The MultiMorbidity Model for care coordination:Towards clinical decision sup- port that enables general practitioners to optimise treatment plans for multimorbid patients

The MultiMorbidity Model

for care coordination

Towards clinical decision support that enables general practitioners to optimise

treatment plans for multimorbid patients

Anna Beukenhorst

Masterthesis

The MultiMorbidity Model for care coordination.

Towards clinical decision

sup-port that enables general practitioners to optimise treatment plans for multimorbid

patients

Author

Anna Beukenhorst

Bloys van Treslongstraat 42-III

1056XC Amsterdam

Netherlands

Supervisors

Danielle Sent, PhD

Assistant professor

Department Medical Informatics, University of Amsterdam

Georgio Mosis, PhD

Concept Business Architect Philips Healthcare,

Primary and Secondary Care Services, Shanghai

SRP Duration

February 2015 - February 2016

SRP Locations

Philips Innovation Campus Shanghai

No.10, Lane 888 Tianlin Road

Min Hang District, Shanghai

P.R.C. 200233

China

Academic Medical Center/University of Amsterdam

Department of Medical Informatics

Meibergdreef 15

1105 AZ Amsterdam

Netherlands

Summary

Multimorbid patients, people suffering from two or more chronic diseases, have complex treatment plans and face high burden of disease. For healthcare professionals that treat patients with chronic dis-eases, clinical practice guidelines (CPGs) have been developed. These describe best practices in chronic disease care. CPGs have, however, a single-disease focus and tend not to account for the effect of recommendations for one disease on other diseases. Multimorbid patients often receive multiple, disease-specific treatment plans. The result is concurrent execution of treatment recommendations from different CPGs, which may cause conflicts. The position of general practitioners (GPs) is suitable for identifi-cation and reconciliation of these conflicts (‘care coordination’). They should be guided in adapting their workflow and decision-making to multimorbid patients. We therefore investigated the challenges of multimorbidity care and propose a clinical decision support system (CDSS), the MultiMorbidity Model (3M), to facilitate care coordination by GPs. The CDSS consists of a workflow model and framework for application of computerised tools that provide information, alerts and support shared decision-making. Firstly, we investigated CPG-based CDSS. We showed that conversion of CPG recommendations to computer-interpretable rules is complicated, labour-intensive and requires expert knowledge from mul-tiple disciplines (Chapter 3).

In a literature review, we found that existing CDSS for multimorbidity care focus on relatively tangi-ble conflicts, such as drug interactions, and inconsistencies related to scheduling medical tasks (Chapter 4). The scope of these CDSS is currently limited to few CPGs, since acquiring knowledge from CPGs still requires substantial manual effort.

In interviews with experts in multimorbidity care (Chapter 5), these tangible inconsistencies were only briefly mentioned. The interviewed healthcare professionals emphasised the difficulty of handling specific comorbidities, frailty and old age. Patient preferences regarding burden of treatment and com-peting demands of comorbidities, regularly conflict with CPG recommendations. These conflicts are discussed gerontology and family medicine research, but these studies are often descriptive and do not provide specific mechanisms for identification and reconciliation.

We propose the MultiMorbidity Model (3M, Chapter 6), a framework for CDSSs that supports GPs in multimorbidity care. It consists of a workflow model of five steps, promoting a holistic view and facilitating identification and reconciliation of various conflicts: Select, Prioritise, Personalise, Simplify, Formulate. Extension of the workflow model with four tools for computerised decision-support should aid GPs by providing relevant (a) medical knowledge, (b) shared decision making tools, (c) alerts for expected conflicts between CPG recommendations and (d) automatic generation of personalised infor-mation for (health illiterate) patients.

In Chapter 7 we report results of a preliminary evaluation of the 3M. Various healthcare profes-sionals developed treatment plans for hypothetical patients, with or without support of the 3M. The results indicate that the 3M encourages prioritisation and concretisation of treatment recommendations. Regarding the effect of the 3M on conflict identification, incorporation of patient preferences and char-acteristics, we could, however, draw no conclusions.

This thesis is a first step towards CDSSs that facilitates GPs in care coordination for multimorbid patients and provides clinical decision support. Future research should focus on further validation of the model, as well as implementation in practice, possibly in collaboration with the Dutch College for

Samenvatting (Nederlands)

Multimorbide pati¨enten lijden aan twee of meer chronische ziekten. Dagelijks moeten zij veel handelingen verrichten om de symptomen van hun ziekten te bestrijden. Hun artsen gebruiken medische richtlijnen, die, gebaseerd op medisch onderzoek, beschrijven hoe een bepaalde ziekte vastgesteld en behandeld kan worden. Deze richtlijnen zijn meestal echter gemaakt voor ´e´en specifieke ziekte en houden geen rekening met effecten van aanbevelingen op andere ziekten. Daardoor krijgen multimorbide pati¨enten vaak meerdere behandelplannen die onderling con-flicteren. Zij zouden gebaat zijn bij ‘zorgco¨ordinatie’: ´e´en behandelplan dat rekening houdt met al hun ziekten, conflictloos is en therapietrouw bevordert. In Nederland zijn huisartsen in een goede positie om deze zorgco¨ordinatie op zich te nemen. Om hun werkwijze optimaal aan te passen op multimorbide pati¨enten binnen de beperkte tijd van een spreekuur, moeten ze ondersteund worden. In deze scriptie onderzoeken we de uitdagingen van zorgco¨ordinatie en stellen we een beslissingsondersteunend systeem voor dat daarvoor ondersteuning biedt.

Eerst onderzoeken we computersystemen die op basis van richtlijnen behandelconflicten ontdekken en oplossen. De omzetting van richtlijn naar computerregels blijkt ingewikkeld en arbeidsintensief te zijn (Hoofdstuk 3). Bestaande computersystemen zijn gespecialiseerd op relatief tastbare conflicten, zoals bijwerkingen van medicijnen en het plannen van medische taken (Hoofdstuk 4). Ze bevatten slechts enkele richtlijnen, omdat het omzetten van kennis uit richtlijnen naar regels veel handmatig werk vereist.

Artsen, gespecialiseerd op zorg voor multimorbide pati¨enten, noemen deze tastbare con-flicten slechts kort (Hoofdstuk 5). Zij benadrukken andere uitdagingen van zorg voor multi-morbide pati¨enten: een holistische aanpak, fragiliteit, ouderdom, pati¨entvoorkeuren die niet overeenkomen met de richtlijn.

We stellen het Multimorbiditeit Model (3M) voor (Hoofdstuk 6), een beslissingsonderste-unend systeem voor huisartsen die multimorbide pati¨enten behandelen. Het 3M is een werkwijze van vijf stappen voor een holistische aanpak en het oplossen van conflicten: Selecteer, Prioriteer, Personaliseer, Simplificeer, Formuleer. Computerprogramma’s bieden verdere beslissingsonder-steuning: (a) selectie en visualisatie van relevante richtlijnen, (b) hulpmiddelen voor gezamen-lijke besluitvorming met de pati¨ent, (c) waarschuwingen in geval van (verwachte) conflicten en (d) automatische personalisatie van medische informatie voor de (laaggeletterde) pati¨ent.

In Hoofdstuk 7 beschrijven we de resultaten van een eerste evaluatie van het 3M. Verschil-lende artsen ontwikkelden behandelplannen voor virtuele multimorbide pati¨enten, met of zonder 3M. De resultaten tonen dat het 3M−model concrete, geprioritiseerde aanbevelingen stimuleert. We konden nog geen conclusies trekken over het effect van het 3M op het ontdekken van con-flicten, het honoreren van pati¨entvoorkeuren en verwerken van pati¨entkarakteristieken.

