Welcome and introductions

2

@Peter Williams

@Rob Hausam

Recording, notes & attendance.

Check questions in Zulip and SNOMED International #snomed-hl7-fhir

Previous Meetings

2

@Peter Williams

@Rob Hausam

• 2024-07-03 - SNOMED on FHIR Meeting (TS & TB)

• 2024-06-25 - SNOMED on FHIR Meeting (TS & TB)

• 2024-06-05 - SNOMED on FHIR Meeting (TS & TB)

Other Meetings

10

@Peter Williams

@Rob Hausam

Recent Events:

• DevDays 2024: Good meetings and SNOMED presentations, from Peter, Dion, Michael Lawley did a VCL presentation (ECL for Valuesets applicable in other terminologies)

Future Events:

HL7 September 21 - 27 Atlanta, Georgia

SI Business Meetings and Expo  19 - 25 October 2024 Seoul, Republic of Korea

Other Regular Meetings:

HL7 Group TSMG (meeting Wed PM ET every other week) - Terminology Service Management Group (HTA Thursday AM is now a subgroup of the TSMG).   2022-05-17 RH Joint session with Vocab at last business meeting.  2022-06-14 Group has approved minimum capabilities for terminology servers.  Now looking at bigger/better HL7 Terminology Server 

HL7 Cross group projects : CDA & FHIR Translation Meetings (weekly): @Jay Lyle Going well, there is an implementation guide and the group is doing connectathons. May in Dallas and Minneapolis.  See https://confluence.hl7.org/display/CGP/C-CDA+to+FHIR+and+from+US+Core+Mapping

Document the use of an issue type to return incomplete validation

40

@Peter Jordan (Unlicensed) 

@michael lawley 

@Daniel Karlsson (Unlicensed) 

HL7 Jira ticket: https://jira.hl7.org/browse/FHIR-44810

The group discussed the topic and documented the discussion in the HL7 Jira ticket: https://jira.hl7.org/browse/FHIR-44810?focusedCommentId=231490

UPDATE June 25 2024: Notes from HL7 WGM in May:  Attendees agreed that we'd like more feedback from the SNOMED on FHIR committee on how this would work in practice in all the cases where it would apply to SNOMED CT.  Input on whether or not machine processable responses should be supported would be appreciated as well.

TBD: Discuss in future calls with @Kai Kewley and @michael lawley present

MedicationRequest Question

20

@Marie-Alexandra Lambot 

There are questions here currently regarding the Medication "line" (medication Statement I think in the international models) / Medication request / Medication dispense. One is, is there a way in FHIR to somehow link two medications to say that people have to take one drug X hours after the first one? 

How can we in MedicationRequest.dosage instruction express that a drug must be taken x hours after another drug which is in another medication line/Request? Can we somehow group two prescriptions to link them and give "global instructions" for both? Or would that pose other problems?

General discussion on work being done in EU

6

@Marie-Alexandra Lambot

@Daniel Karlsson (Unlicensed) 

See https://build.fhir.org/ig/hl7-eu/laboratory/branches/master/index.html

https://github.com/hl7-eu/laboratory

https://confluence.hl7.org/display/HEU/Laboratory+Report+Implementation+Guide%2C+Edition+1

https://build.fhir.org/ig/hl7-eu/xpandh-hdr/

ML: https://csct.be/projects.html

https://nuva.mesvaccins.net/

2023-09-19 ML CSCT now has official "Not for Profit" status as an entity.

2024-05-29 RH The HL7 Europe Laboratory Report FHIR IG will be balloted as a universal IG either in Sep 2024 or Jan 2025.

Identifying Modules - is most dependant the right thing to do?

@Jon Zammit 

Problem that changing the Canadian identifying module to the most dependant one (French Module) is not recognised by Snowstorm

The "most dependant module" idea doesn't continue to hold when we think about extensions that have multiple, equally dependant modules.   Also some external content (like LOINC) could be used by multiple countries and be the most dependant module, but could not be used as an identifier - it would no longer be unique.

2024-07-09 Review the guidance in the extensions practical guide to define how to select the identifying module, even when is not the most dependent module. The agreement of the group seems to be to separate the notion of package composition from module dependency. The terminology server will use the provided module as an identifier of packaged, which contents are defined elsewhere. The extension maintainer will choose which moduleId will identify his package, and needs to make this available to implementers. Maybe this is a use case for annotations. 

Alternative approach AP: NHS has created a composition module, that is created only for the purpose of referring to the other 5 NHS modules, and be used as the identifier of the package.

