Development and Validation of Prognostic Nomograms to Predict Overall and Cancer-Specific Survival for Patients with Adenocarcinoma of the Urinary Bladder: A Population-Based Study

Bladder Cancer

J Invest Surg. 2020 Aug 27:1-8. doi: 10.1080/08941939.2020.1812776. Online ahead of print.


BACKGROUNDS: Adenocarcinoma of the bladder (ACB) rarely occurs but is associated with poor outcome. We aim to establish reliable nomograms for estimating cancer-specific survival (CSS) and overall survival (OS) of ACB patients.

METHODS: ACB patients were retrieved from the Surveillance, Epidemiology, and End Results (SEER) database (2004-2015). A total of 1,149 patients were randomly divided into training cohort (n = 692) and validation cohort (n = 457). Multivariate Cox proportional hazards regression models were employed to identify independent prognostic factors. Nomograms predicting OS and CSS were constructed utilizing screened factors. The performance of nomograms was internally and externally validated by calibration curves, the receiver operating characteristic (ROC) curves, concordance index (C-index), and decision curve analysis (DCA).

RESULTS: OS nomogram incorporated age, race, histologic grade, American Joint Committee of Cancer (AJCC) stage, metastasis, surgery, chemotherapy, and tumor size. The C-indices were 0.754 (95% CI: 0.732-0.775) for training set and 0.743 (95% CI: 0.712-0.767) for validation set. Meanwhile, the calibration plots for 3- and 5-year OS displayed fine concordance between actual and predicted outcomes. In addition, higher areas under the curve (AUCs) were seen in training cohort (3-year: 0.799 vs. 0.630; 5-year: 0.797 vs. 0.648) and validation cohort (3-year: 0.802 vs. 0.662; 5-year: 0.752 vs. 0.660). Finally, DCA curves of the nomograms exhibited larger net benefits than AJCC stage. CSS nomogram showed similar results.

CONCLUSION: Our study constructed and validated nomograms with improved discriminative abilities and clinical benefits to predict the survival outcomes of ACB patients. The models might assist clinicians in optimizing therapeutic management on individual levels.