Prediction of Survival After Radical Cystectomy for Invasive Bladder Carcinoma: Risk Group Stratification, Nomograms or Artificial Neural Networks?

Purpose: We compared 3 predictive models for survival after radical cystectomy, risk group stratification, nomogram and artificial neural networks, in terms of their accuracy, performance and level of complexity. Materials and Methods: Between 1996 and 2002, 1,133 patients were treated with single stage radical cystectomy as monotherapy for invasive bladder cancer. A randomly selected 776 cases (70%) were used as a reference series. The remaining 357 cases (test series) were used for external validation. Survival estimates were analyzed using univariate and then multivariate appraisal. The results of multivariate analysis were used for risk group stratification and construction of a nomogram, whereas all studied variables were entered directly into the artificial neural networks. Results: Overall 5-year disease-free survival was 64.5% with no statistical difference between the reference and test series. Comparisons of the 3 predictive models revealed that artificial neural networks outperformed the other 2 models in terms of the value of the area under the receiver operator characteristic curve, sensitivity and specificity, as well as positive and negative predictive values. Conclusions: In this study artificial neural networks outperformed the risk group stratification model and nomogram construction in predicting patient 5-year survival probability, and in terms of sensitivity and specificity.