Stream: genomics
Topic: Stress Testing IG Implications with Examples
Jamie Jones (May 05 2020 at 15:59):
We need to fully define the concepts mentioned in our Implication profiles, and are openly calling for example statements from reports needing to model with these structures. Let's gather our requirements here and move forward.
Jamie Jones (May 05 2020 at 16:01):
We are semantically aligning both somatic and germline use cases here
Arthur Hermann (May 05 2020 at 16:21):
I completely agree with you @Jamie Jones - I am not very technical, but I do have access to clinical folks who use the information we are developing and really want to work both with the labs and those clinical folks to make sure we have examples that are clear and resonate with both of them. @Rachel Kutner - I think this is what you and Epic also want... if not, then we can clear that up! @Bret H I believe this is also what you are looking for.
Today's call, where we discussed: "In the Genomics FHIR IG the fields of associated cancer and associated phenotype components do not have clear guidance on how to distinguish 'risk of' versus 'context.' Perhaps, the use of the component in therapeutic implication profile indicates 'context,' whereas the use of the component in diagnostic implication profile indicates 'risk of?'
If necessary, maybe a flag would be appropriate? or some other modifier element could be added?" is a good example
Kevin Power (May 05 2020 at 16:48):
To tie back to V2, here are the attributes related to this topic and how they are defined (NOTE - Not a surprise, but no mention of cancer specific terms):
http://www.hl7.org/documentcenter/public/standards/dstu/V251_IG_LRI_R1_STU3_2018JUN.pdf
Report Section 1 -Variables That Apply To The Overall Study
51967-8 / Genetic disease assessed
Coding systems:
- SCT (SNOMED-CT)
- I9CDX
- I10C
- MedGen-Dis
- HPO (Human Phenotype Ontology)
Applies only to studies that target a disease. While this can be supplied by either the placer or the test performer, this question is typically answered by the placer. Any or allof the above coding systems could be used in the message.It will be up to the message generator to specify the coding system within the message. We encourage the use of SNOMED CT in this field because it is the preferred direction in the US, which is in the example values OBX-5 column.However, the LHC-forms demo of this draft specification shows the content from NCBI MedGen, because it is the most complete with respect to genetic diseases, and public. Further, MedGen includes mappings to SNOMED CT when available.
Interpretations
81259-4 / Probable Associated Phenotype
Coding systems:
- SCT (SNOMED-CT)
- I9CDX
- I10C
- MedGen-Dis
- HPO (Human Phenotype Ontology)
The disorder with which this variant is associated. Allows same coding systems as for disease assessed. The message implementer inserts the approved coding system in CWE.3.See descriptions of the coding systems in Table A.1 of the Appendix.
Arthur Hermann (May 05 2020 at 16:49):
Thanks @Kevin Power - very helpful!!
Arthur Hermann (May 05 2020 at 16:51):
@Rachel Kutner - the hardest part of this will be scheduling a meeting (I suggest an hour) with an agenda that we agree upon prior to get this moving... can you take lead on this? Or how can I help?
Jamie Jones (May 05 2020 at 16:54):
Open question1: is a cancer-specific field necessary?
Kevin Power (May 05 2020 at 16:55):
In FHIR, we have little to no definition on the components themselves. But, we do have other guidance.
This describes the 'Inherited Disease' use case:
http://build.fhir.org/ig/HL7/genomics-reporting/general.html#genomic-implications
It includes the following statement:
For inherited diseases, a diagnostic implication indicates the likelihood of inheritance of a particular disease (the associated-phenotype)
The page for PGx:
http://build.fhir.org/ig/HL7/genomics-reporting/pharmacogenomics.html
It includes this section:
Communicating disease or tumor context
In general, treatments and medications have a set of indications for use which are not explicitly stated. However, if a pharmacogenomics statement has a specific context of a disease (indication) or a tumor indication use the components "associated phenotype" or "associated cancer" respectively.
To explicitly link medication or treatment implication to oncology, for example, populate the associated-cancer component. E.g. If the associated-cancer component is populated, then effect-medication-efficacy pertains to the medication's efficacy on treating the cancer. For oncology related reporting in general see Somatic Reporting.
Somatic testing page is here:
http://build.fhir.org/ig/HL7/genomics-reporting/somatics.html
It includes the following guidance:
Diagnostic Implication indicates an association between the identified variant (linked via Observation.derivedFrom) and a particular type of cancer. The evidence-level and clinical-significance components can convey the evidence-based categorization of the variant for the given cancer type per the AMP/ASCO/CAP guidelines. The prognosis component can also be populated to convey a general indication on the overall outcome for the cancer patient, as described in a linked relatedArtifact.
Therapeutic implication indicates an association between the identified variant (linked via Observation.derivedFrom) and the efficacy of a particular medication or therapy for the specified cancer type. Because this implementation guide is intended for international use, it does not mandate the use of any particular code systems for medications, therapies, or cancer descriptions. Implementations should use the code systems most typically used in their jurisdictions or that are mandated by national FHIR profiles. The associated cancer component can be used to communicate that the variant has repercussions on treatments for cancer. If specimen has specimen.type of ‘tumor’ then the implication and variant are specific to treating the tumor.
Liz Amos (May 05 2020 at 17:57):
Jamie Jones said:
Open question1: is a cancer-specific field necessary?
To make your question a little more specific, could we"remove 'associated-cancer' from the Diagnostic Implication" profile ? Is this basically the question?
