The Development of the Clinical Trial Ontology to standardize dissemination of clinical trial data. Ravi Shankar

Transcription

1 The Development of the Clinical Trial Ontology to standardize dissemination of clinical trial data Ravi Shankar

2 Open access to clinical trials data advances open science Broad open access to entire clinical trial data is on the rise Tremendous opportunity to evaluate new hypotheses that were not originally formulated in the studies by performing cross analysis of clinical trials by combining with other open biomedical datasets We are developing standards-based methodologies for uniformly representing clinical trial data to support dissemination, integration and analysis.

3 Industry and Government Opening the Clinical Trial Data Vault

4 clinicalstudydatarequest.com SAS Datasets Multiple proprietary data models

5 SAS datasets EXCEL, ASCII files Multiple proprietary data models

6 immport.niaid.nih.gov MySQL dumps Proprietary data model

7 New Open Science by Integrating Open Clinical Trial Data with Biomedical Datasets Looking at data is good; reanalyzing it is great Clinical trial results can be reproduced and expanded Discover new markers for personalized medicine While individual studies are great, pooling studies may be more powerful Pooling experimental results enables new questions Let s learn why trials work and fail Integration and reanalysis requires uniform data representation and easy access to data in the analytical environments

8 Clinical Trial Data To support the integration and analysis of clinical trials data We are creating uniform representations of clinical trial data. We are enabling easy access to uniformly represented clinical trial data in analytical environments. represent Uniformly Represented Clinical Trial Data export bridge Open Clinical Trial Repositories Dissemination Analytical Environments (R, SAS) Analysis

9 The 5 Major Steps from Raw Data to to Publish to Ready-for-analysis 1. Scope the clinical trial data types that we want to include for dissemination purposes. 2. Define a set of clinical trial domain entities (an ontology) required to uniformly represent the clinical trial data. 3. Using the ontology, encode specific clinical trial data to create uniformly represented data. 4. Publish the encoded clinical trial data. 5. Create bridges between the encoded data and analytical environments to provide easy access to data for analysis.

10 Towards Minimum Information About Clinical Trial for Dissemination Study Summary Study Objectives and Outcome Measures Study Design Schedule of Activities Case Report Form Templates Study Subjects Clinical Data Experimental Data Adverse Events Analysis Data

11 Using Semantic web standards to represent and disseminate clinical trial data Semantic web standards allow data to be shared and reused over the web OWL Web Ontology Language SPARQL RDF Query Language RDF Resource Description Framework URI Uniform Resource Identifier

12 Classes (OWL ontology) Instances (RDF triples) CDISC 2013

13 Building a Standards-Based Clinical Trial Ontology a Formal Description of the Clinical Trial Domain Leverage of data elements defined in CDISC Standards e.g. Study, Arm, Study Subject Link with appropriate reference biomedical ontologies Clinical Trial Ontology (CTO) PRM TDM CDASH SDTM Data Dissemination Elements ODM ADaM SEND BRIDG NCI Thesaurus SNOMED CT Use Upper-level, Structural and Domain Ontologies BFO RDF Data Cube GO NDF-RT BAO PRO CL IAO OGMS

14 CDISC Foundational Standards PRM (Protocol Representation Model) Study objectives, methods, organization,... CDASH (Clinical Data Acquisition Standards Harmonization) Content standards for case report forms LAB (Lab Data Model) Interchange of laboratory data between labs and CROs SDTM (Study Data Tabulation Model) Structure for data submission to regulatory agencies ADaM (Analysis Data Model) SDTM complement to detail statistical analysis CDISC

15 A list of terms from CDISC PRM and SDTM standards

16 Basic Formal Ontology (BFO) CDISC

17 independent continuant organism dependent continuant quality temperature occurrent process process profile course of temperature changes John John s temperature John s temperature history CDISC 2013

18 Building down the Clinical Trial Ontology participates in CDISC

19 Describing the Clinical Trial Domain as a Set of OWL Axioms Descriptions of a clinical trial entity such as its parent entity, its relationships with other entities and value restrictions are formally expressed as logical axioms to create the Clinical Trial Ontology InterventionalStudy is a subtype of Study Study has title that is of type string is divided into relates Study to an Arm Study Subject is a subtype of Person with role as Subject Role

20 Representing Clinical Trial Data as a Collection of RDF Triples In RDF, data is represented as triples predicate Subject Object arm1 Using the entities in the CTO ontology (Study, title, is divided into, Planned Activity, etc.), the entire CCTPT_SW01 data can be represented as a collection of RDF triples title CCTPT_SW01 type of type of Arm Study CCTPT_SW01 CCTPT_SW01 Is divided into arm1 A double-blind randomized trial of steroid withdrawal label... arm1 Steroid withdrawal arm

