A formal computable definition of a graph of resources - that is, a coherent set of resources that form a graph by following references. The Graph Definition resource defines a set and makes rules about the set.
5.10.1 Scope and Usage
The GraphDefinition resource provides a formal computable definition of a graph of resources - that is, a coherent set of resources that form
a graph by following references. The Graph Definition resource defines a set and makes rules about the set.
The GraphDefinition resource can be used to:
Summarize a set of profiles on resources
Define a graph of resources to return in a query
Define a graph of resources to include in a document
Document rules about the relationship between a set of resources e.g. must all resources concern the same patient?
5.10.2 Boundaries and Relationships
There is a close relationship between Profiles and
GraphDefinitions:
A StructureDefinition defines a profile, and profiles can make rules about the relationships between resources. A carefully defined set of profiles implies part of what is in a GraphDefinition
A GraphDefinition defines rules about the relationships between resources, and in so doing, implies some constraints that could or should be represented in their profiles, if they are defined
Like CompartmentDefinition, GraphDefinition allows defining a grouping of resources and may be used for security access permissions.
However, GraphDefinitions do not define additional query capabilities and are much more limited in how membership in the collection is defined
(all resources that are part of a Compartment must expose a search parameter of the same name).
Profiles and Graph Definitions can be used together, or separately. When used together, they should
be consistent. Note, though, that a graph definition may contain a subset or a superset of the relationships
explicitly described in the profiles it refers to.
It is possible that in some circumstances, a graph definition makes incompatible rules with
the Profiles it refers to - in this case, no graph of resources will meet the constraints expressed.
Applications should - but are not required to - detect when such incompatibilities arise.
Definition of a graph of resources + Warning: Name should be usable as an identifier for the module by machine processing applications such as code generation
Canonical identifier for this graph definition, represented as a URI (globally unique) + Warning: URL should not contain | or # - these characters make processing canonical references problematic
@prefix fhir: <http://hl7.org/fhir/> .
[ a fhir:GraphDefinition;
fhir:nodeRole fhir:treeRoot; # if this is the parser root
# from Resource: .id, .meta, .implicitRules, and .language
# from DomainResource: .text, .contained, .extension, and .modifierExtension
fhir:url[ uri ] ; # 0..1 Canonical identifier for this graph definition, represented as a URI (globally unique)
fhir:identifier ( [ Identifier ] ... ) ; # 0..* Additional identifier for the GraphDefinition (business identifier)
fhir:version[ string ] ; # 0..1 Business version of the graph definition
# versionAlgorithm[x]: 0..1 How to compare versions. One of these 2
fhir:versionAlgorithm[ a fhir:string ; string ]
fhir:versionAlgorithm[ a fhir:Coding ; Coding ]
fhir:name[ string ] ; # 1..1 IName for this graph definition (computer friendly)
fhir:title[ string ] ; # 0..1 Name for this graph definition (human friendly)
fhir:status[ code ] ; # 1..1 draft | active | retired | unknown
fhir:experimental[ boolean ] ; # 0..1 For testing only - never for real usage
fhir:date[ dateTime ] ; # 0..1 Date last changed
fhir:publisher[ string ] ; # 0..1 Name of the publisher/steward (organization or individual)
fhir:contact ( [ ContactDetail ] ... ) ; # 0..* Contact details for the publisher
fhir:description[ markdown ] ; # 0..1 Natural language description of the graph definition
fhir:useContext ( [ UsageContext ] ... ) ; # 0..* The context that the content is intended to support
fhir:jurisdiction ( [ CodeableConcept ] ... ) ; # 0..* Intended jurisdiction for graph definition (if applicable)
fhir:purpose[ markdown ] ; # 0..1 Why this graph definition is defined
fhir:copyright[ markdown ] ; # 0..1 Use and/or publishing restrictions
fhir:copyrightLabel[ string ] ; # 0..1 Copyright holder and year(s)
fhir:start[ id ] ; # 0..1 Starting Node
fhir:node( [ # 0..* Potential target for the link
fhir:nodeId[ id ] ; # 1..1 Internal ID - target for link references
fhir:description[ string ] ; # 0..1 Why this node is specified
fhir:type[ code ] ; # 1..1 Type of resource this link refers to
fhir:profile[ canonical(StructureDefinition) ] ; # 0..1 Profile for the target resource
] ... ) ;
fhir:link( [ # 0..* Links this graph makes rules about
fhir:description[ string ] ; # 0..1 Why this link is specified
fhir:min[ integer ] ; # 0..1 Minimum occurrences for this link
fhir:max[ string ] ; # 0..1 Maximum occurrences for this link
fhir:sourceId[ id ] ; # 1..1 Source Node for this link
fhir:path[ string ] ; # 0..1 Path in the resource that contains the link
fhir:sliceName[ string ] ; # 0..1 Which slice (if profiled)
fhir:targetId[ id ] ; # 1..1 Target Node for this link
fhir:params[ string ] ; # 0..1 Criteria for reverse lookup
fhir:compartment( [ # 0..* Compartment Consistency Rules
fhir:use[ code ] ; # 1..1 where | requires
fhir:rule[ code ] ; # 1..1 identical | matching | different | custom
fhir:code[ code ] ; # 1..1 Patient | Encounter | RelatedPerson | Practitioner | Device | EpisodeOfCare
fhir:expression[ string ] ; # 0..1 Custom rule, as a FHIRPath expression
fhir:description[ string ] ; # 0..1 Documentation for FHIRPath expression
] ... ) ;
] ... ) ;
]
Definition of a graph of resources + Warning: Name should be usable as an identifier for the module by machine processing applications such as code generation
Canonical identifier for this graph definition, represented as a URI (globally unique) + Warning: URL should not contain | or # - these characters make processing canonical references problematic
@prefix fhir: <http://hl7.org/fhir/> .
