of-contact for patients’ non-emergent healthcare services, coordinating and referring patients to specialist care when needed. To the Swedish public, healthcare centers serve as their primary care institutions which employ healthcare professionals in several categories such as general medicine, psychology and physiotherapy.
3.1.1 The Governance System
The Swedish healthcare system is mainly funded by tax money. As such, healthcare providers are regulated by law and policy in regards to how to treat patients and how to get reimbursed for their provided care. The general governance structure of Sweden’s public institutions is divided into three levels; national, regional and local. On a nation-wide level the Ministry of Health and Social Affairs oversees all questions regarding social welfare. Covering the full range of healthcare from primary care to hospital care, the National Board of Health and Welfare (NBHW) operates under the Ministry on a nation-wide level. NBHW works with establishing standards, principles and guidelines for Swedish healthcare practices. Furthermore, on a nation- wide level, the government decides on legislation and the state budget from which money is being directed to the regions and municipalities in state grants.
Healthcare organized in the 21 regional county councils totals 313.6 billion SEK annually as of 2018 and in terms of governance of primary care, they are the most important institutions.
The county councils are responsible for delivering and organizing much of the healthcare in each region and they have authority to individually regulate primary care in its geographical area.
This leads to primary care in Sweden being conducted based on 21 different preconditions. On the local level, 290 local municipalities have governing power, however their role in terms of healthcare is marginal and of no essence in terms of primary care. In coordinating the county councils and the municipalities, all regional and local organizations are members of the Swedish Association of Local Authorities and Regions (SALAR). SALAR supports the county councils and municipalities through coordination in e.g. collective agreements or knowledge exchange in e.g. best-practice guidance.
Returning to nation-wide law, the main governing factor of Swedish primary care on a broad level is the Health and Medical Services Act. The law stipulates that primary care should be provided without delimitation in terms of illnesses, age or patient groups. Furthermore, according to law, the responsibilities of primary care comprise treatment, caregiving, preventive efforts and rehabilitation where hospital care or other specialized care is not required.
3.1.2 The Act on System of Choice in the Public Sector
According to the Health and Medical Services Act, regions are obliged to organize the primary care as a system of choice. This means that citizens must be allowed to freely choose between all available primary care providers and that the county councils must treat all providers within each of the respective regions equally. As citizens freely choose where to satisfy their healthcare needs, the public reimbursement from the county council will also follow the citizens and go to their choice of healthcare provider. One important effect of this system of choice is that privately operated healthcare providers can compete with publicly operated healthcare providers on the same conditions according to law. This was enabled in the beginning of 2009 as the Act on System of Choice in the Public Sector was adopted into law.
Mandated by the Health and Medical Services Act as the regulation to apply in enabling a system of choice, the Act on System of Choice in the Public Sector dictates the tendering process in primary care. As such, each region is required to publish a tender document that outlines the terms of entering into an agreement with the county council which eventually enables any healthcare provider to be publicly reimbursed. Such a tender document regulates the scope of the assignment in terms of services, requirements and reimbursement. All healthcare providers that meet the requirements of the tender document in a specific region have the right to establish themselves in primary care with public reimbursement and serve the citizens of that region. As of 2017, 43% of primary healthcare centers in Sweden are privately operated having fulfilled the respective requirements of the county councils according to the Act on System of Choice in the Public Sector [18]. In doing so, these healthcare providers take part in fulfilling the regions’
responsibility of providing primary care to the Swedish population. One important factor when observing the market competition however, is that all healthcare providers that are publicly reimbursed can not compete through pricing as patients’ co-pay and the reimbursement itself are standardized within each region.
3.1.3 Public Reimbursement
When outlining the public reimbursement system of primary care in Sweden, the circumstance of regional governance should be underlined. As a consequence of the regional sectioning, there are 21 different bases of how healthcare providers are reimbursed wherein no reimbursement model is the same. However, there are three general themes of how primary care providers are reimbursed: capitation, fee-for-service and performance-based.
