An Ontology-driven Proposal for

Semantic Interaction among Heterogeneous

Health Information Systems

Idoia Berges, Jes´us Berm´udez and Arantza Illarramendi

University of the Basque Country, Paseo Manuel de Lardizabal, 1

20018 Donostia-San Sebasti´an, Spain

Abstract. The adoption of Electronic Health Records (EHRs) has brought mul-

tiple advantages to the healthcare area. However, the goal of achieving semantic

interoperability of EHR information between heterogeneous Health Information

Systems has not been accomplished yet. In such scenario, the purpose of this pa-

per is twofold: On the one hand, the presentation of our ontology-based approach

to the problem of EHR interoperability (restricted to the case of medical observa-

tions), which goes one step further with respect to other approaches for the same

goal, and on the other hand, the presentation of two additional features that com-

plement our approach: path mappings for transforming individuals that represent

EHR information and rules for medical knowledge sharing.

1 Introduction

In 2009 the European Community presented a longer-term research and deployment

roadmap that provides the key steps for achieving semantic interoperability in the area

of healthcare[1]. The motivation for that is that nowadays the idea of one person re-

ceiving health assistance from the same medical institution throughout all his life is no

longer realistic. Thus, medical institutions must be prepared to receive patients from

other regions or countries without the quality of service being affected. The incorpora-

tion some years ago of Electronic Health Records to the institutions may be seen as the

ﬁrst step towards the goal, since, apart from local advantages over manual records such

as avoiding legibility problems due to poor handwriting which may lead to misunder-

standings, they favour a fast exchange of clinical data between different organizations.

However,the fact that most institutions have developedtheir health information systems

in an autonomous way has resulted in a proliferation of heterogeneous health informa-

tion systems, each one with its own proprietary models for representing and storing

EHR information, which difﬁcults the task of interoperating with each other.

In many areas, the adoption of knowledge representation standards stands out as the

most usual approach to solve interoperability problems. This happens also in the health-

care area, where some standards such as openEHR

, CEN-13606

and HL7-CDA

are

www.openehr.org

www.en13606.org

www.hl7.org

Berges I., Bermúdez J. and Illarramendi A..

An Ontology-driven Proposal for Semantic Interaction among Heterogeneous Health Information Systems.

DOI: 10.5220/0003350500440053

In Proceedings of the International Workshop on Semantic Interoperability (IWSI-2011), pages 44-53

ISBN: 978-989-8425-43-0

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

under development for this purpose. All those three follow a dual model-based method-

ology for representing the information that may appear in an EHR: On the one hand, the

Reference Model is a stable model which deﬁnes the basic structures for representing

EHR information (such as List, Table, etc.). On the other hand, the Archetype Model

deﬁnes speciﬁc knowledge elements (such as Respiration Rate) by using and constrain-

ing the elements of the Reference Model. Although the idea of using a standard may

seem suitable for the goal, we think that interoperability does not mean to have a unique

representation but a semantically acknowledgeable equivalent one. This would relieve

medical institutions from being forced to use one standard in the representation of their

knowledge and moreover, since several standards are being developed for the same pur-

pose, the interoperability problem will remain unsolved unless these standards merge

into a single one.

In this paper we present a proposal to move towards the notion of full semantic

interoperability of EHRs, which states that when one particular system receives some

EHR information from another institution, the received information can be seamlessly

integrated into its underlying repository because the differences in the language, in the

representation of the information and in the storing systems do not cause any misun-

derstanding[1]. Our solution is based on the use of semantic technologies, and more

precisely on OWL2[2] ontologies and corresponding reasoners.

In the area of EHRs semantic interoperability a certain number of related works can

be found at present. The works mentioned next also rely on semantic technologies that

facilitate semantic interoperation between heterogeneous information systems as op-

posed to other formats for interchanging data such as XML which do not deal with the

semantics of the exchanged data[3]. [4] provides a solution to achieve semantic inter-

operability between systems that have been developed under the HL7 reference model.

However, this proposal requires that the source system has some prior knowledge about

the target system and moreover, it does not tackle the communication between sys-

tems that use proprietary EHR speciﬁcations. In [5] ontology mappings are proposed

between pairs of archetype-based models. Finally, in [6] a model-driven engineering

approach that transforms archetypes of the CEN-13606 standard into OWL models is

presented.

