Efficient Computation of High-Dimensional Penalized Piecewise Constant Hazard Random Effects Models.

Area of Science:

• Biostatistics
• Genomics
• Survival Analysis

Background:

• Proportional hazards mixed effects models (PHMMs) are crucial for analyzing time-to-event data with clustered correlations in biomedical research.
• High-dimensional data presents challenges in specifying and computationally handling fixed and random effects in PHMMs.

Purpose of the Study:

• To develop a computationally efficient method for variable selection in high-dimensional survival data.
• To approximate PHMMs with a more tractable piecewise constant hazard mixed effects survival model.
• To enable simultaneous selection of important fixed and random effects.

Main Methods:

• Approximation of PHMMs using a piecewise constant hazard mixed effects survival model.
• Parameter estimation via a modified Monte Carlo expectation conditional minimization (MCECM) algorithm.
• Incorporation of a factor model decomposition for random effects to enhance scalability.

Main Results:

• The proposed method effectively performs simultaneous variable selection on fixed and random effects.
• The factor model decomposition aids in scaling the variable selection to higher dimensions.
• Demonstrated utility through simulations and application to a pancreatic cancer gene expression dataset.

Conclusions:

• The developed approach offers a scalable and effective solution for variable selection in high-dimensional survival analysis.
• The method aids in identifying key features influencing survival outcomes, particularly in complex datasets like gene expression studies.