[ a fhir:GraphDefinition;
fhir:nodeRole fhir:treeRoot; # if this is the parser root
# from Resource: .id, .meta, .implicitRules, and .language
# from DomainResource: .text, .contained, .extension, and .modifierExtension
fhir:url[ uri ] ; # 0..1 Canonical identifier for this graph definition, represented as a URI (globally unique)
fhir:identifier ( [ Identifier ] ... ) ; # 0..* Additional identifier for the GraphDefinition (business identifier)
fhir:version[ string ] ; # 0..1 Business version of the graph definition
# versionAlgorithm[x]: 0..1 How to compare versions. One of these 2
fhir:versionAlgorithm[ a fhir:string ; string ]
fhir:versionAlgorithm[ a fhir:Coding ; Coding ]
fhir:name[ string ] ; # 1..1 IName for this graph definition (computer friendly)
fhir:title[ string ] ; # 0..1 Name for this graph definition (human friendly)
fhir:status[ code ] ; # 1..1 draft | active | retired | unknown
fhir:experimental[ boolean ] ; # 0..1 For testing only - never for real usage
fhir:date[ dateTime ] ; # 0..1 Date last changed
fhir:publisher[ string ] ; # 0..1 Name of the publisher/steward (organization or individual)
fhir:contact ( [ ContactDetail ] ... ) ; # 0..* Contact details for the publisher
fhir:description[ markdown ] ; # 0..1 Natural language description of the graph definition
fhir:useContext ( [ UsageContext ] ... ) ; # 0..* The context that the content is intended to support
fhir:jurisdiction ( [ CodeableConcept ] ... ) ; # 0..* Intended jurisdiction for graph definition (if applicable)
fhir:purpose[ markdown ] ; # 0..1 Why this graph definition is defined
fhir:copyright[ markdown ] ; # 0..1 Use and/or publishing restrictions
fhir:copyrightLabel[ string ] ; # 0..1 Copyright holder and year(s)
fhir:start[ id ] ; # 0..1 Starting Node
fhir:node( [ # 0..* Potential target for the link
fhir:nodeId[ id ] ; # 1..1 Internal ID - target for link references
fhir:description[ string ] ; # 0..1 Why this node is specified
fhir:type[ code ] ; # 1..1 Type of resource this link refers to
fhir:profile[ canonical(StructureDefinition) ] ; # 0..1 Profile for the target resource
] ... ) ;
fhir:link( [ # 0..* Links this graph makes rules about
fhir:description[ string ] ; # 0..1 Why this link is specified
fhir:min[ integer ] ; # 0..1 Minimum occurrences for this link
fhir:max[ string ] ; # 0..1 Maximum occurrences for this link
fhir:sourceId[ id ] ; # 1..1 Source Node for this link
fhir:path[ string ] ; # 0..1 Path in the resource that contains the link
fhir:sliceName[ string ] ; # 0..1 Which slice (if profiled)
fhir:targetId[ id ] ; # 1..1 Target Node for this link
fhir:params[ string ] ; # 0..1 Criteria for reverse lookup
fhir:compartment( [ # 0..* Compartment Consistency Rules
fhir:use[ code ] ; # 1..1 where | requires
fhir:rule[ code ] ; # 1..1 identical | matching | different | custom
fhir:code[ code ] ; # 1..1 Patient | Encounter | RelatedPerson | Practitioner | Device | EpisodeOfCare
fhir:expression[ string ] ; # 0..1 Custom rule, as a FHIRPath expression
fhir:description[ string ] ; # 0..1 Documentation for FHIRPath expression
] ... ) ;
] ... ) ;
]
URL should not contain | or # - these characters make processing canonical references problematic
exists() implies matches('^[^|# ]+$')
5.10.4.3 Using GraphDefinitions
The GraphDefinition resource can be used to:
Summarize a set of profiles on resources
Define a graph of resources to return in a query
Define a graph of resources to include in a document
Document rules about the relationship between a set of resources e.g. must all resources concern the same patient?
