OHDSI Home | Forums | Wiki | Github

Measurement hierarchy and LOINC place in it

Vocabulary Users
#1

Recently, we have introduced important changes to the representation of LOINC in OMOP vocabularies, some of them: added LOINC Parts and linked LOINC Parts with Lab Tests (1, 2), made common hierarchy between LOINC and SNOMED based on SNOMED Components and LOINC Parts 3, changed classification of radiology concepts from Measurement to Procedure 4, made discouraged concepts without replacement Standard.

Despite the huge number of changes and work done, there are still some things that we would like to discuss: during the recent LOINC refresh we were looking into the github issue #517.

While analyzing the problem, we encountered the following:

The hierarchy is built on the basis of one loinc source table LOINC_XXX_MULTI-AXIAL_HIERARCHY.CSV (exclusion: LOINC Group concepts). This table contains 171180 rows and defines the complete hierarchy for each concept. 55770 concepts are Standard Lab Tests and their immediate parents have class LOINC Hierarchy. But immediate parents of these LOINC Hierarchy concepts are LOINC Components (54926 concepts), and only 2008 concepts have LOINC Hierarchy parent. The hierarchy can be described with help of this table:

Loinc concepts Direct parents Second level parents
Valid Lab Tests: 55 770 LOINC Hierarchy: 55 770 LOINC Hierarchy: 2008
Valid Lab Tests: 55 770 LOINC Hierarchy: 55 770 LOINC Components: 54 926

As you can see, LOINC Components and LOINC Hierarchy concepts are deeply interconnected.

OHDSI relies on source hierarchical relationships in building concept ancestor, hence in concept_ancestor tables it is possible to go from urinalysis to counting WBC in blood specimen, and it is also possible in the LOINC multiaxial hierarchy we took as source. Note that LOINC Components are non-standard concepts and hence absent in concept_ancestor.

|714.5544970926636x428.74266851302525
|714.5544970926636x428.742668513025251600×959 151 KB

However, there are many hierarchical concepts that have only one parent and this parent is LOINC Component (eg. 37070204, 37060474, 37062605, 37034848, 37076837). They are the link between Classification concepts.

Here what we can do:

  1. Leave it as it is. In this case, unrelated concepts will remain linked as they are in source LOINC files.
  2. Deprecate links between LOINC Components and LOINC Hierarchy concepts. In this case, we will lose a huge part of the hierarchy.
  3. Deprecate links between LOINC Components and LOINC Hierarchy concepts, but make LOINC Components Standard. In this case, we will have duplicates with SNOMED concepts.
  4. Deprecate links between LOINC Components and LOINC Hierarchy concepts, but make LOINC Components classification concepts. But, in this case, we will have conflict with SNOMED. You can read more about it here.

@Alexdavv @Dymshyts @Christian_Reich @zhuk @Vojtech_Huser @Polina_Talapova @Vlad_Korsik @aostropolets @hripcsa @rimma @Denys_Kaduk @Chris_Knoll @Eduard_Korchmar

2 Likes
#2

I don’t have a clear answer.
Just a comment1: Effort to coordinate two standards (LOINC and SNOMED CT) is a challenge. I don’t know all the details but I think we embarked on some of it ( I think…)
OHDSI is not a fully resourced SDO (Standard Development Organization).

Another comment2 has to do with which relationship are driving hierarchies. Again, I may now know all the details of how things are set up. But… can we keep the relationship knowledge (for expert queries) but turn off the is_hierarchical flag…

#3

I’m not deeply knowledgeable about the LOINC terminology. So, I’m unsure which option will fulfill or break our use case.

We want to make it easier for researchers to identify concepts of interest. Wildcard, string text searches are the enemy of efficiency and accuracy. Enemy is a strong word, but the bane of existence for the analyst helping the researcher further define an attribute of interest. Which option allows for finding all ‘serum cholesterol’ Measurements? Can we do that with Concept Ancestor? Or, per the GitHub issue #517 linked above, maybe we find all ‘cholesterol’ Measurements and then filter by the system from which it was collected? Were the substance and system data added per Anna’s suggestion?

Also, what does this mean?

#4

Also, there is a possibility to leave the existing structure as is, AND, using LOINC Components, to create the ‘OMOP Measurement’ vocabulary - clean and beautiful, by analogy with ‘OMOP Vaccine’ which had been planned by the vocabulary team but not represented here, in the forum.

t