Semantic Interoperability Solution for Multicentric Breast Cancer

Trials at the Integrate EU Project

Sergio Paraiso-Medina

, David Perez-Rey

, Raul Alonso-Calvo

, Brecht Claerhout

Kristof de Schepper

, Philippe Hennebert

, Jérôme Lhaut

, Jasper Van Leeuwen

and Anca Bucur

Grupo de Informática Biomédica, DIA & DLSIIS, Facultad de Informática,

Universidad Politécnica de Madrid, Campus de Montegancedo S/N, 28660 Boadilla del Monte, Madrid, Spain

Custodix NV, Kortrijksesteenweg 214b3, 9830 Saint-Martens-Latem, Belgium

Institut Jules Bordet, 121 boulevard de Waterloo, 1000 Brussels, Belgium

Phillips Research, Healthcare Information Management, High Tech Campus 34, 5656 AE Eindhoven, The Netherlands

Keywords: Semantic Interoperability, Clinical Trials, SNOMED, HL7, INTEGRATE Project, Breast Cancer.

Abstract: The introduction of –omic information within current clinical treatment is one of the main challenges to

transfer the huge amount of genomic-based results. The number of potential translational clinical trials is

therefore experiencing a dramatic increase, with the corresponding increment on patient variability. Such

scenario requires a larger population to recruit a minimum set of patients that may involve multi-centric

trials, with associated challenges on heterogeneous data integration. To ensure sustainability on clinical trial

management, semantic interoperability is one of the main goals addressed by international initiatives such as

the EU funded INTEGRATE project: “Driving Excellence in Integrative Cancer Research”. This paper

describes the approach adopted within an international research initiative, providing a homogeneous

platform to manage clinical information from patients on breast cancer clinical trials. Following the project

“leitmotif” of reusing standards supported by a large community, we have developed a solution providing a

common data model (i.e. HL7 RIM-based), a biomedical domain vocabulary (i.e. SNOMED) as core dataset

and resources from the semantic web community adapted for the biomedical domain. After one year and a

half of collaboration, the INTEGRATE consortium has been able to develop a solution providing the

reasoning capabilities required to solve clinical trial patient recruitment. The next challenge will be to

extend the current solution to support a cohort selection tool allowing prospective analysis and predictive

modeling.

1 INTRODUCTION

Current oncology treatments are introducing a large

number of new variables to current clinical

guidelines. Molecular tests, in addition to traditional

clinical variables, are inducing an explosion in the

number of potential –if not always actual– clinical

trials. In addition, the high specificity of eligibility

criteria, especially molecular criteria, focusing on

sometimes rare gene mutations, makes patient

recruitment more difficult, increasing the need for

multi-centric and international initiatives.

Information systems with complex data from

different institutions have to deal with additional

heterogeneities on different biomedical vocabularies,

data models, security procedures, legislation, etc.

Integration processes, carried out manually until

now, are becoming less and less manageable with

the dramatic increase of variables and centers

involved.

Within such scenario, interoperability among

different systems (i.e. communication and

understanding of data transferred) is essential to

provide a sustainable solution. This work has been

carried out within the three-year EU funded

INTEGRATE (INTEGRATE, 2012) research

project. The main goal of the INTEGRATE platform

is to provide solutions to clinical researchers and the

pharmaceutical industry for sharing of data and

knowledge, support for molecular testing scenario

for patient enrolment in trials, querying trial data,

and building and sharing of predictive models for

response to therapies.

Paraiso-Medina S., Perez-Rey D., Alonso-Calvo R., Claerhout B., de Schepper K., Hennebert P., Lhaut J., Van Leeuwen J. and Bucur A..

Semantic Interoperability Solution for Multicentric Breast Cancer Trials at the Integrate EU Project.

DOI: 10.5220/0004223400340041

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2013), pages 34-41

ISBN: 978-989-8565-37-2

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

This paper describes the approach proposed to

provide semantic interoperability to the

INTEGRATE platform. The Background section

describes previous projects facing similar

challenges. Then, the Semantic Interoperability

Layer section presents the required components and

the query expansion mechanism implemented to

retrieve data with semantic reasoning capabilities.

The Results section describes preliminary

performance tests using real clinical data for patient

recruitment. And finally, the Conclusions section

describes the main contribution of this work and

future lines.

