Access the U.S. National Provider Identifier Registry API
Use R to access the U.S. National Provider Identifier (NPI) Registry API (v2.1) by the Center for Medicare and Medicaid Services (CMS): https://npiregistry.cms.hhs.gov/. Obtain rich administrative data linked to a specific individual or organizational healthcare provider, or perform advanced searches based on provider name, location, type of service, credentials, and many other attributes.
npi
provides convenience functions for data extraction so you can spend less time wrangling data and more time putting data to work.
Analysts working with healthcare and public health data frequently need to join data from multiple sources to answer their business or research questions. Unfortunately, joining data in healthcare is hard because so few entities have unique, consistent identifiers across organizational boundaries.
NPI numbers, however, do not suffer from these limitations, as all U.S. providers meeting certain common criteria must have an NPI number in order to be reimbursed for the services they provide.
This makes NPI numbers incredibly useful for joining multiple datasets by provider, which is the primary motivation for developing this package.
Installation
There are three ways to install the npi
package:
- Install from CRAN:
- Install from R-universe:
- Install from GitHub using the
devtools
package:
Usage
npi
exports four functions, all of which match the pattern “npi_*“:
npi_search()
: Search the NPI Registry and return the response as a tibble with high-cardinality data organized into list columns.
npi_summarize()
: A method for displaying a nice overview of results from npi_search()
.
npi_flatten()
: A method for flattening one or more list columns from a search result, joined by NPI number.
npi_is_valid()
: Check the validity of one or more NPI numbers using the official NPI enumeration standard.
Search the registry
npi_search()
exposes nearly all of the NPPES API’s search parameters. Let’s say we wanted to find up to 10 providers with primary locations in New York City:
The full search results have four regular vector columns, npi
, enumeration_type
, created_date
, and last_updated_date
and seven list columns. Each list column is a collection of related data:
basic
: Basic profile information about the provider
other_names
: Other names used by the provider
identifiers
: Other provider identifiers and credential information
taxonomies
: Service classification and license information
addresses
: Location and mailing address information
practice_locations
: Provider’s practice locations
endpoints
: Details about provider’s endpoints for health information exchange
A full list of the possible fields within these list columns can be found on the NPPES API Help page.
If you’re comfortable working with list columns, this may be all you need from the package. However, npi
also provides functions that can help you summarize and transform your search results.
Working with search results
npi
has two main helper functions for working with search results: npi_summarize()
and npi_flatten()
.
Summarizing results
Run npi_summarize()
on your results to see a more human-readable overview of your search results. Specifically, the function returns the NPI number, provider’s name, enumeration type (individual or organizational provider), primary address, phone number, and primary taxonomy (area of practice):
Flattening results
As seen above, the data frame returned by npi_search()
has a nested structure. Although all the data in a single row relates to one NPI, each list column contains a list of one or more values corresponding to the NPI for that row. For example, a provider’s NPI record may have multiple associated addresses, phone numbers, taxonomies, and other attributes, all of which live in the same row of the data frame.
Because nested structures can be a little tricky to work with, the npi
includes npi_flatten()
, a function that transforms the data frame into a flatter (i.e., unnested and merged) structure that’s easier to use. npi_flatten()
performs the following transformations:
- unnest the list columns
- prefix the name of each unnested column with the name of its original list column
- left-join the data together by NPI
npi_flatten()
supports a variety of approaches to flattening the results from npi_search()
. One extreme is to flatten everything at once:
However, due to the number of fields and the large number of potential combinations of values, this approach is best suited to small datasets. More likely, you’ll want to flatten a small number of list columns from the original data frame in one pass, repeating the process with other list columns you want and merging after the fact. For example, to flatten basic provider and provider taxonomy information, supply the corresponding list columns as a vector of names to the cols
argument:
Validating NPIs
Just like credit card numbers, NPI numbers can be mistyped or corrupted in transit. Likewise, officially-issued NPI numbers have a check digit for error-checking purposes. Use npi_is_valid()
to check whether an NPI number you’ve encountered is validly constructed:
Note that this function doesn’t check whether the NPI numbers are activated or deactivated (see #22). It merely checks for the number’s consistency with the NPI specification. As such, it can help you detect and handle data quality issues early.
Set your own user agent
A user agent is a way for the software interacting with an API to tell it who or what is making the request. This helps the API’s maintainers understand what systems are using the API. By default, when npi
makes a request to the NPPES API, the request header references the name of the package and the URL for the repository (e.g., ‘npi/0.2.0 (https://github.com/ropensci/npi)’). If you want to set a custom user agent, update the value of the npi_user_agent
option. For example, for version 1.0.0 of an app called “my_app”, you could run the following code:
Package Website
npi
has a website with release notes, documentation on all user functions, and examples showing how the package can be used.