Comparative performance of comorbidity indices for estimating perioperative and 5-year all cause mortality following radical cystectomy for bladder cancer

Abstract

Purpose: Radical cystectomy continues to be associated with a nonnegligible risk of perioperative death and all cause mortality in the years after surgery remains relatively high. We investigated the comparative ability of various comorbidity indices to predict perioperative and 5-year all cause mortality after radical cystectomy.

Materials and methods: We evaluated 891 patients who underwent radical cystectomy between 1994 and 2005. The associations of American Society of Anesthesiologists (ASA) score, Charlson comorbidity index, Elixhauser index and ECOG (Eastern Cooperative Oncology Group) performance status with outcomes were assessed using Cox regression models. Model performance was compared with area under receiver operating curves.

Results: A total of 33 (3.7%) patients died within 90 days of radical cystectomy. On multivariate analysis locally advanced pathological tumor stage (HR 4.86, p = 0.002) as well as Elixhauser index (HR 1.48, p = 0.002), ASA score (HR 3.17, p = 0.001) and ECOG (HR 2.40, p <0.0001) were significantly associated with 90-day perioperative mortality. Median followup after radical cystectomy was 10.1 years, during which time 576 patients died. Charlson comorbidity index (HR 1.23, p <0.0001), Elixhauser index (HR 1.28, p <0.0001), ASA score (HR 1.44, p = 0.007) and ECOG (HR 1.97, p <0.0001) were independent predictors of 5-year all cause mortality. Moreover Charlson comorbidity index (AUC 0.798, p <0.0001), Elixhauser index (AUC 0.770, p = 0.03) and ECOG (AUC 0.769, p = 0.03) significantly enhanced the performance of a base model which did not include comorbidity status (AUC 0.757) to predict 5-year all cause mortality.

Conclusions: Comorbidity status is predictive of perioperative death and 5-year all cause mortality after radical cystectomy and, therefore, should be incorporated into patient counseling and risk stratification models. Further prospective studies are warranted to overcome the retrospective limitations in determining the relative prognostic value of various comorbidity indices.