The purpose of this paper is twofold: On the one hand, the presentation of our

ontology-based approach to the problem of EHR interoperability (restricted to the case

of medical observations) and on the other hand, the discussion of some features that

complement the current approach -which may be also relevant to other ontology-based

interoperability solutions. More speciﬁcally, we would like to stress two of them: ﬁrst,

the usefulness of deﬁning a new category of mappings between the elements of two

ontologies -called path mappings- which indicate some kind of relationship between

two property paths in the ontologies and facilitate the transmission of the information

about individuals between two ontologies. Secondly, the convenience of incorporating

SWRL[7] rules to the ontologies to deﬁne and share medical knowledge among institu-

tions.

The rest of the paper is divided as follows: ﬁrst, a general overview of our ontology-

based approach for semantic interoperability of EHRs is presented in section 2. Section

3 tackles the complementary features pinpointed above. Finally conclusions are dis-

cussed in section 4.

2 Overview of the Framework

In general, an EHR includes clinical statements such as observations, laboratory tests,

diagnostic imaging reports, treatments, therapies, administered drugs and allergies. In

this paper we focus on the exchangeability of medical observation statements, which

are used to record all notionally objective observations of phenomena and patient-

reported phenomena, such as physical examinations, laboratory results or basic infor-

mation about the patient (weight, sex,...). In Fig.1 the architecture of our proposal is

shown.

Fig.1. Architecture of the solution.

This proposal is sustained in one of the approaches for interoperability among sys-

tems described in [8]: Using a canonical model to which the particular systems are

linked. More precisely, we deal with a Canonical ontology that represents medical ob-

servations in a canonical way, that is using a general representation that is independent

from the different conceptualizations of them that can exist. In that ontology we pro-

pose a subdivision of medical observations into two groups: simple observations and

composite observations. Simple observations have a single value and unit of measure-

ment. Additionally, we have also identiﬁed three properties that may be relevant at the

time of characterizing an observation: the protocol, which records information about

how the observation process was carried out, either by indicating a particular clinical

protocol (e.g. the Balke protocol for treadmill graded exercise testing) or the medical

device used for taking the measurement (e.g. a stethoscope); the anatomical site, to in-

dicate the speciﬁc body location in which the observation was taken; and the state of

the patient, which is intended to record the state of the subject of the observation during

the observation process. In order to represent the information about the protocol in a

controlled way, we advocate for using the terms of an ontology that comprises classes

from the Device and Procedure categories of SNOMED-CT[9]. Moreover, in order to

represent anatomical information, the terms of the Foundational Model of Anatomy

ontology[10] are suggested. Finally, we have developed one ontology for describing

information about the state of the patient, which contains 121 classes divided into 28

categories to represent states such as the level of exertion (low, medium, high intensity)

or the position of the patient (standing, sitting,...).

On the other hand, composite observations are composed of two or more observa-

tions, either simple or composite. They are intended to represent observations of phe-

nomena such as the Blood Pressure, which is composed of the systolic and diastolic

blood pressures. Below, we present some OWL2

axioms that represent the general

terms for representing medical observations

c:Observation ≡ c:Simple Obs ⊔ c:Comp Obs

c:Simple Obs ≡ =0 c:comp

c:Simple Obs ⊑ =1 c:value ⊓ ≤ 1 c:unit ⊓ ≤ 1 c:protocol.c:Protocol ⊓

∀c:state.c:State ⊓ =1 c:site.c:AnatomicalSite

c:Comp Obs ≡ ≥ 2 c:comp.c:Observation

Speciﬁc observations are described as specializations of these general terms.

Other main components of the proposal are Application ontologies, which repre-

sent the observations as they are understood in one particular health information sys-

tem. When such a system wants to join the framework, the following steps must be

followed: ﬁrst its Application ontology has to be deﬁned on top of its underlying data

repository. One module named Internal2OntoModule has been developed for that task.

In some cases the module will receive as input a database schema and after apply-

ing a set of rules founded on schema features (tables, keys, inclusion, exclusion and

functional dependencies, null values and semantic integrity constraints), it constructs

the corresponding ontology components (classes, properties, relations and restrictions).

