OHDSI Home | Forums | Wiki | Github

Proposal to store radiology diagnosis report in OMOP CDM


In the last Psychiatry Working Group meeting, we ran into the issue of how to store radiology diagnostic report in OMOP CDM. Examples of such report are below:

  • Bilateral basal ganglia lacunar infarction

  • Mild cerebral atrophy in bilateral medial prefrontal lobes

  • White matter demyelinating

  • Bilateral frontal fissure widened slightly

  • Intracranial aneurysm of the left vertebral artery

Please note: these are not original CT scan finding but the diagnosis report based on CT examination. If we parse these reports, some commonality of the attributes can be seen and are listed below:

Radiology diagnostic report Laterality Body site Severity Condition/Diagnosis
Bilateral basal ganglia lacunar infarction bilateral basal ganglia lacunar infarction
Mild cerebral atrophy in bilateral medial prefrontal lobes bilateral medial prefrontal lobes mild cerebral atrophy
White matter demyelinating white matter demyelinating
Bilateral frontal fissure widened slightly bilateral frontal fissure slightly widened
Intracranial aneurysm of the left vertebral artery left vertebral artery Intracranial aneurysm

As you can see, these diagnostic reports have more granularity than what SNOMED can currently handle. Besides the condition/diagnosis, they may also include other attributes such as severity, laterality, size as well as more detail anatomy/topology. As this is similar to tumor diagnosis that has modifiers of the diagnosis, we propose to use the same mechanism (measurement table) to handle these radiology diagnostic reports just as how it is done for tumor. Specifically, load diagnosis into CONDITION_OCCURRENCE table and put other attributes (laterality, severity, size and body site etc.) into MEASUREMENT table and link them via the two modifier columns (modifier_of_event_id and modifier_of_field_concept_id)

Below displays the result of this proposal using “Mild cerebral atrophy in bilateral medial prefrontal lobes” as an example:

The idea is probably not new. Some relevant discussion can be found here and here.

Please feel free to shed any insight on this.

I agree that we need to discuss about this in the community.

Technically, it’s condition in OMOP originated from radiology report in some databases (rather than ‘radiology’ diagnosis).
I think we can leverage the wisdom of oncology WG about this issue. How do you think, oncology folks?

@QI_omop @SCYou

I think how @QI_omop expressed reusing the relationship between a diagnosis (CONDITION_OCCURRENCE) and refining tumor characteristics (MEASUREMENT via
modifier_of_field_concept_id) for radiology findings makes sense. In the OMOP Oncology Extension, we have mostly discussed pathology reports that express a diagnosis and the additional findings, but radiology reports have a very similar structure.

Please bring this to the OMOP Oncology CDM/Vocabulary Subgroup meeting.

I’ve mapped some Diagnostic descriptions kindly provided by @Vines.
Please see the table attached:
radiology report.xlsx (10.1 KB)

In green there are suggested modifiers.

  • Instead of ‘laterality’ I used ‘Topographical modifier’ as there might be: frontal, dorsal, caudal as well as left, right.
  • so, there are:
    4021920 - Topographical modifier;
    3040950 - Body site;
    4077563 - Severity;
    4175951- Morphologically abnormal structure.
    All these belong to Observation domain. And I can’t find proper modifiers in Measurement domain.
    So, we have to change domain for these concepts OR add modifiers to OBSERVATION table as well.

CONDITION_OCCURRENCE.condition_concept_id are put in CONDITION_OCCURRENCE and connected to MEASUREMENT (or Observation) like it’s shown in @QI_omop’s diagramm above.
Where it maps to a diagnosis directly, I put diagnosis (4278243 - Cerebral arteriosclerosis (disorder)),
Where not - I put generic term (4041437 - Radiologic finding).

So there are two concerns:

  • Modifiers belong to Observation domain
  • condition_concept_id inconsistency.

Please let me know your thoughts

There are two issues that need to be solved, as usual: one is vocabulary and the other representation in the model. @meerawhy has explored RadLex, a specialized radiology terminology, recommended by MSK Radiology Informatics. It is very comprehensive, and covered every nuanced aspect of breast radiology. RadLex procedures are linked to LOINC (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6016707/). Other aspects of RadLex like anatomic site or finding are not linked to any standard terminology. Our team is exploring it now, and we’ll make a proposal of incorporating it into OMOP vocabulary. @meerawhy, please chime in if I missed anything.

@rimma I’m second to incorporating more RadLex into OMOP vocabulary.

Just had a brief discussion with @Vines about this topic

This is not a concern at all if CDM v6 is used.
Observation table has these modifiers, so Variable - Value pairs can be easily connected to the Condition table, which will contain the clinical finding per se (Cerebral arteriosclerosis, for example). But then we need to use different Type Concept to distinguish Imaging clinical finding from a diagnosis.

Is there any conclusion about the topic?