A comparison between artificial neural networks and urologists' assessment of outcome in bladder cancer. II. Survival in muscle-invasive (T2-T4) tumours

Currently, we lack accurate methods of predicting survival in patients with muscle-invasive bladder cancer. Data relating to 40 such patients (out of a comprehensive database of 212 patients) was retrospectively analysed by artificial neural networks (ANNs). A total of 15 different factors including clinicopathological and molecular markers of mixed prognostic significance were used in the analysis. The accuracy of the ANN in predicting 12-months cancer-specific survival for T2-T4 cancers was 82%. This was subsequently compared with the predictions of four experienced urologists who analysed the same data blindly. The corresponding mean accuracy for the urologists was 65%.