Deze thesis is een eerste stap naar beslissingsondersteuning voor huisartsen, om hen te on-dersteunen bij zorgco¨ordinatie voor multimorbide pati¨enten. Vervolgonderzoek zou zich kunnen richten op verdere validatie van het model en de mogelijkheden voor implementatie in de huis-artsenpraktijk, bijvoorbeeld in samenwerking met het Nederlands Huisartsengenootschap.

List of abbreviations and definitions

Multimorbid patients People that are diagnosed with two or more chronic diseases, such as diabetes, cardiovascular disease and arthritis.

Care coordination

Development of one, comprehensive treatment plan that addresses al a multimorbid patient’s conditions and is shared with the patient and all caregivers

3M MultiMorbidity Model

CDSS Clinical decision support system CLP Constraint Logic Programming

COPD Chronic obstructive pulmonary disorder CPG Clinical practice guideline

DM Diabetes mellitus type II FD Foundation doctor (‘basisarts’)

GP General Practitioner

HF Heart Failure

HT Hypertension

MI Medical Intern (‘co-assistent’)

NHG ‘Nederlands Huisartsen Genootschap’ or Dutch College of General Practitioners

OA Osteoarthritis

OM Ontomorph

RBC Rule-based combinations

Contents

Summary vi

Samenvatting (Nederlands) viii

List of abbreviations and definitions ix

Contents x

1 Introduction 1

2 Preliminaries 3

2.1 Multimorbidity and chronic diseases . . . 3

2.1.1 Wagner’s Chronic Care Model . . . 3

2.1.2 Self-management and therapy adherence . . . 4

2.2 Clinical Practice Guidelines (CPG) . . . 5

2.3 Clinical Decision Support Systems (CDSS) . . . 5

2.4 General practitioner: future care coordinator? . . . 6

2.4.1 Individual Care Plan of the NHG . . . 6

3 Case study: computerising guidelines and automatic identification of inconsistencies 7 3.1 Methods . . . 7

3.2 Results . . . 8

3.3 Discussion . . . 8

3.4 Conclusion . . . 9

4 Literature review: CDSS for identification and reconciliation of conflicts between concurrently executed clinical practice guidelines 11 4.1 Methods . . . 11

4.2 Results . . . 11

4.2.1 Ontomorph . . . 12

4.2.2 Constraint Logic Programming . . . 12

4.2.3 Rule-based Combinations . . . 13

4.2.4 Transition-based Medical Recommendations model for Inferring Interactions (TMR4I) . . . 13

4.3 Discussion . . . 13

4.4 Conclusion . . . 15 5 Expert interviews: personalising treatment plan of the multimorbid patient 17

Contents

5.1 Methods . . . 17

5.2 Results . . . 18

5.2.1 Complexity of competing demands and comorbidity . . . 18

5.2.2 Incorporating patient preferences and patient characteristics when managing com-plexity . . . 19

5.3 Discussion . . . 19

5.4 Conclusion . . . 20

6 The MultiMorbidity Model (3M) 21 6.1 MultiMorbidity Model . . . 21 6.2 Workflow . . . 22 6.3 Select . . . 23 6.4 Prioritise . . . 23 6.5 Personalise . . . 24 6.6 Simplify . . . 24 6.7 Formulate . . . 25 6.8 Computerisation of the 3M . . . 26 6.9 Discussion . . . 27

7 Evaluation of the MultiMorbidity Model 29 7.1 Material & methods . . . 29

7.1.1 Cases . . . 29

7.1.2 Questionnaire . . . 30

7.1.3 Analysis . . . 31

7.1.4 Interviews . . . 31

7.2 Results . . . 32

7.2.1 Results per experience level . . . 32

7.2.2 3M versus control group . . . 33

7.2.3 Interview results . . . 33

7.3 Conclusions and discussion . . . 34

8 Discussion 37

9 Acknowledgements 42

References 43

A Literature review: search strategy 51

One

Introduction

The trend of ageing populations coincides with

increased numbers of multimorbid patients [1, 2] − suffering from two or more chronic diseases [3]. With our increasing life expectancies, the preva-lence of chronic disease and multimorbidity is ex-pected to rise [4, 5].

In the Netherlands, the percentage of multimor-bid patients has been estimated at 13% to 29.7% of the complete population and 59.2% to 78% in people aged 80 or older1 _{( [2, 6], 2012 respectively}

1998). In primary care practices, the proportion of multimorbid patients has been rising over the last decennia [5].

Multimorbidity has major consequences: it is associated with disability, functional decline and poor quality of life [7]. In addition, it is associ-ated with increased costs, since multimorbid pa-tients use more healthcare resources: they receive more hospital care, municipal care and outpatient care [8, 9, 10, 11].

Multimorbid patients often face a high burden of treatment: they visit multiple healthcare profes-sionals [9] and are prescribed with multiple drugs [12]. This may result in poor coordination of care, adverse drug events, anxiety and depression [13, 14]. Chronic diseases have no cure, but lifestyle changes, such as healthy diet, physical exercise and intake of medication, may delay disease progres-sion [15]. A major challenge for patients is therapy adherence [16], conducting behaviour that corre-sponds with the recommendations from a health-care provider [17].

Historically, healthcare and medicine have been organised in specialties, with disciplinary special-ists treating single conditions. They mostly use

1_{The prevalence of multimorbidity depends on the}

reg-istries used and on the chronic diseases included in the def-inition.

disease-specific clinical practice guidelines (CPGs) that describe best practices in diagnosing and treat-ing patients. As a result, they may lack recom-mendations for multimorbid patients, specifying ef-fects of (treatment for) one condition on another [14]. Hence, when multiple CPGs are concurrently executed, conflicts between recommendations may arise [18, 19, 20]. These conflicts, such as adverse drug events (ADEs) or contradictory lifestyle rec-ommendations [21, 22], may lead to hospitalisa-tion or even death [18, 23]. Currently, commu-nication between care givers, devision of holistic treatment plans, identification and reconciliation of conflicts and support for self-management is often sub-optimal.

Patients [24], physicians [25, 26, 27, 28, 29] and health policy makers [30, 31, 32] have advocated the need for care coordination, especially in multi-morbidity care. Care coordination facilitates de-livery of the appropriate healthcare services at the right time, in the right order, and in the right set-ting [33]. Within the context of multimorbidity, care coordination refers to development and of one, comprehensive treatment plan that is shared with the patient and all caregivers [24, 25, 26].

Currently, multimorbid patients often receive multiple, possibly conflicting treatment plans from various healthcare professionals [14]; these profes-sionals cannot adequately exchange information. Patients therefore desire support by a care coor-dinator that communicates an individualised care plan, adapted to all conditions [24]. Physicians em-phasize the need to complement specialist care with care coordination: providing and adapting a com-prehensive therapy plan for all conditions [25, 26], intensified collaboration between specialists [27, 28] and attention for the various aspects of wellbeing [29]. Healthcare organisations such as the OECD

1. Introduction

[30] and WHO [31] also acknowledged this problem, and published on the reforms required to “reorient health systems to meet the challenge of multimor-bidity” [34].

General practitioners (GPs) are especially well-suited for care coordination, following their general-ist and patient-centered approach, frequent contact with multimorbid patients and position as gate-keeper for secondary care [26, 35, 36]. Entrusting GPs with this responsibility rather than special-ists would fit the current healthcare reform in the Netherlands, towards decentralisation of care [37]. The government’s policies aim to move long-term care from institutions to the home, and care pro-vision from (expensive) specialists to generalists to volunteer caregivers.

For these GPs, however, multimorbidity care de-manding too: consultation time is limited, as evi-dence on interactions between multiple conditions [38]. Due to limited time, GPs employ an additive-sequential model : they prioritise medical problems and manage these sequentially until the consult is over [39]. Remaining problems are deferred to later appointments, hampering a comprehensive approach. Research indicates that optimal multi-morbidity care requires GPs to adopt a different workflow [40].