Kevin Power (May 05 2020 at 18:42):
Liz Amos said:
Jamie Jones said:
Open question1: is a cancer-specific field necessary?
To make your question a little more specific, could we"remove 'associated-cancer' from the Diagnostic Implication" profile ? Is this basically the question?
I do like the idea of considering this differently for the different profiles (not that we will reach different conclusions, but maybe we will? Regarding Diagnostic Implication, the two component approach seems to make it easier for downstream consumers of the data to more easily distinguish between implications related to inherited disease or cancer. There are other ways to do that, but they will not be as straightforward. Having said that, I don't know for sure that is a real requirement?
Bret H (May 05 2020 at 19:09):
just an FYI. I'll be contacting folks by email later this evening. Scope: Examples of use of associated phenotype and associated cancer in reports. Including, example PDF of deidentified report, statement from report that is relevant, FHIR representation of the statement. First pass is limited to statements about the variant where the phenotype information is either contextually related to the variant result or about the variants contribution to risk (or 'probable diagnosis').
Bret H (May 05 2020 at 19:09):
...ps if the therapeutic implication profile information is unclear - let me know. in full disclosure, I authored the words in the current draft. I'm either biased or to blame : ^ )
Liz Amos (May 05 2020 at 19:20):
Kevin Power said:
Liz Amos said:
Jamie Jones said:
Open question1: is a cancer-specific field necessary?
To make your question a little more specific, could we"remove 'associated-cancer' from the Diagnostic Implication" profile ? Is this basically the question?
I do like the idea of considering this differently for the different profiles (not that we will reach different conclusions, but maybe we will? Regarding Diagnostic Implication, the two component approach seems to make it easier for downstream consumers of the data to more easily distinguish between implications related to inherited disease or cancer. There are other ways to do that, but they will not be as straightforward. Having said that, I don't know for sure that is a real requirement?
I just want to be concrete in the ask/action - I'm not finding where else "associated-cancer" shows up in the IG. I think it was just a leftover from when Jamie had consolidated the implication profiles (i.e. remnants from Somatic Diagnostic Implication).
Kevin Power (May 05 2020 at 19:26):
It is included directly by name and generically in some of the text I included above from the IG. For example, in the DiagnosticImplication side on the Somatics page:
Diagnostic Implication indicates an association between the identified variant (linked via Observation.derivedFrom) and a particular type of cancer.
The particular type of cancer would refer to the 'associated-cancer' component.
On the TheraputicImplication side, it is directly referenced here:
Communicating disease or tumor context
In general, treatments and medications have a set of indications for use which are not explicitly stated. However, if a pharmacogenomics statement has a specific context of a disease (indication) or a tumor indication use the components "associated phenotype" or "associated cancer" respectively.
To explicitly link medication or treatment implication to oncology, for example, populate the associated-cancer component. E.g. If the associated-cancer component is populated, then effect-medication-efficacy pertains to the medication's efficacy on treating the cancer. For oncology related reporting in general see Somatic Reporting.
Jamie Jones (May 05 2020 at 19:35):
It was previously one concept used in each somatic use case: diagnosis, prognosis, predictive (what we'd now call therapeutic)
Kevin Power (May 05 2020 at 20:00):
Did we still call it 'associated-cancer' before the split?
Jamie Jones (May 05 2020 at 20:16):
Believe so, it was a required component in the abstract somatic-implication profile
Arthur Hermann (May 05 2020 at 22:11):
Bret H said:
just an FYI. I'll be contacting folks by email later this evening. Scope: Examples of use of associated phenotype and associated cancer in reports. Including, example PDF of deidentified report, statement from report that is relevant, FHIR representation of the statement. First pass is limited to statements about the variant where the phenotype information is either contextually related to the variant result or about the variants contribution to risk (or 'probable diagnosis').
Brett - are you able to include me in this email?
Kevin Power (May 05 2020 at 22:13):
Oh yea - it was in the SomaticImplication profile (the base for Diagnostic, Prognostic, Predictive):
http://hl7.org/fhir/uv/genomics%2Dreporting/STU1/somatic-implication.html
Bret H (May 06 2020 at 03:23):
(deleted)
Bret H (May 06 2020 at 04:17):
email for example tiger team sent. The email is From: Bret H <bheale@gmail.com> sent Date: Tue, May 5, 2020 at 10:13 PM
Subject: hit example squad associated phenotype/cancer. Let me know if you did not get it and would like to volunteer to build examples too.
Kevin Power (May 07 2020 at 20:25):
Just so we don't lose momentum on this, wanted to see if we made any progress?
@Arthur Hermann / @Rachel Kutner / @Bret H
Rachel Kutner (May 07 2020 at 20:27):
I haven't had a chance to look at this so far this week, due to other engagements. I'll be checking in on it tomorrow, but not sure how much time I'll be able to spend coordinating. Once I've read through everything I'll respond in more detail.
Arthur Hermann (May 08 2020 at 22:06):
@Kevin Power @Bret H created an excellent letter which he is sending through InterMountain. I will do at KP as welll.... then we need to come back together with the information we gain along with @Rachel Kutner to be clear we all agree on final resolution :grinning_face_with_smiling_eyes:
Rachel Kutner (May 08 2020 at 22:16):
Sounds great! Thanks so much Arthur and Bret for spearheading this, it sounds like there's a lot of interest in it. I'm working on getting up to speed on all the messages.
