SEER Survival Analysis Tool

The SEER Survival Analysis Tool predicts mortality in patients with retinoblastoma using a random survival forest model trained on Surveillance, Epidemiology, and End Results (SEER) program data, providing prognostic insights for clinicians.

Key Features:

• Random Survival Forest Model: Employs a random survival forest (RSF) algorithm with optimized parameters (mtry = 10, nodesize = 25, ntree = 160) to predict patient mortality.
• Feature Importance Ranking: Calculates feature importance under the random forest framework to identify prognostic factors, with Stage identified as the most crucial variable, followed by M stage and T stage.
• Multi-timepoint Prediction: Predicts mortality at 3, 5, and 10 years with area under the receiver operating characteristic curve (AUC) evaluation.
• Internal Validation: Validates predictive performance using a 7:3 training-validation split with consistency index (C-index) and calibration curves.
• Thirteen Prognostic Features: Incorporates stage, T-stage, M-stage, surgery radiation sequence, radiation, systemic therapy surgery sequence, age, chemotherapy, surgery, race, residence, sex, and primary sequence.

Scientific Applications:

• Retinoblastoma Prognosis: Evaluates prognosis of high-risk retinoblastoma patients using simple variables from the SEER database.
• Clinical Decision Support: Provides experience for the diagnosis and treatment of retinoblastoma patients in clinical practice.
• Mortality Risk Stratification: Identifies high-risk patients based on tumor stage and metastasis status to guide therapeutic interventions.

Methodology:

Data were extracted from the SEER database using SEER Stat Version 8.4.3 for 577 patients diagnosed with retinoblastoma from 2000 to 2019. The cohort was randomly split 7:3 into training and validation sets. Feature importance was calculated under the random forest framework. Optimal mtry (10) and nodesize (25) were determined by out-of-sample error, and optimal ntree (160) by learning curve. Model performance was evaluated by C-index, calibration curves, and AUC.