Comorbidity Index Selection Interpreter
Comorbidity Index Selection Interpreter
Charlson and Elixhauser were derived from different data against different outcomes, and so were their several weightings. This names which published index was derived for the question you are asking — and says when none was.
Which comorbidity index was derived for this question
Data source and outcomeComorbidities taken from ICD-coded discharge abstracts, adjusting for mortality within one year
What each index was derived against
- Charlson 1987
- chart review of 559 general medical admissions in 1984, fitted to one-year mortality; validated against ten-year survival in 685 patients with breast cancer. Nineteen conditions, weights 1 to 6, maximum 33 with the hierarchy applied
- Quan 2011
- ICD-coded discharge abstracts from Calgary in 2004, fitted to mortality within one year of discharge, validated against in-hospital mortality in six countries with c-statistics from 0.727 to 0.878. Twelve non-zero weights, maximum 24
- Elixhauser 1998
- a statewide California administrative database of 1,779,167 stays, against length of stay, charges and in-hospital death. Thirty conditions, and deliberately NOT an index: Elixhauser is a condition list, not a score. The 1998 publication kept the comorbidities as a set of separate covariates on purpose, because each one affected outcomes differently across patient groups; turning the list into one number means choosing a published weighting, and the published weightings are different instruments giving different numbers.
- van Walraven 2009
- about thirteen years of admissions (1996–2008) at one Canadian hospital, fitted to in-hospital death. Twenty-one of the thirty conditions weighted, range −19 to 89
- AHRQ’s two indices
- an in-hospital mortality index derived from about 23.8 million 2018 discharges across 45 states, and a separate 30-day readmission index, both over a refined 38-measure version of the list. Their weights are not printed on this site because they could not be extracted from AHRQ’s own user guide
- Sharma 2021
- 6,094,672 adult cases from 102 Swiss general hospitals, 2012–2017, fitted to in-hospital death. Included here because it is the clearest published demonstration that weights are local: refitting van Walraven’s weighting in a new country improved the c-statistic from 0.863 to 0.867
- what no index gives you
- a patient-level probability. A score is not a diagnosis and a cohort figure is not this patient’s outcome: a stratum in which 52 per cent died within a year describes that stratum, not which 52 per cent.
Worked example
Comorbidities taken from ICD-coded discharge abstracts, adjusting for mortality within one year
Administrative ascertainment with a one-year mortality outcome is the exact question Quan's weights were fitted to, in Calgary 2004 discharge data, and they were published with ICD-9 and ICD-10 code lists
Change the outcome to in-hospital mortality and the answer becomes the Elixhauser condition set, or van Walraven's weights — fitted to in-hospital death, c-statistic 0.863 against 0.850 for Charlson weights in 6,094,672 Swiss admissions
Change the outcome to 30-day readmission and the answer becomes the AHRQ Elixhauser readmission index, which exists because readmission is not mortality
Change the ascertainment to chart review, keeping the one-year outcome, and the answer becomes Charlson 1987, as originally derived — the only one of these built from charts
Chart review with an in-hospital or readmission outcome returns No published index was derived for this combination, which is the honest answer and the reason this page exists. Nothing on this site was derived for it, and a borrowed index should be declared as borrowed
What each index was derived from, and against what
| Index | Ascertainment | Derivation data | Outcome fitted |
|---|---|---|---|
| Charlson 1987 | Chart review | 559 general medical admissions, one month of 1984 | Mortality at one year |
| Quan 2011 updated Charlson | ICD codes | Calgary discharge abstracts, 2004 | Mortality within one year of discharge |
| Elixhauser 1998 (as 30 covariates) | ICD-9-CM codes | Statewide California database, 1,779,167 stays | Length of stay, charges, in-hospital death |
| van Walraven 2009 point system | ICD codes | One Canadian hospital, about 13 years to 2008 | In-hospital death |
| AHRQ mortality index | ICD-10-CM codes | About 23.8 million discharges, 45 states, 2018 | In-hospital death |
| AHRQ readmission index | ICD-10-CM codes | HCUP databases | Readmission within 30 days |
| Sharma 2021 Swiss weights | ICD codes | 6,094,672 cases, 102 Swiss hospitals, 2012–2017 | In-hospital death |
Discrimination, where it has been compared head to head
| Model | c-statistic | Cohort |
|---|---|---|
| Age, sex and hospital type alone | 0.757 | 6,094,672 Swiss admissions, in-hospital mortality 2.3% |
| Plus Charlson weights | 0.850 | the same |
| Plus van Walraven Elixhauser weights | 0.863 | the same |
| Plus locally refitted Swiss weights | 0.867 | the same |
| Quan updated Charlson weights | 0.727 – 0.878 | six countries, in-hospital mortality |
| Charlson 1987 weights | 0.723 – 0.882 | the same six countries |
An index predicts what it was fitted against
There is no best comorbidity index, and the question that produces a useful answer is not which index is best but which one was derived for the question being asked. Two things define that: how the comorbidities are found, and which outcome is being adjusted for. The Charlson index is the only one of the common choices built from chart review, and it was fitted to one-year mortality in 559 general medical admissions in 1984. Everything else — Quan’s re-derived Charlson weights, the Elixhauser condition set, van Walraven’s point system, AHRQ’s two indices, Sharma’s Swiss weights — was derived from ICD-coded discharge abstracts, and most of them against in-hospital death.
