By Sarianne Gruber
Twitter: @subtleimpact
Before we had patient panels and electronic health records, there were patient study logs and regulatory binders. Yet, the same questions are asked today when analyzing study outcomes as yesteryear. How do you assess a treatment, procedure or process for a specific disease in your population sample with the presence of confounding diseases? The answer is a Comorbidity Index. It is a method of categorizing comorbidities of patients. Each comorbidity category has an associated weight, based on the adjusted risk of mortality, and the total of all the weights is the assigned comorbidity score per patient. Whatever the disease of interest is, whether it is to determine the likelihood for readmission, the potential risk for early disease onset, or clinical outcomes and hospital cost, a comorbidity index score is commonly used as a predictor.
An Index with Longevity
Almost thirty years later, articles continue to be written on the Charlson Index as recent as this month’s Medical Care publication by Austin et al. on “Why Summary Comorbidity Measures Such as the Charlson Comorbidity and Elixhauser Score Work”. Their results “provide an analytical proof of the utility of comorbidity summary measures when used in place of individual comorbidities”. Researcher Dr. Hude Quan continues to re-evaluate and update the indexes with more recent data. In his 2011 study, “Updating the Charlson Comorbidity Index and Score for Risk Adjustment in Hospital Discharge Abstracts Using Data from 6 Countries” published in the American Journal of Epidemiology, he noted that “since the publication of Charlson et al.’s original article in 1987, the paper has been cited nearly 5,000 times”. He also stated that “the index has been validated for its ability to predict mortality in various disease subgroups, including cancer, renal disease, stroke, intensive care, and liver disease. These studies consistently demonstrate that the Charlson index is a valid prognostic indicator for mortality.” Dr. Quan’s new Charlson scores for 12 comorbidities “showed good-to-excellent discrimination in predicting in-hospital mortality in data from 6 countries”.
Over the years, there have been enhancements to the Index presented in various research studies.
Here is a helpful chronological list:The original Charlson Index was developed with 19 categories in 1987
- The Index was modified to 17 categories by Deyo et al. in 1992,
- The list of specific ICD diagnosis codes was modified by Romano et al. in 1993.
- The ICD-10 coding update by Halfon et al. in 2002 and Quan et al. in 2005.
- The original weights developed for use with the Index were modified by Halfon et al. 2003 and Schneeweiss et al. in 2003.
Dr. Mary Charlson and her associates Dr. Peter Pompei, Dr. Kathy Ales and Dr. C. Ronald MacKenzie, published their milestone work “A New Method of Classifying Prognostic Comorbidity in Longitudinal Studies: Development and Validation” in the Journal of Chronic Disease in 1987. The Cornell University Medical College research team’s objective was to develop a prospective methodology for comorbidity classification for longitudinal studies. They set out to create “a prognostic taxonomy for comorbid conditions which singly or in combination might alter the risk of short term mortality for patients enrolled in longitudinal studies”. I decided it was about time I read the original study.
Creating Training and Testing Populations
The dilemma the team was trying to solve was how to account for comorbid diseases and aging in a longitudinal study. Charlson and team decided the best starting point was to conduct a prospective study. They collected the number and severity of comorbid diseases from cohort of 604 patients, admitted for various causes to New York Hospital, during a one month period. Demographic and clinical characteristics, complications, arrests, deaths, status was collected at discharge. Survival was tracked in months from the date of admission to the hospital to the date of death or to 1 year after admission for 93% of the patients studied. For a validation dataset, they used a historical study of 685 patients, who had been treated for primary breast cancer at Yale New Haven Hospital between 1962 and 1969. The study included the number and severity of comorbid diseases at the 5 year and 10 year follow–ups. Survival was also measured in months, with data on deaths due to either to breast cancer or to comorbid disease. Survival analysis was performed on comorbid death only.
Developing a Comorbidity Index
The first step was to classify patients according to each comorbid disease so that a patient with diabetes and hypertension would be classified in both categories. Next, comorbid condition survival rates were calculated for both the training and testing population by Cox’s regression method for life table data. Adjusted relative risk estimates of comorbid death were used to control for (1) the effect of coexistent diseases, (2) illness severity and (3) reason for admission. In addition, to use the comorbidity index in prospective studies, the team decided to use the Hutchinson and Thomas method to create a scoring system that combined both age and comorbidity.
It was important that an index took into account both the number and the seriousness of comorbid diseases. The team decided to use the adjusted relative risks as weights for the different comorbid diseases. To make the weighting system simple, weights from 1 to 6 (low to high) were assigned based on a stratification of relative risk scores of the conditions. Each condition was then assigned a weight (see Table 1). If a patient has multiple conditions such as Dementia (weight = 1), Any tumor (weight = 2) and Moderate or severe liver disease (weight = 3), the comorbidity index score equals 6.
Table 1 Assigning Weights for Diseases
| Conditions | Assigned Weights for Diseases |
| Myocardial infarctCongestive heart failurePeripheral vascular diseaseCerebrovascular disease
Dementia Chronic Pulmonary disease Connective tissue disease Ulcer disease Mild liver disease Diabetes |
1 |
| HemiplegiaModerate or severe renal diseaseDiabetes with end organ damageAny tumor
Leukemia Lymphoma |
2 |
| Moderate or severe liver disease |
3 |
| Metastic solid tumorAIDS |
6 |
Validating the Comorbidity Index
The prevalence of comorbid disease in the breast cancer patients was significantly lower than the cohort of medical patients. The only significant predictors of risk of comorbid death were age and comorbidity. In the medical patient sample age was not a predictor of comorbid death. The team was not surprised since the follow-up period was only one year. However, it is noted that in “longitudinal studies with follow-up periods of 5 yr or more, both age and comorbidity should be taken into account as predictors of death from comorbid disease” and the recommendation was to create a combined age-comorbidity variable. The validation risk scores with 10 year survival rates are presented in the original paper, and it states that “the actual survival and the estimated survivals are quite close”.
“A New Method of Classifying Prognostic Comorbidity in Longitudinal Studies: Development and Validation” authored by Charlson et al. was the first of its kind “that tackles the issue of validating a method of measuring the prognostic impact of comorbid disease”. I highly recommend giving this paper a read even if you have to dust off a statistics text book.