Model Binding

@Jay Lyle 

Model Binding, specifically model binding for elements within resources (rather than values).

Previous binding discussions:   Bindings to FHIR Clinical Resources. See also terminology binding

See demo which shows various elements in FHIR resources and how those would be populated from a SNOMED concept:  https://ihtsdo.github.io/sct-implementation-demonstrator/#/context. but working back from that, we could identify the SNOMED Attribute Types, which are relevant for the various elements in each resource.

JL see https://hl7.org/fhir/R4/condition-mappings.html#sct-attr or, more challenging: https://build.fhir.org/observation-mappings.html#sct-attr

Also discussion on negation in Condition, Allergy and Family History (see 160266009 |No family history of clinical finding (situation)|)

Also on the subject of the purpose and value of putting time into this:  if mapped to the appropriate SNOMED Attribute Types in compatibility with the Concept Model, that would allow an automatic transformation to a Post Coordinated concept - or possibly identifying as a Pre-coordinated one.

New Slimline Agenda 

@Peter Williams 

See discussion pages for links to archived discussion:  Discussions

Problem lists in FHIR

@Marie-Alexandra Lambot 

How are Problem Lists (aka concern list) shared in FHIR? 

PJ Transfers of patient notes / records contain problem lists.   Suggestion to use Observations.

AP Shared https://developer.nhs.uk/apis/digital-maternity-1-0-0/explore_problems_list.html based on List Resource

Background reading: Medical Records That Guide And Teach also "Medicine in Denial" (also updated - 

"Ending Medicines Chronic Dysfunction")

Affiliate license

@Peter Jordan (Unlicensed) 

Questions around the Affiliate license in Zulip 

See affiliate license - https://www.snomed.org/_files/ugd/900274_689013e9e0c74d23892abe9caee02612.pdf

2024-07-09 New section in IHTSDO Freshdesk with Licensing FAQs: https://ihtsdo.freshdesk.com/support/solutions/4000002199

Exposing Relationship Grouping in FHIR

@Peter Jordan (Unlicensed) 

@michael lawley 

Zulip discussion on exposing relationship role grouping: https://chat.fhir.org/#narrow/stream/179202-terminology/topic/SNOMED.20relationship.20groups

Ontoserver uses their own extension: https://ontoserver.csiro.au/docs/6/ext-subproperties-cs.html

OWL representation groups explicitly with 609096000 |Role group (attribute)|

ALO Since you already have access to the normal form as a property, role grouping could be determined from this.

Asserter / recorder

@Marie-Alexandra Lambot 

How should we deal with the situation where the person entering data into some system may not be responsible for the record.

PJ See Immunization Performer - each one has an action and an actor

https://www.hl7.org/fhir/provenance.html ( fully featured & flexible but heavy handed approach)

See also https://www.hl7.org/fhir/R5/participant.html and https://build.fhir.org/procedure.html

R6 (unchanged from R5) states:  9.2.4.8 Use of Condition.asserter 
If the data enterer is different from the asserter and needs to be known, this could be captured using a Provenance instance pointing to the Condition. For example, it is possible that a nurse records the condition on behalf of a physician. The physician is taking responsibility, despite the nurse entering it into the medical record.

DK Would expect to see multiple Condition resources where multiple parties assert a diagnosis

JL See discussed list of possible participants https://confluence.hl7.org/display/CGP/Provenance+Domain also (DK) see ValueSet https://hl7.org/fhir/R5/valueset-participation-role-type.html

Group agreed that the current wording in the specification is unclear.

Introduction to SNOMED with FHIR YouTube recordings of Dev Days Presentations:

https://www.youtube.com/watch?v=40Lvv2t8OxU

https://www.youtube.com/watch?v=x6UqJtcJwmQ

https://www.youtube.com/watch?v=8cXRAtNoSvc

SNOMED Concept Model & Resources elements mapping

@Jay Lyle 

Information on Mapping SNOMED Concept Model to FHIR Resource properties

@Andrew Perry (Unlicensed) There is info in the FHIR home page: Bindings to FHIR Clinical Resources There discussions in termInfo about semantic overlaps but they were not exhaustive. UK has specific advice about what attributes to avoid.

Round table updates

@Marie-Alexandra Lambot 

Belgium Update + TX IG Support Update - authorities looking for independence from HL7 servers, partially due to performance considerations.   RH Suggested self hosted TxFHIR.org server code (needs built locally unless Windows install, see https://github.com/HealthIntersections/fhirserver ) available (or Ontoserver of course - as of v6.15).

