OHDSI Home | Forums | Wiki | Github

MedDRA Overhaul Discussion

Vocabulary Users
#1

After extensive deliberations, including discussion during the Common Data Model WG - VOCABULARY Subgroup meeting on October 3, 2023, we’re presenting the following plan to the community for the MedDRA overhaul in 2024 and MedDRA future in OMOP:

  1. We will focus on the Preferred Term (PT). It means that we cover with “Maps to” links to SNOMED mostly PT level of MedDRA concepts because:
  • PT concepts are used in RWD much more often than LLT terms
  • PT concepts are more convenient for analytics needs
  • in most cases, LLTs only represent the synonyms of PT (with some sort of granularity)

  1. We utilise available mapping candidates from various existing sources (OHDSI internally built mappings, UMLS thesaurus, MedDRA-SNOMED / SNOMED-MedDRA mappings from MSSO, ICD10-MedDRA mappings), including newly created mappings, to establish “Maps to” links between MedDRA and SNOMED.

  2. MedDRA remains a classification vocabulary (important: please see the details below in 6) and will be used in two directions:

  • to map source data with MedDRA codes to OMOP standard codes during ETL (using ‘Maps to’ / ‘Maps to value’ links)
  • to use MedDRA codes as classification concepts for concept set construction
  1. In case of complete semantic correspondence between the MedDRA and SNOMED concepts, the MedDRA concept is positioned above the SNOMED concept when constructing the hierarchy.

In case of incomplete correspondence, various scenarios are possible, where the MedDRA concept in the hierarchy can be either higher (downhill mapping to SNOMED) or lower (uphill mapping to SNOMED) than the corresponding SNOMED concept. See the schema attached.

image
image1333×692 74.1 KB

  1. It is essential to understand that when using MedDRA concepts to construct concept sets based on the MedDRA-SNOMED hierarchy, there might be cases where not all underlying child concepts from SNOMED are fully included. In such instances, it is recommended to conduct additional checks for the potential inclusion of additional SNOMED concepts when constructing the final concept set.

  2. Using MedDRA concepts as classification will be available for PT and LLT concepts with valid “Maps to” links (and ‘is a / subsumes’ links accordingly constructed on their basis). Other PT and LLT MedDRA concepts without corresponding “maps to” links will be considered non-standard (previously they were classification concepts).

  3. At higher levels of MedDRA (SOC, HLT, and HLGT), “maps to” links between MedDRA and SNOMED are not planned due to the complex differences in the hierarchy of these two vocabularies on higher levels. These levels will stay classificational.

For example, MedDRA PT concept “HIV infection” (concept id 36009928) and related SNOMED concept “Human immunodeficiency virus infection” (concept id 439727) belong to different axes of hierarchy:

MedDRA: Immune system disorders(SOC)-Infections and infestations(SOC)-Immunodeficiency syndromes(HLGT)-Viral infectious disorders(HLGT)-Retroviral infections(HLT)-Acquired immunodeficiency syndromes(HLT)

SNOMED: Clinical finding - Disease - Disorder due to infection - Viral disease - Disease due to Retroviridae - Disease due to Lentivirus

So, SOC, HLT, and HLGT MedDRA levels are planned to be used only to identify relevant PT/LLT concepts.

In conclusion, MedDRA will continue to serve as a classification vocabulary and support two primary use cases: source data mapping and concept set construction (using OMOP standard concepts only).

The changes will be included in the planned February 2024 release.

1 Like
#2

point 7. about SOC, HLT, LHGT - makes perfect sense to me.
Classification concepts are a bit described in (I think now outdated) old documentation here: documentation:vocabulary:standard_classification_and_source_concepts [Observational Health Data Sciences and Informatics]

And very briefly in the latest documentation:
https://ohdsi.github.io/CommonDataModel/cdm54.html#CONCEPT

standard_concept This flag determines where a Concept is a Standard Concept, i.e. is used in the data, a Classification Concept, or a non-standard Source Concept. The allowable values are ‘S’ (Standard Concept) and ‘C’ (Classification Concept), otherwise the content is NULL.

image
image938×375 23.9 KB

It seems some Athena included vocabularies have only source concepts while other may have source+classification.

#3

We are pleased to announce the latest release of MedDRA 26.1 with many updates and enhancements, including additional mapping to standard OMOP vocabularies and, for the first time, a hierarchy between MedDRA and SNOMED. For more detailed information on the changes and updates, please refer to the release notes

1 Like
#4

Hi, @Mikita_Salavei ! which parameters do I need to enable to ensure I have the latest mapping set of MedDRA-SNOMED?

#5

Hi, @Askar_Tursyn. Currently the latest mapping set of MedDRA-SNOMED presents in Athena. But MedDRA is an EULA required vocabulary. If you have a license for MedDRA you can click on “License required” in download section of Athena (Athena) to use this vocabulary.

#6

My apologies for lazy posting. I was planning to ask about the specific values in the table “concept_relationship” that I have to set up. Specifically the columns “relationship_id” and “domain”. Thank you.

