Deep learning-based survival analysis of bladder cancer patients in the Putuo District, Shanghai, China.

Background: Bladder cancer poses significant health risks and necessitates effective public health management. Objective: To develop a deep-learning survival prediction model using TabNet and compare its performance with logistic regression. Methods: Data on bladder cancer patients were collected from the Putuo District subset of Shanghai Cancer Registration and Reporting System. A total of 620 patients were included, divided into a training cohort (n=434) and a validation cohort (n=186). Logistic regression analyses were conducted to identify risk factors, while the TabNet framework was used to develop a deep learning-based model. Model performance was evaluated using ROC curves, decision curve analysis, and calibration curves. Shapley Additive Explanations (SHAP) was applied to interpret feature importance. Results: Baseline characteristics showed no significant differences between the training and validation cohorts (P>0.05). The TabNet model demonstrated high discriminative ability in predicting both 5-year OS and CSS within the training cohort, with net benefits surpassing those of logistic regression, and showed good calibration. In the validation cohort, the TabNet model exhibited excellent performance in predicting 5-year OS and CSS. SHAP analysis revealed that age, T stage, and N stage were the most influential factors. Conclusion: The TabNet model showed robust performance in predicting bladder cancer survival, offering valuable insights for community-based management and follow-up strategies.