Using MedDRA for analysis and source code mapping coverage

sblyman · January 21, 2015, 12:42pm

I have a couple of questions related to the intended use of MedDRA in the OMOP vocabulary and wanted to see if anyone in the OHDSI community had some insight. Looking at the concept relationship and ancestor data in V5, the assumption is that a user referencing a MedDRA term in an analytic query would select condition records in the CDM via the mapping with SNOMED-CT.

Has there been any research done on the coverage of MedDRA term mapping to the source codes indirectly via SNOMED-CT? Many of our users will continue to use MedDRA for drug safety surveillance activities and it would be beneficial to have some factual evidence of the type and quality of coverage (e.g. overall mapping coverage indirectly through SNOMED and the coverage of the most commonly used terms such as listed events or those contained in the IME list). Research seems to validate direct mappings for ICD-9 to MedDRA and ICD-9 to SNOMED without significant loss of data integrity (Reich et al., 2012). Also, has there been any research on the quality of direct mapping of Read to SNOMED or ICD-10 to SNOMED (other than the coverage percentages listed in Appendix C of the OMOP V4 Vocabulary specification)?

Mark_Danese · January 21, 2015, 4:54pm

Perhaps this has already been done, but it would interesting to compare selected, “validated” definitions in their source vocabulary (for example, ICD9 or MedDRA) and compare the ascertainment to two different approaches – a direct 1:1 mapping (converting the source to the SNOMED concept) and an indirect mapping where you search in the concept MAP (SNOMED) for the same term. What is the agreement among these approaches? How does it vary for key conditions that people use all the time? I don’t know that anybody has done this, but it would be a great experiment for the IMEDS lab and a nice publication. It would be a bridge from current practice for new users to the OMOP process. I am sure Christian and Patrick have done some work on this, or perhaps even published a paper.

Christian_Reich · January 22, 2015, 7:21am

I checked briefly the current situation:

There are 95,214 MedDRA concepts

Of these, 53,057 have a mapping to a SNOMED concept

Keep in mind that the mapping is not done for LLT, only PT (and, through the hierarchy, above). I can’t currently count the coverage for those only, because due to a glitch the last release didn’t have the concept class information. Give us a couple days and we get back to you on that one.

What’s the IME list?

sblyman · January 22, 2015, 12:08pm

The IME (Important Medical Event) list is a list of MedDRA PTs that was created by a working group in the EMA. This list is used to facilitate data analysis in relation to pharmacovigilance activities. The current list can be downloaded from here.

Patrick_Ryan · January 31, 2015, 8:57pm

To @sblyman’s first question:

Has there been any research done on the coverage of MedDRA term mapping to
the source codes indirectly via SNOMED-CT? Many of our users will continue
to use MedDRA for drug safety surveillance activities and it would be
beneficial to have some factual evidence of the type and quality of
coverage (e.g. overall mapping coverage indirectly through SNOMED and the
coverage of the most commonly used terms such as listed events or those
contained in the IME list). Research seems to validate direct mappings for
ICD-9 to MedDRA and ICD-9 to SNOMED without significant loss of data
integrity (Reich et al., 2012). Also, has there been any research on the
quality of direct mapping of Read to SNOMED or ICD-10 to SNOMED (other than
the coverage percentages listed in Appendix C of the OMOP V4 Vocabulary
specification)?

I’m not aware of anyone who has done this, but definitely sounds like a
worthwhile analysis problem that would warrent publication. Note, we
consider MedDRA to be a classification system atop SNOMED, so the real
question is what proportion of SNOMED concepts (and proportion of data
mapped to those concepts) have a relationship with a MedDRA concept. This
would help address the completeness question, but there’d be further
opportunities to undertake some review to address the correctness
question. Beyond just doing an ‘all concepts’ approach to the summary,
it’d be quite interesting to see WHERE the concepts/data are mapped to
within MedDRA, to determine if the completeness is differential in any
particular areas.

