125

icd

Comorbidity Calculations and Tools for ICD-9 and ICD-10 Codes

Calculate comorbidities, Charlson and van Walraven scores, perform fast and accurate validation, conversion, manipulation, filtering and comparison of ICD-9 and ICD-10 codes. This package enables a work flow from raw lists of ICD codes in hospital databases to comorbidities. ICD-9 and ICD-10 comorbidity mappings from Quan (Deyo and Elixhauser versions), Elixhauser and AHRQ included. Common ambiguities and code formats are handled.

CRAN GitHub Travis Appveyor codecov.io CII BestPractices CRAN RStudio mirror downloads last calendarmonth CRAN RStudio mirror downloads lastweek

Comorbidities from ICD-9 and ICD-10 codes, manipulation and validation

Introduction

Calculate comorbidities, Charlson and van Walraven scores, perform fast and accurate validation, conversion, manipulation, filtering and comparison of ICD-9 and ICD-10 codes. This package enables a work flow from raw lists of ICD codes in hospital databases to comorbidities. ICD-9 and ICD-10 comorbidity mappings from Quan (Deyo and Elixhauser versions), Elixhauser and AHRQ included. Common ambiguities and code formats are handled.

icd is used by many researchers around the world who work in public health, epidemiology, clinical research, nutrition, journalism, health administration and more. I’m grateful for contact from people in these fields for their feedback and code contributions, and I’m pleased to say that icd has been used in works like the Pulitzer finalist work on maternal death by ProPublica.

Features

  • find comorbidities of patients based on ICD-9 or ICD-10 codes, e.g. Cancer, Heart Disease
    • several standard mappings of ICD codes to comorbidities are included (Quan, Deyo, Elixhauser, AHRQ, PCCC)
    • very fast assignment of ICD codes to comorbidities (using novel matrix multiplication algorithm and C++ internally)
  • use your existing data format, minimizing requirements for pre-processing
  • summarize groups of ICD codes in natural language
  • Charlson and Van Walraven score calculations
  • Hierarchical Condition Codes (HCC) from CMS
  • Clinical Classifcations Software (CCS) comorbidities from AHRQ
  • Pediatric Complex Chronic Condition comorbidities
  • AHRQ ICD-10 procedure code classification
  • annual revisions of ICD-9-CM and ICD-10-CM
  • correct conversion between different representations of ICD codes, with and without a decimal points, leading and trailing characters (this is not trivial for ICD-9-CM). ICD-9 to ICD-10 cross-walk is not yet implemented
  • comprehensive test suite to increase confidence in accurate processing of ICD codes
  • all internal ICD and comorbidity data is extracted directly from public data or code, allowing end-to-end reproducibility
  • used, tested and benchmarked against other comorbidity calculators on hardware from laptops to big servers

Examples

See also the vignettes and examples embedded in the help for each function for more. Here’s a taste:

library(icd)
 
# Typical diagnostic code data, with many-to-many relationship
patient_data
#>   visit_id  icd9
#> 1     1000 40201
#> 2     1000  2258
#> 3     1000  7208
#> 4     1000 25001
#> 5     1001 34400
#> 6     1001  4011
#> 7     1002  4011
#> 8     1000  <NA>
 
# get comorbidities using Quan's application of Deyo's Charlson comorbidity groups
comorbid_charlson(patient_data)
#>         MI   CHF   PVD Stroke Dementia Pulmonary Rheumatic   PUD LiverMild
#> 1000 FALSE  TRUE FALSE  FALSE    FALSE     FALSE     FALSE FALSE     FALSE
#> 1001 FALSE FALSE FALSE  FALSE    FALSE     FALSE     FALSE FALSE     FALSE
#> 1002 FALSE FALSE FALSE  FALSE    FALSE     FALSE     FALSE FALSE     FALSE
#>         DM  DMcx Paralysis Renal Cancer LiverSevere  Mets   HIV
#> 1000  TRUE FALSE     FALSE FALSE  FALSE       FALSE FALSE FALSE
#> 1001 FALSE FALSE      TRUE FALSE  FALSE       FALSE FALSE FALSE
#> 1002 FALSE FALSE     FALSE FALSE  FALSE       FALSE FALSE FALSE
 
