How to extract ICD-8 and ICD-10 codes from a large dataset using SAS?

05 May 2013 10 452 Report

Does anyone know how to extract ICD-8 and 10 codes from data using SAS? I want to extract causes of death and categorise them into explained and unexplained.

Mark E Cohen Popular answer

I'm not sure that I understand your question but we do some similar types of things with ICD-9 codes. It's first necessary to add the full text description to the file and then search for key words using the SAS substr or similar function. Once the list is of manageable size, I might order by frequency, and then evaluate each description to create you categories... but I could be misunderstanding your question.

Mark E Cohen

Yuri V Sebastião

It shouldn't be too complicated with SAS, but I think that I didn't fully understand your question either, you're trying to see if the cause of death variable(s) have a real ICD10 code as opposed to some other diagnosis and then basically separate the ICD10 codes (explained) from the diagnoses that don't match a real code (unexplained)? If this is the case, I think you would have to follow the approach explained in the answer above.

Sinéad M O Neill

Hi Mark and Yuri. I think I have since sorted this. I need to catergorise the causes of stillbirth into explained and unexplained to do a competing risks analysis. I will be given one variable (eg. All_Causes) which will consist of the many ICD-8 and ICD-10 codes for each particular cause of stillbirth. I think what I need to do is simply say if All_Causes = XX, or XX or XX then Explained= 1; Else Explained=0 (with XX being the different ICD codes). And then do the same for unexplained causes. Does this make sense?!

Sinéad M O Neill

Also, the SAS substr function...... is this a way of extracting particular 'words' from within a variable? So could I otherwise use this to extract the ICD codes?

Mark E Cohen

This will be a labor intensive process no matter how you approach it. I would begin by listing the text descriptions in frequency order and see if you identify key words in the text that would allow you to assign explained or unexplained status. Once you have those key words, then search the text in all the codes using the SUBSTR or similar function (if you don't already have it the book "SAS functions by example" describes all the string functions that might be useful for you). Hopefully you won't have any codes that have key words for both categories. Once that's settled it should be a simple matter to make the assignments.

Sinéad M O Neill

Thanks Mark. Will do

Alberto Vidal-Diez

I agree with Mark's method, you can get the list of descriptions for all codes. The problem is that although the code might be I744, sometimes it is I744-, so you normally use SUBSTR(DIAG_1,1,4) ="I744" to play only with the first 4 characters. I have worked with causes of death and sometimes are not so reliable, so you might want to check all causes of death rather the main one. If the patient had AIDS, and he died because of Pneumonia, should they put that he died because of AIDS or Pneumonia as main cause? There are many examples like that and some things that might not be coherent.

For your if statement you can use IN. Something like

IF SUBSTR(COD,1,4) IN ("I711","I722","I734") THEN explained_flg=1;

If you have a whole set of disease under the same codes, for instance I71x, then you could use SUBSTR(COD,1,3) = "I71" to save your time.

Maryam Sina

Hi Alberto;

your suggestion (IF SUBSTR(COD,1,4) IN ("I711","I722","I734") THEN explained_flg=1) does not work, it only works for numerical codes. I think we need a command that differentiates numerical from alphabets variables.

just wondering if anyone has an idea for identifying ICD 10 codes that start with a letter.