The Transparent Representation of Medical Decision Structures

Based on the Example of Breast Cancer Treatment

D. Andrzejewski

, L. Tetzlaff

, J. DeBoer

E. Beck

and N. Haeusler

University of Applied Computer Science Brandenburg, Department Informatics and Media,

Magdeburger Straße 50, Brandenburg, Germany

Städtisches Klinikum Brandenburg GmbH, Geburtshilfe und Frauenheilkunde, Brandenburg, Germany

Keywords: Medical Decision Structures, Representation, Decision Support, DMN.

Abstract: Choosing the appropriate treatment for patients have a direct influence on each patient`s future. A doctor´s

expertise, the patient´s preferences, and the current medical research have a highly influence on the choice of

the treatment. Doctors shall be aware of their own decision patterns, the most influenced factors and the

relevant literature by choosing the optimal patients treatments. By considering quality management and

certifications, transparent representations of internal processes with simple decision-making notes are

required. In support of the hypothesis, a decision analysis was conducted based on the S3 guideline for

diagnosis, treatment and follow up care of breast cancer. A notation is required, which combines the process

modeling and the representation of (medical) decisions.

1 INTRODUCTION

A transparent representation of processes is a useful

gadget for physicians and other medical

professionals, especially considering the

requirements for quality management and

certifications. Medical decisions are generally based

on physicians’ expertise, patients’ preferences and the

relevant recommendations based on the best available

medical evidence. However, most decision structures

are only “present in the medical experts head”, which

means, they are not at hand in a standardized,

structured and thus “non-expert-understandable”

form. The representation of decisions has been

modelled in the past with The Decision Model (von

Halle, Goldberg, 2009). In our partner clinic

“Städtisches Klinikum Brandenburg GmbH” the

representation of processes and decisions will be

modelled mostly with Microsoft Visio. This

representation has the main problem that the clarity

and comprehension suffer. The Object Management

Group developed the Decision Model Notation

(DMN) as possibility to represent decisions

transparently in an easy way of understanding and

handling (Object Management Group, 2014).

Therefore this paper tries to answer the question if a

representation of medical decisions with DMN is

possible. Also the supporting data are required for a

transparent decision representation.

2 MATERIAL AND METHODS

2.1 Decision Model Notation (DMN)

The Object Management Group (OMG) published in

January 2014 the first Beta version of Decision Model

Notation (DMN). The primary goal of this notation is

design readable, understandable and transparent

decision models for every kind of user.

DMN is a new standard to combine business

decision design and decision implementation. Those

decisions need to be analysed and represented (Object

Management Group, 2014). The three aspects of

modelling are as followed: Business Processes (e.g.

modelled in BPMN), Decision Requirements

Diagram (modelled in DMN) and the Decision Logic.

A Business Process Model is e.g. a transparent

representation of an internal procedure of a company.

A Decision Requirements Diagram is a diagram to

represent the decision with different elements (like

knowledge elements or input data elements) and the

Decision Logic represents the analysed rules in form

617

Andrzejewski D., Tetzlaff L., DeBoer J., Beck E. and Haeusler N..

The Transparent Representation of Medical Decision Structures Based on the Example of Breast Cancer Treatment.

DOI: 10.5220/0005283306170621

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2015), pages 617-621

ISBN: 978-989-758-068-0

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

of a decision table. Figure 1 shows the three aspects

of modelling.

Figure 1: Three aspects of modelling (OMG, 2014).

It is very important to understand that there are two

different standards for modelling decisions which are

defined as followed:

 Business process modelling, using e.g.

Business Process Model and Notation

(BPMN).

 Decision logic, which defines business rules by

the way of decision tables and can be used for

individual decisions.

In BPMN it is possible to define business rules via the

BPMN Business Rule task. This task is the link to the

DMN notation. At this time there is currently no tool

which combines the modelling with BPMN and

DMN, except the Signavio Process Editor. The

intension of DMN is to combine business process

models and decisions logics (Object Management

Group, 2014). The following three aspects of

modelling are relevant:

 Business process models are defining tasks in

procedures, which may include decisions to be

reached.

 Decision requirements diagrams (DRDs)

specify the kind of decision to be made and the

information required. DRDs consist of

different elements: the specified decision,

required business knowledge, an Input-Data-

Model, the knowledge source and the different

connectors to model the flows which combine

these elements (information, knowledge and

authority requirements).

 Decision logics should represent the necessary

decisions in such detailed manner that valid

decisions in an automated fashion can be made

details for validate decisions and automate the

decisions also.

