Practical considerations on the use of the Charlson comorbidity index with administrative data bases

Abstract

To develop a measure of the burden of comorbid disease from the MED-ECHO data base (Québec), the so-called Charlson index was adapted to International Classification of Disease (ICD-9) codes. The resulting comorbidity index was applied to the study of inpatient death in 33,940 patients with ischemic heart disease. Multiple logistic regression was used to relate inpatient death to its predictors, including gender, principal diagnosis, age, and the comorbidity index. Various transformations of the comorbidity score were performed, and their effect on the predictive accuracy was assessed. The comorbidity index was constantly and strongly associated with death. From a statistical viewpoint, the best results were obtained when the index was transformed into four dummy independent variables (the area under the receiver-operating curve is then 0.87). In a validation analysis performed on 1990-1991 MED-ECHO data (36,012 admissions with ischemic heart disease), the comorbidity index has the same statistical properties. We conclude that the Charlson index may be an efficient approach to risk adjustment from administrative data bases, although it should be tested on other conditions.