HERC - Resources - FAQ - Finding and Using Healthcare Data

I. Finding and Using Healthcare Data

6. How do you Crosswalk HCUP to the Medicare PPS report?

Researchers may wish to link the HCUP database to Medicare data. Link to details on HCUP. HCUP identifies hospitals with HOSPID while Medicare uses the PPS identifying number. For our purpose, we wanted to link the two so that we could cost-adjust the HCUP charges.

First, note that the crosswalk between HCUP and Medicare is not perfect. Each dataset (HCUP, AHA, and Medicare cost report) has a different method for identifying hospitals. In one dataset, a hospital system may be counted as a single entity, whereas in another system, it may be counted as multiple, separate entities. This can lead to discrepancies.

Below are the detailed steps involved in crosswalking the data.

Step 1: Link HCUP to AHA

HCUP provides a fairly good link between the HCUP hospital identification number (HOSPID) and AHA identification number (AHAID).There were 906 HCUP hospitals. A total of 755 (83.3%) of these hospitals had AHAID, HOSPITAL NAME, HOSPITAL STATE, or HOSPITAL ZIP CODE data. Without these hospital identifiers, it is impossible to do the crosswalk.

Step 2: Link HCUP to Medicare using AHA identification number (AHAID)

In 1995, the Agency for Health Research and Quality crosswalked the HCUP to Medicare cost report. AHRQ developed a crosswalk file using 1993 AHA files as an intermediate step. Although slightly out of date, this file is still helpful for making the crosswalk today. This file matched 604 of the 755 hospitals (80.0%). For these 604 cases, the PPS field was populated. Details on the AHA survey can be found at: http://www.hospitalconnect.com/healthforum/hfstats/datasources.html

Step 3: Link HCUP to the MCR using name and zip code

HCFA has a provider file (Provider of services listing). This file has detailed information on each hospital in 1999 ( http://cms.hhs.gov/data/download/default.asp). This file is updated quarterly and the June 1999 list was used for the match.

First, this file was matched to HCUP using exact name. If there was a match and the PPS identifier was missing from step 2, then the missing PPS identifier was replaced with the PPS identifier in the HCFA file.

Second, the file was matched using the first 5 characters of the hospital name and hospital zip code. If there was a match and the PPS identifier was missing from prior steps, then the missing PPS value was replaced with the PPS value in the HCFA file.

Third, this file was matched to HCUP using the first 5 characters of the name and the first three numbers in the zip code. If there was a match and the PPS value was missing from prior steps, then it replaced the missing PPS value with the PPS value in the HCFA file.

Results of the matching:

From Step 2, we had PPS numbers for 604/755 cases (80.0%).
After Step 3, we had PPS numbers for 720/755 cases (95.4%).
Of the 720 hospitals, 104 had 2 valid PPS numbers. This means that HCUP and Medicare had different methods for coding hospitals. As mentioned earlier, there is no perfect crosswalk.

Step 4: Link HCUP to Medicare cost reports using the PPS identifier Generating a dataset from the PPS reports

Pull variables (fields): PPS number (1), MSA (72), charges (2135), costs (2138) for PPS reports #13, #12, and #11.
Start with Medicare cost report version 13, I merged it to the Medicare cost report version 12, and then to the Medicare cost report version 11, only new cases. This resulted in a file with information on 6,594 hospitals. Approximately 90% of these hospitals existed in all three Medicare cost reports.

Step 5: Linking HCUP with PPS dataset

The crosswalk file generated in Step 3 was then merged with the file generated in Step 4. This match was successful in 688 out of a possible 720 (95.6%) hospitals. In other words, this match represents a 91.1% match with the 755 HCUP hospitals that had identifying information and a 76.0% match with all 906 HCUP hospitals in the 1996 HCUP files.

The final crosswalk file is called XWALK. It was created predominantly using Stata 6 and SAS 6.12 (UNIX). It can be obtained from HERC. See http://www.stattransfer.com for transferring this dataset between formats. Summary: With the HCUP to Medicare crosswalk file, it is easy to link information with the Medicare data to the HCUP data. While such a crosswalk is feasible, remember that an exact 1 to 1 crosswalk does not exist.