To facilitate care coordination, defined as devel-opment of a comprehensive treatment plan for all a multimorbid patient’s conditions, the GP should be supported in decision-making, specifically in the application of multiple disease-specific guide-lines, incorporation of patient preferences and op-timal communication of treatment plan and self-management activities.

This support could be provided by clinical deci-sion support systems (CDSS), information technol-ogy that support healthcare professionals in provid-ing care. Computerised CDSS have the additional advantage of providing support at the point of care and dynamically select and present appropriate in-formation [41].

The goal of this thesis is to design a CDSS con-cept that supports general practitioners in coordi-nating care for multimorbid patients.

In Chapter 2 we provide background information on multimorbidity care and introduce relevant ter-minology. In the first part of this thesis, the chal-lenges of care coordination in multimorbid patients are investigated. The conflicts caused by

concur-rent execution of disease-specific clinical practice guidelines are investigated with a case study (Chap-ter 3) and a li(Chap-terature review (Chap(Chap-ter 4); expert interviews elucidate additional challenges of multi-morbidity care (Chapter 5).

The second part of this thesis describes the MultiMorbidity Model (3M) that enables general practitioners to optimise treatment plans for multi-morbid patients (Chapter 6), and a preliminary evaluation of the model (Chapter 7). The results and implications of the research executed in this thesis, the potential of the MultiMorbidity Model and suggestions for future research are discussed in Chapter 8.

Two

Preliminaries

In this chapter, we provide background information on themes, concepts and terms discussed in this thesis. We define multimorbidity and briefly describe the state of the art in multimorbidity care, specifically Wagner’s chronic care model (2.1.1), self-management (2.1.2) and the position of general practitioners in multimorbidity care in the Netherlands (2.4). In addition, we describe the role of clinical practice guidelines (2.2) and clinical decision support tools (2.3) in chronic disease management. Where applicable, we focus on the situation of multimorbidity (care) in the Netherlands.

2.1

Multimorbidity and chronic

diseases

Multimorbidity is defined as the presence of two or more chronic diseases [3]. Chronic diseases, such as cardiovascular disease, diabetes and COPD, are characterised by a long duration, slow progression and the lack of a cure. Research indicates that chronic diseases often occur in clusters; specific combinations of chronic diseases (e.g. cardiovascu-lar disease and diabetes mellitus) have an especially high prevalence [1, 2, 42, 43]. In elderly patient that suffer from multimorbidity, polypharmacy − prescription with five or more drugs − is common [22].

The prevalence of multimorbidity has been rising with the global life expectancy [5]. Especially in the elderly population, multimorbidity is prevalent: in the Netherlands, two third of elderly with a chronic disease, is multimorbid [2].

Multimorbidity is associated with lower life ex-pectancy and higher use of medical resources [9]. In the Netherlands, multimorbid patients visit the GP multiple times a year, following the need for monitoring chronic conditions [10]. 75% of them also visits one or multiple medical specialists per year [44]. Primary caregivers are responsible for a large part of chronic disease management,

includ-ing most follow-up visits. For COPD and diabetes patients, 70% to 80% of their care is managed by the general practitioner [45]. When an elderly pa-tient shows unexplained symptoms, the papa-tient is referred to a geriatrician. When chronic diseases cause (acute) complications in organ systems, for example in diabetes patients that develop cataract, specialists are involved.

2.1.1

Wagner’s Chronic Care Model

As discussed in the Introduction, multimorbid pa-tients frequently visit various healthcare profession-als and are at risk of treatment conflicts between the treatment plans proposed by these profession-als [20]. In addition, addressing patient values and preferences is pivotal, since this has been shown to impact healthcare decisions and effectivity of longterm care [19].

Many chronic disease management programmes aiming at optimising multimorbidity care, are based on the Chronic Care Model (CCM) of Wag-ner et al. [46]. The pillars of the CCM are re-sources and policies, organisation of healthcare, self-management support, delivery system design, decision support and clinical information systems. The CCM models chronic disease care as an inter-action between a well-informed patient (and

fam-2. Preliminaries

ily/informal caregivers) and a team of healthcare professionals (see Figure 2.1) [47]. The patient is responsible for treatment execution and health-related decisions (Self-management ).

Figure 2.1 – Chronic Care Model

These pillars are the elements required for im-proving chronic care. The CCM describes what re-sources and policies should be available to the com-munity; how decision-support and self-management support and clinical information systems should be implemented; and what health system designs con-tribute to productive chronic disease care.

The CCM is, however, not a replicable interven-tion, but a framework that facilitates generation of general ideas and their translation to local applica-tions [48]. The majority of evaluaapplica-tions of the CCM report an improvement in at least one outcome measure and a reduction in cost [28]. On the other hand, several limitations, such as unfavourable fi-nancial incentives and costs of reorganisation ham-per implementation.

2.1.2

Self-management and therapy

adherence

One of the pillars of the CCM, and one of the major challenges in multimorbid patients’ daily lives is self-management: a patient’s health-related decisions and behaviours [49]. Adequate self-management reduces chronic disease symptoms and delays disease progression [50]. This requires be-haviour change, which makes it an especially diffi-cult part of multimorbidity care. Therapy adher-ence is defined as the extent to which a person’s

be-haviour corresponds with agreed recommendations from a healthcare provider [17].

Research indicates that people often do not ad-here to pharmacological and non-pharmalogical treatment recommendations. When people need to change multiple lifestyle factors, adherence is especially low [16].There are various views − not mutually exclusive − on determinants of therapy adherence. The WHO describes the three pillars of long-term therapy adherence: information, motiva-tion and behavioural skills [51]. This is also known as Fisher & Fisher’s Information Motivation Be-havioral Skills model.

Others distinguish between factors that are re-lated to patients (lack of involvement in treatment process), factors that are related to physicians (in-effective communication) and factors that are re-lated to the healthcare system (limited time for consultation) [52].

The NHG (Dutch College of General Practition-ers − Nederlands Huisartsengenootschap) states that adequate self-management is associated with various patient characteristics: problem-solving skills, decisiveness, resolution, self-tailoring, usage of appropriate other (information) resources and collaboration with healthcare professional [15].

Research subjects report a variety of barriers to therapy adherence, such as lack of time, coexist-ing diseases and weather conditions. Excessive us-age of cars, consumption of high-calorie food, fre-quent social gatherings and stress [53].These bar-riers are sometimes associated with cultural and demographic factors. Other determinants of ther-apy adherence are patient characteristics (socio-economic status, health literacy, health skills, edu-cation level) and other diagnosed diseases (depres-sion, cognitive decline) also influence likelihood of therapy adherence [16].

A major effector of low treatment adherence is health illiteracy. Health illiterate people have a limited understanding of health and limited abil-ity to take responsibilabil-ity for their health. Health illiterate patients often have [15]:

− Limited understanding of how the body works − Deviant views on obesity, social problems or

healthy diet

− Organisational or financial impediments − Lack of social support, because of norms

2.2. Clinical Practice Guidelines (CPG)

Health literacy is associated with poorer health-related knowledge and comprehension [54]. In ad-dition, it leads to poorer health-related outcomes, among which adequate intake of medication and prpoer interpretation of health massages [55]. Cur-rent primary care CPGs state that GPs should give health illiterate patients at most three action-able, specific and short recommendations without metaphors; show illustrations [15].

