Feature request: charlson_sundararajan2004

[dewittpe]

During the R/Medicine 2026 conference, @DesiQuintans asked

Is there support for ICD code versions from other countries, e.g. ICD-10-AM from Australia?

@dewittpe replied:

Not yet, but I would love to add them to the package! Is there a good reference where I can get the ICD-10-AM specific codes?

@DesiQuintans replied with links and notes:

https://www.jclinepi.com/article/S0895-4356(04)00164-7/abstract
The comorbidity package has Quan-adjusted Elixhauser and Charlson for ICD-10-AM, and I contributed chronic condition "u-codes" to it also. https://github.com/ellessenne/comorbidity/blob/master/data-raw/make-mapping.R

Adding the Sundararajan et.al. (2004) codes will be easy enough. The harder part to start is getting a good reference for the ICD-10-AM codes.

[dewittpe]

To get the sundararajan_2004 method into the current workflow I would need a machine readable reference for ICD-10-AM. My initial searches suggest such a reference may not exist.

Alternative: extend the initial ICD to comorbidity mapping in medicalcoder::comorbidities() to use regex instead of the database joins optionally for the existing methods and manditory for methods based on ICD systems not included in the internal ICD database.

[DesiQuintans]

Hi Peter, what requirements do you have for a "machine readable reference"? A dataframe of codes, dates instated/obseleted, descriptions, etc?

The electronic code lists can be licensed from IHACPA (an Australian government agency) at no cost for non-commercial research, but the application takes a long time to process. I am happy to place a licensing application for medicalcoder since I am located in Australia, but it may take some back-and-forth.

[dewittpe]

@DesiQuintans good question.

What I would be hoping for are data sets similar to what Centers for Medicare and Medicaid Services (CMS) provide for ICD-10-CM: data-raw/icd/icd10/cms

Or from the Centers for Disease Control (CDC): data-raw/icd/icd10/ftp.cdc.gov/pub/Health_Statistics/NCHS/Publications/ICD10CM/2026

That data is then formatted into the internal datasets for medicalcoder through a few steps in the data-raw directory and then in the .onLoad() call

Right now the mapping between ICD codes and comorbidities is done via database style joins. A lot of regex was evaluated in the construction of the internal tables for the linking ICD codes to comorbidites.

If a set of ICD-10-AM codes could be added to medicalcoder that would be ideal. Do you know if the licensed codes from IHACPA could be added to an R package and published? My guess is no, but I hope I'm wrong.

All that said, I think there is a viable alternative which could be built and add more features to medicalcoder in general.

Right now the internal tables linking ICD to comorbidities look like this:

head(medicalcoder::get_charlson_codes())
#   icdv dx full_code code condition charlson_cdmf2019 charlson_deyo1992 charlson_quan2005 charlson_quan2011
# 1    9  0     38.48 3848       pvd                 0                 1                 0                 0
# 2    9  1     003.1 0031      aids                 1                 0                 0                 0
# 3    9  1     007.2 0072      aids                 1                 0                 0                 0
# 4    9  1     007.4 0074      aids                 1                 0                 0                 0
# 5    9  1       010  010      aids                 1                 0                 0                 0
# 6    9  1     010.0 0100      aids                 1                 0                 0                 0

If additional tables of the form

icdv dx regex condition charlson_cdmf2019 charlson_quan2011 charlson_sundararajan2004
10   1  ^I50  chf                       1                 1                         1

are added and medicalcoder::comorbidities() gains an option to specify exact matching on ICD codes (current functionality) or regex then the need for a full ICD code base is eliminated. The only downside I see is that the initial identification of comorbidities by regex will be slower than the exact matching, but that's not a deal breaker. Better to have to wait a few extra seconds to get a result than get nothing at all.

[dewittpe]

I've been building out this on the feature-45 branch.

I think I may have found a source to build the machine readable ICD-10-AM codes:

IHACPA Twelfth Edition has a couple pdfs I can scrape.

At the moment I have regex built and I'm working on documentation and testing. There will be a new argument in the comorbidities() call allowing end users to select if they want the mapping between ICD codes and comorbidities to be done via a prebuilt table (default) or via regular expressions.

[dewittpe]

@DesiQuintans I just pushed to the feature-45 branch a lot of dev work. I have a lot left to do, but I think you should be able to use method = "charlson_sundararajan2004" in comorbidities(). If you get a chance to start testing it out I would love to know how it goes and if it works as expected. I have been able to add ICD-10-AM codes (src = "ihacpa") to the internal database too.

[dewittpe]

Released this feature in v0.9.0