Charlson Index Comorbidity Adjustment for Ischemic

Stroke Outcome Studies

Larry B. Goldstein, MD; Gregory P. Samsa, PhD; David B. Matchar, MD; Ronnie D. Horner, PhD

Background and Purpose—The Charlson Index is commonly used in outcome studies to adjust for patient comorbid
conditions, but has not been specifically validated for use in studies of ischemic stroke. The purpose of the present study
was to determine whether outcomes of ischemic stroke patients varied on the basis of the Charlson Index.
Methods—The Department of Veterans Affairs (VA) Stroke Study prospectively identified stroke patients admitted to 9
VA hospitals between April 1995 and March 1997. The Charlson Index was scored on the basis of discharge
International Classification of Diseases, 9th Revision, Clinical Modification coding and dichotomized (low comorbidity
0 or 1 versus high ⱖ2) for analysis. Validity was assessed on the basis of modified Rankin score at hospital discharge
(good outcome 0 or 1 versus poor ⱖ2 or dead) and 1-year mortality, adjusting for initial stroke severity.
Results—Of the 960 enrolled ischemic stroke patients, 23% had a Charlson Index of 0, 34% 1, 22% 2, 12% 3, and 8% ⱖ4.
Forty-eight percent of those with a low Charlson Index had a good outcome at discharge versus 37% of those with a
high Charlson Index (P⬍0.001). For 1-year mortality, the proportions were 16% versus 26%, respectively (P⬍0.001).
Logistic regression adjusting for initial stroke severity showed that those with a high Charlson Index had 36% increased
odds of having a poor outcome at discharge (P⫽0.038) and 72% greater odds of death at 1 year (P⫽0.001). Every
1-point increase in Charlson Index was independently associated with a 15% increase in the odds of a poor outcome at
discharge (P⬍0.005) and a 29% increase in the odds of death by 1 year (P⬍0.001).
Conclusions—These data support the validity of the Charlson Index as a measure of comorbidity for use in ischemic stroke
outcome studies. (Stroke. 2004;35:1941-1945.)
Key Words: stroke 䡲 outcome 䡲 morbidity 䡲 mortality 䡲 outcome assessment
Downloaded from http://ahajournals.org by on April 23, 2023

S everal studies have developed mathematical models for

predicting the outcomes of acute ischemic stroke patients.
These types of prognostic models generally require data
obtained either prospectively or through medical record
review and allow for statistical adjustments on the basis of the
patients’ relevant comorbid conditions. Outcome studies that
use large databases also need to consider other diseases that
patients may have that could impact the analyses. Direct
access to the patients’ medical records is often impractical or
impossible. These databases generally include listings of
comorbid conditions categorized using the International
Classification of Diseases, 9th Revision, Clinical Modifica-
tion (ICD-9-CM) coding system. The Charlson Index is a
comorbidity scoring system that includes weighting factors
on the basis of disease severity (Table 1).1 The system was
developed originally as a prognostic indicator on the basis of
patients with a variety of conditions admitted to a general
medical service and then validated in an independent cohort
of women with breast cancer. The Charlson Index has been
used subsequently to account for the impact of comorbid
conditions on studies of conditions such as ischemic stroke.
Despite being used for this purpose, the Charlson Index has
not been specifically validated for ischemic stroke outcome
studies. The purpose of the present study was to determine
whether a modified version of the Charlson Index on the basis
of hospital discharge ICD-9-CM codes reflected short-term
functional status and 1-year mortality in a cohort of patients
with acute ischemic stroke independent of stroke severity.
Subjects and Methods
Study Design
Data collected as part of the Department of Veterans Affairs (VA)
Stroke (VASt) Study were used for this analysis. The overall study
design has been published previously.2 Briefly, the VASt Study was
a 9-site nationwide prospective observational cohort study of 1073
acute stroke patients hospitalized within the VA Health Administra-
tion between April 1, 1995, and March 31, 1997. For confidentiality,
individual sites are not identified. Data were collected through
medical record review using a standardized chart abstraction proto-