I was going to suggest a quick call (15-20 min?) early next week with @Bret H, @Arthur Hermann , and myself to get on the same page and determine if there are any loose ends or gaps to fill. @Arthur Hermann what do you think?
Arthur Hermann (May 08 2020 at 22:22):
@Rachel Kutner that would be fine with me if we feel we need to.. @Bret H - what do you think?
Kevin Power (May 08 2020 at 22:24):
We could have this be a topic on the Monday FHIR subgroup call?
Rachel Kutner (May 08 2020 at 22:27):
@Bret H that could be good. It wouldn't need to be much time, but an update might be good for the group.
WIll we want labs to review once the pieces are modeled?
Kevin Power (May 08 2020 at 22:48):
We always welcome lab review, but getting attendance from lab personnel with the right expertise is a constant struggle. If you can bring some to the group, please do :slight_smile:
Bret H (May 10 2020 at 13:56):
@Kevin Power the work of making examples is best done outside of a committee. The results would be better presented. I have production work to do, but can say a few words Monday. I don't thin it really worth mentioning other than it is ongoing. We started on Wednesday last week. It would be a mistake to review currently.
Bret H (May 10 2020 at 13:57):
Right now, I am asking folks to simply find the statements to be modeled.
Bret H (May 10 2020 at 13:58):
No example statements have been added: https://docs.google.com/spreadsheets/d/1j6YfGPsBht63Dm_KFHMnCLzom1XzFkT7mnMTagiEc-s/edit#gid=0
Bret H (May 10 2020 at 13:58):
@Rachel Kutner @Arthur Hermann @May Terry please document statements you've found: https://docs.google.com/spreadsheets/d/1j6YfGPsBht63Dm_KFHMnCLzom1XzFkT7mnMTagiEc-s/edit#gid=0
Kevin Power (May 10 2020 at 14:05):
Fair enough.
Arthur Hermann (May 11 2020 at 19:22):
@Kevin Power I may be able to bring a lab person to specific sessions if we know we are focusing on issues for which we need their input. I know this can be hard to do, since topics are fluid, but let me know..
Kevin Power (May 11 2020 at 19:32):
If they can weigh in on the work you are doing now, that would be very helpful.
Arthur Hermann (May 12 2020 at 17:49):
@May Terry thank you for all of the wonderful information you added to the spreadsheet!!!! Great helpful info!
May Terry (May 12 2020 at 17:50):
Thanks, although, I wasn't the one who added all that info. Perhaps it was Rachel or Liz?
Arthur Hermann (May 12 2020 at 21:24):
LOL - indeed it was Liz - thank you Liz . :ok: .. but thank you May for everything you do as well :grinning_face_with_smiling_eyes:
Rachel Kutner (May 13 2020 at 15:13):
I got a response from our contact at Tempus. I'll add the PDFs she sent to the drive folder and the docs sheet soon. She offered to also send the HL7 message and JSON formats of the reports, which I'll also add to the drive once I have them.
Here's her message:
Hi Rachel,
I have some fake patients/sample PDFs to start with from our clinical science team. These examples were created a while ago so it may take me a bit of time to track down or replicate the discrete JSON/HL7 data. In the meantime, take a look at these examples and I'll send you discrete data once I have it.
- Prostate cancer case with abiraterone resistance and a diagnostic TMPRSS2-ERG fusion
- Sarcoma case with a diagnostic EWSR1-ERG fusion:
- Non-small cell lung cancer case with resistance to EGFR inhibitors:
- Colorectal cancer case with resistance to EGFR monoclonal antibodies and a CHEK2 variant associated with increased risk
Let us know if you have any questions!
Hopefully these are new examples.
May Terry (May 13 2020 at 15:15):
Thanks Rachel! It's also a good list since it'll also stress-test how to represent fusion genes.
Arthur Hermann (May 13 2020 at 15:18):
I received a Strata report - marked up by our Oncologists - noting the specific areas they are interested in. I will add to the spreadsheet right now..
Kevin Power (May 13 2020 at 22:25):
@Arthur Hermann -- I took a quick look at the Strata report you uploaded to the Examples folder, but it is pretty blurry? What is the title of the box with the red #2 on it? I think it says "Associated Standard Care Therapy" but I am not sure? The box in the example is empty, but what would they normally deliver in that box? Recommended therapies I assume?
Arthur Hermann (May 13 2020 at 23:44):
@Kevin Power i I am sorry it is so blurry.... they sent me an image pasted into a an email which I then used some tools to make more legible... I will see if I can do any better., Yes that second box is where a recommended therapy would be shown (in this case none)
Kevin Power (May 14 2020 at 02:49):
Thanks @Arthur Hermann
Jamie Jones (May 15 2020 at 20:39):
http://hapi.fhir.org/baseR4/Observation/?component-code-value-concept=http://loinc.org|81252-9$https://www.ncbi.nlm.nih.gov/clinvar|653974&_revinclude=Observation:derived-from shows both a diagnostic and therapeutic implication for a clinvar variant. Attempting to capture the usage of 'associated-phenotype' with text:
https://www.ncbi.nlm.nih.gov/clinvar/variation/653974 suggests that the patient has a likely pathogenic variant and is at risk for Malignant Hyperthermia (see https://www.ncbi.nlm.nih.gov/medgen/?term=CACNA1S[gene]).
vs
https://cpicpgx.org/guidelines/ suggests that the patient has a variant that is high risk with HALOGENATED VOLATILE ANESTHETICS for Malignant Hyperthermia (see https://cpicpgx.org/guidelines/cpic-guideline-for-ryr1-and-cacna1s/).
