Presented at CDISC 2009 in Baltimore, it explores what the Semantic Web can bring to Healthcare. Can it be deployed right now? With ease? CDISC sets standards for the exchange of clinical trial data. Once deployed, they remove much of the redundancy and paper processing that characterizes a typical trial today. Its membership includes government regulators like the US FDA, all the major drug companies and their IT vendors.
2. Hoot72.org
• “Demonstrate the power the
Semantic Web brings to Health-Care
and how easy it is to deploy today.”
• Incubate: open source, docs
• Not Green Field - 40+ years of
Health IT
3. : Just Another Format?
• Technology: Web Stack++
• Reuse: HTTP, URIs, not HTML
• + RDF, OWL, SPARQL
• Get Link docs -> Query Link data
• One more reuse: Link ANYWHERE
• Begone CD-ROM: no islands
• WW: new adds on, reuse, open
4. A Linkable Patient
type
Patient
about
personName
observation observationValue
familyName
givenName Doe
CodingSystem Code
middleName
Code CodingSystem
John Fitzgerald
LN 30949-2 005 NIP
Identifiers and Time not shown
URI: http://www.facilityx.com/cdrs/112123449
5. Now Just Ask ...
All Patients with adverse outcome from
vaccine ...
SELECT DISTINCT ?givenName ?familyName
WHERE {
?patient hoot72:personName
[ hoot72:givenName ?givenName ;
hoot72:familyName ?familyName ] .
[ hoot72:about ?patient ;
?assert [ hoot72:nameOfCodingSystem "LN" ;
hoot72:simpleIdentifier "30949-2" ] ]
}
6. Move out and up
• Question: Patients taking “Weight Loss Drugs”
• Patient Web: very particular
• Patient drugs as NDC codes: DESOXYN
TABLETS (00074337701) ...
• Too big a gap?
7. Ontologies Link!
Obese StanDrug: C0025611 Methamphetamine
May Treat
Name Name
SameAs
Stanford Drug
Ontology Methamphetamine RxNorm:6816
Ingredient
NDC: 00074337701
Patient Joe
SameAs
Hoot72 Name
Patient Medication
Graph Desoxyn 5MG Tablet
NDC: 00074337701
RxNorm
* Dotted: composite of links to save space
** w3c HCLS Example
8. The Ontologies?
• “an implementable model of the entities that need
to be understood in common in order for some
group of software systems and their users to
function and communicate at the level required for
a set of tasks” -- Alan Rector
• “Shared Knowledge” for Machines
• Links, hierarchies, equivalence ...
• The “middleware” of the Semantic Web
• OWL (WOL) - Web Ontology Language
10. Not just “Standards”
SameAs
CodingSystem CodingSystem Code
Text
Code
Local 182253 MRSA Culture LN 13317-3
Local Code LOINC Code
• Enable standard, off-the-shelf queries
• Definition is incremental
11. CDISC: it’s the content
• Roadmap: “The separation of content
standards from the means of
transporting that content”
• Terms: to OWL and Endpoints
• “BRIDGing” in OWL
• Trials as querable Graphs (vs docs)
12. Many Users, Contributors
Patient Researcher
Linked Health Data
Doctor Informatics
Insurance Manager
One Semantic Web for Health-Care
13. But ... “Patient Gap”
• “Trapped”, “Silo’ed”