Received March 5, 2004; final revision received May 7, 2004; accepted May 14, 2004.
From the Durham VA Medical Center (L.B.G., D.B.M.), the Department of Medicine (Neurology [L.B.G.] and General Internal Medicine [G.P.S.,
D.B.M.]), the Stroke Policy Program, Center for Clinical Health Policy Research (L.B.G., G.P.S., D.B.M.), and the Center for Cerebrovascular Disease
(L.B.G., G.P.S., D.B.M.), Duke University, Durham, NC; and the Office of Minority Health and Research (R.D.H.), National Institute of Neurological
Disorders and Stroke, Bethesda, Md.
Correspondence to Dr Larry B. Goldstein, Director of Duke Center for Cerebrovascular Disease, Head of Stroke Policy Program, Center for Clinical
Health Policy Research, Box 3651, Duke University Medical Center, Durham, NC 27710. E-mail golds004@mc.duke.edu
© 2004 American Heart Association, Inc.
Stroke is available at http://www.strokeaha.org DOI: 10.1161/01.STR.0000135225.80898.1c

1941
 1942 Stroke August 2004

TABLE 1. Charlson Index Categories, ICD-9 Codes, and TABLE 1. Continued

Frequencies (nⴝ960)
Frequency,
Frequency, Condition Weight ICD-9 Codes n (%)
Condition Weight ICD-9 Codes n (%) Nonmetastatic solid tumor 2 153.9 (2) 173.9 (2) 46 (4.8)
Myocardial infarct 1 410.11 (2) 411.81 (3) 92 (9.6) 154.8 (1) 185. (25)
410.71 (6) 412. (51) 162.3 (6) 200.13 (1)
410.72 (2) 413.1 (1) 162.4 (1) 238.4 (6)
410.81 (2) 413.9 (13) 162.9 (2)
410.91 (6) 429.79 (2) Leukemia 2 204.10 (1) 205.00 (1) 3 (0.3)
411.1 (4) 204.11 (1)
Congestive heart failure 1 402.01 (1) 428.0 (73) 80 (8.3) Lymphoma, multiple myeloma 2 202.80 (1) 1 (0.1)
402.11 (2) 428.1 (1)
Metastatic tumor 6 197.0 (2) 198.5 (7) 16 (1.6)
402.91 (1) 429.3 (2)
197.4 (1) 198.89 (1)
Peripheral vascular 1 440.20 (9) 440.9 (3) 55 (5.7) 197.7 (2) 199.0
disease 440.21 (2) 443.81 (2) 198.3 (2) 199.1 (1)
440.23 (1) 443.9 (34)
AIDS 6 042 (1) 1 (0.1)
440.24 (2) 444.22 (2)
Dementia 1 290.0 (12) 291.81 (4) 120 (12.5) The Charlson Index categories were modified as described in the text.
290.10 (1) 293.0 (2) “Weights” refer to the weight given to the indicated category used in computing
290.12 (1) 293.83 (5) a patient’s total Charlson Index. The ICD-9-CM codes reflect the codes that had
290.40 (28) 294.0 (1) been assigned at the time of hospital discharge. Numbers in parentheses
291.0 (3) 294.1 (2) indicate the frequencies that each code was assigned. The final column gives
291.2 (1) 294.8 (12) the number of patients and percent in each of the indicated Charlson Index
291.8 (1) 438 (47) categories.

Chronic pulmonary 1 416.0 (2) 493.20 (3) 147 (15.3)

disease 416.8 (2) 403.90 (7) col. The institutional review board for each site approved the study
416.9 (1) 495.1
protocol.
491.20 (3) 496. (113)
491.21 (3) 508.0 (1) Subjects
491.8 (1) 515. (2) A research assistant identified patients with possible stroke within 48
491.9 (1) 516.3 (1) hours of hospital admission by screening admission logs. The
diagnosis was then confirmed by review of medical records and
Downloaded from http://ahajournals.org by on April 23, 2023

492.0 (1) 518.1 (1)