Jamie Jones (May 15 2020 at 20:42):
We need to examine more statements here, I'm not a huge fan of the semantic structure on the therapeutic case. I think this "mad lib" style exercise is great and I think we should strive for the sentence structure to be consistent within each profile regardless of which components are filled out. Likely will need some tweaking.
Kevin Power (May 15 2020 at 21:03):
So for diagnostic-implication, your formula is:
[variant] suggests that the patient has a [clinical-significance] variant and is at risk for [associated-phenotype] (see [RelatedArtifact])
Bret H (May 15 2020 at 21:05):
what does it say in ig for pgx in build? @Kevin Power
Jamie Jones (May 15 2020 at 21:06):
first element in both is technically another relatedartifact (but with type 'derived from'). the variant works there though
Kevin Power (May 15 2020 at 21:07):
Bret H said:
what does it say in ig for pgx in build? Kevin Power
We aren't talking about Therapeutic yet @Bret H ? But what does it say? :slight_smile:
Jamie Jones (May 15 2020 at 21:09):
"The component Effect Medication High Risk indicates if the associated genetic findings pose a particular risk for the patient independent of metabolism or efficacy. i.e. Does the medication have an unusual (and potentially dangerous) effect on patients with these genetic characteristics. For example, in the case of a known adverse event, such as Stevens-Johnson syndrome (SJS), the high-risk component would be used with the associated-phenotype to communicate the risk of SJS"
Kevin Power (May 15 2020 at 21:12):
It also says this:
In general, treatments and medications have a set of indications for use which are not explicitly stated. However, if a pharmacogenomics statement has a specific context of a disease (indication) or a tumor indication use the components "associated phenotype" or "associated cancer" respectively.
Jamie Jones (May 15 2020 at 21:13):
so we are suggesting presence of 'high-risk' changes interpretation of 'associated-phenotype'
Kevin Power (May 15 2020 at 21:14):
Yup, was just typing that very sentiment :slight_smile:
Kevin Power (May 15 2020 at 21:14):
Seems risky if you pardon the pun
Jamie Jones (May 15 2020 at 21:17):
I don't like that pattern... does 'adverse-event-phenotype' merit its own component?
Kevin Power (May 15 2020 at 21:19):
I wasn't sure what to call it, but I think we do need a new component. Even wonder if we should more clearly indicate the various sorts of contexts we are talking about. Up until seeing more of these examples, I have sort of liked the generic components, but once again, they led to confusion and delay.
Kevin Power (May 15 2020 at 21:26):
On Diagnostic, should we have 'risk-of-(disease/diagnosis/condition)' and it could include all sorts of things (not break out between phenotype versus cancer.
On Therapeutic, we could have 'therapuetic-context' (what would this treatment be for) and 'adverse-event-context' (what could happen because of this therapy).
Holes in that I am sure (it is late on a Friday afternoon), but that feels like the right direction to me.
Jamie Jones (May 15 2020 at 21:28):
therapeutic suggestion i see the value in. for diagnostic, a more general concept seems good--we could note that it covers the current loinc as well
Kevin Power (May 15 2020 at 21:29):
Actually, maybe keeping the 'associated-*' on Diagnostic is OK?
Kevin Power (May 21 2020 at 20:36):
@Bret H @Rachel Kutner @Arthur Hermann -- Any updates on this initiative?
Arthur Hermann (May 21 2020 at 20:59):
You beat me to it @Kevin Power ! I was just going to check with @Rachel Kutner and @Bret H about this. We have our spreadsheet updated (I have added the input from our Kaiser Oncologists and added a single Strata report which they like very much. So - Rachel and Bret, what are our next steps to complete this process?
Rachel Kutner (May 22 2020 at 13:55):
I still need to upload the example reports I have to the spreadsheet. I'm not the most well-versed with FHIR, but it sounded like the next part we wanted to do was to try and model the pertinent pieces of each report using the different methods and get feedback on how well it represents the information in the report.
Jamie Jones (May 22 2020 at 14:13):
If we can get folks to select a short list of individual statements of interest I would have no problem doing a lift to FHIR, which we can then comment on more broadly (I think we'll see there will be some room for decision-making in the encoding)
Jamie Jones (May 22 2020 at 14:17):
After I've posted a couple in FHIR and patterns for the encoding are more obvious, folks should be able to easily edit them "Mad Lib" style and we can see if we run into problems from there
Kevin Power (May 22 2020 at 15:40):
@Jamie Jones - I wonder if some of the things you did to extract attributes from our profiles could be repurposed to create a "Mad Lib" like form/spreadsheet for anyone to do an initial mapping of these examples?
Rachel Kutner (May 26 2020 at 16:02):
I uploaded the labs I received from Tempus - they were 4 versions of similar sample reports, but had a few interesting pieces that seemed relevant, which I added to the spreadsheet.
Arthur Hermann (May 26 2020 at 16:07):
I would like to have a very clear statement about what the problem is,what needs to be fixed and why - before we move any further. @Bret H @Rachel Kutner @Kevin Power @Jamie Jones . Rachel - you originally brought this issue up. We now have a wealth of information both from current oncology genomic reports added by many gracious people and from the Kaiser side - some comments as to what the oncologists want to see in their report (plus a very blurry Strata report as an example. @Rachel Evans - can you please provide an answer to my questions above? Bret H - would also like your input and of course Kevin and Jaime - feel free to weigh in as well... Thank you! Arthur
Rachel Kutner (May 26 2020 at 18:20):
@Arthur Hermann I'm not sure I'm the best to clarify the question. I think @Bret H might have a better handle on it. I thought it was to answer the question of whether there is enough context within the diagnostic/therapeutic components to not need to explicitly flag an associated phenotype/cancer as "therapeutic" context vs "diagnostic" context.
