R4 Ballot #1 (Mixed Normative/Trial use)

This page is part of the FHIR Specification (v3.3.0: R4 Ballot 2). The current version which supercedes this version is 5.0.0. For a full list of available versions, see the Directory of published versions . Page versions: R5 R4B R4 R3 R2

FHIR Infrastructure Work GroupMaturity Level: N/ABallot Status: Informative

One common use of slicing is to describe different constraints on different kinds of patient contact details. In this example, Patient.telecom is defined as: ContactPoint [0..*] where the ContactPoint has system, value and use.

Consider the case where the profile should say:

  • There must be a single home phone number
  • The patient may have a work phone number
  • The patient may have an email address
  • No other types of contact are allowed

An example of a patient resource that meets these rules:

<Patient>                         
... snip ...
  <telecom>                       
    <system value="phone" />      
    <use value="home" />          
    <value value="5551234567" />  
  </telecom>
  <telecom>                       
    <system value="email" />      
    <value value="someone@acme.org" />  
  </telecom>
... snip ...
</Patient>

To do this, the profile that implements these rules needs to do the following:

  • On the base Patient.telecom element: define that slicing is discriminated by system and use, and that the slices are closed ("No other types of contact are allowed"). Order is left unfixed
  • Then define 3 slices:
    1. home phone: fixed values for system and use, and slice cardinality 1..1
    2. work phone: fixed values for system and use, and slice cardinality 0..1
    3. email: fixed value for system, cardinality 0..0 for use, and slice cardinality 0..1

In a StructureDefinition, this will look like:

  <!-- setting up the slicing -->
  <element> 
    <path value="Patient.telecom"/>
    <slicing>
      <discriminator value="system"/>
        <type value="value"/>
        <path value="system"/>
      </discriminator/>
      <discriminator value="system"/>
        <type value="value"/>
        <path value="use"/>
      </discriminator/>
      <rules value="closed"/>
    </slicing>
    <!-- net cardinality rules -->
    <min value="1"/>
    <max value="3"/>
  </element>
  
  <!-- first slice: home phone -->
  <element> 
    <path value="Patient.telecom"/>
    <name value="HomePhone"/> <!-- mandatory - gives the slice a name -->
    <min value="1"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Patient.telecom.system"/>
    <min value="1"/>
    <fixedCode value="phone"/>
  </element>
  <element> 
    <path value="Patient.telecom.value"/>
    <min value="1"/>
  </element>
  <element> 
    <path value="Patient.telecom.use"/>
    <min value="1"/>
    <fixedCode value="home"/>
  </element>
  
  <!-- second slice: work phone -->
  <element> 
    <path value="Patient.telecom"/>
    <name value="WorkPhone"/> <!-- mandatory - gives the slice a name -->
    <min value="0"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Patient.telecom.system"/>
    <min value="1"/>
    <fixedCode value="phone"/>
  </element>
  <element> 
    <path value="Patient.telecom.value"/>
    <min value="1"/>
  </element>
  <element> 
    <path value="Patient.telecom.use"/>
    <min value="1"/>
    <fixedCode value="work"/>
  </element>
  
  <!-- third slice: email -->
  <element> 
    <path value="Patient.telecom"/>
    <name value="Email"/> <!-- mandatory - gives the slice a name -->
    <min value="0"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Patient.telecom.system"/>
    <min value="1"/>
    <fixedCode value="email"/>
  </element>
  <element> 
    <path value="Patient.telecom.value"/>
    <min value="1"/>
  </element>
  <element> 
    <path value="Patient.telecom.use"/>
    <max value="0"/>
  </element>

Note: much of the definition detail has been left out, and only the parts relevant to the pattern are shown. Also, providing a fixed value makes the minimum cardinality irrelevant, but it is shown here for completeness.