2 BACKGROUND

The latest advances on breast cancer research have

produced a wealth of data. To obtain a proportional

increment of knowledge, previous projects have

been focused on cancer research studies and

heterogeneous biomedical integration such as

caBIG, i2b2, OMOP and ACGT.

The cancer Biomedical Informatics Grid (caBIG)

(Eschenbach von and Buetow, 2006) is an open

source information network deployed in 2003,

allowing cancer researchers to share tools, data and

applications, and to agree upon common standards

and needs. It is based on Open Grid Services

Architecture (OGSA and OGSA-Data Access

Integration, OGSA-DAI (Antonioletti et al., 2005))

and the open source grid computing project, Globus

Toolkit from Globus Alliance (Globus, 2012).

Applications developed for caBIG are highly

dependent on the GRID-based middleware, which

difficult reusing applications outside the caBIG

framework.

Informatics for Integrating Biology and the

Bedside (i2b2) is a framework based on the

Research Patient Data Registry developed at the

Massachusetts General Hospital (Murphy Shawn N

et al., 2010). i2b2’s main goal is to allow researchers

to use the clinical data for discovery research. It is

designed as a set of services, denominated cells,

which fit together in an integrated environment

(called a hive). Every cell is a SOA service like a

file repository, ontology management, data

repository, etc. The data repository cell is designed

as the data warehouse to provide the information of

the users. i2b2 includes an ontology cell to define

the vocabulary, but this vocabulary only allows one

type of relationship, thus medical ontologies like

SNOMED CT (SNOMED, 2012) cannot fully stored

within this implementation.

Observational Medical Outcomes Partnership

(OMOP) was a clinical project to analyze healthcare

databases for studying issues and effects of medical

products (Stang et al., 2010). One of the advantages

of the OMOP is the simple data model. But similar

to i2b2, it has some problems including an ontology

vocabulary like SNOMED CT in the model. In fact,

OMOP only provides a dictionary that performs the

mapping between different data sources and the

database.

Projects described above have been very

valuable in obtaining practical results, but few of

them have exploited the benefits of the current

semantic web tools. In this context, emerged ACGT

(FP6-2005-IST-026996), an EU funded Project

devoted to the development of a technological

platform for supporting clinical trials on cancer

(Martin et al., 2011). The platform included an ad

hoc ontology built specifically for ACGT, i.e. the

Master Ontology on Cancer (MO) (Brochhausen et

al., 2011), as data model and domain vocabulary. A

semantic mediation layer was developed to

dynamically translate queries in terms of the MO to

the concrete schemas of data sources. While

providing an efficient layer for the integrated access

to a set of disparate resources, the complexity of the

MO (over a thousand classes with hundreds of

properties), hindered its use as schema for users to

build meaningful queries.

Further resources, developed within the semantic

web community, have been adapted to the

biomedical domain. Nowadays, there are classifiers

to solve the management of large ontologies such as

SNOMED CT, CEL (Baader et al., 2006),

SNOROCKET (Lawley and Bousquet, 2010) and

ELK (Kazakov et al., 2012) instead of general

purpose reasoners such as Pellet (Parsia and Sirin,

2004), Fact++ (Tsarkov and Horrocks, 2003) or

Hermit (Shearer R et al., 2008). Classifiers are

reasoners with specific algorithms of inferring

optimal for certain types of ontologies. Comparing

classifiers to deal with SNOMED CT, ELK was

more efficient than CEL and SNOROCKET

(Kazakov et al., 2011). And although classified

ontologies are around 75% lighter than the original,

semantic repositories are required to efficiently store

them. Examples include Sesame (Broekstra et al.,

2002), Virtuoso (Erling and Mikhailov, 2009) or

OWLIM (Kiryakov et al., 2005).

These repositories are used in the semantic web

to store data represented in markup languages, with

efficient search engines to extract domain

knowledge in terms of relationships.

SemanticInteroperabilitySolutionforMulticentricBreastCancerTrialsattheIntegrateEUProject

Figure 1: Overview of the INTEGRATE platform general architecture. eCRF: Electronic Case Report Form; EHR:

Electronic Health Record; DWs Datawarehouses.

The review of the state of the art suggested that

semantic resources can still be largely exploited to

improve the semantic interoperability of integrative

solutions.

