BlaCaSurv

BlaCaSurv predicts five-year survival probabilities for bladder cancer patients using machine learning models trained on SEER (2000–2017) data.

Key Features:

• Data source: Patient records from the Surveillance, Epidemiology, and End Results (SEER) database covering 2000–2017.
• Cohort: 52,529 microscopically confirmed adult bladder cancer patients.
• Predicted outcome: Five-year survival probability.
• Dataset split: Random 70/30 division into training and test sets.
• Algorithms evaluated: Logistic regression, support vector machine (SVM), gradient boosting, random forest, and K nearest neighbor (KNN).
• Model inputs: Patient features were used as predictors to train the models.
• Performance assessment: Models were compared using standard performance metrics and discrimination measures.
• Selected model: Gradient boosting demonstrated superior predictive ability and discrimination and was selected as the best-performing algorithm.

Scientific Applications:

• Individual prognosis: Estimate five-year survival probabilities for bladder cancer patients.
• Clinical decision support: Provide prognostic information to inform treatment planning and patient counseling.
• Comparative model evaluation: Evaluate and compare machine learning algorithms for bladder cancer survival prediction.

Methodology:

Analysis used SEER (2000–2017) data of microscopically confirmed adult bladder cancer patients randomly split 70/30 into training and test sets; logistic regression, SVM, gradient boosting, random forest, and KNN models were trained on patient features to predict five-year survival and compared using standard performance metrics, with gradient boosting selected as best-performing.