EHR Orientation

Eric Prud'hommeaux, <W3C/MIT>
With material from:

Charlie Meed (NIH)
John Glaser (CEO Siemens Health IT)
Peter Szolovits (MIT)
Josh Mandel (Harvard University/Boston Children's Hospital)

Health Care & Life Sciences IG, <public-semweb-lifesci@w3.org>

Health Care Environment

Many standards organizations.
Many legal requirements/policies.
Evolving foundational understanding.
Communications impedence.
Information^h^hData overload.

Pharmaceutical Environment

High failure rate
Blockbuster exhaustion
implies person-centric medications
(and decreased cohort sizes for evaluation studies)
Elevating criteria
Patenty expiration
Inefficient competetion

Outsource risk to startups
VCs make the early go/no-go decisions.

Pharmas left with manufacturing development and distribution.

Low Bars for Information Re-use

Affordable Care Act demands - "meaningful use".
1st stage is simply comprehensive EMRs, e.g. HTML in Clinical Document Architecture.
@@counterparts in other countries@@

Practically, we want information available to
improve clinical research and clinical care.
The hard job is communicating context.

Everyone thinks their model should be the over-arching model
(or they go work on another one).

What are the characteristic of a model?

What Information?

personal: name, demographics
historical: behavior, family history
financial
institutional: ids of patient and practitioners
conditions/diagnoses
allergies
medications
immunizations
physiology: vitals
lab results
images
gentics
plans/progress
discharge/transfer notes

Clinical Information

ID: 23453
name: George Andrew Tour
demographics
    DOB: 1934-05-15
    gender: Male
presriptions
  1 no earlier than: 2009-05-15
    ndc: 00007414020
    generic name: COREG TABLETS
  …
conditions
  1 no later than: 2009-05-15T12:00:00
    ICD9: 410
    common name: hypertension
  …
encounters
  1 date: 2009-05-15
    reported symptoms: none
    working diagnosis: none
    tests
      1 no earlier than: 2009-05-15T12:00:00
        CPT: 85025
        common name: Blood count; complete (CBC), automated (Hgb, Hct, RBC, WBC and platelet count) and automated differential WBC count
          1 CPT: 85014
            LOINC: 41654-5
            common name: Hematocrit
            value: 38.0
            units: percent
          2 CPT: 78720
            LOINC: 30428-7
            common name: Mean Corpuscular Volume
            value: 97.4
            units: fl
  …

Information Without Context - Noise

ID: 23453
name: George Andrew Tour
demographics
    DOB: 1934-05-15
    gender: Male
presriptions
  1 no earlier than: 2009-05-15
    ndc: 00007414020
    generic name: COREG TABLETS
  …
conditions
  1 no later than: 2009-05-15T12:00:00
    ICD9: 410
    common name: hypertension
  …
encounters
  1 date: 2009-05-15
    reported symptoms: none
    working diagnosis: none
    tests
      1 no earlier than: 2009-05-15T12:00:00
        CPT: 85025
        common name: Blood count; complete (CBC), automated (Hgb, Hct, RBC, WBC and platelet count) and automated differential WBC count
          1 CPT: 85014
            LOINC: 41654-5
            common name: Hematocrit
            value: 38.0
            units: percent
          2 CPT: 78720
            LOINC: 30428-7
            common name: Mean Corpuscular Volume
            value: 97.4
            units: fl
  …

Which communities manage what?

ID: 23453
name: George Andrew Tour
demographics
    DOB: 1934-05-15
    gender: Male
presriptions
  1 no earlier than: 2009-05-15
    ndc: 00007414020
    generic name: COREG TABLETS
  …
conditions
  1 no later than: 2009-05-15T12:00:00
    ICD9: 410
    common name: hypertension
  …
encounters
  1 date: 2009-05-15
    reported symptoms: none
    working diagnosis: none
    tests
      1 no earlier than: 2009-05-15T12:00:00
        CPT: 85025
        common name: Blood count; complete (CBC), automated (Hgb, Hct, RBC, WBC and platelet count) and automated differential WBC count
          1 CPT: 85014
            LOINC: 41654-5
            common name: Hematocrit
            value: 38.0
            units: percent
          2 CPT: 78720
            LOINC: 30428-7
            common name: Mean Corpuscular Volume
            value: 97.4
            units: fl
  …

US Federal Drug Administration (National Drug Codes)

ID: 23453
name: George Andrew Tour
demographics
    DOB: 1934-05-15
    gender: Male
presriptions
  1 no earlier than: 2009-05-15
    ndc: 00007414020
    generic name: COREG TABLETS
  …
conditions
  1 no later than: 2009-05-15T12:00:00
    ICD9: 410
    common name: hypertension
  …
encounters
  1 date: 2009-05-15
    reported symptoms: none
    working diagnosis: none
    tests
      1 no earlier than: 2009-05-15T12:00:00
        CPT: 85025
        common name: Blood count; complete (CBC), automated (Hgb, Hct, RBC, WBC and platelet count) and automated differential WBC count
          1 CPT: 85014
            LOINC: 41654-5
            common name: Hematocrit
            value: 38.0
            units: percent
          2 CPT: 78720
            LOINC: 30428-7
            common name: Mean Corpuscular Volume
            value: 97.4
            units: fl
  …

