Evaluation of two competing methods for calculating Charlson's comorbidity index when analyzing short-term mortality using administrative data

The performance and predictive power of the Deyo-Charlson and the Romano-Charlson comorbidity indices were compared when short-term mortality after hospitalization was the outcome of interest. These indices are commonly used to adjust for the effect of comorbidities when using administrative data in comparative studies. In hospital Medicare clam data for patients admitted to one of six medical categories (back pain, stroke, pneumonia, hip replacement, transurethral radical prostatectomy, or lysis of peritoneal adhesion), were selected for analyses. Logistic regression models were employed to evaluate the relative importance and the explanatory power of these indices for predicting mortality 30, 90, and 180 days after admission. The contribution of each index to mortality was assessed via the likelihood ratio chi-square statistic (G2), and the area under the receiver operator characteristic (ROC) curve was used to assess predictive power. Indices were evaluated using weights suggested by Charlson et al. and using empirically derived weights. Both indices improved the base model equally, although the predictive power of both indices was poor with values of the C statistic ranging from 0.60 to 0.78. Our results suggest limited applicability of these approaches when examining short-term mortality. A slight increase in predictive power was observed when indices were calculated using empirical weights derived from our data.