Stream: genomics
Topic: How to represent the test identifier
Bob Dolin (Jul 30 2021 at 18:08):
Is there a way in the current IG, to represent the vendor-specific test identifier(s) used?
Jamie Jones (Jul 30 2021 at 18:29):
ServiceRequest can have 0..* Identifiers, but we don't have a recommendation to put those on results or reports directly
Bob Dolin (Jul 30 2021 at 18:30):
Right, but isn't that the request? I don't think the request would contain the actual test identifier.
Kevin Power (Jul 30 2021 at 18:31):
@Bob Dolin - Do you mean an 'order catalog' identifier, like a code to represent what is being ordered? Or something that identifies a specific order for a patient?
Jamie Jones (Jul 30 2021 at 18:31):
If it is a report level ID, could put it on the DR. Or follow emerge and use InstantiatesCanonical
Bob Dolin (Jul 30 2021 at 18:41):
I think there are a lot of reports that summarize the findings from > 1 test. I'm finding a lot of tests really bundle up complex methodologies into the test ID, so it's important to track. Would this be a good item for a jira ticket?
Bob Dolin (Jul 30 2021 at 18:42):
@Kevin Power Not so much what is being ordered, but an identifier of the test that was used to actually perform the sequencing.
Jamie Jones (Jul 30 2021 at 18:42):
In that case, Observation.method can have multiple Codings :)
Bret H (Oct 11 2021 at 21:36):
look at GTR https://www.ncbi.nlm.nih.gov/gtr/ @Bob Dolin for codes for some frequently used tests
Bret H (Oct 11 2021 at 21:38):
here's sort of an example: Overall methodology may be reported using hla-genotyping-results-method. Here a codeable concept may be used to refer to method entered into a public registry, such as the NCBI Genetic Test Registry, or a local private registry.
Structure of hla-genotyping-results-method
Name Flags Card. Type Description & Constraintsdoco
.. method 0..1 CodeableConcept URL = http://hl7.org/fhir/StructureDefinition/hla-genotyping-results-method
method: The platform, methodology and software applied at the time of the genotyping.
Use on Element ID DiagnosticReport
doco Documentation for this format
Example of hla-genotyping-results-method to report a lab test registered in the NCBI Genetic Test Registry
<extension uri="http://hl7.org/fhir/StructureDefinition/hla-genotyping-results-method">
<valueCodeableConcept>
<coding>
<system value="http://www.ncbi.nlm.nih.gov/gtr/"/>
<code value="GTR000000000.0"/>
</coding>
<text value=”NGS based HLA-A genotyping”
</valueCodeableConcept>
</extension>
Bret H (Oct 11 2021 at 21:39):
from https://www.hl7.org/fhir/genomics.html recycling old things
Bret H (Oct 11 2021 at 21:40):
or use observer
Bob Dolin (Oct 11 2021 at 21:41):
@Bret H I don't think coded methods are sufficient. I think we need a way to represent the specific pipeline / configuration of the test used to generate the observation
Bret H (Oct 11 2021 at 21:41):
Use the observer - it is for organizations
Bret H (Oct 11 2021 at 21:41):
Now you're getting into granular details of representing the test methodology
Bob Dolin (Oct 11 2021 at 21:42):
It would be helpful to be able to explicitly reference the performers test identifier
Bret H (Oct 11 2021 at 21:42):
certainly more of a ObsersvationDefinition
Bob Dolin (Oct 11 2021 at 21:42):
The challenge is that there can be a lot of details about the test that provide important context for interpreting the observations
Bret H (Oct 11 2021 at 21:43):
The lab interprets the observations
Bret H (Oct 11 2021 at 21:43):
or do you not trust the lab? just being cheeky
Bret H (Oct 11 2021 at 21:43):
Who needs the intricate details of the test methodology? whose the target auidence?
Bob Dolin (Oct 11 2021 at 21:43):
Let's say NGS is performed, and I don't see any variants in geneX. Does that mean there aren't any?
Bret H (Oct 11 2021 at 21:44):
That's region studied
Bob Dolin (Oct 11 2021 at 21:44):
Let's say NGS is performed, and the lab reports SNVs, MNVs, and Indels, but doesn't report any CNVs. Does that mean there aren't any?
Bob Dolin (Oct 11 2021 at 21:45):
Yes, region studied is one of the test details that we've culled out as being important. But there are others.
Bret H (Oct 11 2021 at 21:45):
In the current state of the art, the test is ordered from a trusted vendor to produce a specific result. It's known ahead of time.
Bret H (Oct 11 2021 at 21:45):
But with broader NGS
Bret H (Oct 11 2021 at 21:45):
I totally see how that knowledge can't scale
Bret H (Oct 11 2021 at 21:46):
Have you checked out the kind of information in GTR?