That matters more than it sounds, because ascertainment and outcome both change the answer. A chart-based and a code-based Charlson score on the same patient are different measurements: coding cannot see the original’s five-year window on the tumour item, cannot separate a compensated from a decompensated cirrhosis without reading the notes, and records only what the discharge summary happened to list. And an index fitted to in-hospital death is fitted partly to length of stay, because a patient who dies after discharge does not count; AHRQ publishes a separate readmission index for exactly that reason.
What the choice between published indices buys is smaller than the literature’s enthusiasm suggests. In 6,094,672 Swiss admissions, age, sex and hospital type alone gave a c-statistic of 0.757 for in-hospital mortality; adding Charlson weights gave 0.850, van Walraven’s Elixhauser weights 0.863, and weights refitted in that same population 0.867. Comorbidity adjustment is worth about 0.09 of a c-statistic; the choice among weightings is worth about 0.017, and local refitting beats every published weighting. Across six countries the two Charlson weight sets traded places, each better in three.
One combination returns no index at all here, and that is deliberate. Chart-based or clinically assessed comorbidity against an in-hospital or thirty-day readmission outcome is something nothing on this site was derived for. The honest handling is to name the index, the weight set, the ascertainment method and the derivation outcome, accept that the published discrimination does not transfer, and consider carrying the conditions as separate covariates — which is what the Elixhauser measures were designed for in the first place, and which needs no weighting at all.
A screening score is not a diagnosis: a published sensitivity is a property of the instrument in the population it was validated in, not a statement about this patient. This page reports published figures and recommends no action. Every weight, cut-off and outcome figure here comes from a named derivation cohort, and cohorts differ in case mix, era, coding and outcome definition; where your own institution’s protocol or analysis plan differs, it takes precedence.
Frequently asked questions
Charlson or Elixhauser?
It depends on the data and the outcome. Charlson is the only one of the two built from chart review and the only one fitted to a post-discharge outcome. Elixhauser was built from administrative codes against in-hospital death and was designed to enter a model as thirty separate covariates. Where both are available, the weighted Elixhauser discriminates in-hospital mortality slightly better — 0.863 against 0.850 in 6,094,672 Swiss admissions.
Which Charlson weights should I use?
The 1987 weights for chart-based ascertainment and a one-year or longer outcome, because that is what they were derived for. Quan’s 2011 weights for ICD-coded data, because they were derived in it and published with code lists. In six-country validation the updated weights were better in three countries and worse in three, so neither is simply the newer and better choice.
Can I use a mortality index to adjust for readmission?
People do, and it is a substitution to declare. AHRQ publishes a separate 30-day readmission index over the Elixhauser list precisely because readmission and death have different predictors. Its weights are not reproduced on this site because they could not be extracted from AHRQ’s own user guide, so go to that documentation.
Is a single comorbidity score better than separate covariates?
Usually not for adjustment, if the sample supports the covariates. Elixhauser’s original publication kept the thirty conditions separate on purpose, because each affected outcomes differently in different patient groups. A weighted summary is less prone to overfitting in a small sample and is easier to tabulate, and it throws that variation away.
Do published weights transport to my population?
Imperfectly, and there is a clean demonstration. Refitting van Walraven’s Elixhauser weighting in Swiss national data raised the c-statistic from 0.863 to 0.867 and improved net reclassification by 1.6 per cent, and the authors’ conclusion was that country- or region-specific weights are worth deriving. Weights encode a case mix, a coding practice and a treatment era.
Related calculators
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):373–83.
- Quan H, Li B, Couris CM, Fushimi K, Graham P, Hider P, Januel JM, Sundararajan V. 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):676–82.
- 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):626–33.
- Sharma N, Schwendimann R, Endrich O, Ausserhofer D, Simon M. Comparing Charlson and Elixhauser comorbidity indices with different weightings to predict in-hospital mortality: an analysis of national inpatient data. BMC Health Serv Res. 2021;21:13.
- Agency for Healthcare Research and Quality, Healthcare Cost and Utilization Project. Elixhauser Comorbidity Software Refined for ICD-10-CM: user guide v2022.1. hcup-us.ahrq.gov (accessed 9 October 2026).
- Wikipedia. Elixhauser Comorbidity Index. en.wikipedia.org (accessed 9 October 2026).
Not medical advice. For healthcare professionals and education. Reference intervals vary by laboratory and assay — always use your own laboratory's. Never base a dose or a treatment decision on this page alone. Full disclaimer at calcengines.com/disclaimer/