ML suggested:  See http://tx.fhir.org/tx-reg/ and https://github.com/FHIR/ig-registry/blob/master/tx-registry-doco.md for alternative terminology server support

MAL: Is every FHIR message validated, and if so, against a separate server?  RH Having the same discussion with US implementation - unsure of consensus.   Group discussion on: if a message fails validation, what are you going to do with it?  Quarantine (can then get stuck)?    Bounce back (not implemented?).   ML Usually systems are not synchronously coupled, so a message rejection is not possible. Watch out for CodeSystem updates eg change to display text for a code suddenly causing large volumes of rejections.  Also watch out for text in ValueSet definitions as it can also go out of date. PWI SNOMED-CT has the option to be more forgiving for inactive descriptions.   

Belgium (via Translation User Group) would like to inactivate descriptions of inactive concepts to help avoid apparent duplications.  ML Ontoserver will detect this and use "best case" instead eg most recently inactivated descriptions.

See also https://build.fhir.org/episodeofcare.html (unchanged since R4) collects together a set of temporally linked encounters.

Valueset Version Validation issue in IPS

@Rob Hausam 

RH: IPS Guide has bindings to value sets expressed with the International Edition version. Validation of content with the Canadian edition fails, even if it refers to concepts in the international edition.

PJ: sharing version in the codeable concept is complex for validation

ML: using the international edition in the valueset definition is limiting in this setting

RH: Version is overloaded to represent edition + release 

Plan: Not using the version in the value set definition. Verify if the validator is firing appropriately. it should be a warning in some cases. What would be the guideline on wether to include version or not when sharing a codeable concept? Share both your national concept + the closest concept in the international edition?

How to package content IGs with Terminology IGs

@Peter Jordan (Unlicensed) 

ML: The challenge of keeping them aligned

DK: we include valuesets but not coded systems in our IGs

FHIR Questionnaires and SNOMED Demo

@Alejandro Lopez Osornio 

Demo: https://ihtsdo.github.io/sct-implementation-demonstrator/#/questionnaires

URI Resolution Update

@Peter Williams 

Requirement to resolve URIs representing post-coordinated expressions.

See https://chat.fhir.org/#narrow/stream/179166-implementers/topic/The.20URL.20is.20not.20valid.20-.20Snomed-ct.20compositional.20grammar.20.7B.20.7D

ML Suggests resolving to a page that displayed information about that expression - formatted nicely, diagram? - and perhaps further capabilities, but not go as far as attempting to classify in the first instance.

ECL would return implicit valuesets

"Post Coordination is modeling and shouldn't be taken lightly" - DK

Incomplete Expansions

@Peter Jordan (Unlicensed) 

Potential for marking the expansion of certain  valuesets as incomplete because theoretically, all possible Post Coordinated Expressions exist but are not included in the expansion.  See https://build.fhir.org/ig/HL7/fhir-extensions/StructureDefinition-valueset-unclosed.html

https://jira.hl7.org/browse/FHIR-40718

DK - flag excludePostCoordinated to include or not include post coordinated values would affect this.

Snowstorm Development Question

@Peter Jordan (Unlicensed) 

Question on the future direction for Snowstorm

PWI : Potentially looking at horizontal scaling by separating out country instances on number of services.   Considering test suites eg Touchstone.   Also support for use with IG Tooling.  PJ suggests could look at Inferno (wrapper for FHIR Validator).

ALO : Snowstorm-lite would allow for quicker scaling with much reduced memory footprint, with a small number of limitations such as reduced ECL capability and restriction to single Snapshots.  See https://github.com/IHTSDO/snowstorm-lite

Update of the Implementation Course - Terminology Services Module

@Jon Zammit 

Regarding Using SNOMED CT with HL7 Standards

• "Simple Map Reference Sets" link is now dead - replace with one of the directional maps?

• Since we can do this with any maps (eg Complex/Extended), could the language in this section be made more general. 

• "Where [sctid] is a value from the table above" need not be limited to the table above, could be any map reference set SCTID.

• In Snowstorm, check how the correlation id is being translated into the 'equivalence' ICD-10 coming back as 'unmatched' due to the values in the ICD-10 map.   In the case of the historical associations, we could populate the equivalence based on the association type

• How is the CTV3 map working?  In that map the target is a SNOMED CT code.  However, might be better use of our time to remove the CTV3 map rather than get it working.  Note that NHS UK were generating CTV3 codes until 2016 (mapping until 2020) so if SI codes are out of step with UK codes, then that's dangerously misaligned.

• Do we want to include the attribute value reference sets here, as the reason  for inactivation gives an indication of which historical association is going to be used.

