Alternative tree-structured survival analysis based on variance of survival time

Area of Science:

• Biostatistics
• Medical Informatics
• Survival Analysis

Background:

• Tree-structured survival analysis (TSSA) is a key method for analyzing survival data in medical research.
• Existing TSSA methods often rely on log-rank statistics or martingale residuals.
• Alternative TSSA construction methods include Lp Wasserstein metrics and impurity-based approaches.

Purpose of the Study:

• To explore the utility of the degree of separation (DOS) index in TSSA.
• To propose a new TSSA algorithm based on the variance of restricted mean lifetimes.
• To compare the performance of the proposed TSSA method against conventional approaches.

Main Methods:

• Developed a TSSA algorithm analogous to least squares regression trees.
• Utilized the variance of restricted mean lifetimes as the primary criterion for node partitioning.
• Applied the algorithm to prospective cohort data from the Study of Osteoporotic Fracture.
• Compared the proposed method's classification rules with existing TSSA methods.

Main Results:

• The proposed TSSA algorithm, based on variance of restricted mean lifetimes, demonstrated competitive performance.
• The new method showed potential as an alternative to log-rank and martingale residual-based TSSA.
• Application to the Study of Osteoporotic Fracture data provided real-world validation.

Conclusions:

• The variance of restricted mean lifetimes serves as a viable index for constructing survival trees.
• The proposed DOS-based TSSA algorithm is a promising alternative for survival data analysis.
• Further simulation studies suggest its competitiveness against established TSSA techniques.