5.10.4.3.1 Summarize a set of profiles on resources
FHIR resources are relatively granular. In many/most cases, many resources
are needed to handle any particular task. A typical example of this is a complex
diagnostic report: it will start with a DiagnosticReport, which will link
to a set of panels (Observation resources), each of which link to a set of
Observation resources for atomic data items.
One way to represent this is to profile each of the resources, creating
hundreds of profiles, and then leave it to the user to infer the overall pattern
of the report from the detailed profiles for each observation in the report.
But it's not easy to see the forest for the trees. A GraphDefinition can
summarize the overall picture and present a summary to the user.
Here's an example of
the kind of summary this represents. (Todo: make this an actual graph definition,
and clone into the main spec)
5.10.4.3.2 Fetching a graph of resources
As another example of using many resources, to completely represent a medication dispense,
an application needs not only the MedicationDispense resource, but also resources to represent the patient,
provider, organizations, and the associated prescription.
A client can retrieve a single resource:
GET [base]/MedicationDispense/example
Then, when it reads the returned resource, it can fetch the referenced resources:
GET [base]/Patient/example
GET [base]/Practitioner/example
GET [base]/MedicationRequest/example
... etc.
This is a very inefficient way to retrieve all the required resources. An alternative
approach is to do a search, and _include the required resources:
GET [base]/MedicationDispense?_id=example
&_include=MedicationDispense:authorizingPrescription
&_include=MedicationDispense:subject
But scaling this approach to fetch a full package with its dependencies
becomes increasingly difficult as the package gets deeper. A graph definition
can be used instead to inform the server what to return with the resource using the $graph operation:
GET [base]/MedicationDispense/example/$graph?graph=med-package
This is a request to return a graph of resource, using the graph definition
'med-package'. In this case, the graph definition would look approximately
like this:
Systems may either provide a pre-defined list of graph definitions that clients may choose
from, or allow clients to define their own GraphDefinition resources and then refer to them.
Server may also allow clients also pass in their own graph definition using a text representation:
GET [base]/MedicationDispense/example/$graph?definition=Patient{managingOrganization:Organization{endpoint:Endpoint}}
See below for further details.
5.10.4.3.3 Building a document
A very similar issue applies when building a document using the $document operation.
A document must include all the resources linked directly from the composition, but
whether to include additional linked resources is at the discretion of the document author. How does
the user inform the $document operation which linked resources to include? One option is a boolean flag
for including all linked data, but this may be extensive - up to an entire patient record - and
may include resources that are not desired.
An operation can use a graph definition as a parameter to the $document operation:
GET [base]/Composition/example/$document?graph=example
This tells the server to include the graph of resources defined
in the example GraphDefinition - in this case,
any resources referred to from lists, when the section content is a list. Alternatively,
servers may allow a client to pass in a definition directly (as shown above) using
the parameter definition.
5.10.4.3.4 Document Relationship rules
One important question about the use of resources is cross-resource consistency.
For example, if an Observation refers to both a Patient and Encounter, does the
Encounter have to refer to the same patient?
In general, the answer to this is that it usually should - the record needs to be
consistent. However there are edge cases where the references may differ. For
example, with regard to patient references, they may differ for:
Health Records concerning mother and baby
Organ Transplants
Counselling, particularly family counselling
Other reasons for the references to differ - mixing records about the same patient
from different servers, or specific records about patients mixed with records
about groups of patients (particularly common in veterinarian care).
The GraphDefinition resource allows for compartment consistency rules
to be made regarding the links between resources. For each link in the graph, the
graph definition can make a rule about the compartment consistency. The rule
can specify one of the following consistencies:
Code
Meaning
identical
The compartment must be identical (the same literal reference)
matching
The compartment must be the same - the record must be about the same patient, but the reference may be different
different
The compartment must be different
custom
The compartment rule is defined in the accompanying FHIRPath expression
Todo: how would this be validated? - where is the graph referred to?
5.10.4.4 Simple Examples
GraphDefinition can walk a graph forward through the links, which is the natural order.
