Optimizing the measurement of comorbidity for a South Australian colorectal cancer population using administrative data

Abstract

Rationale and objectives: In epidemiological research, it is essential to account for the confounding effects of factors such as age, stage, and comorbidity for accurate prediction of cancer outcomes. There are several internationally developed and commonly used comorbidity indices. However, none are regarded as the gold-standard method. This study will assess and compare the predictive validity of established indices for use in a South Australian (SA) colorectal cancer (CRC) population against a local index. Furthermore, the prognostic influence of comorbidity on survival is investigated. Methods: A population-based study of patients diagnosed with CRC from 2003 to 2012 and linked to in-hospital data to retrieve comorbidity information was conducted. The predictive performance of established indices, Charlson comorbidity index (CCI), National Cancer Institute comorbidity index (NCI), Elixhauser comorbidity index (ECI), and C3 index was evaluated using the Fine and Gray competing risk regression and reported using measures of calibration and discrimination, area under the curve (AUC), and Brier score. Furthermore, to identify the optimal index, a local CRC comorbidity index (CRCCI) was also developed and its performance compared with the established indices. Results: Comorbidity models adjusted for age, sex, and stage showed that all indices were good predictors of mortality as measured by the AUC (CCI: 0.738, NCI: 0.742, ECI: 0.733, C3: 0.739). CRCCI had similar mortality prediction as established indices (CRCCI: 0.747). There was a significant increase in cumulative risk of noncancer and CRC-specific mortality with increase in comorbidity scores. The two most prevalent comorbidities were hypertension and diabetes. Conclusions: The existing indices are still valid for adjusting for comorbidity and accurately predicting mortality in an SA CRC population. Internationally developed indices are preferred when policymakers and researchers wish to compare local study results with those of studies (national and international) that have used these indices. Comorbidity is a predictor of mortality and should be considered when assessing CRC survival.