3 SEMANTIC

INTEROPERABILITY LAYER

Semantic interoperability among applications and

tools is an essential requirement to achieve the main

goal of the INTEGRATE project, i.e. data sharing

for breast cancer clinical trials. The general

architecture of the proposed platform includes

different services, presented in Figure 1.

The INTEGRATE platform has to deal with two

main scenarios, each belonging to one of two

operational domains: (i) Trial Conduct Domain and

(ii) Research Domain. In the Trial Conduct Scenario,

patients are recruited into clinical trials from each

site, and relevant information from CRFs and EHRs

sources is homogeneously represented. In this

scenario, patients should remain identified until the

end of the clinical trial, and therefore, information

should remain distributed at each site. The Research

Domain, however, deals with encoded data to allow

researchers to perform retrospective analysis of

multiple clinical trials or predictive modeling.

To provide homogeneous access to different data

sources, the semantic interoperability layer should

provide a Common Information Model (CIM) to

represent the information. Thus, a common query

endpoint can be provided to retrieve semantically

uniform data. Components required for this task are

shown in Figure 2.

Figure 2: Semantic interoperability layer components.

The CIM proposed for the INTEGRATE

platform semantic layer is comprised by two

components: (i) the Core Dataset (CD) and (ii) the

Common Data Model (CDM). CDM refers to the

schema of the Data Warehouse and CD is the

domain vocabulary of the INTEGRATE platform.

This vocabulary, previously transformed into a

XML-based ontology representation language, is

stored in a semantic web repository. The CD will be

used to extract domain knowledge to retrieve data

HEALTHINF2013-InternationalConferenceonHealthInformatics

stored within the CDM.

In the following sections, a detailed description

of components previously mentioned is provided.

The query expansion subsection describes the

proposed method to semantically retrieve

information from the CDM by considering relations

contained within the domain vocabulary (CD).

3.1 Common Data Model

CDM proposed at INTEGRATE is the structure of

the Data Warehouse, and a wrapper that offers a

SPARQL endpoint. The CDM resolves

heterogeneity problems in different data sources.

Therefore, the CDM acts as the central data model

of the semantic interoperability layer. Information

from the different data sources is stored in different

Data Warehouses (distributed across the

institutions). This information is extracted,

transformed and loaded into those Data Warehouses

by the mapping tools.

HL7 RIM was the standard selected to develop

the common data model for the INTEGRATE

platform. HL7 RIM (HL7, 2012) includes most

common healthcare domains and serves as a general

data model for healthcare administrative and clinical

information. A relational database, based on HL7

RIM was therefore developed. Messages, documents

and rules conforming to that model are also defined

by HL7.

SPARQL was the query language selected to

query the information loaded in the Data

Warehouse. It is “de facto” standard, and W3C

recommendation, for querying RDF in the semantic

web. SPARQL also facilitates the federation of

queries in different data sources. To obtain a

SPARQL endpoint the D2R Server (Bizer C and

Cyganiak R, 2006) has been applied to publish

relational databases on the Semantic Web.

3.2 Core Dataset

Since the HL7 RIM data model does not specify the

vocabulary to be used for semantic representation of

concepts, a choice of domain ontology had to be

made to act as a “lingua franca” and to facilitate

extracting and exchanging information. We

considered different candidates such as SNOMED

CT, LOINC (McDonald CJ et al., 2003) or

MedDRA (Brown EG et al., 1999). SNOMED CT

(one of the largest medical ontology, developed,

distributed and maintained by IHTSDO) was

selected for the INTEGRATE platform. SNOMED

CT also provides mechanisms for identifying post-

coordinated concepts and adding new concepts with

extensions.

SNOMED CT consists of over 400.000 medical

concepts, with about one million descriptions and

more than one million relationships. Therefore, this

large amount of information implies a great

complexity to be managed. We used classifiers to

infer implicitly stated knowledge from explicitly

represented information, thereby eliminating

inconsistencies, incongruities and all types of

information not expected.

Among available classifiers, ELK has been

selected to classify the CD, filtering required

relationships to improve performance. SNOMED

CT was firstly transformed into the Ontology Web

Language (OWL) (McGuinness DL and Van

Harmelen F, 2012). Once a classified version of

SNOMED CT was obtained, it was necessary to use

a semantic repository to store it. Sesame was

selected in this case.

3.3 Query Expansion

To retrieve semantically uniform information from

the CDM, a query expansion method has been

proposed. The objective in this case is to exploit

relationship information contained within the CD

when querying the platform. A data flow of this

process is showed in Figure 3.