• Check the implicit maps returned eg https://browser.ihtsdotools.org/fhir/ConceptMap?_format=json

DK: How do we express equivalence now?  Should that be done with two complementing simple directional maps, or would we use a complex map with a correlation type of equivalence?

Alt Identifier File

@Peter Williams 

@Kai Kewley 

Using the Alternate Identifier File in implicit maps.

No refset is available.  Can we use the scheme id?

On non receipt of a scheme id, we discussed that all 'mapped' values could be returned?

So here for example, is the scheme &url=http://snomed.info/sct?fhir_cm=<scheme sctid> needed? It would cause issues by being looked up as a refset first, and code would have to be ready to fall back to also check scheme ids.

ML suggest we need the URL parameter to indicate that the alternate identifier file should be used.  Is there a URI for the Alternate Identifier File itself?  No.

The scheme id is more specific.  It is redundant because we already have the targetSystem URI.   Could it be used for anything else or confused with anything else if we tried to use it?  

URI suggestion:   http://snomed.info/sct?fhir_alt

This will return all scheme known in the alternate identifier map.   ^ requires that we have the target system URI as an annotation (or additional relationship) on the SNOMED schema concept.

So here fhir_alt follows the HL7 defined pattern as per fhir_cm

The targetSystem would need to mapped to the URI (annotation/additional relationship) which gives the URI for that scheme.  705114005 |LOINC Code System|.   The additional relationship file will need to be parsed for this metadata which is needed for the FHIR server to perform these mappings between URIs and Scheme SCTIDs.

Another Question:  the source and target imply a direction and the alt identifier file can be considered to be an equivalence (Equal in R4)

R4 (also  has Product that could be used to specify the Scheme as a return value, see https://hl7.org/fhir/R4/conceptmap.html

2024-02-20 ML Is FHIR clear itself on how alternate codes should work?  Not a problem at the CodeableConcept level (where multiple codes are expressed), but at the CodeSystem level so how would we see it in a $lookup (presumably as a property).  PJ Subsumption would need to work in the orginal codesystem.   PWI doesn't want to reference things like LOINC codes (NNNNN-N) in the SNOMED CodeSystem because at that point (if we were to add multiple external codeSystems) then we can't tell which alternate CodeSystem it originated in.  DK Languages Group defined ECL solution  for alternate identifiers including the schema eg LOINC#12345-6, see 6.1 Simple Expression Constraints PWI so that does seem like it could exist in the SNOMED CT namespace.  ML So if we looked up that code in SNOMED CodeSystem we'd get back the relevant SNOMED code (with a property of "AlternateIdentifier" as a Coding where the system would be "http://loinc.org", and the code 12345-6).   DK At this point we've removed the need to do a $translate operation at all (or do we support both?)

IG Tooling Compatibility

Requirements for making a Terminology Server compatible with IG Tooling

See Zulip discussion here:  https://chat.fhir.org/#narrow/stream/179202-terminology/topic/Terminology.20server.20requirements.20for.20IG.20builds.2E.2E.2E

2023-07-25

• @michael lawley is implementing the requirements in Ontoserver

• Graeme is preparing the Terminology Server registry, probably a JSON document, that would make available different terminology servers, listing the capabilities, including which code systems on each server

2023-08-22

• @michael lawley the scope has expanded, advancing on the implementation. The registry will document which terminology server meet requirements, it is a centralized resource, maintained by HL7. There is document that describes this. An implementation guide developer would use this information to choose a compatible terminology server as the backend of the IG valuesets. There is a test suite that servers can apply to demonstrate compatibility, Graeme will be the curator of the list, the submission process for a new server is under development.

2023-10-03 ML Ontoserver still a work in progress in this area.

Observables model

@michael lawley 

Question on the Observables Model - was discussed in both the MAG and the EAG.   @Yongsheng Gao currently looking at proposals and next discussion will be at all staff meeting in June.

There was an inconsistent role grouping pattern in the observables, and the question was trying to find the reason for this. This is being evaluated.

@Peter Williams There is a decision to move to the grouped pattern

@Andrew Perry (Unlicensed) looking for opportunities to contribute to the observables and LOINC modeling decisions

2023-09-19 Update:  Direction of travel here suggests moving to have all attributes in a single role group, but preparatory hierarchy movements will be required first - Evaluation Procedures moving to Observables.  Discussion expected at October MAG meeting, followed by a briefing note if no major objections raised.   AP UK have concerns about problems caused with existing counterparts between the two subhierarchies.  UK did respond to RFC last week.