492.8 (5)
discussion with the patient’s attending physician. This patient iden-
tification method was supplemented by review of the hospital
Connective tissue 1 710.0 (2) 715.91 (2) 61 (6.4) discharge files for diagnoses of intracerebral hemorrhage (ICD-
disease 710.1 (1) 715.95 (1) 9-CM code 431) or acute cerebral infarction (ICD-9-CM code 434 or
710.4 (1) 715.96 (2) 436). In addition, all discharge ICD-9-CM codes were recorded.
712.38 (1) 715.98 (21) Patients whose stroke was iatrogenic because of brain trauma or
714.0 (10) 716.98 (2) neoplasm, occurred during hospitalization for other medical condi-
715.09 (9) 716.99 (2) tions, was an extension of a previous stroke, or occurred ⬎7 days
715.36 (2) 719.98 (1) before admission were excluded. The present analysis focused only
715.89 (2) 728.89 (2) on patients with ischemic stroke. Those with hemorrhagic stroke
were not included.
Ulcer disease 1 531.40 (1) 533.20 (1) 23 (2.4)
532.40 (1) 533.70 (4)
532.70 (1) 533.90 (12) Demographic Variables and Clinical Outcome
532.90 (3) Demographic characteristics included age, sex, and self-reported
race/ethnic group (white, black, Hispanic, other). Stroke subtype was
Mild liver disease 1 571.2 (2) 571.5 (1) 6 (0.6) assigned using the Trial of ORG 10172 in Acute Stroke Treatment
571.3 (1) 573.8 (2) (TOAST) criteria with the aid of a computerized system to optimize
Diabetes 1 250.00 (194) 250.21 (1) 312 (32.5) reliability.3 Stroke severity was determined retrospectively with the
250.01 (67) 250.80 (4) Canadian Neurological Scale.2,4,5 The patient’s functional status at
250.02 (29) 250.81 (2) hospital discharge (modified Rankin score [mRS]6) was recorded
250.03 (12) 250.82 (1) prospectively, and mortality rates at 1 year were ascertained.
250.20 (1) 250.83 (1)
Diabetes with end-organ 2 250.31 (1) 250.61 (8) 42 (4.4)
Modified Charlson Index
damage 250.40 (2) 250.62 (1) Comorbid conditions were classified with a modified version of the
250.41 (2) 250.63 (1)
Charlson Index on the basis of hospital discharge ICD-9-CM codes.
(Table 1)1 Each of the indicated diagnoses is assigned a weight and
250.50 (6) 250.71 (1)
summed to provide a patient’s total score. The original Charlson
250.51 (6) 250.72 (1)
Index includes cerebrovascular disease (weight 1) and hemiplegia
250.52 (1) 250.90 (1)
(weight 2). Because these items are reflected in the condition being
250.60 (10) 250.92 (1) evaluated (ie, stroke), they are not included. In addition, the Charlson
Moderate or severe renal 2 581.9 (1) 583.81 (4) 18 (1.9) Index has separate categories for mild versus moderate or severe
disease 582.81 (1) 585. (8) liver disease as well as moderate or severe renal disease. Because
582.9 (3) 586. (1) disease severity cannot be derived from ICD-9-CM codes, liver
disease codes were all categorized as reflecting mild disease, and all
renal disease codes were categorized as reflecting moderate or severe
 Goldstein et al Charlson Index for Ischemic Stroke 1943

TABLE 2. Charlson Index Categories, Frequencies of ICD-9

Codes, and Proportions Within Each Category
Frequency in Proportion
Total Cohort, Within
Condition Code n⫽960 (%) Category (%)
Myocardial infarct 412 5.5 54
Congestive heart failure 428.0 8.4 91
Peripheral vascular disease 443.9 4.1 66
Dementia 438 5.3 45
290.40 3.1 26
Chronic pulmonary disease 496 12.8 76
Figure 1. Distribution of modified Charlson Index scores. Connective tissue disease 715.98 2.3 34
715.09 1.0 15
disease (Table 1). Patients with diabetes and renal disease were
included in the diabetes with end-organ damage category. Ulcer disease —
Mild liver disease —
Statistical Analysis Diabetes 250.00 22.1 63
The modified Charlson Index was dichotomized (low comorbidity 0 250.01 7.5 21
or 1 versus high ⱖ2) for analysis. The mRS was also dichotomized
as commonly used in stroke outcome studies (good outcome, mRS 0 Diabetes with end-organ damage 250.6 1.0 24
or 1 versus poor outcome, mRS ⱖ2 or dead). The relationships Moderate or severe renal disease —
between the unadjusted modified Charlson Index and functional
Nonmetastatic solid tumor 185.0 3.1 55
status at hospital discharge as well as 1-year mortality were deter-
mined. Multivariable logistic regression was then used to determine Leukemia —
the independent effect of Charlson Index on each outcome, control- Lymphoma, multiple myeloma —
ling for stroke severity.
Metastatic tumor 198.5 1.0 41