More details about the nature of the rules are described in a previous paper of our re-

search group[12]. In other cases, the input will be an archetype description (e.g. of a

EHR standard) written in Archetype Description Language[13] which is transformed to

OWL. Moreover,this module is responsible for creating the Σ links (Fig.1) that regulate

the information ﬂow between the underlying repositories and the Application ontology,

following the guidelines in [14]. Then, the particular system must import the above de-

scribed Canonical ontology and create the integration mapping that relates the terms of

its Application ontology with the terms in the Canonical ontology. A MappingModule

has been developed for this purpose.

Once a particular system A has joined the framework it is prepared to send infor-

mation about observations stored in its underlying repository to another system B in

the framework. Thanks to the Σ links between the underlying repository and the Ap-

plication ontology of system A, the information to be sent is converted into instances

(individuals) of the classes of that Application ontology. Then, all the implicit knowl-

edge (regarding the individuals) that can be inferred from the Applications layer is

made explicit with the help of a reasoner. At this point, the integration mapping that

For the sake of conciseness we use the Description Logic (DL)[11] representation of axioms.

Throughout the paper, the namespace ’c’ will be used for referring to terms in the Canonical

ontology, and namespaces ’a’ and ’b’ will be used for referring to the ontologies of some

speciﬁc systems A and B

has been deﬁned between the Application ontology of system A and the Canonical on-

tology comes into play and as a reasoning result, the individuals are also classiﬁed as

instances of the concepts of the Canonical ontology. All the inferred knowledge about

the individuals is then sent to system B, which asserts it into its ontology. Since system

B has also imported the Canonical ontology, this is a straightforwardprocess. Moreover,

thanks to the integration mapping between the Application ontology of system B and

the Canonical ontology, the individuals are then recognized as instances of the speciﬁc

terms of system B. Finally, the Σ links between the Application ontology of system B

and its underlying repository allow to assert the knowledge into the latter. The whole

process described above is directed by a reasoner.

To sum up, the main features of the framework presented in this section are the

following:

– It is extensible to any model, either standard or proprietary.

– It is not based on peer-to-peer transformations but on the semantic acknowledge-

ment of one instance of a class in the source ontology as instance of another class

in the target ontology.

– The features of any speciﬁc system remain unknown to the other systems in the

framework. Acknowledging and using the Canonical ontology as a shared model is

enough.

– Reasoning plays a major role in several parts of the framework.

3 Additional Features for Complementing the Proposal

In this section two additional features that complement our approach will be discussed:

path mappings for transforming information about individuals between two ontologies

and rules for knowledge sharing. The usefulness of these features may be also relevant

to other ontology-based interoperability solutions. Prior to that, subsection 3.1 is in-

tended to present the deﬁnitions of the elements that will appear in the examples of the

following subsections.

3.1 Scenario for the Examples

The Revised Trauma Score (RTS)[15] is a physiological scoring system for predicting

death taking into account three measures: the Glasgow Coma Scale value, the Systolic

Blood Pressure and the Respirations Rate. Moreover, the Glasgow Coma Scale(GCS)

is a neurological scale that aims to give a reliable and objective way of recording the

conscious state of a patient[16]. It is calculated from the result of three tests: the eye,

motor and verbal responses.

Subset S

of the Canonical Ontology. It contains the deﬁnitions of the observations

Revised Trauma Score and Glasgow Coma Scale.

c:RTS ≡ c:Comp Obs ⊓ ∃c:comp.c:GCS ⊓ ∃c:comp.c:SysBP

⊓∃c:comp.c:RespRate

c:GCS ≡ c:Comp Obs ⊓ ∃c:comp.c:EyeR ⊓ ∃c:comp.c:VerbalR

⊓∃c:comp.c:MotorR

c:value ∈ owl:DatatypeProperty

Subset S

of the Application Ontology of a Speciﬁc System A. In this subset only

the observations related to the Revised Trauma Score are considered.

a:RTS ≡ ∃a:hasEyeResp.a:EyeResp

⊓∃a:hasMotorResp.a:MotorResp

⊓∃a:hasVerbalResp.a:VerbalResp

⊓∃a:hasSysBP.a:SysBP ⊓ ∃a:hasRespRate.a:RespRate

a:hasValue ∈ owl:DatatypeProperty

Notice the difference in the representation of the a:RTS class with regard to the c:RTS

in the Canonical ontology. While in the latter the class c:GCS is used in the deﬁnition,

in the former the ﬁve values that ultimately are necessary to calculate the RTS score are

indicated directly.

