Transcript Clarity Medication Mapping

Clarity Medication Mapping to
NDF-RT
Design and Current Status
Outline
•
•
•
•
•
•
Brief Tour of RxNorm Tables Used
Design
Current Status (Results)
Next Steps
Code Walkthrough
Discussion
RxNorm Concept Names and Sources
(rxnconso)
http://www.nlm.nih.gov/research/umls/rxnorm/docs/2012/rxnorm_doco_full_2012-3.html
(starting section 12.4)
• “Primary” table – consists of all RxNorm Concepts
– Example: A medication and synonyms - there may be several rows for a single
concept.
– Disulfiram (generic) and Antabuse (brand name) are both the same concept and
have the same RxCUI.
• RxCUI: Concept Unique Identifier (unique per concept, may be
many rows with the same CUI)
• RxAUI: Atom Unique Identifier (unique per entry in the table)
• SCUI: Source-asserted Concept Identifier
– The identifier as provided by the source (NDDF, NDFRT, RXNORM, etc.)
• TTY: Term Type (preferred term, synonym, ingredient, etc.)
RxNorm Tables: Others
• Simple Concept and Atom Attributes (RXNSAT)
– Example: Used to match NDC and find VA Class types
• Related Concepts (RXNREL)
– Example: Parent/Child relationships of VA classes,
“ingredient_of”, etc.
• Source Information (RXNSAB)
– Source abbreviation/full name (NDFRT/National Drug File),
version, etc.
• Documentation for Abbreviated Values (RXNDOC)
– Full name for abbreviations used in other tables
RxNav: Relationships
http://rxnav.nlm.nih.gov
VA Class Ontology
http://bioportal.bioontology.org/ontologies/47101/?p=terms&conceptid=N0000029067
Map Clarity Medications to RxCUI:
GCN/NDC
• Clarity Medication List
– clarity.clarity_medication
• GCN (Generic Code Sequence Number - First
Databank Inc.)
– clarity.rx_med_gcnseqno
– rxnorm.rxnconso (code column when sab = NDDF, and tty
!= ‘IN’)
• NDC (National Drug Code)
– clarity.clarity_ndc_codes
– rxnorm.rxnsat (atv column where atn = NDC)
The Leftovers: Match with MedEx NLP
http://knowledgemap.mc.vanderbilt.edu/research/content/medex-tool-finding-medication-information
• For the medications that don’t match using
GCN/NDC, use MedEx (NLP)
– map directly to RxCUI via the drug name in clarity
– “NAME” (arbitrarily preferred)
– “GENERIC_NAME”
• Issues
– Closed source (though, open source soon as per authors)
– Windows Only right now (Linux binaries won’t run with our
current configuration on our servers)
– Not integrated into our ETL (“manual technical-debt”)
– Linking results with input is problematic
Map to Drug Form and VA Class
• Map Medications to Semantic Clinical Drug
and Form (SCDF) or Semantic Branded Drug
and Form (SBDF)
– Example Clarity Medication: “ANTABUSE 250 MG
PO TAB”
– Example SBDF: “Disulfiram Oral Tablet”
• Map Medications to Veterans Administration
class (VA Class)
• Example: “[AD100] ALCOHOL DETERRENTS”
Resulting I2B2 Hierarchy
The Leftovers:
No SCDF, SBDF, or VA Class!
• Some medications didn’t map directly to SCDF,
SBDF, or VA Class
– Sometimes, it was because the drug mapped to an
ingredient.
– Example: “CEFAZOLIN INJ 1GM IVP” (medication id
210319, MedEx mapped to RxCui 2180 “CEFAZOLIN” an
“ingredient”)
The Leftovers:
Map via “ingredient” relationships
• Use “ingredient_of” and “constitutes”
relationships
• Use “isa” relationships to get SCDF/SBDF
• Help! Results in 21.7 Million results from 20,354
Medications!
– A huge number of components, packs, and associated
SCDFs/SBDFs
• Reduce this by mapping to the SCDF/SBDFs we
already have mapped from direct links
– Is there a better way?
RxNav (Cefazolin)
Putting Relationships Together
i2b2 Ontology
• Use prior mappings (Medications to
SCDF/SBDF and Medications to VA Class) to
then map the SCDF/SBDF to VA class.
• Create table with parent/child relationships
– Use these relationships to build i2b2 compatible
ontology
Resulting I2B2 Hierarchy
Results
Based on June 2012 data (Cimarron)
• “Round 1”:
– GCN + NDC Mapping
– 89.4% of medication observations covered
(100,395,527 total facts, 10,636,780 missing facts)
• “Round 2”:
– Added MedEx NLP
– linking missing medications to SCDF/SBDF via
"ingredient_of" relationship.
– 94.39% of medication observations (100,395,527 total
facts, 5,630,904 missing facts)
Next Steps
• Peer review of the code!
• Manual mapping of some top concepts
– Problem children thus far:
http://informatics.kumc.edu/work/attachment/ticket/1246/unmapped_meds_20120823.csv
• Review in more detail code from Dustin Key
from Group Health (ghc.org)
– Basic approach is the same as per overview
• How to test/validate?
References
RxNorm documentation
http://www.nlm.nih.gov/research/umls/rxnorm/docs/2012/rxnorm_doco_full_2012-3.html
KUMC Work Ticket
http://informatics.kumc.edu/work/ticket/1246
UMLS Reference Manual
http://www.ncbi.nlm.nih.gov/books/NBK9676/
RxNav
http://rxnav.nlm.nih.gov/
BioOntology.org
http://bioportal.bioontology.org
Paper: “Enabling Hierarchical View of RxNorm with NDF-RT Drug Classes”
http://www.ncbi.nlm.nih.gov/pubmed/21347044
MedEx
http://knowledgemap.mc.vanderbilt.edu/research/content/medex-tool-finding-medication-information
Code Walkthrough!
epic_med_mapping.sql