PRACTICAL CONSIDERATIONS ON THE USE OF THE CHARLSON COMORBIDITY INDEXWITH ADMINISTRATIVE DATA-BASES

To develop a measure of the burden of comorbid disease from the MED-EC HO data base (Quebec), the so called Charlson index was adapted to Int ernational Classification of Disease (ICD-9) codes. The resulting como rbidity index was applied to the study of inpatient death in 33,940 pa tients with ischemic heart disease. Multiple logistic regression was u sed to relate inpatient death to its predictors, including gender, pri ncipal diagnosis, age, and the comorbidity index. Various transformati ons of the comorbidity score were performed, and their effect on the p redictive 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 du mmy independent variables (the area under the receiver-operating curve is then 0.87). In a validation analysis performed on 1990-1991 MED-EC HO data (36,012 admissions with ischemic heart disease), the comorbidi ty index has the same statistical properties. We conclude that the Cha rlson index may be an efficient approach to risk adjustment from admin istrative data bases, although it should be tested on other conditions . Copyright (C) 1996 Elsevier Science Inc.