• Ontologies Left Waiting
• EMRs Hold Back
15. Enabler: the Silo’s chat
“HL7 version 2 is a major
breakthrough and market 2.2
success. More than 93% 2.1
3.0 2.3
hospitals in US are using this 2.5
2.4
standard” - Health Level
Horizon (HLH) Project
2.3.1
2.1 2.2 2.3 2.3.1 2.4
2.5 3.0
Source: Neotool, V3 vs V2
16. HL7 “tweet” ...
MSH|^~&|REGADT|MCM|IFENG||199112311501||ADT^A04^ADT_A01|000001|P|2.4|||
EVN|A04|199901101500|199901101400|01||199901101410
PID|||191919^^GENHOS^MR~371-66-9256^^^USSSA^SS|253763|MASSIE^JAMES^A||
19560129|M|||171 ZOBERLEIN^^ISHPEMING^MI^49849^""^||(900)485-5344|
(900)485-5344||S^^HL70002|C^^HL70006|10199925^^^GENHOS^AN|371-66-9256||
NK1|1|MASSIE^ELLEN|SPOUSE^^HL70063|171
ZOBERLEIN^^ISHPEMING^MI^49849^""^
|(900)485-5344|(900)545-1234~(900)545-1200|EC1^FIRST EMERGENCY
CONTACT^HL70131
NK1|2|MASSIE^MARYLOU|MOTHER^^HL70063|300
ZOBERLEIN^^ISHPEMING^MI^49849^""^
|(900)485-5344|(900)545-1234~(900)545-1200|EC2^SECOND EMERGENCY
CONTACT^HL70131
NK1|3
NK1|4|||123 INDUSTRY WAY^^ISHPEMING^MI^49849^""^||(900)545-1200|
EM^EMPLOYER^HL70131|19940605||PROGRAMMER|||ACME SOFTWARE COMPANY
PV1||O|O/R||||0148^ADDISON,JAMES|0148^ADDISON,JAMES||AMB|||||||
0148^ADDISON,JAMES|S|1400|A|||||||||||||||||||GENHOS|||||199501101410|
PV2||||||||199901101400|||||||||||||||||||||||||199901101400
ROL||AD|CP^^HL70443|0148^ADDISON,JAMES
OBX||NM|3141-9^BODY WEIGHT^LN||62|kg|||||F
James was admitted ... his wife is his emergency contact ... hereʼs his weight ...
18. Observation
PID|||1234^^^^SR~1234-12^^^^LR~00725^^^^MR||Doe^John^Fitzgerald^JR^^^L|
...
OBX|4|CE|30949-2^Vaccination adverse event outcome^LN|1|005^required
hospitalization^NIP|
type
Patient
about
personName
observation observationValue
familyName
givenName Doe
CodingSystem Code
middleName
Code CodingSystem
John Fitzgerald
LN 30949-2 005 NIP Identifiers and Time not shown
20. Which Represents ...
CDR/S EMR
Research HL7...
URL
SPARQL
Personal
Ontology
Report Represent Produce
21. Reality: from Vets
• Concrete EMR - VistA
• VA: Largest U.S. Care Provider
• 128 VistAs, federated, 14+ Million in MPI - ICNs
• Available under FOIA
• The Proof
• Mapper Subscribes for HL7 (30/120 packs)
• Maintains a CDR/S for VistA (1 or more)
25. Every EMR, an EndPoint
• EMR links to the cloud, natively
• Mini-Austin: MPI only (old VA Approach)
• Lucky: MUMPS repositories
• Network-Format ala Semantic Web
• VistA’s FileMan (no scale to test)
• If only you could SPARQL them ...
26. FMQL: SPARQL-like
SELECT ?name ?diagnosis ?age ?history FILE
"PATIENT" WHERE {?r "NAME" ?name ;
"DIAGNOSIS" ?d . ?d "DIAGNOSIS" ?diagnosis ;
"AGE AT ONSET" ?age ; "HISTORY" ?history }
• Specification in progress
• Initial goal: limited Patient, meta data dumps
27. Summary
• Semantic Web growing in Health-Care
• But a “Patient Gap”
• Different ways to bridge
• CDISC can drive it forward
• More: http://www.hoot72.org
Notes de l'éditeur
There are green field demos already
Like “goodness”
NOT HTML ... Querying like DB querying, not page fetching
FORM-CONCEPT-QUERY
run thru on baltimore becomes linkable data about baltimore
web didn’t make hyperlinks or protocols or page layout
SEM WEB: ONE MORE WEB THING ... the power, the scale was link anywhere
Nodes and Literals ... Codes would break down
SIMPLE ENCOUNTER
Detailed discussion of semi-structured (not going to get into this aspect)
Observable (OBX|4|CE|30949-2^Vaccination adverse event outcome^LN|1|005^required hospitalization^NIP|)
NIP= National Immunization Program within the Center for Disease Control
ala DATABASE, SEEMS TRIVIAL
One standard code is 30949-2.
