Tree-structured prediction for censored survival data and the Cox model

Area of Science:

• Biostatistics
• Medical Informatics
• Clinical Research

Background:

• Survival data analysis is crucial for understanding disease progression and treatment efficacy.
• Prognostic classification and subgroup analysis are key components of clinical research.
• Existing methods may not fully capture complex interactions within survival data.

Purpose of the Study:

• To explore the utility of prediction trees in survival data analysis.
• To demonstrate the application of prediction trees for prognostic classification.
• To showcase the use of prediction trees in detecting treatment-covariate interactions for subgroup analysis.

Main Methods:

• The study outlines the Recursive Partitioning and Amalgamation (RECPAM) approach for tree-growing.
• Prognostic classification and subgroup analysis are formulated within the RECPAM framework.
• Cox proportional hazards models with a priori strata are utilized.

Main Results:

• Prediction trees effectively summarize prognostic information from covariates.
• Trees are useful in detecting and displaying treatment-covariate interactions.
• Two data analysis examples illustrate the practical application of the methods.

Conclusions:

• Prediction trees are a versatile tool for survival data analysis.
• The RECPAM framework provides a robust method for prognostic classification and subgroup analysis.
• Cross-validation is discussed as a model selection criterion.