7 Provider Data
A claim can name the clinician who performed a service, the organization that billed for it, the facility where it happened, and the clinician who ordered it. All of them are providers, but they play different roles. Choosing the wrong one changes the answer to questions about cost, referral patterns, access, and quality.
In this chapter, we’ll connect the identifiers in claims to information about clinicians and organizations. We’ll also separate what a provider directory tells us from what we still need to establish about specialty, location, and affiliation.
7.1 Start with the provider’s role
Before looking up a provider, decide which relationship matters to the question. A generic provider_id column hides distinctions that are often central to the analysis.
| Provider role | What it represents | Example use |
|---|---|---|
| Billing | The individual or organization billing for the service | Spending by billing practice |
| Rendering | The clinician or other provider identified as performing the service | Services performed by clinician |
| Attending | The clinician identified as responsible for the care reported on the institutional claim or encounter, including outpatient care | Analysis of attending responsibility |
| Operating | The clinician identified as performing an operation | Surgical practice patterns |
| Referring or ordering | The provider who referred the patient or ordered the service | Referral and ordering patterns |
| Service facility | The location or entity identified for where care occurred | Site-of-care analysis |
The available roles vary by claim type and extract. A role may appear on the header, the line, or both. Preserve these distinctions when mapping the source. CMS’s institutional billing instructions define the attending and operating roles. The Key Data Elements chapter places provider fields alongside the other information needed to interpret a claim.
For example, a physician can render services while a group practice bills for them. Grouping spending by the billing provider answers a different question from grouping it by the rendering clinician. Neither field automatically identifies the clinician’s employer or the organization responsible under a value-based care agreement.
7.2 What an NPI identifies
The National Provider Identifier, or NPI, is a 10-digit identifier used for healthcare providers in standard administrative transactions. CMS assigns NPIs through the National Plan and Provider Enumeration System, usually called NPPES. The digits do not encode a provider’s specialty or geography. Store the identifier as text and join it to reference data to learn about the provider. CMS’s NPI overview describes the standard.
There are two entity types:
- Type 1 identifies an individual, such as a physician or nurse practitioner. An individual generally retains the same NPI across jobs and practice locations.
- Type 2 identifies an organization, such as a group practice, hospital, or supplier. Organizational subparts can have their own NPIs, so a health system may be associated with many NPIs.
An incorporated clinician can have an individual NPI and a separate organizational NPI for the corporation. These represent different entities. Likewise, a billing NPI is not necessarily Type 2: an individual can bill in their own name. The CMS identifier FAQs explain entity types and subparts.
An NPI is not evidence that a provider is licensed, credentialed, in network, or currently practicing at a particular address. CMS explicitly cautions that NPI issuance does not validate licensure or credentialing. Use the relevant licensing, credentialing, or network source when the analysis depends on one of those facts.
7.3 Working with NPPES
NPPES is a useful starting point because it gives us a common identifier and public information about individual and organizational providers. Its records include names, entity type, addresses, taxonomy codes, and administrative dates.
CMS offers the NPI Registry for lookups and downloadable files for population-scale work. The NPPES download page publishes monthly files, weekly incremental updates, and deactivation information. Companion files include additional organizational names and non-primary practice locations. Use the current file documentation when building an import; layouts can change.
Three modeling decisions make this data easier to use.
First, keep one provider identity record per NPI, with repeated taxonomies and locations in related tables. Joining every taxonomy directly to every claim line can multiply spending. If a provider has three taxonomies, a careless join turns one $100 line into three rows totaling $300.
Second, keep the source and observation date of directory attributes. A current practice address is not proof of where a service occurred three years ago. Saving successive directory snapshots can show when an attribute changed in the data, but the first time we observe an address is not necessarily the date the provider moved there.
Third, measure match coverage. Count claims with missing NPIs, invalid formats, and NPIs absent from the reference file. Report the associated spending as well as the row count. An unmatched provider on a few expensive institutional claims can matter more than thousands of unmatched low-cost lines.
