Specification of a Cardiovascular Metadata Model: A Consensus Standard Rebecca Wilgus, RN MSN 1 ; Dana Pinchotti 2 ; Salvatore Mungal 3 ; David F Kong, MD AM 1 ; James E Tcheng, MD 1 ; Brian McCourt 1 1 Duke Clinical Research Institute, Durham NC 2 American College of Cardiology Foundation, Washington DC 3 Duke Cancer Center, Durham NC
Current Landscape Quality Measurement & Patient Safety Clinical Decision Support Public & Population Health Research Business, Operations/ Administration Clinical Data Don Mon, AHIMA
Where we need to be Single Source Patient Clinician Healthcare Data Systems Multiple Uses Data Uses Patient care Quality Improvement Research Reimbursement Post Marketing Safety Decision Support Administration & Mgt. Public Health Reporting McCourt, B. 26SEP2011
CV DAM Project Team
Data Standards Development Methodology 1. Agree on fundamental principles 2. Explicitly define scope 3. Identify, prioritize adoption of prior work, ID gaps 4. Prepare materials for review Aggregate, filter, normalize 5. Clinical expert review Adopt, harmonize, then author 6. Metadata annotation 7. Consensus & Publication 8. Stewardship & Maintenance Nahm, M. & McCourt, B. 26SEP2011
Methodology Rationale Strong project management Structured, coordinated processes Proven track-record Nahm, M. & McCourt, B. 26SEP2011
Fundamental Principles for Cardiology Data Standards Each release will expand the content or extend the applicable use cases of prior efforts Clinical data elements with harmonized, consensus definitions Technical representation for multiple standards CDISC SDTM HL7 RIM Terminologies (i.e., SNOMED-CT) Published metadata Vocabularies (NCI EVS, cadsr)
Scope of the CV DAM Cardiovascular Data CV DAM R1 CV DAM R2 CV DAM R3 Future Non-specialty data Common cardiovascular clinical observations - Sub-specialty domains CDISC CTN BP ACC / AHA/STS FDA *NCRI Grant ACC/AHA/STS registries Cardiac Imaging Demographics Concomitant Medications Adverse Events Vital Signs 18 total domains and growing ACS History & Symptoms Top 100 EHR data elements CV Outcomes TIA / Stroke STEMI/NSTEMI (ACTION) Carotid Artery Stenting and Endarterectomy (CARE) Cardiac Cath and PCI (CathPCI Cardioverter defib procedures (ICD Registry) Congenital Heart Conditions (IMPACT) Echocardiography Nuclear Cardiology Cardiac CT Cardiac MR CV Clinical - Data Elements - Event definitions - Clinical terminology and data definitions CDISC - SDTM standard for FDA submission - Controlled Terminology alignment - CRF templates - Stds adoption by researchers *National Cardiovascular Research Infrastructure HL7 - Mappings to HL7 standards - Adoption support for EHR s - CCHIT EHR Certification (future)
Clinical Data Element Development Process Domain Experts Data Standards Workgroup Informatics Experts 1. Identify DE Sources 2. Aggregate, align for review 3. Select or author definitions (incl. valid values) 4. Annotate with vocabulary, relationships and mapping to technical representations 5. Iterate until clean 6. Public comment & ballot 7. Publish 8. Maintain
Anatomy of a Data Element Class Attribute Name Data Type Alias Heart Failure \ NYHA Class Definitions Coding Instructions Permissible values Representation Maps Tagged Values Name: HL7 RIM Value:RIM Mapping: observation.value Condition: Where observation.code = "heart failure class" Tagged Values Name: CDISC SDTM CDISC SDTM: FA.FATESTCD = HFCLASS, FA.FATEST = Heart Failure Class, FA.OBJ = Heart Failure WHERE MH.TERM = Heart Failure Tagged Values Name: CADSR Local Value Domain Value: NYHAClassType Tagged Values: Name: Property Concept Code Value C-66909 Tagged Value: PropertyConceptPreferredName Value NYHA Class I (C66904) Tagged Value: Name: PropertyQualifierConceptCode Value NYHA Class II (C66905) Tagged Value: Name PropertyConceptCide Citation Nahm, M. & McCourt, B. 26SEP2011
CV DAM