The semantic interoperability layer receives a

SPARQL query. If the original query does not need

query expansion for any concept, then the query is

sent directly to the CDM. If the query requires to be

expanded, it is enriched with concepts from the CD

and sent to the CDM. The CD Sesame repository

receives the concept that may require expansion and

includes the corresponding information in the

original query. Queries are enriched from “is_a” tree

structures as showed in Figure 5.

Figure 3: INTEGRATE query expansion method.

SemanticInteroperabilitySolutionforMulticentricBreastCancerTrialsattheIntegrateEUProject

Sesame returns the subclass information of an

Entity in terms of “is_a” relationships. Then, the

Conjunctive Normal Form (CNF) from this concept

is used to expand the original query as shown below.

CONCEPT = Entity OR Subclass

Subclass

OR … OR Subclass

In post-coordination cases, the original query

considers concepts that have to be expanded and the

relationship among them (represented by CDM

structure). Similar to simple query expansion, the

CD Sesame repository receives the concepts that

may require expansion and includes the

corresponding information in the original query

from a tree structure such as Figure 8.

A CNF of the post-coordination concept is then

built (as shown below) and sent to the CDM as

previously explained. Thus,

(CONCEPT = Entity A OR SubclassA

SubclassA

OR … OR SubclassA

)

AND

(RELATIONSHIP = Entity B OR

SubclassB

OR SubclassB

OR … OR

SubclassB

)

Finally, the expanded query is sent to the D2R

wrapper of the CDM and executed in the data

warehouse.

4 RESULTS

A prototype of the semantic interoperability layer,

implementing the query expansion method was

developed using SOAP services and JAVA. Data

constructed based on actual data from 50 patients

from TOP (TOP, 2012) and NeoALTTO (Neo-

ALTTO, 2012) clinical trials, were loaded into the

HL7 RIM-based common data model and used to

test the INTEGRATE semantic interoperability

layer. Queries were built to match eligibility criteria

(EC) from those clinical trials.

The following eligibility criteria did not require

query expansion:

 Inclusion criterion 2: Age of patient <= 70 years

 Inclusion criterion 3: Female patient

 Inclusion criterion 6: Patients with fixed samples

from the primary tumor

 Inclusion criterion 8: Patients who signed the

informed consent

 Inclusion criterion 10a: Patients with

ANC>=1500 mm

 Inclusion criterion 10b: Patients with GOT <=

1.5N

 Inclusion criterion 10c: Patients with

GPT>=1.5N

These EC queries were built using SPARQL and

directly executed on a D2R server that was mapped

onto the INTEGRATE CDM.

The following eligibility criteria did require

simple query expansion or post-coordination query

expansion:

 Exclusion criterion 6: Patients with previous

treatment with anthracyclines

 Exclusion criterion 1: Patients with metastatic

breast cancer

 Inclusion criterion 1: Patients with

Histologically-confirmed breast cancer

The first of these eligibility criteria required to

retrieve information about patient with previous

treatment with Anthracyclines. But it is unlikely that

actual data contains any drug labeled as

Anthracyclines, since it is a family of drugs. In our

case, data contained subclasses of Anthracyclines

such as Daunorubicin, Epirubicin, Idarubicin, etc.

Instead, the query given to the semantic

interoperability layer is showed in Figure 4.

Figure 4: Original SPARQL query on Anthracycline

eligibility criterion.

Figure 5: Anthracycline tree structure.

This SPARQL query, with a function called

isAnySubclassOf, is sent to the semantic

interoperability layer. This function triggers the

query expansion method with the required concept.

HEALTHINF2013-InternationalConferenceonHealthInformatics

Then, the CD retrieves the subclasses of

Anthracycline treatment in SNOMED, obtaining the

following tree structure (Figure 5).

A CNF is then built with the previous tree

structure. This CNF replaces the function

isAnySubclassOf in the given query, which is then

sent to the CDM, as shown in Figure 6.

Figure 6: Final (expanded) SPARQL query on

Anthracyclines eligibility criterion.

Thereby, information about patients with any

kind of anthracyclines treatment is retrieved from

the CDM.

Figure 7: Original SPARQL query on breast cancer (post-

coordinated) eligibility criterion.

Figure 8: Breast Cancer tree structures.

