Relative risk trees for censored survival data

Area of Science:

• Biostatistics
• Survival Analysis
• Machine Learning

Background:

• Estimating relative risk for censored survival data is crucial in medical research.
• Existing tree-structured methods have limitations in handling complex survival data.
• Recursive partitioning offers a flexible framework for survival analysis.

Purpose of the Study:

• To develop a new method for obtaining tree-structured relative risk estimates.
• To adapt the Classification and Regression Tree (CART) algorithm for survival data.
• To evaluate the performance of the proposed method against existing techniques.

Main Methods:

• A recursive partitioning algorithm is employed, utilizing the first step of a full likelihood estimation.
• The algorithm adopts key aspects of the Classification and Regression Tree (CART) methodology.
• Performance is assessed through simulation studies.

Main Results:

• The developed method generates tree-structured relative risk estimates.
• Simulations demonstrate the technique's performance.
• Comparisons are made with established tree-structured survival analysis methods.

Conclusions:

• The proposed method offers a viable approach for tree-structured relative risk estimation in survival analysis.
• This technique provides an alternative to existing methods for analyzing censored survival data.
• Further investigation into its application and performance is warranted.