Ontology Driven CPG Authoring and Execution via a Semantic Web Framework.
ABSTRACT Clinical Practice Guidelines (CPG) are used by healthcare practitioners to standardize clinical practic e and to provide evidence mediated health-care. Currently, ther e have been considerable efforts to computerize CPG so as to operationalize them within Clinical Decision Support Systems (CDSS) and to deploy them at the point of care. In our work, we take a semantic web approach - employing a domain ontology, a patient ontology, decision rules and a rule execution engine - towards the computerization and execution of CPG for CDSS. We present an ontology-driven approach for computerizing CPG and executing them based on individual patient instances. In our work we extend the Guideline Element Model (GEM) for computerizing CPG. We have (i) defined a CPG ontology based on the Document Type Definition (DTD) of GEM for ontologically representing a GEM encoded CPG; (ii) developed CPG decision logic definition tool and defined CPG rule syntax that allows practitioners to abstract and define decision logic rules based on the CPGs decision-variables inherent within the CPG; (iii) developed a forward-chaining CPG execution engine that executes the set of CPG execution logic rules using the JENA reasoning system; and (iv) implemented an automated justification tree generation module that provides the inference trace for the solution in order to assist practitioners in understanding the rationa le for the proposed recommendations. In practice, given a patient instance our CDSS is able to derive CPG based clinical recommendations. We will present a working prototype of our CPG-based CDSS for the EU Radiation Protection 118 Referral Guideline for Imaging (RPG).
- [Show abstract] [Hide abstract]
ABSTRACT: Clinical practice guidelines (CPGs) are knowledge source of evidence based best clinical practices in healthcare. Evidence based medical knowledge improves decision making at point of care. In view of that, efforts are being put forward to develop, disseminate, and implement clinical practice guidelines. But it has been argued that to make effective use of such important knowledge sources at point of care, it is necessary to computerize them so as to use them with clinical decision support systems. We present in this paper CPGs Knowledge Retrieval (CKR) framework that deploys our context, semantics, 'statistical analysis retrieval techniques' and indexing strategy. This framework retrieves the portions of CPGs knowledge content which are contextually sensitive and semantically relevant to the problem at hand.Applications of Digital Information and Web Technologies, 2009. ICADIWT '09. Second International Conference on the; 09/2009
Conference Paper: Pervasive nursing and doctoral assistant — PINATA[Show abstract] [Hide abstract]
ABSTRACT: The ratio of nurses and doctors to patients keeps diminishing due to increasing population health needs, however it is expected that the quality in healthcare services increases. By merging Ambient Intelligence (Ami) and semantic web technologies, PINATA aspires to address this issue. PINATA utilises pervasive devices to aid doctors and nurses to focus on the patient and thus improve the quality of healthcare services. In this paper we go over comparable Ami system architectures; summarise the physical and logical design of PINATA; provide details of the knowledgebase modelled in RDF/S and the ontologies designed for units of interest, including resources, and context-related concepts. These individual models are unified into a connected, context-rich data model through another set of classes and properties. The results presented were based on several tests done and are very promising.Pervasive Computing Technologies for Healthcare (PervasiveHealth), 2011 5th International Conference on; 06/2011
- [Show abstract] [Hide abstract]
ABSTRACT: Providing patient-centric health care services is the goal of health-care institutions. However, due to human-related aspects, this goal is frequently undermined. PINATA offers an automated patient-centric system based upon Pervasive Ambience Intelligence techniques and enriched with Semantic Web technologies. The system makes use of RFID sensors to track the movements of patients and medical staff in order to direct staff effectively. An automated camera system monitors the patients and alerts hospital staff in case of emergencies. Through handheld devices hospital staff is automatically provided with relevant patient information gathered from various sources. PINATA is based on a Service Oriented Architecture and makes use of domain specific ontologies.01/2011;
Ontology Driven CPG Authoring and Execution via a Semantic Web Framework
Sajjad Hussain and Syed Sibte Raza Abidi
Faculty of Computer Science, Dalhousie University, Halifax B3H 1W5, Canada
Clinical Practice Guidelines (CPG) are used by
to provide evidence mediated health-care. Currently, there
have been considerable efforts to computerize CPG so as
to operationalize them within Clinical Decision Support
Systems (CDSS) and to deploy them at the point of care.