# or go straight to the Charlson scores:
charlson(patient_data)
#> 1000 1001 1002 
#>    2    2    0
 
# for more examples, see this and other vignettes
vignette("introduction", package = "icd")

Relevance

ICD-9 codes are still in heavy use around the world, particularly in the USA where the ICD-9-CM (Clinical Modification) was in widespread use until the end of 2015. ICD-10 has been used worldwide for reporting cause of death for more than a decade, and ICD-11 is due to be released in 2018. ICD-10-CM is now the primary coding scheme for US hospital admission and discharge diagnoses used for regulatory purposes and billing. A vast amount of electronic patient data is recorded with ICD-9 codes of some kind: this package enables their use in R alongside ICD-10.

Comorbidities

A common requirement for medical research involving patients is determining new or existing comorbidities. This is often reported in Table 1 of research papers to demonstrate the similarity or differences of groups of patients. This package is focussed on fast and accurate generation of this comorbidity information from raw lists of ICD-9 codes.

ICD-9 codes

ICD-9 codes are not numbers, and great care is needed when matching individual codes and ranges of codes. It is easy to make mistakes, hence the need for this package. ICD-9 codes can be presented in short 5 character format, or decimal format, with a decimal place separating the code into two groups. There are also codes beginning with V and E which have different validation rules. Zeroes after a decimal place are meaningful, so numeric ICD-9 codes cannot be used in most cases. In addition, most clinical databases contain invalid codes, and even decimal and non-decimal format codes in different places. This package primarily deals with ICD-9-CM (Clinical Modification) codes, but should be applicable or easily extendible to the original WHO ICD-9 system.

ICD-10 codes

ICD-10 has a somewhat simpler format, with consistent use of a letter, then two alphanumeric characters. However, especially for ICD-10-CM, there are a multitude of qualifiers, e.g. specifying recurrence, laterality, which vastly increase the number of possible codes. This package recognizes validity of codes by syntax alone, or whether the codes appear in a canonical list. The current ICD-10-CM master list is the 2016 set. There is no capability of converting between ICD-9 and ICD-10, but comorbidities can be generated from older ICD-9 codes and newer ICD-10 codes in parallel, and the comorbidities can then be compared.

How to get help

Look at the help files for details and examples of almost every function in this package. There are several vignettes showing the main features. Many users have emailed me directly for help, and I’ll do what I can, but it is often better to examine or add to the list of issues so we can help each other. Advanced users may look at the source code, particularly the extensive test suite which exercises all the key functions.

?comorbid
?comorbid_hcc
?explain_code
?is_valid
 
# first show the list
vignette(package = "icd")
vignette("pccc", package = "icd")

Note that reformatting from wide to long and back is not as straightforward as using the various Hadley Wickham tools for doing this: knowing the more detailed structure of the data let’s us do this better for the case of dealing with ICD codes.

Development version

The latest version is available in github icd, and can be installed with:

    install.packages("devtools")
    devtools::install_github("jackwasey/icd")

Contributing and Building

A substantial amount of code has now been contributed to the package. Contributions of any kind to icd are very welcome. See the [GitHub issues page]](https://github.com/jackwasey/icd/issues) to see open issues and feature requests. Documentation, vignettes and examples are very welcome, especially if accompanied by some real-world data.

To build icd, Rcpp must be compiled from source. This happens automatically on Linux, but on Mac and Windows, the following may sometimes be required, especially after upgrading R itself. This is a limitation of the R build system.

install.packages("Rcpp", type = "source")

News

Version 3.3

  • Fixes #160, #161
  • Includes replication code for benchmarking, showing enormous speed of 'icd' over alternatives.
  • Improved referencing and documentation
  • Functions to print vectors of ICD codes more cleanly when they have 'icd' classes or attributes.

