icd-package: icd: Comorbidity Calculations and Tools for ICD-9 and ICD-10...

Description Details Author(s) See Also


Calculate comorbidities, medical risk scores, and work very quickly and precisely with ICD-9 and ICD-10 codes. This package enables a work flow from raw tables 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. Comorbidity computation includes Hierarchical Condition Codes, and an implementation of AHRQ Clinical Classifications. Risk scores include those of Charlson and van Walraven. US Clinical Modification, Word Health Organization, Belgian and French ICD-10 codes are supported, most of which are downloaded on demand.



comorbid determines comorbidities for a set of patients with one or more ICD codes each. All the comorbidity functions guess which columns are the identifiers and which are ICD code fields. You may also specify these. Most 'long' or 'wide' data can simply be passed directly, e.g.: comorbid(patient_data_frame).

  • The AHRQ comorbidity mappings from ICD-9 and ICD-10 are provided as icd9_map_ahrq and icd10_map_ahrq. The easiest way to use them is by calling the function comorbid_ahrq directly with your icd_long_data format patient data.

  • Quan revised both Deyo/Charlson and Elixhauser ICD-9 and ICD-10 to comorbidity mappings. These are presented as: icd9_map_quan_deyo, icd10_map_quan_deyo, icd9_map_quan_elix, and icd10_map_quan_elix. Like the AHRQ mappings, these are all carefully extracted from the original publications or source code. These mappings can be used on patient data by calling comorbid_quan_deyo and comorbid_quan_elix

  • There is no canonical Charlson ICD-9 or ICD-10 mapping, so comorbid_charlson uses the thoroughly researched and widely cited comorbid_quan_deyo method by default.

  • The original Elixhauser mappings are provided by the lists icd9_map_elix and icd10_map_elix, and Elixhauser comorbidities can be calculated directly from patient data using comorbid_elix.

  • The US Center for Medicare and Medicare services (CMS) publishes Hierarchical Condition Codes (HCC) which are essentially comorbidity maps with very many comorbidities, complicated by a single- or multi-level system. These categories can be computed using comorbid_hcc.

  • AHRQ also publishes Clinical Classification Software (CCS) which provides another set of disease groups, and this SAS code is implemented in 'icd' by comorbid_ccs

Risk scoring systems

charlson calculates Charlson scores (Charlson Comorbidity Indices) directly from your patient data. If you already calculated the Charlson comorbidities, it is more efficient to use charlson_from_comorbid. Similarly, van_walraven calculates Van Walraven scores (based on the Elixhauser comorbidities, instead of Charlson), and van_walraven_from_comorbid if you already calculated Elixhauser comorbidities


is_valid checks whether ICD codes are syntactically valid (although not necessarily genuine ICD diagnoses). In contrast, is_defined checks whether ICD-9 codes correspond to defined diagnoses or diagnostic groups in the hierarchy of ICD codes. is_leaf (and for the US, is_billable) determine whether given codes are leaves in the hierarchy, or not. icd may need to download data due to package size or copyright restrictions on redistributing data, and needs a cache directory and your permission to do this. Use set_icd_data_dir to do this, e.g., set_icd_data_dir() for a default location in a suitable place for your OS, e.g. ~/.local/icd/icd-ver-num on Linux. Use download_all_icd_data to download everything at once, which will take a few minutes on a broadband connection.

Validation depends on the class of code, and is different if the code is from France, Belgium, the USA, or the World Health Organization (WHO). Use the functions as.icd10who, as.icd10fr, as.icd10be, and as.icd10cm to set the class of a set of ICD codes. This doesn't affect comorbidity calculations, but will change the result of the above validation functions, and


There are many functions to convert ICD-9 codes or their components between different formats and structures. The most commonly used are: decimal_to_short, short_to_decimal to convert, e.g., 002.3 to 0023 and back again. See convert for other options. Conversion between ICD-9, ICD-10, and ICD-11 codes is not yet supported, but is the subject of an upcoming US National Institutes of Health (NIH) hackathon.


You can find children of a higher-level ICD-9 code with children and find a common parent to a set of children (or arbitrary list of ICD-9 codes) with condense. sort (sort.icd9, sort.icd10cm, etc.) sorts in hierarchical, then numerical order, so '100.0' comes before '100.00', for example.

The comorbidity functions in icd accept 'wide' or 'long' format data, but you may wish to use wide_to_long and long_to_wide, which convert between these two most common data structures containing patient disease data. This is more sophisticated and tailored to the problem than base reshaping or packages like dplyr, although these could no doubt be used.

These functions use base R functions to avoid dependencies, so are not very fast for very big datasets.

Explanation and decoding

Use explain_code to convert a set of ICD codes into human-readable descriptions. See above for discussion on WHO, French, Belgian and US ICD code classes. This function can can also reduce the codes to their top-level groups if all the child members of a group are present. diff_comorbid allows summary of the differences between comorbidity mappings, e.g. between revisions by different authors.


Jack O. Wasey jack@jackwasey.com

See Also

Useful links:

icd documentation built on July 2, 2020, 4:07 a.m.