Re: Is there a sas code that can clean up ICD-9 codes?

Thank you for your code!
I will check if it is helpful for me.

I posted my question because I was bothered the most by the following
finding:
Medicaid data has '3153','31530','31531','31532'.

The CMS label you mentioned below has only 2 lables as valid ones--the
last two: '31531','31532'. These two belong to higher aggregation
classification code , 74 other mental.

The first three are not valid ICD-9 codes. However, I think that they
should not be classified 'missing'.
I want to include them as 74 other mental.

Do you think that your code help me in fixing that type of errors?

Thanks again,
Duckhye

>>> Jonathan Kerman <jkerman@JHMI.EDU> 02/22/05 2:20 PM >>>
I have been doing a fair amount of work with ICD-9 codes. I could
provide
you with some of what I did. Basically, I checked a text string to
make
sure it has the format of a valid ICD-9 code, or reject it. I also
built a
dataset that maps ICD-9 codes to text descriptions.

I got my input files from the Centers for Medicare and Medicaid
Services at
URL: http://www.cms.hhs.gov/paymentsystems/icd9/
After I read these text files, one for diagnosis codes and one for
procedure
codes, into SAS datasets, this is the code I used to build a single
lookup
table. It places a decimal point at the proper place, and creates the
mapping of codes to descriptions. (I found that there are still some
codes
that are not found in these CMS files.)

See if this helps...

------------------------------------------------------

libname comp "[location with SAS datasets]";

/* Build a lookup table with standardized, short descriptions of ICD-9
codes
taken from CMS
http://www.cms.hhs.gov/paymentsystems/icd9/ */
data comp.icd9lookup;
length CODE tmp $ 6 DESCRIPTION $ 30;
set comp.icd9diag (in=diag) comp.icd9proc (in=surg);

if diag then do;
if substr(code, 1, 1) = 'E' then do; /* for 'E' code diagnoses,
the
format is different */
substr(tmp, 1, 4) = substr(code, 1, 4);
if length(code) > 4 then do;
substr(tmp, 5, 1) = '.';
substr(tmp, 6, 1) = substr(code, 5, 1);
end;
end;
else do; /* for numeric ICD-9 diagnoses and for 'V' codes use this
format */
substr(tmp, 1, 3) = substr(code, 1, 3);
if length(code) > 3 then do;
substr(tmp, 4, 1) = '.';
substr(tmp, 5, 2) = substr(code, 4, 2);
end;
end;
end;
else if surg then do; /* for procedure codes, use this format */
substr(tmp, 1, 2) = substr(code, 1, 2);
if length(code) > 2 then do;
substr(tmp, 3, 1) = '.';
substr(tmp, 4, 2) = substr(code, 3, 2);
end;
end;

code = tmp;
drop tmp;
run;

We found a great deal of variability in codes from claims data. In our
most flexible mathcing algorithm, we tried adding one and then two zeros to
the end of codes that didn't match, and subtracting zeros off the end of
codes that didn't match. So for

315

we would try

315, 3150, and 31500

For

31500

we would try

31500, 3150, and 315

And for

3150

we would try

3150, 31500, and 315

The easiest way to do this is with formats, but you could also use SET
codes KEY=lookupvalue. If the codes were always numeric (not the case
here) you could also use SET xxx POINT=numericlookupvalue. In V9, you
could use a hash table; I don't know whether that would be faster than
using a format, but my guess is that it would.



At 03:21 pm 2/22/2005 -0600, Duck-Hye Yang wrote:

>Thank you for your code!
>I will check if it is helpful for me.
>
>I posted my question because I was bothered the most by the following
>finding:
>Medicaid data has '3153','31530','31531','31532'.
>
>The CMS label you mentioned below has only 2 lables as valid ones--the
>last two: '31531','31532'. These two belong to higher aggregation
>classification code , 74 other mental.
>
>The first three are not valid ICD-9 codes. However, I think that they
>should not be classified 'missing'.
>I want to include them as 74 other mental.



-----
Jack Hamilton
jfh@alumni.stanford.org
Sacramento, California