Survival Prediction in Patients With Bladder Cancer Undergoing Radical Cystectomy Using a Machine Learning Algorithm:...

Area of Science:

• Oncology
• Medical Informatics
• Artificial Intelligence

Background:

• Traditional statistical models struggle with bladder cancer survival prediction post-cystectomy.
• Radical cystectomy has a high rate of metastasis (50% within 2 years).
• Artificial intelligence (AI) integration can enhance prognostic accuracy and treatment personalization.

Purpose of the Study:

• Develop and evaluate a machine learning algorithm for predicting disease-free survival (DFS), overall survival (OS), and cause of death in bladder cancer patients.
• Utilize a comprehensive dataset of clinical and pathological variables for prediction.
• Enhance prognostic accuracy and personalize treatment strategies for bladder cancer patients.

Main Methods:

• Retrospective analysis of 370 bladder cancer patients undergoing radical cystectomy.
• Employed the CatBoost algorithm for regression (DFS, OS) and binary classification (tumor-related death).
• Assessed model performance using Mean Absolute Error (MAE) and F1-score, with 5-fold cross-validation and SHAP values for interpretability.

Main Results:

• CatBoost model achieved MAE of 18.68 months for DFS and 17.2 months for OS (improved to 14.6 months after feature filtering).
• For tumor-related death classification, the model achieved 78.6% recall and 0.44 F1-score.
• Key predictors included clinical/pathological tumor stage, systemic immune-inflammation index (SII), and bladder tumor position.

Conclusions:

• The AI model shows significant promise in predicting survival and cause of death for bladder cancer patients post-cystectomy.
• Clinical and pathological tumor staging, SII, and tumor position are crucial predictive factors.
• AI offers an objective, data-driven tool to enhance personalized prognostic assessment and guide clinical decisions.