Exclusion criterion number 1 and inclusion

criterion number 1 consult information relates to the

post-coordinated concept breast cancer. This term

links the SNOMED CT concept of Neoplasm with

its target site, in this case Breast. Figure 7 shows the

SPARQL query provided to the semantic

interoperability layer, where the function

isAnySubclassOf triggers the query expansion

method with the SNOMED CT code for Neoplasm

and the SNOMED CT code for the target site

Breast.

Figure 9: Final (expanded) SPARQL query on Breast

Cancer (post-coordinated) eligibility.

As it can be seen in Figure 8 and Figure 9, the

Neoplasm concept and its target site Breast are

expanded by two tree structures. Both structures

replaced the concepts of Breast Cancer in the

original query, by any subclass of Neoplasm (x

subclasses) and Breast (and subclasses). The

expanded query was sent to the CDM, retrieving

results of patients, and fulfilling the initial semantic

capabilities required by a patient recruitment tool for

clinical trials.

5 CONCLUSIONS

This paper has presented the semantic

interoperability approach developed within the

INTEGRATE project. The main challenge was to

provide a homogenous and powerful solution to

facilitate the collaboration among a complex set of

tools required to manage post-genomic clinical trials

on breast cancer.

After a comprehensive review of the literature in

the area, we identified two main issues that have not

been always addressed in previous efforts: (i) solve

post-genomic clinical trial heterogeneities by (ii)

exploiting semantic web technologies. Semantic web

technologies have been extensively developed

during the last years, together with data models and

domain ontologies that required long periods of time

to develop. Specific characteristics of the biomedical

research nowadays require advanced methods to

solve basic problems of interoperability. These

interoperability issues are essential to enhance

SemanticInteroperabilitySolutionforMulticentricBreastCancerTrialsattheIntegrateEUProject

patient care, by supporting integrative prospective

analysis and predictive modeling over multi-centric

datasets.

Patient data from multi-centric and international

clinical trials have been used to test the proposed

solution, suggesting the suitability of the proposed

solution. Next steps of the project will be focused

on: (i) testing performance for large amounts of

patient data (cohort selection), (ii) a cache

implementation to support large reasoning, (iii)

formalization of post-coordination related reasoning

and (iv) extension of such reasoning with new types

of relationships for the query expansion method.

After a year and a half of the INTEGRATE joint

effort, we have already undertaken essential issues

to improve the management of post-genomic clinical

trials. We have adapted and successfully applied

semantic web technologies to the complex domain

of biomedical research. The next steps aim to

support a crucial challenge nowadays, enhancing the

translation of –omic research to improve clinical

practice in oncology patients.

ACKNOWLEDGEMENTS

The present work has been funded by the European

Commission through the INTEGRATE project

(FP7-ICT-2009-6-270253).

REFERENCES

Antonioletti, M., Atkinson Malcolm, Baxter R., Borley A.,

Chue N. P., Collins B., Hardman N., et al., 2005. The

Design and Implementation of Grid Database Services

in OGSA-DAI. In Concurrency and Computation:

Practice and Experience, Volume 17, Issue 2-4, Pages

357-376.

Baader F., Lutz C and Suntisrivaraporn B., 2006. CEL – a

polynomial.time reasoner for life science ontologies.

In Lectures Notes in Computer Science, Volume

4130/2006, 287-291, DOI: 10.1007/11814771_25.

Bizer C. and Cyganiak R., 2006. D2R Server – Publishing

relational databases on the semantic web. In the 5

International Sematic Web Conference (ISWC).

Brochhausen M., Spear A. D., Cocos C., Weiler G.,

Martin L., Anguita A., Stenzhorn H., Daskalaki E.,

Schera F., Schwarz U., Sfakianakis S., Kiefer S., Dörr

M., Graf N. and Tsinakis M. ,2011. The ACGT Master

Ontology and its applications--towards an ontology-

driven cancer research and management system. In J

Biomed Inform; 44(1): 8-25.

Broekstra J, Kampman A and van Harmelen F. 2002.

Sesame: An Architecture for Storing and Querying

RDF and RDF Schema. In: Proceedings of the First

International Semantic Web Conference (ISWC2002).

Number 2342 in Lecture Notes in Computer Science

(LNCS), Springer-Verlag (2002) 54–68.