I also thought there may have been some piece about determining whether associated cancer should be combined with associated phenotype and if the components provide sufficient context to represent that accurately.
I remember asking about how we expect molecular-based diagnostic results to be represented and thought the answer was they would be sent in a separate part of the FHIR message, not in the Observation (e.g. as a Condition), though I'm a bit fuzzy on those details.
Arthur Hermann (May 26 2020 at 19:01):
Thank you @Rachel Kutner ! Much of that resonates with me as to the original set of questions. I too would like to get @Bret H input and @Kevin Power input if he wishes to chime in. I think it is very helpful for us to have a clear problem statement. I believe the information we have gathered should allows us to answer those questions, even if in some cases, that answer is that the working group needs to do additional work in this area.
Kevin Power (May 26 2020 at 21:42):
I say what @Rachel Kutner outlined is generally right. Her first two are really corrections/clarifications to our existing IG. I think the third statement is good to evaluate, but see it as a new concept to our IG, so personally I see it as a lower priority.
Bob Dolin (May 26 2020 at 21:48):
I thought another question was whether we need to differentiate between variants relevant to existing conditions/phenotypes that a patient has vs. variants that are associated with conditions through a knowledgebase that the patient may or may not have.
Kevin Power (May 27 2020 at 00:04):
Agree @Bob Dolin - Relates back to the question of 'do our phenotype and cancer components represent the context well enough or not'
Arthur Hermann (May 27 2020 at 01:04):
@Kevin Power @Bob Dolin @Rachel Kutner @Bret H - I agree with Kevin's validation (and answer to Bob D) about the question we are trying to answer. So I will formalize it again: We are trying to validate that there is sufficient context within the IG's current diagnostic/therapeutic components to represent an associated phenotype/cancer as "therapeutic", without needing to explicitly flag it as being in the "therapeutic" context vs "diagnostic" context. If the answer is no, then we need to address this in some way to allow for this specificity. Rachel - I agree with Kevin that the issue re: molecular-based diagnostic reports is a separate question, and that unless there is an urgent need, that should become follow-on work that that Working Group will deal with later. However the first issue does need to be clarified now, based on the input of the Oncologists I have been in contact with. Thanks to all!
Bret H (Jun 08 2020 at 13:24):
@Bob Dolin @Kevin Power @Jamie Jones the reason for study is where one would send the lab a diagnosis or problem of interest that the patient may have. Our implications can or might not indicate a relationship to those. But our current build would rely upon the receiving system to notice that the diagnosis or problem (or medication or therapy) is already on the patient's chart....sure we could add an indicator that the request contained a specific problem, condition, medication or therapy. But consider a condition that is not on the patients list - would not the receiving system be responsible for noting and doing something with it?
Bret H (Jun 08 2020 at 13:24):
FYI: I've been remiss in adding examples. schools out for the kids so I might have my laptop back
Bret H (Jun 08 2020 at 13:26):
also, as a reminder - we're not working on solving issues here as much as identifying where the current build is deficient. Try to stick with discussions that involve existing elements in the build.
Bret H (Jun 08 2020 at 13:27):
once the issues are identified - I suggest we start a new thread and close this one
Arthur Hermann (Jun 08 2020 at 16:42):
@Bret H @Kevin Power @Jamie Jones @Rachel Kutner - I am happy to address this in any way the group thinks is most efficient and allows for proper input. Bret - I understand you comment above - but until we have some examples using the components in our IG so we can see where the current build is insufficient, I personally have a hard time knowing what we need to address (if anything).
Bret H (Jun 15 2020 at 14:38):
@Arthur Hermann there's a good number of examples on the list now
Arthur Hermann (Jun 30 2020 at 16:11):
@Jamie Jones @Rachel Kutner @Bret H
@Kevin Power Please see above. I am also copying info from different threads so we have all of our info together in on place. My question is do we have a proposed solution, or do we need to meet and/or discuss more to propose solution for a vote?
Kevin Power (Jun 30 2020 at 16:24):
@Arthur Hermann -- For what it is worth, I do not see a proposed resolution in those comments, but maybe I am missing it? Do you have a proposal from those comments?
Arthur Hermann (Jun 30 2020 at 16:26):
@Kevin Power - I think we have in this mess a proposal... that is why I am asking the rest of the team to chime in.. if no response, I will work with @Jamie Jones to clarify - because I think that his research had turned up what (with some polishing) will become the proposal
Jamie Jones (Jun 30 2020 at 16:37):
Thanks all for contributing examples! I will take a pass down the sheet and outline how I'd see (versions of) them modeled in FHIR, and we can go over gaps, ambiguities, and inconsistencies. The 'Mad Lib' approach may be useful here and I encourage participation on that end.
Original focus was keyed in on if 2 separate components for associated cancer and associate phenotype is beneficial or even needed. The granularity between conditions and phenotypes is a concern. One proposal here is to have one broad component that is clearly defined and can cover both if they are used the same way (thoughts on example/preferred binding??)