For the astute: better if code became URI.
Beyond an isolated set of patients
FOLLOW THE LINKS: Typical Report: chase type (Ontology) in a world of (EMR) particulars. Stanford Drug Ontology gives compounds that treat conditions. RxNorm relates compounds to branded drugs. Hoot72 Clinical Data has branded drugs.
From: HCLS == the w3c Health Care and Life Sciences Interest Group
Patient Data is secure - “intranet” LINK OUT
RxNorm not yet an ontology but has web api so can represent it as a SPARQL end point
Simplied to fit. Ingredient = consists of to ingredient to brand name etc.
Ala Semantic Web: pretty loose definitions. Philosophy.
Dumb AI
from obese to desoxyn ... we need entities
Middleware - format gets out of the way. IT gets out of the reformatting business.
NOT CLASS HIER.
Growing in number ... Billions of triples, ready to be leveraged, all these URIs.
Gen purpose (demographics) and PubMed, Drug Bank, GeneID, Diseasome
Arrows representing linking out to another conceptual scheme
1. STANFORD and BIO-MED guys big ...
2. ONTOLOGY == “TYPE” vs “THING” ... TYPE AND THING
CDASH ODM (machine readable)
CDISC SDTM and other terminology goes through an extensive process of definition, development, and review before it is declared ready for release. Terminology that has completed this process is tagged as "Production," and now includes some 50 SDTM codelists with about 2,200 terms covering demographics, interventions, findings, events, trial design, units, frequency, and ECG terminology. This terminology is maintained and distributed as part of NCI Thesaurus
CDC Example.
We will have standard and local ontologies, standard and local queries
there may be several different ways to express the same concept. Human users may be able to recognise that these are essentially the same, but the rules for doing so must be made explicit to be usable by computer. -- Why is Terminology hard?, Alan Rector
ME to learn of CDISC work. See how to leverage all the work.
CONTENT KING
FDA: “Improve Interoperability: The Target EA establishes enterprise-wide standards that promote platform and vendor independence, enabling greater interoperability across disparate applications, both internal and external”
DOCUMENT model vs GRAPH model
Trial == Snapshot. Extrapolated from individual observations (weight gain etc)
... Here at the Drug Information Association (DIA), you can see a “live” implementation of the interoperability that is possible between Electronic Health Record (EHR) systems and Electronic Data Capture (EDC) systems used for clinical research, which leverages the Integrating the Healthcare Enterprise’s (IHE) Retrieve Form for Data Capture (RFD) integration profile along with CDISC’s ODM and CDASH standards
Contrast to RDIF XFORMs.
ALL ROSY - CONCEPT and PATICULAR, TYPE and THING
The big picture ... Concept and Concrete, Users and Contributors in one web
Trial Recruitment, Drug Safety, Outcomes research
HL7 holds our health data
HL7 everywhere means v2. Small V3.
Of course, more structured than your average tweet
Pick out message type, patient name, contact relationship, body weight observation
SO MANY CODES IN HEALTHCARE
Unload the Truck
What we've done
Key is automatic i.e. requirement
Mapping is on the site.
Moving beyond rough logs.
STILL TOO ACADEMIC
But don't just want a script EMR
Get Real
The Integration Control Number (ICN) - ASTM e1714-95 standard for a universal health identifier.
Like the efforts in the showcase to interop EMRs
Already done for us - or at least we know it works
Note Multiple VistAs
HL7 is triggered. Data there and WHEN it is there.
GE TOO - FLU TO CDC
open use docs first
DO SEGWAY
MUMPS (Massachusetts General Hospital Utility Multi-Programming System)
EMR NOT LEFT OUT OF THE PICTURE, not just a “old” aside.
Looking in the code, you could see ...