Brown E. G., Wood L. and Wood S., 1999. The Medical

Dictionary for Regulatory Activities (MedDRA). In

Drug Safety, Volume 20, Number 2, 109-117.

Erling O. and Mikhailov I., 2009. RDF support in the

Virtuso DBMS. In Studies in Computational

Intelligence, Volume 221/2009, 7-24.

Eschenbach A. von and Buetow K., 2006. Cancer

Informatic Vision: caBIG. In Cancer Informatics, vol

2. Pages 22-24.

Globus Toolkit from Globus Alliance. Available at:

http://www.globus.org/toolkit/ [16 July 2012]

HL7, Reference Information Model. Available at:

http://www.hl7.org/implement/standards/rim.cfm [16

July 2012]

INTEGRATE Driving Excellence in Integrative Cancer

Research. Available at: http://www.fp7-integrate.eu/

[16 July 2012]

Kazakov Y, Krötzsch M and Simancik F. 2011.

Concurrent classification of EL ontologies. In Lectures

Notes in Computer Science, Volume 7031/2011, 305-

320.

Kazakov Y., Krötzsch M. and Simancík F. 2012. ELK: a

reasoner for OWL EL ontologies. In Tech. rep.

Kiryakov A., Ognyanov D. and Manov D., 2005. OWLIM

– A pragmatic semantic repository for OWL. In

Lectures Notes in Computer Science, volume

3807/2005, 182-192.

Lawley M. and Bousquet C., 2010. Fast Classification in

Protege: Snorocket as an OWL2 EL Reasoner. In

Australasian Ontology Workshop.

Martin L., Anguita A., Graf N., Tsinakis M., Brochhausen

M., Rüping S., Sfakianakis S, Senqstaq T., Buffa F.

and Stenzhorn H., 2011. ACGT: advancing clinic-

genomic trials on cancer – four years of experience. In

Stud Health Technol Inform, 169:734-8.

McDonald C. J., Huff S. M., Suico J. G., Hill G., Leavelle

D., Aller R., Forrey A., Mercer K., DeMoor G., Hook

J., Williams W., Case J. and Maloney P., 2003.

LOINC, a universal standard for identifying laboratory

observations: a 5-year update. In Clinical Chemistry,

vol 49. No. 4, 624-633.

McGuinness D. L. and Van Harmelen F., 2004. OWL

Web Ontology Language Overview. W3C

Recommendation, available at: http://www.w3.org/

TR/owl-features/ [16 July 2012]

Murphy Shawn N, Weber G, Mendis M, Gainer V, Chueh

HC, Churchill S and Kohane I. 2010. Serving the

enterprise and beyond with informatics for integrating

biology and the bedside (i2b2). In J Am Med Inform

Assoc 2010; 17:124-130.

Neo-ALTTO, Neo-Adjunvant Lapatinib and Trastuzumab

Treatment Optimisation Trial. Available at:

http://www.alttotrials.com/neoaltto.php [16 July,

2012]

Parsia B. and Sirin E., 2004. Pellet: An OWL DL

reasoner. In ISWC 2004, 2004. ISWC.

Shearer R, Motik B and Horrocks I. 2008. HermiT: a

HEALTHINF2013-InternationalConferenceonHealthInformatics

highly-efficient OWL reasoner. In Proceedings of the

5th International Workshop on OWL: Experiences and

Directions (OWLED 2008): 26-27.

SNOMED Clinical Terms Core Content. Available at:

http://www.ihtsdo.org/snomed-ct/ [16 July 2012]

Stang P. E., Ryan P. B., Racoosin J. A., Overhage J. M.,

Hartzema AG, Reich C, Welebob E., Scarnecchia T

and Woodcock J., 2010. Advancing the science for

active surveillance: rationale and design for the

Observational Medical Outcomes Partnership. In Ann

Intern Med. 2010 Nov 2;153(9):600-6.

Tsarkov D. and Horrocks I., 2003. Reasoner prototype:

Implementing new reasoner with datatypes support. In

WonderWeb Project Deliverable.

TOP, Jules Bordet Institute. Topoisomerase II Alpha Gene

Amplification and Protein Overexpression Predicting

Efficacy of Epirubicin (TOP). Available at:

http://clinicaltrials.gov/ct2/show/NCT00162812 [16

July, 2012]

SemanticInteroperabilitySolutionforMulticentricBreastCancerTrialsattheIntegrateEUProject