CareSet Dataset: MrPUP

MrPUP Dataset Documentation

The Medicare Referring Provider Utilization for Procedures (MrPUP) dataset details the healthcare procedures that Medicare providers referred in the outpatient setting on a per-year basis, and is particularly useful in understanding how specific doctors use blood tests, CT scans, and MRIs to diagnose patients.

To purchase or download the MrPUP dataset, please go to the CareSet Data Store.

MrPUP Summary Statistics

  • Total Row Count – 7,404,936
  • Distinct Referring NPI Count – 616,606
  • Distinct HCPCS Code Count – 4,238
  • Max Patient Count – 10,158
  • Min Patient Count – 11
  • Transaction Line Service Total – 648,797,587

MrPUP Data Fields

  • Carrier_claim_referring_physician_npi – Referring provider National Provider Identifier (NPI)
  • Line_center_healthcare_common_procedure_coding_system – Healthcare Common Procedure Coding System (HCPCS) code for the referring NPI
  • Patient_count – Count of unique beneficiaries for the referring NPI and HCPCS code
  • Transaction_line_service_total – Total number of transactions for the referring NPI, HCPCS code and patient count


What kinds of procedures are represented in MrPUP ?

The dataset is especially rich when considering laboratory and imaging studies, because Medicare will not pay for many of these procedures unless the referring NPI field is filled out. For these procedures one can expect to have pretty good coverage of actual referring NPIs.

However, there is other data that appears here, based on the fact that the referring NPI field can always be filled in.


Is the data coverage duplicative with HOP?

Strictly speaking, not at all. All versions of the HOP teaming algorithm explicitly exclude consideration of the referral NPI field, in order to avoid double counting relationships between two providers. However, it is typical for provider A (referring) to provider B (performing) on one claim to preceded by a different claim where provider A was the billing NPI.

There are a few circumstances where this will not happen however. First, it is possible for a provider to be a “referring” provider for Medicare but not actually bill CMS directly. Second, it is possible that the earlier claim does in fact exist, but is on the other side of the Dec 31st/Jan 1st year cutoff date used for most of our datasets. There are also likely to be some rare cases of fraud where the performing provider arbitrarily listed a referring NPI for a provider who “seemed plausible”.

How to normalize the data in MrPUP per provider?

The rates of referral vary wildly in the MrPUP data file. This is largely driven by the number of Medicare patients that a provider is treating. It is very different thing for a provider to be referring 15 of procedure X if that provider is treating 1000 Medicare patients vs 100 Medicare patients.

There is no easy way to get data to understand what the “payor mix” is for a given patient. However, it is possible to know how many Medicare patients a provider is treating per year using the per-provider summary portion of the Medicare summary PUF. You can find this data with the standard medicare utilization data.

Here is a direct link to the Socrata dashboard for the 2016 Medicare Physician and Other Supplier National Provider Identifier (NPI) Aggregate Report, the variable you will want to attend to is the “Number of Medicare Beneficiaries”. Using this you can calculate on a per-provider basis the percentage of their patients that are being referred for a given procedure.