Version 3.2.1

  • Pediatric Complex Chronic Condition functions more consistent with the other comorbidity functions
  • Minor bug fixes
  • Fixes #157
  • Eigen library is multithreaded, so OpenMP dependencies now dropped from this package, simplifying and speeding up build, dropping configure script, etc..

Version 3.2.0

  • added Pediatric Complex Chronic Condition comorbidity classification
  • added ICD-10 procedure code data, and AHRQ procedure code classification -- see ICD-10 vignette
  • added ICD-10-CM data from US CMS for 2015 to 2018
  • fixed 588 instead of 588.0 in Quan Elixhauser ICD-9 comorbidity map
  • fix regression where codes appearing in two comorbidities would only count once
  • default best-effort ordering of identifiers (e.g. Visit ID, Encounter, Patient ID) is done. There are many cases where this is not possible (disordered input data), or obvious (the IDs are given as a factor, where the factor level order does not resemble the order of appearance in the vector). For performance, this can be turned off using restore_id_order = FALSE in comorbidity calculations.
  • added icd9_comorbid_charlson and icd10_comorbid_charlson as synonyms for the Quan/Deyo comorbidity calculations. comorbid_charlson will infer the ICD type.
  • depend on new package icd.data for large, rarely modified data files

Version 3.1.2

  • Documentation improved
  • https://jackwasey.github.io/icd/ now has pkgdown generated site in gh-pages branch.
  • Allow more functions to work without having called library(icd) first

Version 3.1.1

  • Includes article submitted to the Journal of Statistical Software, brings in minimal new dependencies with rticles and tinytex
  • Drop RODBC dependency. Uranium sample data can still be reproduced offline.

Version 3.0

  • major improvement in comorbidity calculation speed using matrix multiplication behind the scenes. ICD-10 codes were slower than ICD-9 code comorbidities before, and are now 1.5 - 2 orders of magnitude faster.
  • Added AHRQ clinical condition software (CCS) classification for ICD-9 and ICD-10, thanks to Vitaly Druker, fixes #127
  • added 'charlson' synonyms for 'quan_deyo', e.g. Charlson comorbidities can now be found using comorbid_charlson(patient_data)
  • Added AHRQ clinical condition software (CCS) classification for ICD-9 and ICD-10, thanks to Vitaly Druker, fixes #127
  • function names have lost the 'icd_' prefix, old names kept for compatiblity, but easier now to write comorbid(patient_data), and icd::comorbid may also be used, and which many consider good practice. explain_icd synonym avoids name conflict with the popular dplyr package, but icd::explain also a nice option.
  • Assume C++11, which is now four years old, enabling code simplification
  • rename all functions to remove the icd_, people can use the concise icd:: if library(icd) was not callled
  • deprecated rarely or unused code, including %i9in%

Version 2.4.1

  • updated HCC to include 2017 data, thanks to @anobel fixes #136
  • fixed github #133 - segfault with empty ICD-10 string in icd10_comorbid
  • add Suggested dependency on microbenchmark, now CRAN is sensibly not requiring Suggests to be installed even for testing
  • re-enabled testthat C++ Catch tests and make examples tests run correctly
  • drop Wecker continuous integration support
  • fixed #108 - when no valid codes are given to icd_explain_table

Version 2.3

  • Fixed nasty bug which failed to sort the names of a vector of ICD codes correctly, resulting in some incorrect icd_explain. Fixes github #123, #124, #126.
  • update stale URLs for example data sources

Version 2.2

  • Removed deprecated functions and data. icd9 still available in CRAN repo, but not being updated. This greatly speeds up and simplifies the test suite.
  • Bug fixes #89 #99 #106 #109 #100 #111 #113 #115 #116 #117 with many thanks to bug reporters and users
  • stopped changing an attribute by reference, thanks to Tomas Kalibera for highlighting.

