BRIDG-based Trial Metadata Repository

Need for Standardized Machine Interpretable Trial Descriptions

J. van Leeuwen

, A. Bucur

, B. Claerhout

, K. De Schepper

, D. Perez-Rey

and R. Alonso-Calvo

HIM department, Philips Electronics BV, High Tech Campus 34, Eindhoven, the Netherlands

Custodix NV, Kortrijksesteenweg 214b3, Sint-Martens-Latem, Belgium

Departamento de Inteligencia Artificial, Universidad Politécnica de Madrid, Boadilla del Monte, Madrid, Spain

Keywords: Clinical Trials, Trial Metadata, Trial Metadata Repository, Semantic Interoperability, Trial Eligibility.

Abstract: Making information about clinical trials accessible in a machine interpretable way could aid applications

both in clinical care and clinical research, such as patient screening, trial recruitment, trial meta-analysis,

trial duplication detection and clinical decision support. We present our standards-based trial metadata

repository that captures structured trial information and application-specific formalisms and execution logic

supporting a range of relevant applications, with a focus on interoperability and machine interpretability to

enable more efficient support for clinical research and faster knowledge transfer into care, We further

exemplify the use of the Trial Metadata Repository for a patient screening application for clinical trials.

Additionally, the mechanisms are described to manage the information model of the repository when the

scope is enlarged to additional contexts.

1 INTRODUCTION

From academic medical research centres to

community hospitals and other stakeholders, the

healthcare industry continues to improve its

capabilities for electronic data capture. Ideally,

clinical care and clinical research would live

symbiotically together, resulting in optimal patient

care –based on the latest validated research

findings– and efficient increase of clinical

knowledge (aided by the accessibility of clinical care

information). Currently however, there is a large

separation between clinical care and clinical

research with information typically silo-ed in the

respective contexts.

The execution of clinical trials is an important

vehicle used in clinical research to progress clinical

knowledge. In addition, treatment in a clinical trial is

often a cancer patient’s best option (Edwards, 1998).

The National Comprehensive Cancer Network

advises that the best management of any cancer

patient is in a clinical trial and encourages

participation in clinical trials (NCCN, 2010).

Unfortunately, comprehensive information about

clinical trials is not easily accessible at the point of

care. Typically, the clinical trial protocol is only

accessible in a non-machine-interpretable form (such

as paper or pdf file). The clinical trial protocol

describes amongst others the purpose of the trial, the

clinical rationale, eligibility criteria, and the

schedule and details of the tests, procedures, and/or

medication.

In addition, information about the results of

clinical trials is inefficiently transferred back to

clinical care. Again, information is accessible in a

non-machine-interpretable form (typically in the

form of literature, papers or guidelines) which is

time-consuming to process for the clinical users and

cannot be used for automatic processing in relevant

applications.

As the information is in a non-machine-

interpretable form, it is also not possible to aid the

clinician by targeting the information to the patient

case at hand in a clinical decision support

application.

Capturing the information about clinical trials in

a machine interpretable way could aid applications

both in clinical care and clinical research. For

instance, it could aid a clinician to find a suitable

trial for a patient (patient screening), it could aid a

clinical researcher and a pharmaceutical company to

efficiently recruit participants (trial recruitment), it

453

van Leeuwen J., Bucur A., Claerhout B., De Schepper K., Perez-Rey D. and Alonso-Calvo R..

BRIDG-based Trial Metadata Repository - Need for Standardized Machine Interpretable Trial Descriptions.

DOI: 10.5220/0004910304530458

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2014), pages 453-458

ISBN: 978-989-758-010-9

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

could increase the medical knowledge by allowing

more efficient data analysis across trials (meta-

analysis), it could prevent duplicating the execution

of trials (duplicate detection) and it could aid a

clinician in finding relevant treatment options for a

patient (clinical decision support). Therefore, an

effective solution is required to represent structure

and store this information. We name the repository

storing this information the trial metadata repository.

By convention we call this information the trial

metadata to differentiate from the term “trial data”

which typically refers to the patient data collected

for a clinical trial.