Another focus is to challenge the current paradigm of relying on the observation code (diagnostic vs therapeutic) to define the context and hence meaning of an associated cancer/phenotype. One proposal is to have components differentiate based on semantic context - i.e., one component on diagnostic imp for the concept of a 'potentially implied condition', two components on therapeutic: 1 for an 'at-risk condition' (e.g. predicted complications) and another for 'contextual condition' (able to link a cancer type or other indicated phenotype to a possible treatment)
Arthur Hermann (Jun 30 2020 at 16:38):
@Jamie Jones @Kevin Power @Bret H - Jamie I am not very technical when it comes to FHIR but if I can help.. just reach out please!
Jamie Jones (Jul 02 2020 at 21:46):
I really liked the Tempus report Rachel added (https://drive.google.com/file/d/1wqoVUMP0dDsM4msSrgolv4At7sVDkoN_/view) so I put most of it in a CSV format that I use to generate FHIR resources for Connectathons (https://docs.google.com/spreadsheets/d/1C3SSYuOAYSp8EPN3gS3aHKP1tAL0U8UcbOJ3GT00t1A/edit#gid=666724439). Please review if you're interested! I made some ... leading choices ;)
Arthur Hermann (Jul 02 2020 at 23:02):
Great work Jamie - thanks! So remember that according to the Kaiser geneticists diagnostic vs therapeutic is essential for the lab's work.. but they don't expect to see this explicitly elucidated in the report. It sounds like Epic has had clients request this, so no harm in exposing it if needed... but we should likely make it optional. It isn't obvious from your spreadsheet... what changes are you suggesting to the model to separate these two concepts - because that would be the suggestion we need to clearly document and get in front of the group for a vote.... again - thank you for all of you wonderful work on this issue!!
Kevin Power (Jul 03 2020 at 17:10):
@Jamie Jones - Great stuff indeed. Curious why you are using 'medication-assessed' rather than 'therapy-assessed'?
Kevin Power (Jul 03 2020 at 17:12):
And I assume things like "Multi-TKI', 'mTOR Inhibitor', 'Anti-androgen', etc are more like drug classes?
Jamie Jones (Jul 03 2020 at 17:12):
We had written for therapy-assessed: "The non-medication therapy whose implication on the cancer outcome is being predicted.E.g. altered diet, radiation therapy, surgery, etc. If multiple medications and/or therapies are present, the implication asserted represents the implication of the combination of interventions."
Jamie Jones (Jul 03 2020 at 17:13):
there's a clinical trial in there for Vigil Immunotherapy + Irinotecan and Temozolomide in Ewing's Sarcoma (NCT03495921). We haven't put enough focus on clinical trials but I think the approach I took for it is decent
Kevin Power (Jul 03 2020 at 17:13):
Wonder if that name will cause confusion, and if we should rename that slice 'non-medication-therapy-assessed' to be more clear?
Kevin Power (Jul 03 2020 at 17:14):
And yes on your approach to the clinical trial -- I think it looks good.
Jamie Jones (Jul 03 2020 at 17:15):
i think all of our slice names will cause confusion to folks at different points, so as long as there is a good definition in the glossary I wouldn't get too hung up on the names themselves :) (though I agree improvements could be made)
Kevin Power (Jul 03 2020 at 17:18):
Fair point, but I do like clear names when they are possible, and especially when they can distinguish themselves from other names.
Jamie Jones (Jul 03 2020 at 17:20):
for these cancer reports, the associated-phenotype field for therapies is really useful to denote the studied cancer type of the medication, as it is common to list therapies approved for cancer types separate from the patient's diagnosis. Not something I had thought of before.
Kevin Power (Jul 03 2020 at 17:23):
One one slight reservation in how you are doing 'medication' -- not sure about just lumping 'drug class' in with 'drug' -- are you imply the drug class would get delivered in someway as well? I am likely way out over my skis here, but are they are only saying this is the class for the drug we are talking about, or are they saying something in this class can be used but here is the one we recommend? Or something else?
Jamie Jones (Jul 03 2020 at 17:28):
Not my expertise by any means. I assume not all mTOR Inhibitors are created equal, for example, the report is only talking about one, Everolimus (but they do explicitly label it as mTOR Inhibitor in their table, so I wanted to make sure it was represented). I recall us punting modeling drug classes, but our guidance is loose and the component is unbound.
Kevin Power (Jul 03 2020 at 17:30):
Sparks some thought anyway :slight_smile:
Jamie Jones (Jul 03 2020 at 17:30):
Yes, also when considering combination therapies (these reports list a few)
Kevin Power (Jul 03 2020 at 17:31):
One last question for you (for now): Why have a Variant row for the 'low coverage region' section and not Region Studied?
AREG;CBLB;FBXW7;FLT4;GFRA2;IGH;KDM5A;NOTCH1;PDPK1;RANBP2;SEMA3C
Jamie Jones (Jul 03 2020 at 17:33):
I'm seriously concerned with our current value structure on Variant and Region studied, as that "variant" I drafted marking them as "Indeterminate" also seems valid. I know we specifically made region-studied for that purpose but folks will likely try this way too unless we call it out
Kevin Power (Jul 03 2020 at 17:35):
The report didn't claim any "observations" in those low coverage region's did they? I am still not clear why you would have tried to denote that as a variant?
Kevin Power (Jul 03 2020 at 17:36):
I think we need to be clear that Variant.value = Indeterminate is more for "I looked for a variant, and for some reason I am not sure if it is there or not"
Jamie Jones (Jul 03 2020 at 17:37):
Isn't that what happened though?
