OHDSI Home | Forums | Wiki | Github

Some ICD10CM codes are not mapped correctly

Vocabulary Users
#1
I65 Occlusion and stenosis of precerebral arteries, not resulting in cerebral infarction
I65.01 Occlusion and stenosis of right vertebral artery
I65.09 Occlusion and stenosis of unspecified vertebral artery

These codes are mapped to following 2 standard SNOMED concept_id:

43022059 Disease of non-coronary systemic artery
255919 Finding of head and neck region

These two SNOMED codes are too high level concepts whose descendants include almost half of the diseases on planet earth and totally messed up my stroke analysis.

Translating source codes to standard concept_ids results in Conditions NOT of interest
#2

Just curious: why don’t simply use SNOMED hierarchy? It’s easy, convenient, transparent, save you a lot of time anyway and, after all, was specifically designed for research. I always wonder why people prefer to use incomplete and imperfect ICD10 (unless they replicate a study that used specific codes).

#3

@aostropolets, I believe @QI_omop is trying to use the SNOMED hierarchy, but when looking for what 165.01 is mapped to (because that may be what the source system is coded in), he’s finding that the only concepts are these generic ‘findings’ and ‘some disease’ concepts.

Why can’t 165.09 at least map to vertebral artery stenosis?

This concept doesn’t seem to have any mapping from ICD9 or ICD10.

1 Like
#4

@aostropolets

Exactly as @Chris_Knoll said, clients gave a list of iCD10CM codes, and wanted us to find out the population for these ICD10CM codes and descendants.

#5

@QI_omop, @Chris_Knoll

Vascular occlusion is known to be a blockage of a blood vessel with a clot or embolus, while stenosis is an abnormal narrowing of a blood vessel due to sclerotic changes. Taking into account this fact, we may suggest that Occlusion and Stenosis in the context of ICD10CM are NOT simultaneous conditions (I65 semantically implies OR between them).

So, if the vocabulary team map I65 to both 374371 1055001 Stenosis of precerebral artery AND 443239 266253001 Precerebral arterial occlusions, this will return obligatory both of them during a study whereas one of them never happened.
Also, it is not allowed to map them only to one concept because this leads to irrevocable data loss.

According to the mapping rules, such ICD concepts are mapped to the closest common ancestors (as 43022059 473449006 Disease of non-coronary systemic artery and 255919 118254002 Finding of head and neck region for I65), which can help to embed them into SNOMED CT hierarchy as SNOMED Extension concepts in the future (read more).

But now the best idea is to exclude I65-like concepts from the study.

#6

Thank you @Polina_Talapova. That is very helpful!

Now before we have the SNOMED Extension, as a temporary solution, would it be possible to publish a list ICD9CM and ICD10CM codes that are I65-like which are mapped to a higher level SNOMED concept? This way, we can advise clients not to use those codes. Currently, when these codes are used, we get almost 50% of our data asset population having stroke.

#7

We should report on how many mappings were modified/fixed in OMOP Vocab in 2020. I think mapping can be wrong and tracking updates to mapping would be a good thing.

Also, how community can submit retirement of specific mapping row and provide a replacement row with suggested mapping. (ideally an R package for that, linking to API).

Each mapping should have a history page where I see what was old mapping and when a new one replaced it. Well, hard to do in community without funds… I know.

2 Likes
#8

Actually, we have tech which would give the full history of mappings (as well as automatically remap). It’s been tested already on ICD and SNOMED

#9

Would it be possible to run it against the ICD - SNOMED mappings we have in OMOP? The “automatically remap” part is gonna be wrong mappings we have, right?

1 Like
#10

Yes, that’s a use case for sure.

What we can do is take a historical mapping of the ICD/SNOMED mappings from OMOP from an earlier point, and then run the AI, and compare with what’s already there in the last instance. That’s the easiest way to compare the AI’s accuracy with what OMOP currently has.

We would expect to see circa 90% accuracy.

Incidentally, this is also the test format we’ll be running for a Drug Pricing company as well.

Share Your Vocab/Concept Updating Process
t