Kevin Power (Oct 11 2021 at 21:47):
Is this a solved problem anywhere (outside of HL7 / FHIR)?
Bret H (Oct 11 2021 at 21:47):
Oh here, I'd say it is specific to genomics as there are an unlimited number of pipelines that could be used
Bret H (Oct 11 2021 at 21:48):
Only so many contrast solutions to use in radiology. And in Immunoflouresence you don't get revelation time.
Bret H (Oct 11 2021 at 21:49):
but it is the kind of information that is definitional. Perhaps extending ObersvationDefinition?
Kevin Power (Oct 11 2021 at 21:49):
Sorry, I meant in the genomics space - but are the details you want represented somewhere, and all we have to do is translate them to FHIR?
Bob Dolin (Oct 11 2021 at 21:49):
Chamala's article talks about how they manage their specific bioinformatics pipeline.
Bob Dolin (Oct 11 2021 at 21:49):
https://www.thieme-connect.com/products/ejournals/abstract/10.1055/s-0040-1721480
Bret H (Oct 11 2021 at 21:51):
have you looked at the Genetic Test Registry? (: ^ )
Bob Dolin (Oct 11 2021 at 21:51):
I don't know of any way that two parties can generate the same VCF from a given FASTQ, unless they are using the same pipeline.
Bob Dolin (Oct 11 2021 at 21:51):
Yes, GTR is very nice, and they have a good set of methodology codes, but GTR is not explicit regarding the pipeline / configurations
Bret H (Oct 11 2021 at 21:51):
here's observation definition that could be extended with pipeline informaiton: https://www.hl7.org/fhir/observationdefinition.html
Bret H (Oct 11 2021 at 21:52):
So perhaps, GTR plus ObservationDefinition - OR talk to GTR about adding the missing data elements? They are a national database specializing in genetic test information
Kevin Power (Oct 11 2021 at 21:52):
This feels like something that is in the sports car version of our IG, and we are about to release the moped :wink:
Bret H (Oct 11 2021 at 21:53):
The hardest part won't be adding it or modeling it. It will be adoption
Bret H (Oct 11 2021 at 21:53):
that's why I'm looking at GTR which already has some test vendors giving groovy details
Bob Dolin (Oct 11 2021 at 21:54):
Many labs will now provide test compendiums - what I'm suggesting is that it would be nice to somehow be able to link an observation to FHIRized test from a given compendium
Bret H (Oct 11 2021 at 21:54):
If they put a description of thier test to be accessible through a URI that would be one way
Kevin Power (Oct 11 2021 at 21:55):
Can you refer to a good example (or 5) @Bob Dolin ?
Bob Dolin (Oct 11 2021 at 21:55):
I'm not sure if this is ObservationDefinition or some other resource, @Bret H ?
Kevin Power (Oct 11 2021 at 21:55):
eMERGE used PlanDefinition
Bret H (Oct 11 2021 at 21:55):
Obs def gives you a way to essentially create a LOINC code and it is in FHIR and can be extended.
Kevin Power (Oct 11 2021 at 21:55):
https://emerge-fhir-spec.readthedocs.io/en/latest/artifacts/plan_definition.html
Bret H (Oct 11 2021 at 21:56):
@Bob Dolin that brings up a good point. Does this need to be machine computable or simply delivered for human readability and access?
Jamie Jones (Oct 11 2021 at 21:56):
eMERGE test description was largely a way to list textual steps taken and disclaimers made. Nothing encoded to enable committing answers to Bob's questions
Bret H (Oct 11 2021 at 21:57):
again, the obs def can be exposed through a FHIR server and referenced at need (rather than sending the same information in a large package more than once)
Bob Dolin (Oct 11 2021 at 21:58):
Thanks @Kevin Power for the link. I have to run off to a meeting, but will try to post some scenarios / examples later today. @Bret H I think there are some aspects of the method that are ideally computable (e.g. region studied).
Bret H (Oct 11 2021 at 21:58):
whatever is done, try to recycle the information it could blow up the data usage. And these things are liable to change, so a permanent link to the version that was used seems intriguing
Bret H (Oct 11 2021 at 21:59):
I don't think it will be efficient to send the whole instance of data for the method eveyrtime a test is reported - but on the business end of the recipient they could do the cleanup and keep one copy of each version....this starts to get quite specialized
Bret H (Oct 11 2021 at 22:00):
too bad general lab does not already have this precedent :sad:
Kevin Power (Oct 11 2021 at 22:00):
I think the key question is --- do we expect it to be 100% computable (someone could recreate the VCF exactly) or 5% computable (a set of strings that would help a human see what was done) or something in between?
Bret H (Oct 11 2021 at 22:03):
I think the obs def is better suited than the care plan. see https://www.hl7.org/fhir/observationdefinition.html#bnc
Jamie Jones (Oct 11 2021 at 22:04):
Semantically it is close, I agree.