I’d suspect that @Christian_Reich would recommend we wait until the next
version of OMOP Vocab V5 is released, but assuming that is soon, I could
imagine doing this type of analysis in time for an AMIA submission. Is
anyone interested?

herrcerd · March 5, 2015, 2:58am

Hi there - this is actually my first post, so thanks to everyone for providing such a great community.

We are actually working on MedDRA to ICD9CM mappings and have produced a number of curated mappings over the years but have recently had a requirement come up to map all of the IME serious MedDRA 17.1 PTs to ICD9CM codes.

We are using BioPortal SNOMED-CT mappings as the bridge vocabulary from ICD9CM to MedDRA. Is there anything in particular I might make a point of documenting to help answer questions posted here?

As an aside RE: correlation - Zhao-Wong (2012) discovered for the E and V ICD9CM codes a 94% relative confidence overlap between ICD9CM and MedDRA - so like many here I have a need to assess or at least measure accuracy/precision of the SNOMED mappings between the two vocabs.

Patrick_Ryan · March 5, 2015, 4:24pm

@herrcerd, Welcome to the OHDSI community, we’re delighted to have you.

Within the standardized vocabularies that we’re using within OHDSI, we have
mappings from ICD-9-CM to SNOMED, and then we have mappings from SNOMED to
MedDRA. We are not using BioPortal as the source of these mappings. You
can see more about it in the github repository,
https://github.com/OHDSI/Vocabulary-v5.0, and also @Christian_Reich can
shed some light.

One important point to make: The relationships within our standardized
vocabularies are directional and in this case hierarchical, not symmetric,
so an ICD-9-CM -> MedDRA relation would mean the ICD-9 code is subsumed by
the MedDRA concept, but wouldn’t necessarily mean that if you wanted a
MedDRA -> ICD-9-CM mapping that the MedDRA concept would be fully satisfied
by the ICD9 code.

iholt · May 26, 2015, 1:42pm

The documentation for Standardized Vocabularies V4.5 mentioned that support for the ICD9 to MedDRA mappings was going to be removed from subsequent releases, but it sounds like it is still in V5. Can anyone confirm this?

Also, if anyone knows if an updated link to the mappings of the ICD9 to MedDRA codes cited in the article published in the Journal of Biomedical Informatics 45 (2012) 689–696, “Evaluation of alternative standardized terminologies for medical conditions within a network of observational healthcare databases” exists, that would be extremely helpful.
The link in the article no longer works and was not forwarded.

Thanks,
Ingeborg

Christian_Reich · May 26, 2015, 2:42pm

@herrcerd:

We will release the improved SNOMED-MedDRA relationships for the IME set shortly (this week).

Christian_Reich · May 26, 2015, 2:47pm

@iholt:

The ICD9CM to MedDRA relationships still exist. Nobody has touched them. However, we don’t really encourage using them. Instead, we the proper OMOP way of doing it is this:

Map ICD9CM to SNOMED represent your data in SNOMED. Those mappings are in a very good shape right now, they will de-coordinate all those pre-coordinated ICD9 terms and also resolve history of and family history of
You will get MedDRA as a classifciation system for those SNOMED concepts with the vocabulary. So, if you want to query the data for all patients that have some condition you have MedDRA codes for there is a single query where you can do that.

Both the ICD9CM to SNOMED and SNOMED to MedDRA relationships are optimized for this way to use the data.

Let me know if you have more questions.

iholt · May 26, 2015, 6:13pm

One more, it seems the instructions (http://www.ohdsi.org/web/wiki/doku.php?id=documentation:vocabulary:vocabulary_creation_etl_process) to recreate the vocabulary step-by-step have disappeared. Will they be back soon or is it possible to link to the older version while the new version is being updated?

Thanks,
Ingeborg

lee_evans · May 26, 2015, 6:25pm

I moved that page to the link below. I made the change because I created a new ETL processes section in the documentation here (http://www.ohdsi.org/web/wiki/doku.php?id=documentation:overview) where the LAERTES ETL process documentation page will reside alongside the Vocabulary ETL process documentation.

http://www.ohdsi.org/web/wiki/doku.php?id=documentation:vocabulary_etl

This will be the permanent location for this page.