This table illustrates the relationship between the instance and the ElementDefinitions:

Path Name Min Max Fixed
<Patient> Patient
Patient.telecom 1 3
<telecom> Patient.telecom HomePhone 1 1
 <system value="phone" /> Patient.telecom.system 1 1 phone
 <value value="5551234567" /> Patient.telecom.value 1 1
 <use value="home" /> Patient.telecom.use 1 1 home
</telecom>
Patient.telecom WorkPhone 0 1
<telecom> Patient.telecom Email 0 1
 <system value="email" /> Patient.telecom.system 1 1 email
 <value value="someone@acme.org" /> Patient.telecom.value 1 1
 </telecom> Patient.telecom.use

Another use of Slicing is for Blood Pressure Measurements, where the profile says:

  • There must be two components
  • The first has LOINC code 8480-6, and a quantity
  • The second has LOINC code 8462-4, and a quantity
  • Other components are allowed (posture, etc., but not profiled in the base blood pressure profile)

An example of an Observation resource that meets these rules:

<Observation>                         
... snip ...
  <component>                       
    <code>                       
      <coding>                       
        <system value="http://loinc.org" />      
        <code value="8480-6" />  
        <display value="Systolic blood pressure" />  
      </coding>                       
    </code>                       
    <valueQuantity>                       
      <value value="120" />  
      <unit value="mmHg" />  
      <system value="http://unitsofmeasure.org" />  
      <code value="mm[Hg]" />  
    </valueQuantity>                       
  </component>
  <component>                       
    <code>                       
      <coding>                       
        <system value="http://loinc.org" />      
        <code value="8462-4" />  
        <display value="Diastolic blood pressure" />  
      </coding>                       
    </code>                       
    <valueQuantity>                       
      <value value="80" />  
      <unit value="mmHg" />  
      <system value="http://unitsofmeasure.org" />  
      <code value="mm[Hg]" />  
    </valueQuantity>                       
  </component>
</Patient>

To do this, the profile that implements these rules needs to do the following:

  • On the base Observation.component element: define that slicing is discriminated by code. Order is left unfixed, and rules are left open.
  • Then define 2 slices:
    1. systolic: fixed values for code, cardinality 1..1, value is a Quantity
    2. diastolic: fixed values for code, cardinality 1..1, value is a Quantity

In a StructureDefinition, this will look like:

  <!-- setting up the slicing -->
  <element> 
    <path value="Observation.component"/>
    <slicing>
      <discriminator value="system"/>
        <type value="value"/>
        <path value="code"/>
      </discriminator/>
    </slicing>
    <!-- net cardinality rules -->
    <min value="2"/>
    <max value="*"/>
  </element>
  
  <!-- first slice: systolic -->
  <element> 
    <path value="Observation.component"/>
    <name value="systolic"/> <!-- mandatory - gives the slice a name -->
    <min value="1"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Observation.component.code"/>
    <min value="1"/>
    <fixedCodeableConcept>
      <coding>                       
        <system value="http://loinc.org" />      
        <code value="8480-6" />  
        <display value="Systolic blood pressure" />  
      </coding>                       
    </fixedCodeableConcept>
  </element>
  <element> 
    <path value="Observation.component.valueQuantity"/>
    <min value="1"/>
  </element>
  
  <!-- second slice: diastolic -->
  <element> 
    <path value="Observation.component"/>
    <name value="diastolic"/> <!-- mandatory - gives the slice a name -->
    <min value="1"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Observation.component.code"/>
    <min value="1"/>
    <fixedCodeableConcept>
      <coding>                       
        <system value="http://loinc.org" />      
        <code value="8462-4" />  
        <display value="Diastolic blood pressure" />  
      </coding>                       
    </fixedCodeableConcept>
  </element>
  <element> 
    <path value="Observation.component.valueQuantity"/>
    <min value="1"/>
  </element>

Note: much of the definition detail has been left out, and only the parts relevant to the pattern are shown. E.g. A real blood pressure profile would fix unit, an overall Observation code etc.

