Enhancing interoperation an i 2 b 2 ETL

Enhancing interoperation: an i 2 b 2 ETL schema for Epic EHRs James R. Campbell MD James Mc. Clay MD Departments of Internal Medicine & Emergency Medicine University of Nebraska Medical Center

Outline ØOrganizing i 2 b 2 for interoperation ØI 2 b 2 Extract, Transfer and Load architecture for Epic EHRs ØData warehouse extracts vs CCDA vs FHIR interface for transportability of code

Operational Expectations of i 2 b 2 load ØQuery / aggregate data across collaborators with little or no mapping Ø(Move towards US standard data model) ØStore facts so that query by value is supported: Numeric, Code lists, structured text ØNarrative reflects content of many EHRs but is not interoperable and should be structured when extracted

ONC Top Level Model for Semantic Interoperability Information model: (Clinical) Ø Demographics: LOINC, Observables HL 7/OMB code set Ø Social and medical history: Findings SNOMED and CT Situations Ø Problem list/encounter diagnoses: SNOMED CT Findings, Events and Situations ICD-10 -CM Ø /Lab results Ø Lab resultsand (observables): Lab LOINC (Laboratory) Observables Radiology other test results: LOINC, Observables SNOMED CT Ø Physical findings: (Clinical) Ø observables Medication orders: Ø orders: Rx. NORM, SNOMED CT Ø Medication Laboratory Orders: Observables Ø Orders: (Laboratory) LOINC Ø Laboratory Immunizations: Ø CVX, MVX Ø Immunizations: Procedures: Ø Ø Procedures: Documents: CPT, HCPCS Ø Documents: LOINC

ONC Terminology Model for Semantic Interoperability Ø Demographics: (Clinical) LOINC, Observables HL 7/OMB code set Ø Social and medical history: Findings SNOMED and CT Situations Ø Problem list/encounter diagnoses: SNOMED CT Findings, Events and Situations / ICD-10 -CM Ø Lab results (observables): (Laboratory) Lab LOINC Observables Ø Physical findings: (Clinical) LOINC, Observables SNOMED CT observables Ø Medication orders: Rx. NORM, SNOMED CT Observables Ø Laboratory Orders: (Laboratory) LOINC Ø Immunizations: CVX, MVX Ø Procedures: CPT, HCPCS Ø Documents: LOINC

I 2 b 2 Star schema: One fact per record (= One question + answer) i 2 b 2 Observation fact Patient Encounter Concept_CD Observation Fact Modifier_CD Instance_num Provider VALTYPE_CD UNITS_CD TVAL_CHAR NVAL_NUM OBSERVATION_BLOB Start_date End_date

Desiderata for interoperability of OBSERVATION_FACTs (OFs) Ø When possible for observables, CONCEPT_CD should align with interoperability reference standards Ø When coded ontologies are the answer, use…MODIFIER_CD for imposing information model context Ø Complex data records (allergy list, medication orders) should be organized into set of facts that are organized by content with MODIFIER_CD linked by INSTANCE_NUM Ø Precoordinate results into CONCEPT_CD only if valueset is small and there is no reference observable code Ø Choice of TVAL_CHAR, NVAL_NUM or BLOB for results should reflect datatypes; use published valuesets always for interoperability

What is the patient hemoglobin? “ 13. 2 mg/dl” i 2 b 2 Observation fact Patient Encounter Concept_CD LOINC: 2951 -2 Provider Observation Fact 13. 2 Mg/dl Modifier_CD 11/1/2016 7: 30 AM End_date

Epic Laboratory in i 2 b 2 Maps LRR CLARITY_COMPONENT 1534435|“Hemoglobin”|”HGB”|LOINC: 718 -7| Record Data ORDER_RESULTS Laboratory results Ø LABS_TRANSFORM. sql

What is the patient problem? “Breast cancer” i 2 b 2 Observation fact Patient Encounter Concept_CD SNOMEDCT: 254837009 (Malignant tumor of breast) Provider Observation Fact Modifier_CD DX|PROBACTIVE Start_date End_date

I 2 b 2 Metadata Ø I 2 b 2 client employs reference ontologies such as ONC terminologies in the user interface: – displays the hierarchical structure of the terminology which may be useful for concept navigation – supports queries of sets of concepts (hierarchical sub-trees) supported by boolean logic

“Severe allergic rx to aspirin with anaphylaxis” i 2 b 2 Observation fact Patient Encounter Concept_CD Observation Fact SNOMEDCT: 39579001 (Anaphylaxis) Modifier_CD Instance: 10030 DX|ALGRX Start_date Provider End_date

Nebraska Medicine i 2 b 2 Data Architecture I 2 b 2 Information Class Standard Metadata Ontology ADT history Epic Facility Cancer registry ICD-O Clinical measurements Social history LOINC; SNOMED CT Demographics LOINC; SNOMED CT Diagnoses (Encounter dx; Problems; Past Med History) (ICD-9 -CM); ICD-10 -CM; SNOMED CT Encounters Epic Facility; Encounter classes Laboratory results LOINC Medications (Orders and Rx; Dispense records) RXNORM; NDC Procedures (Professional services; Hospital procedures; Procedure history ) CPT; (ICD-9 -CM); ICD-10 -PCS; HCPCS; SNOMED CT

Loading an i 2 b 2 warehouse from Epic with ONC terminology Ø Install and maintain terminology maps in Epic and i 2 b 2 Ø Refresh research Clarity (Epic includes mapping data) Ø Run ETLs and load (identified) staging tables Ø Obfuscate dates, anonymize patients and encounters and populate (identified and deidentified) OBSERVATION_FACT tables Ø Install and maintain i 2 b 2 standards metadata & metadataxml for browsing and query Ø Extract CDMV 3 tables (SAS) from OBSERVATION_FACT

I 2 b 2 Load Documentation on the Epic User. Web https: //datahandbook. epic. com/Reports/Details/9000400 Ø UNMC i 2 b 2 ETL Procedures 20160803. docx Ø Identified (idwk) dataset procedures: – Heronloader data extracts and table builds(ETLs) – Blueheronmetadata build Ø De-identified (deid) dataset procedures: – Heronloader extracts and table builds – Blueheronmetadata – CDMV 3 extracts from deid Ø Python scripts: – Fact counter code

FHIR datatypes supported by Epic Ø Ø Ø Ø Ø *Adverse reaction *Allergy/Intolerance *Conditions Devices Document Reference *Family History Goals Immunizations Lab results *Medications *Prescription *Patient *Practitioner Procedures *Substance *Social history; smoking status *Vital signs *2015 release

Questions? Comments?