Jamie Jones (Jul 03 2020 at 17:38):
I agree those should be region-studied, but take a look at https://drive.google.com/file/d/16n9SfVqxvMvQORM_Hz5V7nDt8h2r3vMy/view?usp=sharing. This one explicitly lists pertinent negatives at the gene level
Kevin Power (Jul 03 2020 at 17:39):
Perhaps this goes back to our lack of clarity around 'defining the variant' - if you defined a specific 'definitional variant' in one of those genes, and sent that as a Variant profiled observation with Indeterminate, I suppose that is OK.
Jamie Jones (Jul 03 2020 at 17:39):
Is that also region studied? we have no "negation" structure on it currently so you'd have to explain it in Observation.note
Kevin Power (Jul 03 2020 at 17:41):
Ahhh, yes, Pertinent Negatives is different from low coverage. Pertinent Negatives is something we should support. Perhaps at the gene level is OK? That sounds like some fun conversations just waiting to happen :slight_smile:
Kevin Power (Jul 03 2020 at 17:43):
And that sentence from the last report has a key word in it:
No pathogenic single nucleotide variants, indels, or copy number changes found in:
They key word is pathogenic -- does that mean those genes might have some variants, but just nothing pathogenic?
Kevin Power (Jul 03 2020 at 17:47):
Well Jamie, once again some great work to both confirm alot of our work AND raise more questions. Can someone else take the second Tempus report that Jamie referenced, and fill out a second spreadsheet formatted like Jamie did the first? Perhaps @Rachel Kutner @Bret H @Arthur Hermann or anyone else?
Arthur Hermann (Jul 03 2020 at 17:48):
@Jamie Jones do we have reports from other labs that might shed light on this issue? I know I can get one from invitae if helpful... I think labs may Nobel retuning this info in many ways, so from my perspective, best to leave as open as possible without creating more confusion for those using IG
Bob Dolin (Jul 03 2020 at 17:48):
My ears perked up on pertinent negatives. What I'm about to say is half baked, but having some fun with it. We're looking at $find-subject-variants, wondering what it would look like if it supported multiple regions. The 400 error code approach no longer works, since some regions will or wont have been studied. So, we're looking at the case where you have three 'region' files - the query regions, the studied regions, and the no-callable regions:
Bob Dolin (Jul 03 2020 at 17:48):
Bob Dolin (Jul 03 2020 at 17:49):
and testing this logic:
Arthur Hermann (Jul 03 2020 at 17:49):
I could do that @Kevin Power but wonder if better to use a different lab’s report?
Bob Dolin (Jul 03 2020 at 17:49):
if conversion_region:
if region_studied:
query_regions =
region_studied.subtract(nocall_region).intersect(conversion_region)
else:
query_regions = conversion_region
Kevin Power (Jul 03 2020 at 17:49):
@Arthur Hermann Whatever lab report you are willing to use :slight_smile:
Bob Dolin (Jul 03 2020 at 17:50):
Anyhow, long story short, is that, even for multi-region queries, I think the above approach lets us differentiate true negatives vs. the absence of variants due to region not covered or region is uncallable.
Kevin Power (Jul 03 2020 at 17:51):
@Bob Dolin - I think that is some fun, but I don't want to distract too much from the 'Stress Testing implications' theme we have going here, if that is OK?
Arthur Hermann (Jul 03 2020 at 17:51):
We also need to focus on answering just a few questions, otherwise this exercise will spin out of control
Bob Dolin (Jul 03 2020 at 17:51):
Ok @Kevin Power . Maybe a future discussion after more fully baked...
Jamie Jones (Jul 03 2020 at 17:52):
Agreed, another topic for pertinent negatives may be in order.
One more take away from the tempus reports for me: Prognostic biomarkers are hard.
Kevin Power (Jul 03 2020 at 17:53):
To echo what @Arthur Hermann said, anyone who does the same sort of analysis Jamie did should have their primary focus on the Diagnostic and Therapeutic implication tabs, for sure.
Kevin Power (Jul 03 2020 at 17:55):
@Jamie Jones - As in this section of the 2nd report you linked?
ADDITIONAL INDICATORS
Unfavorable prognosis KRAS p.G12C Gain-of-function
Consensus, non-small cell lung cancer: NCCN
Kevin Power (Jul 03 2020 at 17:58):
If so, is that our Diagnostic implication, with 'prognosis' component = Unfavorable?
Jamie Jones (Jul 03 2020 at 17:58):
That, OR:
I think therapeutic implication is supposed to have 1..* either medication or therapy associated, but we didn't explicitly add an invariant for it and both are currently 0..*, so could model that as a therapeutic implication without a therapy
Kevin Power (Jul 03 2020 at 17:59):
Yea, maybe me over skis again, but that feels more Diagnostic than Therapeutic?
Jamie Jones (Jul 03 2020 at 18:00):
I agree. BUT I don't like putting nsclc as associated-phenotype in diagnostic because every other instance of that pattern uses that field for an implicated diagnosis, not the context
Kevin Power (Jul 03 2020 at 18:03):
Hmm, maybe? Does not feel like a stretch to me (using associated-phenotype), but can let others weigh in.
Jamie Jones (Jul 03 2020 at 18:05):
Oh, I have completely overrun my skis, but I think the statement is "patients with nsclc AND this variant have a bad prognosis", not "this somatic variant is associated with nsclc AND a bad prognosis". I don't like the presence of the prognosis component changing the meaning of the associated-phenotype component.
