Stream: implementers
Topic: Mapping snomed to icd-10
Eugenii (Jul 19 2021 at 08:08):
Good day!
I would like to create a database with mapping SNOMED to ICD-10. For some concepts there are additional age conditions. For example, Bacterial sepsis (10001005) with the additional condition Age of onset of clinical detection (445518008) <29.0 days. There can be many such additional conditions for age (<1.0 years,> = 1.0 years, <2.0 years,> = 2.0, <5.0 years,> = 5.0 years, etc. .).
Do you have any ideas how to work with conditions by age, how to store data in a database by conditions by age?
Lloyd McKenzie (Jul 19 2021 at 12:59):
Are you storing actual patient data and are trying to categorize it? In general, you can't store temporal categorizations because they're continuously changing. Something that was <29.0 days last week might be >29.0 days this week. So typically you just store the data with the date and use your query process to categorize at time of query execution.
Eugenii (Jul 20 2021 at 06:16):
Not certainly in that way. I want to tie time frames to mapping.
For example, there is Corneal degeneration with SNOMED code 111521006.
If the patient is >65 years old, the concept is mapped to H18.419 in ICD-10. In other cases to code H18.40 in ICD-10.
Thus, it is necessary to somehow store data at time intervals.
I can create my own code system with concepts for time periods. How good is this solution?
Jim Steel (Jul 20 2021 at 06:20):
Yeah, not great. There is work afoot in vocab to modify ConceptMap and $translate to allow dependsOn and product elements that are non-code typed. That would help
Eugenii (Jul 20 2021 at 12:27):
Thanks for the answer!
Do you have any results? I would like more specifics.
Lloyd McKenzie (Jul 20 2021 at 15:19):
Actually, you wouldn't want "if the patient is >65 years old" at the time of mapping, you'd want "if the patient was > 65 years old" at the time the SNOMED code was asserted (which could have been 10 years ago)
Lloyd McKenzie (Jul 20 2021 at 15:19):
So the mapping would have to be tied to a mathematical expression between 2 dates and shouldn't be dependent on when the mapping was run.
Daniel Venton (Jul 20 2021 at 17:52):
Are there other considerations in your snomed => icd 10? For instance, do you have to know location in order to do a translation. Take 254837009 (breast cancer) does that become C50.012, C50.812 or C50.912 OR if the patient is male, perhaps C50.021.
What it sounds like you are trying to do is convert a snomed ct based conditions list (patient chart) to an ICD10 based conditions list (patient chart). Without all of the details of the patient and probably a set of rules that are individualized to each snomed ct code, you aren't going to automate this process.
If code = 254837009 AND gender = male AND laterality = right AND location = central Then ICD C50.121
If code = 111521006 AND Age at diagnosis > 65y Then H18.419
If code = 111521006 AND Age at diagnosis <=65y Then H18.40
The list goes on....
Did you ask yourself why there wasn't already a Snomed ct => icd10 dictionary. This is why, it's not a dictionary it's a formula, likely a different formula for every single value.
Daniel Venton (Jul 20 2021 at 18:02):
There may be a simple conversion dictionary but it probably excludes all the extra stuff that icd10 brings along.
For instance:
If code = 254837009 Then ICD 10 Breast Cancer (Unspecified laterality, Unspecified location, Unspecified Gender). Unfortunately there doesn't appear to be such a code. (unless you want to code it as "History")
Peter Jordan (Jul 20 2021 at 22:35):
The SNOMED CT to ICD-10 Map developed by SNOMED International, is a complex, unidirectional map. Many of the challenges involved can be seen in this excellent presentation from the Australian Digital Health Agency.
Michael Lawley (Jul 21 2021 at 10:20):
Moreover if you are working with data that uses ICD10 codes without applying these rules, then they have a different meaning/interpretation and technically should not be consider ICD10 any more (not that this has ever stopped people from doing so :shrug: )
Daniel Venton (Jul 21 2021 at 12:19):
In the presentation that @Peter Jordan links to, one of the "map" tools specifically calls out that the output of the map is only to be used for research and never for anything patient facing. I suppose that is an indication of how reliable the map is.
Last updated: Apr 12 2022 at 19:14 UTC
