Semantic Web Infrastructure for Health Informatics

SMBM
06 October, 2014

Eric Prud'hommeaux, <eric@w3.org>
- Sanitation Engineer, W3C
Charlie Mead, <charles.normanmead@gmail.com>
- Co-chair W3C Health Care Life Sciences

Data Value

Data utility depends on whether it

Directs our actions
- diagnoses
- drug development
- research
- policy
Affects our confidence
Costs less than the value it provides

Data dimensions

Size of population
Coverage of population
Phenotypic breadth

Size/Structure (false) dichotomy

Fiscal Pressure

US spends 18% of GDP on health care.
Closest is France at 8%.
US's outcomes are well-worse than France's.
100,000 deaths/year due to adverse drug reactions.
High variability of care.
Penny-wise/pound-foolish decisions.

What to want

Preventable deaths prevented.
Accountability.
Data for far-sighted decisions.
Democratize innovation.
Personalized medicine
(cheap enough to save small populations of unimportant people)
SCIENCE (e.g. systems biology)
"predictive, preventative, participatory, personalized" (Lee Hood)

What's needed for clinical practice

drug lists
allergies
immunizations
vitals
encounters
procedures
labs
problem statements
diagnosis
treatment plans
outcomes
genomics

Meaningful Use Objectives:

Maintain up-to-date problem, medication, allergy lists
Record demographics, date/cause of death, vitals,
Predict (to avoid) drug-drug and drug-allergy interaction
Computerized Provider Order Entry (CPOE) for prescriptions
Logic: clinical decision support, quality metrics

What's needed for research over populations

treatment research

epidemiology

efficacy…

practice data +

history
demographics
adverse events
outcomes

to build

precise cohorts,
for population research,
with differentiating factors,
like environmental etc.

Representation

Clinical care:

sheets of paper
EMR systems with wide schemas
exchange messages

Clinical trials:

inclusion/exclusion criteria

sheets of paper
terminology codes
realized as DB queries

protocol specifies a small number of specific acquisitions with special terminology codes.

Data scale

Growing data size:

imaging: PET scans, CT scans, FMRI
molecular diagnostics
genome (3 billion base pairs)
biome (300 billion base pairs)

Growing data vectors:

next-generation (post-Sanger) sequencing. (name stuck to Moores law?)
online genomics services: Navigenics, 23andMe and Helix Health
online support communities: CureTogether, patientslikeme
virtual visits: American Well, Cisco HealthPresence, RelayHealth, Teladoc, Medfusion Virtual Office
remote clinical devices: RP-7i (robo-nurse), iPill
commodity clinics: minute clinic (CVS), take care clinic (Walgreens), RediClinic (H-E-B), Target Clinic, The Little Clinic
consumer devices: asthma-charter, USB chip for STDs, glucose meter, urinalysis

Policy

Clinical care

US: Afordable Care Act mandates patient portability in the form of HL7's C-CDA standard.
Europe: Innovative Medicines Initiative focusing on clinical data re-use
- EHR4CR: Electronic Health Records for Clinical Research
- SALUS: Post-market patient safety

Clinical trials

The European Medicines Agency (EMA) pushing publication of clinical-trial data for public health,
Goal of understandable clinical data formats.
FDA goal of improved use of submitted clinical trial data through Therapeutic Areas.
Pharmaceutical companies have attempted to use submission standards to create study repositories.

How's that going?

Clinical care

Too early to tell.

263870008 (SNOMED code for Sequestration)

Clinical trials

Around 2009, CDISC members identified a need for intra and inter pharma sharing of trial metadata:

questions
value sets

They were going to call it SHARE.

4 years later,

GSK started their own.
CDISC2RDF Roche, AZ mapping SDTM to RDF
rolled into PhUSE Emerging Technologies.
Abbvie, Boehringer Ingelheim, AstraZeneca, BMS, GSK, J&J, Lilly, Pfizer, Roche, Sanofi formed TranCelerate.

Mission

Foster innovation
(help mad scientists).
Reduce impedance
(fix plumbing)
Reveal structure
(it's there but opaque)
Reveal semantics
- chemistry: side chains, INCHI codes.
- NLP: grammatical context
- medical: observation codes

Clinical Informatics: information model

General-purpose graph structure for medical information:

activity times and states.
mood: considered, planned, accomplished:
Factors commonalities in e.g. Prescription vs. Injection.
acts: observations, substance administrations, procedures ...
entities: people, organizations, places.
roles: patient, nurse, surgeon.