For another example, consider slicing extensions. The base extension on every element is defined as a list (0..*) of extensions, and each extension has a url that identifies it, and a value. Consider an example where a profile defines that for a particular element (named Patient), there are two extensions, with URLs http://acme.com/a and http://acme.com/b. In addition, the profile allows other extensions to be used.

Technically, the profile achieves this by "slicing" the extension list, into two slices, and saying that the slicing is "open" - that there can be other slices introduced. Here are the relevant parts of the Profile on patient:


<StructureDefinition xmlns="http://hl7.org/fhir">
  <!-- snip -->
    <baseType value="Patient" />
    <baseDefinition value="http://hl7.org.fhir/StructureDefinition/Patient" />
    <derivation value="constraint" />
    <snapshot>
      <element>
        <path value="Patient"/>
        <!-- snip -->
      </element>
      <element>
        <path value="Patient.extension"/>
        <!-- this first element defines the slicing, and carries the base definition forward -->
        <slicing>
          <!-- Extensions are always discriminated by URL -->
          <discriminator value="system"/>
            <type value="value"/>
            <path value="url"/>
          </discriminator/>
          <ordered value="false"/>     <!-- we don't care what order they appear in -->
          <rules value="open"/>        <!-- other extensions can be used -->
        </slicing>
        <!-- -- snip definition -->
      </element>
      <!-- first extension -->
      <element>
        <path value="Patient.extension"/>
        <!-- snip most of definition -->
        <type>
         <code value="Extension"/>
         <!-- the profile for an extension is a reference to the extension definition itself -
           this implies a profile, and happens to fix the @url value to the desired URL -->
         <profile value="http://acme.com/a"/>
        </type>
      </element>
      <!-- second extension -->
      <element>
        <path value="Patient.extension"/>
        <!-- snip most of definition -->
        <type>
         <code value="Extension"/>
         <!-- the profile for an extension is a reference to the extension definition itself -
           this implies a profile, and happens to fix the @url value to the desired URL -->
         <profile value="http://acme.com/b"/>
        </type>
      </element>
      <!-- snip rest of profile -->
    </snapshot>
</StructureDefinition>

Here is a patient example that conforms to this profile:

<Patient xmlns="http://hl7.org/fhir">
  <!-- two extensions, the order doesn't matter -->
  <extension url="http://acme.com/b">
    <!-- this has the right url, and so matches the second slice -->
    <!-- snip whatever value extension would have -->
  </extension>
  <extension url="http://acme.com/a">
    <!-- this has the right url, and so matches the first slice -->
    <!-- snip whatever value extension would have -->
  </extension>
  <!-- the rest of patient -->
</Patient>

In this example, a profile on a diagnostic report says that it must have four observations, each with a different LOINC code (e.g. a classic lab panel). In this case (taken from the Example Lipid Profile), the structure that applies to DiagnosticReport will say that there are four slices on DiagnosticReport.result, each conforming to a different structure, which are also contained in the same profile. Each of those structures will constrain the LOINC code in the observation.

<!-- first structure, the DiagnosticReport -->
<StructureDefinition xmlns="http://hl7.org/fhir">
  <!-- snip -->
  <url value="http://acme.org/fhir/StructureDefinition/lipid-report"/>
  <name value="LipidProfile"/>
  <baseType value="DiagnosticReport"/>
  <baseDefinition value="http://hl7.org/fhir/StructureDefinition/DiagnosticReport"/>
  <derivation value="constraint" />