NPPES is maintained through provider submissions and updates. Treat its attributes as reported directory information. Reconcile conflicting records with source-specific evidence instead of assuming that whichever file was downloaded most recently must describe the service date best.
7.4 Taxonomy and specialty
The Health Care Provider Taxonomy code set classifies provider types, classifications, and areas of specialization. Codes are ten-character alphanumeric identifiers maintained by the National Uniform Claim Committee, or NUCC. A provider can report multiple taxonomies and identify a primary one.
Taxonomy is self-selected. It does not establish board certification, and it does not describe every service a provider performs. NUCC explains how providers select codes and the code set’s scope in its taxonomy guidance, which also links to the code lookup.
For analytics, choose a specialty definition that fits the question. If the question is how much was billed by providers reporting a primary care taxonomy, NPPES may be sufficient. If the question is whether a clinician qualifies for a particular contract or attribution method, use that method’s specialty rules and supporting data.
Keep the reported taxonomy separate from a derived analytic specialty. A mapping that groups several taxonomies into “primary care” is an analytical choice. Preserve the original codes, the mapping version, and how multiple taxonomies were resolved so that someone else can reproduce the result.
7.5 Organizations, locations, and affiliations
An organization, a billing entity, and a physical location are different things. One practice may operate at several addresses under the same billing NPI. One building may contain several unrelated practices. A hospital campus can generate claims under multiple NPIs.
Suppose two clinicians share a mailing address in NPPES. That may be a centralized administrative office. It does not prove that they work at the same clinic. Conversely, clinicians with different mailing addresses may belong to the same organization. Address matching can generate candidates for review, but it should not silently become an ownership rule.
For a facility analysis, use the service location recorded on the claim when available and assess how reliably it identifies the site. For a group or health system analysis, maintain a relationship table that records the provider, related organization, relationship type, source, and applicable dates. Keep contractual participation, employment, ownership, and observed billing relationships distinct.
This matters during acquisitions and practice changes. Applying today’s group roster to five years of claims can make historical spending appear to move between organizations even though the care itself has not changed.
7.6 Other provider data sources
NPPES is not the only source, and combining sources is often necessary.
Medicare enrollment data. PECOS, the Provider Enrollment, Chain, and Ownership System, supports Medicare provider enrollment. Although the full administrative system is not a public directory, CMS publishes selected information through the Medicare Fee-for-Service Public Provider Enrollment files. These include enrollment and relationship information for the population described in the public-file methodology. They describe currently approved enrollments at the extract date, not a complete history of participation. They are useful for Medicare enrollment questions; they are not a complete roster of every provider in every payer’s network.
Payer and program rosters. These can establish participation in a network, group, or care arrangement for a specified period. Confirm what participation means and whether the roster describes current or historical status. A claim paid at an in-network rate and a provider listed on today’s roster are different pieces of evidence.
Licensing and commercial reference data. State licensing authorities and licensed provider datasets can supply details that a public enumeration file does not establish. Evaluate their covered population, update process, historical depth, and permitted uses. “Available for research” does not necessarily mean openly downloadable or freely redistributable.
7.7 Where tax identifiers fit
A Taxpayer Identification Number, or TIN, identifies a taxpayer. Claims extracts may include a billing tax identifier, which can help connect claims to a contractual or financial entity. It can represent an organization or, in some cases, an individual. Handle it as sensitive data and use a protected internal key in shared analytical outputs.
An NPI and a TIN are not interchangeable. Several clinicians may bill under one TIN, and a clinician may have billing relationships with multiple TINs over time. Network contracts and performance arrangements also differ in the entities they recognize. Use the contract’s actual definition instead of assuming every agreement operates at the TIN level.
The useful result is a provider model that preserves identity, role, and relationships separately. Then “spending by provider” becomes a question we can define precisely: spending for which services, attributed to which role, grouped under which organization, during which period?