In our work, we take a semantic web approach - employing
a domain ontology, a patient ontology, decision rules and
a rule execution engine - towards the computerization and
execution of CPG for CDSS. We present an ontology-driven
approachfor computerizingCPG andexecutingthem based
on individual patient instances. In our work we extend
the Guideline Element Model (GEM) for computerizing
CPG. We have (i) defined a CPG ontology based on the
Document Type Definition (DTD) of GEM for ontologically
representing a GEM encoded CPG; (ii) developed CPG
decision logic definition tool and defined CPG rule syntax
that allows practitioners to abstract and define decision
logic rules based on the CPGs decision-variables inherent
within the CPG; (iii) developed a forward-chaining CPG
execution engine that executes the set of CPG execution
logic rules using the JENA reasoning system; and (iv)
implemented an automated justification tree generation
module that provides the inference trace for the solution in
order to assist practitioners in understanding the rationale
for the proposed recommendations. In practice, given a
patient instance our CDSS is able to derive CPG based
prototype of our CPG-based CDSS for the EU Radiation
Protection 118 Referral Guideline for Imaging (RPG).
We will present a working
Clinical Practice Guidelines (CPG) are systematically
developed disease-specific recommendations to assist
clinical decision-making in accordance with the best
evidence [5, 8]. Despite the increased efforts by medical
specialists to developmedical guidelinesthey are still under
utilized at the point of care due to a variety of behavioural,
operational and technical reasons.
recognized that the incorporation of CPG in the clinical
workflow will have an impact of clinical decision-making.
Clinical Decision Support Systems (CDSS) provide an
apt medium for the computerization and execution of
CPG. Yet, to date, there are numerous challenges in both
the (a) computerization/codification of CPG in a formal,
executable format; and (b) the systematic execution of the
based recommendations for various patient care tasks.
Functionally speaking, CDSS compare a patient’s
medical condition with a medical knowledge base and
then guide a practitioner by offering patient and disease-
specific advice . In order to support CPG based
CDSS, various CPG modelling methodologies have been
developed to convert text-based CPG into an electronic
format (Computerized Clinical Practice Guideline; C-CPG)
that is both understandable by users and computers. Some
of the prominent attempts include GEM , EON, GLIF
, GUIDE and Prodigy .
We believe that C-CPG can serve not only as a
framework for representing CPG electronically, but the
computerization of the CPG can lead to the development
of evidence-based CDSS that incorporates both domain
knowledge and disease-specific recommendation/actions.
By design, CPG follow a decision logic that is structured in
an algorithmic format intended to support clinical decision
making. We argue that the decision logic in a CPG can be
used to generate explicit symbolic clinical decision-support
rules for discharging specific clinical recommendations.
The domain knowledge of the CPG can be represented
using an Domain Ontology such that the properties in a
Domain Ontology and relationships among them is explicit
described and serves to both complement and validate the
clinical decision support rules employed by the CDSS.
To develop CPG-guided CDSS, the challenges are (a)
the apt transformation of the CPG inherent decision logic
into both medically salient decision rules; (b) and to ensure
the validity of the transformed knowledge and to provide
trust in the recommended actions; and (c) to execute
Yet, it is widely
Figure 1. System Design
the computerized CPG to derive decision support.
this paper, we present an ontology-driven approach for
computerizing CPG and executing them based on patient
data. In our work we extend the Guideline Element
Model (GEM) for computerizing CPG which currently
does not support CPG execution logic definition and CPG
execution.We have (i) defined a CPG ontology based
GEM representation in order to ontologically represent a
CPG; (ii) developed a CPG execution logic definition tool
that allows practitioners to define logic rules based on the
CPG decision-variableswithin the CPG; and (iii) developed
an CPG execution engine that executes the set of CPG
execution logic rules using the JENA reasoning system .
Given a patient instance, our CPG-based CDSS is able to
operationalizethe CPG and derive CPG based interventions
recommendations. We will present a working prototype of
our CPG-based CDSS for the EU Radiation Protection 118
Referral Guideline for Imaging (RPG).
2. Problem Description and SolutionApproach
In order to establish a CPG-guided CDSS, we take the
following challenges into consideration:
• How to encode CPG in a computerized format whilst
encoding the underlying semantics.
• How to transform the CPGs inherent decision logic
into medically salient decision rules.
