npi

Access the U.S. National Provider Identifier Registry API

Status at rOpenSci Software Peer Review Project Status: Active – The project has reached a stable, usable state and is being actively developed. lifecycle R-CMD-check Codecov test coverage DOI

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:

  1. Install from CRAN:
install.packages("npi")
library(npi)
  1. Install from R-universe:
install.packages("npi", repos = "https://ropensci.r-universe.dev")
library(npi)
  1. Install from GitHub using the devtools package:
devtools::install_github("ropensci/npi")
library(npi)

Usage

npi exports four functions, all of which match the pattern “npi_*“:

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:

nyc <- npi_search(city = "New York City")
# Your results may differ since the data in the NPPES database changes over time
nyc
#> # A tibble: 10 × 11
#>       npi enume…¹ basic    other_…² identi…³ taxono…⁴ addres…⁵ practi…⁶ endpoi…⁷
#>  *  <int> <chr>   <list>   <list>   <list>   <list>   <list>   <list>   <list>  
#>  1 1.19e9 Indivi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#>  2 1.31e9 Indivi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#>  3 1.64e9 Indivi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#>  4 1.35e9 Indivi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#>  5 1.56e9 Indivi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#>  6 1.79e9 Indivi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#>  7 1.56e9 Indivi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#>  8 1.96e9 Organi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#>  9 1.43e9 Indivi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#> 10 1.33e9 Indivi… <tibble> <tibble> <tibble> <tibble> <tibble> <tibble> <tibble>
#> # … with 2 more variables: created_date <dttm>, last_updated_date <dttm>, and
#> #   abbreviated variable names ¹​enumeration_type, ²​other_names, ³​identifiers,
#> #   ⁴​taxonomies, ⁵​addresses, ⁶​practice_locations, ⁷​endpoints

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:

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):

npi_summarize(nyc)
#> # A tibble: 10 × 6
#>           npi name                                 enume…¹ prima…² phone prima…³
#>         <int> <chr>                                <chr>   <chr>   <chr> <chr>  
#>  1 1194276360 ALYSSA COWNAN                        Indivi… 5 E 98… 212-… Physic…
#>  2 1306849641 MARK MOHRMANN                        Indivi… 16 PAR… 212-… Orthop…
#>  3 1639173065 SAKSHI DUA                           Indivi… 10 E 1… 212-… Nurse …
#>  4 1346604592 SARAH LOWRY                          Indivi… 1335 D… 614-… Occupa…
#>  5 1558362566 AMY TIERSTEN                         Indivi… 1176 5… 212-… Psychi…
#>  6 1790786416 NOAH GOLDMAN                         Indivi… 140 BE… 973-… Intern…
#>  7 1558713628 ROBYN NOHLING                        Indivi… 9 HOPE… 781-… Nurse …
#>  8 1962983775 LENOX HILL MEDICAL ANESTHESIOLOGY, … Organi… 100 E … 212-… Intern…
#>  9 1427454529 YONGHONG TAN                         Indivi… 34 MAP… 203-… Obstet…
#> 10 1326403213 RAJEE KRAUSE                         Indivi… 12401 … 347-… Nurse …
#> # … with abbreviated variable names ¹​enumeration_type,
#> #   ²​primary_practice_address, ³​primary_taxonomy

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:

npi_flatten() supports a variety of approaches to flattening the results from npi_search(). One extreme is to flatten everything at once:

npi_flatten(nyc)
#> # A tibble: 48 × 42
#>           npi basic_fi…¹ basic…² basic…³ basic…⁴ basic…⁵ basic…⁶ basic…⁷ basic…⁸
#>         <int> <chr>      <chr>   <chr>   <chr>   <chr>   <chr>   <chr>   <chr>  
#>  1 1194276360 ALYSSA     COWNAN  PA      NO      F       2016-1… 2018-0… A      
#>  2 1194276360 ALYSSA     COWNAN  PA      NO      F       2016-1… 2018-0… A      
#>  3 1306849641 MARK       MOHRMA… MD      NO      M       2005-0… 2019-0… A      
#>  4 1306849641 MARK       MOHRMA… MD      NO      M       2005-0… 2019-0… A      
#>  5 1306849641 MARK       MOHRMA… MD      NO      M       2005-0… 2019-0… A      
#>  6 1306849641 MARK       MOHRMA… MD      NO      M       2005-0… 2019-0… A      
#>  7 1326403213 RAJEE      KRAUSE  AGPCNP… NO      F       2015-1… 2019-0… A      
#>  8 1326403213 RAJEE      KRAUSE  AGPCNP… NO      F       2015-1… 2019-0… A      
#>  9 1326403213 RAJEE      KRAUSE  AGPCNP… NO      F       2015-1… 2019-0… A      
#> 10 1326403213 RAJEE      KRAUSE  AGPCNP… NO      F       2015-1… 2019-0… A      
#> # … with 38 more rows, 33 more variables: basic_name <chr>,
#> #   basic_name_prefix <chr>, basic_middle_name <chr>,
#> #   basic_organization_name <chr>, basic_organizational_subpart <chr>,
#> #   basic_authorized_official_credential <chr>,
#> #   basic_authorized_official_first_name <chr>,
#> #   basic_authorized_official_last_name <chr>,
#> #   basic_authorized_official_middle_name <chr>, …