Components of the CV DAM: Use Cases Activity Diagrams UML model Context Document Project description Workflows & processes Conventions utilized & rationale Guidelines for interpretation
Use Cases
Use cases
Activity Diagrams
Activity Diagrams
UML Models
UML Model Leverages the UML formalism for representation of data elements Provides great flexibility and extensibility for metadata to be captured for each data element CDISC SDTM HL7 RIM Select SNOMED-CT mappings Concept codes from NCI EVS & cadsr
CV DAM UML Model Requirements 1. Intuitive for domain experts (clinicians) & technologists 2. Easy to update and maintain 3. Supports multiple diagrams reusing the same data elements 4. Supports metadata for multiple implementations 5. Supports complex relationships that exist among clinical data elements 6. Supports the capture of metadata at the class and attribute level 7. Complies with the NCI s modeling guidelines
Data Element Library
Advantages of the Atomic Class Model:
Characteristics of a Well Modeled Data Element: Data element is essential to the collection or evaluation of data within the use cases Data element is broadly applicable or is so critical that it s absence will bring into question the integrity of the standard Data element has a consensus definition. Consensus means that most WG members have researched, discussed and agreed on the recommended definition, data type, and permissible values. Notes from these discussions are documented. Data type is specified Permissible values (response options) are identified and defined Bibliographic citations are provided or the WG noted they authored the definition
Characteristics of a Well Modeled Data Element: Data elements (and permissible values) have been decomposed into atomic, clinical concepts Atomic concepts have been matched with terms in NCI EVS or new concepts with draft definitions are provided. Technical representations for standards of interest are specified (i.e., HL7 RIM or CDISC SDTM A minimal set of associations of importance to the domain are identified. Associations provide context for the data element (i.e., is an indication for, a finding about, a complication of ). Tells others how or where the domain experts think the data element fits in. Data elements that are derived from other data elements (i.e., max SBP) are identified and the data elements needed for the derivation are fully developed
Structure of the UML Model 349 Attributes 166 Atomic Classes 18 Parent Classes 4 Packages
Details of the UML Model
CV DAM Reports
Limitations of the DAM Tools to develop and represent content have limitations Competing philosophies regarding representation of data standards Terminologies often lack precise clinical definitions that are applicable across use cases Functional models exist for each system a consistent set of requirements that are common across all models are lacking
Advantages of the DAM Precisely defined, consensus, clinical content Common platform of clinical requirements Machine readable links Includes metadata and technical representations that inform multiple implementations Publically reviewed and vetted
Future Uses of CV DAM HL7: CV EHR Functional Profile RIM-derived Messages Structured Reports CDISC SDTM: FDA Clinical Trials Data Warehouse McCourt, B. 10MAY2011
Future Content Development Areas CV Imaging (in progress) FDA CV Endpoints Women s Heart Disease
Thank-you! NCRI Principal Investigators Robert Harrington, MD, FACC Duke Eric Peterson, MD, FACC Duke John Rumsfeld, MD, FACC ACCF National Heart, Lung, and Blood Institute grant: 1RC2HL101512-01 ACC Governance Work Group H. Vernon Anderson, MD, FACC Mark Kremers, MD, FACC Martha Radford, MD, FACC Matthew Roe, MD, FACC Richard Shaw, PhD, FACC James Tcheng, MD, FACC William Weintraub, MD, FACC
Thank-you!
Questions? Rebecca Wilgus: rebecca.wilgus@duke.edu Brian McCourt: brian.mccourt@duke.edu Dana Pinchotti: dpinchotti@acc.org