Use of a self-report-generated Charlson Comorbidity Index for predicting mortality.

BACKGROUND: The Charlson Comorbidity Index, a popular tool for risk adjustment, often is constructed from medical record abstracts or administrative data. Limitations in both sources have fueled interest in using patient self-report as an alternative. However, little data exist on whether self-reported Charlson Indices predict mortality.
OBJECTIVES: We sought to determine whether a self-reported Charlson Index predicts mortality, its performance relative to indices derived from administrative data, and whether using study-specific weights instead of Charlson's original weights enhances model fit.
METHODS: We surveyed 7761 patients admitted to a university medical service over the course of 4 years and extracted their administrative data. We constructed 6 different Charlson indices by using 2 weighting schemes (original Charlson weights and study-specific weights) and 3 different datasources (ICD-9CM data for index hospitalization, ICD-9CM data with a 1-year look-back period, and patient self-report of comorbidities.) Multivariate models were constructed predicting 1-year mortality, log total costs, and log length of stay.
RESULTS: The 6 measures of the Charlson index all predicted 1-year mortality. Models with age and gender, with or without diagnosis-related group, had approximately the same predictive power regardless of which of the 6 Charlson indices were used. Nevertheless, there were small improvements in model fit using administrative data versus self-report, or study-specific versus original weights. All models obtained areas under the receiver operating curve of 0.70 to 0.77.
CONCLUSIONS: Overall, self-reported Charlson indices predict 1-year mortality comparably with indices based on administrative data. Administrative data may offer some small improvements in predictive ability and may be preferred when readily available.