The objective of this article was to evaluate the use of the Charlson comorbidity index (CCI) to predict inpatient death in Ribeirão Preto, São Paulo State, Brazil. 54,680 hospitalizations from January 1996 to December of 1997 were analyzed. Two International Classification of Diseases adaptations of CCI were compared, and the 30 clinical conditions assessed by the Charlson index were reviewed. Logistic regression was used to evaluate the models' capacity to predict inpatient death. The baseline model included: age, sex, and principal diagnosis. Differences in ICD adaptations showed little effect on the models' discriminatory capacity. Revision of the 30 clinical conditions increased the model's discriminatory capacity to predict death (C statistic = 0.73), as compared to the model with the original CCI (C statistic = 0.72). All tested models had a reduced effect on the baseline model's discriminatory capacity (C statistic = 0.70). The results show the importance of Brazil having an information system that allows a complete description of hospital morbidity in order to monitor health service performance.
Hospital Mortality; Comorbidity; Clinical Diagnosis