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:

# Flatten basic provider info and provider taxonomy, preserving the relationship
# of each to NPI number and discarding other list columns.
npi_flatten(nyc, cols = c("basic", "taxonomies"))
#> # A tibble: 20 × 26
#>           npi basic_fi…¹ basic…² basic…³ basic…⁴ basic…⁵ basic…⁶ basic…⁷ basic…⁸
#>         <int> <chr>      <chr>   <chr>   <chr>   <chr>   <chr>   <chr>   <chr>  
#>  1 1194276360 ALYSSA     COWNAN  PA      NO      F       2016-1… 2018-0… A      
#>  2 1306849641 MARK       MOHRMA… MD      NO      M       2005-0… 2019-0… A      
#>  3 1306849641 MARK       MOHRMA… MD      NO      M       2005-0… 2019-0… A      
#>  4 1326403213 RAJEE      KRAUSE  AGPCNP… NO      F       2015-1… 2019-0… A      
#>  5 1326403213 RAJEE      KRAUSE  AGPCNP… NO      F       2015-1… 2019-0… A      
#>  6 1326403213 RAJEE      KRAUSE  AGPCNP… NO      F       2015-1… 2019-0… A      
#>  7 1346604592 SARAH      LOWRY   OTR/L   YES     F       2016-0… 2018-0… A      
#>  8 1346604592 SARAH      LOWRY   OTR/L   YES     F       2016-0… 2018-0… A      
#>  9 1427454529 YONGHONG   TAN     <NA>    NO      F       2014-1… 2018-1… A      
#> 10 1558362566 AMY        TIERST… M.D.    YES     F       2005-0… 2019-0… A      
#> 11 1558713628 ROBYN      NOHLING FNP-BC… YES     F       2016-0… 2018-0… A      
#> 12 1558713628 ROBYN      NOHLING FNP-BC… YES     F       2016-0… 2018-0… A      
#> 13 1558713628 ROBYN      NOHLING FNP-BC… YES     F       2016-0… 2018-0… A      
#> 14 1558713628 ROBYN      NOHLING FNP-BC… YES     F       2016-0… 2018-0… A      
#> 15 1558713628 ROBYN      NOHLING FNP-BC… YES     F       2016-0… 2018-0… A      
#> 16 1558713628 ROBYN      NOHLING FNP-BC… YES     F       2016-0… 2018-0… A      
#> 17 1639173065 SAKSHI     DUA     M.D.    YES     F       2005-0… 2019-0… A      
#> 18 1639173065 SAKSHI     DUA     M.D.    YES     F       2005-0… 2019-0… A      
#> 19 1790786416 NOAH       GOLDMAN M.D.    NO      M       2005-0… 2018-0… A      
#> 20 1962983775 <NA>       <NA>    <NA>    <NA>    <NA>    2018-0… 2018-0… A      
#> # … with 17 more variables: basic_name <chr>, basic_name_prefix <chr>,
#> #   basic_middle_name <chr>, basic_organization_name <chr>,
#> #   basic_organizational_subpart <chr>,
#> #   basic_authorized_official_credential <chr>,
#> #   basic_authorized_official_first_name <chr>,
#> #   basic_authorized_official_last_name <chr>,
#> #   basic_authorized_official_middle_name <chr>, …

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:

# Validate NPIs
npi_is_valid(1234567893)
#> [1] TRUE
npi_is_valid(1234567898)
#> [1] FALSE

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:

options(npi_user_agent = "my_app/1.0.0")

Package Website

npi has a website with release notes, documentation on all user functions, and examples showing how the package can be used.

Reporting Bugs

Did you spot a bug? I’d love to hear about it at the issues page.

Code of Conduct

Please note that this package is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.

Contributing

Interested in learning how you can contribute to npi? Head over to the contributor guide—and thanks for considering!

How to cite this package

For the latest citation, see the Authors and Citation page on the package website.

License

MIT (c) Frank Farach

This package’s logo is licensed under CC BY-SA 4.0 and co-created by Frank Farach and Sam Parmar. The logo uses a modified version of an image of the Rod of Asclepius and a magnifying glass that is attributed to Evanherk, GFDL.