Concept maps are a human-intuitive way to communicate structures and relationships.
We need machines to work-with and enforce that structure.
Success metric: minimize implementation guidelines because the information is all in schema:
# Assessments are observations about other observations.
core:Assessmentrdfs:subClassOfcore:ClinicalObservation ,
[ owl:onPropertycore:hasObservation ; owl:minCardinality 1 ] .
# Differentiate observations before and after an intervention.
core:beforeInterventionrdfs:subPropertyOfcore:hasObservation.
core:afterInterventionrdfs:subPropertyOfcore:hasObservation .
# Specialize assessment to include before and after observations.
core:SingleOutcomeAssessmentrdfs:subClassOfcore:Assessment ,
[ owl:onPropertycore:beforeIntervention ; owl:minCardinality 1 ] ,
[ owl:onPropertycore:hasIntervention ; owl:minCardinality 1 ] ,
[ owl:onPropertycore:afterIntervention ; owl:minCardinality 1 ] .
# A kidney graft outcome assessment is based observations of graft viability.
renalX:KidneyGraftOutcomeAssessment ;
rdfs:subClassOfcore:FunctionOutcomeAssessment ,
[ owl:onPropertycore:hasResultValue ; owl:allValuesFromxplant:GraftViability ] .
see also intro slide 7, 29, 33
Intuitive models
Subject clinical observations
Subject interventions
Effects for subject
Effects for population
Comes from BRIDG and the meta-model. (intro slide 15)
Q: Is this a model to capture the results of valid protocol, or a way to generate valid protocols?
Should the models be detailed and restrictive enough to replace much of the existing protocols?
Alternative models
What alternatives did we have to BRIDG?
O-RIM - HL7 Reference Information Model + Datatypes, etc
Endpoint: evaluate outcome assessments over a population.
Outcome assessment: assess state of individual before/after intervention.
Based on observations or assessments.
Assessment: evaluate 1 or more observations.
Observation: atomic data about an individual.
TBC - Anticoagulants Classes
class hierarchy
inheritance diagram
TBC - StrokeOutcomeAssessment
A ancoag:StrokeOutcomeAssessment looks for a stroke (after some intervention).
A ancoag:StrokeAssessment is hemorrhagic, ischemic or uncertain.
An ancoag:IschemicStrokeAssessment is diagnosed by one of four observations.
Distributed development
BRIDG + core provides a canvas to capture clinical trial exposures, observations and assessments in a structured way.
As described in the Machine-readable Structure slide, TA ontologies have to follow a rigid structure of endpoints measuring outcome assessments, which are in turn based on observations and assessments.
A purpose-built .TA (pronounced "dot T A") language has a grammar which follows (enforces) this structure.
Most errors one can make in a TA ontology document are prohibited by the grammar. intro slide 17
Simplified development + factoring of shared libraries enables concurrent development.
Definitions File
Standard column headers provide:
Name: how a row is addressed in .TA files.
Definition: a skos:Definition for the concept.
Definition Source: whether it comes from SNOMED, EVS, LOINC, etc.
Code: a controlled code in some code system.
Code System: the authority administering that code.
Code Extension: whether the code should be used directly or specialized.
Commodity tooling
HL7 started with the UML and added hundreds of extensions, requiring in-house maintenance of custom tools.
OWL now captures most of these extensions.
Basing on RDF and OWL means there are 10s of reasoners and maybe 50-100 SPARQL engines.
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:9606 .
# 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
PREFIX foaf: <http://xmlns.com/foaf/0.1/>
SELECT ?endpoint ?count ?labels {
{
SELECT ("GraftSurvival" AS ?endpoint) (COUNT(*) AS ?count) (GROUP_CONCAT(?name) AS ?labels)
WHERE {
?s bridg:Subject.performingBiologicEntity [ foaf:name ?name ] ;
core:hasOutcomeAssessment [
axplant:GraftSurvivalAssessment ]
}
}
UNION
{
SELECT ("GraftLoss" AS ?endpoint) (COUNT(*) AS ?count) (GROUP_CONCAT(?name) AS ?labels)
WHERE {
?s bridg:Subject.performingBiologicEntity [ foaf:name ?name ] ;
core:hasOutcomeAssessment [
axplant:GraftLossAssessment ]
}
}
}
Result
Clinical observations queriable:
across studies.
across TAs.
?arm :studySubject ?adminDrug
over EDC systems.
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
see also intro slide 20
Commodity base ontologies
Reusing other ontologies enables sharing of queries and applications (minor effect) and teaching and comprehension investment (major effect). Experts in this field should recognize structures and ideally point out improvements.
BRIDG is complex but principled.
Pharma is investing in BRIDG; investment in TA ontologies is an investment in in-house solutions, e.g. SHARE.
Usable by domain experts and RDF techies.
BRIDG is extended by deriving our classes from BRIDG classes.
Such an enumeration must be exhaustive, including even typos, which is an unreallistic requirement. The opposite tactic, more appropriate for instance validation, is to assume that names are distinct unless they are asserted to be equivalent.
Most clinical systems use some variant of ISO 21090 "CD" for codes from controlled terminologies:
code
code system
display name
code system name
# Example instance data -- Sue's baseline measured in cerebral spinal fluid
:labObs1234# Rheumatoid Factor Observations are defined in the Rheumatoid Artheritis ontology
ara:RheumatoidFactorObservation ;
# All observations have a time
core:hasObservationTime"2013-07-07T19:02:00Z"^^xsd:dateTime ;
# We reuse the hl7 schema to capture the convention of a code and a code system (à la ISO 21090)
hl7:coding [ dt:CDCoding.code "14034-3" ; dt:CDCoding.codeSystem "2.16.840.1.113883.6.1" ;
dt:CDCoding.displayName
"Rheumatoid factor:Arbitrary Concentration:Point in time:Cerebral spinal fluid:Quantitative" ;
dt:CDCoding.codeSystemName "LOINC" ] ;
# Instance data will of course have results...
core:hasResultValue [ data:value 65 ; data:unitsucum:u_mL ].
The code system and the code create a unique identifier.
