Data Definition, Data Capture, Quality Assurance and Adherence to Informatics Standards (including UDI) in CRNs James E. Tcheng, MD, FACC, FSCAI Professor of Medicine Professor of Community and Family Medicine (Informatics) james.tcheng@duke.edu The image part with relationship ID rid8 was not found in the file.
A Few Highlights Califf call to solve the data sharing problem
A Few Highlights Platt an enormous lift to understand the meaning of data at an individual site, and to keep data coordinated (in reference to SENTINEL) Matheny translation of data between data models is not particularly difficult (in reference to SENTINEL, PCORNet, i2b2, OHDSI) problem is CDM implementation and its continued coordination
If you don t know where you are going, chances are you will end up somewhere else. -- Yogi Berra (1925-2015)
Informatics and CRNs Data definition (CDEs, CDMs) Data capture (distributed across roles and integrated into workflows) Data quality (from document model to data model - not charting by exception) UDI (and GUDID, AUDI)
Informatics, CRNs and Opportunities Common data models (generic data aggregation) SENTINEL, PCORNet, i2b2, OHDSI Data transfer between data models Data representation / transfer between systems HL7 FHIR portfolio Native, interoperable standardization Minimum core (domain-specific) data elements UDI: reference data in GUDID, AUDI databases Common core data elements Common administrative / demographic data elements Structured reporting
Informatics, CRNs and Opportunities Common data models (generic data aggregation) SENTINEL, PCORNet, i2b2, OHDSI Data transfer between data models Data representation / transfer between systems HL7 FHIR portfolio Native, interoperable standardization Minimum core (domain-specific) data elements UDI: reference data in GUDID, AUDI databases Common core data elements Common administrative / demographic data elements Structured reporting
THE Foundational Issue Tower of Babel Pieter Bruegel the Elder and Pieter Bruegel the Younger, 1563
The Foundational Solution Native, Interoperable Data Standardization Electronic health information Electronic Health Records Procedure reporting systems Registries, CRNs Clinical contexts
Data Challenge: Multiple Masters Recipients Producers (aka Customer #1) Clinical care Health system Payers Patients Federal, state programs FDA Registries Research Oh yes clinicians
Customer #2: Database Developer So you want to build a database Database field label (i.e., what do you call the data element, aka the address of the data in the database) Data type / format (e.g., text string, integer, date, constrained list, ) Business rules (e.g., range limits, consistency checks, )
The Approach Native, Interoperable Data Standardization ü Device evaluation (FDA/MDEpiNet) coronary stents, peripheral artery revasc, bariatric devices, EP CIEDs, ü And their registries ACC NCDR (CathPVI, ICD), SVS VQI, ACS MBSAQIP, à NQRN ü Minimum core (domain-specific) CDEs ü Common CDEs Electronic health information Electronic Health Records Procedure reporting systems Registries, CRNs Clinical contexts
Selecting Domain-Specific Concepts Clean clinical concepts shared across clinical care, procedure documentation, research, and regulatory spaces unique to the discipline and needed for analysis: Define the clinical state of the patient Clinical presentation, risk factors Describe technical aspects Anatomy, procedure details Represent outcomes of interest Think PARSIMONY!
Common Clinical Data Elements Clinical concepts shared across clinical, research, and regulatory contexts NOT unique to the discipline and that already have bindings to standardized terminologies): Demographics, administrative data (ONC) Vital signs, tobacco use history (ONC) Procedure codes (CPT) Laboratory data (LOINC) Medications (RxNorm) UDI and reference device data (GUDID)
Key CDE Metadata Question or prompt May have associated controlled terminology Value, result or answer May have associated controlled terminology HCV status: 1. Clinical concept label (e.g., human prompt for CRF, data entry screen) 2. Db field label (all caps, no spaces, underscores only, limited chars ) 3. Clinical definition of the concept, synonyms thereof 4. Data type / format (e.g., free text, constrained list, integer, ) 5. Allowed values (aka permissible values = value set; VSAC?) 6. Allowed values definitions 7. Business rules (e.g., range / edit checks, consistency, validation) 8. SDO binding(s) 9. Published reference(s)
From Concepts to Action Creating the ecosystem MDEpiNet Projects (BUILD, RAPID, CATNIP, EP PASSION, etc.): identify domain-specific CDEs Clinical concept label, db label, etc. Re-populate GUDID, create AUDI MDEpiNet, NQRN identify common CDEs Leverage FHIR Implement across quality registries of the NQRN Informatics specify / model the technical representation (CIIC, CIMI, NLM, NCI EVS, others)
Informatics, CRNs and Opportunities Native, interoperable CDE standardization Process to identify candidate concepts (domains) Process to develop key metadata (clinicians) Process to formally model (informaticians) Engagement of societies (registry owners) Engagement of documentation system owners (commercial) Engagement of EHR / EHI owners (commercial) UDI
Informatics, CRNs and Opportunities Native, interoperable CDE standardization UDI and the intent to use GUDID (and AUDI) as reference data sources Process to improve GUDID data quality by device category Process to identify clinically relevant concepts for AUDI by device category Process to deliver / create AUDI Other lessons of the Learning UDI Community
Questions? Never, ever think outside of the box!!!
Dammit, Jim, I m a Doctor, Not a Computer!
What is Structured Reporting? Specific data captured by the person closest to that data, integrated into clinical workflow (e.g. MA, tech, RN, pt) Informatics formalisms: universal, well-defined common data elements; data model that parallels (i.e., is representational of) clinical care model Data is compiled by the computer to produce most of the content in a report; MD validates data, focuses on cognitive assessment and recommendations Output: the structured report ROI: data quality /quantity, redundancy / repetition, time to final reports, FTE requirements à augmented knowledge, financial gains
How Is Structured Reporting Done? Engineered, best-practice workflows Just in time, context specific, high usability, point of care data capture via forms Lots of business rules Optimized IT form factors Computer is a compiler In other words Command of who does what when, where, and how
Common CDE Definition Process 1. Determine common data concepts across MDEpiNet projects (and their registries) From registry and industry CRFs 2. Environmental scan to catalog efforts external to MDEpiNet ONC Common Clinical Dataset, PCPI NQRN Registries on FHIR (Seth Blumenthal), HL7 Common Registry Framework, Common Data Models (SENTINEL, PCORNet, OMOP/ODASI), multiple other efforts 3. Develop spreadsheet columns of data element concepts Apply db developer lens: What do I need to build the data element in a database? e.g., Long name, db name, data type, data field constraints, allowed / permissible values, definitions, references 4. Identify and develop / specify candidate terms Domain analysis: cross reference of CDE concepts against existing standards FHIR will likely have specified some of this already at a pragmatic, best fit level ONC Common Clinical Dataset Standards also references existing standards 5. Develop technical representation, bind to existing standards, and test Native interoperability level not just FHIR translation level VQI and CathPVI registries first demonstrations Then ACC NCDR and PCPI NQRN family of registries 6. Develop as a full technical standard logical modeling by CIMI, approval via CIIC Engage EHR, CVIS, EHI system vendors in healthcare