Capitation accounts for the largest share of the total public reimbursement. The exact share of capitation reimbursement relative to the total reimbursement is in the range 50-100 percent, but varies between different regions [19]. The fundamental policy of capitation is that each citizen can choose to enlist with a specific healthcare provider. The caregiver is then responsible to carry out the enlisted patients’ healthcare needs. For this responsibility, the healthcare providers are paid a monthly sum for each enlisted patient, regardless of how much care the enlisted patients seek. This is what’s known as the capitation reimbursement. The actual sum that is paid out for each patient differs, both between regions and within regions as there are three main components that are used to calculate the capitation reimbursement for each provider’s patients.
These are 1) Adjusted Clinical Groups (ACG), 2) Care Need Index (CNI) and 3) the belonging of patients into different age-groups. ACG is a system developed by researchers at the John Hopkins University in Baltimore, which assigns patients an ACG value based on their previously recorded medical diagnoses. CNI is an index based on seven socio-economic factors which have been found to relate to patients’ healthcare needs. Both the adoption of, and eventual weight assigned to each one of these three components differ between regions.
A reimbursement structure that relies on capitation creates incentives for healthcare providers to increase cost efficiency, patient safety and preventive care [20]. In terms of triage, this means that capitation reimbursement also incentivizes that triage should be performed such that the resulting care provided to the patient is as cost efficient as possible.
Fee-For-Service reimbursement is paid to the healthcare provider for each meeting. As the name suggests, this component of public reimbursement is paid to healthcare providers for each completed consultation. In many of the regions, patients themselves bear a part of the cost for their consultation. If such a co-pay is paid, this is part of fee-per-service reimbursement.
Apart from regional variations, the sum of the fee-for-service reimbursement in many cases differ between healthcare professions and whether the patient has enlisted at the provider or not. For example, as of 2019, one third of the county councils do not pay any fee-per-service reimbursement for enlisted patients [21]. The fee-for-service reimbursement can however also reduce the public reimbursement to the healthcare providers. In several regions, the county councils apply a penalty fee for providers whose enlisted patients have consultations with other caregivers.
The implications of the fee-for-service reimbursement in terms of financial incentives in conducting triage are not as straight-forward as the ones of the capitation reimbursement.
There are no general conclusions to be drawn in terms of triage. As different regions apply different reimbursement policies, the most beneficial level of care in terms of gross profits will differ between regions. More specifically, the determining factor in terms of financially incentivizing triage to a certain level of care is found in the difference of reimbursement sums for consultations at different levels of care. Regarding this, there are insights to be found when comparing the reimbursement policies of different regions. As one example, Region Stockholm pays a fee-per-service reimbursement for doctor meetings of 260 SEK and for nurse meetings of 230 SEK. This difference of 30 SEK will, at the current market levels, never outweigh the
higher salary cost for doctors. Thus, the reimbursement policy in Region Stockholm incentivizes triage to nurses, which is in line with the county council’s guidelines in their tender document where healthcare providers are told to treat patients according to the LEON principle, which suggests nurses as the primary endpoint of a triage system. In contrast, Region ¨Osterg¨otland, instructing providers to employ the MEON principle pays 250 SEK more for doctor meetings than for nurse meetings, approximately equalling the gross profit of the two respective meeting types at different levels of care.
Performance-Based is the third major theme that makes up the public reimbursement system. Relative to the other two major themes, performance-based public reimbursement generally constitutes a small share of the total reimbursement. The sum of performance-based reimbursement, just like the two other major themes differs between regions, but can be based on patient satisfaction, care coordination, continuity, enrollment in national registers or compliance with evidence-based guidelines to name a few examples. Many regions employ coverage ratio as one performance measure, aiming at incentivizing healthcare providers to treat patients in primary care when possible rather than referring them to specialist care.
Two-thirds of regions also reimburse healthcare providers on geographical basis, for example compensating players in areas of low population density or with long distances to the closest hospitals. Note that this reimbursement in Region ¨Osterg¨otland is only paid for consultations with patients enlisted at another healthcare provider. The fee-per-service reimbursement of enlisted patients’ consultations is 0 for all levels of care, thus incentivizing nurse meetings based on gross profit.