  <!-- snip -->
  <snapshot>
    <!-- snip elements -->
    <element>
      <!-- first definition for result -->
      <path value="DiagnosticReport.result"/>
      <slicing>
        <!-- this is sliced by the code value of the target of the reference -->
        <discriminator value="system"/>
          <type value="value"/>
          <path value="reference.code"/>
        </discriminator/>
        <!-- have to be in the specified order -->
        <ordered value="true"/>
        <!-- this profile says, no other observations allowed -->
        <rules value="closed"/>
      </slicing>
      <!-- snip definition -->
    </element>
    <!-- first slice: Cholesterol -->
    <element>
      <path value="DiagnosticReport.result"/>
      <name value="Cholesterol"/>
      <!-- snip definition parts -->
      <type>
        <code value="Reference"/>
        <!-- this element must conform to the "Cholesterol" structure -->
        <profile value="http://acme.org/fhir/StructureDefinition/Cholesterol"/>
      </type>
    </element>
    <!-- next 3 slices all the same, but different names for profile -->
    <element>
      <path value="DiagnosticReport.result"/>
      <name value="Triglyceride"/>
      <!-- snip definition parts -->
      <type>
        <code value="Reference"/>
        <!-- this element must conform to the "Triglyceride" structure -->
        <profile value="http://acme.org/fhir/StructureDefinition/Triglyceride"/>
      </type>
    </element>
    <element>
      <path value="DiagnosticReport.result"/>
      <name value="LDLCholesterol"/>
      <!-- snip definition parts -->
      <type>
        <code value="Reference"/>
        <!-- this element must conform to the "LDLCholesterol" structure -->
        <profile value="http://acme.org/fhir/StructureDefinition/LDLCholesterol"/>
      </type>
    </element>
    <element>
      <path value="DiagnosticReport.result"/>
      <name value="HDLCholesterol"/>
      <!-- snip definition parts -->
      <type>
        <code value="Reference"/>
        <!-- this element must conform to the "HDLCholesterol" structure -->
        <profile value="http://acme.org/fhir/StructureDefinition/HDLCholesterol"/>
      </type>
    </element>
    <!-- snip elements -->
  </snapshot>
</StructureDefinition>

<!-- now, the second structure, for the Cholesterol profile -->
<StructureDefinition>
  <url value="http://acme.org/fhir/StructureDefinition/Cholesterol"/>
  <name value="Cholesterol"/> 
  <baseType value="Observation"/>
  <baseDefinition value="http://hl7.org/fhir/StructureDefinition/Observation"/>
  <derivation value="constraint" />
  <snapshot>
    <!-- snip elements -->
    <element>
      <!-- this the element definition for name. Because of the
        slicing / discriminator rules in the LipidReport profile
        that references it, it is required to fix the value of
        the name element -->
      <path value="Observation.code"/>
      <definition>
         <!-- there are actually 3 ways to fix a CodeableConcept
         to a single fixed value. Here, we used the simplest one -->
         <valueCodeableConcept>
           <!-- just fix the value to the right code -->
           <coding>
             <system value="http://loinc.org"/>
             <code value="35200-5"/>
             <display value="Cholesterol"/>
           </coding>
         </valueCodeableConcept>
      </definition>
    </element>
    <!-- snip elements -->
  </snapshot>
</StructureDefinition>

<!-- Triglyceride profile -->
<StructureDefinition>
  <url value="http://acme.org/fhir/StructureDefinition/Triglyceride"/>
  <name value="Triglyceride"/> 
  <baseType value="Observation"/>
  <baseDefinition value="http://hl7.org/fhir/StructureDefinition/Observation"/>
  <derivation value="constraint" />
  <snapshot>
    <!-- snip elements -->
    <element>
      <!-- this the element definition for name. Because of the
        slicing / discriminator rules in the LipidReport profile
        that references it, it is required to fix the value of
        the name element -->
      <path value="Observation.code"/>
      <definition>
         <!-- there's actually 3 ways to fix a CodeableConcept
         to a single fixed value. Here, we used the simplest one -->
         <valueCodeableConcept>
           <!-- just fix the value to the right code -->
           <coding>
             <system value="http://loinc.org"/>
             <code value="35217-9"/>
             <display value="Triglyceride"/>
           </coding>
         </valueCodeableConcept>
      </definition>
    </element>
    <!-- snip elements -->
  </snapshot>
</StructureDefinition>