• How to execute the computerized CPG to achieve
• How to ensure the validity of the transformed
knowledge and to provide trust in the recommended
We take a semantic web approach to meet the
above challenges.The Semantic Web [13, 14] is
a logic-based architecture which provides a framework
for both representing and operationalizing different data
sources, where data is enriched by their semantics and
ontologies.For these enrichments, the Semantic Web
supports standards/languages, which define data objects
and relations between them using metadata vocabularies
and concept hierarchies. Resource Description Framework
(RDF) [13, 14] is used to annotate data objects (resources)
in terms of their properties and property values as RDF
triples, in an RDF graph.
graph is defined by its Domain (rdfs:domain) and Range
(rdfs:range). The Web Ontology Language (OWL) [13,
14] is a language used for defining and instantiating
ontologies in a Semantic Web.
includes descriptionsand relationshipsamongRDF classes,
properties and their instances.
semantically annotating text-based CPGs into RDF and
defining their concept hierarchies and properties in OWL,
a better cooperation between CPGs in the Semantic Web
can be achieved.
A property in an RDF
An OWL ontology
We believe that by
To develop a semantic web based CDSS, using a
computerized CPG, we developed three ontologies.
These ontologies model the entire working environment
recommendations. The three ontologies are:
CPG Ontology that models the computerized structure of
the CPG. In this case we model the CPG using the GEM
structure therefore our CPG ontology is based on GEM
DTD (described in section 4) for representing the CPG
semantically; (ii) A Domain Ontology that models the
medical knowledge pertaining to the CPG. The Domain
Ontology represents both the concepts described in a CPG
and the relationships between these concepts as OWL
classes and properties, respectively (see section 3); and
(iii) A Patient Ontology that models the patient in terms
of various health information elements that may constitute
the longitudinal medical record of patient.
Table 1. EU Radiation Protection 118 Referral Guideline for Imaging (Table A excerption)
Headache: chronic XR skull, sinus, C
spine (I) (B)
Not indicated routinelyRadiology
signs/symptoms. See A13 below.
oflittle usein theabsence offocal
Pituitary and Juxta-
children seeCT (II) or MRI (0) Not indicated routinely
Some exceptions for specialists or if evidence of raised
intracranial pressure, posterior fossa or other signs.
Demonstration of microadenomas may not be helpful for
management. CT if MRI not available. Urgent referral when
vision deteriorating. Some centers use specific NM agents.
SXR (I)Not indicated routinely
Patients who require investigation need MRI or CT.
fossaMRI (0)MRI much better than CT. CT images often degraded by
beam hardening artifacts.
CT adequate for most cases; MRI sometimes necessary and
may be more appropriate in children. US first choice for
infants.NM used in some centres, especially for shunt
CT (II)Indicated (B)
childrenseeXR (0)Indicated (C)XR can demonstrate whole valve system.
ontology allows to generate standardized descriptions of
a patient, which in turn serve as patient instances used to
execute the decision logic. CPG decision logic is captured
and represented as JENA rules (see section 5.2) that
constitute elements from the Domain Ontology.
Our CDSS is divided into two main modules, namely
CPG Authoring System and CPG Rule Authoring and
Execution System. The CPG Authoring System requires a
text-basedCPG and a domainontologyeliciting the domain
concepts pertinent to the CPG in question. The properties
in the Domain Ontology are used to semantically annotate
the decisions variables in the CPG Ontology (described
in section 4.2). We encode the text-based CPG into the
CPG ontology and annotate the decision variables and
logic structures in the CPG Ontology based on the Domain
CPG Rule Authoring and Execution System provides
a framework for defining the decision logic rules in a
CPG and executing them based on the patient clinical
data. We have developed a simple rule syntax that allows
practitioners to define decision rules based on the CPG
decision variables (see section 5.1).
transform the decision rules into JENA rule syntax, which
can then be inputted to an inference system JENA (see
section 5.2 and 5.3).JENA uses the rule set to infer
recommendations based on patients clinical situation. The
patients clinical situation is representedin terms of a patient
ontology that incorporates patient properties such as age,
gender, medical history; and the values to these properties
serveas inputto thereasoningsystem. Thearchitecturealso
supports the generation of an automated derivation trace
of inferred recommendations for enhancing the plausibility
of the judgement (see section 5.4). Figure 1 shows the
architecture of our CPG guided CDSS. .
3. Domain Ontology: EU Radiation Protection
118 Referral Guideline for Imaging (RPG)
The Domain Ontology models the medical knowledge
pertaining to the CPG. It represents both the concepts
described in a CPG and the relationships between these
concepts as OWL classes and properties, respectively. It
stores the clinical scenario and treatment as instances
of the Ontology. We argue that the Domain Ontology
must be valid and complete in order to generate correct
recommendations to the patient.
working of our CPG-guided CDSS, we have used the EU
Radiation Protection 118 Referral Guideline for Imaging
(RPG)  as the Domain Ontology for explaining our
CPG Authoring and Execution System . RPG Domain
Ontology will be referred later on in the paper.