Kevin Power (Oct 11 2021 at 22:06):
Yea, but I struggle to see more that a few attributes we would ever use on ObsDef to define what we are trying to define. It seems like they are trying to keep PlanDef more generic than just a Care Plan:
This resource allows for the definition of various types of plans as a sharable, consumable, and executable artifact. The resource is general enough to support the description of a broad range of clinical and non-clinical artifacts such as clinical decision support rules, order sets, protocols, and drug quality specifications.
Arthur Hermann (Oct 11 2021 at 22:10):
I concur with the question about use case. @Bob Dolin - I think you are trying to solve for something that is as Kevin said - the race car version of our IG. Nothing wrong with that, as long as this is clearly stated and understood. I also feel that it is important to be sure we focus on the "moped" version of the IG first. Bob - am I correct in thinking that your desire for this information comes from your VCF translator and and/or development of a GACS? If so - those are excellent use cases - but not the current focus of our IG (at least as I understand it). Unless there is a clear way to "point to this data" - e.g. through an endpoint or find a set of resuable data hosted somewhere that could be utilized. I have other thoughts, but I think it would be best to have a discussion about this (outside of a normal meeting ) to brainstorm.
Bret H (Oct 11 2021 at 22:12):
Well obs def says "An instance of ObservationDefinition represents a set of constraints applicable to the value produced by observations of this kind. This set of constraints may have two utilities: 1) It guides the entry of the result of an observation of the kind defined. 2) It helps the consumer of the results of observations of that kind, to interpret properly these results."
Bret H (Oct 11 2021 at 22:13):
we'd have to profile it to add things like region studied
Bret H (Oct 11 2021 at 22:14):
Might be nice to try something with extensions to get the feel for the data provided. Shame not to use GTR though
Bret H (Oct 11 2021 at 22:15):
when I last worked with them 5+ years they were very, very amenable to incorporating new data elements.
Bret H (Oct 11 2021 at 22:15):
(that was at least one GTR director ago, though)
Bob Dolin (Oct 11 2021 at 22:25):
Kevin Power said:
I think the key question is --- do we expect it to be 100% computable (someone could recreate the VCF exactly) or 5% computable (a set of strings that would help a human see what was done) or something in between?
5% would be good - so far, I've only come across a couple aspects of the test that I find computationally useful.
Bob Dolin (Oct 11 2021 at 22:26):
from a test compendium perspective: region studied; what was the (coded) methodology (e.g. NGS vs. FISH); what types of variants is the test / pipeline designed to detect.
Bob Dolin (Oct 11 2021 at 22:26):
from a test instance perspective: which studied regions were uncallable
Bob Dolin (Oct 11 2021 at 22:51):
and from an observation perspective: ability to reference the test that generated the observation
Bob Dolin (Oct 18 2021 at 18:14):
@Jamie Jones On the PGx page (http://build.fhir.org/ig/HL7/genomics-reporting/pharmacogenomics.html), Figure 1 shows the report body containing a 'Text Performed Methodology', represented as a PlanDefinition. If the report contained observations from more than one test, how would you link a region-studied or variant observation to the right one?
Kevin Power (Oct 18 2021 at 18:32):
Since PlanDefinition came from eMERGE, I am tagging @Mullai Murugan for input. My short answer - I don't see a way. But, technically using PlanDef isn't part of the base spec, eMERGE did it via their own extension. Text from the IG:
In this report example, the text which explains the methodology of the test, was modeled using the PlanDefinition resource (#5 in the schema). The clinical genomics work group has not yet determined the value of providing a profile on PlanDefinition, and requests community feedback on encoding testing methodology details.
Bret H (Nov 09 2021 at 23:43):
I'd link the region studied (which is a profile of Observation) to the Variant observations that are relevant as one way
Bret H (Nov 10 2021 at 16:45):
FYI: https://build.fhir.org/ig/HL7/fhir-order-catalog/index.html
Bob Dolin (Nov 10 2021 at 18:00):
Thanks @Bret H . It looks like they, like eMerge, use the PlanDefinition for the lab order compendium. For genetics, we'd potentially want to figure out: [1] what are the priority structured data elements to capture about a test type; [2] what are the priority structured data elements to capture about a test instance; [3] how to indicate that a (variant) observation was generated from a particular test.
Bob Dolin (Nov 10 2021 at 18:01):
On top of that, as you mention, do a revision of region studied along the way, since the region studied observations should potentially also linkable back to a test instance.
Bob Dolin (Nov 10 2021 at 18:01):
Maybe after we get the current release published, we should put in a Jira on this?
Bret H (Nov 10 2021 at 22:41):
Yep.
Last updated: Apr 12 2022 at 19:14 UTC