Christian_Reich · May 27, 2015, 1:33am

@iholt:

Why would you want to reproduce it? You can just download the meal ready-to-eat: http://www.ohdsi.org/web/athena/.

iholt · June 1, 2015, 1:27am

I’ve run the Vocabulary-v5.0\MedDRA installation scripts (ctl+load_stage.sql) and haven’t been been able to produce any mappings. Step 8 of the load_stage.sql script is “Create a relationship file for the Medical Coder”, this file has the format of the information I am looking for but is not complete. I was expecting the mapping to be done by the program not by a medical coder (or at least for it to make use of mapping done previously by a coder). Before I go back and review in more detail the steps to see where I things may have gone wrong on my side, I want to be sure that v5 contains the actual mapping information between SNOMED and MedDRA and not just the set up to have it mapped by another party which the comment in the load_stage.sql and the results of step 8 seems to indicate.

Thanks,
Ingeborg

Patrick_Ryan · June 1, 2015, 1:43am

Ingegorg: I’m not sure if this answers your question, but the easiest way
for you to get the SNOMED->MedDRA mapping is to just download it directly
from here: http://www.ohdsi.org/web/athena/. It sounds like you are
looking into the ETL scripts that are used to generate the mappings, but
the finished product of that semi-automated process is the release of
ATHENA.

Cheers,

Patrick

iholt · June 1, 2015, 2:47pm

Thanks for your response Patrick. I did download from athena but couldn’t find either a file that contained the SNOMED->MedDRA mapping or a way to link the tables to produce the mapping. If there is file that contains the direct mapping and I missed it, could you provide more information about the name and/or location of this file.

When I attempted to use the data files from athena (using the create_source_tables.sql and ctl files (Oracle) provided by github) to do the mapping I was unsuccessful despite the tables loading without any errors. Although the relationship table contains RELATIONSHIP_IDs for 8 MedDRA mappings (4 if we don’t count the reverse mappings) only MedDRA - SMQ/SMQ - MedDRA was contained in the final concept_relationship table and that isn’t of help to me. My assumption was due to licensing issues , MedDRA would need to be done through the ETL by a user who had access to the MedDRA files. I have the MedDRA data files (from the MSSO) and I was able to run the MedDRA ctl files and load_stage.ctl successfully but that only produce a template mapping file with the MedDRA information and no corresponding ICD9 information and not the mapping I was looking for (either SNOMED to MEdDRA or ideally, ICD9-MedDRA).

Any ideas on how to remedy this is appreciated. I am not ruling out that my implementation is faulty but it would be helpful to receive some assurance that someone (non-OHDSI) was successfully able to extract this information from v5.0.

Thanks again for you response,
Ingeborg

Christian_Reich · June 1, 2015, 5:02pm

@iholt:

Do you still have the Oracle database with the Athena tables? If so, run the following:

select c1.concept_code as meddra_code, c1.concept_name as meddra_name, r.relationship_id as rel, r.invalid_reason as r_ir, 
  c2.concept_code as snomed_code, c2.concept_name as snomed_name
from concept c1
join concept_relationship r on r.concept_id_1=c1.concept_id and r.invalid_reason is null
join concept c2 on c2.concept_id=r.concept_id_2 and c2.vocabulary_id='SNOMED'
where c1.vocabulary_id='MedDRA'

That will give you our mapping from MedDRA to SNOMED. Change the names of the vocabularies to “ICD9CM” where needed.

Let me know what you get. If you get an empty file, you probably missed to click on “MedDRA” when requesting the files from Athena.

iholt · June 2, 2015, 1:56pm

I wanted to follow up and see if the improved SNOMED-MedDRA relationships for the IME set was released and, if so, where on your site they would be- eg if they will be released as csv or tab-delim file with direct mappings or something that requires more assembly? Thanks.

Christian_Reich · June 2, 2015, 11:51pm

Ingeborg:

It will be released tomorrow or the day after.

And it will be in the exact same form as I alluded to. You need to donwload the vocabulary and run those queries.

iholt · June 25, 2015, 1:19pm

Thanks for the query, I was able to get the mappings I needed!