Results AIDS —
Of 1073 enrolled patients, 960 had an ischemic stroke. The
individual sites enrolled 61 to 118 patients each. Because in each category accounted for by each code. Figure 2 gives
Downloaded from http://ahajournals.org by on April 23, 2023

there was no difference in outcomes among the participating the univariable relationships between the modified Charlson
institutions, the data were collapsed across sites for further Index and the mRS at hospital discharge and 1-year mortality.
analysis. Logistic regression adjusting for initial stroke severity
The patient mean age was 68.1 (⫾9.5) years with 68.8% showed that those with a high Charlson Index (ie, ⱖ2) had
whites, 27.7% blacks, and 3.2% Hispanics. Comorbid condi- 36% increased odds of having a poor outcome (mRS ⱖ2) at
tions by history included diabetes in 36.0%, hypertension in discharge (P⫽0.038) and 72% greater odds of death at 1 year
58.1%, ischemic cardiac disease in 26.0%, atrial fibrillation (P⫽0.001). The relationship between Charlson Index and
in 12.7%, and congestive heart failure in 7.7%. More than discharge status remained if the Rankin Index was considered
76% of the patients had a history of cigarette smoking. On the as a full rather than a dichotomized scale. Every 1-point
basis of the TOAST classification scheme, 13.5% of patients increase in the Charlson Index was associated independently
had large-vessel atherothrombotic, 18.8% cardioembolic, and with a 15% increase in the odds of a poor outcome at
24.7% lacunar subtypes, with the remainder having strokes of discharge (P⬍0.005) and a 29% increase in the odds of death
undetermined or multiple causes. Because this was a veteran by 1 year (P⬍0.001). A high Charlson Index was also
population, only 1.6% of the patients were women. The mean associated with 32% increased odds of death at hospital
admission Canadian Neurological Scale score was 8.4 discharge (P⫽0.266) and 60% increased odds of 30-day
(⫾2.5). Figure 1 gives the distribution of Charlson Index
scores for the entire cohort (57.4% of patients had Charlson
scores of either 0 or 1. Overall, 7.9% of patients died by 1
month after stroke, and 20.8% died by 1 year. Among patients
surviving at discharge, 15.8% had an mRS of 0; 31.9%, 1;
3.5%, 2; 13.0%, 3; 27.6%, 4; and 8.1% had an mRS of 5.
Table 1 gives the proportion of patients having each
condition on the basis of the indicated discharge ICD-9-CM
codes. Other ICD-9-CM codes that may appropriately reflect
diagnoses within the indicated categories exist but were not
used for any patient in this cohort. Table 1 also gives the
frequencies that each ICD-9-CM code was assigned. Table 2
gives ICD-9-CM codes used in ⬎1% of the patients in the
cohort, the proportions of patients (n⫽960) assigned each Figure 2. Relationships between Charlson Index and mRS at
code, and the proportion of patients with a condition included hospital discharge and 1-year mortality.
 1944 Stroke August 2004
mortality (0.066), but the differences were not significant
after adjustment for stroke severity (P⬍0.001). Inclusion of
age as a covariate yielded essentially equivalent results.
(Compared with models that included only severity, the odds
ratios for the Charlson Index in models that controlled for
both age and severity increased by 0.9% to 3.5%.)
Discussion
These data show that both functional outcome at the time of
hospital discharge and 1-year mortality rates are associated
independently with the number and severity of patients’
comorbid diseases as reflected in a modified version of the
Charlson Index. This analysis specifically extends the useful-
ness of the Charlson Index as a measure of comorbidity for
studies focused on acute ischemic stroke.
There are a variety of parameters that may reflect the
validity of a scale. Content validity indicates how well a scale
includes domains thought to be relevant to a condition.
