comorbidity

Version 1.0.3 | 2026-10-12

Quick Start · Installation · Worked Examples · Key Options · Stata-Tools · Report a bug

comorbidity scans wide-format ICD-10 diagnosis fields and creates patient-level Charlson, Elixhauser, or custom weighted scores. It is for analysts who need to turn one or more diagnosis variables per encounter into condition indicators and a reproducible comorbidity score.

Quick Start

After installing codescan and comorbidity, score three patients with the original Charlson weights and collapse to one row per patient:

clear
input long pid str6 dx1 str6 dx2
1 "I21"  "I50"
1 "E119" ""
2 "C780" ""
3 ""     ""
end

comorbidity dx1 dx2, id(pid) charlson(original) collapse band
list pid charlson mi chf dm_uncomp metastatic, noobs abbreviate(16)
matrix list r(bands)

  +----------------------------------------------------+
  | pid   charlson   mi   chf   dm_uncomp   metastatic |
  |----------------------------------------------------|
  |   1          3    1     1           1            0 |
  |   2          6    0     0           0            1 |
  |   3          0    0     0           0            0 |
  +----------------------------------------------------+

Patient 1 scores 3 (myocardial infarction, heart failure, uncomplicated diabetes), patient 2 scores 6 (metastatic solid tumor), and patient 3, with no codes, scores 0.

(note: condition pvd matched 0 observations)
(note: condition cvd matched 0 observations)
(note: condition dementia matched 0 observations)
(note: condition copd matched 0 observations)
(note: condition rheumatic matched 0 observations)
(note: condition peptic matched 0 observations)
(note: condition liver_mild matched 0 observations)
(note: condition dm_comp matched 0 observations)
(note: condition hemiplegia matched 0 observations)
(note: condition renal matched 0 observations)
(note: condition cancer matched 0 observations)
(note: condition liver_severe matched 0 observations)
(note: condition hiv matched 0 observations)
(note: mi and chf overlap in 1 obs, 100% of smaller group)

codescan: 17 conditions, 2 variables, N =          3 pid values

  Condition              Matches   Prevalence
  --------------------------------------------
  Myocardial Infarction        1        33.3%
  Congestive Heart Fai~        1        33.3%
  Peripheral Vascular ~        0         0.0%
  Cerebrovascular Dise~        0         0.0%
  Dementia                     0         0.0%
  Chronic Pulmonary Di~        0         0.0%
  Rheumatic Disease            0         0.0%
  Peptic Ulcer Disease         0         0.0%
  Mild Liver Disease           0         0.0%
  Diabetes without Com~        1        33.3%
  Diabetes with Compli~        0         0.0%
  Hemiplegia or Parapl~        0         0.0%
  Renal Disease                0         0.0%
  Any Malignancy               0         0.0%
  Moderate or Severe L~        0         0.0%
  Metastatic Solid Tum~        1        33.3%
  HIV/AIDS                     0         0.0%

  Collapsed to          3 unique pid values

comorbidity: charlson (original), N =          3
  score: mean =   3.00, range = [     0.00,      6.00]
  bands n (<0 / 0 / >0-<3 / 3-<5 / 5+):         0         1         0         1         1

r(bands)[5,2]
                      n    percent
score_nega~e          0          0
      score0          1  33.333333
    score1_2          0          0
    score3_4          1  33.333333
  score5plus          1  33.333333

_{Output copied from demo/console_output.md, written by demo/demo_comorbidity.do.}

Features

• One row per patient by default. collapse is the default output shape; merge puts the patient-level indicators and score back on the encounter rows instead.
• Named, published schemes. charlson(original) and charlson(quan2011) use the Quan et al. (2005) ICD-10 Charlson definitions; elixhauser(vanwalraven) uses the Quan Elixhauser definitions with van Walraven weights. The reserved AHRQ scheme names exit with r(198) rather than run an approximation.
• Hierarchy rules in the open. Built-in indexes apply stated supersession pairs (complicated over uncomplicated diabetes, metastatic over non-metastatic cancer, and so on) by default; nohierarchy returns the unadjusted indicators.
• Your own weighted index. custom() reads name, pattern, and weight from a .csv or .dta file; the schema, names, and weight range are checked before any diagnosis is scanned.
• Inputs are protected. replace can overwrite earlier condition and score variables but never id(), a diagnosis field, date(), or refdate().

