GSC

GSC predicts individualized survival outcomes for patients with glioblastoma multiforme (GBM) using statistical and machine learning models to inform prognostic estimates.

Key Features:

• Data source and population: Models were developed using the Surveillance Epidemiology and End Results (SEER) database (2005–2015) comprising 20,821 surgically treated, histopathologically confirmed GBM patients.
• Input variables: Models were trained on 13 features spanning demographic, socioeconomic, clinical, and radiographic variables.
• Predicted outputs: The tool produces estimates of overall survival, one-year survival status, and personalized survival curves.
• Algorithms evaluated: Fifteen statistical and machine learning algorithms were compared for prediction performance.
• Best-performing model: The accelerated failure time (AFT) model achieved superior performance with a concordance index of 0.70 and outperformed Cox proportional hazards regression.
• Evaluation criteria: Models were assessed for discrimination, calibration, interpretability, predictive applicability, and computational efficiency.

Scientific Applications:

• Clinical prognosis: Provides individualized survival estimates to support clinical decision-making and patient counseling for GBM.
• Model development framework: Serves as a comparative framework for developing and validating predictive survival models in oncology.

Methodology:

Models were trained and compared across fifteen statistical and machine learning algorithms using 13 demographic, socioeconomic, clinical, and radiographic features from SEER (2005–2015) for 20,821 surgically treated, histopathologically confirmed GBM patients, producing overall survival, one-year survival status, and personalized survival curves, with the accelerated failure time model achieving a concordance index of 0.70 and compared against Cox proportional hazards regression.