Convergent validity is demonstrated when scales or items that
are thought to measure the same construct have high corre-
lation coefficients. Divergent validity is demonstrated when
items or scales thought to measure different constructs have
low correlation coefficients. In this case, there is no “gold
standard” measure of aggregate comorbidity for comparison,
so these types of validity cannot be assessed. The present
study does provide evidence of both known groups and
predictive validity of the Charlson Index in this setting.
Known groups validity tests for differences in scores from
patients known to be clinically different (in this study, they
Downloaded from http://ahajournals.org by on April 23, 2023
are reflected in discharge status and 1-year mortality). Pre-
dictive validity reflects how well a scale predicts future
events.
It should be noted that this analysis was designed to
determine whether the Charlson Index would provide a valid
comorbidity adjustment for stroke outcome studies and not
intended to provide a clinically applicable prognostic model.
A variety of such models exist that were developed to address
particular questions.7–11 In addition, several recent studies
that generally require prospective data collection have de-
vised and tested prognostic models for use in specific
settings. For example, the Combined Lysis of Thrombus in
Brain Ischemia Using Transcranial Ultrasound and Systemic
TPA investigators assessed the relative contributions of
recanalization during the first hours of middle cerebral artery
occlusion and the clinical and neuroradiological data in predicting
good outcome (mRS ⱕ2) 3 months after stroke.12 Recanalization
within 300 minutes (odds ratio [OR], 4.11; 95% CI, 2.42 to 6.95),
baseline National Institutes of Health Stroke Scale (NIHSS) score
(OR, 0.35; 95% CI, 0.16 to 0.78), stroke volume on brain computed
tomography scan (Alberta Stroke Program Early CT score;13 OR,
2.98; 95% CI, 1.13 to 7.85), systolic blood pressure (OR, 0.32;
95% CI, 0.13 to 0.76), and proximal occlusion (OR, 0.25; 95%
CI, 0.10 to 0.61) were associated independently with good
outcome at 3 months. The German Stroke Study Collaboration
validated 2 models predicting complete functional recovery
(Barthel Index ⬎95) and mortality at 3 months on the basis of
patients’ age and NIHSS score obtained within 6 hours of
symptom onset.14 Severely affected patients were not included in
the analysis. The models explained 51% of the variance for
complete functional recovery and 30% of the variance in
mortality.
Although the Charlson Index would be useful to adjust for
comorbidity in studies with prospective data collection, the
present analysis supports its use in outcome studies involving
large databases that often lack clinical information. These
databases record comorbid diseases using ICD-9-CM codes.
As shown in Table 2, a few ICD-9-CM codes accounted for
the majority of conditions in each Charlson Index category. In
addition to those codes actually used in stroke patients
included in this cohort as reflected in the tables, other
ICD-9-CM codes are available that can also reflect diseases
included in the Charlson Index. It is well recognized that
ICD-9-CM codes may be inaccurate, including their applica-
tion to stroke conditions.15–17 There was no attempt to
confirm the accuracy of ICD-9-CM discharge coding for
comorbid conditions as applied in the present study because
such confirmation would not be feasible in large database
outcome studies in which records generally cannot be linked
to a patient’s medical records. Moreover, there may be
undercoding of comorbid conditions among patients who die
in the hospital. Finally, this modified version of the Charlson
Index ignores previous strokes, the impact of which is only
partially reflected in the Canadian Stroke Scale. Additional
