Shape restricted additive hazards models: Monotone, unimodal, and U-shape hazard functions

Area of Science:

• Statistics
• Biostatistics
• Survival Analysis

Background:

• Semiparametric additive hazards models are crucial for survival data analysis.
• Estimating models with unspecified unimodal hazard functions presents challenges for traditional methods like the proportional hazards model.
• The complexity of partial likelihood in proportional hazards models limits their application in settings with unknown modes.

Purpose of the Study:

• To develop an efficient estimation method for semiparametric additive hazards models with unspecified unimodal hazard functions.
• To address the computational inefficiencies of standard quadratic programming methods in high-dimensional survival data.
• To extend the proposed method for time-dependent covariates with monotone or U-shaped hazard functions.

Main Methods:

• A quadratic loss function is defined, enabling a parameter-free global Hessian matrix.
• A quadratic programming method is proposed for profiling potential modes.
• The quadratic pool adjacent violators algorithm is introduced to reduce computational costs.
• The method is extended to handle time-dependent covariates.

Main Results:

• The proposed quadratic pool adjacent violators algorithm significantly improves computational speed compared to standard quadratic programming.
• Simulation studies demonstrate reductions in bias and mean square error.
• The method was successfully applied to analyze data from a cardiovascular study.

Conclusions:

• The novel quadratic pool adjacent violators algorithm offers an efficient and accurate solution for estimating semiparametric additive hazards models with unimodal hazard functions.
• This method provides a valuable tool for survival data analysis, particularly in complex scenarios involving unknown hazard shapes.
• The approach is robust and applicable to time-dependent covariates, broadening its utility in biostatistical research.