Improving the Performance of Claims-Based Comorbidity Measures in Predicting In-hospital Mortality

Background: Comorbidity is an important controlling factor in health services research using administrative data. Our aim was to improve the performance of various claims-based comorbidity measures in predicting in-hospital mortality.

Methods: Five different comorbidity measures, including three different adaptations of the Charlson comorbidity index (CCI), one method developed by Elixhauser et al., and the Chronic Disease Score (CDS), were selected for investigation in this retrospective cohort study. New empirical weights were derived from Cox's regression models of all inpatients in 2001. The performance of comorbidity methods with new weights was validated and compared using the c-statistics derived from multiple logistic regression models among each of the following inpatients in 2002: AMI (n=8,961), type 2 DM (n=32,687), and CHF (n=15,972).

Results: We derived new weights for five comorbidity measures from 1,214,014 inpatients in 2001. For all the adaptations of CCI, some comorbid conditions had smaller weights than the original weights, such as rheumatologic disease, ulcer disease, hemiplegia or paraplegia, and AIDS, while dementia had larger weight than the original. The empirically derived weights of the CCI and the Elixhauser method outperformed than the methods with the original definition among each of study populations in 2002.

Conclusions: The findings of this study have indicated that CCI should be reassessed in cohorts with larger sample size. The empirically derived weights of CCI performed better than the original one, and suggest that the new weights should be used in further researches.