Subset S

of the Ontology of a Speciﬁc System B. In this subset only the observations

related to the Glasgow Coma Scale are considered.

b:GCS ≡ ⊓∃b:hasEyeResponse.b:EyeResponse

⊓∃b:hasVerbalResponse.b:VerbalResponse

⊓∃c:hasMotorResponse.c:MotorResponse

b:hasValue ∈ owl:DatatypeProperty

Integration Mappings.

Finally, let us imagine that the following integration mappings

have been established by the MappingModule:

= hS

, S

{a:RTS ≡ c:RTS, a:EyeResp ≡ c:EyeR, a:RespRate ≡ c:RespRate, a:hasEyeResp ⊑ c:comp,

a:hasSysBP ⊑ c:comp, a:hasRespRate ⊑ c:comp, a:hasValue ≡ c:value}i

= hS

, S

{b:GCS ≡ c:GCS, b:EyeResponse ≡ c:EyeR, b:hasEyeResponse ⊑ c:comp,

b:hasValue ≡ c:value}i

3.2 Path Mappings

As stated in the previous section, the MappingModule is in charge of creating the in-

tegration mapping between the Canonical Layer and the Applications Layer. Thanks

to this integration mapping instances that initially belong to the Application Layer can

be recognized as instances of the Canonical Layer (and viceversa). For example, if the

aforementioned integration mapping I

is considered, given the triples

(b:indGCS rdf:type b:GCS) (b:indGCS b:hasEyeResponse b:indER)

(b:indER rdf:type b:EyeResponse) (b:indER b:hasValue 4)

the reasoner will infer the following statements, which classify all the information about

the individuals b:indGCS and b:indER in the Canonical layer:

For the sake of visual clarity, in this integration mapping we indicate only the axioms related

to the eye response component of the Glasgow Coma Scale. Please assume that the other two

components are treated accordingly.

(b:indGCS rdf:type c:GCS) (b:indGCS c:comp b:indER)

(b:indER rdf:type c:EyeR) (b:indER c:value 4)

This is a quite straightforward process since the representation of the concept GCS

is similar in both the Canonical ontology and the Application ontology of system B (i.e.

in both cases the class GCS is directly related to each of its three components via an

object property). The problem arises when the representation of the same concept in the

source and target ontology is more heterogeneous than just different names for classes

or properties. Let us compare the deﬁnitions of classes a:RTS and c:RTS in section

3.1. Looking at the description of a:RTS, it can be seen (Fig.2) that any individual

belonging to that class will be directly related to an individual of the class a:EyeResp

via the role a:hasEyeResp (assume the same intuition for the case of the motor and

verbal responses). However, in the case of the descriptions in the Canonical ontology, it

turns out that classes c:RTS and c:EyeR are not directly related, but indirectly: ﬁrst

c:RTS is related to the class c:GCS via the role c:comp and then the class c:GCS is

related to the class c:EyeR again via the role c:comp. Then it could be stated that there

is a simple path between classes a:RTS and a:EyeResp (Fig.2a) and a composite path

between classes c:RTS and c:EyeR (Fig.2b).

Fig.2. Structurally different but semantically equivalent ontology paths.

Intuitively, those two paths could be regarded as equivalent, since their only dif-

ference is from the structural point of view caused by the heterogeneous origin of the

ontologies, not from a semantic point of view. Let us denote those equivalences with the

following statements where the expressions on both sides of the ≡

symbol represent a

path. Each expression begins with a class name that is followed by (one or more) pairs

propertyName[className]:

a:RTS.a:hasEyeResp[a:EyeResp] ≡

c:RTS.c:comp[c:GCS].c:comp[c:EyeR]

a:RTS.a:hasMotorResp[a:MotorResp] ≡

c:RTS.c:comp[c:GCS].c:comp[c:MotorR]

a:RTS.a:hasVerbalResp[a:VerbalResp] ≡

c:RTS.c:comp[c:GCS].c:comp[c:VerbalR]