3.1.4 Public Reimbursement of Non-Resident Patients
All the above themes of reimbursement however only applies to their full extent when providers treat patients within the one of the 21 regions that they are established in. Since patients are still legally allowed to seek care in regions outside their home region, the regions through SALAR have established a common policy for reimbursement of non-resident patient consultations. This policy enables healthcare providers to be reimbursed on a fee-for-service basis for treating non-resident patients. Since digital healthcare meetings do not require physical presence at a certain location by neither the healthcare professional not the patient, this is what essentially has enabled digital providers to operate on a national level with public reimbursement. Through establishing themselves as a healthcare provider in one region according to the Act on System of Choice in the Public Sector, digital healthcare providers have been treating patients from outside their established region as non-resident patients. As such, the digital practice has enabled a national reimbursement model not intentionally created for this purpose. As of 2020, the public reimbursement for digital meetings with non-resident patients according to the agreement in SALAR are as follows: [22]
• Consultations with doctors: 500 SEK
• Consultations with psychologists, councelors or psychotherapists: 425 SEK
• Consultations with other care professions: 275 SEK
3.1.5 Triage As a Requirement For Reimbursable Digital Healthcare
In the wake of digital healthcare consultations emerging in Sweden, questions have risen regarding the ability of digital consultations to substitute physical ones. To ensure that the tax money spent on publicly reimbursed digital healthcare meetings serve its purpose, SALAR has set up five criteria which establish the principles for reimbursing digital healthcare. [23] The principles were then attached to the recommendations sent to the county councils regarding reimbursement of non-resident patient consultations. The five principles are:
• The meetings must be preceded by ID check through strong authentication.
• The meetings must be preceded by an assessment to exclude symptoms and diagnosis that should be taken care of by physical care or that no not need medical attention.
• The meeting must constitute of “eligible healthcare” according to the definition by NBHW, and thus not be a question of advisory services.
• The meeting must fulfill the same requirements on EMR usage and reporting as equivalent meetings in primary care according to the criteria set out by the county council.
• The digital healthcare provider is liable to pay for and to have routines in place for referring patients in need of lab tests and other medical services.
In terms of digital triage, the second criteria requires at least some form of triage to occur before a digital meeting in order to be eligible for reimbursement.
3.2 Guiding Principles in Swedish Healthcare
3.2.1 Prioritization and the Ethical Platform
According to a government bill adopted by the Swedish Parliament in 1997, prioritizing in Swedish healthcare must be performed based upon an ethical platform of three principles. These are, in hierarchical order: [24]
1. The Principle of Human Rights, meaning that all human are equal regardless of personal characteristics or societal function.
2. The Principle of Need and Solidarity, meaning that resources should be distributed according to needs.
3. The Principle of Cost Efficiency, meaning that when faced with a choice between different actions, a reasonable relationship between costs and effect should be strived after.
Effects are to be measured by improved health and improved quality of life.
As mentioned previously, patients within digital healthcare tend to have a comparable need of care, meaning that no prioritization must be done according to needs. As such, prioritization of healthcare resources, i.e. primary care triage in our setting of digital healthcare, should be performed whilst striving after a reasonable relationship between costs on one side, with improved health and improved quality of life on the other side.
3.2.2 Principles of Cost Efficiency
Returning to the Health and Medical Services Act, Swedish law stipulates that publicly reimbursed healthcare should be organized such that it promotes cost efficiency. The NBHW also lists efficient healthcare as one of six areas which constitute good-quality health and medical care in Sweden [25]. Efficient healthcare according to NBHW is defined to describe that ”available resources are utilized in the best possible way to reach intended targets”. This means, NBHW continues, that ”healthcare is organized and supplied in cooperation between the the players in the healthcare system based on the severity of the patient’s illness and the cost efficiency of the treatment”. Distinguishing efficiency from productivity, the NBHW underlines that efficiency sets results in relation to costs whilst productivity sets efforts in relation to costs and thus, is only one of multiple factors that constitute efficiency.