2023-10-03 SI responding to briefing note feedback (including encouragement from UK to engage as a matter of priority).  Discussion expected in both MAG and EAG at October Business Meetings. Two Schools of Thought:  LOINC style single observable hierarchy (order / result) vs. procedure + observable pairs for expressing ordering and results.

Post R5 Concept Map - support for SNOMED Map features.

@michael lawley 

Concept maps and SNOMED Maps  - how to materialize as both R4 and R5.

Metadata needed:  target CodeSystem URI,  source and target ValueSets.

Do we assume that a simple map is an equivalence.

For complex maps, how are the additional fields represented?   Eg in property, dependsOn and product (these are typed).   See 5.2.3.3 Complex and Extended Map from SNOMED CT Reference Sets  

CC @Kai Kewley 

Update 2023-06-13 Now working in confluence Concept Map in R5

2023-07-11 Work being done (ML) around clarity and specification for the parameters used in implicit Concept Maps.

2023-07-25 @michael lawley and Dion had a discussion on metadata and put together a proposal of a JSON based metadata, presumably to be presented to the MAG (clarified 2023-10-03)

2024-02-14 DK Enquired on status here.  KK has coded some support in R4 but uses comma delimited messaging.   ML Using extension to allow properties in R4 (fully available in R5).  See extension:  http://hl7.org/fhir/5.0/StructureDefinition/extension-ConceptMap.group.element.target.property

See example 1 and example 2 shows multiple groups matching.

Demos showing machine processing of ICD-10 rules:

https://github.com/IHTSDO/snomed-to-icd-10-mapper

https://ihtsdo.github.io/iid-icd-maps/

Discussion on whether or not the Terminology Server should have responsibility for applying the values to the rules to fully identify the correct target of the $translate operation. DK would like to see choice for client to calculate the rules.

2024-02-20 DK Sweden looking for FHIR based SNOMED to ICD-10 map.  Vendor had decided against FHIR Concept map and implemented a proprietary solution (although if standard existed, they might have gone that way, and would have been an R5 solution if so)

Expand Alliance

@Marie-Alexandra Lambot 

XpanDH - Expand Alliance European Project.  Interoperability in Europe, targeting providing 6 x 40 hours of medical interoperability training to 40 students to fill gaps in existing offerings.  HL7 Europe (consisting of all HL7 European country affiliates) also participating.   

2023-05-30 Submission made (SNOMED International is a co-partner).  Response expected by the end of the Summer.

2023-10-03 ML News update on course proposals expected for November.  See https://cordis.europa.eu/project/id/101095594

Language behaviour compliance in Snowstorm

@michael lawley 

@Kai Kewley 

BCP-47 requires a dash every 8 digits so SCTIDs need to be broken up to be compliant.

Also the "x-sctlang" is required. 

Also discussed GG's request to @Kai Kewley in the behaviour around language headers and how they interact with the displayLanguage in the HTTP GET request.

R5 changes suggested around specifying language using BCP-47 

@Rob Hausam 

@Daniel Karlsson (Unlicensed) 

See Tuesday meeting item #6

Changes for R5  and check on use of dialect in x form in ECL Specification Appendix C:  Appendix C - Dialect Aliases

Note that where a language reference set pulls from more than one language, the BCP-47  tag could potentially missing out the <lang> element and start directly with "x-".  Options for referring to multiple languages: could be done either with a * (see https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Accept-Language#syntax ) or with a comma separated list to indicate a priority order (although this isn't supported by the FHIR parameter (unless the server went out of it's way to allow for this), or use the accept-language header).  Use case is in the displayLanguage input parameter to a ValueSet $expand operation

2022-08-31 ML CSIRO attempting to get R5 compliant Ontoserver ready in time for HL7 Connectathon.

2022-10-04 With this page moving to terminology.hl7.org it is no longer subject to the ballot process.  Instead a "TSMG" ticket should be created.  @Daniel Karlsson (Unlicensed) and @Rob Hausam will progress this. 

Also tracking R5 changes to concept map - see http://build.fhir.org/valueset-concept-map-relationship.html.   @Peter Williams to check SI intention for R5 compliance.

2022-10-18 DK EHealth Working Group on Semantics - discussion on cross border infrastructure based on CDA.  Suggestion that it could move to FHIR for new specifications/developments eg Lab Reports.

2022-11-02 RH Rob will check the status and how we can help it move

2023-02-07 RH R5 Ballot is complete.   Deadline to complete resolutions is Feb 24th.   Expectation to publish late April / early May.