2.2

Clinical Practice Guidelines

(CPG)

Clinical Practice Guidelines (CPGs) help health-care professionals gather, evaluate and imple-ment best practices in medical care [56]. CPGs aim reduce inappropriate variation among health-care professionals and increase quality and cost-effectiveness of care. They often consist of sections on epidimiology, diagnosis, treatment recommenda-tions and underlying scientific evidence [57]. Due to a variety of barriers, such as lack of knowledge and lack of agreement with recommendations, health-care professionals do not always adhere to CPGs [58, 59, 60]. The relation between guideline ad-herence and quality of care is unequivocal: effects of guideline adherence on patient outcomes vary greatly among studies [61, 62]. The Dutch body for guideline development in primary care states that ”guidelines support doctors in decision making, but are not binding (...) deviation from guidelines is often necessary.” [63].

Two characteristics hamper application of CPGs on multimorbid patients. Firstly, CPGs might not be optimally applicable to these patients. Part of our knowledge on best practices is generated by randomised clinical trials (RCTs), the golden stan-dard in healthcare research. Their results, how-ever, not be generalizable to all patient groups [64]. Limitations of RCTs include a focus on relatively short-term periods, exclusion of multimorbid pa-tients and underrepresentation of women and el-derly [65, 66].

Secondly, most CPGs are disease-specific. When various healthcare professionals provide multimor-bid patients with treatment recommendations of multiple disease-specific CPGs this can lead to problems. These may include adverse drug events, increased treatment complexity, burden of disease

and cost of treatment [14, 18, 19].

In the Netherlands, CPGs for primary care are developed by the Nederlands Huisartsen Genootschap (NHG), the Dutch College of Gen-eral Practitioners. Most of their CPGs are disease-specific1, but they recently developed several CPGs with a broader scope, such as the CPG Polyphar-macy in elderly [12].

2.3

Clinical Decision Support

Systems (CDSS)

To promote adherence to CPGs and support healthcare professionals in providing care accord-ing to best practices, clinical decision support sys-tems (CDSS) are often implemented. These may present relevant medical knowledge, guide work-flow, answer questions, retrieve information, per-form calculations or group elements [41, p. 32]. CDSS have many potential and demonstrated ben-efits in preventing medical errors, reinforcing best practices and increasing time-efficiency [41, p. 59-69].

In this thesis, CDSS refers to computerised clin-ical decision support systems, for which informa-tion and communicainforma-tion technologies are used. A computer-based CDSS usually consists of a knowl-edge base, consisting of facts in the form of rules (‘knowledge-based’) or algorithms and relations (‘datadriven’) and an execution engine that applies knowledge on the data input (e.g. the patient’s electronic medical record) [67].

Various technologies are used in computerised CDSS, such as machine-learning, ontologies and de-cision trees. Some CDSS can learn from data, but currently, most CDSS are rule-based [41, p. 128]. It is suggested that this is caused by limited availabil-ity of structured data and limited dissemination of algorithms for pattern discovery among clinicians. In addition, rule-based CDSS can explain the rea-soning behind specific advice.

In clinical practice, many CDSS have been de-veloped to promote guideline adherence and pro-vide healthcare professionals with up-to-date guide-line information. CPGs are then used as source of knowledge for rule-based CDSS. The Dutch Col-lege of General Practitioners developed NHGDoc,

1_{Disease-specific CPGs of the NHG often do contain}

2. Preliminaries

a guideline-based expert system that facilitates guideline adherence and registration.

For usage in CDSS, guideline information needs to be converted to computer-interpretable rules. Several ontologies have been developed to struc-ture medical concepts and procedures for genera-tion of computer-interpretable rules. Examples of frequently used ontologies include OWL, GASTON and GLIF [68].

A variety of techniques, such as semantic nets, decision rules and expressions or workflow mod-els, is available [41, p. 267 and p. 281]. Genera-tion of computer-interpretable rules requires major manual effort. The Logical Elements Rule Method specifies seven steps for a usable and reliable trans-formation of clinical rules.

In his bible for CDSS developers, Greenes [41, p. 122] states that development of CDSS for chronic disease management is especially challenging, since the number of exceptions and alternatives is very high. Reviews of ontologies describe that the na-ture of chronic disease management often ham-per computability. Uncertainty, recommendations over multiple encounters and representing knowl-edge from multiple specialties hamper the com-putability of chronic disease CPGs [69].

2.4

General practitioner: future

care coordinator?

In the Netherlands, multimorbid patients fre-quently visit the GP: for check-up visits, medi-cation refills and handling changes in health sta-tus [44]. Building a personal relationship with the patient, and acknowledging a patient’s prefer-ences are, especially in patients with limited life expectancy, one of the pillars of GP care [37]. GPs receive information from specialists that treat their patients, and may mobilise home-care ser-vices, such as the Thrombosis service, social work-ers or visiting nurses.

The NHG states that the chance of succesful self-management is highest when a patient commits to a plan that has been developed together with the GP and consists of concrete statements [15]. In order to increase motivation to adhere to recommendations, the NHG advices that:

− Advice should be practical and executable within a patient’s (financial) abilities

− Recommendations should correspond with prob-lems that the patient experiences and goals the patient wants to achieve

− Health goals should be formulated with the pa-tient

− Advice should be adjusted to the patient’s pri-orities and way of life

The NHG promotes shared decision-making, a strategy for making reference-sensitive healthcare decisions [70], especially in chronic disease man-agement [71]. In chronic disease management, preference-sensitive decisions need to be made. A variety of treatments may be applicable on a par-ticular patient, resulting in different outcomes.

2.4.1

Individual Care Plan of the

NHG

The NHG states that patients with a chronic dis-ease should receive different care than GPs tradi-tionally deliver [45]. A generalist approach with attention for a patient’s personal circumstances, guidance in the broader care ecosystem and col-laboration with other healthcare professionals are of increased importance.

The NHG is currently working on the ”Indi-vidueel Zorgplan” (individual care plan, ICP [15]). The ICP is a component of GP information systems for dynamic registration of patients with (multiple) chronic diseases. In the ICP, GPs will document: ∗ The activities that are needed to achieve the

pa-tient’s goals;

∗ The person that executes the activities (patient, volunteer caregiver, healthcare professional); ∗ What healthcare professional monitors the

activ-ities;

∗ When the activities should be executed;

∗ How and when the action plan will be evaluated. This reflects acknowledgement of the need for care coordination for multimorbid patients by the NHG. With these developments in Dutch primary care in mind, we will investigate conflicts in multi-morbidity care and functionalities of CDSS re-quired for supporting GPs in identifying and rec-onciling these problems.

Three

Case study: computerising guidelines and automatic

identification of inconsistencies

Clinical practice guidelines (CPGs) describe best practices in clinical care for a specific dis-ease. For multimorbid patients, however, concurrent execution of multiple disease-specific guidelines may cause conflicts [14]. Computerised clinical decision support systems based on multiple chronic disease guidelines could alert care coordinators to these conflicts, improving appropriate adherence to and deviation from CPGs [67]. Here fore, recommendations from chronic disease CPGs need to be converted to computer-interpretable rules.

In this chapter, CPG recommendations for three chronic disease guidelines are converted to computer-interpretable rules. This should result in an overview of problems that arise when developing a computerised tool that supports care coordinators in identification and reconcili-ation of conflicts between concurrently executed chronic disease guidelines.

3.1

Methods

For the case study, we studied a realistic hypotheti-cal patient suffering from arthritis (OA), hyperten-sion (HT) and diabetes mellitus type 2 (DM). This combination of conditions has been frequently used in studies on drug interactions [18], since the combi-nation of conditions is prevalent among multimor-bid patients [2].

Firstly, we identified recommendations from the sections ’Recommendations regarding Treat-ment’ of relevant CPGs (Guideline arthritis [72], Guideline cardiovascular risk management [73] and Guideline diabetes mellitus type II [74], all devel-oped by the Dutch Association for General Prac-titioners) that refer to incompatible drug prescrip-tions and contradictory advices that were identified previously [18].