<!-- LDLCholesterol profile -->
<StructureDefinition>
  <url value="http://acme.org/fhir/StructureDefinition/LDLCholesterol"/>
  <name value="LDLCholesterol"/> 
  <baseType value="Observation"/>
  <baseDefinition value="http://hl7.org/fhir/StructureDefinition/Observation"/>
  <derivation value="constraint" />
  <snapshot>
    <!-- snip elements -->
    <element>
      <!-- this the element definition for name. Because of the
        slicing / discriminator rules in the LipidReport profile
        that references it, it is required to fix the value of
        the name element -->
      <path value="Observation.code"/>
      <definition>
         <!-- because of the way that LDL cholesterol measurements works
           (well, in this context- it varies), there's 2 different LOINC
           codes for either measured or calculated. So here, we bind to
           a value set -->
       <binding>
        <conformance value="required"/> <!-- must be required if this is a discriminator -->
        <!-- snip the actual value set reference, but it refers to a value
             set with two LOINC codes, one for each kind of LDL, which in
             this case are LOINC codes 18262-6 and 13457-7 -->
       </binding>
     </definition>
    </element>
    <!-- snip elements -->
  </snapshot>
</StructureDefinition>

<!-- HDLCholesterol profile -->
<StructureDefinition>
  <url value="http://acme.org/fhir/StructureDefinition/HDLCholesterol"/>
  <name value="HDLCholesterol"/> 
  <baseType value="Observation"/>
  <baseDefinition value="http://hl7.org/fhir/StructureDefinition/Observation"/>
  <derivation value="constraint" />
  <snapshot>
    <!-- snip elements -->
    <element>
      <!-- this the element definition for name. Because of the
        slicing / discriminator rules in the LipidReport profile
        that references it, it is required to fix the value of
        the name element -->
      <path value="Observation.code"/>
      <definition>
         <!-- there's actually 3 ways to fix a CodeableConcept
         to a single fixed value. Here, we used the simplest one -->
         <valueCodeableConcept>
           <!-- just fix the value to the right code -->
           <coding>
             <system value="http://loinc.org"/>
             <code value="2085-9"/>
             <display value="LDL Cholesterol"/>
           </coding>
         </valueCodeableConcept>
      </definition>
    </element>
    <!-- snip elements -->
  </snapshot>
</StructureDefinition>

Here is an instance that meets the rules for this profile:

<!-- first, the diagnostic report -->
<DiagnosticReport xmlns="http://hl7.org/fhir">
  <!-- snip -->
  <!-- here's the set of results. We don't know what
    slices they are or anything until we go off, find
    the references, and look in them -->
  <result>
    <reference value="Observation/cholesterol"/>
  </result>
  <result>
    <reference value="Observation/triglyceride"/>
  </result>
  <result>
    <reference value="Observation/ldlcholesterol"/>
  </result>
  <result>
    <reference value="Observation/hdlcholesterol"/>
  </result>
  <!-- snip -->
</DiagnosticReport>

<!-- Observation, id = cholesterol -->
<Observation xmlns="http://hl7.org/fhir">
  <!-- the observation starts with the name, as specified
    by the profile for the first slice  -->
  <code>
    <coding>
      <system value="http://loinc.org"/>
      <code value="35200-5"/>
      <display value="Cholesterol"/>
    </coding>
  </code>
  <!-- snip -->
</Observation>

<!-- Observation, id = triglyceride -->
<!-- this code matches the second slice. good -->
<Observation xmlns="http://hl7.org/fhir">
  <code>
    <coding>
      <system value="http://loinc.org"/>
      <code value="35217-9"/>
      <display value="Triglyceride"/>
    </coding>
  </code>
  <!-- snip -->
</Observation>