One of the nation-wide organizational measures to promote cost efficiency is the capitation reimbursement model previously discussed. Another cost efficiency measure which could have large effects on triage systems is the LEON principle. An acronym for the Swedish translation of Lowest Efficient Level of Care, the LEON principle could serve as a guiding principle for what level of care a patient should be referred to. In the 2016 final report of the government inquiry into efficient care, led by G¨oran Stiernstedt, the LEON principle is described as what should be the ”obvious strategy” in healthcare. The report further defines the principle as the process of ”directing duties to the profession that can perform them for the lowest total cost with maintained or improved quality”. At the same time, the report also notes that this principle does not seem to be a given fact in regionally funded healthcare such as primary care. [26]
The lack of adopting the LEON principle in Swedish primary care is reflected by a mapping of the currently valid tender documents as written by the county councils. Only in Stockholm are healthcare providers explicitly mandated to treat patients according to the LEON principle.
However, such a mapping also reveals that other principles of cost efficiency have been adopted within Swedish primary care. The following alternatives of the LEON principles are currently present in the regions. Note that V¨armland only applies BEON as a policy in publicly operated primary care, meaning that the policy is not present in the tender documents.
Policy ... Efficient Level of Care
# of County
Councils County Councils
LEON Lowest 1 Stockholm
MEON Most 3 Kronoberg, V¨asterbotten, ¨Osterg¨otland
BEON Best 2 Blekinge, V¨armland
NEON Closest 1 J¨amtland/H¨arjedalen
Table 1: Mapping of Principles of Cost Efficiency in Primary Care
The consequences of the different policies are unclear as no tender document presents a clear definition of their policy. According to the definition of the LEON principle presented in the government inquiry final report, one could argue that all policies essentially are the same without any further explanation. There does however seem to be a distinction made amongst healthcare professionals, especially between the most well-known LEON principle and the MEON principle, most frequently present in the tender documents from the county councils. For example, in 2019 the founder and CEO of Min Doktor Magnus Nyhl´en wrote a debate article on the Swedish debate platform Dagens Samh¨alle Debatt, arguing for substituting the LEON principle for the MEON principle, in opposition to the previously released report by the government inquiry [27].
As there are no commonly adapted definitions of neither the LEON, the MEON, the BEON nor the NEON principles, it is difficult to draw any further conclusions about their consequences on triage in primary care. What can be said with certainty is that in this debate, there are two main schools of thought when applied to primary care. One perspective is held by those who argue that nurses should serve as the first point of contact [28]. The other side argues that patients shouldn’t have a designated first point of contact but rather be referred to the most appropriate level of care immediately when such a judgment can be made ahead of the first meeting with a healthcare professional [29] [30].
3.3 Digital Primary Care Practice
In June 2016, a report was published by McKinsey which concluded that digitalization of the Swedish healthcare system potentially could lead to 25% cost savings by 2025 [31].
Approximately 11% of the potential cost savings were attributed to consultations at distance.
The practice of digital primary care has grown rapidly during the second half of the 2010’s.
Although many of the privately operated digital healthcare services were founded some years before, the starting point for nation-wide expansion was when Region J¨onk¨oping equated digital healthcare consultations with physical ones in the spring of 2016. The digital healthcare services could then partner up as subcontractors to healthcare providers in Region J¨onk¨oping with agreements in place according to the Act on System of Choice in the Public Sector. By doing so, their digital healthcare services could be used by any Swedish patient according to the Health and Medical Services Act, allowing all citizens to freely choose between all available
primary care providers [32]. Consultations with patients residing outside Region J¨onk¨oping would then be reimbursed according to the policy on non-resident patients. During 2018, 4.6 percent of all primary care consultations with doctors in Sweden were performed through privately operated digital services, with the patient formally treated as a non-resident patient.
3.3.1 Major Players and Triage
As the digital healthcare providers have grown in number of yearly consultations, the diversity of said providers has also grown. For example, the digital healthcare practice has been adopted by many county councils in charge of their publicly operated healthcare providers. When the digital healthcare practice first started, doctors were the only healthcare profession available through the digital services. The services today involve more professions than just doctors, such as nurses, psychologists and midwives. As of 2019, the three largest privately operated digital healthcare providers in terms of number of digital meetings are KRY, Min Doktor and Doktor.se. There are a few similarities between all three players. First of all, they all operate using a smartphone application as their primary contact method. Also notably, they all operate on a nation-wide level through Region S¨ormland, much due to the fact that the county council in S¨ormland enables the digital healthcare providers to offer their publicly reimbursed services without co-pay for the patients. Furthermore, the three all employ at least both doctors and nurses, and thus, use some triage system to guide patients to the appropriate level of care.