1 shows an excerpt on the RPG, whereas an exemplar
fragment of RPG Domain Ontology is shown in Figure 21.
Given a clinical problem (e.g.
various investigationmethods and their associated radiation
dosages are proposed.For each investigation method,
To demonstrate the
1Taken from 
Figure 2. A fragment of RPG Domain Ontology
a recommendation is made along with the grade
indicating the type of available evidence on which the
recommendationis based upon. So, the correlationbetween
the first three columns seem straightforward. The comment
column contains discussions, cautions and alternatives of
the investigation methods. Reflecting such knowledge in
the ontology is the most challenging task.
in Table 1, a comment can simply state the effectiveness
of investigation method (i.e.
in the absence of focal signs/symptoms”) or can suggest
alternatives with implication of probabilistic reasoning.
The RPG Domain Ontology is mainly divided into three
categories of concepts: Clinical Problems, Investigations
and Recommendations. All distinct clinical problems are
represented as instances of respective clinical problem
classes.These classes are arranged as sub-classes
under the class Clinical Problem.
radiological procedures in RPG and represented in RPG
Domain Ontology similar to the Clinical Problems. For
a given clinical problem, one or more investigations are
recommended. Recommendations are treatments based on
investigationsand clinical problems. Eachrecommendation
is represented along with the grade of its evidence. An
indicated recommendation is most likely to contribute
to the diagnosis and management, while a not indicated
routinely recommendation emphasizes the limitations of an
”Radiography of little use
4. CPG Authoring System
In order to support CDSS, various CPG computerization
methodologies have been developed to convert text-based
CPG into an electronic format (for instance GLIF, GEM,
GASTON), most CPG methodologies do not extend as
far as the execution of the computerized CPG based on a
patient’s case; this is typically due to the absence of a CPG
execution engine. In our work we aim to execute the CPG,
and in this regard the challenges that we addressed are:
(a) the apt transformation of the CPGs inherent decision
logic into both medically salient decision rules; (b) to
ensure the validity of the transformed knowledge and to
the computerized CPG to derive clinical decision support.
In order to meet above mentioned challenges, we design
a CPG Authoring framework that incorporates a CPG
ontology, based on the GEM DTD, to represent the CPG
in concert with the domain ontology. The CPG authoring
tool allows the user to computerize a text-based CPG by
annotating the decision variables inherent within the CPG
based on the concepts described in the domain ontology.
4.1Guideline Elements Model (GEM)
The Guideline Elements Model (GEM) provides
a platform for annotating text-based clinical practice
guidelines (CPG) in Extensible Markup Language (XML)
documents and representing main/key features in those
CPG . For this purpose, GEM incorporates over 100
elements in order to fully implementation heterogeneous
parts that make up the content of a clinical practice
guideline.These elements are organized in the GEM
hierarchy. The Knowledge Component elements are the
most important elements of GEM and represent procedural,
conditional or imperative knowledge found in a CPG. The
KnowledgeComponentelement contains Recommendation
sub elements, each of which describes the recommended
actions for patients. These recommendations can be either
imperative or conditional. Imperative recommendationsare
those that are applicable to the entire eligible population.
In contrast, conditional recommendationsdescribe decision
variables that need to be considered and actions to be
undertaken if the decision variables meet a certain criteria.
An important sub-element of conditional recommendations
is the logic element. The logic element states explicitly the
conditionsthat are requiredfor certain actions to take place.
4.2 CPG Ontology
We developed the CPG Ontology based on the GEM
DTD using an Ontology Editor Prot´ eg´ e .
a new property variable.name in the decision.variable
class into the CPG Ontology, where the rdfs:domain
of variable.name is the decision.variable class in CPG
Ontology and rdfs:range is all the properties in the
pre-defined Domain Ontology.
decision.variable with a property variable.name with the
property value from the Domain Ontology, each decision
variable in the GEM encoded CPG is represented by a
resources and properties in the Domain Ontology. Figure 3
demonstrates the semantic annotation of decision variables
dv1 ...dv11 and action variable a1 with the list of
variable names, which are represented as properties in the
RPG Domain Ontology.
In Figure 3, each of the decision variables in the
CPG Ontology are annotated with the properties in the
RPG Domain Ontology based on concepts presented
in Table 1.For example, decision varible dv1 is
annotated with a property hasClinicalProblem, where the
property value describes various clinical problems in the
RPG test-case as shown in the first column of Table 1.
