Misusing the Reimbursement Map

April 10th, 2013 / By Ron Mills, PhD

The so-called Reimbursement Mapping is an ICD-10-code-to-ICD-9-code/cluster crosswalk available on the CMS website. It was created by taking each ICD-10-CM/PCS code and looking it up in the ICD-10–to–ICD-9 GEM. When only one ICD-9 translation was found there, it was left just as it is in the GEM entry, paired with the source ICD-10 code in the Reimbursement Map. When multiple alternatives were found, the ICD-9 alternative most frequently coded was used (based on ICD-9 Medicare data for everything but obstetrics and newborns and several years of commercial data for obstetrics and newborns).

The Reimbursement Mapping was developed by 3M under contract to CMS, in response to industry requests for a 10-to-9 crosswalk that could be used for payment. CMS did not create it for itself and has announced at every opportunity that CMS will not be using the mapping for any purpose whatsoever.

I can think of only two legitimate uses for the so-called Reimbursement Mapping:

  1. As an example of an ICD-10–to–ICD-9 mapping, illustrating that every ICD-10 code has to appear on the left side, that not every ICD-9 code will be used, that sometimes the best mapping is to multiple ICD-9 codes appearing simultaneously on the record (a cluster). Indeed, this is the purpose for which it was created—as an example of what such a crosswalk could look like.
  2. At the eleventh hour, when an organization has to continue to use an ICD-9 system and has run out of time, budget, or clued-in vendors needed to create a map appropriate to their system, they choose to cover their fannies by using this thing they got from the CMS website.

Recently, someone asked what would be the top DRG shifts if they were to use the Reimbursement Map to map incoming ICD-10 codes into ICD-9 to feed into an MS-DRG grouper. What is going on here? Mapping from 10 to 9 to get by with an ICD-9 system that won’t be converted is legitimate… but the MS-DRGs have been and will be converted… by at least five different vendors that I know of…including implementations in the public domain available cheaply through NTIS. Why would anyone need to map ICD-10 codes to get MS-DRGs?

Okay, perhaps their MS-DRG system is written in MUMPS or for the Pick OS or something and the dude who built it is like, “Hey, I’m retirin’͞I ain’t gonna deal with no ICD-10.” So they’re stuck. Well, there is still a better solution than the Reimbursement Map. I’ll even tell you how to build it from almost-free parts. Get the ICD-9 MS-DRG diagnosis table from the NTIS distribution of the ICD-9 grouper. Get the GEMs from the CMS website. Get the ICD-10 diagnosis table from its corresponding NTIS distribution. For every ICD-10 diagnosis, find via the GEMs the ICD-9 codes it might map to. Pick the one whose DRG attributes in the ICD-9 table most closely match those in the ICD-10 table. Repeat for procedures.

Here’s where I admit that I’m partly to blame for the misperception that the Reimbursement Map might be an okay solution for ICD-10 grouping. A couple of years ago, we published a paper [Mills, Ronald E., et al., Impact of the Transition to ICD-10 on Medicare Impatient Hospital Payments, Medicare & Medicaid Research Review, 2011: Volume 1, Number 2] where we modeled the financial impact on MS-DRG reimbursement of the switch to ICD-10. In that paper, we also modeled using the Reimbursement Map as an illustration of how much worse mapping was compared to conversion. Back then, we predicted a DRG shift rate of a little over 1 percent for the converted grouper but well over 3 percent for the Reimbursement Map—enough difference, we thought, to make the point that you don’t want to use it. It wasn’t reported in the paper, but we also tried a DRG-specific map built according to the recipe I gave above, and its DRG shift rate was around 2 percent.

By far the most harebrained misuse of the Reimbursement Map I’ve come across is when people try to flip it around to make an ICD-9–to–ICD-10 map. It isn’t even a map when you do that—there are lots of ICD-9 codes that are not mapped to, so when flipped, they won’t even be represented. The GEMs are not mirror images of each other—that is why we have two. “If it rains, I will carry my umbrella” does not imply that, “If I carry my umbrella, it will rain.” Mapping from ICD-9 to ICD-10 is, in our opinion, not possible—certainly not advisable. Yes, we take ICD-9 coded records and create ICD-10 coded records from them to test our software, but we never claim that the ICD-10 records so created are equivalent in any way to their ICD-9 progenitors—only that the ICD-10 record is a plausible example of how the chart that gave us the ICD-9 codes might be coded in ICD-10. For that exercise, we wouldn’t touch the Reimbursement Map with a ten-foot pole.

Ron Mills is a Software Architect for the Clinical & Economic Research department of 3M Health Information Systems.