1.1 State of Practice

The realization that information about clinical trials

should be publically available is nowadays common

ground. The World Health Organization (WHO)

publishes the WHO Trial Registration Data Set

(International Clinical Trials Registry Platform,

2013) which specifies the minimum amount of trial

information that must appear in a trials registry. The

WHO site also contains a collection of links to trial

registries (which are typically organized on a

geographical level).

In addition, various countries have made

legislation enforcing companies to publish clinical

trial information.

In the current state of practice, many trials

publish information on clinicaltrials.gov (i.e.

disease, intervention, eligibility criteria, etc.).

Unfortunately, these initiatives are focused on a

textual distribution of the information. We argue that

the information should be made accessible in a

machine interpretable way, allowing for

contextualization of the information given and

enabling a wide range of applications that rely on

access to structured trial information, such as

clinical decision support, trial recruitment, meta-

analysis of trial results, duplicate trial design

detection, etc.

Existing initiatives like linkedct.org (which aims

at publishing an open Semantic Web data source for

clinical trials data) are of limited use as the

information is post-processed from clinicaltrials.gov

and is rather course grained. To illustrate this, the

following criteria text excerpt has been retrieved

from linkedct.org, which is available as blob only

(i.e. not structured):

“DISEASE CHARACTERISTICS: - Histologically

proven metastatic renal cell carcinoma not

amenable to complete surgical resection and

progressive despite immunotherapy -

Bidimensionally evaluable clinically or

radiographically - HLA 6/6 or 5/6 matched family

donor available - No CNS metastases PATIENT

CHARACTERISTICS: Age: - 18 to 80 Performance

status: - ECOG 0 or 1 Life expectancy: - At least 3

months Hematopoietic: - Not specified Hepatic: -

Bilirubin no greater than 4 mg/dL - Transaminases

no greater than 3 times upper limit of normal Renal:

- Creatinine no greater than 2.5 mg/dL - No

malignancy-associated hypercalcemia (< 2.5

mmol/L) Cardiovascular: - Left ventricular ejection

fraction greater than 40% Pulmonary: - DLCO

greater than 65% of predicted Other: - Not pregnant

- HIV negative - No major organ dysfunction that

would preclude transplantation - No other

malignancies except basal cell or squamous cell skin

cancer - No psychiatric disorder or mental

deficiency that would preclude study participation

PRIOR CONCURRENT THERAPY: Biologic

therapy - See Disease Characteristics Chemotherapy

- Not specified Endocrine therapy - Not specified

Radiotherapy - Not specified Surgery - Not specified

Other - At least 1 month since prior treatment for

renal cell carcinoma.” (Hassanzadeh, 2013). This

unfortunately does not allow for contextualization or

processing.

At the same time, the Biomedical Research

Integrated Domain Group (BRIDG) Model

initiative (Biomedical Research Integrated Domain

Group Model, 2013) is gaining traction. The BRIDG

model is a domain analysis model which aims to

provide a shared view of the dynamic and static

semantics for the domain of protocol-driven research

and its associated regulatory artifacts. The BRIDG

model is a collaborative effort spanning important

and relevant standardization bodies like the Clinical

Data Interchange Standards Consortium (CDISC),

the HL7 Regulated Clinical Research Information

Management Technical Committee (RCRIM) Work

Group, the US National Cancer Institute (NCI), and

the US Food and Drug Administration (FDA). This

collection of stakeholders ensures a wide variety of

viewpoints on the model, which increases the

potential for stability of the model. In addition the

BRIDG model has the promise of easing future

interoperability as the various standardization bodies

are defining their new standards based in the BRIDG

model. As the BRIDG model is a domain analysis

model (and a conceptual model for clinical

research), it cannot be used “as is” to implement a

physical design or to generate code. Rather it can be

leveraged to further build out detailed logical models

and physical designs.