KRY is currently the largest player in the digital healthcare industry. The majority of the digital consultations at KRY are conducted through synchronous video meetings between patients and healthcare professionals. In addition to their digital service, KRY has since December 2018 established physical presence, enlisting patients in two regions; Region Sk˚ane and Region S¨ormland. In terms of triage, KRY claims to have developed a digital triage system in which the assessment is performed on the basis of artificial intelligence [33]. When initially launching the triage system, the assessment relied on pre-defined rules intended to imitate the judgment of a nurse, similar to the triage handbooks presented in Section 2.2.2.
Gathering data from their service, KRY today aims to use machine learning to teach the triage system what level of care that is most appropriate, based on the patients’ previous healthcare consultations, demographics and recorded outcomes for similar symptoms. With such a triage system, KRY looks to facilitate processes where patients always are guided to the appropriate level of care at the first consultation.
Doktor.se is the most newly founded of the three major players, launched 2016. Similarly to KRY, Doktor.se conducts the majority of their digital healthcare consultations synchronously, although conducting text-based chat-like consultations and audio-calls without video to a larger extent than their competitor. Doktor.se operates physical healthcare centers in three regions;
Region Sk˚ane, Region S¨ormland and Region Uppsala. Furthermore, they have physical presence through a cooperation with the pharmacy Apoteket Kronan, operating a handful of small clinics
at the pharmacies. These clinics are not on their own complete healthcare centers with ability to enlist patients. Regarding triage, Doktor.se advocates for a system with more traditional triage performed by a triage officer, i.e. a nurse. As such, nurses are always the initial level of care when seeking healthcare at Doktor.se and consultations with doctors are only booked by a nurse. When Region Stockholm in 2019 requested information from the industry regarding how to act in order to ensure that digital offerings became an integrated part of the healthcare in the region, Doktor.se was one of many respondents. In the response, Doktor.se suggested that all triage had to be performed by a nurse and that using doctors as the first level of care should be made non-applicable for public reimbursement. Furthermore, Doktor.se suggested in the same document that the region should institute a public reimbursement policy for only the triage, i.e.
guiding patients to the appropriate level of care without treatment.
Min Doktor was initially founded with a focus towards insurance patients as there were no path towards publicly reimbursed digital healthcare when the company started up. Min Doktor performs the major part of the digital healthcare consultations asynchronously through text, but consultations through audio or video also occur. In Region S¨ormland, Min Doktor has one physical healthcare center but the physical footprint of Min Doktor is mainly focused to pharmacies. In 2018, retailer ICA Gruppen bought a 42 percent share of Min Doktor through their pharmacy subsidiary Apoteket Hj¨artat. Since then, Min Doktor has operated Apoteket Hj¨artat’s local clinics located in proximity to ICA’s grocery stores. Besides light healthcare services, Min Doktor clinics also perform a number of tests and vaccinations. In guiding patients to either nurses or doctors, Min Doktor publicly endorse the MEON principle, aiming to help the patients in short and efficient care chains. Equal to KRY, patients to Min Doktor answer a set of predetermined questions based on their symptoms before the meeting which is the basis of the assessment in the triage system. The specifics of the assessment and judgment in the triage system are not publicly known, however the use of asynchronous meetings certainly allows for a less automated solution.
3.3.2 Criticism
The digital healthcare practice has been subject to much criticism. A wide range of critical opinions have been heard in the public forum. The main factors of criticism have regarded the medical quality of the services, the reimbursement policy used by digital healthcare providers and that digital healthcare does not follow the principle that healthcare should be given according to needs [34] [35].