Version 2.1

  • Many thanks to @anobel for providing parsing and lookup of Hierarchical Condition Codes.
  • Many thanks to @edlee123 for contributing explain_table which tabulates results of looking up various information about a list of ICD codes. This is a new feature which may be changed as it is used in the real world, and more tests are developed.
  • removed stringr family of dependencies: it was often slower on benchmarking than built-ins, and no clear benefits other than internally consistent syntax, and stringr updates caused CRAN warnings due to a documentation change.
  • Fixes #86, thanks to @anobel
  • Fixes #91
  • Fixes #93 thanks to @nancyorgan
  • Fixes #97, thanks to @michaelgao8
  • Speed improvements, mostly simply from using base R hashed Environments instead of lists

Version 2.0

  • ICD-10 support, including ICD-10 comorbidity mappings, validation and explanations (code to description). ICD-10-CM 2016 is included.
  • Extended the already comprehensive test suite to cover ICD-10 and fixed various bugs. Thanks to @wmurphyrd for providing some Elixhauser test data.
  • Package icd9 should now be uninstalled.
  • Simplify functions
  • Automatic detection of ICD version when not specified
  • Automatic detection of use of short or decimal format codes
  • e.g. icd9ExplainShort becomes icd_explain
  • Optional use of the S3 class system so data can be described as ICD-9 or ICD-10, so appropriate functions are used without guesswork. This means that you can optionally label your data with the appropriate class, e.g. as.icd10("A01") or as.icd9cm("0101"). This will help avoid mistakes when working with mixed data.
  • ICD-9 and ICD-10 sub-versions, particularly for distinguishing of ICD-9 (WHO) from ICD-9-CM, and same for ICD-10. This can also be extended to other countries, e.g. ICD-10-CA, or other code types altogether.
  • Soft deprecated all icd9 prefix functions, now this package equally covers ICD-10. New naming scheme follows Hadley Wickham's preferred coding style, using underscores. Most public functions begin with icd_. Package data, and version specific functions, are named with icd_, icd9_, icd10_, prefixes. E.g. icd10_chapters and icd9cm_hierarchy. All deprecated functions will still work, but they give warnings (sometimes many). The warnings can be turned off with an option. The original test suite from icd9 runs and passes on the icd package, with only minimal changes.
  • Completely deprecated some previously soft deprecated functions, e.g. icd9ValidDecimal
  • Works with very latest testthat which has backward-incompatible changes
  • Minimized dependencies, but icd does import stringi via stringr to give cleaner string processing. Base string processing is still used as it is often faster. magrittr is now too useful not to import, has no dependencies of its own, and is imported by stringr anyway. CRAN now also seems to need base packages to be listed as imports.
  • More automated testing with Travis and Wercker, Codecov and Coveralls.

Version 1.3

  • With many thanks to @wmurphyrd, Quan's revised scoring system for Charlson comorbidities is now included.
  • Re-enabled OpenMP, and also use GNU C++ standard library parallel extensions (which also use OpenMP) when available. Thoroughly tested with various docker compiler configurations and no memory or undefined behavior problems appear.
  • Use fastmatch for fast factor generation, but with the tweak of not sorting the levels. This had been by far the slowest step in generating comorbidities.

Version 1.2.2

  • Completed fix from version 1.2.1 by temporarily disabling OpenMP on all platforms. Still runs quickly due to other optimizations made in supporting the OpenMP. It'll be back, but only when I or someone else can create a docker or VM image which runs Clang 3.7 or greater, OpenMP, with/without LLVM C++ standard library. See #75

Version 1.2.1

  • Partly fixed obscure memory access violation error seen only withwhen using OpenMP clang 3.7 on fedora, and maybe OS X.