Similarly, decision variables dv2, dv4, dv5, dv10,
dv11 are annotated with properties, where the property
values describe one (represented by applyOnlyTo) or more
investigations based clinical problems as shown in the
second column of Table 1.
annotated and describe the recommendations along with
their grades based on investigations and clinical problems
as shown in the third column of Table 1. Action variable
a1 is annotated with a property isRecommended, where
the property value repesents the list of investigations along
their methods and recommendations along their grades.
By annotating each
Finally, dv8, dv9 are
Patient Ontology models the Electronic Patient Record
Figure 3. Annotation of Decision Variables
information and patient clinical situations.
generate standardized descriptions of a patient, which in
turn serve as patient instances used to execute the decision
logic of a CPG within the CDSS.
It allows to
5. CPG Rule Authoring and Execution System
The functionality of the CPG Rule Authoring and
Execution system is to encapsulate the clinical decision
makinglogicinherentwithina CPG intermsoflogicalrules
that can be executed by reasoning engines, to derive CPG-
based recommendations for specific patient conditions. To
achievethis functionalitywe built twosub-modulesnamely,
Rule AuthoringModule and ExecutionModule for defining
decision logic rules embedded in a CPG, and executing
them based on clinical investigations and patient profiles,
The Rule Authoring Module provides an interface (see
section 6) to the practitioners to define decision logic rules
a CPG Rule Syntax. This is achieved as follows:
• A CPG ruleis writteninthelogictagofCPG Ontology
and comprise decision variables present in the CPG
• Each of the decision variables has a property
variable.name, where the property value (of each
variable.name) corresponds to a property in the
Domain Ontology (see Example 4.1).
Upon completion of the rule authoring process, the rule
authoring module transforms the CPG rules into the JENA
Logic := IF Decision Variable List THEN Action Variable
| Action Variable
Decision Variable List := dv Rel Node, dv Rel Node, .., dv Rel Node
//where <dv,dv name> ∈ V
//where <a,a name> ∈ A
Action Variable := a Rel Node
Rel := < | <= | > | >= | =
Node := ?
| ’a literal’
| [dv1 dv2 ...
Algebra := Value | Value + Value | Value
Value := dv | number
// a plain string literal
Value | Value * Value
// dv must already been declared before
Figure 4. CPG Rule Syntax
syntax for their execution with the JENA inference engine
(see section 5.3).
The Execution Module invokes the JENA inference
engine to execute a CPG, and infer recommendations
based on patient clinical situations, various treatment plans
modelled in CPG and its Domain Ontology. We model
instances from the Domain Ontology, CPG Ontology and
Patient Ontology as RDF graphs, which serve as the
knowledge base for JENA. JENA inference engine starts
with the knowledge base and the JENA rule set and
builds an inference model.
for querying inferred recommendations using backward
logic programming engine (described in section 5.4).
Furthermore, we use the JENA inference model and their
supported modules for presenting the derivation trace for
inferred recommendations. The detailed process steps are
described in following subsections.
The model is then used
5.1CPG Rule Syntax
V is the set of pairs <dv,dv name> of all decision
variables and their names, where dv is a decision.variable
instance in the CPG Ontology and dv name is the
variable.name property value of dv,
A is the set of pairs <a,a name> of all action variables
and their names, where a is an action.variable instance
in the CPG Ontology and a name is the variable.name
property value of a.
Rules in the logic element of CPG Ontology can be
written in the CPG rule syntax as shown in Figure 4. Each
rule is a forward rule, which has a list of decision variables
(body) and an action variable (head) of the rule, followed
by IF and THEN, respectively. In the decision variable list,
each variable dv is an equality or inequality relation with
eitheri)a variable,ii) astring,iii) alist of(alreadydeclared)
decision variables or iv) an algebraic (binary) formula. An
example CPG rule is described in Example 4.1.
5.2JENA Rule Syntax
JENA is a general purpose rule-based reasoner used to
implement both the RDF and OWL reasoners and also can
be applied to general purposes . This reasoner supports
chaining, backward chaining and a hybrid execution
model. JENA is comprised of two internal rule engines,
namely,forwardchainingRETE engineand backwardlogic
programming engine and can run as a backward chaining
reasoningsystem. An informal descriptionof the simplified
text rule syntax (as mentioned in JENA documentation )
is shown in Figure 5. The ”,” separators are optional. The
functor in an extended triple pattern is used to create and
access structured literal values. An example rule written in
JENA rule syntax is shown in Example 4.1.