BRIDG currently spans the following specialized

HEALTHINF2014-InternationalConferenceonHealthInformatics

454

subdomains: Protocol representation focusses on

planning and design of a research protocol, Study

conduct focusses on the execution of a research

protocol (study conduct and results from the study

activities), Statistical analysis describes the planning

and performance of the statistical analysis of data

collected during clinical trial research and their

relationships, Adverse event focusses on all safety

related activities, e.g. detection, evaluation, follow-

up and reporting, regulatory focusses on

submissions to regulatory authorities.

2 THE TRIAL METADATA

REPOSITORY

In order to support the efficient dissemination of

clinical trial information, interoperability and

machine-interpretability of content should be

important features of a trial metadata repository.

In the context of the INTEGRATE project

(www.fp7-integrate.eu) – a collaboration project

aiming to develop innovative infrastructures to

enable data and knowledge sharing and to foster

large-scale collaboration in biomedical research –

the need arose for a trial metadata repository. The

developed trial metadata repository leverages the

BRIDG domain analysis model for its information

model to facilitate interoperability.

The trial metadata repository uses the BRIDG

domain analysis model by subsetting the model

(selecting the concepts and relations necessary for

the use cases) and subsequently extending the set

with application specific concepts.

The information model for the trial metadata

repository is expressed in the Unified Modeling

Language (UML) (www.uml.org) – a modeling

language that includes a set of graphical notation

techniques to create visual models of object oriented

software-intensive systems (Unified Modeling

Language, 2013) – as is the BRIDG model.

Figure 1 shows an excerpt of the information

model, depicting a version of a study protocol and

its relations to the inclusion and exclusion criteria.

The model is extended with application specific

information where our use cases require trial

metadata (currently) not covered by the BRIDG

model.

Figure 1: Information model excerpt.

BRIDG-basedTrialMetadataRepository-NeedforStandardizedMachineInterpretableTrialDescriptions

455

Figure 2: Script example evaluating the eligibility of a patient requiring the GPT lab value to be less than 1.5 * max.

The trial metadata repository has an administrative

interface enabling users to inspect, input and update

trial descriptions in the trial metadata repository.

This enables to efficiently populate the repository

with new trials, implement changes (e.g. for

approved trial amendments), extend the solution

with new formalisms and deploy the trial metadata

repository for new applications.

2.1 Trial Eligibility Evaluation

A current application for which the trial metadata

repository is essential addresses the use case of a

clinician assessing the eligibility of a patient for

enrolment into a particular clinical trial. In order to

enrol into a clinical trial, the patient must meet the

eligibility criteria of the trial (covering specific

criteria like cancer type, previous treatments, health

status, etc.). The eligibility criteria and their

verification are typically not integrated into the

clinical information systems, but usually the trial

description (with the eligibility criteria) is

distributed as a read-only document (be it

electronically or in print). To verify clinical trial

eligibility, the clinician has to browse through the

clinical information systems in order to find the

required patient information so eligibility status can

be assessed (which can be a time consuming

activity).

In order to provide maximum benefit, the trial

metadata repository should not only provide easy

access to the trial information, but it should also

provide a means to easily connect the information

with patient data. For the trial eligibility use case,

the trial metadata repository heavily leverages

classes from the Protocol Representation of BRIDG

to capture information about clinical trials such as

name and description of the trial, recruitment status,

inclusion and exclusion criteria, current patient

accrual, target patient accrual, due date, etc.

In addition, the information model has been

extended with the ability to associate statements in

different formalisms with each criterion. The

required formalism might differ depending on how

the patient data is stored (syntax and semantics) and

how the patient data is accessible (e.g. using

def query="""SELECT DISTINCT ?obs_id ?value ?max

WHERE {

?instParti a hl7rim:participation;

hl7rim:participation_entityId "$patientID";

hl7rim:participation_act ?instAct.

?instAct hl7rim:act_observation ?instObs.