Version 1.2

  • Make annual revisions of ICD-9-CM available. The package includes data from each year which CMS has published (versions 23 to 32). The default is to use version 32. More work will be needed to make it straightforward to use an arbitrary version when running icd9 commands. These are avaiable in the package data icd9Billable. See vignette for examples.
  • Inexplicably, the only canonical list which includes both ICD-9-CM codes and headings is an RTF file. icd9 can now parse this eclectically formatted document to extract all the headings, so it is not possible to do icd9Explain on a non-billable four-digit code, e.g. 643.0 (Mild hyperemesis of pregnancy). Previously on three-digit and billable (i.e. lead node) codes were used. In principle, the RTF parsing code could be run on previous versions going back to about year 2000. It seems that most years are the same or expand previous years, although there are a few deletions. Ideally, we would know what year/version a given ICD-9 code was coded under, and then validate or interpret accordingly. This can indeed be done for billable codes, but until the RTF is parsed for previous years, not for headings.
  • Condense and range functions has been completely reworked now that all the intermediate heading codes are available. This means there will be slight differences in the results produced, and there are still very minor quirks, but the tests cover thoroughly all known ICD-9-CM codes in existence from any available version.
  • The package data was refreshed with comorbidities derived from source ranges specified by the original authors. These deliberately produce valid but non-existent codes, and these are now slightly different with the range work. This should not have any impact on comorbidity assignmentments from real ICD-9 codes, but keeps the package consistent with itself.
  • OpenMP and C++11 are now both enabled on platforms which allow this (i.e. everything except Solaris), which gives performance improvements.
  • Van Walraven comorbidity score (analagous to Charlson score, but based on Elixhauser comorbidities) added by @wmurhpyrd, with thanks.
  • Dropped most included data from the package, as most can be retrieved from reliable web sites. The data is still in the github repo, but is downloaded automatically when needed (which is only working in the package source tree.)
  • stopped exporting 'parts' functions, as these complicate the namespace and are unlikely to end-user, but still available with icd:::icd9PartsToShort etc.
  • code clean-up with excellent lintr package from @jimhester
  • bug fixes
  • more tests, with coverage at about 85% with the full test suite
  • included first thousand patients from a public domain dataset from Vermont, available as vermont_dx.

Version 1.1

  • C++ OpenMP parallel processing of comorbidities for further many-fold speed improvement, scaling to cores available in a machine, allocating about a million rows of comorbidities in a second or two, on a moderate workstation. Some speed is lost by (optionally) allowing disordered input visit IDs.
  • C++ optimization using simpler, faster STL structures, working with R factors (converted automatically) so comorbidity assignment is done using integer logic only
  • drop unwanted old vignette from the build. Fixes #42
  • refinement of many functions to specifically accept factors or character vectors, but not integers in most cases. Fixes #38
  • more natural use of matrix for comorbidities, with option to return data frames. Allow matrix or data.frame in fuctions which accept comorbidities
  • bug fixes with thanks to @wmurhpyrd. Fixes #44, fixes #46
  • fixes #41
  • extensive use of the excellent and very efficient checkmate package for validating function arguments
  • improved test coverage, demonstrated by the excellent covr package and coveralls.io . Fixes #4

Version 1.0

  • Calculate Charlson scores
  • Sum distinct comorbidities or diagnoses by patient
  • Core rewrite in C++ for 50+ times speed improvement. 100,000 patients assigned comorbidities in ~2 seconds.
  • Simplified handling validation of codes. No longer done in every function.
  • Most functions now guess the ICD-9 code type automatically (e.g. 00321 vs 003.21)
  • Reduced external dependencies down to Rcpp and checkmate (a very lightweight and fast function argument checker)
  • Bug fixes (see github)
  • API changes
    • no more validation except in the icd9IsValidXxx functions. Removed stopIfInvalidIcd9, icd9InvalidActions
    • internalized utility functions. They are also packaged and tested in jwutil
    • deprecated icd9ValidXxx in favour of icd9IsValidXxx
    • deprecated icd9ComorbditiesXxx replacing with briefer icd9ComorbidXxx
    • stopped exporting benchmarking and SAS code processing.