For that reason, we have decided to incorporate a new kind of mappings to our

framework: the so called path mappings, which establish equivalence or subsumption

relations between two ontology paths. Path mappings are useful at the time of trans-

forming individuals from one ontology so that they meet the requirements of the target

ontology. The implementation of path mappings is done by using SWRL rules. For

example, the path mappings shown before would be implemented using the following

rule:

a:RTS(?r) ∧ c:RTS(?r) ∧ a:hasEyeResp(?r,?e) ∧ a:hasMotorResp(?r,?m)

∧ a:hasVerbalResp(?r,?v) ∧ swrlx:createOWLThing(?g,?r)

→ c:comp(?r,?g) ∧ c:GCS(?g) ∧ c:comp(?g,?e)∧ c:comp(?g,?m) ∧ c:comp(?g,?v)

Let us look at what happens when system A wants to send the following triples

about a RTS reading to another system

(a:indRTS rdf:type a:RTS) (a:indGCS a:hasEyeResp a:indER)

(a:indER rdf:type a:EyeResp) (a:indER a:hasValue 4)

Following our proposal, thanks to the integration mapping I

in the ﬁrst place

the individuals will be classiﬁed in the Canonical ontology. For example, individuals

a:indRTS and a:indER will be recognized as instances of the classes c:RTS and

c:EyeR respectively. In order to comply with the speciﬁcation of the class c:RTS,

there should be an individual of the class c:GCS that acts as a connection between

individuals a:indRTS and a:indER. That individual is created by the SWRL rule

above, which ﬁres as soon as a:indRTS is recognized as instance of the class c:RTS

(because the rest of the clauses in the body of the rule are also fulﬁlled).

3.3 Knowledge Rules

Up to know, we have presented a solution that allows the interoperability of data about

medical observationsbetween Health Information Systems. However we think that once

this framework is set up, its potential could be enhanced to solve a more ambitious

problem: the possibility of deﬁning and sharing medical knowledge among those sys-

tems. It is widely known that EHRs hold great potential for clinical decision support,

for example by translating practice guidelines into automated reminders and actionable

recommendation[17] which can improve the quality and safety of healthcare as sub-

stantial evidence suggests[18]. Usually, medical experts are in charge of performing

those translation tasks and of incorporating them into their systems, without sharing

them outside their local context. However, it would be interesting to have the option

to spread that knowledge from one ontology to another. For example, widely accepted

knowledge directives could be integrated into the Canonical ontology, and due to the

mappings between the Canonical ontology and the Application ontologies, spread to

the diverse Application ontologies. This could incorporate valuable knowledge into the

systems without too much effort on their part. Accordingly, a speciﬁc system could

deﬁne knowledge directives in its Application ontology and spread them to the Canon-

ical ontology and in consequence to other Application ontologies, although in this case

some supervision should be carried out to avoid that one system infects other systems

with knowledge that is relevant locally but not necessarily relevant for other systems.

An appropriate way of modelling knowledge directives related to diagnoses and

treatment of illnesses is by using rules expressed in a declarative form, since they are

Accordingly for the remaining components of a:RTS.

suitable for obtaining conclusions from a set of data. More precisely, we have chosen

again SWRL as language for representing these rules. For example, one of the rules

that could be deﬁned in the Canonical ontology is shown next. This rule is intended to

calculate the Glasgow Coma Scale value of a patient as the sum of the values of each

of the three components (Eye, motor and verbal response). The result is stored as the

value of the c:value property of the corresponding c:GCS reading.

c:GCS(?g) ∧ c:comp(?g,?e) ∧ c:EyeR(?e) ∧ c:value(?e,?ev) ∧ c:comp(?g,?m)

∧ c:MotorR(?m) ∧ c:value(?m,?mv) ∧ c:comp(?g,?v) ∧ c:VerbalR(?v) ∧ c:value(?v,?vv)

∧ swrlb:add(?emv,?ev,?mv) ∧ swrlb:add(?emvv,?emv,?vv) → c:value(?g,?emvv)

4 Conclusions

In the ﬁrst part of this paper we have presented a framework which has the following

main features: First, it favors the notion of semantic interoperability among health in-

formation systems by using formal ontologies as canonical conceptual models, which

allow to focus on semantic aspects that are independent of the languages or technologies

used to describe EHRs. This reasoning-drivenapproach avoids the need of peer-to-peer

transformations and in addition, the features of any speciﬁc system remain unknown to

the other systems in the framework. Moreover, it favors the notion of extensibility to

different models, since any medical institution can create its own Application ontology

and relate it to the terms of the Canonical layer via an integration mapping. Finally it

facilitates this seamless adaptation by providing of one module that facilitates the task

of obtaining the deﬁnitions of the Application ontology from a particular underlying

system and another module that facilitates the task of linking deﬁnitions of the Appli-

cation ontology to deﬁnitions of the Canonical ontology. In the rest of the paper some

features that complement the proposal have been discussed: the usefulness of path map-

pings in the transformation of ontology individuals and the convenience of using rules

for knowledge representation and sharing.

