Clinical terminology for personalized medicine: Deploying a common concept model for SNOMED CT and LOINC Observables in service of genomic medicine James R. Campbell MD W. Scott Campbell PhD Departments of Internal Medicine and Pathology University of Nebraska Medical Center Observables Project Group and iPALM SIG; IHTSDO

Outline  ONC terminology model for clinical Interoperability  Terminology challenges created by Personalized Medicine  LOINC – SNOMED CT harmonized model for observable entities  Application of harmonized model to genomic observables/findings  Anatomic and molecular pathology project at UNMC  Deploying observables model extensions for AP/MP:  Ontology model application  Content development  Publication for comment and evaluation

ONC Terminology Model for Semantic Interoperability  Demographics: LOINC, HL7/OMB code set  Social and medical history: SNOMED CT  Problem list/encounter diagnoses: SNOMED CT / ICD-10-CM  Lab results (observables): Lab LOINC  Physical findings: LOINC, SNOMED CT observables  Medication orders: RxNORM, SNOMED CT  Laboratory Orders: LOINC  Immunizations: CVX, MVX  Procedures: CPT, HCPCS  Documents: LOINC

Office National Coordinator : Objectives of Interoperability  Facilitate transitions of care  Promote free flow of results data  Engage patients and families in managing their EHR  Support Public Health  Enhance clinical research within Learning Healthcare Environment

Interoperable Learning Health Ecosystem

Revolution in healthcare  Sequencing of human genome has led to flood of new observational data appearing in scientific literature and now…clinical records  Majority of this clinical data is unstructured and not useful for decision support in the EHR or clinical research  NLP is a second best approach to re-use of this data  Understanding of genetic and molecular basis of human disease now having impact on therapy  Challenge for precision medicine - to have sufficiently detailed molecular genetics observations and diagnostic data for clinical care and research

Limitations of ONC terminologies Research community  NCBI: Gene Ontology; HGNC; OMIM; Orphanet; Protein Ontology; Cell ontology Clinical community  SNOMED CT: –No concept model for subcellular anatomy or molecular structure –No concept model for Observable entity or molecular basis of disease in Clinical findings –Little content, all primitives  LOINC 2.54: –1275 PCR; 1406 MOLGEN; 116 FISH; 1502 CELL MARKERS –Concept model inadequate to fully define what is being result –Provides only tag-level interoperability of molecular data  No meaningful bridge joining genetic research findings with clinical concept models

LOINC – SNOMED CT Harmonization of Observable entities  2008 agreement between Regenstrief (RI) and IHTSDO –Extend and harmonize a shared concept model for |Observable entity| –Map and instantiate LOINC parts in SNOMED CT content –Jointly publish expression data set defining (lab) LOINC concepts within the harmonized concept model –Technology preview alpha January 2016  What is needed to support clinical decision making and research in molecular genetics is a unified domain ontology for Observable entities spanning clinical and research content including genomics

UNMC: Project for structured encoding of AP/MP cancer reports  Objective: Detailed structured reporting of all anatomic and molecular pathology observations for all CAP synoptic cancer worksheets (82 types of malignancies)  Proposal: Analyze detailed semantics of CAP worksheets; apply harmonized concept model in dialogue with Observables project to develop terminology requirements; deploy as real-time structured reporting from COPATH system interfaced to tissue biobank and EPIC. Integrate with Technology Preview from IHTSDO to develop draft domain ontology for Observable Entities.  Tooling: Nebraska Lexicon© extension namespace; SNOWOWL authoring platform; SNOMED CT International + US Extension + Technology preview + Nebraska Lexicon; ELK DL classifier  Penciled into IHTSDO workplan for 2017

UNMC: Project for structured encoding of AP/MP cancer reports  Objective: Detailed structured reporting of all anatomic and molecular pathology observations for all CAP synoptic cancer worksheets (82 types of malignancies)  Proposal: Analyze detailed semantics of CAP worksheets; apply harmonized concept model to develop terminology requirements; deploy as real-time structured reporting from COPATH system interfaced to tissue biobank and EPIC  Tooling: Nebraska Lexicon© extension namespace; SNOWOWL authoring platform; SNOMED CT International + US Extension + Technology preview; ELK DL classifier  Penciled into IHTSDO workplan for 2017

Observables Project: Concept model draft

Development for AP: Techniques and Properties

Expanded Techniques  |Technique(qualifier value)| extend the value set of LOINC parts for the (optional) LOINC attribute of Method

Measurement properties  |Measurement Property (qualifier value)| extend the value set of LOINC parts for attribute Property

Developments for MP: Genes and proteins

Application to MP: Genes and proteins

MP: Immunohistochemistry

Development for MP: Datatypes requirements

Molecular Pathology Finding Fully defining observations of sequence variants

Patient Condition

Terminology summary: Colorectal and Breast Cancer

Observable Entity: Domain Ontology Development

Issues with DL classification across Lab Tech Preview  Testing of model has not extended to permission to define existing grouper concepts in SNOMED CT, ergo DL classification not yet useful across all subsets  LOINC often does not uniformly specify methods and so techniques are sparsely populated in tech preview; clinically useful & understandable Domain Ontology probably requires them

Domain Ontology: Terminology Use Cases  Retrieve all colon cancer tissue specimens which had BRAF gene testing (IHC or sequencing)  Retrieve all breast cancer cases that tested ER-, PR- and Her2/Neu-  Identify all colon cancer cases with high degree microsatellite instability by PCR or absence of mismatch repair protein by IHC  Identify all cases with Stage IIIa non-small cell lung cancer that had EGFR mutation testing by any method

Questions:  What other use cases for clinical genomics should drive the development or application of the observables convergent model?  Requirements for binding the evolving science of genomics to the SNOMED CT / LOINC clinical ontologies?