#7

Thank you to those that participated in this overhaul! Unfortunately we still currently have a host of MeDRA codes that returned no matches or relationships to standard terminology in Athena after the overhaul even though they have semantically similar if not exact matches in SNOMED. Admittedly most of these are LLTs but curious if there is a continued effort to align these two vocabularies? Thank you.

#8

Hi, @cpathdolson. As part of the MedDRA Overhaul plan, which was discussed on the Vocabulary WG and OHDSI forum (MedDRA Overhaul Discussion), our primary focus is on mapping PT (Preferred Term) concepts. We only map LLT (Lowest Level Term) concepts that have been previously encountered in real data. In most cases, you can use the existing mappings of their corresponding PTs for LLT mapping, as LLT concepts in MedDRA are typically semantic equivalents of PTs. Occasionally, LLTs may be more granular in meaning than the PTs they belong to.

We continue to work on alignment between MedDRA and SNOMED vocabularies, but most effort will be put on maintenance, while new mappings are welcomed from contributors.

We would greatly appreciate it if you could share how LLTs are used in your research.

#9

Thank you @Mikita_Salavei! Critical Path Institute has a steady flow of RWD contributed so we run into all kinds of interesting situations mapping data. I was recently working on some rare disease data in which the contributor used MedDRA codes in their in SDTM datasets. Some of the codes were mapped through automation but others had to be mapped manually even when standard concepts that were semantically similar existed. Should I submit each instance of these for review by the community for future LLT mappings or would it be better to just continue mapping the LLTs (and a few a PTs) by hand? Here’s a few more details of project where this particular datasets is from: Rare Disease Cures Accelerator-Data and Analytics Platform

I really appreciate the quick response on an old thread! Thank you.

#10

Just to clarify, I am mapping data from SDTM (with MedDRA codes) into OMOP CDM concept IDs.

#11

Thank you, @cpathdolson, for clarifying your use-case.
OHDSI Vocabulary team continues to do some new MedDRA-SNOMED mappings for new release in August 2024 but mostly with emphasis on PT terms.
It would be advantageous for the whole to utilize any mappings for future use within the OHDSI community. Аnd if you could share your insights or mappings themselves, it would be wonderful.

#12

Hi there, @Mikita_Salavei - how complete is the MedDRA → SNOMED mapping considered to be? i.e., if I am missing a mapped OMOP standard concept for a given MedDRA concept, how likely is it that a standard concept simply does not exist vs that the mapping is incomplete?

#13

Hi @katy-sadowski. MedDRA-SNOMED mapping is currently ongoing, with the dictionary release in August 2024 adding about 3,000 more MedDRA terms mapped to SNOMED (of which 2,700 are PT level, which is now a priority). Finally, in August release will be approximately 18.4 thousand PT and 11.1 thousand LLT MedDRA concepts with valid mapping to SNOMED.
In general, new mappings will be added sequentially depending on community priority and available resources. Therefore, the lack of mapping of a specific MedDRA concept at this time does not mean that it is impossible to find a match for it in SNOMED.
If you have a portion of independently completed MedDRA-SNOMED mappings not currently available in Athena, we would be happy to incorporate them into the next release for the entire OHDSI as a community contribution.

#14

Thanks for letting me know. I have not completed any independent mappings.

I did find this paper which gave some very interesting stats about mapping coverage of MedDRA in OMOP (unfortunately it doesn’t specify what version of the vocabulary was used): Evaluating MedDRA-to-ICD terminology mappings | BMC Medical Informatics and Decision Making | Full Text (biomedcentral.com)

I think that statistics on unmapped codes like the ones provided in the paper would be an extremely useful resource for the vocabulary team to provide (not just for MedDRA). It could give more confidence in some cases, and appropriate caution in others.

Here, interestingly, only a small percent of exact matches were found to be missing; most misses were due to differing levels of granularity/scope. (They make another hop to ICD in the paper, but the principle is the same as they’re going via standard concepts.)

#15

It is interesting indeed to see how other groups interpret MedDRA - ICD - SNOMED relationships. In our work, we use UMLS relationships as one of the mapping candidate sources (we also use MedDRA - SNOMED, jump through ICD and AI-augmented semantic equivalents). @Mikita_Salavei has presented the results of our most recent work during the European symposium. According to our experience, we could not blindly rely on external sources, because these mappings very often break OMOP conventions (like mapping events to Meas Value domain) or have poor semantic equivalence. Therefore we combine them, analyse and then produce curated mappings. And everyone can help build them through community contributions.

We’ve been thinking about these statistics a lot. As part of the enrichment of the vocabularies with metadata, we plan to include this information - something like ‘concept not mapped because of no good standard equivalent’ vs ‘concept not mapped yet’. We are not there as of now, but we are moving towards this goal. We will publish the metadata for some of the relationships during the upcoming August release.

1 Like
#16

Oh I love this idea! The community would get a ton of value out of this (and I think it could also help direct community mapping efforts in the right direction!).

Thanks also for sharing your Symposium poster - I’ll check it out. I look forward to the August release and continued improvement of the vocabs. Really appreciate your hard work on this invaluable resource :slight_smile:

1 Like
t