5.3Transformation of CPG rules into
X is the set of JENA ?varname
?= V ∪ A
D :??→ X is a function which takes either a decision
variable or action variable and returns a unique JENA
variable, which represents resource of its variable name.
R :??→ X is a function which takes either a decision
variable or action variable and returns a unique JENA
variable, which represents value of its variable name.
ν :??→ X is a function which takes either a decision
variable or action variable and returns the encoded value.
Each CPG rule have a Decision Variable List
and an Action Variable, which are followed by the
IF and THEN and serve as head and body of the CPG rule,
respectively. Translation of CPG rules into JENA rules is
performed by a Transformation Algorithm. It parses the
head and body of a CPG rule and translate the decision
variable relations (in the body) and action variable (in the
Rule := bare-rule | [ bare-rule ] | [ ruleName :
bare-rule := term, ...
| term, ...term <- term, ...
hterm := term | [ bare-rule ]
term := (node, node, node)
| (node, node, functor)
| builtin(node, ...
functor := functorName(node, ...
node := uri-ref // e.g.
| ’a literal’
| number// e.g.
term -> hterm, ...hterm
// forward rule
// backward rule
// triple pattern
// extended triple pattern
// invoke procedural primitive
node)// structured literal
// a plain string literal
// a typed literal, xsd:* type names supported
42 or 25.5
Figure 5. JENA Rule Syntax
Let R be a CPG rule, B be a body (premises) and H be a head (conclusion) of JENA rule.
1. Parse the body of a CPG rule R.
(a) For each decision variable relation in the body of R.
i. If the decision variable dv has a relation with its value v and annotated with a variable name dv name then add the triple
(D(dv) dv name v) in B.
ii. If the decision variable dv has a relation with another (declared) decision variable dv′and annotated with a variable name
dv name then add the triple (D(dv) dv name ν(dv′)) in B.
iii. If the decision variable dv has a relation with a variable ? and annotated with a variable name dv name then add the
triple (D(dv) dv name R(dv)) in B.
iv. If the decision variable dv has a relation with a variable list of variable [dv1...dvn] and annotated with a variable name
dv name then add the triple (D(dv) dv name List(ν(dv1)...ν(dvn)) in B.
(b) For decision variables with inequality relation and algebraic formula, repeat steps (i - iv) and add JENA built-in functors for
inequality and algebraic formula in B.
2. Parse the head of a CPG rule R : analogous to step (1).
Figure 6. CPG Rule Transformation Algorithm
head) into JENA rule syntax, recursively. Main steps of
this algorithm are outlined in Figure 6 and illustration is
presented in Example 4.1.
Let R be a decision rule written for the RPG test-case in
CPG rule syntax.
R = IF dv3=Patient, dv1=?, dv6>45, dv5=?,
dv8=indicated, dv9=?, dv10=?, dv11=?
dv7=[dv5 dv11 dv8 dv9]
The above CPG rule is translated into JENA rule and
shown as follows:
Transform (R) = [conditional1:
(?X1 rdf:type rpg:Patient) , (?X1
rpg:hasClinicalProblem ?X3) ,
greaterThan(?X6, 45) , (?X1 rpg:age
?X6) , (?X3 rpg:hasSolution ?X4), (?X4
rpg:hasRecommendation rpg:indicated) ,
(?X4 rpg:hasRecommendationGrade ?X7) ,
(?X4 rpg:hasInvestigationDetails ?X8)
, (?X8 rpg:hasInvestigationMethod ?X9)
(?X1 rpg:isRecommended List(?X4 ?X9
In Example 4.1, dv1=?
relation, where dv1 is annotated by a variable name
hasClinicalProblem (as shown in Figure 3) and
has equality relation with a variable ?.
variable relation dv1=? is translated into JENA syntax as
(?X1 rpg:hasClinicalProblem ?X3) and added
in the body of the JENA rule via step (iii) of the
transformation algorithm (as shown in 6).
is a decision variable
Figure 7. Derivation Trace for Recommendations for Jane
the decision variable relation dv6>45 is translated
(via step i) into greaterThan(?X6, 45) , (?X1
rpg:age ?X6).Note that JENA built-in functor
greaterThan(?X6, 45) is added (via step (b)) due to
5.4Automated Derivation Trace Module
We developed an automated derivation/justification
module which generates the justifications behind inferred
recommendations based on the CPG and the patient data.