Figure 2 visualizes the elements of DMN:

Figure 2: Elements of the DMN (OMG, 2014).

In summery a decision structure can be represented

with a Decision Requirements Diagram (DRD) and a

Decision Logic. Those can be combined also with a

business process model in the relevant task.

2.2 Breast Cancer – Tumour Board

and Relevant Decisions

Breast cancer is the most common malignant

neoplasia in females in Germany, accounted for

approximately 70.000 new cases per year. After

confirming the diagnosis by histopathological

examination, the attending physician will inform the

patient about the further steps, which follow after

diagnosis. In some cases prior to this, the patient´s

case will be discussed in a pre-operative tumour

conference. However, according to the S3-guideline

(Kreienberg et.al, 2012) all cases will be discussed in

the postoperative tumour board. The attending

physician presents the individual case to the members

of the board and recommends the patient-individual

treatment choices. Then the board discusses the

different options and finally decides which

therapeutic regiment is the most promising, based on

the recommendations of the S3-guideline. This

decision will take place in form of a formal

consensus. The decision is then formally documented

and forms the basis for the further treatment. In figure

3 this process is visualized as BPMN model.

But which parameters are relevant for such a

decision? According to Kreienberg et.al the patient´s

age, her menopausal status, the expression of estogen

and/or progesterone receptor and HER2/neu status of

HEALTHINF2015-InternationalConferenceonHealthInformatics

618

Figure 3: Tumour board modelled in BPMN.

the tumour, Grading and the tumour and lymph node

classification are required.

For this paper the systemic treatment of the patient

will only be considered – consisting of either

chemotherapy or endocrine therapy, a combination of

both modalities, and/or the anti HER-2/neu (Human

epidermal growth factor receptor 2) treatment. Based

on the above listed tumour and patient related

parameters according to Kreienberg R. et.al: the

following rules concerning therapeutic decisions can

be derived:

 The patient`s menopausal status is required for

the choice of the endocrine therapy

 A primary (neoadjuvant) chemotherapy should

be recommended in cases of triple negative

breast cancer (i.e. estrogen receptor and

progesterone receptor and HER-2/neu negative

tumours)

 A primary (neoadjuvant) chemotherapy is

mandatory in cases of inflammatory breast

cencer or advanced tumour stages (T4)

 A primary (neoadjuvant) chemotherapy could

be recommended in cases with a positive HER-

2/neu status and/or a pathological tumour size

of >= pT1c and a high grading (G2 or G3, or a

positive nodal status and/or a positive

HER2/neu status and/or a positive or negative

hormone receptor (HR) status.

 In cases of positive HR status and a grading of

G2 further tests like uPA/PAI1 or gene

expression signatures can be performed

 Endocrine treatment (ET) is required in cases

of positive HR-status

 Anti HER-2/neu antibody treatment is required

in cases with positive HER2/neu status (which

is always applied in combination or sequential

to a chemotherapy regimen).

Tumour related parameters are important to describe

the biology and pathology of the tumour. Gathered

with patient related parameters, as the patient´s age

and menopausal status, they are required to model

valid DRDs. These parameters are the basement of

the breast cancer treatment and they are represented

as Input-Data-Model in a DRD. The most relevant

knowledge source is the S3 guideline, published by

the German Cancer Society. The current statements

of the guideline however may be modified by other

relevant literature (Kreienberg R., 2012, Woecke A.,

2010, Wolters R, 2011).

3 RESULTS

The results are represented in two major sections –

the graphical representation of the treatment

decisions for breast cancer modelled in DMN and a

representation of the deduced rules in a decision table.

For a representation of the treatment in DMN three

elements are required: decision element, knowledge

source and Input-Data-element. The decision element

will be labelled with the name “(systemic) therapy

TheTransparentRepresentationofMedicalDecisionStructuresBasedontheExampleofBreastCancerTreatment

619

decision for breast cancer”, the two Input-Data-

elements will be labelled with “Tumour and Patient

related parameters” and the knowledge source will be

labelled with “S3 guideline”. The following figure 4

shows the treatment of breast cancer in DMN.

Figure 4: therapeutic decision model of breast cancer

(DMN).

A draft of the decision logic is shown in table 1.