?instObs hl7rim:observation_code "34608000";

hl7rim:observation_id ?obs_id;

hl7rim:observation_valueST ?value;

hl7rim:observation_refRangeMax ?max;

}""";

Map<String,String> values = semanticLayerWSClientImpl.executeQuery(target,

query);

QueryResult[] result = new QueryResult[1];

result[0] = new QueryResult();

if(values == null){

result[0].setResult(MatchResult.UNDETERMINED);

}else{

if (values.get("value")==null || values.get("max")==null) {

result[0].setResult(MatchResult.UNDETERMINED);

return result; }

String id = values.get("obs_id");

result[0].setResult(MatchResult.NONMATCH);

if (new Float(values.get("value")) < (1.5* new Float(values.get("max"))) ){

result[0].setResult(MatchResult.MATCH);}

Evidence evidence = new Evidence();

evidence.setEvidenceId(id);

result[0].setEvidence(evidence);

};

return result;

HEALTHINF2014-InternationalConferenceonHealthInformatics

456

webservices/odbc/sparql/etc.). For the trial eligibility

application, the trial metadata repository stores

executable logic (in the example below a groovy

script). In order to bridge the gap between trial

information and patient data, standard

ontologies/terminologies are used in the formalisms

to achieve shared semantics.

The INTEGRATE trial eligibility application is

composed of the following components: The

application front-end: the user interface for the

clinical care giver; the criterion matcher: retrieves

the executable logic for a criterion and executes it to

assess eligibility, the trial metadata repository:

contains the trial information and executable logic

for the eligibility criteria, and the semantic data

access service: (semantically enabled) data access

services for patient data.

Figure 3: architecture of the trial eligibility application.

The development of the trial metadata repository

is based on a model driven approach. This choice

has been made as it is expected that the information

model will be regularly extended according the

needs of additional use cases. Technically, a series

of code generators have been created that transform

the UML description of the information model into

the actual trial metadata repository.

The code generators take as input an xml

serialization (xmi) of the UML model and

subsequently:

 create the underlying database

 expose the content via webservices

 create an administrative web interface

 create the documentation of the information

model.

The components comprising the trial eligibility

demonstrator are loosely coupled and the

architecture is a Service Oriented Architecture. The

deployment of the components is flexible and the

trial metadata repository can be deployed in various

context – e.g. in a hospital context, enterprise

context or across enterprises.

In the trial eligibility application, executable

logic is used as formalism. Each criterion is

associated with a groovy script which will be

executed by the criterion matcher. The script

typically contains queries to retrieve the necessary

patient data and subsequently uses that data to

evaluate the criterion. As example a script (Figure 2)

is described that evaluates a laboratory value (the lab

value for GPT (glutamate pyruvate transaminase, an

enzyme measurement associated with liver

functioning) should be lower than 1.5 times the

upper value of the normal range of the laboratory).

The script contains a SPARQL query to retrieve the

relevant patient data. Shared semantics are used in

this demonstrator, binding ah HL7v3 RIM based

information model with an core dataset consisting of

SNOMED, MedDRA and LOINC, and in the

example an observation with SNOMED code

"34608000” (the GPT lab value) is retrieved. The

criterion matcher inserts the correct patient id and

retrieves the observation id (obs_id), the value

(value) and the upper value of the normal value of

range of the lab (max). Subsequently, the scripts

evaluates the criterion (by evaluating whether value

< i.5* max.) Finally, it returns whether there is a

match, a non-match, or that the criterion could not

be evaluated.

2.2 Current and Future Work

We plan to further extend the trial metadata

repository for other use cases, to capture information

about the structure of the clinical trial (a shared

standard representation of the components of a trial)

and to capture the results of trials.

As discussed in (Speedie et al., 2008): “A

standard representation of Clinical trial information

is necessary to clearly and accurately communicate

the structure of a trial for uniform implementation at

multiple sites. One of the challenges in such

multisite trials is consistent implementation, when

numerous individuals at the different sites are

charged with executing the trial. Inconsistencies can

arise from different understandings of the protocol’s

elements. Consistency is supported by a common

understanding of the relevant aspects of the trial.