This is to provide a trace of the rule execution to the
medical practitioner so that he/she may be able to interpret
the logic behind a certain recommendation; without such
justifications the system will turn into a ’black-box’
which is not appreciated by medical practitioners.
derivation includes the linear representation of premises
(facts) under which the JENA rules are satisfied and the
conclusions based on those rules.
derived patient recommendation (derived facts) from the
JENA model (knowledge base) and generates facts which
recursively. This module terminates, if all the premises are
ground instances (known facts).
In the RPG test-case, based on the encoded JENA
rule in Example 4.1, automated derivation trace for the
recommendations for a patient ”Jane” is generated and
shown in Figure 7.
This module takes a
6. CPG Authoring and Execution System
We used the above presented approaches to establish
an Ontology-DrivenCPG Authoring and Execution System
(CPG-EX). The System inputs a text-based CPG and loads
its pre-defined Domain Ontology and the Patient Ontology.
It encodes the CPG in terms of the CPG Ontology. It uses
the Domain Ontology to semantically annotate the decision
variables in the CPG Ontology(as describedin section 4.2).
Subsequently, it transforms the CPG rules encoded in CPG
Ontology into JENA rule syntax and passes the rules to
the JENA reasoning system. Finally, the CPG-EX system
invokes JENA reasoner with the patient instance and rule
set for inferring recommendations and other information
based on the patient profiles. The system also generates
derivations traces for inferred recommendations in order to
enhance plausibility of those recommendations.
The CPG-EX interface (as shown in Figure 8) is
composed of three panels. On the left panel, user can load a
CPG. The middle panel has the CPG Ontology structure for
semantically annotating the CPG into CPG Ontology. The
to the user:
• Duplicate Button: Duplicates selected tag/element in
the CPG Ontology structure.
• Save Button: Save the CPG Ontology structure into a
file in RDF/XML format.
Figure 8. CPG-EX Interface
• Run Button: Run the CPG on patient instances by
transforming CPG logic tags into JENA rules.
• Query Button: Query the result after running CPG.
Right panel is used for assigning instances to each tag
in the CPG Ontology structure. It consists of Ontology
Instances text box, Variable Name List and a Decision
Variable List. The Variable Name List displays all the
properties stored in the Domain Ontology. The Variable
Name List becomes active only when a variable.name tag
(from CPG ontology) is selected. User can select a variable
name form the Variable Name List and assign it to a
decision variable. Decision Variable List shows the list of
annotated decision variables associated with their variable
names. Query Button initiates a Query Window (as shown
information and inferred recommendations. Furthermore,
it allows the user to view the derivation trace for inferred
CPG Authoringis donebyextractingtextualinformation
of the text-based CPG (shown in the left panel of CPG-
EX) and annotating them based on the classes and their
properties in the CPG Ontology (shown in the middle panel
ofCPG-EX).The annotatedtextis assignedto the Ontology
Instance text box (shown in the right panel of CPG-EX)
based on the classes/properties in the CPG Ontology. Rule
Authoring is performed by defining decision rules in the
logic tag of CPG ontology. Decision rules are written in
CPG syntax. Rule Authoring process can be outlined by
following main steps:
1. Select the list of decision variables from the Decision
Variable List, which represents the body (premises) of
the rule and followed by IF.
2. Select the action variable from the Decision Variable
List, whichrepresentsthe head(conclusion)of the rule
and followed by THEN.
3. For each decision variable and action variable in the
either a variable, a value, a binary algebraic formula,
another decision variable or list of decision variables
(see Example 4.1).
We tested our system on the clinical practice Guideline,
EU Radiation Protection 118
Imaging(RPG) . We usedthe pre-definedRPG Ontology
(see section 3) and Patient Ontology, which are defined
and described in . The CPG-EX system encoded the
text-based RPG into the CPG Ontology and transformed
the CPG rules into JENA. Finally, it invokes the JENA
reasoner to query inferred recommendations and other
information based on selected patient profiles. An example
recommendation for the patient Jane based on RPG test-
case is shown in Figure 7.
Referral Guideline for
7. Discussion and Concluding Remarks
In this paper we developed a CDSS by exploiting C-
CPG and Domain Knowledge represented as ontologies.
