Integrating preliminary test and Stein-type techniques to improve estimation in the time-dependent Cox model

Area of Science:

• Statistics
• Biostatistics
• Survival Analysis

Background:

• Shrinkage estimation is well-established for linear and static Cox models.
• Integrating shrinkage into time-dependent Cox models is challenging due to evolving risk sets and nonhomogeneous data.
• Previous methods lacked a unified theoretical framework for dynamic semiparametric survival settings.

Purpose of the Study:

• To develop a unified theoretical framework for shrinkage estimation in time-dependent Cox proportional hazards models.
• To extend classical Stein-type theory to dynamic semiparametric survival data.
• To achieve a principled balance between estimation efficiency and robustness.

Main Methods:

• Extension of classical Stein-type theory to dynamic semiparametric survival data.
• Development of a positive-rule Stein estimator.
• Theoretical analysis of unbiasedness and variance attenuation properties.
• Monte Carlo simulation studies and empirical application.

Main Results:

• The positive-rule Stein estimator maintains unbiasedness under valid restrictions.
• It adaptively reduces variance inflation when restrictions are approximately correct.
• Demonstrated substantial efficiency gains compared to unrestricted and adaptive LASSO estimators.
• Validated theoretical advantages through simulations and a real-world dataset.

Conclusions:

• The proposed unified framework effectively addresses shrinkage estimation in time-dependent Cox models.
• The positive-rule Stein estimator provides a robust and efficient statistical strategy.
• This approach enhances estimation accuracy in complex survival data analysis.