Version 0.5

  • Filter icd9 data for validity or existence. Fixes issue #27
  • Guess whether a code is short or decimal. Fixes #22
  • Enable ranges for Exxx codes. Fixes #9
  • Explain and condense long lists of ICD-9 codes using maximum number of higher-level descriptions without being overly broad. Fixes #3
  • Include high level descriptions, chapters, majors, which do not have CMS descriptions because they are not themselves billable codes. Fixes #2
  • E codes <800 can be valid in 2014. Fixes #1
  • Bug fixes, vignette and other documentation updated.
  • Test suite ever more comprehensive.
  • Rename Elixhauser arguments and functions to "Elix" for clarity of code, and ease of typing.

Version 0.4

  • Present-on-arrival field handled gracefully
  • Revised co-morbidity handling, with functions specific to each mapping
  • Standardized naming of fields across Elixhauser and Charlson based mappings for clarity, consistency and comparaibility
  • Handle mild and severe co-morbidities without double-counting. Fixes issue #23
  • Enable magrittr-style piping as option to clarify chains of operations. Issue #5
  • many more test cases, including specific tests for individual values in each computed mapping, and also for the new comorbidity functions
  • Vignette updated to demonstrate clearer new functions

Version 0.3

  • Addition of several utility functions, mostly kept for internal use, but which clarify the code
  • Consistent behavior with invalid codes, allowing ignore, silent conversion to NA, warn with conversion to NA, or stopping.
  • Outline of code to read top level ICD-9 definitions from canonical CDC RTF file.
  • Progress to
  • Made functions magrittr friendly
  • Bug fixes, refactoring, more tests
  • Include parsing of hierarchy
  • Internal function which reads a file from a zip at a URL
  • The included pre-parsed data are slightly updated to include some missing top-level numbers when all the children were also present. ("043" %i9s% "04499" should equal "043" %i9s% "044")

Version 0.2.1

  • trivial changes to appease CRAN

Version 0.2

  • Added more co-morbidity mappings, Elixhauser original and Quan/Elixhauser.
  • Exposed more functions to public API
  • Better conversion of ICD-9 codes to textual representation
  • Expanded vignette
  • Bug fixes and test cases
  • Improved documentation

Version 0.1.1

  • Include missing import for memoise
  • optionally process AHRQ comorbidity subcategories into parent categories
  • include the processed binary data. Unlikely to change often, and not too big.

Version 0.1

  • First release. Fully functional, no known bugs, but will benefit from addition of planned features described in README.

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.
install.packages("icd")

3.3 by Jack O. Wasey, 5 months ago

https://jackwasey.github.io/icd/

Report a bug at https://github.com/jackwasey/icd/issues

Browse source code at https://github.com/cran/icd

Authors: Jack O. Wasey [aut, cre, cph] , William Murphy [ctb] (Van Walraven scores) , Anobel Odisho [ctb] (Hierarchical Condition Codes) , Vitaly Druker [ctb] (AHRQ CCS) , Ed Lee [ctb] (explain codes in table format)

Documentation:   PDF Manual  

GPL-3 license

Imports checkmate, magrittr, methods, stats, Rcpp, utils

Depends on icd.data

Suggests knitr, roxygen2, rmarkdown, rticles, testthat, tinytex, xml2

Linking to Rcpp, RcppEigen, testthat

System requirements: pandoc >= 2.2 to build the JSS vignette

Suggested by icd.data, nhds.

See at CRAN

R links

R homepage

Download R

Mailing lists

R documentation

R manuals

R FAQs

The R Journal

CRAN links

CRAN homepage

CRAN repository policy

Submit a package

METACRAN stuff

About METACRAN

At github

Report a bug