World Health Organization (International Classification of Diseases)

ID: 23453
name: George Andrew Tour
demographics
    DOB: 1934-05-15
    gender: Male
presriptions
  1 no earlier than: 2009-05-15
    ndc: 00007414020
    generic name: COREG TABLETS
  …
conditions
  1 no later than: 2009-05-15T12:00:00
    ICD9: 410
    common name: hypertension
  …
encounters
  1 date: 2009-05-15
    reported symptoms: none
    working diagnosis: none
    tests
      1 no earlier than: 2009-05-15T12:00:00
        CPT: 85025
        common name: Blood count; complete (CBC), automated (Hgb, Hct, RBC, WBC and platelet count) and automated differential WBC count
          1 CPT: 85014
            LOINC: 41654-5
            common name: Hematocrit
            value: 38.0
            units: percent
          2 CPT: 78720
            LOINC: 30428-7
            common name: Mean Corpuscular Volume
            value: 97.4
            units: fl
  …

LOINC

ID: 23453
name: George Andrew Tour
demographics
    DOB: 1934-05-15
    gender: Male
presriptions
  1 no earlier than: 2009-05-15
    ndc: 00007414020
    generic name: COREG TABLETS
  …
conditions
  1 no later than: 2009-05-15T12:00:00
    ICD9: 410
    common name: hypertension
  …
encounters
  1 date: 2009-05-15
    reported symptoms: none
    working diagnosis: none
    tests
      1 no earlier than: 2009-05-15T12:00:00
        CPT: 85025
        common name: Blood count; complete (CBC), automated (Hgb, Hct, RBC, WBC and platelet count) and automated differential WBC count
          1 CPT: 85014
            LOINC: 41654-5
            common name: Hematocrit
            value: 38.0
            units: percent
          2 CPT: 78720
            LOINC: 30428-7
            common name: Mean Corpuscular Volume
            value: 97.4
            units: fl
  …

CPT

ID: 23453
name: George Andrew Tour
demographics
    DOB: 1934-05-15
    gender: Male
presriptions
  1 no earlier than: 2009-05-15
    ndc: 00007414020
    generic name: COREG TABLETS
  …
conditions
  1 no later than: 2009-05-15T12:00:00
    ICD9: 410
    common name: hypertension
  …
encounters
  1 date: 2009-05-15
    reported symptoms: none
    working diagnosis: none
    tests
      1 no earlier than: 2009-05-15T12:00:00
        CPT: 85025
        common name: Blood count; complete (CBC), automated (Hgb, Hct, RBC, WBC and platelet count) and automated differential WBC count
          1 CPT: 85014
            LOINC: 41654-5
            common name: Hematocrit
            value: 38.0
            units: percent
          2 CPT: 78720
            LOINC: 30428-7
            common name: Mean Corpuscular Volume
            value: 97.4
            units: fl
  …

SNOMED-CT, etc:

ID: 23453
name: George Andrew Tour
demographics
    DOB: 1934-05-15
    gender: Male
presriptions
  1 no earlier than: 2009-05-15
    ndc: 00007414020
    generic name: COREG TABLETS
  …
conditions
  1 no later than: 2009-05-15T12:00:00
    ICD9: 410
    common name: hypertension
  …
encounters
  1 date: 2009-05-15
    reported symptoms: none
    working diagnosis: none
    tests
      1 no earlier than: 2009-05-15T12:00:00
        CPT: 85025
        common name: Blood count; complete (CBC), automated (Hgb, Hct, RBC, WBC and platelet count) and automated differential WBC count
          1 CPT: 85014
            LOINC: 41654-5
            common name: Hematocrit
            value: 38.0
            units: percent
          2 CPT: 78720
            LOINC: 30428-7
            common name: Mean Corpuscular Volume
            value: 97.4
            units: fl
  …

Which leaves the information framework:

ID: 23453
name: George Andrew Tour
demographics
    DOB: 1934-05-15
    gender: Male
presriptions
  1 no earlier than: 2009-05-15
    ndc: 00007414020
    generic name: COREG TABLETS
  …
conditions
  1 no later than: 2009-05-15T12:00:00
    ICD9: 410
    common name: hypertension
  …
encounters
  1 date: 2009-05-15
    reported symptoms: none
    working diagnosis: none
    tests
      1 no earlier than: 2009-05-15T12:00:00
        CPT: 85025
        common name: Blood count; complete (CBC), automated (Hgb, Hct, RBC, WBC and platelet count) and automated differential WBC count
          1 CPT: 85014
            LOINC: 41654-5
            common name: Hematocrit
            value: 38.0
            units: percent
          2 CPT: 78720
            LOINC: 30428-7
            common name: Mean Corpuscular Volume
            value: 97.4
            units: fl
  …

Modeling the framework

wing it- make up new names as they're needed.
not distributable.