Requirements

• Stata 16 or later
• The codescan dependency
• One or more wide-format diagnosis variables and an identifier for the patient or analytic unit

Installation

Install the scanner dependency first:

net install codescan, from("https://raw.githubusercontent.com/tpcopeland/Stata-Tools/main/codescan") replace

Then install comorbidity:

capture ado uninstall comorbidity
net install comorbidity, from("https://raw.githubusercontent.com/tpcopeland/Stata-Tools/main/comorbidity") replace

Commands

How It Works

comorbidity passes the diagnosis variables to codescan with a built-in or user-supplied code dictionary, then applies the selected weights to the resulting binary condition indicators.

• charlson(original) and charlson(quan2011) use the Quan et al. (2005) Charlson ICD-10 definitions with the corresponding weight scheme.
• elixhauser(vanwalraven) uses the Quan et al. (2005) Elixhauser ICD-10 definitions and van Walraven weights.
• custom(filename) reads condition names, ICD-10 patterns, and weights from a .csv or .dta code file.
• collapse produces one row per id() and is the default when neither collapse nor merge is specified; merge returns patient-level indicators and the score on the encounter rows.
• Built-in Charlson and Elixhauser indexes use hierarchy rules by default. Use nohierarchy when you need the unadjusted indicators; custom indexes do not apply a built-in hierarchy.

Worked Examples

1. Charlson scores and score bands

band adds patient-level counts and percentages for negative scores, 0, greater than 0 to less than 3, 3 to less than 5, and 5 or greater to r(bands). These bands are exhaustive for integer and fractional scores. The nonnegative cutpoints reproduce Charlson et al. (1987); for Elixhauser and custom scores they are descriptive categories rather than validated risk strata.

clear
input long pid str6 dx1 str6 dx2
1 "I21" "I50"
1 "E119" ""
2 "C780" ""
3 "" ""
end

comorbidity dx1 dx2, id(pid) charlson(original) band
list pid charlson mi chf dm_uncomp metastatic, noobs
matrix list r(bands)

2. Merge a generated score back to encounter rows

Use generate() to prefix the component indicators and score variable. replace permits the example to overwrite the placeholder score variable.

clear
input long pid str6 dx1 str6 dx2
1 "I21" "I50"
1 "E119" ""
2 "C780" ""
end
gen double cmb_score = -99

comorbidity dx1 dx2, id(pid) charlson(quan2011) merge generate(cmb_) replace
list pid dx1 dx2 cmb_score cmb_mi cmb_chf cmb_dm_uncomp cmb_metastatic, noobs

3. Elixhauser van Walraven scoring

The van Walraven score can be negative because its published weight vector includes negative weights.

clear
input long pid str6 dx1 str6 dx2 str6 dx3
1 "I50" "C780" "F11"
2 "E66" "F32" ""
3 "I10" "I11" "E112"
end

comorbidity dx1 dx2 dx3, id(pid) elixhauser(vanwalraven) generate(elx_) replace band
list pid elx_score elx_chf elx_metastatic elx_drug elx_obesity elx_depression, noobs
matrix list r(bands)

4. Restrict matching to a date window

Supply encounter and reference dates with date() and refdate(), then set the lookback and lookforward windows in days.

clear
input long pid str6 dx1 int dxdate int refdate
1 "I50" 21910 21915
1 "I21" 21885 21915
2 "I50" 21550 21915
end
format dxdate refdate %td

comorbidity dx1, id(pid) charlson(original) date(dxdate) refdate(refdate) lookback(30) lookforward(10) inclusive
list pid charlson mi chf, noobs

5. Use a custom weighted code file

The custom file must contain name, pattern, and weight columns. Each detected condition must have exactly one matching weight.

clear
input long pid str6 dx1 str6 dx2
1 "I21" "I50"
2 "E119" ""
end

tempfile custom_codes
preserve
clear
input str12 name str20 pattern double weight
"mi" "I21|I22" 10
"chf" "I50" 2
"dm" "E11" 4
end
save "`custom_codes'.dta", replace
restore

comorbidity dx1 dx2, id(pid) custom("`custom_codes'.dta") replace
list pid custom mi chf dm, noobs

Demo

The repository demo exercises the built-in indexes, merge and collapse output, generated names, date windows, hierarchy control, custom weights, and verbose scanning. From the repository root, run:

stata-mp -b do comorbidity/demo/demo_comorbidity.do

The checked-in console transcript is demo/console_output.md.

