Explainable machine learning predicts overall survival in female bladder cancer patients after radical cystectomy

Comprehensive Summary

The study examined whether machine learning models are capable of accurately predicting overall survival in female bladder cancer patients after radical cystectomy. The authors analyzed 4,536 female patients, using data from the SEER database. They identified key clinical and pathological factors associated with survival, which were T stage, N stage, age, tumor size, marital status, chemotherapy, and the number of examined lymph nodes. Based on these factors, five machine learning models were developed and compared: SVM, KNN, Random Forest, XGBoost, and Gradient Boosting Decision Tree (GBDT). From these models, the GBDT performed best with high accuracy in predicting 1-, 3-, and 5-year survival. This performance remained consistent when the model was tested in an external group of 67 patients. To better understand what drove these predictions, SHAP analysis showed that tumor stage (T and N stage) and use of chemotherapy had the strongest influence on survival estimates.

Outcomes and Implications

This study shows that machine learning models can help predict survival outcomes in female bladder cancer more accurately than traditional statistical methods. The study depicted a successful case of combining established clinical factors with AI, highlighting the potential for more personalized follow-up planning and treatment decisions after surgery. The use of SHAP also improves transparency, making the predictions easier for clinicians to understand and trust. However, the study is retrospective and relies mainly on SEER data, which limits generalizability. Studies using larger and more diverse patient groups will be needed before these models can be used more broadly in clinical care.