HL7 approach: Reference Information Model (RIM) @@image allowed?
constrained by Refined Message Information Models (R-MIMs).
Implementable Technology Specifications (ITS) is syntactic restrictions (schema) while R-MIM is a model restriction?

Some desirable features:

document-self-description
distributed extensibility
using web technology
minimal version disruption
Don't lose the information/tooling which was stable through revisions.

Proliferation of EHR/EMR Systems

EMR uptake low and fragmented.

Most clinics have fewer than 50 beds.
Most physicians work in practices of two or three.
300 health record vendors in US.
~25% market share to the top 7.
¼ disappear and are replaced every year.

Can EHRs be as practical as Quicken?

Clinical Care Outcomes

US spends the most, ranks ~30th @@Michelle.
High variability: births, heart attacks, etc.
Paid by volume (birth, surgery) with no quality feedback.
Operating margin of 1-2%.

Clinical Research Re-use

Very low, have to focus on success stories.

@@Lena?

Goals

Conventional uses:
- Billing
- Supporting public health, from epidemiology to public health worker interventions.
- Computerized Provider Order Entry (CPOE)
Clinical care:
- Diagnostic support
- Adverse events detection
- Syndromic suvrveillance
Clinical research:
- Phenotype/GWAS studies
- Biomarker discovery

Re-use Approaches

Brute force -- acquire data and read lots of documentation.

Force into common framework
(either the publisher/custodian or the user may get stuck with this job).

Traditional approaches

Mimic DB, Machine-learning, @@Peter

Traditional exchange records

Moving data between Clinical Data Repositories (CDRs):

looser:
- HL7 V2
by restriction:
- HL7 V3 in all its forms
  e.g. Clinical Document Architecture
by extension:
- OpenEHR
- Continuity of Care Record

HL7 V2

@@example of Hematocrit and Mean Corpuscular Volume with random Z-segment extensions - Lloyd? Charlie?@@

Ungoverned extensibility via z-segments.
High update for
- Admission Discharge and Transfer (ADT)
- financial
- administratie
- lab messages

HL7 Clinical Document Architecture

@@pack some HTML and coded data in here@@

The "always-human-readable" aspect is key to success.

ASTM Continuity of Care Document

@@p...@@

Recent and evolving schemes

I2B2
Indivo
SMART
Translational Medicine Ontology
HL7 FHIR
BRIDG
LS-DAM

I2B2 Star Schema

I2B2 Observation Structure

observation_fact

encounter_num
concept_cd
provider_id
start_date
modifier_cd
patient_num
valtype_cd, tval_char, nval_num, valueflag_cd, quantity_num, units_cd, end_Date, locataion_cd, confidence_num, observation_blob

I2B2 Factoids

observation_fact

concept	name_char	valtype_cd	nval_num	units_cd
4544-3	Haematocrit	N	0.55	N/A
787-2	Mean Cell Volume	N	99	fL

The value of the latent interoperability of data depends on the quality of the data.

SMART Factoids

<obs9> sp:labName [
    a sp:CodedValue ;
    sp:code <http://purl.bioontology.org/ontology/LNC/787-2> ;
    dcterms:title "Serum sodium" ;
    sp:codeProvenance [
        a sp:CodeProvenance ;
        sp:sourceCode <codes#_02341> ;
        dcterms:title "Mean Corpuscular Volume" ;
        sp:translationFidelity <http://smartplatforms.org/terms/code/fidelity#automated>l .
    ]
] .

Early TMO

  [ trans:testCode [
    trans:x_3 "MCV" ;
    trans:LOINC "787-2" ;
    trans:description "Mean Cell Volume" ;
    trans:value 97.4 ;
    trans:units "fl" ] ] .

FHIR

Early stages but lots of energy
Isolating resources in a REST-like fashion.
Currently (today) modeling RIM in OWL
and restricting FHIR resources from there.
!! RDF representation of the data is done when the modelling is done !!
Documents map from XML Schema to the names in the FHIR restrictions.