Acknowledgements

This work is supported by the Spanish Ministry of Education and Science (TIN2007-

68091-C02-01) and the Basque Government (IT-427-07). The work of Idoia Berges is

supported by the Basque Government (Programa de Formaci´on de Investigadores del

Departamento de Educaci´on, Universidades e Investigaci´on).

References

1. Kalra, D., Lewalle, P., Rector, A., Rodrigues, J.M., Stroetmann, K.A., Surjan, G., Ustun, B.,

Virtanen, M., Zanstra, P.E.: Semantic Interoperability for Better Health and Safer Healthcare.

Technical report, European Commission (2009)

2. OWL2 Web Ontology Language (2009) http://www.w3.org/TR/owl2-overview/.

3. Hefﬂin, J., Hendler, J.: Semantic Interoperability on the Web. In: Proceedings of Extreme

Markup Languages 2000, Graphic Communications Association (2000) 111–120

4. Kilic, O., Dogac, A.: Achieving Clinical Statement Interoperability using R-MIM and

Archetype-based Semantic Transformations. IEEE Transactions on Information Technol-

ogy in Biomedicine, 13 (2009) 467–477

5. Bicer, V., Kilic, O., Dogac, A., Laleci, G.B.: Archetype-Based Semantic Interoperability

of Web Service Messages in the Health Care Domain. Int’l Journal on Semantic Web &

Information Systems 1 (2005) 1–22

6. Mart´ınez-Costa, C., Men´arguez-Tortosa, M., Valencia-Garc´ıa, R., Maldonado, J., Fern´andez-

Breis, J.T.: A Model-driven Approach for Representing Clinical Archetypes for Semantic

Web Environments. Journal of Biomedical Informatics 42 (2009) 150–164

7. Swrl (2009) Available at http://www.w3.org/Submission/SWRL/.

8. Kashyap, V., Sheth, A.P.: Semantic and schematic similarities between database objects: A

context based approach. The Very Large Databases Journal 5 (1996) 276–304

9. SNOMED (2009) Available at http://www.ihtsdo.org/snomed-ct/.

10. Rosse, C., Mejino, J.L.V.: A Reference Ontology for Biomedical Informatics: the Founda-

tional Model of Anatomy. Journal of Biomedical Informatics 36 (2003) 478–500

11. Baader, F., Calvanese, D., McGuinness, D., Nardi, D., Patel-Schneider, P., eds.: The De-

scription Logic Handbook. Theory, Implementation and Applications. Cambridge University

Press (2003)

12. Blanco, J.M., Illarramendi, A., Go˜ni, A.: Building a Federated Relational Database Sys-

tem: An Approach Using a Knowledge-Based System. International Journal on Cooperative

Information Systems 3 (1994) 415–456

13. The openEHR Foundation: Archetype Deﬁnition Language (2007)

14. Mena, E., Illarramendi, A., Kashyap, V., Sheth, A.: OBSERVER: An Approach for Query

Processing in Global Information Systems based on Interoperation across Pre-existing On-

tologies. International journal on Distributed And Parallel Databases 8 (2000) 223–272

15. Champion, H.R.: A Revision of the Trauma Score. The Journal of Trauma 29 (1989) 623–

629

16. Teasdale, G., Jennett, B.: Assessment of Coma and Impaired Consciousness. a Practical

Scale. Lancet 2 (1974) 81–84

17. Hoffman, L.: Implementing Electronic Medical Records. Communications of the ACM 52

(2009) 18–20

18. Kawamoto, K., Houlihan, C.A., Balas, E.A., Lobach, D.F.: Improving clinical practice using

clinical decision support systems: a systematic review of trials to identify features critical to

success. BMJ 330 (2005) 765–768