Secondly, we programmed drug conflicts in JBOSS Drools, an open source Java-based ruleR

engine. In Drools, relational rules can be declared in easily interpretable syntax. It has been previ-ously used for development of CDSS [75, 76]. The

objective was to develop a program that contains CPG knowledge and gives alerts when patient infor-mation indicates that conflicting CPG recommen-dations are applicable.

To be able to use CPG recommendations, we converted these to Drools rules. CPGs distin-guish between clinical context (e.g. [presence of] osteoarthritis), decision nodes (e.g. contraindica-tions for NSAID?) and action steps (e.g. pre-scribe ibuprofen) [68]. Representing CPGs as flowcharts from these nodes is therefore a com-mon approach for computerising and analysing CPGs [77]. The pathways represented in the flowchart were converted to IF-THEN statements (”IF osteoarthritis AND NOT contraindications for NSAIDs DO prescribe ibuprofen”) were pro-grammed in JBOSS Drools. The resulting appli-R

cation was tested by entering the three diagnoses of the hypothetical patient and having the application select and execute applicable rules.

3. Case study: computerising guidelines and automatic identification of inconsistencies

3.2

Results

When converting CPGs to Drools rules it was dif-ficult to distinguish between clinical context, deci-sion node and action step, because a decideci-sion node is sometimes also clinical context, or vice versa.

In the CPGs, potential conflicts between treat-ment recommendations were treat-mentioned explicitly, implicitly but specific or implicitly and general. Several examples of these conflicts are shown in Ta-ble 3.1.

Table 3.1 – Three types of conflicts and examples from the OA, DM and HT guidelines

• Explicit

”If choosing one kind of NSAID, please pay attention to comorbidities (cardiovascular, gastro-intestinal), side effects and interactions (acetylsalicylacid) and previous reactions on NSAIDs. Sometimes it is recommended to prescribe stomach-protecting medication.” [72]

• Implicit and specific effects of comorbidities on blood pressure [73]: Table of ’Medication and substances that increase blood pressure’ with drugs and nutrients that increase blood pressure (e.g. NSAIDs, oral anticonceptives, alcohol) ”Elderly patients and patients with diabetes or kidney failure are more sensitive to

hypertensive effects of NSAIDs.” • Implicit and general statements on inconsistencies [72, 73, 74]:

”in case of contra-indications or side effects” ”take contra-indications into account when choosing...”

We found that, without expert clinical knowl-edge, it was impossible to derive inconsistencies in the form of ”Combination drug A and drug B lead to interaction X”. Wordings such as ”please pay attention to” and general terms such as contra-indications, side-effects and comorbidities could not be specified without help of a clinician.

For example, when combining the OA guideline and DM guideline, an inconsistency occurs. The OA guideline recommends avoiding gastrointestinal bleedings, hence not administering aspirin, while the DM guideline recommends avoiding thrombi by administering aspirin.

This particular inconsistency is hard to derive based on the Dutch guidelines. The DM guide-line states: ”Diabetic patients with cardiovascular comorbidity that are not prescribed with coumarin derivatives, receive once daily 80 mg acetylsalicylic acid” [74]. The OA guideline states: ”Indepen-dent of treatment consequences of the diagnosis, the GP can treat arthritis symptoms with NSAIDs (...) Pay attention to comorbidities (cardiovas-cular, gastrointestinal), side effects and interac-tions (acetylsalicylic acid) (...)”. These statements will lead to a conflict, but it is not explicitly de-scribed. When extending the knowledge base with the guideline Polypharmacy in Elderly [12], the conflict could be easier derived, since it states:”Add a proton pump inhibitor when (...) patient is 60-70 and uses acetylsalicylic acid and a NSAID; when a patient is older than 70 and uses NSAID; when a patient is older than 80; (...)”.

When computerising CPG recommendations with Drools, these ambiguous concepts impeded generating computer-interpretable rules.

In addition, the lack of a beforehand defined on-tology for modeling CPG concepts, hindered com-puterising CPG recommendations. In Drools it is necessary appoint concepts to a class. With trial and error, classes were defined and redefined (such as ’disease’, ’drug’, ’care action’), because assump-tions for the first recommendation did not hold for others. We did not succeed in reasoning developing a consistent ontology for appointing care actions, intentions and recommendations to classes in a way that facilitated reasoning.

3.3

Discussion

When identifying applicable CPG recommenda-tions for a hypothetical patient with a common combination of three chronic diseases, several con-flicts arise. Side-effects of drug concon-flicts may have serious implications on a multimorbid patient’s other conditions.

We found that these conflicts are often only im-plicitly described. Identifying and reconciling a conflict solely based on CPG information is not pos-sible because of ambiguous terminology. Upon re-quest, medical experts indicated that clarification of CPGs is not needed for them, since they have additional knowledge from own experience,

text-3.4. Conclusion

books and reference documents. We found, how-ever, that the ambiguities and implicit phrasing do hamper computerisation of recommendations by people without a degree in medicine.

An exception is the Multidisciplinary Guideline Polypharmacy in Elderly patients, applicable on “elderly multimorbid patients that are prescribed with more than five drugs and have at least one of five risk factors for drug conflicts” [78] devel-oped by the Dutch college of GPs. In this CPG, frequently prescribed medications, their contra-indications, side-effects, criteria for prescription and replacement, potential inappropriate medica-tions and potential errors of omissions are specif-ically described. This CPG does introduce other ambiguities, such as ”signals of reduced adherence to therapy” and arbitrary age limits.

Conversion of CPGs into flowcharts and then in computer-interpretable rules proved to be very dif-ficult. We found that the following is needed: ∗ Clinical expert(s) that clarify ambiguous

termi-nology

∗ Clinical expert(s) that specify workflow

∗ An ontology to represent medical knowledge and medical procedures

∗ Informatics or knowledge engineering expert(s) that understand reasoning by computers. Our findings are echoed in other publications. Jafarpour et al. [79] state that paper-based CPGs are often complex and difficult to understand for non-clinicians. Articles that provide methods to convert CPGs to computer-interpretable guide-lines, also require a combination of clinical experts and knowledge modelling or medical informatics ex-perts [80, 81]. Arts et al. [82] found, when devel-oping CDSS on stroke prevention, that expert pan-els were necessary to specify ambiguous terminol-ogy. They suggested, in unison with bodies that develop methods to appraise quality and imple-mentability of guidelines (Appraisal of Guidelines, Research and Evaluation assessment [83]; Guide-Line Implementability Appraisal [84]), that these expert panels should clarify terminology when cre-ating CPGs.

In an interview, the guideline development ex-pert of the NHG stated, however, that his organi-sation’s responsibility is in offering information to physicians, not in developing CPGs for computeri-sation. Past collaborations with ICT companies to

computerise CPGs were unsuccessful:“In the end, flowcharts never work.” The interviewee conveyed that the subject of merging therapy plans for multi-morbid patients and reconciling inconsistencies is very important to the organisation for Dutch GPs. They are researching the potential for CDSS and technologies for this purpose, but this is expected to be a long-term project, with major challenges concerning technology, privacy and interoperabil-ity.

3.4

Conclusion

In conclusion, we have shown that development of a rule-based execution engine that identifies conflicts between recommendations of three spe-cific chronic disease CPGs, is complicated, labour-intensive and requires expert knowledge from mul-tiple disciplines.

To assess the potential of decision-support for conflict identification and reconciliation, we will ex-amine current developments in CPG-based CDSS for multimorbidity care in Chapter 4. We will anal-yse methodologies and evaluate strengths and lim-itations for supporting GPs in their role of care coordinator.

