By: Ron Mills
“Mike” wrote Rhonda:
I saw a good article you wrote in The Journal of AHIMA … been very helpful … but the bigger question we are encountering is the continuity of reporting from ICD9 to ICD10. I am hoping you may be able to point me in the direction of any work that has been done on supporting data warehousing, reporting and information challenges when the data spans 2013 and the ICD-9 and ICD-10 code sets.
Rhonda passed it on to me and I responded:
Mike, you’ve got a good* problem on your hands. Here’s what I’ve been doing with the mixed ICD-9/ICD-10 databases I’ve been dealing with, those behind the extension of our products to handle ICD-10. I keep the database structure the same, but recode the ICD-9 codes so they don’t overlap with ICD-10. (See my “False Friends” blog for details.) I put “9$” in front of ICD-9 diagnoses and “I$” in front of ICD-9 procedures, so they can live in the same places as ICD-10 — the ICD-9 becomes an extension of ICD-10 in this system. Since there are roughly 140,000 ICD-10 codes and only about 17,000 ICD-9 codes, that is only a 12% increase in the number of codes being considered. Since code types are not ever mixed at the record level, a code type indicator in the record serves as well, but may require more complicated selection logic.
Since rarely does analysis go down to the individual code level, I make sure any analysis tool I use is upgraded to handle both ICD-10 and ICD-9. For example, analysis by DRG is accomplished by feeding the ICD-9 codes to an ICD-9 grouper and the ICD-10 codes to an ICD-10 grouper, and the groupers make sure the DRGs produced are as consistent as possible. Other vendor’s tools should be available soon which do the same.
When analysis is required at the code level, the record selection (the “where” clause) is merely extended to include both ICD-9 and ICD-10 codes characterizing the patients to be included. We use, through our tools, the General Equivalence Mappings (GEMs) on the CMS web site, to help our analysts decide which ICD-9 and ICD-10 codes to use to select cases and make inferences.
In order to test our ICD-10 products, we’ve had to build a system to translate ICD-9 test cases into “equivalent” ICD-10. It took months of programming and clinical effort to produce something that worked well, and the experience taught us that the result is only useful for testing inpatient groupers and their derivatives. A “one size fits all” forward translation of ICD-9 worth using would require a massive clinical and engineering investment, with no guarantee of success and formidable legal liabilities.
Some other companies claim to have forward mappers but I’ve not seen their products in any detail. I’m skeptical but I’m not in a position to claim they haven’t pulled it off. So you might want to give them a look.
Meanwhile, hang on to your ICD-9 coded data. It ain’t ICD-10 but it’s what you’ve got.
*good. Mike wrote back that I should have said “interesting problem” instead of “good problem.” A good problem, he wrote, “is when I’m given both a Mercedes and a Lexus but I have only one space in my garage.”
And now I’ll indulge myself in some esprit d’escalier. I should have brought up the “garbage codes” argument. Coders often refer to codes with “unspecified” or “NOS” or “NEC” in them as “garbage codes”, the ones you use when the documentation doesn’t specify something it should have. (I would not suggest it may also mean “the ones you use when you’re in a hurry and you know the reimbursement system doesn’t care.”) Whereas ICD-10-CM diagnoses carry on the tradition of garbage codes from ICD-9-CM, ICD-10-PCS procedures do not. So adding the 3,838 ICD-9-CM procedures codes to the 71,918 ICD-10-PCS procedure codes, an increase of only 5%, provides a rich set of garbage codes for the era before ICD-10-PCS precision became the norm. If you have mixed ICD-9 and ICD-10 in your database and you want to cast a wide net, include the garbage codes. If your focus is narrow, include the good codes, and if your focus is so narrow that only ICD-10 can represent it, use only the good ICD-10 codes.
Ron Mills is a Software Developer for the Clinical & Economic Research department of 3M Health Information Systems.