|
Charlson Index (CI)
Please note that this section is an archive (last updated in June 2006). [disclaimer]
Sections:
Overview |
Instrument Reviews |
Construct Overviews |
Book Compendium Reviews |
Internet Site Reviews
|
|
|
|
Practical Information
|
Instrument Name:
|
Charlson Index (CI)
|
Instrument Description:
|
The Charlson Index (CI) was developed to predict the risk of mortality from co-morbid illness within a one-year period for use in longitudinal studies. Severity weights were developed (based on estimation of relative risks of death) for 19 comorbid conditions. This list originally contained 30 clinically important illnesses, but conditions whose relative risk was less than 1.2 were dropped. (Ref: 1). The CI does not take into account age as a predictor of death. The developers of the CI developed a combined age-comorbidity variable by adding 1 to the CI for every decade over 40.
|
Price:
|
Not applicable.
|
Administration Time:
|
Not applicable.
|
Publication Year:
|
1987
|
Item Readability:
|
Readability is not an issue since the index is researcher-generated, not patient self-report.
|
Scale Format:
|
The CI uses a weighted scoring approach. Weights were empirically derived.
|
Administration Technique:
|
Researcher-generated.
|
Scoring and Interpretation:
|
Researchers determined that age and comorbidity had a significant impact on death rates. To determine the risk of mortality of an individual, illnesses are given risk scores from the list of comorbid illnesses.
When the index was developed, researchers assessed the number and seriousness of certain illnesses over a one-year time span. Relative risks were determined for each illness and those with scores less than 1.2 were deleted. Relative risk scores at or above 1.2 were assigned weight scores as follows:
| Relative Risk Scores
|
Weighted scores
|
| ≥ 1.2 < 1.5
|
1
|
| ≥ 1.5 < 2.5
|
2
|
| ≥ 2.5 < 3.5
|
3
|
| 6 or more
|
6
|
|
Forms:
|
No information found.
|
|
|
Research Contacts
|
Instrument Developers:
|
Mary E. Charlson, Peter Pompei, Kathy L. Ales, and C. Ronald MacKenzie
|
Instrument Development Location:
|
Cornell University Medical College
1300 York Avenue
New York, NY 10021, U.S.A.
|
Instrument Developer Email:
|
No information found.
|
Instrument Developer Website:
|
No information found.
|
|
|
Annotated Bibliography
|
1. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A New Method of Classifying Prognostic Comorbidity in Longitudinal Studies: Development and Validation. J Chron Dis 1987;40(5):373-83. [PMID: 3558716]
| |
Purpose: To develop a scale to classify comorbid conditions that might alter the risk of 1-year mortality for patients enrolled in longitudinal studies. |
| |
Sample: The CI was developed based on data from a sample of 607 patients admitted to the New York Hospital-Cornell Medical Center during a 1 month period. The medical records of 3 patients could not be located, resulting in an initial sample size of 604 patients. One-year follow-up data were obtained for 559 (93%) of the initial sample of 604 patients. The authors call this sample the ¡¡ãtraining population.¡¡À Cross-validation of results was undertaken in a second sample comprised of 685 women with breast cancer who received their first breast cancer treatment between January 1, 1962 and December 31, 1969 at Yale New Haven Hospital. The authors call the cross-validation sample the ¡¡ãtesting population.¡¡À This population was followed at 5- and 10-year follow-up. Five-year follow-up data were obtained for 684 of the 685 patients, and 10-year follow-up data were obtained for 681 of the 685 patients. Deaths were attributed to breast cancer or to comorbid disease. |
| |
Methods: The authors compared several approaches for developing a co-morbidity index that would be predictive of mortality. The most predictive was the CI in which co-morbid conditions are assigned severity weights. The results were cross-validated in a sample of women with breast cancer. The CI retained its predictive value in this sample. Since the cross-validation sample was followed for a longer time (10 years), the investigators also experimented with an age-risk index. A combined, age-comorbidity variable was obtained by adding 1 point to the CI for risk of death from comorbid disease for every decade over 40 years of age. The combined variable improved predictive value in the testing population. |
| |
Implications: Prospective studies often use strict eligibility criteria to limit the impact of comorbid conditions on the study outcomes. The CI is an alternative method that allows researchers to stratify patients based on severity of comorbidities. For studies with longer follow-up times, including an age-risk factor can improve predictive validity. |
2. Deyo RA, Cherkin DC, Ciol MA. Adapting a Clinical Comorbidity Index for Use with ICD-9-CM Administrative Data. J Clin Epidemiol 1992;45(6):613-619. [PMID: 1607900]
| |
Purpose: To determine if an adapted version of the CI based on ICD-9-CM codes from an administrative database would be effective in predicting non-mortality outcomes (e.g., postoperative complications, length of stay, hospital charges, etc.) in a population who had undergone lumbar spine surgery. |
| |
Sample: Medicare (Part A) claims data from the Health Care Financing Administration (HCFA) were used to identify 27,111 patients with a mean age of 71.8 years. Women comprised 57.1% of the sample and the racial composition was 92.7% white, 3.7% black and 3.6% other race or ethnicity. |
| |
Methods: The original CI was adapted from its designated use with medical records for use with ICD-9-CM codes from the Medicare Part A administrative claims data. Most of the diagnoses and procedures listed in the CI could be matched with similar ICD-9-CM diagnoses and procedures. Researchers obtained a list from HCFA of patients that underwent lumbar spine surgery in 1985, linked data for hospitalizations one year before the procedure, and Medicare mortality data. Patients were excluded if a diagnosis of neoplasm, spinal infection, spine fracture, ankylosing spondylitis was discovered or if patients underwent a second major surgical procedure.