<!-- Observation, id = hdlcholesterol -->
<!-- this code matches the fourth slice. good -->
<Observation xmlns="http://hl7.org/fhir">
  <code>
    <coding>
      <system value="http://loinc.org"/>
      <code value="2085-9"/>
      <display value="HDL Cholesterol"/>
    </coding>
  </code>
  <!-- snip -->
</Observation>

<!-- Observation, id = ldlcholesterol -->
<!-- this code matches the third slice. good -->
<Observation id="ldlcholesterol">
  <code>
    <coding>
      <system value="http://loinc.org"/>
      <code value="13457-7"/>
      <display value="LDL Chol. (Calc)"/>
    </coding>
  </code>
  <!-- snip -->
</Observation>

Note that this version is not valid, because the slices are not in the correct order:

<!-- first, the diagnostic report -->
<DiagnosticReport xmlns="http://hl7.org/fhir">
  <!-- snip -->
  <!-- here's the set of results. We don't know what
    slices they are or anything until we go off, find
    the references, and look in them -->
  <result>
    <reference value="Observation/cholesterol"/>
  </result>
  <result>
    <reference value="Observation/triglyceride"/>
  </result>
    <!-- Oops the last two are out of order -->
  <result>
    <reference value="Observation/hdlcholesterol"/>
  </result>
  <result>
    <reference value="Observation/ldlcholesterol"/>
  </result>
  <!-- snip -->
</DiagnosticReport>

Most uses of Composition involve conformance to a profile that specifies which sections will exist, and what their contents will be. This is yet another example of slicing. A typical document content profile might specify a section structure something like this:

  • Reason for visit Narrative, LOINC Code 29299-5
  • Medications, LOINC Code 46057-6
    • Prescribed Medications, LOINC Code 66149-6
    • OTC medications, 66150-4 (optional)
  • Vital Signs, LOINC Code 8716-3

Real profiles will contain lots of detail about the sections, but these are omitted here in the interests of clarity.

An example of a Composition that meets these rules:

<Composition>                         
... snip ...
  <section>                       
    <code>                       
      <coding>                       
        <system value="http://loinc.org" />      
        <use value="29299-5" />          
        <value value="Reason for visit Narrative" />  
      </coding>                       
    </code>                       
    ... snip ...
  </section>
  <section>                       
    <code>                       
      <coding>                       
        <system value="http://loinc.org" />      
        <use value="46057-6" />          
        <value value="Medications section" />  
      </coding>                       
    </code>                       
    ... snip ...
    <section>                       
      <code>                       
        <coding>                       
          <system value="http://loinc.org" />      
          <use value="66149-6" />          
          <value value="Prescribed medications" />  
        </coding>                       
      </code>                       
      ... snip ...
    </section>
    <section>                       
      <code>                       
        <coding>                       
          <system value="http://loinc.org" />      
          <use value="66150-4" />          
          <value value="O medications" />  
        </coding>                       
      </code>                       
      ... snip ...
    </section>
  </section>
  <section>                       
    <code>                       
      <coding>                       
        <system value="http://loinc.org" />      
        <use value="8716-3" />          
        <value value="Vital signs" />  
      </coding>                       
    </code>                       
    ... snip ...
  </section>
</Composition>

To do this, the profile that implements these rules needs to do the following:

  • On the base Composition.section element: define that slicing is discriminated by code, and that the slices are closed and ordered
  • Then define 3 slices on Composition.section:
    1. reason-for-visit: fixed code, cardinality 1..1
    2. medication: fixed code, cardinality 1..1
    3. vital-signs: fixed code, cardinality 1..1
  • Then, in the medication slice, slice the Composition.section.section: define that slicing is discriminated by code, and that the slices are closed and ordered
  • Then define 2 slices on the Composition.section.section in medication:
    1. prescribed: fixed code, cardinality 1..1
    2. otc: fixed code, cardinality 0..1

In a StructureDefinition, this will look like:

