Under the Streetlight

September 13th, 2013 / By Ron Mills, PhD

Towards midnight, a cop comes upon a guy crawling around on his hands and knees under a streetlight. The cop asks, “What are you doing?” “Looking for my car keys.” The cop thinks maybe he can help, so he asks, “Exactly where were you when you dropped them?” The guy looks up, considers and points into the dark, halfway down the block. “Somewhere there.” The cop asks, “Then why are you looking here?” Answer: “The light is better here.”

We’ve developed a tool that lets a payer or provider input the number of claims they submitted or processed under MS-DRGs or APR-DRGs in one year. The tool then computes estimates of the changes in reimbursement that might occur if those claims were coded in ICD-10. It also provides a table by DRG of all estimated shifts to other DRGs under ICD-10, their probability of occurrence, and projected reimbursement change. For some of those shifts, we provide a clinical explanation of the coding reason behind the shift and, when possible, improved ICD-10 coding practices that would avoid the shift.

We were asked why this table did not contain explanations for likely shifts. Why, for example, there was no explanation for the 2.71% of DRG 11 (tracheostomy for face, mouth & neck diagnoses w MCC) claims that were expected to shift to DRG 3(ECMO or tracheostomy with mechanical ventilation for 96+ hours or PDX excluding face, mouth & neck w major O.R), with a decrease in reimbursement around $25,000 per claim.

Here’s the process we used. We took a database of 14 million real patient records and ran them through the ICD-9 and ICD-10 groupers. We noted the DRG shifts, sorting them by decreasing order of frequency. That is, frequency in the entire 14 million records — not frequency within the DRG (the 2.71% above) but frequency in the entire population – 2.71% times the percentage of DRG 11 (tiny, tiny) in the entire population. Then we took the top five shifts and analyzed them — a process that takes a team of coding and grouper experts from a few minutes to a few hours each shift. Once analyzed, we either corrected the grouper (rarely) so that the shift did not occur, improved the programs that created the 9-to-10 translated records (slightly less rarely) so the ICD-10 records are more like what we expect coders to code, or (usually) wrote an explanation for the shift. Then we recomputed and did likewise for the next five.

We did this until the highest number of unexplained DRG shifts were 200 in a population of 14 million. We fixed or explained shifts up to this point, stopping because beyond that we would only be improving the unexplained shifting by about 0.001 percent. By that time we had explained 70% of the MS-DRG shifts by volume. This is much less than 70% of the different kinds of MS-DRG shifts, but represents 70% of what a typical hospital might observe. Then we applied what we had learned for MS-DRGs to APR-DRGs.

So, working the DRG 11 example: There were only 1,700 claims in the 14 million record database for this DRG. Hence, the probability of seeing the shift is (1,700/14,000,000) times (.0271) = .0000033. A hospital with 10,000 patients a year and an average case mix should expect to see this shift about once every 30 years.

DRG 11 and others like it may be under the streetlight, but the keys are with the DRGs you actually expect to provide starting on October 1, 2014.

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