Tells you where to write but not how to write:

terminology
data types
value sets

information model (RIM)

Clinical Informatics: terminology model

Taxonomies used for:

problem statements
tests performed
medications administered
reactions observed
diagnoses

terminology model

Clinical Informatics: connected models

information model
(Concepts and Relationships)
- Observations
- AdverseEvents
- Substance Administrations
on patient Isabella Jones
terminology model
(Terms and Relationships)
- Pneumonia a Pulmonary Disease
- Serum Sodium a lab test
- Penicillin an antibiotic
value sets
- Pneumonia, SLE, CHF
- mild, medium, severe
datatypes
- dates
- times
- intervals

Decentralization of Semantics

Information model: implemented by database administrators.
Terminology model: implemented by domain expert dictionaries.

Separation:

Evolves naturally
Enables decentralization
Enables interoperability with other information models.

Three forms of standards:

syntactic standards:
fax-level interoperability shares data between humans.
HTML's ubiquitous rendering is a great success.
semantic standards:
code systems, data types and value sets.
human eyes not required.
Semantic Web-based standards:
use RDF, OWL, Web architecture etc. in definitions and data.
leverage growing body of standards and expertise.

Semantic standards:

Reduce the n² pressure.

hundreds of partially-overlapping formats.
no forum for collaborative mapping.
many mappings are use-case-dependent.
reduce risk to entrepreneurs.
foster interesting innovation.

Interoperability value proposition

Reduce redundancy
Commoditize systems
Effective quality metrics
Cheaper protocol development and distribution
Cheaper and more comprehensive drug reasearch

Interoperability state of the art

Short-term commercial interests oppose interop.
Lock-in provides stability.
Use cases percieved as sci-fi.

Technical barriers:

Standardization (precise communication) is expensive.
Multiple intersecting domains with intersecting authorities:
- drugs, diseases, tests, diagnoses (Terminology Model)
  pathology: SNOMED (IHTSDO), tests: LOINC, drugs: RxNorm, NDC, CPT (AMA), NPU (IUPAC), ICDⁿ (WHO), research results: MeDRA (US FDA), HCPCS, billing: NDC, efficacy: CQM (CMS) …
- graph of what happened to whom and why (Information model)
  Reference Information Model (HL7), SDTM (CDISC), MIMIC II database schema, I2B2, SMART, OpenEHR, all 700+ EMRs, closed (EPIC, Cerner) or open (CommCare)
- datatypes and units
  ISO21090, XML Schema, UCUM
Legacy systems (success of HL7 V2).
Insufficient or overly complex implementation guidelines.
Insufficient information model, inappropriate model, inconsistent data types.

Don't have fun

	Learn	Invent
idiom	embrace	extend
costs	apparent	hidden

SemWeb makes invention too easy and safe.

Try stuff out, ask questions, fix other peoples' stuff.

When I understand a model well enough to criticize it,
only then do I understand it well enough to use it.

Payback

Reduce impedance.
Improve innovation.
Improve accountability.
Improve value.
Improve uptake.

Priority TAs (Therapeutic Areas)

FDA (CDER) project billed as:

Standardizing key study data specific to therapeutic areas (TAs)
facilitate clinical research and the evaluation of medical products
prioritized by:
- particular medical need
- existing data standardization projects
- greater drug development pipeline activity
each scoped to be done in 1 year
"subsequent projects in those areas build on the results"
all due December 2017

Statement of Work

Establish common study data standards:

For data elements, terminologies, and data structures.
For efficacy review and evaluation of new medicinal products.
(Efficacy data includes: primary, secondary, and exploratory outcome measures).
Enable automation clinical study data to support regulatory decision-making.
Speed the delivery of new therapies to patients.

High-level goals

Code everything!

minimize:

ambiguity - one way to write two things.
diffusion - two ways to write one thing.

cope gracefully with:

diversity - huge range of assertions.
stratification - assertions in intersection of multiple terminologies.

interoperate with EHRS:

FHIR
OpenEHR
C-CDA
Vista
vendors: EPIC, Cerner, Siemens, ...

RDF toolbox available for reuse

Existing Models/Standards

BRIDG
SDTM
ODM
SDM
HL7 RIM
CDASH
etc.

Controlled Terminologies:

MedDRA
LOINC
SNOMED
etc.

Strategy

Use BRIDG where possible.
- Collective effort of CDISC, HL7, NCI and FDA
- Represents clinical trial planning, specialization, scheduling and execution.
- Available in OWL
Specialize bridg:PerformedObservation
- PerformedObservation - all RenalTransplantation studies share a notion of a Serum Creatine test.
Represent some protocol and stats
factor factor factor…

Portable BRIDG Representation

factoring

renal.ttl

:SerumCreatinineLevel a owl:Class ;
    rdfs:subClassOf 
        [ owl:onProperty bridg:resultIn ; owl:allValuesFrom bridg:PerformedClinicalResult ],
        [ owl:onProperty bridg:resultIn ; owl:cardinality 1 ]  .

transplant.ttl

:GraftBPARAssessment a owl:Class ;      
    rdfs:subClassOf core:NegativeOutcome ;
    owl:equivalentClass [
        a owl:Class ;
        owl:intersectionOf (
          [ a owl:Restriction ;
            owl:onProperty core:afterIntervention ;
            owl:someValuesFrom [
                a owl:Restriction ;
                owl:onProperty core:hasPathologyFinding ;
                owl:hasValue :BanffIII
            ]
          ]
          [ a owl:Restriction ;
            owl:onProperty core:hasResultValue ;
            owl:hasValue :NonFunctioningGraft
          ]
        )
    ]
.