Jamie Jones (Jul 03 2020 at 18:06):
adding a 'contextual-cancer' component to both implications would solve this problem and make all these examples more clear. Might open up outside cancer space too, though, so should look at other uses.
Kevin Power (Jul 03 2020 at 18:07):
I see the distinction you are making. I actually read it as "this somatic variant is associated with nsclc AND a bad prognosis", hence why I was OK with associated-phenotype.
Kevin Power (Jul 03 2020 at 18:11):
Jamie Jones said:
adding a 'contextual-cancer' component to both implications would solve this problem and make all these examples more clear. Might open up outside cancer space too, though, so should look at other uses.
Would it always be 'contextual-cancer' ? We would have to be really clear in our definitions to say when to use what.
Jamie Jones (Jul 03 2020 at 18:18):
I'm going to try and dig in to a different set of reports after lunch, it certainly helps the cancer case as from these 2 reports you can see very similar KRAS Gain of Function mutations (p.G12C and p.G12D) have completely different therapeutic recommendations based on the indicated cancer types (lung vs colon)
Arthur Hermann (Jul 03 2020 at 18:21):
I can't fine any other reports that we have which provide this detailed info. I just sent emails two Invitae and a KP geneticist and sent the Tempus report asking if they have anything like that report. @Jamie Jones what other reports are you going to dig into.. more tempus reports? I just want to be sure we are not using one lab only as our basis for understanding here.. BTW - in terms of being over our skis, I can send our findings to some oncologists and validate that our thinking is either correct or incorrect, once we are ready.. based on all of this - I don't see it being helpful to create another spreadsheet - especially if Jamie is doing so. I will be happy to do it, with any new lab reports I can get - make sense?
Arthur Hermann (Jul 03 2020 at 18:22):
@Jamie Jones if you want me to try to dig into some reports instead of you... point me to some reports that make sense to use... again, I am concerned if we only use Tempus
Jamie Jones (Jul 03 2020 at 18:23):
I agree we may not need another sheet, and we need to validate against other labs' reports. Lots of good examples on https://docs.google.com/spreadsheets/d/1j6YfGPsBht63Dm_KFHMnCLzom1XzFkT7mnMTagiEc-s/edit#gid=0 now thanks to the efforts from this group!
we've got color, arup, mayo, partners, and strata (may be others). I'll take a look at strata next, with a magnifying glass if needed :)
Kevin Power (Jul 03 2020 at 18:32):
Considering our focus on implications, we would be remiss if we don't do 1-2 PGx reports, so that would be good if someone is simply looking for something to do :slight_smile:
Kevin Power (Jul 03 2020 at 18:33):
And if we use the same sheet, I guess that is OK, we just need to make it clear what rows in the sheet goes with what report.
Arthur Hermann (Jul 03 2020 at 18:39):
@Jamie Jones I am having trouble "translating" some of these reports into the same fields you have in your xls... maybe just my misunderstanding of report? if you want to walk me through one I could help with others.. if this is not worth your time, then that is fine.. call me at 925 588 8766 if you wish to walk through one
Kevin Power (Jul 03 2020 at 18:40):
Which report are you trying? I can give you a few hints maybe?
Jamie Jones (Jul 03 2020 at 18:42):
I added a "report" column
Kevin Power (Jul 03 2020 at 18:57):
Might want to change the name of the spreadsheet as well.
Jamie Jones (Jul 03 2020 at 19:28):
I'll throw in some from this ARUP P450 genotype panel: https://drive.google.com/file/d/1Ung6sH9jW7iQZlvApu5Is0iem3BwUhZj/view
Arthur Hermann (Jul 03 2020 at 19:48):
I am trying to do one with Mayo Wilson's Disease report to see if I can still map
Arthur Hermann (Jul 03 2020 at 20:16):
@Kevin Power @Jamie Jones could one of you give me a call at 925 588 8766 - trying to work with report having trouble.. could use some input
Arthur Hermann (Jul 03 2020 at 20:29):
@Jamie Jones are you working ONLY with Cancer reports? In other words, would a somatic change unrelated to cancer still fit into the spreadsheet you have created? I am having great trouble finding many of the fields
Jamie Jones (Jul 03 2020 at 20:42):
not all fields have columns yet, that is a good point, and we provide multiple ways to represent info. I have been normalizing to one representation but I can add more columns in --sheet may get pretty wide though
Jamie Jones (Jul 03 2020 at 20:46):
https://docs.google.com/spreadsheets/d/1zgT0GYQZHjUYD7N0mYNzJozqdPieW7G2yglh3Ayl6gU/edit#gid=1105995725 has all of our components and http://build.fhir.org/ig/HL7/genomics-reporting/Glossary.html is very helpful as well
Arthur Hermann (Jul 04 2020 at 00:22):
Love your “cheat sheet in google sheets.. VERY helpful.. thanks Jamie! :tada:
Rachel Kutner (Nov 18 2020 at 15:06):
@Jamie Jones Where can I find the drive folder with all the examples we have? I know it was sent before, but I can't figure out how to find it - and wanted to look through our examples to see if there are any with Structural Variants included.
Jamie Jones (Nov 18 2020 at 15:08):
https://drive.google.com/drive/folders/18T4RS0VnrJdLS3k79skbrZ0cYyL1U53t
Last updated: Apr 12 2022 at 19:14 UTC