Four

Literature review: CDSS for identification and

reconciliation of conflicts between concurrently

executed clinical practice guidelines

When treating multimorbid patients, one or more healthcare practitioners retrieve recommen-dations from multiple chronic disease clinical practice guidelines (CPGs). Combining these recommendations may introduce inconsistencies, for example when a drug, prescribed for one condition, has an adverse effect on another condition [18]. With the growing number of multi-morbid patients, identification and reconciliation of these inconsistencies becomes increasingly important [85]. Computerised clinical decision support systems (CDSS) have been used to alert healthcare professionals to adverse drug events at the point of care [69]. Similar alerting sys-tems could be designed for identification and reconciliation of conflicts in multimorbidity care. We execute a literature review to examine existing CDSS for identification and reconciliation of conflicts between concurrently executed CPGs. We analyse the methodology of development of four CDSS, their strengths and limitations.

4.1

Methods

A literature review in three databases was per-formed to identify publications on CDSS for identi-fication and reconciliation of conflicts arising from concurrent execution of CPGs. The search strategy is described in detail in Appendix A.

204 articles fulfilled the search criteria. Selection on relevance was carried out as follows. First, ti-tles and abstracts were screened; articles that did not focus on concurrent execution of multiple CPGs were excluded. We examined full text of the re-maining 20 publications. We excluded 12 articles: 2 focused on other challenges in multimorbidity care (provision of lifestyle recommendations respectively scheduling of tests for chronic disease patients); 6 focused role of ontologies and knowledge represen-tation in CDSS development; 2 articles were re-views of CDSS systems; 2 articles discussed devel-opment of a patented CDSS system, and therefore

did not specify techniques used for development. The remaining 8 articles described four CDSS sys-tems with focus on conflicts in treatment plans for multimorbid patients.

4.2

Results

Four CDSS systems met our inclusion criteria: On-toMorph, constraint logic programming, the TM4I model and rule-based combinations. The major ob-jective of all these approaches is to propose a (part of a) treatment plan for multimorbid patients. We will describe the four CDSS in more detail.

A multitude of ontologies has been developed that offer a refined and comprehensive method of CPG representation. They represent both declar-ative knowledge (medical statements and proposi-tions) and procedural knowledge (workflow struc-tures and actions).

4. Literature review: CDSS for identification and reconciliation of conflicts between concurrently executed clinical practice guidelines

4.2.1

Ontomorph

The objective of the Ontomorph approach is to pro-pose a treatment plan, consisting of several tasks, that do not conflict and that are time- and resource-efficient [79].

Firstly, clinical tasks are extracted from two CPGs and converted to computer-interpretable rules with an OWL-based CPG ontology. An on-tology is a methodology for CPG representation. It consists of rules to represent declarative knowl-edge (medical statements and propositions) and procedural knowledge (workflow structures and ac-tions). OWL (Web Ontology Language) is a W3C-standard for web ontologies, for which CPG con-cepts are converted to RDF triples and XML files [86].

Secondly, clinicians are interviewed on ‘merg-ing criteria’: rules for concurrent execution of tasks from multiple guidelines. Clinicians identified 4 types of workflow constraints, operational con-straints, temporal constraints and 2 types of medi-cal constraints, listed in Table 4.1.

Table 4.1 – Constraints between task TAand TB of

con-currently executed guidelines A and B (Jafarpour [79])

Category Example Workflow Constraint

Simultaneous Action

Constraint TAsimultaneous with TB

Precedence Constraint TAbefore TB

Identical Action Constraint Results T_Areused instead of T_B Combination Constraint New task TX instead of TA+TB

Operational Constraint Institution-specific Temporal Constraint TAone week after TB

Medical Constraint

Task Substitute TX has same result as TAand does not conflict with TB

Use Results Constraint Results TAexpire after one month

Workflow constraints are rules that specify whether a tasks should be combined with, substi-tuted by, executed simultaneously with or executed before or after a task from another guideline. Op-erational constraints refer to limitations for com-bining tasks at a specific medical institute; tempo-ral constraints specify the time required between the first and second task of two guidelines. Med-ical constraints are divided in Task Substitutes (a substitute for a task of guideline A that does not conflict with a task of guideline B) and use results constraints (rule that specifies expiry date of task results).

Thirdly, an OWL-based CPG execution engine is build, specifying rules written in Semantic Web

Rule Language (SWRL), based on the clinicians’ ‘merging criteria’ [79]. If a constraint is applicable on two CPG tasks, the engine will propose the tasks or scheduling of tasks that fulfils the constraint. When combining the guidelines for duodenal ulcer (DU) and transcient ischemic attack (TIA), for ex-ample, a Task Substitute is defined. If the task ‘give aspirin’ (TIA) is combined with the task ‘don’t give aspirin’ (DU), the first task is substituted by ‘don’t give aspirin’ [79].

4.2.2

Constraint Logic Programming

Constraint Logic Programming (CLP) is an ap-proach that combines logic programming with con-straint satisfaction problems [87].

Firstly, two CPGs are converted to a logical model [85]. The CPGs are represented as an ‘Activ-ity Graph’ (AG), a flowchart that specifies context (Patient is diagnosed with TIA), decision nodes (Are neurological symptoms resolved? ) and actions (Administer aspirin). The AG is then converted to an ‘Enhanced Path Table’ (EPT) with Boolean variables for all decisions and tasks. CPG recom-mendations and workflow are formulated as con-straints. Conflicts between the two CPGs are sepa-rately defined as constraints and stored in a Knowl-edge Base (KB) for the pair of guidelines.

Secondly, a mitigation algorithm is developed, consisting of rules that identify a conflict (‘inter-action operators’) and rules that specify possible guideline modifications (‘revision operators’). Re-vision operators only concern CPG modifications, since patient variables cannot be altered. The algo-rithm focuses on adverse interactions only, follow-ing the Boolean nature of CLP variables (compara-ble with Zamborlini’s ‘Contradiction to same care action’ in Table 4.2 ).

The execution engine receives Boolean patient variables (Are neurological symptoms resolved?, value true or false ), Boolean CPG variables (Administer aspirin, value true or false ) and a set of constraints defined by the CPG models. The engine detects ‘points of contention’, violations of the constraints, with the interaction operators. Subsequently, solutions to the conflicts are deter-mined with revision operators.

CLP could be used as alerting tool for physicians that concurrently apply multiple CPGs. The out-put would be a warning about possible adverse

in-4.3. Discussion

teractions and suggestions for clinically appropriate solutions [87].

4.2.3

Rule-based Combinations

The rule-based combinations method has been de-veloped for identification and reconciliation of drug conflicts between recommendations of two concur-rently executed CPGs [88].

The first step is knowledge acquisition: a health-care professional and two knowledge engineers list all possible drug conflicts that can occur when com-bining two specific CPGs.

Then the drug conflicts (excluding those that are very complex or related to drug dosages) are con-verted to IF/THEN-statements.

Execution software combines the treatment rec-ommendations of two CPGs, following IF/THEN-statements in case of conflicts. Output is a final treatment plan without interaction, consisting of ATC-codes of drugs that should be prescribed. Al-though the software can only combine CPGs pair-wise, a final treatment plan based on two CPGs could again be combined pairwise with a new CPG.

4.2.4

Transition-based Medical

Recommendations model for

Inferring Interactions (TMR4I)

The Transition-based Medical Recommendations model has been developed for the automatic in-ference of interactions between recommendations (TMR4I, [89]). Its scope is currently limited to conflicts between CPG statements on drug pre-scription, but the authors it could be used for non-pharmacological treatment recommendations as well.

The developers firstly defined meta-rules for identification and reconciliation of three categories of drug conflicts using SPARQL queries (SPARQL is a W3C-standard for semantic queries). The meta-rules define how a conflict is identified, and how drugs with similar effects but without con-flicts can be selected from CPG-knowledge. The categories of conflicts within CPGs are repetition interactions, contradiction interactions and alter-native interactions, shown in Table 4.2. In addi-tion, the TM4I accounts for external interactions, retrieved from DrugBank, an online database with drug-drug interactions.