Aside from needing a standard representation of

clinical trial information to help run a trial, such a

representation is essential for combining results

from multiple heterogeneous clinical trials in a

meta-analysis, where small differences in trial

design and outcome measures may lead to

inaccuracy in the overall effect estimate. The ability

to determine which elements of two or more trials

BRIDG-basedTrialMetadataRepository-NeedforStandardizedMachineInterpretableTrialDescriptions

457

are similar and which are different is critical to

detecting such differences. Without a standard

method of representing the components of a trial, it

is necessary to depend entirely on the interpretations

of readers regarding the comparability of trial

elements. There is an overlapping and equally

important issue of the standard representation and

reporting of clinical data for the purposes of

comparing the results of multiple clinical studies.

Essential to the task of conducting a systematic

review of clinical trials is the need to objectively

evaluate the quality of the trials. For this task, it is

important to be able to understand the design

elements of a given trial and be able to compare it

with others of known quality. These comparisons

require identification and description of trial

components such as treatment allocation strategies,

in clear and unambiguous terms, to make valid

judgments about the overall trial quality. The lack of

a standard representation of trial design features

impedes this process by making it more difficult to

locate and characterize the important elements of a

trial that are used in critical appraisals of trial

evidence. A standard, computable representation

would improve the ability to evaluate the quality of

clinical trials and provide a basis for doing so in an

automated fashion”.

These additions could increase the medical

knowledge by allowing more efficient data analysis

across trials (meta-analysis), it could prevent

execution of similar trials (duplicate detection) and it

could aid a clinician in finding relevant treatment

options for a patient (clinical decision support).

With the addition of more and more types of

information to the information model, it becomes

important to manage data capture in a coherent way.

Information should be sound and complete. In order

to aid the user in capturing sound and complete

information, views will be provided on the

information. Views will be pertinent to specific

application areas - like trial recruitment (with a

focus on eligibility criteria) or trial (meta-)analysis

(with a focus on the components of a trial design).

These views will be captured in the UML model

by the use of tagged values (allowing to add

additional information to UML elements),

leveraging a dedicated UML profile.

3 CONCLUSIONS

The information about clinical trials that is currently

locked away in non-machine-interpretable form

(typically .pdf or paper) can deliver a lot of value to

a wide variety of application in the clinical care and

clinical research domains. We have presented our

ongoing work to unlock this information. For this,

we have devised an information model leveraging

the BRIDG model (to ensure future interoperability)

and implemented a trial metadata repository that

stores in a structured, semantics-aware way relevant

trial information. This information model has been

extended to allow storing different application

specific formalisms and execution logic, for instance

to describe machine-interpretable eligibility criteria.

These additional elements have the role to support a

variety of applications that need access to trial

information.

Finally, we describe how the information model

can be extended for different clinical trial contexts

while ensuring maintainability.

In our future work we will extend the current

solution to support a wide range of applications in

clinical research and clinical care.

ACKNOWLEDGEMENTS

This work has been partially funded by the European

Commission through the INTEGRATE project

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

REFERENCES

Hassanzadeh, O. (2013). Retrieved 11 12, 2013, from The

Linked Clinical Trials (LinkedCT) project:

http://data.linkedct.org/resource/eligibility/52af484a1a

ea77150c3a9f454226b302/

(2013). Retrieved 11 12, 2013, from Biomedical Research

Integrated Domain Group Model:

http://bridgmodel.nci.nih.gov/

International Clinical Trials Registry Platform. (2013).

Retrieved 11 12, 2013, from WHO:

http://www.who.int/ictrp/network/trds/en/index.html

Unified Modeling Language. (2013). Retrieved 11 12,

2013, from Wikipedia: http://en.wikipedia.org/wiki/

Unified_Modeling_Language

Edwards, M. (1998). Access to quality care: Concensus

statement of the american federation of clinical

oncologic societies. Annals of Surgical Oncology,

657-659.

NCCN. (2010). NCCN Clinical Practie Guidelines in

Oncology - Breast Cancer.

Speedie, S., Taweel, A., Sim, I., Arvanitis, T., Delaney,

B., & Peterson, K. (2008). The Primary Care Research

Object Model (PCROM): A Computable Information

Model for Practice-based Primary Care Research. J

Am Med Inform Assoc., 661–670.

HEALTHINF2014-InternationalConferenceonHealthInformatics

458