allisonheath commented 3 years ago

Requester information

Please provide the following information:

Name: Allison Heath
Affiliations: CHOP / KFDRC
Request Details

Please provide the following information about what you wanting to accomplish with your model change request:
Purpose: Determining if the base FHIR Specimen resource is the right place for representing specimens used for research.
Who it benefits: All platforms have some form of representing specimens in different granularities
Use case: For KFDRC we get genomic data that are the result of extracting DNA/RNA from specimens originating from blood, saliva, tumors, lesions and occasionally other solid tissue types (mainly bone right now). We currently make this information searchable on our portal as well as utilize it to determine which files to input into pipelines (e.g. tumor/normal pairs for somatic calling). For most studies we only have an identifier for the aliquot that was shipped to the sequencing center and so have kept our specimen modeling quite simple. Occasionally, we have information about multiple aliquots coming from same/different original specimens and this is important information to be able to provide for analysis. Especially with cancer, there might be multiple specimens taken from different sites and you want to be able to distinguish those down stream. Ideally, using the Specimen resource would give the flexibility to just put in the simple case, but augment additional information about what Specimens they were derived from when that information is available.

cc @liberaliscomputing @nicholasvk @youngnm @baileyckelly

RobertJCarroll commented 3 years ago

Do you have a list of fields from your source data? Some idea of prevalence and importance would also be helpful to review the fit!

bwalsh commented 3 years ago

https://www.hl7.org/fhir/valueset-body-site.html points at snomed codes. This is an important item for consensus. Likewise for collection method. https://www.hl7.org/fhir/valueset-specimen-collection-method.html

bwalsh commented 3 years ago

Tumor/normal is a common use case. Is this supported?

RobertJCarroll commented 3 years ago

Hi @bwalsh - Those are "Example" bindings, which means there is no weight given to the provided valuesets. You can tell because of the (Example) listed after the valueset links on the Specimen resource page. More info is available here: https://www.hl7.org/fhir/terminologies.html#strength

Re consensus, we may provide a list of our own in the IG, but it's not strictly necessary. This is where I expect we'll need to provide some of the terminology services you brought up on the call- there are likely structured links among common vocabs.

RobertJCarroll commented 3 years ago

Re tumor/normal- nothing obvious stands out to me here, but it seems like it really ought to exist. Not sure if I'm missing something.

allisonheath commented 3 years ago

Right now we've been using NCIt for specimen terminologies, as we reused a lot from the GDC. We have also done some light mapping to Uberon for anatomical site for comparative/model organism work. Not opposed to snomed (other than the licensing issues that pop up from time to time), especially as we're likely going to start getting data from clinically sequenced samples.

Also I guess to note - we keep the consent linked to the specific specimen, this is the level that dbGaP keeps it at and helps to distinguish if different specimens were take from the same person under different consents.

On a per-specimen basis, yes it can handle normal samples and tumor samples. But @bwalsh are you talking about knowing what specimen pairs were (or should be) used for analysis? Or being able to query to find availability of tumor/normal pairs? Or something else?

And here are our current list of fields:

entity	property
biospecimen	age_at_event_days
biospecimen	composition
biospecimen	concentration_mg_per_ml
biospecimen	consent_type
biospecimen	dbgap_consent_code
biospecimen	external_sample_id
biospecimen	kf_id
biospecimen	method_of_sample_procurement
biospecimen	ncit_id_anatomical_site
biospecimen	ncit_id_tissue_type
biospecimen	participant_id
biospecimen	sequencing_center_id
biospecimen	source_text_anatomical_site
biospecimen	source_text_tissue_type
biospecimen	source_text_tumor_descriptor
biospecimen	spatial_descriptor
biospecimen	uberon_id_anatomical_site
biospecimen	visible
biospecimen	volume_ul
aliquot	analyte_type
aliquot	concentration_mg_per_ml
aliquot	external_aliquot_id
aliquot	external_sample_id
aliquot	shipment_date
aliquot	shipment_origin
aliquot	volume_ul
aliquot	kf_id

RobertJCarroll commented 3 years ago

For CMG, we advocated for UBERON codes.

