From Stage to Systems: Integrating Standard Laboratory Markers into Bladder Cancer Prognostic Models

Background/Objectives: Bladder cancer (BC) remains a globally prevalent malignancy with significant clinical heterogeneity. While the TNM staging system is the diagnostic standard, its ability to provide individualized, comprehensive prognostic insights is limited. This study evaluates whether integrating routine laboratory markers into machine learning (ML) frameworks can enhance mortality risk prediction beyond traditional staging. Methods: Routine electronic health records from 337 adult patients with confirmed BC (2010–2023) were analyzed. Six ML architectures were evaluated for mortality risk prediction: logistic regression (GLM), elastic net logistic regression (GLMNET), generalized additive models (GAM), extreme gradient boosting (XGBoost), gradient boosting machines (GBM), and support vector machines (SVM). Results: All models demonstrated stable performance, with the SVM achieving the highest pooled AUC (0.679, 95% CI 0.57–0.77). Feature importance analyses identified tumor grade and routine blood markers, specifically platelet count, bicarbonate (CO2), and random glucose, as the highest-ranking predictors, demonstrating a greater relative feature weight than individual traditional anatomical staging parameters within the models. Conclusions: Integrating standard hematologic and metabolic panels into ML models offers a potential, cost-effective avenue to augment risk stratification, warranting further external validation before clinical implementation.