Command Reference

comorbidity varlist [if] [in], id(varname) charlson(scheme) [options]
comorbidity varlist [if] [in], id(varname) elixhauser(scheme) [options]
comorbidity varlist [if] [in], id(varname) custom(filename) [options]

Exactly one of charlson(), elixhauser(), or custom() is required. collapse and merge are mutually exclusive, and collapse is selected automatically when neither is specified.

Key Options

Stored Results

After a successful run, comorbidity stores:

Assumptions and Limits

• The input diagnosis fields are wide-format variables containing ICD-10 codes that can be matched by codescan.
• The package-defined Charlson hierarchy supersedes uncomplicated diabetes with complicated diabetes, mild liver disease with severe liver disease, and any malignancy with metastatic cancer.
• The package-defined Elixhauser hierarchy supersedes uncomplicated hypertension with complicated hypertension, solid tumor without metastasis with metastatic cancer, and uncomplicated diabetes with complicated diabetes. These hierarchy rules are package conventions rather than rules established by the cited scoring papers.
• The Quan ICD-10 Charlson dictionary combines solid tumors, leukemia, and lymphoma into one cancer indicator. charlson(original) therefore reproduces Charlson's weights but cannot separately add those three original conditions when more than one is present.
• The quan2011 weight vector is cross-validated against R comorbidity 1.1.0. That parity check is not an independent audit of the primary paper's exact condition-level weight table.
• elixhauser(ahrq_mortality) and elixhauser(ahrq_readmission) are reserved but not implemented in this release and exit with error code r(198).
• Custom code files are limited to .csv and .dta extensions and must provide one nonmissing numeric weight per unique condition name. The schema, output names, and aggregate numeric range of the weights are checked before diagnosis scanning begins.
• replace cannot overwrite id(), diagnosis variables, date(), or refdate() through either the score name or a custom condition name.

References

• Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373–383. doi:10.1016/0021-9681(87)90171-8
• Quan H, Sundararajan V, Halfon P, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43(11):1130–1139. doi:10.1097/01.mlr.0000182534.19832.83
• Quan H, Li B, Couris CM, et al. Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. Am J Epidemiol. 2011;173(6):676–682. doi:10.1093/aje/kwq433
• van Walraven C, Austin PC, Jennings A, Quan H, Forster AJ. A modification of the Elixhauser comorbidity measures into a point system for hospital death using administrative data. Med Care. 2009;47(6):626–633. doi:10.1097/MLR.0b013e31819432e5

QA

QA suites and how to run them are documented in qa/README.md.

Version History

• 1.0.3 (2026-09-30): Preserve native S_FN and S_FNDATE presence and exact bytes on every route. Stata extended macro functions omit these native file macros; capture and restoration now use Mata.

• 1.0.2 (2026-09-30): Preserve exact custom double weights and merge sort metadata and the caller’s estimation sample; correct the Quan paralysis boundary; accept case-distinct output names; reject invalid negative windows; protect caller installations during QA

• 1.0.1 (2026-08-30): Prevent custom-weight overflow from changing caller data; strengthen method caveats, released-package wording, and QA oracles

Author

Timothy P Copeland, Karolinska Institutet

Questions and bug reports: GitHub issues. Please include the first line of which comorbidity and your Stata version.

To cite: Copeland TP. comorbidity: Charlson and Elixhauser scores from ICD-10 fields. Stata-Tools, https://github.com/tpcopeland/Stata-Tools/tree/main/comorbidity. Please give the version you used (which comorbidity prints it).

License

MIT (see LICENSE)