FHIR as RDF

      [ a :Result ;
        :name [
          a :Coding ;
          :code "787-2" ;
          :system <urn:hl7-org:sid/loinc> ;
          :display "Mean Cell Volume" ] ;

        :valueQuantity [
          :value 99 ;
          :units "fL" ;
          :code "fL" ; # @@ same as units?
          :system <urn:hl7-org:sid/ucum> ] ;

        :flag "+" ;
        :status "final" ;
        :referenceRanges [ 
          a :ReferenceRange ;
          :rangeInterval [
            :low [
              :value 80 ;
              :units "fL" ;
              :code "fL" ;
              :system <urn:hl7-org:sid/ucum> ] ;
            :high [
              :value 98 ;
              :units "fL" ;
              :code "fL" ;
              :system <urn:hl7-org:sid/ucum> ]
          ]
        ]
      ]

HL7 V3 Lab R-MIM for Mean Cell Volume loinc:787-2

  # 
  _:mcvComp a M0:Component2 ;
            M0:Component2.typeCode "COMP"  ;
            M0:Component2.contextControlCode "AP"  ;
            M0:Component2.observationEvent _:mcvObsEvt .
  _:mcvObsEvt a M0:ObservationEvent ;
              M0:ObservationEvent.classCode "OBS"  ;
              M0:ObservationEvent.moodCode "EVN"  ;
              M0:ObservationEvent.code _:mcvObsEvt_CD .
  _:mcvObsEvt_CD a DT:CD ;
                 DT:CD.code "787-2"  ;
                 DT:CD.codeSystem "urn:hl7-org:sid/loinc"  ;
                 DT:CD.displayName _:mcvDispName .
  _:mcvDispName a DT:ST ;
                  DT:ED.value "Mean Cell Volume"  .
  _:mcvObsEvt M0:ObservationEvent.statusCode [ a DT:CD ; DT:CD.code "completed" ] .
  _:mcvObsEvt M0:ObservationEvent.value [
      a DT:PQ ;
      DT:QTY.originalText [ a DT:ED ; DT:ED.value "99 fL" ] ;
      DT:PQ.unit [ a DT:CD ; DT:CD.code "fL" ] ;
      DT:PQ.value "99" ] .
  _:mcvObsEvt M0:ObservationEvent.interpretationCode [
               a DT:DSET ;
               DT:COLL.items _:b0x1184590 ] .
  _:b0x1184590 a rdf:Bag ;
               rdf:li [
      a DT:CD ;
      DT:ANY.nullFlavor [ a DT:CD ; DT:CD.code "UNC" ] ;
      DT:CD.code "+" ] .
  _:mcvObsEvt M0:ObservationEvent.referenceRange [
      a rdf:Bag ;
      rdf:li [
          a M0:ReferenceRange ; M0:ReferenceRange.typeCode "REFV" ;
          M0:ReferenceRange.interpretationRange _:mcvRefIntrpLow ] ;
      rdf:li [
          a M0:ReferenceRange ; M0:ReferenceRange.typeCode "REFV" ;
          M0:ReferenceRange.interpretationRange _:mcvRefIntrpHigh ]
  ].
  _:mcvRefIntrpLow a M0:InterpretationRange ;
                   M0:InterpretationRange.classCode "OBS"  ;
                   M0:InterpretationRange.moodCode "EVN.CRT" ;
                   M0:InterpretationRange.value [ a DT:PQ ; DT:PQ.value "80" ;
      DT:QTY.originalText [ a DT:ED ; DT:ED.value "80 fL" ] ;
      DT:PQ.unit [ a DT:CD ; DT:CD.code "fL" ] ] ;
              M0:InterpretationRange.interpretationCode [
      a DT:CD ; DT:CD.code "L" ; DT:CD.codeSystem "2.16.840.1.113883.5.83"
  ] .
  _:mcvRefIntrpHigh a M0:InterpretationRange ;
                    M0:InterpretationRange.classCode "OBS" ;
                    M0:InterpretationRange.moodCode "EVN.CRT" ;
                    M0:InterpretationRange.value [ a DT:PQ ; DT:PQ.value "98";
      DT:QTY.originalText [ a DT:ED ; DT:ED.value "98 fL" ] ;
      DT:PQ.unit [ a DT:CD ; DT:CD.code "fL" ]
  ] .
  _:mcvRefIntrpHigh M0:InterpretationRange.interpretationCode [
      a DT:CD ;
      DT:CD.code "H" ;
      DT:CD.codeSystem "2.16.840.1.113883.5.83"
  ] .

Gap between HL7 V3 R-MIM and TMO

former reflects lineage, later references lineage
generic framework vs. "curried" predicates (e.g. #Systolic_BP_mmHg) for simple surface for programmers/users

Compare/Contrast

SMART, Indivo, FHIR leverage web architecture.
resource-based, SMART using OAuth.
SMART, Indivo, FHIR, TMO designed with extebsibility in mind.
Any publicly-visible standards democratize health care, enable "app stores".
Ones leveraging conventional technologies lower the barrier to entry.
(SMART is designed explicitly for this.)

Exploitation

How can we take advantage of "semantic" representations?

@@Cecil's V2->OWL->JESS@@

@@Clinical Observation Interoperability Demo@@