Besides rather simples rules like: “If the patient’s

tumour expresses HER-2/neu, then an anti HER-

2/neu antibody treatment is required” and “If an anti-

HER-2/neu treatment is indicated this has to be

applied simultaneously or sequentially to

chemotherapy”, there are a number of more or less

complex rules. The example for the easy rules

extending the decisions elements to: hormone

receptor status (positive or negative), HER2/neu

status (positive or, negative), Grading (G2, G3) and

therapy (chemo- and anti HER2 therapy (CT+T),

chemo-, anti HER2 and hormone therapy

(CT+T+ET)).

Table 1: example for easy rules.

ER/PgR

status

HER2/

neu

status

Grading Therapy

negative positive G3 chemo- and

anti HER2

therapy

positive positive G2 chemo-, anti

HER2 and

hormone

therapy

… … … …

Examples of more complex rules are e.g. those, based

on the current recommendation of the St. Gallen

consensus meeting (Goldhirsch, 2011, Kreienberg

2012) as shown in table 2.

In cases of an intermediate grading (G2) and a

positive HR-status the decision rules are modified by

introducing the concepts of luminal A and B like

tumors. Luminal B-like tumors are defined either by

Table 2: Risk and therapy decision on breast cancer

subtypes based on (Goldhirsch, 2011).

Sub-

type

Lumi

-nal

Lumi

-nal

B like

Lumi

-nal

B like

Non

lumina

Triple

negativ

ER/

PgR

+ + + - -

HER2

- - + + -

Ki-67

Low

14%/

High

14%/

n/a

therap

CT +

T +

CT +

ET: Endocrine Therapy, CT: Chemotherapy, T: Trastuzumab

a positive HER-2/neu status or are Ki-67 proliferation

index of > 14%. In cases of luminal A like tumors an

endocrine regimen is sufficient, in cases of luminal B

like tumors a combination of endocrine and

chemotherapy is required. Table 3 transforms the

criteria of the St. Gallen consensus in a clear cut

decision table.

Table 3: St. Gallen Consensus meeting – decision table.

ER/

PgR

status

HER2/

neu

Status

Grading Ki67 Therapy

positive negative G2 high CT + ET

positive negative G2 low ET

… … … … …

The example tables can be modelled as a decision

table in DMN as decision logic. This DMN table

representation is not considered in this paper.

4 CONCLUSIONS

A mandatory regarding the increasing complexity of

clinical decisions as well as patient self-

determination. This aim in our view can be achieved

by modelling decision requirements diagrams using

the Decision Management Notation. The

representation of business rules is likewise achievable

by means of defining a decision logic- therefore the

decision rules must be analysed in the first instance

HEALTHINF2015-InternationalConferenceonHealthInformatics

620

and then represented in DMN. DMN results from a

combination of Business Process Management and

Decision Management (OMG, 2014). Business

Process Management is required for the transparent

representation of business processes and procedures

– the Business Process Model and Notation (BPMN)

being one of the possibilities for modelling the

processes. For the analysis of business rules a

structured approach is required. According to Taylor

there are four principles to be kept in mind when

analysing decision rules (Taylor, 2011).

1. Model the decisions, which are in the experts

head

2. Be flexible, transparent and agile for changes

3. Be predictable

4. Combine all three rules and continue with the

first task

The recommendations are proven as good approach

for the analysis of business rules. However, to our

knowledge, there are no recent papers concerning this

approach in medicine. Further work is now centred on

the validation of our so far analysed and modelled

decision rules. Physicians and other medical staff can

benefit from transparent representation of the

processes and the modelled decision rules.

REFERENCES

Kreienberg R. et.al, (2012) Interdisziplinäre S3 Leitlinie zur

Diagnostik, Therapie und Nachsorge des

Mammakarzinoms.

Goldhirsch et al, (2011) Strategies for subtypes – dealing

with the diversity of breast cancer: highlights of the St

Gallen International Expert Consensus on the Primary

Therapy of Early Breast Cancer 2011.

von Halle B und Goldberg L, (2009) The Decision Model,

Edition, 2009.

Woecke A. et al, (2010) Impact of Guideline Conformity on

Breast Cancer Therapy: Result of a 13-Year

Retrospective Cohort (2010).

Wolters R. et al, (2011) Auswirkungen leitlinienkonformer

Therapie auf das Überleben von Patientinnen mit

primärem Mammakarzinom – Ergebnisse einer

retrospektiven Kohortenstudie (2011).

Taylor J., (2011) Decision Management Systems, 1

Edition, 2011.

Object Management Group (OMG), Decision Model

Notation (DMN) Beta Version 1.0, 2014.

TheTransparentRepresentationofMedicalDecisionStructuresBasedontheExampleofBreastCancerTreatment

621