The proposed attempt could also be applied to other similar
domains (such as workflows) where the activities are based
on a design that entails a decision logic that is structured in
an algorithmic format.
enriched by the Algorithm tag, which represents sequential
stages in health management described by a CPG. The
Algorithm tag is comprised of Action Step, Conditional
Step, Branch Step and Synchronization Step. We believe
that exploitingthe Algorithmtag is beneficialfor improving
the CPG-EX and formally representing the sequential
stages of recommendations and then executing them based
of patient clinical situations and recommendation stages.
CPG rule authoring module provides an interface for
defining rules in CPG rule syntax.
syntax helps practitioners to define abstract decision rules
and refrains from the formal details.
syntax is not as expressive as formal languages such as
N3 and RDF . This is a common trade-off and
challenge while building rule authoring interfaces. There
has been an attempt to define an abstract syntax ARDEN
for representing medical decision logic . It would be
interesting to investigatethe rationale and expressivenessof
ARDEN syntax and then exploit this syntax for authoring
the decision rules for CPGs in CPG-EX.
So far, we have proposed an approach which takes into
account one CPG and its corresponding domain ontology
for establishing the CPG execution system. However in
practice, practitioners may require more than one CPG to
consult and advice better and feasible recommendations.
Hence, it is important to investigate and establish such a
framework, which allows multiple CPG with their inherent
logic and decision structures to support in finding patient
specific recommendations in a consistent way .
We believe that whilst working with a CDSS, the
presentation of inferred information without providing
justifications reduces the plausibility of the inferred
recommendations to the practitioner. Although CPG-EX
provides justifications/proof trees at their calculi levels,
these proofs are not understandable and difficult to read.
In order to share these proofs among users and other
of proofs at the higher level of abstraction and granularity
In this paper we established a CDSS and showed how
the Domain Ontology of a CPG can be linked with CPG
Ontologyfor semantically annotatingthe decision variables
and decision logic elements in CPG Ontology.
we defined a transformation of CPG Rules (annotated in
the CPG ontology) into JENA Rules, which are used by
JENA inference engine to infer recommendationsand other
information based on patient profiles. We implemented an
automated derivation/justification generation module that
Although, this rule
practitionersin understandingthe rationale forthe proposed
Future developments will involve:
methods for incorporating multiple CPGs to ensure
more feasible clinical decision making, ii) adapting
the Algorithm tag for representing sequential steps
for recommendations, iii) exploiting ARDEN syntax
for authoring CPG decision rules and iv) presenting
justifications for clinical recommendations in natural
language at higher level of abstraction and granularity.
 S. Abidi.
integration of medical knowledge resources.
International Symposium on Computer-Based Medical
Systems, Dublin, June 23-24 2005.
 H. Chen, S. Devitt, and J. D. Roo. Rpgontology: A semantic
webized knowedge base for radiation protection guidelines.
First International Conference on Semantic, Knowledge and
Grid, Beijing, China, Nov. 27-29 2005.
 E. Commission. Radiation protection 118 referral guideline
for imaging. 2000.
 M. Eccles, E. McColl, N. Steen, N. Rousseau, J. Grimshaw,
and D. Parkin.Effect of computerized evidence based
guidelines on management of asthma and angina in adults
in primary care: cluster randomized controlled trial. BMJ,
(325):1–7, October 2002.
 M. J. Field and K. N. Lohr. Clinical Practice Guidelines:
Directions for a New Program. National Academy Press,
Institute of Medicine, Washington, DC, 1990.
 G. Hripcsak.The arden syntax for medical logic
modules: introduction. Computers in Biology and Medicine,
 R. A. Jadad and R. B. Haynes. Collaboration-advances and
challenges in improving evidence-based decision making.
Med Decision Making, (18):2–9, 1998.
 H. Knublauch, R. W. Fergerson, N. F. Noy, and M. A.
Musen.The prot´ eg´ e owl plugin: An open development
environment for semantic web applications. In McIlraith et
al, pages 229–243.
 M. Peleg, A. Boxwala, and O. Ogunyemi.
evolution of a guideline representation format.
Annual Symposium, pages 645–649, 2000.
 M. Peleg and S. Tu.Comparing computer-interpretable
guideline models: A case-study approach. J Am Med Inform
Assoc, (10):52–68, 2003.
 R. N. Shiffman, B. T. Karras, A. Agrawal, R. Chen,
L. Marenco, and S. Nath. Gem: A proposal for a more
comprehensive guideline document model using xml. J Am
Med Informatics Assoc, (7):488–98, 2000.
Recommendation, February 2003.
 W3C. Rdf primer. http://www.w3.org/TR/rdf-primer, W3C
Recommendation, February 2004.
Medical knowledge morphing: Towards the