Explainable machine learning model predicts response to adjuvant therapy after radical cystectomy in bladder cancer

Area of Science:

• Urology
• Oncology
• Medical Informatics

Background:

• Radical cystectomy (RC) is standard for bladder cancer, but recurrence is common.
• Adjuvant therapies improve outcomes, yet response varies due to tumor heterogeneity.
• Predictive models are crucial for personalized treatment post-RC.

Purpose of the Study:

• Develop and validate machine learning models to predict adjuvant therapy response in bladder cancer patients.
• Identify key pathological, demographic, and molecular features influencing treatment outcomes.
• Enhance individualized treatment strategies for bladder cancer post-RC.

Main Methods:

• Retrospective analysis of bladder cancer patients undergoing RC.
• Utilized LASSO regression for feature selection and nine machine learning algorithms for model development.
• Assessed model performance using AUC and SHAP for interpretability.

Main Results:

• A random forest model achieved high predictive performance (AUC=0.92 training, 0.74 testing).
• Vascular invasion, perineural invasion, and PD-L1/HER2 expression were key predictive features.
• Decision curve analysis indicated favorable clinical utility.

Conclusions:

• Machine learning models integrating diverse features show potential for predicting adjuvant therapy response.
• External validation revealed performance limitations, emphasizing the need for further research.
• Prospective, multi-center studies are recommended to improve model generalizability and clinical application.