Initial Prototype for Clinical Data
Normalization and High Throughput Phenotyping
SHARPn F2F
June 30,2011
PurposeDemonstrate a proof of concept solution, based on
new tools, technology, models and methods.
The prototype demonstrates:– The ability to push unsolicited data using NwHIN exchange
protocols– Conversion and normalization of HL7 2.x lab messages to XML
clinical element model (CEM) instances– Conversion and normalization of HL7 2.x medication orders to
CEMs.– Extraction of medication CEM instances from narrative clinical
documents using NLP processing– Persistence of CEM instances in a light weight SQL database– Phenotype processing across the CEM database utilizing the Drools
rules engine
High Level Architecture Diagram
1. Use Data from IHC (De-Identified) HL7 2.x messages2. Send data into Mirth Connect on the IHC side3. Create NwHIN Document Submission (XDR) message using HL7 2.x message as payload4. Send Document Submission (XDR) message from Mirth to IHC NwHIN Aurion Gateway5. Send XDR message from IHC Aurion Gateway to SHARP NwHIN Aurion Gateway6. Send XDR message from SHARP NwHIN Aurion to Mirth Connect6a. Send Mayo HL7 2.x Lab Messages & Clinical Documents to Mirth Connect 7. Process HL7 2.x messages and/or clinical documents in the UIMA Pipelines, to normalize and transform into Clinical Element Model (CEM) instances8. Send the resulting XML instance of Clinical Element Model (CEM) to Mirth Connect9. Persist Clinical Element Model (CEM) instances to MySql database.10. Perform phenotype processing on the CEM instance database.
SHARP Processing Sequence
10
IHC(Backend CDR
Systems)
MirthConnect
IHC NwHINAurion
Gateway
SHARP NwHINAurion
Gateway
MirthConnect
UIMAPipeline
CEM Instance Database
1
2
3
4
5
6
7
8
9
6a
Firewall Firewall
MayoEDT System
Mirth Connect
Enables information flow and transformationMirth channel receives message from some source,
transforms it, and routes it to one or more destinations
Product is open sourceNwHIN with Aurion/CONNECT can be source or
destination of a channelUsed to store CEM Instances to the databaseCan be used to route data to other locations or
databases
Mayo EDTMayo EDT
cTAKEScTAKEScTAKEScTAKES
cTAKEScTAKES
CDA for Meds
HL7 for labsMirthMirth
CEM
SharpDbSharpDb
High level flow - Mayo
cTAKES(NLP)
cTAKES(NLP)
Mayo EDTMayo EDTAdminDiagnosis
processorAdminDiagnosis
processorTabular data CEM
Custom UIMA
pipeline
Custom UIMA
pipeline
Custom UIMA
pipeline
Custom UIMA
pipeline
Configurable UIMA
pipeline
Configurable UIMA
pipelineCEM
Medication to CEM - Mayo data
CDACDACDA-
InitializerCDA-
Initializer
SentenceAnnotatorSentenceAnnotator
TokenizerAnnotatorTokenizerAnnotator LVGLVG
ContextDependentTokenizer
ContextDependentTokenizer
POSTaggerPOS
Tagger ChunkerChunker
DictionaryLookup
Annotator
DictionaryLookup
Annotator
DrugMention
Annotator
DrugMention
Annotator
DrugCEMCAS
Consumer
DrugCEMCAS
Consumer
MirthMirth
SharpDbSharpDb
cTAKES UIMA Annotators (NLP)
Patient count – 10000CDA document count - 360452CEM count for medication – 3442000
IHC-Medication, Mayo, IHC LAB to CEM
HL7MedsHL7
Meds
HL7Initializer
HL7Initializer
IHC-GCNTO-
RXNORMAnnotator
IHC-GCNTO-
RXNORMAnnotator
DrugCEMCAS
Consumer
DrugCEMCAS
Consumer
MirthMirth
SharpDbSharpDb
HL7LabsHL7Labs
HL7Initializer
HL7Initializer
Generic-LAB-
Annotator
Generic-LAB-
Annotator
LABCEMCAS
Consumer
LABCEMCAS
Consumer
MayoLOINC
resource
MayoLOINC
resource
IHCLOINC
resource
IHCLOINC
resource
IHCRXNORMresource
IHCRXNORMresource
New UIMA Process Nodes
SharpDB a CEM Instance Database
Phenotyping (Drools)
Business Logic
Clinical Element
Database
List ofDiabetic Patients
Data Access Layer
Transformation Layer
Inference/ workflow
Engine (Drools)
Service for Creating Output (File, Database,
etc)
Transform physical representation Normalized logical representation (Fact Model)
Completed Work Installation of informatics “SHARP” Cloud system at Mayo Installation and configuration of tools on IHC side and
SHARP Cloud “Tracer Message” processing
– Used to test communication throughout system– Successful transfer using NwHIN/Aurion of test message
between IHC & Mayo 30 de-id IHC patients through pipeline/Drools end-to-end
– 134 Thousand CEMS generated Extraction and message generation for 10,000 patients Processing of 10,000 patients Meds, Labs, Billing data
– 15 Million CEMS generated Conversion to selected CEM models via UIMA framework Persisted from CEM to MySQL
Completed Work (Cont.)
Produced New XML Schemas for CEM Models– Standard lab panel– Ambulatory medication order– Administrative diagnosis
These three models were usedfor the prototype experiment.
Excerpt of Lab CEM instance
CEM Search Tool:http://intermountainhealthcare.org/cem
• Mirth Enhancements– Implemented NwHIN XDR connector capability– Implemented UIMA connector capability– Created NwHIN Aurion XDR adapter
• Channels Created
Completed Work (Cont.)
Sample XDR Channel Channel that receives HL7 2.x message, places the message as the payload of an XDR message and sends it to a remote NwHIN gateway
ReceiveXDRMessage Receives an XDR message from Mirth and extracts the HL7 2.x message
CemAdminDxtoDatabase Receives an XML instance of the administrative diagnosis CEM and persists it to the database
CemLabToDatabase Receives an XML instance of a standard lab panel CEM and persists it to the database
CemMedicationToDatabase
Receives an XML instance of a medication CEM and persists it to the database
Dual Security Certificate Exchange
SHARP/Mayo Cloud
Intermountain Healthcare
Mirth Aurion Gateway
SHARP Aurion Gateway
IHC Proxy
SHARP Proxy
Mirth
Internet
Thank You!
Calvin BeebeChristopher ChuteCraig ParkerCui TaoCyndalynn TilleyDavid MeadDingcheng LiDonna IhrkeGerald BortisGuergana Savova
James MasanzJeff FerraroJohn HolmanJon TeichrowKevin BruceKyle MarchantLes WestbergMargarita SordoMat BockolMichael Turk
Mitch DempseyNathan DavisPei ChenSean MurphySridhar DwarkanathStan HuffSusan WelchTim PetersTom OnikiVinod Kaggal