Results: The adapted Charlson comorbidity index was significantly related to in-hospital complications, blood transfusion, mean length of stay in days, total hospital charges, discharge to nursing home and 6-week post-operative mortality. Even when the analyses were repeated using only cross-sectional diagnoses recorded during the index 1985 hospitalization, the magnitude of outcome differences was maintained. |
| |
Implications: Researchers concluded that their ICD-9-CM based version of the CI is valuable when used with ICD-9-CM codes from administrative databases. This adapted instrument may be used in future studies to examine the outcomes of other medical procedures or medical services using administrative databases. |
3. Romano PS, Roos LL, Jollis JG. Adapting a Clinical Comorbidity Index for Use with ICD-9-CM Administrative Data: Differing Perspectives. J Clin Epidemiol 1993;46(10):1075-79. [PMID: 8410092]
| |
Purpose: To critique Deyo¡¯s et al. (ref. #2) adaptation of the CI for use with ICD-9 data and to compare Deyo¡¯s ICD-9-CM assignments to ones the authors developed previously (Dartmouth-Manitoba assignments), also based on ICD-9-CM codes. |
| |
Summary: The authors point out that there are differences between Deyo¡¯s ICD-9-CM definitions of comorbidities and those defined in the CI and conclude that the correspondence between the CI and the ICD-9-CM ¡¡ãis not intuitively obvious.¡¡À The authors also critique the CI itself noting, for example, the limits of its generalizability and the presence of ¡¡ãtoo few observations to rule out significant interactions among patients with multiple comorbidities.¡¡À |
| |
Implications: Tailored risk models, developed for a particular set of data and research question, have greater clinical face validity and statistical strength than the CI or either of the ICD-9-CM adaptations of the CI. The ICD-9-CM adapted CI may not be appropriate for modeling the impact of comorbidities on short-term outcomes using data from large administrative databases. |
|
|
top
|
Factors and Norms
|
Factor Analysis Work:
|
No information found.
|
Normative Information Availability:
|
No information found.
|
|
|
Reliability Evidence
|
Test-retest:
|
No information found.
|
Inter-rater:
|
No information found.
|
Internal Consistency:
|
No information found.
|
Alternate Forms:
|
No information found.
|
|
|
Validity Evidence
|
Construct/ Convergent/ Discriminant:
|
Researchers used the log-ranking test to compare the difference between comorbidity groups (mild, moderate, severe). It was found that the groups were significantly different (X=163, p < 0.0001). (Ref: 1) Furthermore, researchers compared the weighted index (CI) to the Kaplan-Feinstein methods and found that both techniques identified patients at low risk of comorbid death (p <0.001 for both methods) (Ref: 1). One study, in which the CI was adapted for use with ICD-9-CM scores, categorized patients into CI scores and compared variables among the categorized groups: mean age, in-hospital complication percentage, percentage of patients receiving blood transfusions, average length of stay, hospital charges, percentage of patients discharged to nursing homes and the percentage of mortality after 6 weeks of operation procedure. (Ref: 2) Results indicated that the differences between the groups were significant p < 0.0005 except for percentage of in-hospital complications (p=0.01) (Ref: 2). Romano et al. argue that the CI may compromise the clinical face validity of a risk adjustment model as well as the statistical strength of the model. This is because important risk factors pertaining to a specific condition or procedure may be overlooked when comorbidities are summarized into an ordinal index. (Ref: 3)
|
Criterion-related/ Concurrent/ Predictive:
|
Researchers found that the number of comorbidity diseases predicted death within a one-year period, p< 0.05. (Ref: 1) The prospective rating of disease severity was found to be the most significant predictor of in-hospital mortality (p<0.0001). (Ref: 1) Specifically, diagnosis with oncologic concerns and AIDS significantly correlated with outcomes of death within one year of hospitalization (p< 0.0001 for both groups) (Ref: 1). Furthermore, patients with tumors (p< 0.01), Leukemia (p< 0.01) Lymphoma (p <0.1) and moderate to severe liver concerns (p <0.01) had a significant increase of deaths within a year from hospitalization. (Ref: 1) The total number of comorbid diseases was a significant predictor of 1-year mortality (p<0.05), and the main differences in outcomes were between patients with no comorbid conditions and those with 1 or more comorbid conditions. (Ref: 1)
|
Content:
|
No information found.
|
Responsiveness Evidence:
|
No information found.
|
Scale Application in VA Populations:
|
Yes. (Ref: 1-2)
|
Scale Application in non-VA Populations:
|
No information found.
|
|
|
Comments
|
The CI would be most useful for studies that are similar to the ones in which the CI was developed and cross validated. Specifically, it is most appropriate for evaluating the impact of comorbid conditions on mortality in longitudinal studies. The practical importance of the CI is that it can avoid the need to use highly restrictive inclusion criteria in longitudinal studies.
The CI has been adapted in two ways: (1) mapping of CI comorbidities with ICD-9-CM codes, and (2) using the index to model the impact of comorbidities on outcomes other than mortality. These adaptations will not necessarily have the same predictive validity as the CI showed in its parent study. An alternative to adapting the CI for use in studies dissimilar to the parent study is to develop risk models tailored to a study's data source and prediction needs. (Ref: 3)
|
|
|
|