RenalTransplant.ttl

<http://www.w3.org/2013/12/FDA-TA/RenalTransplantation> a owl:Ontology ;
    owl:imports <http://www.w3.org/2013/12/FDA-TA/core> ,
        <http://www.w3.org/2013/12/FDA-TA/renal> ,
        <http://www.w3.org/2013/12/FDA-TA/transplant> .

Result

RenalTransplant is tiny
trivial interop with:
- ComplicatedUTI - renal functions
- RheumatoidArthritis - Immunosuppression
- all - bridg:MedicationAdminstration

SELECT ?outcomeType ?dose
       (AVG(?endpoint-time) AS ?rate)
 WHERE {
    # drug of interest
    ?adminDrug dt:CD.displayName "Upsidasium" ;
        … codingSystem … .

    # subjects in studies about drug
    ?arm :studySubject ?adminDrug ;
        :studyParticipation ?subject .

    # demographic selection
    ?subject :taxon ncbitax:7609 .

    # outcomes assessing prescription performance
    ?outcome :intervention ?p ;
        :value [ a ?outcomeType ] .

    # ... of that drug on that subject (participation)
    ?p a :Prescription ;
        :involvedSubject ?subject ;
        :medication ?adminDrug .
} ORDER BY ?dose

EHR Interop

:subjectspostOpDay3GFR a rrej:RenalFunctionObservation ;
    mm:observationTime "2013-07-08T14:50:00"^^xsd:dateTime ;
    rrej:gfrFlowRate [ data:value 12.0 ; data:units ucum:mL-per-minute ].

Would be in a table with codes in an EMR:

obs id	when	coding system	ObsCode	value	units	addressing…performer…
1234	2013-07-08T14:50:00	CPT	82565	12	mL-per-minute
5678	2013-07-08T14:50:00	SNOMED 2014-01-01	241374009	12	mL-per-minute
4321	2013-07-08T14:50:00	LOINC	48643-1	12	mL-per-minute

or C-CDA:

<observation classCode="OBS" moodCode="EVN">
  <code code="48643-1" codeSystem="2.16.840.1.113883.6.1" codeSystemName="LOINC"
        displayName="Glomerular filtration rate/1.73 sq M.predicted.black" />
  <effectiveTime value="201307081450"/>
  <value type="PQ" value="12.0" unit="mL/min"/>
  <interpretationCode/>
  <referenceRange />
</observation>

SemWeb plumbing work

Information model efforts

FDA TA - US Food and Drug Administration models for "Therapeutic Areas"
HL7 RIM (MIF) - Information model from HL7.
Mostly used for state transfer (Admission, Discharge, Transfer).
HL7 FHIR - Hacker-focused state representaiton
Open EHR - Clinical Information Modeling Initiative
Vista - Veterans' Aministration EHR
Now being integrated with DOD EHRs.

Needed: EPIC Cerner SIEMENS AllScripts HP athenahealth NextGen eClinicalWorks McKesson Greenway Praxis e_MDCureMD MediTab NextTech ADP ...

Motivations

?
?
!

Questions

What's next?
- validation
  - demo
  - requirements
  - W3C WG charter, homepage
- mappings
  - SALUS
  - Yosemite
  - EHR4CR homepage, report
- privacy

Semantic Web Infrastructure for Health Informatics

SMBM06 October, 2014

Data Value

Fiscal Pressure

What to want

What's needed for clinical practice

What's needed for research over populations

treatment research

epidemiology

efficacy…

Representation

Clinical care:

Clinical trials:

Data scale

Growing data size:

Growing data vectors:

Policy

Clinical care

Clinical trials

How's that going?

Clinical care

Clinical trials

Mission

Clinical Informatics: information model

Clinical Informatics: terminology model

Clinical Informatics: connected models

Decentralization of Semantics

Three forms of standards:

Semantic standards:

Interoperability value proposition

Interoperability state of the art

Don't have fun

Payback

Priority TAs (Therapeutic Areas)

Statement of Work

High-level goals

RDF toolbox available for reuse

Strategy

factoring

renal.ttl

transplant.ttl

RenalTransplant.ttl

Result

EHR Interop

SemWeb plumbing work

Motivations

Questions

SMBM
06 October, 2014