Table 4.2 – Interactions between drug prescription rec-ommendations of concurrently executed guidelines

(Zam-borlini [89])

Category Example

Repitition Administer aspirin in multiple CPGs Contradiction

to same care action Administer aspirin/stop aspirin

to similar transitions Lower bloodpressure/avoid lowering blood pressure

to inverse transitions Lower bloodpressure/increase bloodpressure

Alternative

to similar transitions Ibuprofen, aspirin and naproxen for inflammation

to inverse transitions No aspirin to avoid bleeding/PPI to avoid bleeding

External

Incompatible drugs Retrieved from Drugbank Alternative drugs Retrieved from Drugbank

Subsequently, CPG recommendations are de-scribed as care actions (e.g. ‘Administer aspirin’), situations (‘Medium risk cardiovascular event’) and transitions (e.g. ‘Medium risk cardiovascular event → Low risk cardiovascular event’). Recommenda-tions conflict when care acRecommenda-tions are contradictory, or when similar or inverse transitions are intended by multiple care actions. Although this categori-sation is made for medication recommendations, it might also be appliciable on non-medication recom-mendations [90].

A web-tool for execution of guidelines was de-veloped. In this tool, clinicians firstly enter all guideline recommendations applicable on a patient. The execution engine creates a new, merged guide-line with all recommendations. With the SPARQL meta-rules, interactions are identified and classi-fied. Then, the engine consults the alternative rec-ommendations, in order to choose solutions for the conflicts. Finally, a list of conflicts and recom-mended solutions is presented to the clinician.

4.3

Discussion

Three CDSS aim at identification of drug con-flicts, while the Ontomorph approach focuses on the scheduling of CPG tasks, including but not limited to drug prescriptions.

When comparing the development method-ology of the discussed CDSS, several steps can be identified, not always executed (in the same order). These steps are depicted in Table 4.3. We will now discuss the steps and the different techniques used for these steps. For readability we will abbreviate the CDSS methodologies’ names:

4. Literature review: CDSS for identification and reconciliation of conflicts between concurrently executed clinical practice guidelines

OM forOntomorph, CLP for Constraint Logic Programming, RBC for Rule-Based Combinations and TM4I for Transition-based Medical Recom-mendations model for inferring interactions.

Figure 4.1 – Steps in development of CDSS for identifi-cation and reconciliation of conflicts between concurrently executed CPGs (Abbreviations: OM = Ontomorph; CLP = Constraint Logic Programming; RBC = Rule-Based Combinations; TM4I = Transition-based Medical

Recom-mendations)

CPG conversion

Two methods start with conversion of CPGs to computer-interpretable rules. The RBC skips this step, while the TM4I incorporates it as second step. The techniques used in this step are summarised in Table 4.3.

Identification of conflicts, conflict constraints and knowledge base

The second step for the OM and CLP is identifica-tion of conflicts. For conflict identificaidentifica-tion in the OM, three general practitioners with experience in multimorbidity care were consulted [86]. The med-ical expert from the CLP team identified conflicts for their approach.

The RBC does not use complete CPGs, but only conflicts; this is therefore the first step of this methodology. Conflicts were identified by one experienced healthcare professional.

Table 4.3 – Techniques used for conversion of CPGs to computer-interpretable rules

Representation Concepts OM Own CPG ontology Tasks

Concepts Decisions CLP Flowchart Actions

Workflow constraints Context Decisions

RBC N.A. N.A.

TM4I UML Care actions

First order logic Situations Transitions

Hereafter, the conflicts have to be converted to computer-interpretable rules (see Table 4.4). Three approaches define conflicts the conflicts that may occur when two specific CPGs are executed, as constraints. These constraints are stored in a knowledge base. This means that, for every pair of CPGs that is included in the CDSS, conflicts have to be manually defined by medical experts and constraints have to be manually defined by knowledge engineers.

Table 4.4 – Techniques used for conversion of conflicts to computer-interpretable rules Technique OM Merge criteria CLP Constraint logic RBC IF/THEN-statements TM4I N.A.

Meta-rules as substitute for manual effort

The TM4I model abstains from the three before-mentioned steps. Instead, they start the CDSS development process with defining meta-rules for general, guideline-independent conflicts (repiti-tions, contradictions and alternatives, see Table 4.2). Subsequently, drug recommendations of CPGs are converted to computer-interpretable rules, using the syntax of the meta-rules. Con-flicts between CPG recommendations can thus be identified with the meta-rules; an external, manually filled knowledge base is not neccesary.

4.4. Conclusion

It is not reported whether medical professionals are required for identification of conflicts between guidelines. The articles suggest that the authors retrieved conflicts from case studies reported in other scientific articles [89, 90].

Execution and output

For all CDSS, execution engines have been devel-oped that check applicable CPG recommendations against conflict constraints (OM, CLP, RBC) or meta-rules (TM4I). OM and CLP can only be used to combine exactly two CPGs. RBC also performs pairwise combination of CPGs, but the result of one execution can be used as input for a new pairwise execution. The TM4I approach can check unlim-ited CPGs for conflicts; currently, five CPGs have been converted to computer-interpretable format.

The output differs per CDSS (see Table 4.5: some have been developed as alerting systems and produce lists of conflicts, while others generate treatment plans.

Table 4.5 – Techniques used for conversion of conflicts to computer-interpretable rules

Technique

OM Due CPG tasks and schedule CLP Conflict alerts (and solution) RBC Complete drug prescription list TM4I Conflict alerts and solution

Implications This literature review shows that con-version of CPGs to computer-interpretable rules is an essential step in development for CDSS that identify and reconcile CPG conflicts. For this con-version, various techniques and standards can be used.

For the identification of conflicts by the CDSS, either meta-rules or conflict constraints need to be formulated. When using conflict constraints, CPG conflicts identified by medical experts are ‘hard-coded’ and stored in a knowledge base. In case of changes to a guideline, or a new combination of diseases, knowledge has to be manually acquired and added. Usage of meta-rules have an advantage. Meta-rules can be reused, because they are appli-cable on many CPGs. Secondly, conflicts do not need to be manually identified per CPG, because

they can be automatically derived from the CPG representation.

When evaluating the potential of these CDSS for implementation in primary care practices, it is im-portant to acknowledge several limitations of the discussed approaches.

Firstly, their scope is limited, both in terms of supported CPGs and identifiable conflicts. Three CDSS can only combine two CPGs; only the TM4I model can be used for combination of more CPGs, although currently only two combinations of two re-spectively three chronic diseases are implemented. Extending the number of supported CPGs requires substantial manual effort; for identification of flicts, conversion of CPGs, and formulation of con-straints, experts in multimorbidity and/or knowl-edge engineering are required. Acquiring knowl-edge on conflicts for all possible combinations of diseases is time-consuming, as is updating CPG knowledge − CPGs are frequently revised. For the TM4I, that has the advantage reusable meta-rules (contrary to the guideline-dependent conflict con-straints of other approaches), the bottleneck will be in converting CPGs to computer-interpretable rules.

In addition, the CDSS focus only on drug con-flicts that do not regard dosage and timing and, in case of the OM, task scheduling. The authors of CLP and TM4I state that they aim to expand their approaches to identify and reconcile more con-flicts. Currently, another disadvantage is the focus on ‘internal conflicts’, i.e. conflicts among CPG recommendations [89]. The CDSS use hardly any patient information, while research indicates that conflicts can also arise from non-pharmacological recommendations and external information such as a patient’s diet or life expectancy [91]. In addi-tion, healthcare professionals have suggested that, in multimorbid patients, conflicts can also arise from poor communication between involved care-givers, or affect more abstract variables such as wellbeing [27, 28, 29].