21 Examples of Temporal Constraints in Clinical Trials Plan Participant is ineligible if he/she had vaccination with a live virus within the last 6 weeks before enrollment Visit 17 must occur at least 1 week but no later than 4 weeks after the end of 2003 ragweed season Screening visit evaluations must occur between 30 days prior to Visit -1 and 45 days prior to Visit 0 Monitor cyclosporine levels 3 times per week while in-patient, then weekly as out-patient CDISC 2013

22 Constraint Expression Ontology CDISC

23 Anchored Duration - A temporal constraint expression

24 The Anchored Duration class is used to specify relative start time of a visit Time Axis Visit 0 time 2 weeks ± 3days Visit 1 start time

25 Multi-dimensional clinical data CDISC

26 Exploring RDF Data Cube for representing and analyzing multi-dimensional longitudinal data CUBE - a set of observations indexed by dimensions (e.g., subjects, months after transplant ) describing measures (e.g., standard height) interpreted according to attributes (e.g., z-score) OBSERVATION measured values (with attributes) at dimensions SLICE optional grouping of observations by fixing dimensions

27 RDF Data Cube Vocabulary Based on Statistical Data and Metadata Exchange (SDMX) 2.0 Information Model CDISC

28 The RDF Data Cube can be sliced and diced to analyze clinical data

29 Building the Clinical Trial Ontology in Protege CDISC

30 Representing Clinical Trial Data as a Collection of RDF Triples In RDF, data is represented as triples predicate Subject Object arm1 Using the entities in the CTO ontology (Study, title, is divided into, Planned Activity, etc.), the entire CCTPT_SW01 data can be represented as a collection of RDF triples title CCTPT_SW01 type of type of Arm Study CCTPT_SW01 CCTPT_SW01 Is divided into arm1 A double-blind randomized trial of steroid withdrawal label... arm1 Steroid withdrawal arm

31 Named RDF Graphs The collection of RDF triples can be viewed as a graph CCTPT_SW01_Graph type of Arm Study arm1 type of is divided into title CCTPT_SW01 label Steroid withdrawal arm A double-blind randomized trial of steroid withdrawal

32 Storing the RDF Encoded Data using 4store, an RDF database Clinical Trial Data represent Clinical Trial Ontology (OWL) OWL / RDF Editor Encoded Clinical Trials (RDF) 4store CDISC Standards, Other Ontologies & Linked Data export bridge Open Clinical Trial Repositories Dissemination Analytical Environments (R, SAS) Analysis

33 Querying 4store for Clinical Trial Data using SPARQL Clinical Trial Data represent Clinical Trial Ontology (OWL) OWL / RDF Editor Encoded Clinical Trials (RDF) 4store CDISC Standards, Other Ontologies & Linked Data export Jena RDF API bridge rrdf Open Clinical Trial Repositories Dissemination Clinical Trial Data Objects in R Analysis

34 Loading the RDF Encoded Clinical Trial Data in R a prototype bridge Create R data classes that correspond to the Clinical Trial Ontology (CTO) entities. CTO Entities Study title (string) R Classes Study <- setrefclass("study", fields = list(title = "character )) Query the RDF triples and populate the R classes with clinical trial data. Using R RDF libraries for SPARQL queries CDISC 2013

35 > cctpt_sw01 = loadstudy(<clinical Trial URI>) A single line of R code can load the entire clinical trial data into R for analysis > cctpt_sw01 = loadstudy(" > cctpt_sw01$title title "A DOUBLE-BLIND RANDOMIZED TRIAL OF STEROID WITHDRAWAL IN SIROLIMUS- AND CYCLOSPORINE-TREATED PRIMARY TRANSPLANT RECIPIENTS" > cctpt_sw01$arms [[1]] Reference class object of class "Arm" Field "name": name "Control arm" [[2]] Reference class object of class "Arm" Field "name": name "Steroid withdrawal arm" CDISC 2013

36 CDISC Standards CDISC

37 CDISC Terminologies NCI EVS CDISC

38 CDISC

39 CDISC

40 CDISC

41 CDISC SHARE an electronic library of standards CDISC

42 CDISC

43 CDISC

44 CDISC

45 CDISC

46 CDISC2RDF AstraZeneca, Roche FDA/PhUSE Emerging Technologies / Semantic Technology Draft representations: CDISC