  <!-- setting up the slicing on Composition.section -->
  <element> 
    <path value="Composition.section"/>
    <slicing>
      <discriminator value="system"/>
        <type value="value"/>
        <path value="code"/>
      </discriminator/>
      <ordered value="true"/>
      <rules value="closed"/>
    </slicing>
    <!-- net cardinality rules -->
    <min value="3"/>
    <max value="3"/>
  </element>
  
  <!-- first slice: reason for visit -->
  <element> 
    <path value="Composition.section"/>
    <name value="reason-for-visit"/> 
    <min value="1"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Composition.section.code"/>
    <min value="1"/>
    <fixedCodeableConcept>
      <coding>                       
        <system value="http://loinc.org" />      
        <use value="29299-5" />          
        <value value="Reason for visit Narrative" />  
      </coding>                       
    </fixedCodeableConcept>
  </element>
  
  <!-- second slice: medications -->
  <element> 
    <path value="Composition.section"/>
    <name value="medications"/> 
    <min value="1"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Composition.section.code"/>
    <min value="1"/>
    <fixedCodeableConcept>
      <coding>                       
        <system value="http://loinc.org" />      
        <use value="46057-6" />          
        <value value="Medications section" />  
      </coding>                       
    </fixedCodeableConcept>
  </element>

  <!-- setting up the inner slicing on medication Composition.section.section -->
  <element> 
    <path value="Composition.section.section"/>
    <slicing>
      <discriminator value="system"/>
        <type value="value"/>
        <path value="code"/>
      </discriminator/>
      <ordered value="true"/>
      <rules value="closed"/>
    </slicing>
    <!-- net cardinality rules -->
    <min value="1"/>
    <max value="2"/>
  </element>

  <!-- first inner slice: prescribed medications -->
  <element> 
    <path value="Composition.section.section"/>
    <name value="prescribed"/> 
    <min value="1"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Composition.section.section.code"/>
    <min value="1"/>
    <fixedCodeableConcept>
      <coding>                       
        <system value="http://loinc.org" />      
        <use value="66149-6" />          
        <value value="Prescribed medications" />  
      </coding>                       
    </fixedCodeableConcept>
  </element>
  
  <!-- second inner slice: over the counter medications -->
  <element> 
    <path value="Composition.section.section"/>
    <name value="otc"/> 
    <min value="0"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Composition.section.section.code"/>
    <min value="1"/>
    <fixedCodeableConcept>
      <coding>                       
        <system value="http://loinc.org" />      
        <use value="66150-4" />          
        <value value="Over the counter medications" />  
      </coding>                       
    </fixedCodeableConcept>
  </element>
  
  <!-- third slice: Vital Signs -->
  <element> 
    <path value="Composition.section"/>
    <name value="vital-signs"/> 
    <min value="1"/>
    <max value="1"/>
  </element>
  <element> 
    <path value="Composition.section.code"/>
    <min value="1"/>
    <fixedCodeableConcept>
      <coding>                       
        <system value="http://loinc.org" />      
        <use value="8716-3" />          
        <value value="Vital signs" />  
      </coding>                       
    </fixedCodeableConcept>
  </element>

For some use cases, it is necessary to further reslice in a derivative profile. This is when Profile A derives from Profile B, Profile A slices an element in Slice A and B, and Profile B further slices Slice A into A1 and A2.

As an example, assume that institution implementation guide specifies that a medication list is represented by a List resource, where the items into the list are split up into 3 slices, based on resource type:

  • MedicationRequest
  • MedicationAdministration
  • MedicationStatement

These slices must come in order.