studies could be conducted in nonveteran populations to
further investigate the utility of case-mix adjustment on the
basis of the modified Charlson Index. Despite these limita-
tions, ICD-9-CM coding remained related independently to
both functional status at hospital discharge and 1-year mor-
tality, supporting its usefulness for this purpose.
Acknowledgments
This work was supported by a grant from the Department of Veterans
Affairs (SDR 93-003 to R.D.H. and D.B.M.). L.B.G. was supported
in part by a National Institutes of Health mid-career development
award in patient-oriented research (K24 NS02165). R.D.H. was
supported in part by a grant from the VA Cooperative Studies/
Epidemiologic Research and Information Center Programs (CSP/
ERIC 602).
References
1. 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;40:373–383.
2. Jones MR, Horner RD, Edwards LJ, Hoff J, Armstrong SB, Smith-
Hammond CA, Matchar DB, Oddone EZ. Racial variation in initial stroke
severity. Stroke. 2000;31:563–567.
3. Goldstein LB, Jones MR, Matchar DB, Edwards LJ, Hoff J, Chilukuri V,
Armstrong B, Horner RD. Improving the reliability of stroke subgroup
classification using the Trial of ORG 10172 in Acute Stroke Treatment
(TOAST) criteria. Stroke. 2001;32:1091–1097.
4. Goldstein LB, Chilukuri V. Retrospective assessment of initial stroke
severity with the Canadian Neurological Scale. Stroke. 1997;48:
1181–1184.
5. Bushnell CD, Johnston DCC, Goldstein LB. Retrospective assessment of
initial stroke severity: comparison of the NIH Stroke Scale and the
Canadian Neurological Scale. Stroke. 2001;32:656 – 660.
6. Rankin J. Cerebral vascular accidents in patients over the age of 60. II.
Prognosis. Scott Med J. 1957;2:200 –215.
7. Johnston KC, Connors AFJ, Wagner DP, Knaus WA, Wang X, Haley
ECJ. A predictive risk model for outcomes of ischemic stroke. Stroke.
2000;31:448 – 455.
8. Hankey GJ, Jamrozik K, Broadhurst RJ, Forbes S, Burvill PW, Anderson
CS, Stewart-Wynne EG. Five-year survival after first-ever stroke and
 Goldstein et al
related prognostic factors in the Perth Community Stroke Study. Stroke.
2000;31:2080 –2086.
9. Stineman MG, Maislin G, Fiedler RC, Granger CV. A prediction model
for functional recovery in stroke. Stroke. 1997;28:550 –556.
10. Modell JG, Foster NL, DuPree ES, Ackermann RJ, Petry NA, Bluemlein
LE, Kuhl DE. Prognostication of recovery following stroke using the
comparison of CT and technetium-99m HM-PAO SPECT. J Nucl Med.
1990;31:61– 66.
11. Loewen SC, Anderson BA. Predictors of stroke outcome using objective
measurement scales. Stroke. 1990;21:78 – 81.
12. Molina CA, Alexandrov AV, Demchuck AM, Saqqur M, Uchino K,
Alvarez-Sabin J. Improving the predictive accuracy of recanalization on
stroke outcome in patients treated with tissue plasminogen activator.
Stroke. 2004;35:151–157.
13. Barber PA, Demchuk AM, Zhang J, Buchan AM. Validity and reliability
of a quantitative computed tomography score in predicting outcome of
Downloaded from http://ahajournals.org by on April 23, 2023
Charlson Index for Ischemic Stroke 1945
hyperacute stroke before thrombolytic therapy. Lancet. 2000;355:
670 –1674.
14. Weimar C, König IR, Kraywinkel K, Ziegler A, Diener HC. Age and
National Institutes of Health Stroke Scale score within 6 hours after onset
are accurate predictors of outcome after cerebral ischemia. Stroke. 2004;
35:158 –162.
15. Benesch C, Witter DMJ, Wilder AL, Duncan PW, Samsa GP, Matchar
DB. Inaccuracy of the International Classification of Diseases
(ICD-9-CM) in identifying the diagnosis of ischemic cerebrovascular
disease. Neurology. 1997;49:660 – 664.
16. Goldstein LB. Accuracy of ICD-9-CM coding for the identification of
patients with acute ischemic stroke: effect of modifier codes. Stroke.
1998;29:1602–1604.
17. Leibson CL, Naessens JM, Brown RD, Whisnant JP. Accuracy of hospital
discharge abstracts for identifying stroke. Stroke. 1994;25:2348 –2355.