4.4

Conclusion

We discussed four existing CDSS that support healthcare professionals in identification and recon-ciliation of conflicts between concurrently executed CPGs. These, especially the TM4I, are promising

4. Literature review: CDSS for identification and reconciliation of conflicts between concurrently executed clinical practice guidelines

first steps towards (semi-automatic) identification and reconciliation of relatively tangible conflicts, but not mature enough for implementation in clin-ical practice [77].

Care coordination for multimorbidity care will, however, require a broader approach, focussing on the complete range of potential conflicts. Devel-opment of a comprehensive treatment plan for all a patient’s conditions, includes addressing non-pharmacological recommendations. For multimor-bid patients, who suffer from chronic diseases, an essential part of that treatment plan consists of recommended lifestyle changes. Research indicates that, even though adequate self-management im-proves symptoms and prognosis, behaviour change is often either not achieved, or not maintained [92]. Various theoretical frameworks indicate that an im-portant factor in persuading someone to perform a behaviour, is simplicity of the desired behaviour [93]: it should fit into a person’s time, budget, so-cial norms, daily routines and required physical and mental effort should be minimised. As multimorbi-dity is associated with functional decline, reduced quality of life and hospitalisation, it might increase mental and physical effort required for behaviour change. In addition, the combination of various independently developed treatment plans may con-sume high amounts of time and budget, hindering a patient’s daily routines. Hence, a comprehensive treatment plan that addresses the user characteris-tics (time, budget, social norms and daily routines) and reduces mental and physical effort, may influ-ence success of self-management.

We therefore will firstly perform expert inter-views to identify sources of conflicts including, but not limited to, pharmacological recommendations. These may be conflicts that arise when combining two CPGs, when combining more than two CPGs; conflicts that arise because CPGs are not optimally adapted to a specific patient; or problems caused by cumulative burden of disease, cumulative burden of treatment.

Five

Expert interviews: personalising treatment plan of the

multimorbid patient

In the previous chapter, we described existing CDSS that alert healthcare professionals to conflicts between multiple chronic disease clinical practice guidelines (CPGs). The scope of these CDSS was limited to relatively tangible conflicts, such as drug conflicts and scheduling conflicts, only among CPGs. Previous research indicates, however, that GPs tend not to adhere to CPGs, because of limited applicability on multimorbid patients and limited consideration of patient preferences and personalisation [63, 94]. Hence, conflicts can also be caused by CPGs when these do not adequately acknowledge individual patient characteristics, need for personalisation and incorporation of patient preferences.

To gain an overview of challenges healthcare professionals encounter in medical practice, it is therefore necessary to peruse other sources. We perform interviews with three healthcare professionals specialised in multimorbidity care to identify other categories of conflicts.

5.1

Methods

To acquire information on inconsistencies in multi-morbidity care, three medical professionals were in-terviewed. Participants were informed about the purpose and content of the interview and agreed to participate by email. All interviews were taken orally: two via telephone and one face-to-face.

Participants were:

(∗) The head of the guideline development group responsible for development of all CPGs for pri-mary care in the Netherlands

(∗) A social-psychiatric nurse working at the geri-atrics department of a national Dutch mental health organization (’GGZ’; secondary care) (∗) A medical specialist and researcher in internal

medicine and geriatrics, working in the Spaarne Ziekenhuis

All participants were experienced clinicians, work-ing at least 20 years in clinical practice. The pur-pose of the interviews was to gain general under-standing of:

1. Types of conflicts that occur when multimor-bid patients are treated for multiple conditions 2. Additional challenges in multimorbidity care in practice, as experienced by medical profes-sionals

Interviews were focused and unstructured: before-hand we prepared topics and general questions for discussion, but interviewees were allowed to raise relevant issues not covered by the interview sched-ule. Topics were: experiences with conflicts be-tween treatment plans, both pharmacological and non-pharmacological; method for solving these con-flicts in clinical practice; adherence to and de-viation from CPGs in multimorbidity care; and sources of complexity in multimorbid patients; cur-rent usage of tools for decision-making; opinion on CDSS.

5. Expert interviews: personalising treatment plan of the multimorbid patient

5.2

Results

The interviewed GP and geriatrician frequently use CPGs, to complement their theoretical knowledge, experience-based methods and other reference doc-uments. Two interviewees emphasised that ambi-guities and vague workflow description are inherent to the purpose of CPGs: these are intended to only support GPs in daily practice.

All three interviewees raised the subject of inter-actions related to drug prescriptions: side effects, adverse drug events and drug-drug or drug-disease interactions. Recent CPGs discuss these interac-tions, as well as polypharmacy. In addition, phar-macies have computerized decision support systems to warn for drug interactions.

Although consciousness hereof has been rising, the interviewees felt that drug inconsistencies still frequently occur. Two interviewees hypothesised that communication between pharmacists, who may discover drug interactions, and medical spe-cialists, who need to change prescriptions, currently is suboptimal. Reconciliation of identified drug conflicts is hampered by the requirement of calling up the prescribing GP or specialist, a cumbersome process for the pharmacist.

Two out of three interviewees raised the prob-lem of unknown inconsistencies. The multimorbid elderly population is highly heterogeneous in terms of combinations of conditions and frailty.

Furthermore, the interviews described various in-teractions, related to patient complexity, patient preferences and competing demands.

5.2.1

Complexity of competing

demands and comorbidity

Multimorbid patients’ health is subject to change: over time, a condition(’s symptoms) may worsen or alleviate. As a consequence, one of the con-ditions might require a more active management, and side effects of treatments may be weighed dif-ferently. Other reasons for a certain prioritisation during a patient visit may be home situation (with or without spouse, volunteer caregivers, meal ser-vice), psychosocial status, or cost of treatment.

All interviewees named several comorbidities as source of additional complexity, such as Alzheimer’s disease or cognitive decline, and other characteristics, such as old age. CPGs do not

al-ways recognise competing demands. In patients with memory problems, for example, number of daily drug dosages is sometimes reduced. The ad-vantage of higher therapy adherence (less forgotten dosages) is prioritised over the advantage of multi-ple dosages.

One of the experts explained the importance of the relation between a patient’s conditions. Dis-eases can be in concordance (share pathophysiology and likely have overlapping management plans [95]) or in discordance (don’t share underlying predis-posing factors and likely have unrelated treatment plans [95]). In general, treatment plans for patients with discordant diseases conflict more often.

Thirdly, frailty is a source of added complexity, especially in older multimorbid patients. The geri-atrician and GP described physical frailty and psy-chological frailty. Physical frailty may be caused by, for example, malnutrition or a decline in strength, balance and fitness. It affects fall risk, speed of recovery and ability to exercise. Psy-chological frailty is caused by cognitive decline and may cause patients to forget to take medica-tion. The social-psychiatric nurse added that social frailty, a decline in social relations and social sup-port, may be a risk factor for depression and reduce physical exercise.

Frailty therefore may require adaptation of treat-ment plans. Herefore, the geriatrician used two tools. Firstly, the CPG ’Comprehensive Geriatric Assessment’ [96] developed by the Dutch associa-tion for clinical geriatrics (NVKG). This multidisci-plinary assessment aims at improving quality of life and independence of elderly patients. It results in an integrated and coordinated therapy plan, with a clear division of responsibilities for execution of the recommendations. In order to map a patient’s needs, physical, psychological, functional and social status are assessed. Other tools of the geriatrician are the Karnofski score (quantifies the functional status of cancer patients and evaluates chance of survival of aggressive treatment) and the Charlsson comorbidity index (predicts mortality and resource use based on a patient’s comorbidities).