Related to triage, the key policy risk for digital healthcare providers is the consequence of a triage system where patients are guided to an inappropriately high level of care. In such cases, questions could be raised whether the digital healthcare providers adhere to the law of organizing the publicly reimbursed healthcare on principles of cost efficiency. As much of the criticism towards the digital practice also has revolved around providers draining the healthcare system of financial resources, a faulty triage system could further fuel such opinions since doctor
meetings have a higher reimbursement sum than nurses at a lower level of care.
Section 4: Theoretical Background
This chapter presents the theoretical background of the MCDM model. It proceeds with the theoretical aspects of using simulation as a tool and presents the performance evaluation framework used to analyze the simulation outcome.
4.1 Multi-Criteria Decision-Making
The intellectual challenge of making decisions in complex environments is as old as mankind and something that most people encounter, both in work and everyday life. The approaches to complex decision-making have varied through history and historically, a common approach has been to seek advice from from oracles, kings or priests. However, simultaneously with the last century’s development of scientific disciplines, the old methods have been replaced with modern science and technology. Today, there are several theories such as linear programming, dynamic programming and inventory optimization amongst others, that have been widely researched and that all have the common element of acting as a tool in search for optimal solutions, or decisions.
One of those methods, that has captured a lot of attention in recent years, is the Multi-Criteria Decision-Making (MCDM) model [36].
Multi-Criteria Decision-Making models have during the last decades become a widespread strategic decision-making tool within several different areas and disciplines. They are used to, given a set of alternatives and decision criteria, numerically rank and evaluate the different alternatives to in the end, be able to choose the best one. They are often used in settings with contradicting criteria, e.g. when balancing decreasing cost and maintaining quality. All decision-making models including numerical analysis of alternatives involve three steps [36]:
1. Determine the relevant criteria and alternatives.
2. Attach numerical measures to the relative importance of the criteria and to the impacts of the alternatives on these criteria.
3. Process the numerical values to determine a ranking of each alternative.
Zimmermann [37] divides MCDM into two categories, Multi-Objective Decision-Making (MODM) and Multi-Attribute Decision-Making (MADM). MODM studies decision problems with a continuous decision space, where mathematical programming problems is a typical example. MADM, which is often referred to as MCDM, focus on decision problems with finite, discrete and predetermined decision alternatives. In the setting of this study where there are only two decision alternatives, send the patient to a doctor or to a nurse, MODM is not relevant and will therefore not be further introduced.
4.1.1 MCDM Definition
MCDM problems are easily expressed in matrix format. Assume we have m alternatives Ai
(where i = 1, 2, . . . , m) and n criteria Cj (where j = 1, 2, . . . , n). Let A be a (m × n) matrix, in which each element aij represents the performance of alternative Ai with respect to criterion Cj. Further, assume weights wj (j = 1, 2, . . . , n) representing the relative weight of each criterion.
Then, a typical decision matrix is constructed as follows [37]:
Criterion
CCC1 CCC2 CCC3 . . . CCCn (w1 w2 w3 . . . wn)
A A
A1 a11 a12 a13 . . . a1n
A A
A2 a21 a22 a23 . . . a2n ... ... ... ... . .. ... A
A
Am am1 am2 am3 . . . amn
Given a decision matrix as per above, the difference between the different MCDM methods is how to process the numerical values in the matrix to determine the rank of each alternative. Some examples of widely used methods are the weighted sum model (WSM), weighted product model (WPM), analytical hierarchy process (AHP) and revised analytical hierarchy process (revised AHP). The following definitions will assume that all criteria are to be minimized. However, this could be replaced by a maximization case.
WSM The WSM model is the most commonly used MCDM approach, especially in single dimensional problems. In this model, the best alternative AAA∗ is the alternative i that minimizes (given that it is a minimization case) the weighted sum:
AAA∗= min
i n
X
j=1
aijwj, for i = 1, ..., m (1)
A vital assumption for this model is the additive utility assumption, which requires all criteria to be measured in the same unit. Hence, in multi-dimensional problems with criteria measured in different units, the WSM model is not applicable.
WPM Unlike the WSM model, WPM eliminates any units of measure and is therefore sometimes referred to as dimensionless analysis. By using relative values rather than absolute values, units are eliminated and hence, it is well suited for both single- and multi dimensional
