Overlap between ICD-9 and ICD-10

[joecandeux]

If one has a data set with ICD-9/10 diagnosis codes in one set of fields (say, dx1-dx25) and ICD-9/10 external cause of injury codes in a different set of fields (say, ecode1-ecode4), ICDPIC traditionally requires renaming the external cause of injury codes to have the same prefix as the diagnosis codes (i.e., rename ecode1-ecode4 to dx26-dx30). However, this causes problems with the transition to ICD-10. ICD-10 has diagnosis codes starting with "E" that ICDPIC confuses with ICD-9 external cause of injury codes. This results in ICDPIC picking up extra mechanisms and intents that don't belong.

The ultimate solution is to keep diagnosis codes and external cause of injury codes separate, which would require a major restructuring of the program. Is there a short-term solution that might work?

I can supply a small example data set that illustrates this problem if desired.

[ablack3]

This is a problem when trying to handle both ICD 9 and ICD 10 at the same time. ICDPICR also requires those E-codes to be renamed with the same prefix as all the other diagnoses codes. I did some thinking about the issue of overlapping codes a couple years ago and put my comments on the topic in the cat_trauma.R file on line 142. At the time I thought we were ok since while some codes exist in both coding systems (E and V codes for example) they are only relevant to injury classification in one or the other. The current behavior of ICDPICR is to treat them as relevant injury codes if they are in the data. (i.e. if your data has codes that start with E and match an ICD9 external cause of injury code then it will be treated as such. This might not always be correct.

This is an important issue and if you have some simple examples I could convey to unit tests that would be much appreciated.

As far as a short term solution, if your E codes are ICD 9 external cause of injury codes then ICDPICR should work as expected provided all the columns have the same prefix. Another suggestion would be to process your ICD9 and ICD10 data in separate steps.

If you don’t mind me asking in your experience how often do researchers have ICD9 and ICD10 codes mixed together in one column? Perhaps I should be putting the burden on the user to specify which codes are ICD9 and which are ICD10 rather than trying to guess.

[joecandeux]

Thanks for the detailed response. I am uploading a small file that illustrates the problem I am having. When you run ICDPICR on it, notice that some of the ICD-10 diagnosis codes that start with "E" are getting copied to the ecode_* variables and being interpreted as ICD-9 external cause of injury codes. They show up in the mechanism variable as "Transport, other".

NRD_Readmissions_W.txt

I would be fine with processing ICD-9 and ICD-10 in separate steps. That would actually solve my problem since the data set I am working on at the moment is only ICD-10 (although that rarely happens generally). Is there an option in ICDPICR to specify only ICD-10? From my brief examination of the code it looks like the ICD-10 option includes both ICD-9 and ICD-10.

The HCUP data sets that I use a lot (NIS, NEDS, NRD, KID) have different variable names for ICD-9 and ICD-10 codes, so processing the two separately is not a problem. I think the NTDB also keeps them separate. I have used a dataset (IBM MarketScan) which stores both ICD-9 and ICD-10 diagnosis codes in the same field. I rarely use that for trauma, though, so I haven't run into problems yet.

[joecandeux]

Sorry, didn't mean to close (GitHib noob).