@allisonheath I think the question was "How do you indicate in FHIR if a Specimen is tumor or normal?".

liberaliscomputing commented 3 years ago

Here is an example of a KFDRC specimen using FHIR:

{
  "resourceType": "Specimen",
  "id": "bs-001-no-phi",
  "meta": {
    "profile": [
      "http://fhir.kids-first.io/StructureDefinition/kfdrc-specimen-no-phi"
    ],
    "versionId": "0.1.0",
    "lastUpdated": "2020-07-20T20:58:16.025096+00:00"
  },
  "extension": [
    {
      "url": "http://fhir.kids-first.io/StructureDefinition/age-at-event",
      "valueAge": {
        "value": 2750,
        "unit": "d",
        "system": "http://unitsofmeasure.org",
        "code": "days"
      }
    },
    {
      "url": "http://fhir.kids-first.io/StructureDefinition/concentration",
      "valueQuantity": {
        "value": 0.05,
        "unit": "mg/mL"
      }
    }
  ],
  "identifier": [
    {
      "system": "https://kf-api-dataservice.kidsfirstdrc.org/biospecimens?study_id=SD_BHJXBDQK",
      "value": "571312"
    }
  ],
  "type": {
    "coding": [
      {
        "system": "http://terminology.hl7.org/CodeSystem/v2-0487",
        "code": "TISS",
        "display": "Tissue"
      }
    ],
    "text": "Solid Tissue"
  },
  "subject": {
    "reference": "Patient/pt-001-no-phi"
  },
  "collection": {
    "quantity": {
      "value": 50.0,
      "unit": "uL"
    },
    "bodySite": {
      "coding": [
        {
          "system": "http://snomed.info/sct",
          "code": "21483005",
          "display": "Structure of central nervous system"
        }
      ],
      "text": "Central Nervous System"
    },
    "method": {
      "coding": [
        {
          "system": "http://snomed.info/sct",
          "code": "129314006",
          "display": "Biopsy - action"
        }
      ],
      "text": "Biopsy"
    }
  }
}

We've added two extensions:

age-at-event: For the X01 studies, we receive de-identified collection dates (e.g. in days), so we had to either expand the collected[x] attribute or create a new extension. Given we don't usually receive identified dates for diagnoses and phenotypes as well, we decided to create this extension to be used for Specimen as well as other profiles such as Condition and Observation.
concentration: This is a different metric from collection.quantity, so we had to create it

Re collection.bodySite: Given the use of SNOMED CT is an example (not required) and the attribute's data type is CodeableConcept, I am down for putting as many codes from authoritative ontologies as possible. So, the above example can be:

{
  "collection": {
    "bodySite": {
      "coding": [
        {
          "system": "http://snomed.info/sct",
          "code": "21483005",
          "display": "Structure of central nervous system"
        },
        {
          "system": "http://purl.obolibrary.org/obo/ncit.owl",
          "code": "NCIT:C12438",
          "display": "Central Nervous System"
        },
        {
          "system": "http://purl.obolibrary.org/obo/uberon.owl",
          "code": "UBERON:0001017",
          "display": "central nervous system"
        }
      ],
      "text": "Central Nervous System"
    }
  }
}

Re tumor/normal: Specimen has a type attribute which is pre-bound to a v2 Specimen Type ValuetSet. Again, the use of this ValueSet is not required, but it is pretty inclusive. The above example also use this ValueSet to code TISS. This ValueSet also includes:

{
  "code": "TUMOR",
  "display": "Tumor"
}

But, it doesn't have a code for Normal. We can do something like the followings:

Tumor tissue

{
  "type": {
    "coding": [
      {
        "system": "http://terminology.hl7.org/CodeSystem/v2-0487",
        "code": "TISS",
        "display": "Tissue"
      },
      {
        "system": "http://terminology.hl7.org/CodeSystem/v2-0487",
        "code": "TUMOR",
        "display": "Tumor"
      }
    ],
    "text": "Tumor Tissue"
  }
}

Normal tissue

{
  "type": {
    "coding": [
      {
        "system": "http://terminology.hl7.org/CodeSystem/v2-0487",
        "code": "TISS",
        "display": "Tissue"
      }
    ],
    "text": "Normal Tissue"
  }
}

RobertJCarroll commented 3 years ago

Ah, thanks. Type makes sense. I didn't think about it further after seeing it included things like blood!

NIH-NCPI / ncpi-model-forge

Representing Biospecimens #22

Requester information

Request Details