OHDSI Home | Forums | Wiki | Github

ICD-9 to ICD-10


(Paula Casajust) #1

Hi!,
I am very new to OMOP and I wanted to ask if anyone could point me in the direction of
how to convert ICD-9 codes to ICD-10. Should I use the mapping of CMS?
Any suggestions would be welcome

https://www.nber.org/data/icd9-icd-10-cm-and-pcs-crosswalk-general-equivalence-mapping.html
https://www.cms.gov/Medicare/Coding/ICD10/index.html


(Dmytry Dymshyts) #2

Welcome to the Community!

Yes, CMS mapping is good.

I’m wondering why do you need these crosswalks?
In OMOP we link both ICD9CM and ICD10CM to SNOMED. So if you want to operate with OMOPed data, you need mappings to SNOMED you can obtain from concept_relationship table.


(Paula Casajust) #3

Thank you @Dymshyts for the information. I need to convert because I need to use a non-fully OMOPed data source for a specific project which does not contain SNOMED.


(Christian Reich) #4

Wo, wo, wo!!! :slight_smile:

The CMS mapping is lousy, but as good as it gets. In general, mapping ICD9CM to ICD10CM is a bad idea. A stop gap. The codes only partially have the same semantic content, roughly 50%-70% map over, depending how much of a stickler you want to be. The rest do not map over well.

We generally do not recommend doing that, no matter how attractive the theoretical idea sounds. Instead, we map to SNOMED, a much more comprehensive vocabulary system with description logic (roughly, all concepts are combinations of attributes, and descendants inherit the attributes from their ancestors and add more).

Can you explain?


(Paula Casajust) #5

Many thanks @Christian_Reich. I am aware that it is not a quite good idea, and that not all the codes have the same semantic content.

I’m not really part of that project. I just was asked about how to convert ICD-9 codes to ICD-10. So I cannot offer much more information about that.


(Dmytry Dymshyts) #6

@Christian_Reich I know CMS ICD9CM to ICD10CM mapping is not good, mostly because there are the cases one to many mapping, so you don’t have exact parent to match to, but if given semantic meaning of concept exists in both vocabularies, they have this mapping as 1 to 1.
BTW, to solve 1 to many mapping issue, you may find common parent of the target concept group, and map source concept to it.


(Paula Casajust) #7

Thanks @Dymshyts, I will consider your suggestions.


t