ICU care at the end of life: Limitations of MEDPAR data

Research Objective

Using MEDPAR data, Wennberg et al documented in separate national and California hospital studies large variations in resource use among chronically ill patients in their last six months of life. With performance improvement as its objective, management at Cedars-Sinai Medical Center (CSMC) initiated a data validation effort to confirm the population and performance attributed to CSMC.

Study Design

We obtained from California's QIO Lumetra the CSMC MEDPAR patient records from 1999 to 2003. Employing a matching algorithm that included last name, first name, admit and discharge dates, and principal diagnosis, we matched to our local administrative data over 99% of the hospitalizations. Because not all attributed patients died in CSMC, we focused our validation on patients who expired here.

Principal Findings

Focusing on the death discharge, we found differences in average ICU days between MEDPAR and our local database. Overall, year to year variance ranged from 0.2 day to 1.7 days (p=0.005). Specifically, in 1999, the MEDPAR data had 5.6 ICU days per discharge while the local data had an average of 3.9 ICU days, a 30% difference (p=0.001). Further examination revealed that the MEDPAR data included days from “Post ICU or Step-Down” UB-92 revenue code 206. In 2000, the difference was only 0.3 day (MEDPAR, 3.8 days and Local, 3.5 days), or 8% (p=0.53). The smaller difference was due to the exclusion of UB-92 Code 210 “General CCU” in the MEDPAR count. This exclusion had an offsetting effect on the inclusion of Code 206. These data issues also impacted the ICU days reporting in the last 6-months of life. We found statistically significant differences in total ICU days in each of the five years (p<0.0001). The most significant difference was observed in 1999, a 4.4 days or 45% difference (p<0.0001) with MEDPAR reporting 9.6 days and local data showing 5.2 days. Similarly, an offsetting impact was observed in year 2000 when there was a 1.9 days difference (p=0.01).

Conclusions

Our study found that the use of MEDPAR data for ICU days analysis is problematic on two fronts. First, MEDPAR inclusion of “Step-Down” activities implies that “Step-Down” resource use is equivalent to that in a general ICU, an implication not supportable clinically or financially. In hospitals with sizable “Step-Down” activities such as CSMC, this will overstate their ICU resource use. Second, year-to-year variation on the construction of this variable in MEDPAR appears to be inconsistent.