ICD-10 financial impact update

August 28th, 2015 / By Ron Mills, PhD

Back in March, I reported at the CMS ICD-10 Coordination and Maintenance meeting that the expected financial impact of the conversion to ICD-10 for a typical Medicare inpatient case mix was -0.04% — that is, about $4 less on each $10,000 of reimbursement. I reminded the audience several times that such a tiny amount is statistically zero, since the study’s sampling error is at least 0.10%.

The report was based on several things particular to the Medicare setting in which I gave the talk:

  • It used MS-DRG version 32, but MS-DRG version 33 is now released and will be the version in use during the first year of coding and reimbursement using ICD-10-CM/PCS.
  • It used a Medicare population (MedPAR data), but some payers or states (like Oregon, where I live) use the MS-DRG grouper on a general population.
  • It did not mention APR-DRGs, which are used instead of MS-DRGs in many state systems or payer contracts.
  • It discussed the DRG shifts going from ICD-9 to ICD-10, but some organizations are contemplating using the CMS Reimbursement Map to translate claims from ICD-10 to ICD-9 after October 1, for the analysis of shifting as it occurs during that first year.

3M has developed tools to address these concerns for providers and payers interested in seeing how their populations will fare with the transition to ICD-10. Using these tools, I have updated the March results and broadened them to other groupers and populations.

At the highest level, DRG financial impact results are reported with two numbers: the percentage of cases with a DRG-shift, and the net reimbursement change percentage. A DRG-shift occurs when the DRG computed for a record coded with the submitted code set (ICD-9 or ICD-10) is different from the DRG computed for the same record after translation (to ICD-10 or ICD-9 respectively). Some shifts will be to higher weighted DRGs, some to lower weighted DRGs. By adding up the differences in the weights and dividing by the total weights for the untranslated DRGs, you get the net expected DRG weight change percentage. Since DRG reimbursement primarily depends on DRG weights (and since these studies do not take into account exceptions like outliers), we equate the net expected reimbursement change to the net expected weight change.

Both the DRG-shift percentage and the net expected reimbursement change percentage depend on the case mix of the population under study. The net expected reimbursement change percentage also depends on the set of DRG weights used. Different states and/or payers have different sets of weights depending on type of DRG, version and effective date. In choosing from scores of published weight sets or the option of uploading my own I used CMS weights for MS-DRG results and a national generic weight set for APR-DRG results.

For reference, the numbers reported in the March talk, using MS-DRGv32, a Medicare population and version 32 CMS weights were: 1.07% MS-DRG shifts, -0.04% net expected reimbursement change.

Using MS-DRGv33, the same Medicare population and version 33 CMS weights, the results are: 1.20% MS-DRG shifts, -0.08% net expected reimbursement change. The difference is due to two new MS-DRGs for hip and knee revision – some knee revisions were specified differently between ICD-9 and ICD-10.

Using the MS-DRGv33 grouper on a general population and version 33 CMS weights, the results are: 1.24% MS-DRG shifts, -0.14% net expected reimbursement change. A general population has more obstetrics cases and effective translation of obstetrics coding between ICD-9 and ICD-10 is difficult.

The APR-DRG grouper has not been changed between versions 31, 32 and 33, except for the addition of new ICD-10-PCS codes released during that time. New codes do not affect the financial impact estimation, however, since the claims data used for the study are older than the new codes. The APR-DRGv33 grouper results on a general population using a national generic APR-DRG weights are: 3.16% DRG/SOI shifts, -0.30% net expected reimbursement change. Note that an APR-DRG weight depends on both the APR-DRG and its Severity of Illness (SOI) score. APR-DRG-only shifts occur in only 1.27% of the population, but an additional 1.89% of the cases shift SOI. The APR-DRG grouper makes a much more comprehensive analysis of each case than other groupers, and hence is more sensitive to subtle differences in coding.

Going in the other direction, using the CMS Reimbursement Map to translate from ICD-10 claims to ICD-9 claims, the results for a general population are: 1.08% DRG shifts, -0.26% net expected reimbursement change. Since ICD-10 is more specific than ICD-9, you would expect that very accurate translation from ICD-10 to ICD-9 should be possible. It may be, but the Reimbursement Map falls short of perfection in three ways:

  • It does not take into account “input clusters” where two or more ICD-10 codes team up to create one ICD-9 “combination” code.
  • It does not re-entangle diagnosis and procedure coding for obstetrics and rehabilitation. ICD-9 has procedure information in diagnoses and diagnosis information in procedures, which ICD-10 does not.
  • It does not re-order diagnoses for anemia and some other kinds of cases where coding guidelines have changed.

Is the 1.08% DRG shift going from ICD-10 to ICD-9 just reversing most of the 1.20% DRG shift going from ICD-9 to ICD-10? If so, the Reimbursement Map would be handy for identifying those ICD-10 coded cases that might have been different had they been coded in ICD-9. To test this, we took a few million ICD-9 coded cases, grouped them with MS-DRGv33, translated them to ICD-10, translated them back to ICD-9 using the Reimbursement Map, grouped them again and compared the two ICD-9 MS-DRGs. The results: 2.03% DRG shifts, -0.36% net expected reimbursement change. The answer is mostly “no”: the Reimbursement Map is adding its own shift issues, not merely reversing those derived from going from ICD-9 to ICD-10.

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


Need more in depth ICD-10 physician training?  Find the tools and resources you need to succeed.