As an example, consider a profile on composition that wants to specify that the graph
follows the section entry into a list:
<!-- this graph starts with a composition. We don't care what the specific profile is
(though the statement above 'this case doesn't cross patients' implies that we do care a little) -->
<start value="composition1"/>
<node>
<nodeId value="composition1"/>
<description value="The base composition"/>
<type value="Composition"/>
<profile value="http://hl7.org/fhir/StructureDefinition/clinicaldocument"/>
</node>
<node>
<nodeId value="list1"/>
<description value="A list resource that a section entry reference points to"/>
<type value="List"/>
</node>
<node>
<nodeId value="resN"/>
<description value="Generic resource that's the target of a list reference"/>
<type value="Resource"/>
</node>
<!-- define the section -> list link -->
<link>
<description value="Link from Composition.section to list"/>
<sourceId value="composition1"/>
<!-- any section entry. Todo: this recurses; are we profiling this at all levels? -->
<path value="Composition.section.entry"/>
<!--
one target. This graph is not making rules about the content of the section entries - that
would be done in a profile. it's just saying, if you see a reference to a list in a section
entry, these are the rules that describe the graph
-->
<targetId value="list1"/>
</link>
<!-- and from the list, any references -->
<link>
<description value="Include any list entries"/>
<sourceId value="list1"/>
<path value="List.entry.item"/>
<targetId value="resN"/>
</link>
The pattern can be extended to require that the none of the resources
in this little graph reference a different patient - that is, it is a
requirement when traversing the links that the patient compartment
remain identical:
<!-- define the section -> list link -->
<link>
<description value="Link from Composition.section to list"/>
<sourceId value="composition1"/>
<!-- any section entry. Todo: this recurses; are we profiling this at all levels? -->
<path value="Composition.section.entry"/>
<!--
one target. This graph is not making rules about the content of the section entries - that
would be done in a profile. it's just saying, if you see a reference to a list in a section
entry, these are the rules that describe the graph
-->
<targetId value="list1"/>
<compartment>
<use value="requirement"/>
<code value="Patient"/>
<rule value="identical"/>
</compartment>
</link>
<!-- and from the list, any references -->
<link>
<description value="Include any list entries"/>
<sourceId value="list1"/>
<path value="List.entry.item"/>
<targetId value="resN"/>
<compartment>
<use value="requirement"/>
<code value="Patient"/>
<rule value="identical"/>
</compartment>
</link>
Another useful approach is to walk any link * (e.g. wildcard support):
<link>
<sourceId value="comp1">
<!--
follow all links that refers to a list
-->
<path value="*"/>
<targetId value="list1">
</link>
<!-- and inside this list, follow any references -->
<link>
<sourceId value="list1">
<path value="*"/>
<targetId value="resN"/>
</link>
It's also possible to build a graph by walking links in reverse.
This technique is useful, for example, when including provenance resources in a document:
<link>
<sourceId value="obs1"/>
<!-- if the path is missing <path value=""/> -->
<targetId value="prov"/> <!-- any provenance -->
<!--
and specify the search parameters to include
all the provenances that refer to this resource
-->
<params value="target={ref}"/>
</link>
5.10.4.4.1 Text Representation
For convenience, a graph definition may be represented using a text short hand form.
A graph definition is introduced by two or more node statements:
node start comp1 'The base composition' = Composition (http://hl7.org/fhir/StructureDefinition/clinicaldocument);
node list1 'A list resource that a section entry reference points to' = List;
node resN 'Generic resource that's the target of a list reference' = Resource;
Each node declaration starts with the fixed word node followed by the nodeId, the node description,
=, the node type, and then, if the node has a profile, the profile value in brackets. One of the
nodes have the key word "start"
Once all the nodes are declared, then the links are declared:
link 'Link from Composition.section to list' = comp1[Composition.section.entry] -> list1 requires identical patient;
link 'Include any list entries' = list1[List.entry.item] -> resN requires identical patient;
The format for a link is:
link '{description}' {min}..{max} = sourceId[{path}:{sliceName}] {targetId}?{params} {resource-compartment-rules};
where:
description, min, max, path, sliceName, params and resource-type-rules are all optional
resource-compartment-rules is present if there's a compartent, and the syntax is {use} {rule} {code}. If the rule is 'custom' then the syntax is '=' '{fhirpath}' '{description}'
In this format, the amount of whitespace, and its form, is irrelevant. For the purposes
of clarity, the statements here are laid out on different lines, but this is not required.
When a GraphDefinition is used as a parameter in a url, no extra whitespace is used, just
a single space, and the descriptions SHOULD be omitted.
Here is a full example that uses all the features of the syntax:
node pat = Patient;
node org = Organization;
node org2 = Organization;
node ept = EndPoint;
node prac = Practitioner;
node grp = Group;
link 'patient managing org' 0..1 = pat[managingOrganization] -> org;
link = org[endpoint] -> ept
link 'groups patient is in' = pat -> grp?item={ref}
link 'registered GP' = pat[generalPractitioner] -> org;
link 'Observations for the patient' = pat -> org?patient={ref};
link = org[performer] -> prac;
link = org[related.where(type='has-member').target] -> org2 requires matching Patient;
link = org[related.where(type='derived-from').target] -> org2 requires identical Patient;
link = org[related.where(type='sequel-to').target] -> org2 requires different Organization;
link = org[related.where(type='qualified-by').target] -> org2 requires custom Patient = path;