Now, a particular application creates an additional profile that specifies the following rules in addition to the base rules of the institution:

  1. the medicationRequests will be sliced into active and non-active, where all the active requests will come first, and then the inactive ones
  2. only active MedicationAdministrations will be encountered
  3. Medication statements are not supported

An example of a List resource that meets these rules:

<List>                         
... snip ...
  <entry>                       
    <item>  
      <reference value="MedicationRequest/ex-active-1"/>  
    </item>  
  </entry>
  <entry>                       
    <item>  
      <reference value="MedicationRequest/ex-active-2"/>  
    </item>  
  </entry>
  <entry>                       
    <item>  
      <reference value="MedicationRequest/ex-inactive-1"/>  
    </item>  
  </entry>
  <entry>                       
    <item>  
      <reference value="MedicationAdministration/ex-any-1"/>  
    </item>  
  </entry>

  ... snip ...
</List>

The StructureDefinition differential for the first profile sets up the slicing, and then defines 3 slices:

  <!-- setting up the slicing -->
  <element> 
    <path value="List.entry"/>
    <slicing>
      <discriminator>
        <type value="profile"/>
        <path value="entry.item.resolve()"/>
      </discriminator>
      <ordered value="true"/>
      <rules value="closed"/>
    </slicing>
  </element>
  
  <!-- first slice: medrequest -->
  <element> 
    <path value="List.entry"/>
    <name value="medrequest"/> <!-- mandatory - gives the slice a name -->
  </element>
  <element> 
    <path value="List.entry.item"/>
    <type>
      <code value="Reference"/>
      <targetProfile value="http://example.org/StructureDefinition/medrequest"/>
    </type>
  </element>
  
  <!-- second slice: medadmin -->
  <element> 
    <path value="List.entry"/>
    <name value="medadmin"/> <!-- mandatory - gives the slice a name -->
  </element>
  <element> 
    <path value="List.entry.item"/>
    <type>
      <code value="Reference"/>
      <targetProfile value="http://example.org/StructureDefinition/medadmin"/>
    </type>
  </element>
  
  <!-- third slice: medstmt -->
  <element> 
    <path value="List.entry"/>
    <name value="medstmt"/> <!-- mandatory - gives the slice a name -->
  </element>
  <element> 
    <path value="List.entry.item"/>
    <type>
      <code value="Reference"/>
      <targetProfile value="http://example.org/StructureDefinition/medstmt"/>
    </type>
  </element>

The derived StructureDefinition differential reslices the first slice into 2 slices, constrains the second slice, and prohibits the 3rd:

  <!-- setting up the slicing - this profile doesn't change the slicing 
    (though it could add new discriminators if it needed) -->
  <element> 
    <path value="List.entry"/>
    <slicing>
      <discriminator>
        <type value="profile"/>
        <path value="entry.item.resolve()"/>
      </discriminator>
      <ordered value="true"/>
      <rules value="closed"/>
    </slicing>
  </element>
  
  <!-- first slice: medrequest/active -->
  <element> 
    <path value="List.entry"/>
    <!-- name/sub-name: splitting an existing slice into new slices  -->
    <name value="medrequest/active"/> 
  </element>
  <element> 
    <path value="List.entry.item"/>
    <type>
      <code value="Reference"/>
      <targetProfile value="http://example.org/StructureDefinition/medrequest-active"/>
    </type>
  </element>
  
  <!-- second slice: medrequest/inactive -->
  <element> 
    <path value="List.entry"/>
    <name value="medrequest/inactive"/>
  </element>
  <element> 
    <path value="List.entry.item"/>
    <type>
      <code value="Reference"/>
      <targetProfile value="http://example.org/StructureDefinition/medrequest-active"/>
    </type>
  </element>
  
  <!-- third slice: medadmin (rule on the second slice in the parent profile) -->
  <element> 
    <path value="List.entry"/>
    <!-- use the same name - constraining the existing slice -->
    <name value="medadmin"/> 
  </element>
  <element> 
    <path value="List.entry.item"/>
    <type>
      <code value="Reference"/>
      <targetProfile value="http://example.org/StructureDefinition/medadmin-active"/>
    </type>
  </element>
  
  <!-- fourth slice: medstmt -->
  <element> 
    <path value="List.entry"/>
    <!-- use the same name - constraining the existing slice -->
    <name value="medstmt"/> 
    <max value="0"/>
  </element>